Modelling our Future, Volume 15: Population Ageing, Social Security and Taxation (International Symposia in Economic Theory and Econometrics) (International ... in Economic Theory and Econometrics)

MODELLING OUR FUTURE POPULATION AGEING SOCIAL SECURITY AND TAXATION

International Symposia in Economic Theory and Econometrics Series Editor: Volume 14:

William A. Barnett Economic Complexity Edited by W. A. Barnett, C. Deissenberg & G. Feichtinger

International Symposia in Economic Theory and Econometrics Volume 15

MODELLING OUR FUTURE POPULATION AGEING SOCIAL SECURITY AND TAXATION EDITED BY

Ann Harding National Centre for Social and Economic Modelling, University of Canberra, Australia

Anil Gupta Applied Research and Analysis Directorate, Health Canada, Canada

Amsterdam – Boston – Heidelberg – London – New York – Oxford Paris – San Diego – San Francisco – Singapore – Sydney – Tokyo

Elsevier Radarweg 29, PO Box 211, 1000 AE Amsterdam, The Netherlands The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, UK First edition 2007 Copyright r 2007 Elsevier B.V. All rights reserved No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means electronic, mechanical, photocopying, recording or otherwise without the prior written permission of the publisher Permissions may be sought directly from Elsevier’s Science & Technology Rights Department in Oxford, UK: phone (+44) (0) 1865 843830; fax (+44) (0) 1865 853333; email: [email protected]. Alternatively you can submit your request online by visiting the Elsevier web site at http://elsevier.com/locate/permissions, and selecting Obtaining permission to use Elsevier material Notice No responsibility is assumed by the publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made Library of Congress Cataloging-in-Publication Data A catalog record for this book is available from the Library of Congress British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library ISBN-13: 978-0-7623-1343-3 ISBN-10: 0-7623-1343-9 ISSN: 1571-0386 For information on all Elsevier publications visit our website at books.elsevier.com Printed and bound in The Netherlands 07 08 09 10 11 10 9 8 7 6 5 4 3 2 1

To James and Jack Sekoranja, with love always

Introduction to the Series The series International Symposia in Economic Theory and Econometrics publishes quality proceedings of conferences and symposia. Since all articles published in these volumes are refereed relative to the standards of the best journals, not all papers presented at the symposia are published in these proceedings volumes. Occasionally these volumes include articles that were not presented at a symposium or conference, but are of high quality and are relevant to the focus of the volume. The topics chosen for these volumes are those of particular research importance at the time of the selection of the topic. Each volume has different co-editors, chosen to have particular expertise relevant to the focus of that particular volume. William A. Barnett Series Editor

Contents About the Editors

xi

Notes on Contributors

xiii

Foreword

xix

Preface

xxi

Chapter 1: Introduction and Overview Ann Harding and Anil Gupta Part I:

1

Pension Analysis Using Dynamic Microsimulation

Chapter 2: Can We Afford the Future? An Evaluation of the New Swedish Pension System Lennart Flood Chapter 3: A Microsimulation Model of Private Sector Pensions in France Thierry Debrand, Sophie Pennec and Anne-Gise`le Privat Chapter 4: Effects of Demographic Developments, Labour Supply and Pension Reforms on the Future Pension Burden in Norway Dennis Fredriksen and Nils Martin Stølen Chapter 5: Macroeconomic Effects of Proposed Pension Reforms in Norway Dennis Fredriksen, Kim Massey Heide, Erling Holmøy and Ingeborg Foldøy Solli Chapter 6: Adding Private Pensions to the Canadian DYNACAN Model Richard J. Morrison

33

55

81

107

143

viii

Contents

Chapter 7: Post-Secondary Education and Training Participation Rates in Australia in the Next 30 Years: A Microsimulation Approach Sandra Roussel

PART II:

171

Taxes, Benefits and Labour Supply

Chapter 8: Lifetime Redistribution Through Taxes, Transfers and Non-Cash Benefits Thomas Pettersson and Tomas Pettersson

205

Chapter 9: Income Distribution and Redistribution in a Medium-Term Perspective in Denmark Frederik Hansen

233

Chapter 10: Population Ageing and Fiscal Sustainability: Integrating Detailed Labour Supply Models with CGE Models Rolf Aaberge, Ugo Colombino, Erling Holmøy, Birger Strøm and Tom Wennemo

259

Chapter 11: Canadian Retirement Behaviour: A Microeconomic Examination Greg Maloney, Mashood Mirza and Franc- ois Paris

291

Chapter 12: STINMOD: Use of a Static Microsimulation Model in the Policy Process in Australia Rachel Lloyd

315

Chapter 13: The Impact of Canadian Population Ageing on Federal Personal Income Tax: Microsimulation Results from 2000 to 2026 Weng-Fong Lu, Wei Li and Earl Bailey Chapter 14: Basic Income or Vital Minimum? A Note on the Distributive Effects of Possible Reforms of the Spanish Income Tax Xisco Oliver and Amedeo Spadaro

335

361

Contents

Part III:

ix

Wealth and Services

Chapter 15: Self Provision in Retirement? Forecasting Future Household Wealth in Australia Simon Kelly

387

Chapter 16: Household Net Wealth in New Zealand: Implications for Retirement Incomes Grant M. Scobie and John Gibson

407

Chapter 17: Population Ageing, Housing Demand and House Prices in Australia: A Microsimulation Analysis Ross S. Guest

435

Chapter 18: Population Ageing and State Concessions Policy Karen Piper and Don Siemon Chapter 19: Future Service Provision at the Local Level: Demographic Modelling and Forecasting in the City of Bayside Simon Byth

459

483

Chapter 20: Characteristics and Travel Patterns of Older Australians: Impact of Population Ageing on Tourism Afzal Hossain, Geoff Bailey and Milly Lubulwa

501

Index

523

This page intentionally left blank

About the Editors Professor Ann Harding Director, National Centre for Social and Economic Modelling, University of Canberra For the past 14 years, Professor Ann Harding has spearheaded the development of highly sophisticated microsimulation models and databases within Australia, so that policy makers can gain much better information about the likely distributional impact of current and proposed policies. She leads the University of Canberra’s National Centre for Social and Economic Modelling, established in January 1993, and is a professor of Applied Economics and Social Policy at the University. In recent years, with substantial grant funding, Professor Harding has steered microsimulation modelling in Australia beyond its traditional ‘tax and social security’ focus to such new areas as health, aged care, housing and regional issues, with the goal of extending sophisticated quantitative decision-support tools to policy makers in these areas. Ann has published widely on income inequality, poverty and the distributional impact of government programmes, and is a prolific contributor to public policy debate in Australia, with her work typically being cited every week in the media. In 1996 she was elected a fellow of the Academy of the Social Sciences in Australia and in 2003 was elected president of the International Microsimulation Association. Professor Anil Gupta Founding Director, Microsimulation Modelling and Data Analysis Group, Health Canada For over 20 years, Anil has been on the forefront of popularizing the use of microsimulation in public policy by building and successfully integrating their use in the policy development process in the areas of taxation and health. He played a pivotal role in major tax policy reforms in Canada. He is the founding director of Microsimulation Modelling and Data Analysis group at Health Canada. Among his many notable endeavours, he has worked with Harvard University to develop a summer course on Tax Analysis and Revenue Forecasting where he taught the same during the summers of 1994–2000, and since 2001 at Duke University. He has helped several

xii

About the Editors

countries in Asia and Central Europe in building tax analysis and forecasting tools. He is a recipient of the prestigious Queen Elizabeth 50th centenary gold medal award for distinguished public service in Canada. Anil is credited with organizing two international microsimulation conferences in USA and Australia. He edited the volume Microsimulation in Government Policy and Forecasting in 2000 under the Elsevier series Contributions to Economics Analysis. Anil is a founding member of the Editorial Board of The International Journal of Microsimulation, and an associate editor of Journal of Statistics & Management Systems (JSMS). Anil holds a PhD in mathematics from the University of Calgary and a masters in systems science from the University of Ottawa.

Notes on Contributors Rolf Aaberge is a Senior Research Fellow at the Research Department of Statistics Norway. He is also a Research Fellow at CHILD, Torino and at IZA, Bonn. His fields of work are labour supply and taxation, public finance, social policy, poverty, income distribution and social welfare and econometrics. Earl Bailey worked as a senior information analyst and project leader in the Canada Public Service for 29 years, working within the Treasury Board Secretariat and then the Canada Revenue Agency, Ottawa until his retirement in August 2005. His projects focused mainly on data mining and information analysis to support policy issues. Geoff Bailey is currently the principal statistical analyst for Tourism Division, Department of Industry, Tourism and Resources, Canberra. He has also worked as an economist for a number of other Australian Government institutions including the Bureau of Tourism Research, the Australian Bureau of Agricultural and Resource Economics and The (Commonwealth) Treasury. Geoff holds an honours degree in Agricultural and Resource Economics from the University of New England. Simon Byth worked for Bayside City Council as its senior research coordinator, Social Research, providing demographic input for local service planning. He holds an MBA from RMIT in Melbourne, Australia, and has consulted with a range of organizations on the implications of local area demography for service planning and strategy development. He is currently working in the Victorian Department of Education and Training. Ugo Colombino holds a PhD from the London School of Economics and is a professor of microeconomics at the Department of Economics Cognetti De Martiis, University of Turin, Italy. He also holds a part time position at the Research Department of Statistics, Norway in Oslo. He is a member of CHILD (Centre for Household, Income, Labour and Demographic economics). His main research interests are in microeconometric modelling and distributional and incentive effects of taxes and social policies.

xiv

Notes on Contributors

Thierry Debrand is a senior researcher at IRDES, the Institute for Research and Information in Health Economics. He received a PhD from the University of Paris 2. His special research interest is on the impact of public decisions on individual behaviour. He has also published in ageing, health and retirement. He is currently responsible for the French part of the International Survey on Health and Retirement (SHARE). Lennart Flood has been professor of econometrics at the School of Business, Economics and Law, Go¨teborg University since 1996. His research covers econometrics of micro data, application of time use data, labour supply, taxes and pensions. He has also been involved in the development of the Swedish microsimulation model SESIM. Dennis Fredriksen is a research fellow at the Research Department of Statistics, Norway. His main research area has been development, maintenance and use of the dynamic microsimulation model MOSART, including computer programming. John Gibson is a professor in the Department of Economics, University of Waikato in New Zealand. He was previously professor and chairperson of the Department of Economics, University of Canterbury and has a PhD from Stanford University. Ross S. Guest is a professor of economics in the Griffith Business School, Griffith University, Australia. He has been at Griffith since 1998, prior to which he was a lecturer in economics at Monash University, Melbourne, from 1991. His main research interests are in the macroeconomics of population ageing, a topic on which he has published extensively. Anil Gupta is the founding director of Microsimulation Modelling and Data Analysis group at Health Canada, where he has worked since 1999. For over 20 years, Anil has been on the forefront of popularizing the use of microsimulation in public policy by building and successfully integrating their use in the policy development process in the areas of taxation and health. He played a pivotal role in major tax policy reforms in Canada. He has worked with Harvard University to develop a summer course on Tax Analysis and Revenue Forecasting where he taught the same during the summers of 1994–2000, and since 2001 at Duke University. He has helped several countries in Asia and Central Europe in building tax analysis and forecasting modelling tools. Frederik Hansen is a senior advisor at the Ministry of Finance, Denmark. He has been active in the modelling of benefits and taxes, their interaction

Notes on Contributors

xv

and income distribution analysis using the Danish Law model system (a microsimulation model) for almost 20 years. Ann Harding is director of the University of Canberra’s National Centre for Social and Economic Modelling (NATSEM), established in January 1993, and is a professor of Applied Economics and Social Policy at the University. For the past 14 years, she has spearheaded the development of highly sophisticated microsimulation models and databases within Australia, so that policy makers can gain much better information about the likely distributional impact of current and proposed policies. In 1996, she was elected a fellow of the Academy of the Social Sciences in Australia and in 2003 was elected president of the International Microsimulation Association. Kim Massey Heide is an economist at the ‘‘Unit for Public Economics’’ at the Research Department, Statistics, Norway. His main research fields are Applied General Equilibrium Analysis of Ageing, Skill-Based Technical Change & Human Capital Investment. Erling Holmøy has been head of the ‘‘Unit for Public Economics’’ at the Research Department of Statistics, Norway since 2005. His main research fields are applied general equilibrium analysis of trade liberalization, tax reforms, energy policy, ageing and pension reforms. Afzal Hossain worked for Tourism Research Australia and published a wide range of research papers, conference papers, niche market report and consultancy reports. He was nominated for the Tourism Global Award for his outstanding research activities in 2004. Currently, Dr Hossain is working in the Transport Research Branch at the Australian Department of Transport and Regional Services. Simon Kelly is an associate professor at the University of Canberra and a principal research fellow at the National Centre for Social and Economic Modelling. He is currently the chief investigator for a five-year research project assessing the social and fiscal policy implications of an ageing population. Simon’s research interests are population ageing, labour force trends, household wealth and retirement savings. Wei Li is a project leader, Compliance Research Division at the Canada Revenue Agency. He was an associate professor of Peking University, visiting fellow of Australian National University, and senior analyst of Human Resource Development, Canada. Rachel Lloyd was a principal research fellow at NATSEM at the University of Canberra when this work was undertaken. At NATSEM, she undertook

xvi

Notes on Contributors

a wide range of microsimulation modelling projects and research into poverty and the socio-economic factors affecting the use of information technology. She is currently working at The Treasury, Australia. Weng-Fong Lu obtained his PhD in agricultural economics from the University of Manitoba, Winnipeg, Canada. He has served the Federal Government of Canada for over 30 years, in applied research in Agricultural Economics, Taxation and Social Welfare. Milly Lubulwa has managed the Analysis Section at Tourism Research Australia (TRA) for over four years. Previously, she worked for more than 10 years at the Australian Bureau of Agriculture and Resource Economics (ABARE) as a manager of the Survey Data Analysis Section. Greg Maloney is a senior researcher at the Canada Revenue Agency. He specializes in a number of research areas, including tax incentives, retirement and, most notably, tax compliance research. Mashood Mirza was a Research Analyst in the Statistics Division at the Canada Revenue Agency. Currently, he is working as a Tax Policy Officer in the Tax Policy Branch at Finance Canada. Richard J. Morrison has spent his career designing and building policy models that governments actually use. The best-known examples include MAPSIT to calculate benefits and taxes as functions of earnings (32 years of use), SIMTAB to perform cross-sectional microsimulation (26 years of use), and DYNACAN to carry out longitudinal simulation of the Canada Pension Plan (12 years of use). Xisco Oliver is a professor of applied economics at the University of Balearic Islands (Spain) where he obtained his PhD in Economics. He has a Master in Applied Economics from the University Pompeu Fabra (Spain). His research has focused on portfolio choice, labour supply and taxation, redistribution of income and social welfare. Frac- ois Paris has held numerous positions at the Canada Revenue Agency over the years. He is now Manager of the Individual Statistics and Modeling Section, which is responsible for the delivery of individual tax data to the various department of Finance at the federal and provincial levels. Sophie Pennec is a researcher with the Institut National d’Etudes De´mographiques (INED), Paris. She is interested in demographic and economic issues and in the construction of dynamic microsimulation models. Sophie is

Notes on Contributors

xvii

currently spending a year with NATSEM at the University of Canberra, having won a French–Australia collaborative grant. Thomas Pettersson is a senior advisor at the Swedish Ministry of Finance. His main interest is income distribution, work incentives and simulation models. Tomas Pettersson is a desk officer at the Swedish Ministry of Finance. His main interest is simulation models, demography and income distribution. Karen Piper has been manager of the Concessions Unit, Department of Human Services, Victoria since 1995. The Unit is responsible for monitoring concessions policy across the Victorian Government and directly responsible for managing the contractual arrangements for the delivery of energy, water and municipal rates concessions. Anne-Gise`le Privat received her PhD from the Institute of political sciences and has been a member of the National Institute of Population Studies. She joined the Caisse nationale d’assurance vieilesse in 1998. Currently, she is demo-economist at the Department of Health. Her research interests include retirement and health. Sandra Roussel was formerly a research economist at the Australian Department of Education, Science and Training. She is currently working at the Australian Department of the Treasury. Grant M. Scobie is a principal advisor in the New Zealand Treasury. He holds degrees from Massey University and the University of New England, and a doctorate in economics from North Carolina State University. He was formerly a professor of economics at the University of Waikato in New Zealand. Ingeborg Foldøy Solli is an economist in the ‘‘Unit for Pubilic economics’’ at the Research Department of Statistics Norway. Her main research fields are applied general equilibrium analysis of ageing and fiscal sustainability, microsimulation modelling and analysis of the Norwegian tax-benefit system. Amedeo Spadaro is a researcher at Paris-Jourdan Sciences Economiques (PSE), at FEDEA (Madrid) and professor of public economics at the University of the Balearic Islands (Mallorca). His current research activities are concerned with labour supply and taxation, public finance, income distribution and social welfare, poverty, econometrics and statistical methods and microsimulation, and also include management of several national and

xviii

Notes on Contributors

international research programmes on evaluation of welfare and tax policies in welfare states. Nils Martin Stølen is a senior research fellow at the Research Department of Statistics, Norway. His main research areas have been in macroeconomics, labour market, wage formation, public finance, taxes and pensions. In the last few years his main effort has been in analyses of the effects of reforms of the Norwegian pension system, using the dynamic microsimulation model MOSART and at present he is the head of the MOSART team. Birger Strom is a Senior Advisor in the Research Department of Statistics Norway. His main working fields include national accounts, programming and use of computable general equilibrium models in studies of tax reforms, environmental policy, energy economics and long-run macroeconomic projections. Tom Wennemo is an advisor in the Research Department of Statistics, Norway at present and his fields of work are estimation and simulation of microeconometric models.

Foreword This book, providing a number of interesting and useful analyses on simulating public spending and tax revenues in face of demographic changes in society, is very welcome. It should be a required reading for anyone working in this field, as well as for policy makers interested in understanding how ageing in society will affect public finances. Demographics are a critical subject for governments to understand in terms of their implications for economic policy. Good research on the impact on ageing in terms of its impact on government spending and taxation will be required in the next several years to lay the foundation for better policies to be put into place so that industrialized economies will be ready to deal with the large number of retirees that may not have the resources to meet their financial needs for income, health expenditures and long-term care. At the same time, the low birth rate has important ramifications for labour force growth in industrialized nations that will need to build up capital resources to ensure economic growth and productivity. Education and child support programmes will be less costly but with a low birth rate, working people will be harder to employ through businesses. Immigration will increasingly be important in a world with greater imbalances in both labour and capital markets. We are venturing into a world of labour shortages rather than focussing on unemployment. This book along with its companion volume Modelling our Future: Population Ageing, Health and Aged Care will provide a ‘‘how-to’’ analysis to experts and officials who need to consider their data base development and testing to determine the impact of ageing on microeconomic policy. Over the coming years, more work will appear in this area — this book will help researchers develop their skills in predicting the impact of demographic changes on governments. It is a good vital read for all. Jack Mintz Deloitte LLP Professor of Taxation J. L. Rotman School of Management University of Toronto and President and CEO C. D. Howe Institute Toronto

This page intentionally left blank

Preface In December 2003, the editors of this volume (and its companion volume) organized a conference entitled ‘‘International Microsimulation Conference on Population Ageing and Health: Modelling Our Future’’ in Canberra, Australia. The conference focused on the recent advances in microsimulation modelling and its direct use in the policy development process around the world. The data challenges associated with the building of these tools received considerable attention at the conference. In addition, two days were dedicated to practical model descriptions and a symposium on modelling efforts in different countries provided further insights into the state of modelling today and its future direction. The wide spectrum of issues covered and the participation of researchers, policy makers and consultants from around the world at the conference provided a venue for intense discussions. Further, an overwhelming response to the technical workshop on practical models motivated the editors of this volume to take this work one step further through this volume and the companion volume. This volume incorporates a selection of refereed papers from those delivered at the Conference, particularly focussing on pensions, social security, taxation, wealth and services. Many of the papers involve applications of dynamic microsimulation models, which have flourished during the past decade as concern about the fiscal and other impacts of population ageing has intensified. The editors would like to thank Rachel Lloyd for her outstanding assistance in the organisation of the conference and in the early stages of producing these two volumes, and would like to also acknowledge the assistance provided by Rebecca Cassells of NATSEM. The financial support provided by the following organizations helped the authors to concentrate on the content of the conference and this was largely responsible for making the conference such a success: Applied Research and Analysis Directorate, Health Canada; Australian Bureau of Statistics; Department of Health and Ageing, Australia; Canada Revenue Agency, Canada; National Centre for Social and Economic Modelling, Australia; SAS; and Statistics Canada, Canada. The editors are grateful to the Hon. Kevin Andrews, Minister for Employment and Workplace Relations, Australia and Dr. Michael Wolfson of Statistics Canada for their keynote addresses, which contributed greatly in steering the conference towards the key public policy issues and challenges facing world governments today. The editors would also like to thank the

xxii

Preface

session chairs for their valuable contributions, including George Rothman, Michael Wolfson, Jane Falkingham, Maria Evandrou, Lennart Flood, Mark Thomann, Agnes Walker, Marion McEwin, Michael Dunn, Marilyn Biviano, Wendy Stone, Rick Morrison, Michael McCracken, Paolo Roberti, Rachel Lloyd, Michel Lamure, Richard Eason, Michael Dunn and John O’Leary. Particular thanks are also due to the referees, who did an excellent job of reviewing the papers. We have included some of their comments in the introductory chapters and very much appreciated their efforts, which greatly improved the quality of the papers and ensured that this became a refereed volume. Very special thanks are also due to Monic Gupta, who edited the volumes with enthusiasm and efficiency. Finally, the authors wish to express to their spouses, Chitralekha Gupta and John Sekoranja, a profound sense of appreciation for their continued support, understanding and patience in the long and arduous task of putting these two volumes together. Their forgiveness and patience brought this work to fruition. Ann Harding and Anil Gupta

Chapter 1

Introduction and Overview Ann Hardinga and Anil Guptab a

National Centre for Social and Economic Modelling, University of Canberra, Australia b Applied Research and Analysis Directorate, Health Canada, Canada

Abstract The impacts of population ageing are being experienced already in most OECD countries, with governments concerned about the social and fiscal implications of population ageing in the coming decades. After a general introduction, the second section of this chapter examines the modelling techniques that can be used to assess the likely impact of population ageing, including an outline of the different types of microsimulation models available. The third section provides an overview of each of the chapters contained within this volume.

1. Introduction Most OECD (Organisation for Economic Co-operation and Development) countries are facing the prospect of rapid demographic change in the decades ahead. In the immediate aftermath of World War II, many countries experienced a baby boom. The ‘baby boomers’, born between and around 1946–1960, form a large population bulge. The oldest are now aged around 60 years and are thus rapidly nearing the ‘official’ retirement age of 65 years. As a result, many OECD countries face the prospect of the mass retirement of this large cohort during the next two decades. Two other factors have also changed greatly in the past 60 years, producing the phenomenon of population ageing. First, the post-war baby boom was followed by the baby bust. In Australia, for example, fertility fell from 3.5 births per woman in the early 1960s to just under two children per woman by the early 1980s. Since then the fertility decline has been much more gradual — but fertility has still inched down to about 1.8 children for each woman today. International Symposia in Economic Theory and Econometrics, Vol. 15 Copyright r 2007 Elsevier B.V. All rights reserved ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15001-2

2

Ann Harding and Anil Gupta

The second key factor has been the continuous increases in life expectancy. Taking Australia again as an example, a baby born at the beginning of the last century could expect to live less than 60 years on average. From a social policy perspective, the problem of funding decades of retirement simply did not exist as, on average, Australians died before they reached retirement age. In sharp contrast, an Australian woman born today can expect to live about 83 years (with men living about five years less on average). And these figures may even be understated, with some demographers predicting more rapid increases in lifespan than current official projections. Not surprisingly, the combination of the baby boom bulge, falling fertility rates and increasing longevity has been ringing alarm bells among governments across the world. For example, while in 1960 there were about 7.3 working age Australians to help support each retiree aged 65 years and over, by 2040 there are forecast to be only about 2.4 working age Australians for each retiree aged 65 and over. Across the OECD nations, most of which have elaborate and complex welfare state programmes, policy makers have been attempting to estimate the likely future fiscal and social consequences of population ageing. In Australia, the Commonwealth Government’s initial assessment of these issues was contained in the 2002 Intergenerational Report. This report suggested that under current projections of fertility rates, labour force participation rates and so on — and assuming that current policy settings remained unchanged — there would be a shortfall between Commonwealth government revenue and outlays of five per cent of Gross Domestic Product by 2042. In today’s dollars, this would translate into a budget deficit of about Au$40 billion. The report concluded that resolving this budget shortfall would require either higher taxes upon future generations or reductions in spending programmes (or some combination of these) (Treasury, 2002, p. 6). This initial analysis for the Commonwealth budget was subsequently extended to examine the implications of population ageing for State and Territory governments. The Productivity Commission (2005) report concluded that the fiscal gap for all levels of government would reach 6.4 per cent of GDP by 2044–2045. While escalating health, aged care and pension costs were seen as playing an important role in creating this fiscal gap, the impact of population ageing on labour supply and economic growth was also expected to be important. Because people aged 55 and over have lower labour force participation rates than those who are younger — and, when they work, are more likely to work part-time rather than full-time — population ageing reduces both labour force and economic growth rates. This concern with the fiscal and other pressures created by population ageing has led to similar efforts in other OECD countries to assess the likely future magnitude of those pressures. Policy makers are questioning whether

Introduction and Overview

3

the generous pay-as-you-go pension schemes that exist in many of these countries will be affordable in the future. Can the future health and aged care needs of the baby boomers be met without draconian increases in the tax burdens of the generations that follow the baby boomers? Will population ageing reduce economic growth to such an extent that it will threaten future growth in tax revenues? Are the labour supply and retirement decisions of the baby boomers being unduly influenced by provisions within tax and cash transfer programmes that now seem outdated, such as financial incentives for early retirement? As the OECD’s chief economist observes: ‘To cope with mounting financial pressures, governments have to make hard choices’ (Cotis, 2003). To assist politicians to make these hard choices, many countries are now developing sophisticated models to help them simulate the impact of policy changes and plan their futures. In December 2003, many of the researchers and government policy makers involved in these initiatives attended the ‘Modelling Our Future’ conference in Canberra, Australia. Selected papers from this conference, which focus on income, social security and taxation, are contained within this volume — while those spanning services such as health and aged care, along with descriptions of microsimulation models from around the world, are contained in the companion volume to this one (Gupta and Harding, 2007). These two volumes are designed to serve a number of purposes. First, for those researchers whose countries have already developed relatively advanced modelling infrastructure, many of the papers in these volumes provide insights into the initiatives being undertaken at the cutting edge of the discipline. They thus provide glimpses of what is already possible at the frontiers of modelling our socio-economic futures — or of what is rapidly becoming possible. Second, while extremely complex models now exist to answer these types of questions in many OECD countries, such models are less common in Japan and in the Asia-Pacific region, as well as in developing nations. A second purpose is thus to illustrate the types of approaches that can be used and the range of policy and other questions that can be answered, for those whose countries have not yet invested the resources needed to develop such modelling infrastructure. To this end, some of the chapters focus on illustrating the use of models, rather than directly simulating our futures. Finally, a third purpose is to provide an overview of the different types of models being used and, in particular, to summarise trends in microsimulation modelling over the past decade. The companion volume to this one also contains descriptions of the microsimulation models being actively used in policy formulation across the world. For those new to modelling, this will provide basic knowledge to help them start developing such models and for experts this will open up international collaboration on a large scale to promote best practice in this ever-growing field.

4

Ann Harding and Anil Gupta

2. Modelling our future There are a wide range of possible methods that can be used to try to simulate our social and economic futures and the impact of policy change. This section first discusses microsimulation techniques, before proceeding to examine cell based and other techniques. 2.1 Microsimulation The defining characteristic of microsimulation models is that they analyse the likely behaviour of and the impact of policy change upon persons (or households, or firms, or other micro-units). They are often constructed on top of microdata, with sample surveys or administrative data forming typical base datasets for such models. In both cases, the microdata usually contain thousands of individual or micro-unit records, with a host of variables describing the demographic, labour force, income and other characteristics of each individual or family. Microsimulation models were the brainchild of Guy Orcutt (1957) who, frustrated by the macroeconomic models of the day, proposed a new type of model consisting of interacting, decision-making entities such as individuals, families and firms. Static Microsimulation Models Microsimulation models have traditionally been divided into two broad categories — static and dynamic — although the boundaries between the two have become increasingly blurred. Static models typically use static ageing techniques to update cross-sectional microdata up to the required point in time. These techniques usually involve reweighting the data and uprating monetary values. When national statistical agencies issue a micro-data file from a national sample survey, they typically attach a weight to each individual record in the file, representing the estimated number of comparable individuals in the entire country with similar characteristics to the individual who took part in the sample survey. This weight is the means used to ‘gross up’ from sample survey results to estimates for the entire population. Reweighting involves replacing these original weights with amended weights, to take account of change in the population structure (by age, gender, labour force status, etc.) between the time the data were collected and the desired time period of the analysis. Uprating involves updating monetary values to reflect changes since the data were collected, such as inflating earnings or private rent paid. After these two steps, static modellers typically impute the receipt of social security and other benefits and/or income tax or other liabilities, by applying the rules for eligibility or liability to each of the micro-units. Steps

Introduction and Overview

5

may also be taken to make the number of programme participants or taxpayers match to external benchmark data by, for example, simulating less than 100 per cent take-up of a particular payment. At this point, a baseline data file has been generated, which usually shows the current incomes and characteristics of each person or family in the microdata file, plus the imputed current incidence of taxes paid to and cash transfers received from government. Most static models also allow the analyst to vary the rules of eligibility or liability, and produce output showing the distributional gains or losses for each micro-unit — and the budgetary impact for government — from the policy change. Importantly, the impact can be assessed for the entire population or for relatively large or small population sub-groups (such as sole parent pensioners with two children or the home-owning elderly living in a particular region). Thus these models permit winner/loser analysis of policy changes and thus allow the user to develop policy parameters in line with governments’ overall policy objectives. It was these unique features which led an exhaustive review in the United States to conclude that no other type of model can match microsimulation in its potential for flexible, fine-grained analysis of proposed policy changes (Citro and Hanushek, 1991, p. 115). Static microsimulation models for assessing the distributional and revenue impacts of possible changes in tax and cash transfer policy have been the ‘bread and butter’ of microsimulation for some decades now. Such models have often played a decisive role in determining the final shape of policy reforms introduced by governments, and they are now widely used across Europe, the US, Australia, and Canada.1 In recent years, their use has expanded to analyse policy issues in the area of health policy and the companion volume to this one covers the latest developments in this important area of public policy. The recent Nordic microsimulation seminar in Oslo (June, 2006) included a special session on health and microsimulation, where researchers from different countries presented efforts in the areas of pharmaceuticals, health status and health promotion.2 Most static microsimulation models still measure the immediate (or firstround) impacts of policy change, before individuals change their behaviour in response to the policy shock. Yet, in some cases, a policy change may be expected to have a major impact upon behaviour — and may even be designed to do so. Consequently, a growing number of modellers have

1 For example, results from the STINMOD static microsimulation model influenced the final shape of the GST-tax reform package introduced in Australia in 2000 (Harding et al., 2000) while, in the US, the TRIM model has influenced public policy for many years (see http://trim3.urban.org/). 2 See http://www.ssb.no/english/research_and_analysis/conferences/misi/

6

Ann Harding and Anil Gupta

attempted to simulate the changes in the behaviour of the individuals directly affected by a policy shock — for example, by allowing labour supply or consumption patterns to vary in response to a tax change.3 Yet even these models often remain very much at the micro level, ignoring macroeconomic consequences. For example, they may assume that a cut in marginal tax rates leads to an increase in the labour supply of particular individuals — but what if there are no suitable jobs available for these individuals to fill? As explained by Citro and Hanushek, the behavioural responses in the short or long term by those directly affected by a policy change will subsequently lead to second-round effects — that is, ‘to effects that alter the nature of factor or product markets or the level and distribution of consumption, production and employment in the economy or in a sector of it affected by the policy change’ (1991, p.178). For example, a cut in marginal income tax rates might initially increase labour supply but subsequently lead to cuts in wage rates and changes in output by industry. Consequently, efforts are now also being made to link microsimulation models to macroeconomic models, to capture these other second-round or feedback effects of a policy change (Arntz et al., 2006 and Fredriksen et al. in Chapter 5 of this volume). Dynamic Microsimulation Models Dynamic models often start from exactly the same cross-section sample surveys as static models. However, rather than using the static ageing procedures described above, dynamic ageing involves ‘updating each attribute for each micro-unit for each time interval’ (Caldwell, 1990, p. 5). Thus, the individuals within the original microdata or base file are progressively moved forward through time: this is achieved by making major life events — such as death, marriage, divorce, fertility, education, labour force participation, etc. — happen to each individual, in accord with the probabilities of such events happening to real people within that particular country. Thus, within a dynamic microsimulation model, the characteristics of each individual are recalculated for each time period. There are two major types of dynamic microsimulation models, both of which are illustrated in these two volumes. Dynamic population models

3

For example, see the MITTS model developed at the Melbourne Institute in Australia (http://melbourneinstitute.com/labour/behavioural/mitts.html) and the wide range of papers presented at the 8th Nordic Seminar on Microsimulation Modelling on labour supply responses (http://www.ssb.no/english/research_and_analysis/conferences/misi). For an example of simulating changes in household consumption as a result of indirect tax changes, see Symons and Warren (1996).

Introduction and Overview

7

involve ageing a sample of an entire population, and typically begin with a comprehensive cross-section sample survey for a particular point in time (such as from a census). To date, such dynamic models have been used for such purposes as the analysis of retirement incomes, future health status, the long-term impact of social security amendments, and the lifetime redistributive impact of the social security system (e.g. Nelissen, 1996; Wolfson, 1988; Bonnet and Mahieu, 2000; Favreault and Caldwell, 2000; Hancock, 2000; and see also O’Donoghue, 2001 for an extremely useful summary of dynamic modelling research). Dynamic cohort models use exactly the same type of ageing procedures, but usually age only one cohort rather than the many cohorts represented in an entire population. Typically, one cohort is aged from birth to death, so that the entire lifecycle is simulated. For some applications, such models are more cost-efficient than ageing an entire population. Such models have been used to analyse lifetime income distribution and redistribution, lifetime rates of return to education, and repayment patterns for student incomecontingent loans (Hain and Helberger, 1986; Harding, 1993a, 1993b, 1995; Falkingham and Hills, 1995; Falkingham and Harding, 1996; Baldini, 2001). Like static models, dynamic models face problems when attempting to incorporate either behavioural change by individuals or second-round macro-economic effects in response to government policy changes. Economists would normally assume that any ‘dynamic’ model would incorporate behavioural change. However, while dynamic microsimulation models typically do capture some types of behavioural change (e.g. a woman within a model might leave the labour force with the birth of her first child), they do not necessarily allow for changes in the behaviour of individuals or macroeconomic change initiated by government tax-transfer policy change. Recent Developments Since the early 1990s, microsimulation has flourished. At least three factors have helped to generate this growth. First, the availability of suitable microdata has improved greatly. In Australia, for example, the first public use cross-sectional microdata files were not made available by the national statistical agency until the mid-1980s (with such microdata usually forming the all important base file of microsimulation models). Similarly, the first comprehensive panel study of individuals did not begin in Australia until 2001 (Weston and Wooden, 2002), thus removing a major impediment to attempts to estimate the crucial transition probabilities that underlie dynamic models. In Europe, cross-national data collection efforts such as the European Community Household Panel vastly improved the availability of both cross-sectional and panel data in many member countries. The efforts of Tim Smeeding to develop internationally comparable household level

8

Ann Harding and Anil Gupta

data through the Luxembourg Income Study, along with the initiatives of the Canberra Group, also contributed to improvements in microdata availability and comparability.4 A second factor has been the growing demand by policy makers for the types of analyses that only microsimulation models can provide. In many countries, social and tax policies have become ever more complex, with the introduction of new, targeted benefits and the endless fine-tuning of existing programmes. As the interactions between the various tax and transfer programmes have increased, it has become more difficult to estimate the impact of policy change from first principles — so that microsimulation models have come to play a vital role in highlighting the (sometimes unexpected) distributional consequences of policy change. In this environment, static microsimulation has blossomed. Sutherland (2000) has played an important role in expanding static microsimulation across Europe, with a series of grants from the European Commission funding the development of EUROMOD for the original 15 EC members and then subsequently its extension to the 10 new EC countries. Numerous new country-specific models have also been developed since the early 1990s. Within Australia, NATSEM’s STINMOD microsimulation model has flourished and been made publicly available — while Wolfson and others within Statistics Canada have continued efforts to make microsimulation accessible to Canadians, including through the publicly available SPSD/ M static microsimulation model (Murphy, 2000). Similarly, Gupta and Kapur (Gupta and Kapur, 1996; Kapur and Gupta, 2007) have developed such models from within the ministries of Finance (TTSIM) and Health (PHARMASIM) for use in the policy development process in the area of taxation and health. Again with on-going government funding, the Urban Institute in the US developed TRIM-3 and made it accessible through the web (O’Hare, 2000). Among the many other country-specific models, many of which are usefully reviewed in Sutherland (1995), are GLADHISPANIA for Spain (Chapter 14 in this volume); the LAW model for Denmark (Pedersen, 2000); MODETE and ASTER for Belgium (Decoster, 2000; Cape´au et al., 2006), the LOTTE model for Norway and the FASIT model in Sweden. Equally important in stimulating the practice of microsimulation has been the growing concern about the social and economic impacts of population ageing. This has encouraged the development of dynamic microsimulation models, which have properties that are uniquely suited to analysing the long-term and future impact of changes in pension and other policies (Harding, 2000). While Orcutt and his fellow researchers had completed

4

See http://www.lisproject.org/links/canbaccess.htm

Introduction and Overview

9

construction of the DYNASIM dynamic microsimulation model of the US by the mid-1970s (Orcutt et al., 1976), dynamic microsimulation subsequently languished both in the US and internationally in the 1980s (Caldwell, 1996, p. 506). Some of the models constructed in the 1980s have apparently ‘died’, including the German SfB3 model (Galler and Wagner, 1986), the Australian HARDING model (1993a), the Canadian dynamic cohort model DEMOGEN (Wolfson, 1988), and the Swedish MICROHUS model (Klevmarken and Olovsson, 1996). (In many cases they have been replaced by newer and more complex models.) Despite the relatively slow progress made in the 1980s, the picture was very different in the 1990s. Often prompted by concerns about the long-term sustainability of public pension programmes, government agencies across the world funded the development of dynamic microsimulation models — including DYNACAN within the Canadian Department of Human Resources, LIFEPATHS within Statistics Canada (Gribble, 2000), MOSART within Statistics Norway, SESIM within the Swedish Ministry of Finance; PenSim within the UK Department of Work and Pensions; Destinie within the French National Statistical Institute (INSEE); PENSIM within the US Department of Labour; and MINT within the US Social Security Administration (with descriptions of most of these models being in the companion volume to this one). Some of these models borrowed heavily from dynamic microsimulation models constructed in the 1980s and extended in the 1990s — with both DYNACAN and MINT, for example, utilizing much of the US CORSIM model developed by Caldwell (1996). Outside government — although generally again with public funding — academics also constructed dynamic microsimulation models to answer similar questions, including the SAGE and PSSRU/NCCSU-aged care models for the UK, DYNAMOD (and now APPSIM) for Australia (Kelly and King, 2001), SVERIGE for Sweden (see the model description chapters for these four models in the companion volume to this one) and SMILE in Ireland (Ballas et al., 2005). Apart from the injection of much-needed funds prompted by concerns about population ageing, another factor assisting the production of microsimulation models was the vast and on-going improvements in computer hardware (and, to a lesser extent, software). These improvements have made the production of microsimulation models and the storage of their results much more feasible. Technology transfer has also been assisted by regular international gatherings of the world community of microsimulators. Four of those gatherings have since resulted in international edited collections, including from the 1993 conference in Canberra, Australia (Harding, 1996); the 1997 conference at Maine in the US (Gupta and Kapur, 2000); the 1998 workshop at Cambridge in the UK (Mitton et al., 2000) and the 2003 conference in

10

Ann Harding and Anil Gupta

Australia (this volume and the companion volume). In addition, microsimulation has been strongly supported within the Nordic countries, resulting in five Nordic seminars on microsimulation models being held between 1995 and 2006. Importantly for the future, at the 2003 conference in Australia the International Microsimulation Association was formed and the International Journal of Microsimulation was launched — and a website has since been established through the efforts of Paul Williamson at the University of Liverpool (www.microsimulation.org). The association thus provides an excellent means for the world community to keep in electronic touch. The Future Microsimulation appears likely to continue to expand in many directions. Geographically, the discipline is still primarily confined to Europe and the English-speaking world. However, there is increasing evidence of expansion into both Asian and developing countries — including China, Russia and Africa and often with encouragement from the UN WIDER institute (Davies, 2004; Xiong et al., 2006; http://models.wider.unu.edu/africa_web/). In a world of limited resources and competing demands, the role of microsimulation is expected to grow more quickly than even in the past two decades. The increased reliance on policies aimed at redistribution of income in a revenue-neutral environment point to targeted policies, which are most efficiently developed using the tools provided by microsimulation. Lately, there have been concerted efforts to promote the use of microsimulation in developing economies, through the teaching of techniques at such centres as the Duke Centre for International Development, NATSEM and the OECD. The subject area of microsimulation also continues to expand. There has been an explosion in the cost of public health care across OECD countries and modern governments are concerned about their ability to continue to deliver health services for the ageing population in an environment where the health workforce itself is ageing. Against this backdrop, models of the health system are becoming more common, as the companion volume to this one makes clear, including of pharmaceutical subsidies and health human resources (Brown et al., 2007; Gupta and Basu, 2007; Kapur and Gupta, 2007). Similarly, a fast-growing new area is spatial microsimulation, for predicting the local effects of policy change and future small area populations and service needs (Williamson et al., 1998; Voas and Williamson, 2000; Ballas et al., 2005; Brown and Harding, 2005; Chin et al., 2005, Chin and Harding, 2006; 2007; Cullinan et al., 2006; King, 2007). Finally, it is clear that there will be continuing efforts to include both the behavioural responses of individuals and macro-economic effects within the ambit of microsimulation (Klevmarken, 1997). Within Australia, the MITTS model is being used to estimate the labour supply effects of policy

Introduction and Overview

11

changes (Creedy et al., 2002) while, in addition to the chapters contained within this volume, there are ongoing efforts to link macro and micro models (Arntz et al., 2006; Davies, 2004). 2.2 Other Modelling Approaches This introduction has focused heavily on microsimulation — as dynamic microsimulation, in particular, provides one of the most appropriate methods for examining the lifetime and future impacts of possible policy responses to population ageing. But there are also many other possible modelling techniques that can help to shed light on our socio-economic futures. Hypothetical or Typical Taxpayer Models Hypothetical models analyse the situation of an individual (or family), whose particular characteristics are defined by the model user. Thus, a hypothetical model might look at the future superannuation of a single woman earning average earnings and working full-time for 20 years and part-time for 20 years. Such models are very useful for analysing policy outcomes for particular types of individuals and can incorporate enormous detail. The RIMHYPO model developed by the Australian Treasury is a good example of this type of model (Gallagher, 1995; http://rim.treasury.gov.au/), as is the Effective Tax Rates Model developed at NATSEM (Toohey and Beer, 2004). Similar models, known as Typical Taxpayer Models (TPU) have been extensively used at the Department of Finance in Canada since the early 1980s. These models are very useful in the early stages of the policy development process, in developing the right parameters to achieve stated policy objectives. Hypothetical models are especially useful for providing information about how a particular type of family or person is likely to be affected by an existing programme or by a possible policy change. What such models cannot tell you, however, is how many such families there are in the population. As a result, such models cannot be used to obtain estimates of aggregate budget outcomes or distributional outcomes for the whole population. Group Models Group (or cell-based) models divide the population into a set of groups defined by characteristics such as year of birth, sex, marital status and labour force status. The extent of diversity is thus effectively constrained by the number and type of groups specified. The output from such models is again very useful to policy analysts seeking to understand the likely impacts of population ageing. A good

12

Ann Harding and Anil Gupta

example of this approach is the RIMGROUP model developed by the Australian Treasury, for analysing the future impact of superannuation and other policy changes (Bacon, 1999). The chapter by Hossain et al. in this volume also provides another example of this type of approach. What are some of the limitations of group models? First, there are constraints on the diversity of the groups that can be considered. Suppose, for example, that a modeller wanting to examine future government health expenditure decided to create a group model with 20 cells — 10 age groups multiplied by two genders. Suppose further that research in that particular country had shown that income also played a major role in the usage of health services and that it was important to also include this factor in any model projecting future outlays. If the population was now divided into 10 income deciles, then there would be 200 groups in the model to keep track of in future years (10 age ranges  2 sexes  10 income groups). Suppose further that whether or not the individuals were still living with a spouse emerged as another key predictor of future expenditure on nursing homes and health. Then, for each of the 200 groups, there would have to be a further division into married/not married, resulting in 400 groups to keep tabs on. It is thus easy to see how group models can easily become intractable, once more sophisticated modelling is desired. Keeping the number of groups down to manageable levels often means failing to capture the diversity present within the population. And to the extent that such diversity and different characteristics are in fact key predictors of whatever the model is looking at, inaccuracy is introduced into the results. Another difficulty is that people can generally not move from one group to another. As a result, for example, such models usually cannot cope with changes in circumstances, such as divorce. They also cannot cope easily with fluctuating individual earnings over a lifetime, with people typically being assigned to a lifetime earnings group and then experiencing the average outcomes for their group. Thus, for some population sub-groups of interest — such as sole parents for example — group models cannot provide comprehensive estimates of likely asset holdings and retirement incomes. Finally, group models provide details of the average experience for each of the groups specified within the model. They do not usually provide details of the degree of dispersion around that mean experience. For example, rather than just showing the average age pension received by a particular group within the model, policy makers might want to know what proportion of that group received no age pension and thus just how skewed the distribution of the pension was. Despite these undoubted disadvantages, group models have one major advantage relative to dynamic microsimulation models: they are much cheaper and quicker to build. Thus, such models provide a very useful tool for gaining insights into the likely impact of population ageing, without

Introduction and Overview

13

having to wait five years for the construction of a full dynamic population microsimulation model. Stock– Flow Models The stock–flow approach is used in the forecasting of health human resources. For example, physician services are a flow over the course of a year, delivered by a stock of physicians working at a certain level of intensity. The stock of physicians is best modelled via a stock/flow approach, a common and well-accepted form of modelling. Stocks represent quantities at a point in time, such as the number of active physicians at the end of a calendar year. Flows, on the other hand, represent quantities that change over a period of time (e.g. annually in a model). Comparing two points in time shows changes in the stock. The change in stock, or net flow, is usually a consequence of a series of contributing flows. For example, for physicians, the key flows are the numbers of physicians entering and exiting the workforce. The flows can be further disaggregated, with outflows being categorised into retirements, emigration and death and inflows into new entrants and immigrants. Inter-provincial migration, on the other hand, can be either an inflow or outflow. Disaggregating flows allows the supply model to be changed at very detailed levels. Consequently, more control translates into more policy levers. These models are especially useful in capturing both lags and leakages from flows. A critical lag, for example, is how long it takes a new medical school student to graduate and start to practice. Leakages are another form of flow. For instance, the number of students who do not complete their training would be considered a leakage. Such stock–flow models are very effective policy tools in the planning of health human resources. Biviano and Tise (2007) describe such a model in the context of nursing resources for the US and Gupta and Basu (2007) elaborate on the building of such tools for Canada. By incorporating both demand and supply in their models, Gupta and Basu are able to project ‘gaps’ in the future supply of physicians and provide policy relevant solutions to fill projected ‘gaps’ using policy levers on the supply side in their model. Econometric and Statistical Models A range of other modelling techniques have also been used in these two volumes to examine the impacts of population ageing, including life cycle models of housing tenure choice and savings behaviour and econometric models estimated from both cross-sectional and longitudinal data. Such models provide useful insights into particular aspects of the consequences of population ageing and are much easier and cheaper to construct than fullfledged dynamic population microsimulation models — although, again, they lack the richness of dynamic microsimulation models.

14

Ann Harding and Anil Gupta

3. Overview of this Book Part I: Pension Analysis Using Dynamic Microsimulation The first five chapters in this book provide excellent examples of one of the key uses of dynamic microsimulation models today: simulating the likely future costs and benefits of current pension schemes and possible reforms to those schemes. Chapter 2 by Flood presents a fascinating analysis of how different cohorts will be affected by pension reform in Sweden — as well as providing a very useful introduction to a typical structure and operation of a dynamic microsimulation model for those unfamiliar with such models. Pension reform in Sweden appears to be more advanced than in many other OECD countries. Concerned that the changing age structure would challenge the financial sustainability of its established pension systems, a new public pension system was introduced in Sweden in 1999. The generous pay-as-you-go defined benefit scheme was largely replaced by a pay-as-you-go defined contribution scheme — in which current contributions were used to finance current pensions being paid, but future pensions were directly linked to contributions paid into each individual’s account. A key function of the SESIM dynamic microsimulation model has been to assess the financial sustainability of this new Swedish pension system and the differential impact on birth cohorts. A crucial public policy issue for countries with social insurance schemes is that many have provided those in their 50 s and early 60 s with strong incentives to retire early. As Cotis notes: Old age pension schemes also stack the cards in favour of people retiring early. If people postpone their retirement by a year, this is rarely reflected in correspondingly higher pensions later on, despite their extra contributions (2003, p. 2). A particularly interesting feature of the Swedish reforms is that the annual benefit level is calculated by dividing the total contributions in the individual account by age-specific life expectancy (which is directly affected by retirement age). Thus, the penalty for an early retirement is high. In addition, increases in life expectancy across Sweden generally will also be reflected in changes in the pension paid to those retiring. Overall, Flood concludes that the new pension scheme is less generous than the old, with later cohorts thus doing less well out of the system. For example, while the middle 50 per cent of the 1940 cohort are expected to receive disposable incomes between the ages of 65 and 69 that reach 83 per cent of their incomes during the five years immediately preceding retirement, the comparable replacement rate for the 1960 birth cohort will be only 74 per cent.

Introduction and Overview

15

Still within Europe, Chapter 3 moves on to examine the likely future impact of the 2003 pension reforms introduced within France, using the ARTEMIS dynamic microsimulation model. Projections suggest that the ratio of retired to active workers within France will rise from 44 per cent in 2000 to 83 per cent in 2040, raising concerns about the financial sustainability of its pay-as-you-go pension system. Most current retirees rely on this system, as nearly 70 per cent of their income is provided by their pension. This chapter by Debrand et al. provides a relatively detailed account of the construction of the ARTEMIS model, which simulates the population of the French mandatory state-monitored pension scheme for private-sector employees (the ‘general scheme’), covering nearly 68 per cent of the labour force in France. The chapter provides another interesting illustration of the use of dynamic microsimulation to inform pension policy, comparing the world existing before and after the 2003 pension reforms. The 2003 reforms lengthened the contribution period required for entitlement to a full pension and introduced new rules for calculating the pension rate when a person retired before 65 without the contribution period required to obtain full rate. Interestingly, after 2008, future changes will be based on the principle of sharing future increases in life expectancy equally between working life and retirement, so that the reform will be continued in such a way as to maintain a constant ratio between length of working life and length of retirement. Debrand et al. show that the 2003 reforms result in estimated total expenditure on the pension system in 2030 being 10 billion euros (or nine per cent) less than under the pre-reform system. Later birth cohorts are expected to receive lower pensions under the reformed system, with the average pensions paid at retirement to those born after 1955 being some 7 to 11 per cent lower than they would have been had the 2003 reforms not occurred. Reform of the public pension system is also high on the policy agenda within Norway. A series of white papers released by the government suggested that expenditure on old age and disability benefits in the National Insurance Scheme would increase from about 9 per cent of GDP in 2002 to about 20 per cent in 2050. Financing this would require an increase in the contribution rate from about 15 per cent today to about 25–30 per cent after 2040. The MOSART dynamic microsimulation model has been playing an active role in policy formulation in Norway, by shedding light on the effects of different reforms discussed by the recent Norwegian Pension Commission (NOU 2002, 2004). A wide range of policy reform options are modelled by Fredriksen and Stølen in Chapter 4, including indexing future pension entitlements by less than wages growth; indexing entitlements by wages growth but indexing pensions after retirement only by the increase in consumer prices; and moving towards a more actuarially fair system (smaller pensions for early

16

Ann Harding and Anil Gupta

retirees or if life expectancy increases). Their analyses suggest that the forecast 27.5 per cent contribution rate in 2040 could be reduced to around 24 per cent if the actuarial element was strengthened (by reducing yearly pension benefits as a consequence of early retirement and higher life expectancy). Indexation of pension benefits by less than wages would also reduce the forecast contribution rate. However, they point out that the various reforms have differential impacts by gender, with some of the proposed reforms disadvantaging women compared with the current system. Chapter 5 also deals with Norway, but provides a very interesting extension to existing dynamic microsimulation modelling practices, by linking the MOSART microsimulation model to a large-scale dynamic CGE model (called MSG6). The chapter by Frederiksen et al. projects the macroeconomic development of the Norwegian economy until 2050 under different public pension systems. The two models are run iteratively to obtain consistency. To the direct (or first-round effects) estimated from the microsimulation model, they add the equilibrium adjustments calculated from their CGE model arising from changes in government expenditures, labour supply incentives and private savings induced by the pension reforms. Frederiksen et al. find that if the existing Norwegian pension system remained unchanged, a broad tax on labour income such as a payroll tax would have to be raised from the present level of about 13 per cent to about 25 per cent in 2050 in order to finance public expenditures. The authors run two possible pension reform scenarios, including the ‘more actuarially fair’ system described in the preceding chapter. Iterating between the microsimulation and CGE models, they find that this actuarially fair pension reform would allow the payroll tax rate to be reduced in 2050 to 10.9 per cent. This is clearly a very different outcome to the expected 25 per cent rate expected in the absence of pension reform. The change is driven by employment being 10.6 per cent higher in 2050 than under the ‘no-reform’ scenario, which in turn stimulates a roughly equivalent percentage increase in GDP. More than half of the increase in employment is due to Norwegians delaying their retirement. This chapter thus represents part of the continuing attempts by researchers to move beyond simply presenting the direct (first round) effects of policy reform, to also including behavioural and general equilibrium effects. In Chapter 6, Morrison reminds us of the numerous challenges faced by dynamic microsimulation modellers — in this case, a lack of the data ideally required to adequately impute private pensions onto the DYNACAN Canadian dynamic microsimulation model. Like the other dynamic models discussed in earlier chapters, the DYNACAN model is a relatively mature dynamic microsimulation model which, by the close of the 1990s, had become a part of the formal process for assessing proposed changes to the Canada Pension Plan (CPP). The CPP is the public, earnings-related,

Introduction and Overview

17

contributory pension component of Canada’s social security system — but most large employers, and virtually all levels of government within Canada, also provide supplementary private pension plans. Accordingly, accurate modelling of retirement incomes and the impact of policy changes requires that these private pensions be included within the DYNACAN model. During the earlier phases of the DYNACAN model, Morrison (2000) made a contribution to dynamic microsimulation modelling via development of ‘alignment’ and ‘variance reduction’ techniques. This chapter extends that contribution by pointing to possible methods for imputing private pensions in the face of very severe data deficiencies. For example, at the time of constructing their prototype module of private pensions, the DYNACAN team did not have access to any data source that indicated whether the private pension amounts received by individuals’ were due to their own pension rights, or were received by them as a survivor’s pension because their spouse had died. They also faced a lack of longitudinal data that would have helped them simulate private pensions. This chapter describes the ingenious modelling solutions developed and implemented by the DYNACAN team, which will provide a guide to others seeking to simulate such pensions in the face of similar data deficiencies. Their future development plans include the use of marginal tax rates in the calculation of CPP benefit to provide the true impact of any policy changes in the CPP programme parameters.

Part II: Taxes, Benefits and Labour Supply As the chapters in Part I illustrated, dynamic microsimulation models are frequently used to analyse the future impact of changes in pension programmes. But the activities of government extend well beyond public pension schemes — for example, to include taxation, social services and the provision of other benefits. Chapter 7 represents an unusual application of a dynamic microsimulation model, examining the likely future demand for post-secondary education and training out to 2027 in Australia. A number of earlier papers projecting participation rates in education and training in Australia have found that the ageing of the population will lead to lower engagement in education. A standard corollary of this finding is that the proportion of government funding devoted to education is expected to decline, particularly as the government comes under increasing pressure to provide more services for the aged. However, while the direct effect of ageing reduces participation in education and training, there may be countervailing effects on educational participation via, for example, technological change, rising levels of prior educational attainment, and changes in the structure of the labour force.

18

Ann Harding and Anil Gupta

To account for some of these additional effects, the microsimulation model detailed in Chapter 7 has been developed by Roussel to analyse the simultaneous impact of demographic, labour force and educational attainment changes on participation rates in education and training. Roussel concludes that the potential for diverting funding from education and training to pensions and health care may be much more constrained than analyses based solely on the ageing of the population have suggested. Chapter 8 also employs a dynamic microsimulation model, but this time the focus is on developing broader measures of economic well-being that include not only taxes and transfers but also non-cash benefits — as well as on extending the measurement period to estimate lifetime redistribution. Pettersson and Pettersson use the SESIM dynamic microsimulation model for Sweden to produce their results, adding to the standard SESIM model the cost of provision of those non-cash benefits that bestow personal benefits upon individuals, including subsidies for education, child care, old-age care, labour market activities, health care and medications. Like such previous studies as Harding (1993a), they find that lifetime income is far more evenly distributed than annual income and that the redistributive impact of the public sector is stronger in any given year than over the lifetime. Over the lifetime, they found that only about one-fifth of all redistribution was interpersonal (e.g. from rich to poor), while the remainder was intrapersonal (e.g. representing transfers from one period of an individual’s lifecycle to another period). They concluded that when analyzing redistributive systems — and when preparing or evaluating reforms — it was essential that the life cycle perspective was considered. The degree of selffinancing in public subsidies and transfers is extensive and only a small part of the redistribution conducted by the public sector results in actual reallocation of resources between individuals. In view of the expected future demographic developments, and the anticipated difficulties in the future financing of the public sector, they suggest that some inter-temporal redistribution could perhaps be handled at an individual level rather than through the state. In Chapter 9, Hansen also moves beyond the ubiquitous annual snapshots of income and distributional impact to look at longer time periods. However, in this case the methodology is not based on a dynamic microsimulation model: instead, Hansen sets up synthetic lifecourses, by matching panel records to other similar panel records, to derive five year and lifetime income estimates. He finds that the Gini-coefficient on five-year average incomes is around 2/3 of the single-year Gini-coefficient — and that the Gini-coefficient on lifetime income is around 50 per cent of the single-year Gini (a remarkably similar result to that found by Harding (1993a) for Australia). Hansen finds that the fraction of the population in relative poverty is lower when measured on longer-run average incomes and persistent relative poverty is almost non-existent in Denmark.

Introduction and Overview

19

In an application that will be of great interest to those accustomed to the ‘morning-after’ snapshots of the impact of policy change emerging from static microsimulation models, Hansen tests the difference made to estimates of the distributional impact of tax reform changes by using multiple years of data rather than one. His results — based on a newly developed multi-year tax model which uses the synthetic panel data — show that there may be fairly large differences in the apparent distributional effects, particularly for individuals aged 20–29 years, depending on whether the effects are measured over the short run or over the medium term. In the following two chapters the focus shifts again, this time to concentrate upon labour supply. As the OECD and others have noted, the possible work disincentive effects created by tax and outlay programmes have emerged as one of the key policy areas for government during the era of population ageing. As some of the chapters in Part I illustrated, ensuring that the provisions of pension and tax programmes do not unduly encourage early retirement or reduce labour supply has become a crucial public policy issue. However, much microsimulation modelling ignores labour supply issues, simulating the impact of policy change before individuals’ change their behaviour. Chapter 10 takes an alternative approach to shed light on this issue, linking a Norwegian Computable General Equilibrium (CGE) model with a detailed microeconometric labour supply model. The chapter by Aaberge et al. breaks new ground in the ongoing attempts to link macro and micro models. The authors are critical of the standard procedure underlying longrun CGE-studies of ageing — of letting a few representative agents determine the aggregate labour supply responses — and have instead developed a micro-econometric model of labour supply for a large representative sample of households. The micro–macro modelling framework works as follows: For given values of the after-tax real wage rate and non-labour income, the microeconometric model simulates the households’ labour supply for a representative sample of households. The assessed percentage change in the supply of man-hours is inserted into the CGE model, in which labour supply is exogenous. The CGE model then computes the equilibrium adjustments in the real wage rate, the revenue neutral tax rate and non-labour income. Next, the changes in these variables are used as a basis for changing the associated variables in the microeconometric model, which then produces new values for households’ labour supply. The authors conclude that while traditional modelling approaches would suggest that the payroll tax rate in Norway would have to double from 13 per cent in 1995 to 26 per cent in 2050 to achieve fiscal sustainability, the required estimated rate falls to 21 per cent when endogenous labour supply is also taken into account. They argue that this could be reduced even

20

Ann Harding and Anil Gupta

further if a flat tax rate was introduced. However, they note that the pattern of labour supply elasticities reveals that what matters in order to bring more people into the labour market is increasing the net wage for individuals living in low- and average-income households, which would suggest tax rate cuts focused on low- and average-income brackets. Overall, they suggest that the fiscal sustainability problems expected in future decades can be reduced to manageable dimensions, provided the tax system is reformed in order to improve the incentives for labour supply. Chapter 11 takes us around the globe to Canada, where much previous research has argued that the Canadian income security system significantly influences individuals’ retirement decisions, by providing labour supply disincentives. However, in Chapter 11, Maloney et al. argue that these studies have all had important limitations and, for example, have not been able to fully capture the effect of private pension plans; non-labour income from non-labour market activities, such as investment income and rental income; and Canada Pension Plan disability effects. Working inside the Canada Revenue Agency, the authors have access to particularly good microdata that allows them to examine these issues more thoroughly. They create an econometric model which draws upon the population microdata, which provides complete information on income security programmes, including entitlements and payouts, as well complete information relating to private pension plans and other sources of non-labour income. Using this more accurate dataset, they conclude that income security wealth has a minimal impact at best on the retirement decision of Canadians. Instead, their findings suggest the decision to retire is influenced by more prevalent factors, such as non-labour income and employer pension plans. The final three chapters in Part II deal with static microsimulation modelling applications. In Chapter 12, Lloyd provides a particularly clear description of the steps involved in the construction of a static microsimulation model — in this case the STINMOD tax and transfer model developed by NATSEM in Australia. For those new to static microsimulation, Lloyd provides some excellent illustrations of how microsimulation models can be used to inform the policy process, with her examples ranging from the distributional impact of a sweeping tax reform package to the costing of paid maternity leave. As noted earlier in this chapter, on occasion static microsimulation models have been reweighted to match future age/gender population benchmarks, to provide insights into the likely future profile of tax or outlay programmes (e.g. Walker et al., 2000, on future pharmaceutical benefits expenditure). Chapter 13 provides us with a further illustration of this technique, where Lu et al. have reweighted their base data out to 2026 benchmarks for Canada.

Introduction and Overview

21

Lu et al. have reweighted the T1 Tax Analysis static microsimulation model of the Canada Revenue Agency. Making assumptions about the likely rate of real income growth in future years, as well as the forecast demographic structure of the population, the authors conclude that total income tax collections will continue to increase over the 2001–2026 period — but that growth rates will slow over the 2011–2026 period. They show that, after 2011, when the proportion of the senior population starts to increase rapidly and the proportion of the working age population starts to drop, Canadian population ageing will have a negative effect on federal personal income tax revenue. Chapter 14 provides us with another illustration of the use of static microsimulation models. In this case Oliver and Spadaro use the GLADHISPANIA static microsimulation model to describe the 1999 Spanish tax system and simulate the impact of a wide range of reforms to that system. A key issue facing policy makers is the complexity of the Spanish income tax system. The authors estimate the distributional impact of replacing the current system with first, a ‘vital minimum’ (which consists of a tax allowance per equivalent adult and a flat tax on all remaining income) and, second, a ‘basic income’ (a universal lump-sum transfer that the government allocates to each household, independent of income and status plus a flat tax on all remaining income). Both of these options are constructed to be revenue neutral. The results show that the basic income option is much more redistributive.

Part III: Wealth and Services One of the key issues associated with population ageing is the extent to which older people will have the financial resources to support themselves and to make a contribution towards their health, aged care and other costs. In Australia, as in other countries, considerable attention has been devoted to the savings and wealth of current retirees and those nearing retirement (e.g. Kelly et al., 2002, 2004a, 2004b). Part of the rationale for attempting to persuade those in their 50 s and 60 s to remain in the labour force is so that they can boost their savings and superannuation. Accordingly, our next three chapters all deal with the crucial issue of wealth. In Chapter 15, Kelly uses the NATSEM dynamic microsimulation model, DYNAMOD, to examine the likely future wealth holdings of the baby boomers in retirement in Australia. Kelly shows that most current retirees have very little income other than the public means-tested age pension. However, he argues that it is important to move beyond income when examining the economic resources of older people, as he estimates that in 2001 older Australians have almost double the average wealth of working age Australians. Unfortunately, the majority of this wealth appears to be

22

Ann Harding and Anil Gupta

home equity, which is thus not producing income to supplement or replace the public pension and which may not be readily accessible for funding retirement needs. Kelly also projects the simulated distribution of wealth for older Australians in 2031, suggesting that while average wealth holdings for all age groups are projected to increase, the share of all household wealth held by those aged 65 and over will increase significantly — from 22 per cent of total wealth in 2001 to 47 per cent by 2031. Thus, Kelly concludes that almost half of all personal wealth will be in the hands of older people by 2031 and that an increased share of this wealth will be held in asset classes other than the family home, thus raising the possibility that older Australians will be able to self-finance more of their retirement income needs in the future. Moving across the Tasman, Chapter 16 examines whether New Zealanders are saving sufficient to finance their retirement. This immediately raises the question of how much saving is ‘enough’. Scobie and Gibson take as their measure of adequacy the capacity to sustain an equal level of consumption before and after retirement. They construct a mathematical model to estimate the saving rates that would be required to build present wealth (as observed in their Household Savings Survey) up to the level which would generate a post-retirement income adequate to sustain pre-retirement consumption. In addition they also derive estimates of the income in retirement that is implied by these estimated replacement rates and then calculate how many people would have incomes in retirement below 60 per cent of the median income of their cohort (i.e., a relative poverty line approach). Their results suggest that, for many low-income New Zealanders, not saving for their retirement is rational behaviour — because the universal non-means-tested public pension, New Zealand Superannuation (which provides an income for a couple equivalent to 66 per cent of the average wage), will finance a standard of living equivalent to their pre-retirement standard of living. They also conclude that only between 10 and 15 per cent of people could be expected to fall below 60 per cent of median predicted retirement income for this cohort. A major question for New Zealanders is thus whether New Zealand Superannuation as it is currently structured will be financially sustainable in the long run. In Chapter 17 the focus is on a subset of all wealth — namely housing wealth. As the earlier chapter by Kelly made clear, most of the wealth of older Australians is currently tied up in their home. Because the family home is such a crucial component of the retirement ‘nest egg’ in so many countries, the prediction by Mankiw and Weil (1989) that the decline in housing demand due to an older population would see real house prices fall by up to 40 per cent in the United States has spawned heated debate ever since. After providing a useful overview of this debate, in Chapter 17 Guest develops a simulation analysis of the impact of population ageing on

Introduction and Overview

23

housing demand and house prices in Australia for the period 2002–2052. The demand for owner-occupied housing is derived from a life-cycle model of housing tenure choice, which is then combined with a simple model of housing supply to analyse the interaction of housing demand, supply and house prices. The model predicts that from the year 2002–2052, population ageing will cause a drop in real house prices by up to 10 per cent in the absence of confounding factors. While the analysis in all the preceding chapters has been at the national level, population ageing is likely to have a pronounced fiscal impact upon provincial and local governments, as well as upon central governments. In Australia, there are eight State and Territory governments across the country — and they play a key role in providing such services as public education and public hospitals to their populaces. These State and Territory governments also frequently provide concessions targeted to lower-income households, to assist with the payment of essential services or state taxes or charges. In Chapter 18, Piper and Siemon describe how, out of its total budget of some $25 billion, the Victorian Government contributes over $700 million on these concessional subsidies (either through outlays or through foregone revenue). The Victorian Concessions Model is a static microsimulation model of the Victorian population’s use of household essential services, the annual bills incurred, and Victorian Government concessions expenditure. It is designed for policy analysis — in particular, allowing users to simulate the effects on household bills or the concessions budget of changes to electricity, gas, and water prices, changes in concession rates and entitlement, and so on. This chapter combines the population estimates arising from the Australian Intergenerational Report with use of the concessions model — and concludes that the number of concession card holders is likely to increase by 20 per cent more than the rate of population growth between 2003 and 2050. The authors note that other possible policy changes that are on the agenda would generate fiscal impacts of a similar magnitude to that expected from population ageing. While the authors conclude that real per capita spending will decline over the next 50 years, because the unindexed ‘cap’ on water and rates concessions will result in an erosion of their real value with inflation, the State government has since increased the cap and announced that it will be indexed — thus generating significant fiscal consequences for the future. In Chapter 19 we move to the third tier of government in Australia — in this case the Bayside City Council in Victoria. The Council has developed a cell-based model of population and housing — the ‘100 Streets’ Model — for monitoring of population and housing developments in the city and for assessing future policy. The model is uniquely informed by local knowledge and the data available to the city council. It operates through three components: population cohort component modelling, household formation modelling and residential housing modelling. Local administrative data is used to

24

Ann Harding and Anil Gupta

update and provide further detail to external data sets. Into the future, it is envisaged that trends identified through the development of the ‘100 Streets’ model will be used to develop further local knowledge of the community and the suburban region. The model has already produced interesting findings, such as that Bayside women in their 30 s have an exceptionally high ‘crude birth rate’, with consequent implications for the planning of services. Finally, while this volume has concentrated upon the consequences for government of population ageing, there will also be profound implications for industry. Some earlier Australian research, for example, has used static ageing procedures to reweight expenditure microdata files to expected future population targets, indicating a rosy future for golfing-related industries and a relatively dim outlook for those producing products and services for children (Harding et al., 2006). In most countries, the business sector has not yet begun to consider the impact of population ageing upon its activity — with larger typefaces for signs and labels, more comfortable clothing and more flexible working arrangements to help retain mature age workers simply representing the tip of the coming iceberg. In this vein, Chapter 20 makes a very useful contribution by examining the likely impact of population ageing upon tourism in Australia. Hossain et al. observe that in 2002 Australian seniors aged 55 years and over spent an estimated $10.8 billion on their domestic trips, accounting for 21 per cent of total expenditure on domestic tourism in that year. The likely change to the demographic mix of Australia’s population reveals the strong potential for the senior travel sector. The authors use a cell-based model to make their projections, combining Australian Bureau of Statistics (ABS) estimates of the future population classified by age and gender with their tourism activity and expenditure data. They conclude that seniors aged 55 years and over will generate around two-thirds of the forecast increase in tourism expenditure between 2002 and 2022. However, they suggest that the largest boost to spending is likely to occur in the next few years — as, after this, the baby boomers will enter the 65–84-year age group, who tend to spend somewhat less when they travel than ‘younger’ seniors.

References Arntz, M., Boeters, S., Gu¨rtzgen, N. and Schubert, S. (2006). Analysing Welfare Reform in a Microsimulation-AGE Model: The Value of Disaggregation. Paper Presented at the 8th Nordic Seminar on Microsimulation Models, Oslo, 7–9 June (available from www.ssb.no/misi). Bacon, B. (1999). Ageing in Australia: Some Modelling Results and Research Issues. Policy Implications of the Ageing of Australia’s Population, Productivity Commission, Melbourne.

Introduction and Overview

25

Baldini, M. (2001). Inequality and Redistribution over the Life-Cycle in Italy: An Analysis with a Dynamic Cohort Microsimulation Model. Brazilian Electronic Journal of Economics, 4(2). Ballas, D., Clarke, G. and Wiemers, E. (2005). Building a Dynamic Spatial Microsimulation Model for Ireland. Population, Space and Place, 11(3), 157–172. Biviano, M. and Tise, S. (2007). What is behind HRSA’S Projected Supply, Demand, and Shortages of Registered Nurses’, in Gupta, A. and Harding, A. (eds), Modelling Our Future: Population Ageing, Health and Aged Care, International Symposia in Economic Theory and Econometrics, North-Holland, Amsterdam. Bonnet, C. and Mahieu, R. (2000). Public Pensions in a Dynamic Microanalytic Framework: The Case of France, in Mitton, L., Sutherland, H. and Weeks, M. (eds), Microsimulation Modelling for Policy Analysis, Cambridge University Press, Cambridge. Brown, L., Abello, A., Phillips, B. and Harding, A. (2007). The Australian Pharmaceuticals Benefit Scheme and Older Australians: Changes in Government Outlays and Consumer Costs from the 2002–03 Federal Budget Measures, in Gupta, A. and Harding, A. (eds), Modelling Our Future: Population Ageing, Health and Aged Care, International Symposia in Economic Theory and Econometrics, North-Holland, Amsterdam. Brown, L. and Harding, A. (2005). The New Frontier of Health and Aged Care: Using Microsimulation to Assess Policy Options. Quantitative Tools for Microeconomic Policy Analysis, Productivity Commission, Canberra (available from www.pc.gov.au/research/confproc/qtmpa/index). Caldwell, S. (1990). Static, Dynamic and Mixed Microsimulation, mimeo, Department of Sociology, Cornell University, Ithaca, New York, July. Caldwell, S. (1996). Health, Welfare, Pensions and Life Paths: The CORSIM Dynamic Microsimulation Model, in Harding, A. (ed), Microsimulation and Public Policy, North-Holland, Amsterdam. Cape´au, B., Decoster, A., De Swerdt, K. and Orsini, K. (2006). Distributional Impact of Shifting the Base of Financing Social Security from Personal Contributions to Indirect Taxes. Paper presented at the 8th Nordic Seminar on Microsimulation Models, Oslo, 7–9 June (available from www.ssb.no/misi). Chin, S.F. and Harding, A. (2006). Regional Dimensions: Creating Synthetic SmallArea Microdata and Spatial Microsimulation Models. Technical Paper no. 33, National Centre for Social and Economic Modelling, University of Canberra, May. Chin, S.F. and Harding, A. (2007). SpatialMSM’, in Gupta, A. and Harding, A. (eds), Modelling Our Future: Population Ageing, Health and Aged Care, International Symposia in Economic Theory and Econometrics, North-Holland, Amsterdam. Chin, S.F., Harding, A., Lloyd, R., McNamara, J., Phillips, B. and Vu, Q. (2005). Spatial Microsimulation Using Synthetic Small Area Estimates of Income, Tax and Social Security Benefits. Australasian Journal of Regional Studies, 11(3), 303–336. Citro, C.F. and Hanushek, E.A. (1991). The Uses of Microsimulation Modelling, Vol. 1: Review and Recommendations. National Academy Press, Washington. Cotis, J. (2003). Population Ageing: Facing the Challenge. OECD Observer, September.

26

Ann Harding and Anil Gupta

Creedy, J., Duncan, A.S., Harris, M. and Scutella, R. (2002). Microsimulation Modelling of Taxation and the Labour Market: The Melbourne Institute Tax and Transfer Simulator. Edward Elgar, Cheltenham. Cullinan, J., Hynes, S. and O’Donoghue, C. (2006). The Use of Spatial Microsimulation and Geographic Information Systems (GIS) in Benefit Function Transfer — An Application to Modelling the Demand for Recreational Activities in Ireland. Paper Presented at the 8th Nordic Seminar on Microsimulation Models, Oslo, 7–9 June (available from www.ssb.no/misi). Davies, J. (2004). Microsimulation, CGE and Macro Modelling for Transition and Developing Economies. Discussion Paper No 2004/08, World Institute for Development Economics Research (UNU/WIDER), Helsinki (available from www. wider.unu.edu/publications/discussion-papers-2004). Decoster, A. (2000). ASTER, in Gupta, A. and Kapur, V. (eds), Microsimulation in Government Policy and Forecasting, North-Holland, Amsterdam. Falkingham, J. and Harding, A. (1996). ‘Poverty Alleviation vs Social Insurance: A Comparison of Lifetime Redistribution, in Harding, A. (ed), Microsimulation and Public Policy, North-Holland, Amsterdam. Falkingham, J. and Hills, J. (1995). The Dynamic of Welfare: The Welfare State and the Life Cycle. Prentice-Hall, New York. Favreault, M. and Caldwell, S. (2000). Assessing Distributional Impacts of Social Security Using Microsimulation, in Gupta, A. and Kapur, V. (eds), Microsimulation in Government Policy and Forecasting, North-Holland, Amsterdam. Gallagher, P. (1995). The Policy Use of the Products of the Retirement Income Modelling Task Force. Conference Paper 95/3, Retirement Income Modelling Task Force, Canberra (available from http://www.rim.treasury.gov.au). Galler, H.P. and Wagner, G. (1986). The Microsimulation Model of the Sfb3 for the Analysis of Economic and Social Policies, in Orcutt, G.H., Merz, J. and Quinke, H. (eds), Microanalytic Simulation Models to Support Social and Financial Policy, North-Holland, Amsterdam. Gribble, S. (2000). LifePaths: A Longitudinal Microsimulation Model Using a Synthetic Approach, in Gupta, A. and Kapur, V. (eds), Microsimulation in Government Policy and Forecasting, North-Holland, Amsterdam. Gupta, A. and Basu, K. (2007). Building Policy-Relevant Health Human Resource Models, in Gupta, A. and Harding, A. (eds), Modelling Our Future: Population Ageing, Health and Aged Care, International Symposia in Economic Theory and Econometrics, North-Holland, Amsterdam. Gupta, A. and Harding, A. (eds). (2007). Modelling Our Future: Population Ageing, Health and Aged Care. International Symposia in Economic Theory and Econometrics, North-Holland, Amsterdam. Gupta, A. and Kapur, V. (1996). Microsimulation Modelling Experience at the Canadian Department of Finance, in Harding, A. (ed), Microsimulation and Public Policy, Elsevier (North-Holland), Amsterdam. Gupta, A. and Kapur, V. (2000). North-Holland, Amsterdam. Hain, W. and Helberger, C. (1986). Longitudinal Simulation of Lifetime Income, in Orcutt, G.H., Merz, J. and Quinke, H. (eds), Microanalytic Simulation Models to Support Social and Financial Policy, North-Holland, Amsterdam. Hancock, R. (2000). Charging for Care in Later Life: An Exercise in Dynamic Microsimulation, in Mitton, L., Sutherland, H. and Weeks, M. (eds),

Introduction and Overview

27

Microsimulation Modelling for Policy Analysis, Cambridge University Press, Cambridge. Harding, A. (1993a). Lifetime Income Distribution and Redistribution. Applications of a Microsimulation Model, North-Holland, Amsterdam. Harding, A. (1993b). Lifetime vs Annual Tax-Transfer Incidence: How Much Less Progressive? Economic Record, 69(205), 179–191. Harding, A. (1995). Financing Higher Education: An Assessment of IncomeContingent Loan Options and Repayment Patterns Over the Lifecycle. Education Economics, 3(2), 173–203. Harding, A. (ed), (1996). Microsimulation and Public Policy. North-Holland, Amsterdam. Harding, A. (2000). Dynamic Microsimulation: Recent Trends and Future Prospects, in Gupta, A. and Kapur, V. (eds), Microsimulation in Government Policy and Forecasting, North-Holland, Amsterdam. Harding, A., Payne, A. and Vu, Q.N. (2006). ‘‘Tomorrow’s Consumers’’, AMP. NATSEM Income and Wealth Report, (15), December (available from www.amp.com.au/ampnatsemreports). Harding, A., Warren, N., Robinson, M. and Lambert, S. (2000). The Distributional Impact of the Year 2000 Tax Reforms in Australia. Agenda, 7(1), 17–31. Kapur, V. and Gupta, A. (2006). A Microsimulation Model for Pharmacare: Development, Analysis and Policy Applications, in Gupta, A. and Harding, A. (eds), Modelling Our Future: Population Ageing, Health and Aged Care, International Symposia in Economic Theory and Econometrics, North-Holland, Amsterdam. Kelly, S., Farbotko, C. and Harding, A. (2004a). The Lump Sum: Here Today, Gone Tomorrow AMP.NATSEM Income and Wealth Report, Issue 7, Sydney, March (available from www.amp.com.au/ampnatsemreports). Kelly, S. and Harding, A. (2004b). Funding the Retirement of the Baby Boomers. Agenda, 11(2), 99–112 (available from www.natsem.canberra.edu.au/publication. jsp?titleID=JA0402). Kelly, S. and King, A. (2001). Australians Over the Coming 50 years: Providing Useful Projections. Brazilian Electronic Journal of Economics, 4(2). Kelly, S., Percival, R. and Harding, A. (2002). Women and Superannuation in the 21st Century: Poverty or Plenty? Competing Visions: Proceedings of the National Social Policy Conference, SPRC Report 1/02 University of New South Wales, pp. 223–249 (available from www.sprc.unsw.edu.au/nspc2001). King, A. (2007). Providing Income Support Services to a Changing Aged Population in Australia: Centrelink’s Regional Microsimulation Model, in Gupta, A. and Harding, A. (eds), Modelling Our Future: Population Ageing, Health and Aged Care, International Symposia in Economic Theory and Econometrics, NorthHolland, Amsterdam. Klevmarken, A. and Olovsson, P. (1996). Direct and Behavioural Effects of Income Tax Changes: Simulations with the Swedish Model MICROHUS, in Harding, A. (ed), Microsimulation and Public Policy, North-Holland, Amsterdam. Klevmarken, N. (1997). Behavioral Modelling in Micro Simulation Models: A Survey. Working Paper No. 31, Uppsala University (available from www.nek.uu.se/Pdf/ 1997wp31). Mankiw, N.G. and Weil, D.N. (1989). The Baby Boom, the Baby Bust, and the Housing Market. Regional Science and Urban Economics, 21, 573–579.

28

Ann Harding and Anil Gupta

Mitton, L., Sutherland, H. and Weeks, M. (eds). (2000). Microsimulation Modelling for Policy Analysis. Cambridge University Press, Cambridge. Morrison, R.J. (2000). DYNACAN, in Gupta, A. and Kapur, V. (eds), Microsimulation in Government Policy and Forecasting, North-Holland, Amsterdam. Murphy, B. (2000). SPSD/M, in Gupta, A. and Kapur, V. (eds), Microsimulation in Government Policy and Forecasting, North-Holland, Amsterdam. Nelissen, J.H.M. (1996). Social Security and Lifetime Income Redistribution, in Harding, A. (ed), Microsimulation and Public Policy, North-Holland, Amsterdam. O’Donoghue, C. (2001). Dynamic Microsimulation: A Methodological Survey. Brazilian Electronic Journal of Economics, 4(2). O’Hare, J. (2000). TRIM3, in Gupta, A. and Kapur, V. (eds), Microsimulation in Government Policy and Forecasting, North-Holland, Amsterdam. Orcutt, G. (1957). A New Type of Socio-economic System. Review of Economics and Statistics, 58(2), 773–797. Orcutt, G., Caldwell, S., Wertheimer, R., Franklin, S., Hendricks, G., Peabody, G., Smith, J. and Zedlewski, S. (1976). Policy Exploration through Microanalytic Simulation. The Urban Institute, Washington, DC. Pedersen, T. (2000). Distributional Outcomes of the Danish Welfare System, in Gupta, A. and Kapur, V. (eds), Microsimulation in Government Policy and Forecasting, North-Holland, Amsterdam. Productivity Commission (2005). Economic Implications of an Ageing Australia, Overview, Canberra (available from www.pc.gov.au). Sutherland, H. (1995). Static Microsimulation Models in Europe: A Survey. Microsimulation Unit Discussion Paper, MU9503, University of Cambridge, Cambridge. Sutherland, H. (2000). EUROMOD, in Gupta, A. and Kapur, V. (eds), Microsimulation in Government Policy and Forecasting, North-Holland, Amsterdam. Symons, E. and Warren, N. (1996). Modelling Consumer Behavioural Response to Commodity Tax Reforms, in Harding, A. (ed), Microsimulation and Public Policy, North-Holland, Amsterdam. Toohey, M. and Beer, G. (2004). Financial Incentives for Working Mothers Under a New Tax System. Australian Journal of Labour Economics, 7(1), 53–69. Treasury (2002), Budget Paper No. 5, Intergenerational Report 2002-03, May. Voas, D. and Williamson, P. (2000). An Evaluation of the Combinatorial Optimisation Approach to the Creation of Synthetic Microdata. International Journal of Population Geography, 6, 349–366. Walker, A., Percival, R. and Harding, A. (2000). The Impact of Demographic and Other Changes on Expenditure on Pharmaceutical Benefits in 2020 in Australia, in Mitton, L., Sutherland, H. and Weeks, M. (eds), Microsimulation Modelling for Policy Analysis, Cambridge University Press, Cambridge. Weston, R. and Wooden, M. (2002). HILDA Has Arrived!, Family Matters, Australian Institute of Family Studies, No 63, Spring/Summer, pp. 66–70. Williamson, P., Birkin, M. and Rees, P.H. (1998). The Estimation of Population Microdata by Using Data from Small Area Statistics and Samples of Anonymised Records. Environment and Planning A, 30(5), 785–816.

Introduction and Overview

29

Wolfson, M. (1988). Homemaker Pensions and Lifetime Redistribution. Review of Income and Wealth, 34(3), 221–250. Xiong, L., Ma, X., Li, Y., Meng, H. and Lin, G. (2007). Microsimulation Model of Medical Insurance Reform of Government Employees and Workers in China, in Gupta, A. and Harding, A. (eds), Modelling Our Future: Population Ageing, Health and Aged Care, International Symposia in Economic Theory and Econometrics, North-Holland, Amsterdam.

This page intentionally left blank

Part I: Pension Analysis Using Dynamic Microsimulation

This page intentionally left blank

Chapter 2

Can We Afford the Future? An Evaluation of the New Swedish Pension System Lennart Flood School of Business, Economics and Law, Go¨teborg University and Swedish Ministry of Finance, Sweden

Abstract In this chapter the Swedish dynamic microsimulation model (SESIM) was used to predict income before and after retirement. Given the interest in the new Swedish pension system, income for birth cohorts covered by the new and the old system were included. The results are presented in terms of replacement rates for both before- and after-tax income. All three components (pillars) of pension income were considered: public, occupational, and private pensions. Special attention was given to the role of other private wealth, both real and financial. The modelling of private wealth in SESIM is described and households’ portfolio allocation in the start-year 1999 is presented. As expected, the results show that the new system is less generous than the old. To achieve a compensation level close to the old system, the retirement age would have to be delayed and the return on savings would have to be high. A sharp reduction in public pensions — larger for high-income earners — is partly offset by an increase in occupational pensions. The results demonstrate the importance of the second and third pillars in the pension system, especially occupational pensions, which will play a crucial role for younger generations. Since these components, as well as part of the public pensions, will increasingly be funded, we can expect large variations in future pension income depending on the returns on these funds.

1. Introduction Along with most other OECD countries, Sweden is faced with a rapid increase in the proportion of elderly in the population, reflecting a combination of the ageing of the post-war ‘‘baby boom’’ generation, increased longevity, and low birth rates. Although the ratio of retired elderly to the International Symposia in Economic Theory and Econometrics, Vol. 15 Copyright r 2007 Elsevier B.V. All rights reserved ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15002-4

34

Lennart Flood

working-age population is already rising, the change is expected to accelerate in the next decade. According to Statistics Sweden (SCB), the number of individuals over 85 years will double in the next 30 years while the working-age population remains roughly the same. Clearly, this change in age structure challenges the financial sustainability of established pension systems. In response, a new public pension system was introduced in Sweden in 1999. The main purpose of this study is to analyze the effect of this new system on household income. A dynamic microsimulation model (SESIM), developed by the Swedish Ministry of Finance, and a rich microdata set Longitudinal Individual Data for Sweden (LINDA) from SCB, were used to compare household income before and after retirement, to decompose the contributions from public, occupational, and private pensions, and to evaluate the importance of other real and financial wealth. Because the simulation model, SESIM, was fundamental to the analysis, it will be presented first, with focus on the modelling of real and financial wealth. Then the new Swedish pension system will be described briefly, followed by the design and the results of the simulations.

2. The Swedish Microsimulation Model SESIM SESIM is a mainstream microsimulation model, in the sense that the variables (events) are updated in a sequence and the space in time between the updating processes is a year. The start year is 1999 and every individual in the initial sample (E100,000) then goes through a large number of real-life events, such as education, marriage, having children, working, retirement, etc. They are assigned a status each year, reflecting their main occupation and source of income: work gives earnings, retirement gives pension, etc. The tax and benefit systems are then applied and after-tax income is calculated. With sufficient repetitions, life-cycle incomes can be generated, and it is easy to compare income before and after retirement. Replacement rates for individuals or households can be defined. Figure 1 presents the sequential structure of SESIM. The first part (top, left-hand side) consists of demographic modules (mortality, adoption, migration, household formation and dissolution, disability pension, and rehabilitation), followed by a module for education (compulsory school, high school [gymnasium], municipal adult education [Komvux], and university.) The next module deals with the labour market including the retirement decision. Retirement has been modelled as an accrual benefit model (see Jansson, 2003) but, in this study, retirement is set to a specific age — the same for all — because we are interested in analyzing the effects of different

An Evaluation of the New Swedish Pension System Figure 1:

35

Structure of SESIM

Demography ⎯ Mortality ⎯ Adoption ⎯ Migration ⎯ Fertility ⎯ Children leaving home ⎯ Cohabitation ⎯ Separation ⎯ Disability ⎯ Rehabilitation

Model population at time t

Next year (t = t + 1)

Model population at time t + 1

Non-cash benefits ⎯ Childcare ⎯ Compulsory education ⎯ Upper secondary education ⎯ University ⎯ Adult education ⎯ Labor market activities ⎯ Old age care ⎯ Health care ⎯ Medication

Education ⎯ Dropout from upper secondary education ⎯ From upper secondary to university ⎯ Dropout from university ⎯ From labor market to university ⎯ From labor market to adult education ⎯ From adult education to university

Labor Market ⎯ Unemployment ⎯ Employment ⎯ Miscellaneous status ⎯ Labor market sector ⎯ Income generation (earnings)

Wealth & Housing ⎯ Financial wealth ⎯ Real wealth ⎯ Income of capital

Taxes, Transfers & pension ⎯ Student loans and allowances ⎯ Income tax ⎯ Real estate tax ⎯ Capital income tax ⎯ Wealth tax ⎯ Maintenance ⎯ Child allowance ⎯ Housing allowance ⎯ Social assistance ⎯ Old age pension ⎯ Disability pension

retirement ages. The labour-market module also includes a model for unemployment and a model for the imputation of the labour-market sector. The sector is required for calculations of occupational pensions. As we will see, occupational pensions are an important component of the total pension. In SESIM, we have implemented the rules for occupational pensions as well as the choice of labour-market sectors. We also allow for change of sectors, and the occupational pension is then adjusted in accordance to the new rules for occupational pensions in that sector.

36

Lennart Flood

At this point, one of nine unique statuses was determined for each individual: 1. Child (0–15), 2. old-age pensioner (from 61), 3. Student (19–45), 4. Disability pensioner (16–63), 5. Parental leave (women 16–49), 6. Unemployed (19–64), 7. Miscellaneous (19–64), 8. Employed (16–67), and 9. Emigrant. For status 8 (employed), an earnings equation was used to determine income and for other statuses, income was imputed according to the relevant rules. Next comes a module on housing and other real and financial wealth, described in detail in the next section. After this, a module applies all relevant tax, transfer, and pension rules, followed by a module for non-cash benefits (discussed in detail in Pettersson & Pettersson, 2003). Finally, disposable income is calculated.

3. Data on Financial and Real Wealth Data on wealth from LINDA1 and complementary data on housing from HEK2 were used to estimate portfolio allocation and the cost of housing. The data are described here in aggregate and by type of asset. Special attention is given to the age profile, as the main purpose of the dynamic microsimulation approach is to construct age profiles for different variables of interest. Data on wealth comes from administrative records, from which the data were drawn. How reliable are the data? One problem is that some assets like cars, boats, and other durables, as well as some assets abroad, are probably underreported — meaning that the data used here to some degree underestimate household wealth. There is also a problem in the ambiguous definition of ‘‘household’’ in administrative data. Finally, and most importantly, there is lack of wealth information over time. We have access to data only for 1999 and 2000, with the implication thus being that we are not able to identify time and cohort effects.3 Furthermore, 1999 and 2000 represent a period of unprecedentedly high levels on Stockholm Stock Exchange. A special effort was made in the construction of accumulated tax-deferred pension savings, accumulating yearly individual values from 1980 to 2000. To the best of our knowledge, this is the first time that this has been done with Swedish data.

1

See Edin and Fredriksson (2000). Household Finances, SCB. 3 Andersson, Berg & Klevmarken (2001) report important cohort effects. 2

An Evaluation of the New Swedish Pension System

37

3.1 Data Description Table 1 summarizes wealth data at the end of 1999, constructed from 771,771 individuals without sample selection, and weighted to obtain total wealth in Sweden. Aggregated total net wealth (total financial assets plus total net real wealth) of Swedish households was SEK 3,073 billion. Sixty five per cent of households had positive net wealth, averaging SEK 568,000, though it is only SEK 347,000 when all households are included. Figure 2 shows how mean individual wealth varied with age. At 40, mean total net wealth was SEK 325,000, while the mean for individuals with positive wealth was almost SEK 600,000. The share of individuals with positive net wealth (right-hand side scale) is also displayed. Approximately 64 per cent of those aged 40 had a positive total new wealth. The overall age profile of total net wealth resembles what could be expected from a life-cycle perspective, but the peak comes rather late (at 56) and then stays at roughly that level until 65. The highest value is actually at age 64 (around 1 million), for those with positive net wealth. After this age it drops rapidly while volatility also increases, because of higher death rates and smaller sample size at higher ages. The lowest share with positive net wealth was at age 24, at 38 per cent, increasing to almost 90 per cent for the oldest. Table 1 also shows the breakdown between real and financial wealth, as does Figure 3. Real wealth is the largest asset in households’ portfolios: SEK 2,233 billion compared to 1,915 billion for financial assets. Note that real wealth measures the market value of housing. These values have been imputed from tax-assessed values. However, households also had aggregate liabilities of SEK 1,075 billion, considered in SESIM as debts on real wealth, reducing it to SEK 1,158 billion. While of course there is also non-mortgage debt, this approximation is fairly realistic.4 Figure 3 shows how individual real wealth, liabilities, and financial assets vary with age. Real wealth and liabilities show clear life-cycle patterns. Real wealth increases steadily from 20 to 55, peaking at SEK 500,000 (including those with zero wealth), but then dropping sharply. Net real wealth shows a few negative values at younger ages but then increases steadily, peaking at SEK 320,000 at age 62 before declining. The age profile of financial assets has a sharper increase for individuals in their early to mid-50 s and then slowly declines.

4 In Linda 2000, about 31 per cent of all individuals older than 18 with no real wealth had debts above SEK 50,000 (excluding student loans), the corresponding figure for those with real wealth was 67 per cent. The average debt for those with real wealth was about SEK 250,000 and for those without only SEK 66,000.

38

Table 1:

Real and Financial Wealth in Sweden, December 1999 Sum (billions,SEK)

Share (%)

Mean40 (1000 SEK)

Share of total sum for top 10%

5%

1%

2,233 1,075 1,158

252 121 131

42.2 50.9 34.9

598 239 472

60.1 57.4 89.6

42.5 39.9 64.8

18.6 18 27.6

1,915

216

71.5

303

72.6

57.4

30.9

365 312

41 35

38.2 27

108 131

75.1 91.3

57 76.9

25.8 42.6

326 419

37 47

36.3 18.4

102 257

83.4 98.8

65.6 94

29.1 70.9

278

31

30.3

104

85.3

66.5

28.4

214

24

5.4

446

3,073

347

65.4

568

Source: Own calculations based on Linda-data for year 1999.

100 69.9

100 51.8

70.6 24.9

Lennart Flood

Total real wealth Total liabilities Total net real wealth Total financial assets Bank deposits Fixed income and other securities Mutual fund shares Swedish quoted shares Pension savings (deductible) Other real and financial assets Total net wealth

Mean (1000 SEK)

An Evaluation of the New Swedish Pension System Mean and Share of Net Wealth, by Age, 1999

1,200

1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00

SEK1 000 : -

1,000 800 600 400 200 0

Share

Figure 2:

39

0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 Age Net wealth

Figure 3:

Net wealth > 0

Share

Real Wealth, Liabilities, and Financial Assets by Age, 1999 600

SEK 1 000 :-

500 400 300 200 100 0 -100

0

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 Age Real wealth

Liabilities

Net real wealth

Financial wealth

Finally, Table 1 shows the breakdown of financial assets, and Figure 4 the age profiles. The largest single item is Swedish-quoted shares, SEK 419 billion in aggregate, the distribution of which is extremely skewed. The top decile owns almost 99 per cent and the top per cent owns more than 70 per cent. Bank deposits, the second largest item (SEK 365 billions), are the least skewed of the financial assets, though the top decile still has 75 per cent. Bank deposits increase steadily with age whereas mutual-fund shares and fixed-income securities reach a peak at 55–60. Swedish-quoted shares increase up to about 60, then level off, and then start increasing again. The later peak at 80 years of age is because of the influence of a high maximum value and a small sample size. Pension savings (SEK 278 billions)

Lennart Flood

40

SEK 1 000

Figure 4:

Financial Wealth, 1999 180 160 140 120 100 80 60 40 20 0 0

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 Age Bank deposits Mutual fund shares Pension savings

Swedish quoted shares Bonds, fixed income securities

peak around age 55 and then drop rapidly, since no new savings are added after retirement. As in the other modules, wealth and housing include a large number of variables modelled or calculated. The calculations are carried out sequentially; the order is given in Figure 5. We will review the process starting with financial wealth and pension savings, then real wealth, cost of housing, and finally, capital income. The process starts (first simulated year 2000) with the diamond-shaped box in the upper left-hand side. This is to check whether an individual had financial wealth in the start data (year 1999): If yes, it is updated using a simple random walk; if no, its probability in 2000 is imputed using model (1) in the second diamond-shaped box in the upper righthand side. Next, the prediction from model (1) is evaluated and a Monte Carlo experiment is applied.5 If a positive wealth is predicted, model (2) is applied to calculate how much. Finally, individual financial wealth is aggregated to household level. Thus, financial wealth is modelled as a two-part model. That is, the probability of financial wealth is estimated independent of the value. The reason for using the two-part model compared to, for instance, generalized tobit or two-stage methods (heckit), is that we are not interested in explaining selectivity. Here, the purpose is to obtain good predictions. It is demonstrated in Manning et al. (1987) that the two-part model performs at

5

Let P denote the predicted value from model (1), which is then compared to a random draw from a uniform (0,1) distribution, say R. If R o P, then the individual was given financial wealth. Most predictions in SESIM follow this process.

An Evaluation of the New Swedish Pension System Figure 5:

t+1

41

Real and Financial Wealth and Cost of Housing in SESIM

i_wealth_financial >0

No

No

Probability i_wealth_financial >0 Logit (1)

Yes

Yes

Random walk: i_wealth_financialt = i_wealth_financialt-1 +εt εt~N(0,σ2)

OLS (2) i_wealth_financial

i_wealth_financial = 0

Aggregate to household wealth: h_wealth_financial = sum of spouses wealth Pension savings: i_wealth_pension_year, yes/ no Logit (3) If yes how much, OLS (4) Stock of pension savings: i_wealth_pension_total

For new households calculate probability of being h_house_owner. Logit (5)

No

h_house_owner

Probability Buying Logit (6)

Yes h_building area: OLS (8) Old owner

No

Yes

New owner h_house_marketvalue: OLS (9)

Probability Selling Logit (7)

Yes

Debt, real and financial wealth: h_house_debt=max(0,h_house_marketvalue-h_wealth_financial) h_wealth_real=max(0,h_house_marketvalue-h_house_debt) h_wealth_financial=max(0,h_wealth_financial-h_housemarketvalue)

h_wealth_real=max(0,h_house_marketvalue-h_house_debt) h_house_netgain=(1-xtaxhouseprofit)*h_wealth_real h_wealth_financial=h_wealth_financial+ h_house_netgain

Cost of housing: h_house_cost

Dissagregate h_wealth_financial into i_wealth_financial Capital Income: i_income_capital=Sf*i_wealth_financial Sf is set to 0.06

42

Lennart Flood

least as well as the tobit type 2. In Flood & Gra˚sjo¨ (2001), the sensitivity of the generalized tobit model is demonstrated, as errors in the specification of the selection equation produce bias in all the estimated parameters. Here we are much more concerned with robustness compared to a potential increase in efficiency. The parameters of the logit (1) and Ordinary Least Squares (OLS) (2) models (presented in Flood, 2003) are generally estimated with both reasonable levels and high precision. The estimated parameters often have the same sign, meaning that the variables have a similar effect on both the probability and the level of wealth. The estimated age profile conditional on positive wealth is strictly increased up to 60 years and is then rather stable up to 80, after which it drops. Education has a strong effect: the odds ratio for lowest education is only 0.44, meaning that the odds of having financial wealth are reduced by 56 per cent for those with the lowest education, all else constant. Naturally there is also a strong effect owing to income: higher income implies a higher probability of wealth. Place of birth has a strong effect: being Swedish born gives 180 per cent higher odds of having wealth, and, conditional on having wealth, Swedish born have on an average almost 20 per cent higher value ([e0.18 1]100). Next, yearly tax-deferred pension savings are imputed for first-time savers using models (3) and (4). After the first year, these amounts, adjusted by the consumer price index, are added to the previous stocks each year until retirement. This process is discussed in more detail in the following section. Next the existence and value of household-owned real wealth is calculated. Home ownership as well as the market value is known in the start data. However, in the simulation, new households are created and, for those, the probability of being a house owner has to be imputed (model 5). Next, we check whether the household owned a house. If yes, the building area (size of the house) is imputed, using model (8). This has to be imputed for all households, since this was not known in the data and this information is needed as an independent variable in other models as well as for calculation of housing cost (housing expenditure: rent, utilities, interest rates etc). If the household does not own a house, the probability that they will buy one, model (6), is calculated. If they buy a house, the building area is imputed (model 8) The next step for new house owners is to impute the market value (model 9), and then to calculate debt, real wealth, and household financial wealth. For old owners the probability of selling is imputed, model (7) — and, given a sale, the net real wealth, net gain from the sale, and financial wealth are all calculated or adjusted. Unless already paid off, home owners are assumed to have mortgages, decreasing by 1/50 each year, with financial assets with real wealth being adjusted accordingly. Finally, household financial wealth is disaggregated

An Evaluation of the New Swedish Pension System

43

into individual wealth in order to calculate yearly income of capital, calculated as 6 per cent of financial wealth.

4. The Pension System in Sweden As mentioned earlier, pressure from an ageing Swedish population forced the introduction of a new public pension system in 1999, consisting of two parts: a notional-defined contribution pay-as-you-go (NDC PAYG) and an advance-funded defined contribution (DC). The former defined-benefit (DB) system is gradually being phased out, so the new system covers individuals born from 1938 to 1953 only partly, while covering individuals born thereafter totally. In the new pension system, employers and employees pay a total contribution of 18.5 per cent on earnings: 16 per cent to the NDC PAYG system and 2.5 per cent to the DC system. Both systems are autonomous from the state budget and self-financing. However, general revenues from the state-budget finances a minimum guarantee benefit for lowincome earners and for lifetime poor. 4.1 The Notional-Defined Contribution Pay-As-You-Go System The NDC PAYG system has the characteristics of a DC system, but in a PAYG setting. One such feature is the full link between contributions and benefits — that is, benefits are projected from contributions paid on all earnings during a lifetime. However, contributions are only recorded in individual accounts and the real contributions are financing payments to today’s pensioners, as in any PAYG setting. However, contributions paid on annual earnings above a ceiling of about SEK 290,000 in 2003 (7.5 Basic Amounts (BA) per year)6 do not qualify for pension rights. Contributions to the individual account represent a promise of future pension and are indexed by average wage growth. Pension holdings and pension payments are indexed at a slower rate than average wage growth, when average wage growth increases faster than wage sum and/or when observed average length of life increases after retirement. A second feature of the NDC PAYG system is that the annual benefit level is calculated by dividing the total contributions in the individual account by age specific and unisex life expectancy, which also includes an expected real rate of return of 1.6 per cent per year. 6

The Basic Amount (BA) is calculated based on changes in the general price level as specified by changes in the Consumer Price Index (CPI), and is set for a full calendar year.

44

Lennart Flood

4.2 The Advance-Funded Defined-Contribution (AF– DC) System The launch of the new defined contribution system in the fall of 2000 entitled the Swedish workforce of more than 4.4 million individuals to invest pension assets in mutual funds. At this time, accumulated contributions from 1995 to 1998 were invested, which approximately corresponded to SEK 56 billion. The individuals could choose to invest in one to five different mutual funds (from 460 available funds in the system). This means that the Swedish system has greater latitude for choice than the U.S. 401(k) plans, which typically include only a few funds. For individuals who do not make an active investment decision, the government provides a publicly managed mutual fund. Continuing annual contributions to the financial account system in the public and mandatory DC system are paid to an individual account once a year. These contributions are invested in mutual funds, based on individuals’ investment decisions. This implies that pension assets will grow at the rate of return of the chosen funds and based on annual contributions. The accumulated capital in the individual account cannot be withdrawn until retirement age, which is flexible from the age of 61. The annuity is calculated by dividing the individual account value by unisex and age-specific life expectancy at retirement day. During the years of retirement, individuals can choose a fixed or a flexible annuity rate: fixed, by moving the assets to the state annuity provider which includes a minimum annual return of 3 per cent; or flexible, by keeping the assets in the fund reflecting the market rate of return.

4.3 Occupational Pensions Most employed individuals are also covered by central agreements between labour unions and employer confederations, which include occupational pensions financed through employer contributions, both in addition to public pensions and covering incomes above the ceiling. Thus, these schemes are mostly important for high-income earners. In principle, four occupational plans are distinguished: blue-collar workers in the private sector, white-collar workers in the private sector, central-government employees, and local-government employees. SESIM includes a model predicting which sector an individual works in, imputed upon entry to the labour market. SESIM also allows for a change of sector, upon which accumulated occupational pension rights are transferred to the new sector. The four occupational systems have recently been reformed, like the public system, from defined benefit to defined contribution. In SESIM, we have implemented both the old and new rules in detail.

An Evaluation of the New Swedish Pension System

45

For blue-collar workers in the private sector, the new system is a fully funded pension scheme, where 3.5 per cent of gross earnings are paid into a personal account in a pension fund. Each worker can chose around a dozen insurance companies to manage his pension fund. White-collar workers in the private sector are covered by a defined benefit scheme as well as a fully funded scheme. The defined benefit scheme is determined by earnings the year before the worker retires. The benefits are 10 per cent of that year’s salary up to 7.5 BA, 65 per cent between 7.5 and 20 BAs, and 32.5 per cent between 20 and 30 BAs. Contributions to the defined benefit scheme have been around 4.5 per cent of gross earnings. The contribution to the fully funded system is approximately 2 per cent of earnings up to 30 BA. The worker is free to choose a company to manage his or her fund. The fully funded system is normally claimed as monthly payments over a five-year period after retirement. After 1992, there exist two schemes for central-government employees: one, fully funded and the other, PAYG. In the fully funded scheme 1.7 per cent of annual salary is paid to a pension fund. The PAYG is determined by average earnings during the five years preceding retirement. The benefits are the same as for white-collar workers, but based on the last five years’ average earnings instead of the last year’s. The pension is reduced proportionally if the requirement of 30 years of contributions since age 28 is not met. The new system for local government employees is fully funded and similar to that of blue-collar workers. 4.4 Private Pension Savings Modelling accumulated tax-deferred pension wealth is a challenge, since neither LINDA nor any other data set has individual data. Private pension savings are only taxed upon withdrawal after retirement. Therefore, the only information available is the yearly deductible savings. Konsumentverket (1999) provides analysis and descriptive statistics. Accumulated savings were thus constructed from repeated LINDA panels. Individual savings are summed up over years, and the resulting stock is increased each year by applying the average return from the life insurance companies. To reduce the starting value problem, we start as early as 1980, when private tax-deferred pension savings was still rather unusual. Table 2 summarizes pension savings, wealth and withdrawals from 1980 to2000. Column 2 gives the share of the whole population, regardless of age, with pension savings. During this period there has been an increase from about 4 to 21 per cent. The share with private pension wealth (column 6) reached 32 per cent in year 2000, the mean value (column 7) was SEK 110,863 and the corresponding mean of yearly savings (column 3) was SEK 6,591. Even if the share of pension savers has increased, the yearly amounts

46

Table 2:

Private Pension Savings, 1989–2000 Share with current pension savings (%)

Mean value given savings (tkr)

Share with income from pension savings (%)

Mean value given income (tkr)

Share with pension wealth (%)

Mean value given pension wealth (tkr)

Aggregate pension wealth (mkr)

Assum-ed return on savings (%)

1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000

4.60 4.70 4.90 3.80 4.40 8.20 8.50 9.70 11.90 14.40 14.50 12.50 12.80 13.30 15.00 16.20 17.30 18.20 19.20 20.50 21.90

3,529 3,962 4,748 6,968 7,846 8,321 9,229 9,969 11,170 12,955 8,138 9,656 8,339 8,465 8,762 6,861 6,764 6,705 6,659 6,785 6,591

0.00 0.00 0.00 0.00 0.00 0.00 0.70 0.80 0.70 0.70 0.70 2.60 2.90 3.30 3.50 4.10 4.10 4.20 4.30 4.50 4.80

936 1,416 981 2,127 1,469 1,427 11,621 14,074 7,676 8,027 8,319 21,013 22,175 23,476 23,572 22,528 23,608 25,272 27,870 30,540 32,598

4.10 4.40 4.70 4.80 5.00 5.40 6.90 8.70 11.00 13.60 15.50 17.20 18.50 19.50 21.40 23.00 24.60 26.10 27.80 29.70 32.00

3,882 8,086 13,136 19,728 28,427 39,080 43,074 46,748 51,523 62,291 69,798 73,414 76,117 79,001 80,551 82,478 85,822 92,546 100,973 104,530 110,863

1,396 3,137 5,449 8,411 12,550 18,828 26,553 36,056 50,285 74,903 95,710 111,944 125,012 136,367 152,702 168,393 187,482 214,326 248,870 275,265 315,101

10 10 10 12 13 15 14 12 14 21 16 10 7 5 7 7 8 11 13 8 12

Source: Author’s calculations are based on the LINDA panel 1980– 2000. Assumed returns (column 9) come from the Swedish Insurance Federation (www.forsakringsforbundet.com). These returns are returns before tax and administrative costs

Lennart Flood

Year

An Evaluation of the New Swedish Pension System

47

have not. Yearly savings reached their highest values in 1989, and since then they have gone down, because of rule changes and lower returns. The share with current income from private pension wealth (column 4) is still fairly small (less than 5 per cent of the population, with mean SEK 32,196). Aggregate Swedish pension wealth (column 8) was low in 1980, indicating little starting value problems, but had increased to SEK 315 billion by 2000. Accumulated pension wealth was thus constructed for every individual, with the starting year 1999. For those who had pension savings in 1999, it was assumed that they continued saving this amount (adjusted for Consumer Price Index ,CPI) every year until retirement. For individuals with no pension savings in 1999, a two-part model for new pension saving in 2000 was estimated, similar to the two-part model discussed earlier. The population at risk are all individuals 18–64 years in 2000 who did not have pension savings in 1999. In forecasting accumulated pension savings, we have to estimate the probability and the amount saved the first time. Then we assume that the individual saves the same amount (adjusted by CPI) each year until the age of 64. These yearly savings are then added to the stock and an assumption on a yearly return is used.7 Various withdrawal options are possible: limited time, whole lifetime, etc. In the current version of SESIM, a five-year period is the norm, but some variation was allowed to match observed profiles in 1999.

5. Comparing Income Before and After Retirement SESIM was used to simulate income before and after retirement during 1999–2041 for different birth cohorts, at five-year intervals, spanning the transition from the old pension system to the new and then completely to the new: those born in 1940 (6/20 on the new system); 1945 (11/20); 1950 (16/ 20); and 1955 and 1960 (completely on the new system). Baseline results assume that everyone has worked at least for the five years immediately before retirement at the age of 65, as well as a 2 per cent annual inflation rate, a 3 per cent real growth rate, and a 5 per cent return on financial assets (relevant for all funded pension systems and private pensions). All cohorts are predicted to have a decline in income at retirement, compared to the incomes that they enjoyed while working. However, if this comparison had included the unemployed, disability pensioners, etc., the

7 Of course this assumption has been introduced as a simplification, but there is some support for it in the data. Comparing the decile mobility for individuals with pension savings in 1995 with the same individuals in 2003 shows that the majority stays in the same decile or moves up or down one decile.

48

Lennart Flood

decline would be less dramatic. There is also a second decline at the age of 70 reflecting the fact that most fully funded systems pay out completely during the first five years after retirement. Table 3 shows the results for the 1940, 1950, and 1960 cohorts, with income replacement rates (relative to taxable income in ages 60–64) by income quartile and type of pension (public, occupational, and private) during the first two five-year periods after retirement. For individuals in quartiles 2 and 3 of the 1940 cohort, total taxable income immediately after retirement was predicted to be 75 per cent of the previous five-year period (public pension 55 per cent, occupational 10 per cent, private 8 per cent, and a residual 2 per cent from earnings on other savings), dropping to 68 per cent during the second five-year period after retirement, when private pensions mostly finish paying out (for this cohort, most occupational schemes have lifetime-defined benefits, so they only drop slightly). For those in quartile one, private pension wealth plays a larger role, as it also does for later cohorts. Lower replacement rates and different composition are predicted for later cohorts: generally much less public pension, and less private, but more occupational.8 For quartile one and two individuals in the 1960 cohort, the drop in the replacement rate from 62 to 53 per cent in the second five-year period after retirement reflects the termination of payouts in many occupational (as well as private) pensions, which have now been mostly converted into fully funded limited term plans. For quartile four individuals in the 1960 cohort (completely in the new system), the replacement rate is only 28 per cent, since much of their pre-retirement income was above the coverage ceiling. However, occupational pensions partly compensate for this. In addition to the public, occupational, and private components of the pension system and any other income from other financial assets, those assets themselves plus the realized net value of real wealth (housing) could be used to defray consumption costs during retirement. This total wealth value was thus simulated by assuming that all houses were sold upon retirement and then added (net of debt and taxes) to financial assets. It was further assumed that one-twentieth of the resulting total was then available for consumption each year (Table 4, last column expressed as a per cent of average income during the last five pre-retirement years). For each birth cohort this wealth available for consumption constituted the largest relative addition to income for those in the lowest income quartile (19–24 per cent).

8

The reduced private pension for later cohorts may reflect underestimation, however, as their savings were mostly imputed in the simulation, whereas more of those for older cohorts were known as of the start date.

Table 3:

1940

1950

1960

Income class

op25 p25–p75 4p75 op25 p25–p75 4p75 op25 p25–p75 4p75

Age 65–69

Age 70–74

Total taxable income (%)

Public pension (%)

Occupational pension (%)

Private pension (%)

Total taxable income (%)

Public pension (%)

Occupational pension (%)

Private pension (%)

110 75 67 82 68 60 77 62 55

85 55 39 59 46 28 55 41 28

8 10 16 12 13 23 13 15 22

16 8 10 10 8 8 7 5 4

99 68 60 72 58 49 67 53 45

86 56 40 60 47 28 55 41 28

7 8 14 9 10 18 11 10 15

5 2 4 1 1 2 0 0 0

Source: SESIM-generated, 1999– 2041 Assumptions: All individuals of age 60– 64 worked, retired at age 65, survived at least until 75; InflationE2 per cent/year, real wage growth E 3 per cent/ year and long-term interest rate 5 per cent/year. The replacement rates are the ratio between average income at ages 65– 69 and 70– 74 and average taxable income at age 60– 64.

An Evaluation of the New Swedish Pension System

Cohort

Replacement Rates for Various Birth Cohorts and Income Quartiles

49

Lennart Flood

50

Table 4: House Ownership Before and After Retirement, and Total Wealth Available for Consumption by Birth Cohort and Income Quartile Cohort

1940

1950

1960

Income class

op25 p25–p75 4p75 op25 p25–p75 4p75 op25 p25–p75 4p75

Share home owners (60–64)

Sharehome owners (65–69)

Potential income from wealth

0.44 0.46 0.54 0.41 0.46 0.54 0.37 0.44 0.49

0.37 0.42 0.48 0.35 0.39 0.47 0.30 0.37 0.42

0.24 0.12 0.12 0.19 0.13 0.10 0.19 0.11 0.10

Source: SESIM-generated, 1999– 2041 Note: The results in the final column are potential income from wealth expressed as a per cent of average income during the last five pre-retirement years. Assumptions: All individuals of age 60– 64 worked, retired at age 65, survived at least until 75, inflation E 2 per cent/year, real wage growth E 3 per cent/year, and long-term interest rate 5 per cent/year.

As also reflected in Table 4, they are also those most likely to sell their homes upon retirement, thus using their home equity as a retirement buffer. Table 5 extends Table 3 from total taxable income to disposable income, including transfers and net of taxes, again reporting replacement rates for the first two five-year post-retirement periods compared to the pre-retirement period. As with taxable income, later cohorts are predicted to have lower replacement rates (columns one and two) — for example, dropping from 76 per cent for quartile two and three, individuals born in 1940 when aged 70–74, to only 63 per cent for similar individuals born in the 1960. Table 5 also shows how replacement rates of disposable income would change if retirement age were lower or higher. Generally, retiring later would raise replacement rates, while retiring earlier would lower them, but there is an asymmetry. While retiring later would raise 1940 and 1960 cohorts replacement rates by about the same amount, retiring early would lower replacement rates for the 1960 cohort more than for the 1940 cohort. In the new pension system, contributions during the whole life matter, and the annual benefit level is calculated by dividing the total contributions in the individual account by age-specific life expectancy — and this is affected by the retirement age. Thus, the penalty for an early retirement is high. Table 5 also shows the effects of high (7 per cent) and low (3 per cent) rates of return on financial assets (last four columns). The changes are much

Table 5: Replacement Rates of Disposable Income for Various Birth Cohorts and Income Quartiles, by Age of Retirement and Period After Retirement, with Normal, Low, and High Returns

1940

1950

1960

Income class

op25 p25–p75 4p75 op25 p25–p75 4p75 op25 p25–p75 4p75

Age of retirement (65)

Age of retirement (67)

Age of retirement (63)

High return (7%)

Low return (3%)

Age 65–69 (1)

Age 70–74 (2)

Age 67–71 (3)

Age 72–76 (4)

Age 63–67 (5)

Age 68–72 (6)

Age 65–69 (7)

Age 70–74 (8)

Age 65–69 (9)

Age 70–74 (10)

102 83 78 86 77 71 77 74 69

104 76 68 82 65 58 70 63 55

111 88 83 88 81 79 85 76 76

112 81 72 86 69 66 77 66 62

97 83 76 77 69 69 69 67 65

100 73 63 74 61 56 66 56 52

109 85 82 93 83 77 86 84 78

107 76 71 88 73 63 79 73 64

104 82 77 81 74 70 70 65 66

100 73 64 76 65 55 63 55 54

Source: SESIM-generated, 1999– 2041 Note: The replacement rates are the ratio of average income during each of the two periods to average income during the last five pre-retirement years. For the variations in the rates of return columns, it is assumed that the individual retired at age 65. Assumptions: All individuals of age 60– 64 worked, retired at age 65, survived at least until 75, inflationE2 per cent/year, real wage growthE3 per cent/year, and long-term interest rate 5 per cent/year.

An Evaluation of the New Swedish Pension System

Cohort

51

52

Lennart Flood

smaller for the 1940 cohort (in the old DB system) than for the 1960 cohort (in the new AF–DC system). For example, the replacement rate for the 1960-cohort quartiles two and three in the first period after retirement is predicted at 84 per cent with high returns, but only 65 per cent with low returns.

6. Summary and Conclusions Income before and after retirement for individuals covered by both the old and new pension systems was predicted until 2041 using the dynamic Swedish microsimulation model, SESIM. The results demonstrate the importance of the second (occupational) and third (private) pillars in the pension system. Occupational pensions will be especially crucial for younger generations — and because they as well as public pensions are increasingly funded, we can expect a large variation in pension income depending on the returns on these funds. As expected, the new system was found to be less generous than the old. To achieve comparable replacement rates, retirement age would have to be delayed and the return on savings would have to be high. A serious simplification in the simulation was that everyone has the same return on savings, so that variability in pension income for later cohorts was underestimated. Modelling the households’ choice of pension funds and allowing for heterogeneity in their returns is an interesting challenge for future research.

References Andersson, B., Berg, L. and Klevmarken, A. (2001). Inkomst- och Fo¨rmo¨genhetsfo¨rdelningen fo¨r dagens och morgondagens a¨ldre. http://www.nek.uu.se/faculty/ klevmark/reswork.html Edin, P.A. and Fredriksson, A. (2000). LINDA — Longitudinal Individual Data for Sweden. http://www.nek.uu.se/ Flood, L. and Gra˚sjo¨, U. (2001). A Monte Carlo Simulation Study of Tobit Models. Applied Economics Letters, 8, 581–584. Flood, L. (2003). Formation of Wealth, Income of Capital and Cost of Housing in SESIM. SESIM Working Paper. http://www.sesim.org Jansson, F. (2003). Modelling the Retirement Decision in Sweden. Paper presented at the International Microsimulation Conference on Population Ageing and Health: Modelling Our Future, Canberra, 7–12 December (available from http:// www.natwsem.canberra.edu.au). Konsumentverket. (1999). Slutrapportering av regeringsuppdrag ro¨rande husha˚llens pensionssparande. Dnr 1999/2339.

An Evaluation of the New Swedish Pension System

53

Manning, W.G., Duan, N. and Rogers, W.H. (1987). Monte Carlo Evidence on the Choice Between Sample Selection and Two-Part Models. Journal of Econometrics, 35, 59–82. Pettersson, T. and Pettersson, T. (2003). Lifetime Redistribution Through Taxes, Transfers and Noncash Benefits. Paper presented at the International Microsimulation Conference on Population Ageing and Health: Modelling Our Future, Canberra, 7–12 December (available from http://www.natwsem.canberra. edu.au).

This page intentionally left blank

Chapter 3

A Microsimulation Model of Private Sector Pensions in France Thierry Debranda, Sophie Pennecb and Anne-Gise`le Privatc a

Institut de Recherche et de Documentation en Economie de la Sante´, France Institut National d’Etudes De´mographiques, France c Caisse Nationale d’Assurance Vieillesse, France b

Abstract As a long-run projection tool, microsimulation models provide a more detailed analysis of pension entitlements than traditional macroeconomic models as they can provide measurements of the effects of pension reform — both at individual level and at the level of the pension system. This paper presents the ARTEMIS (Analysis of Retirement pension from the privaTE Sector by MIcroSimulation) microsimulation model developed by the French national pension fund (Caisse Nationale d’Assurance Vieillesse – CNAV). It provides projections up to 2030 for the main pension fund for employees, and the effects on both contributors and pensioners. The results illustrate how the recent French reforms of pensions of private-sector employees will affect different cohorts retiring in the coming years, in terms of both retirement age and pension. They have only marginally slowed down the increase in the number of pensioners — as their effects on retirement age are relatively modest — but they have a stronger effect on the average level of pensions.

1. Introduction The large number of baby-boom cohorts approaching retirement age around the years 2005–2006, combined with population ageing, means that France, like most developed countries, is facing the problem of the financial sustainability of its pay-as-you-go pension system. Most retirees rely on this system, as nearly 70 per cent of their income is provided by their pension. According to the latest report by the Pensions Advisory International Symposia in Economic Theory and Econometrics, Vol. 15 Copyright r 2007 Elsevier B.V. All rights reserved ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15003-6

56

Thierry Debrand et al.

Council (Conseil d’Orientation des Retraites — COR), the ratio of retired to active workers will rise from 44 per cent in 2000 to 83 per cent in 2040 (COR, 2001). In this context, microsimulation provides an important tool for analysing the effects of various social security pension reforms, by evaluating their advantages or their economic costs and consequences. Moreover, changes in the pension scheme will not affect all people in the same way, as pensions are calculated according to a set of characteristics relative to each individual’s career (career length, wage profile, etc.). The major advantage of microsimulation models is that, in addition to pointing out the effects of major structural phenomena (increase in life expectancy, increase in female employment, fall in total duration of employment, etc.), they offer the opportunity to forecast the impact of reforms both at the individual level and at the level of the pension system as a whole. This chapter describes the first version of the microsimulation model ARTEMIS (Analysis of Retirement pension from the privaTE Sector by MIcroSimulation) that enables us to forecast old age pensions from now until 2030. This model simulates the population of the French mandatory state-monitored pension scheme for private-sector employees, often known as the ‘‘general scheme’’, and managed by the Caisse Nationale d’Assurance Vieillesse (CNAV).1 Although the French pension system is occupation-based, this scheme covers nearly 68 per cent of the labour force in France. In the first part of this chapter, we focus on the main dataset used in the model. It is a comprehensive source established on the basis of administrative records that span the full careers of people insured by the CNAV since the 1940s. The second part gives details about the model’s structure. The third part illustrates some results — in particular, the effects of the latest reforms of the French pension scheme.

1

The general scheme is the first pillar of the pension system for private-sector employees. It has about 15 million contributors and 9 million pensioners. There is a second pillar which consists of two supplementary schemes: one for all workers called the Association pour le Re´gime Comple´mentaire des salaries (ARRCO), and a second called the Association Ge´ne´rale des Institutions de Retraite des Cadres (AGIRC) which is for executives only. For the other members of the working population, there are two other major categories of basic retirement schemes: one for employees of the public sector or of large state-owned enterprises (it represents about 21 per cent of the labour force) and one for self-employed workers (about 11 per cent of the labour force).

A Microsimulation Model of Private Sector Pensions

57

2. Data The first stage in implementing a microsimulation model for retirement pensions is to determine the starting population. One of the model’s key features is its use of administrative data. This feature has three main advantages over the survey data that are often used in studies concerning the future of pensions. The first advantage of administrative data is that the information does not rely on the memory of each person surveyed. As a consequence, data concerning the years when pension fund contributions were made and contribution amounts are of better quality and more accurate. Second, the file created using these types of data is considerably more extensive than a sample based on a survey, and this enables us to reduce the uncertainties inherent in microsimulation. Finally, the data cover career history and, in particular, lifetime individual wages — thus making it possible to simulate retirement and pensions taking real wages into account and to simulate future wages based partly on existing career profiles. This data source is used both to create the starting population and to analyse wages and generate wage profiles that are used to forecast wage careers into the future. The scope of the general scheme is extremely broad. Since 1975, one threemonth contribution period has been sufficient to entitle contributors to a pension. In order to compile the most representative initial database for the population covered by this scheme, we aggregated two files at the starting date of the simulation (31 December 2001): the ‘‘contributors’’ file, which is the dataset of contributors (current and past) to this pension scheme, and the ‘‘pensioners’’’ dataset, listing the retirees from this scheme.

2.1 Samples of Contributors and Pensioners The 2002 contributors dataset is a 1/20th sample of a national data repository2 containing 2,075,046 accounts. It is a particularly rich source of information, composed of individual data on wages earned in the private sector from 1947 to 2001. It includes many elements required to determine the amount of pensions awarded by the general scheme. 2

This repository, created in January 1999, groups together three national databases: the national identification management system (SNGI — Syste`me National de Gestion de l’Identification, providing the national identification number, NIR, and vital statistics on birth, death, marriage, children, etc.), the national career management system (SNGC — Syste`me National de Gestion des Carrie`res) and the national file management system (SNGD — Syste`me National de Gestion des Dossiers).

58

Thierry Debrand et al.

The wages used to estimate the years of contribution are gross wages limited to a ceiling set by the French Social Security (Se´curite´ sociale).3 In addition to the years validated by employment,4 there are assimilated periods validated even when the person is not working (in the event of sick leave, maternity leave, leave due to accidents at work, disability, unemployment, military service, etc.). The latter provide important additional information about individual careers, particularly with regard to disability allowances, unemployment benefits, and contribution credits for nonworking parents (AVPF).5 Some demographic data are also included in this file, though they are of poor quality as they are not used for administrative purposes. They are needed only at retirement age to determine eligibility and the amount of the pension. The pensioners sample is a 1/90th sample of information on all persons who had retired before 31 December 2001 (i.e. 117,105 individuals in all). The main elements provided by this file are: the amount of pensions received, the number of quarters validated (contribution period), the average annual wage (on which the pension is based), and supplementary benefits (10 per cent pension increase for parents of three or more children, extra benefits, etc.). 2.2 Simulation Base The initial population comprises samples of both contributors and pensioners. Some corrections and imputations were made as follows: With regard to contributors, corrections must be made to take account of non-registered deaths (namely deaths that occurred while abroad) and of the share of contributors who will not receive retirement pension because they have not contributed for long enough. Vital statistics were essential, particularly for imputing family status (marriage status, number of children, 3

The Social Security ceiling is used as a reference for calculating the overall sum of contributions that will be used to determine pensions; it is reassessed every year. In 2000, the annual ceiling was set at 26,892 euros. 4 The validation rule used by the general scheme requires a person to work at least 200 hours at the minimum wage (SMIC) in order to validate contributions for one quarter, based on a year divided into four quarters. This rule is implicitly based on a part-time job. In 2000, in order to validate four quarters, annual wages had to be at least 4,966 euros. 5 Contribution credits for non-working parents (Assurance Vieillesse des Parents au Foyer, AVPF) are one of the main family benefits of the French pension system. Under this system, recipients of certain family benefits obtain a period of free contributions to the pension scheme to compensate for a temporary reduction in activity that has reduced both the contribution period and the wage level.

A Microsimulation Model of Private Sector Pensions

59

etc.), as this information is not accurately recorded in CNAV files. The total duration of labour force participation was also corrected using data from the Patrimoine (financial assets) survey carried out in 1998 by the French Statistics Office. The other mandatory basic pension schemes (i.e. for publicsector employees and self-employed workers) do not all regularly send information to the CNAV about the activity of people currently affiliated or who have been affiliated with them but who are or were also affiliated to the general scheme.6 This leads to underestimation of the contribution periods and therefore needs to be corrected. With regard to pensioners, we refer only to persons receiving benefits as former contributors to the scheme.7 The initial file was created by combining the two corrected CNAV administrative files (i.e. the contributor and pensioner files). Thus, the new file obtained reflects the situation on 31 December 1994 for contributors and on 31 December 2001 for pensioners — the aim being to provide a backward simulation period, making it possible to adjust the model to statistics recorded for the 1995–2002 period for new pensioners and for the 2001–2002 period for all pensioners. Overall, the initial base is representative of French private-sector employees and comprises 1,319,316 contributors from the 1935–1970 cohorts and 101,282 pensioners from the 1899–1934 cohorts.

3. Model Description The ARTEMIS microsimulation model is dynamic and cross-sectional. It is dynamic because accurate simulation of future changes in contributor and pensioner populations calls for a model that is capable of integrating relatively subtle hypotheses of demographic and economic change. This cannot be achieved with the necessary degree of flexibility using a static model. The model is cross-sectional, because one of the objectives is to simulate the total burden of pensions paid out on a year-by-year basis. As we have access to the biographies of the individuals present at the initial date, it is also possible to study the results by birth cohort — and so the model can also be used as a cohort model. The model is designed to simulate personal pension entitlements. It comprises three modules: demography, labour force participation, and retirement (Appendix 1). The demography module is limited for the time being to mortality; the other demographic variables (fertility, marriage) are imputed 6

Some schemes do not send this information until insured persons claim their pension. 7 We distinguish between benefits acquired in one’s own name through paid labour (personal entitlement) and those acquired as a spouse (entitlement as a spouse).

60

Thierry Debrand et al.

in the initial simulation database. The economy module (labour) simulates access to and withdrawal from the labour force (based on the situation at the beginning of the year) and wages. Particular attention has been paid to the determination of wages as they are a core element of the model. The third module (retirement) is specifically designed to determine pension levels. 3.1 Demography Module Several demographic events are required to determine personal pension entitlements: mortality, fertility, and marriage. The last two are needed to take into account two major family advantages linked to old age pensions: a 10 per cent increase in benefits for people having raised three or more children and a two-year increase (eight quarters) in the contribution period for women for each child raised. Thus, fertility has a double impact: first, on the number of years that women contribute to the scheme and second, on the level of pensions for parents of three or more children. Union history is also included in order to simulate men’s fertility based on that of their spouse, since specific data on male fertility are not yet available. In this first version, matrimonial status and fertility are imputed in the initial database. This simulation method has the advantage of quickly providing useful elements for calculating family benefits. In the near future, it will be replaced by a two-stage method: an imputation up to the simulation start date and a dynamic, stochastic method for the prediction process. Though this imputation takes into account both the intensity and the tempo of fertility, and can therefore affect the wages and labour force participation of women, the dynamic implementation will factor in two-way effects (fertility on wages and wages on number of children). Owing to some constraints within our main data source, the covariates included to determine the probability of demographic events were limited to variables provided in the file. Since variables such as education or socioeconomic status (SES) are not available, this first version of the model cannot simulate fertility or mortality differentials according to education or SES, and the model cannot provide accurate results according to these variables. Methods to address such problems are under investigation. Mortality A study by Gle´nat (2003) shows that the mortality of pensioners belonging to the general scheme is the same as that of the overall French population. To estimate mortality in the model, we can therefore use life tables and mortality forecasts established by the French statistics office (Brutel, 2001). According to the national population projections, mortality is expected to continue decreasing at the same pace as that observed over the past 30 years.

A Microsimulation Model of Private Sector Pensions Table 1:

61

Life Expectancy at Birth and at Age 60 2000

2010

2020

2030

Life expectancy at birth Men Women

74.7 82.2

76.8 84.4

78.7 86.2

80.5 87.8

Life expectancy at age 60 Men Women

19.7 25.1

21.2 26.6

22.7 28.0

24.1 29.3

Source: Brutel (2001); central hypothesis

Based on this assumption, life expectancy at birth (Table 1) will rise from 74.7 years in 2000 to 80.5 years around 2030 for men (and from 82.2 to 87.8 for women). For the 1995–1998 period, the mortality data are taken from the life tables based on vital statistics. The implementation of mortality in the model is simple and uses the 2000–2050 yearly probabilities of dying by sex and age provided by the national statistics office. Marriage Marriage is introduced into this first version to provide a means of determining male fertility, as a man’s lineage is defined by that of his wife. Marital status can be divided into two categories (never married and ever married) and is determined using the proportion of never-married people (i.e. proportion of single people at age 50) within each cohort. The long-term trends indicate that the proportion of never married is remarkably stable (Toulemon, 1997). Among the birth cohorts currently reaching retirement age, the proportion is 8 per cent among women and 10 per cent among men, according to the 1999 census. For people who are designated as married or likely to marry during their lifecourse, the spouse’s year of birth is estimated using a Gaussian distribution whose parameters reproduce the age differences between spouses as observed in the 1999 Survey of Family History (Toulemon and Mazuy, 2001), with a mean age difference of three years. For the youngest birth cohorts, we designate the proportion likely to remain single on the basis of the 1965 cohort data, which is probably unrealistically high, but this makes it possible to take overall fertility into account (including births out of wedlock). Fertility The simulation is carried out in two stages: the first stage consists of determining whether or not the woman will have children during her lifetime and, if so, how many. This is estimated on the basis of the woman’s birth

Thierry Debrand et al.

62 Table 2:

Distribution of Women by Completed Fertility

Birth cohort

1935 1955 1965

Number of children 0 (%)

1(%)

2 (%)

3 (%)

4+(%)

11.5 12.3 16.2

16.6 19.9 21.8

28.8 37.3 33.5

19.7 18.8 18.2

23.4 11.7 10.3

Source: Survey of Family History, 1999.

cohort. Then, for those women who will have children, we establish the fertility timing for the various births based on the total number of children, the woman’s age, the woman’s age at the time of the previous birth and the birth interval (the time since the previous birth). Knowing the children’s birth year is useful, since one of the explanatory variables of labour market transitions for women is the number of children under the age of three. A man’s fertility is presumed to be equal to that of his wife. The data come from the 1999 Survey of Family History (Table 2). We calculate the number of children and the birth intervals for birth cohorts whose fertility is completed (i.e. 1935–1965). For younger cohorts, we assume their fertility to be the same as the 1965 cohort. This assumption is reasonable, as the youngest cohort reaching retirement age at the end of the simulation period is the 1970 cohort. 3.2 Activity Module (Economic) The economic part concerns activity and wages. Activity is simulated in the same way as demographic trends (i.e. stochastically), using event occurrence probabilities. Wages, on the other hand, are determined using an econometric equation, as they lie at the heart of this pension microsimulation model. Activity Activity is determined in several stages. First, the activity rates specific to the general scheme must be defined by determining the respective share of the working population under this scheme and under other schemes. The estimate based on 1997 and 1998 INSEE labour force surveys enables us to determine the percentage of active individuals in the basic schemes which do not transmit information to the CNAV. We thus distinguish between four situations on the labour market: active under the general scheme, active outside the general scheme, unemployed and inactive. For movements in and out of employment, the largest task is to estimate the transition probability specific to the private sector according to age and sex. For women, the number of

A Microsimulation Model of Private Sector Pensions

63

children aged below three is an additional variable affecting these employment transitions. Once activity has been simulated we determine the wage of workers contributing to the general scheme and increment the various counters for duration of employment, including periods covered by contribution credits (unemployment with benefit entitlement and maternity leave). 3.3 Modelling and Econometric Estimation of Wages Wages are modelled by measuring both the influence of workers’ individual characteristics (gender, contribution credits for the receipt of unemployment or disability benefit, and contribution credits for non-working parents (AVPF)) and that of macroeconomic changes (unemployment rate, productivity, and minimum wage) over the last 50 years (Debrand and Privat, 2002a). With the integration of macroeconomic variables, we aim to create a link between macroeconomic changes that are traditionally considered in pension forecasts and individual wages. Wages are incorporated in the model by means of an equation estimated using data from the sample of contributors. Unlike most studies on the link between work experience and wages, which are obliged to base their estimates on cross-sectional data (Jarousse and Mingat, 1986), we can make use of longitudinal data on individual wages. This provides us with true career profiles for which the real wage differences according to the level of experience can be calculated for the same individuals observed at different ages. However, our data source imposes a restriction of another type, since the recorded wages are limited to the Social Security ceiling. The change in wage profiles by birth cohort over the last 50 years (1947–2000) is analysed on a sub-sample of 110,378 persons aged between 16 and 59 (i.e. 1,910,442 observations). We observe that the median gross annual wage profile by age is concave, whatever the birth cohort concerned (see Figures 1 and 2). In other words, wages grow much faster at the start of a career than at the end. Moreover, the concavity of these curves appears to increase over the birth cohorts, while the wage differences between men and women are tending to decrease. However, for people of the same birth cohort and age, women’s wages are always below those of men (Debrand and Privat, 2002b). Owing to the very different shape of the wage curves, the functions of earnings based on the individual data introduced into the microsimulation model are differentiated by gender. The real individual wage is represented by an earnings function derived from the human capital theory (Becker, 1962, 1975). In the database of contributors to the state pension scheme, the notion of work experience, generally determined as the difference between a person’s age and the age when they completed their education, is unknown (since we do not know how long they remained in the education system). In what follows, as in certain other studies faced with the same problem, we use

Thierry Debrand et al.

64 Figure 1:

Median Gross Annual Wage Profile by Birth Cohort (Euros ‘000), Men

Median wages (euros 2000)

25 000 20 000 15 000 10 000 5 000 0 16-19

20-24

25-29

30-34

35-39 Ages

40-44

45-49

50-54

G1935-G1939

G1940-G1944

G1945-G1949

G1950-G1954

G1960-G1964

G1965-G1969

G1970-G1974

G1975-G1979

55-59

G1955-G1959

Note: G1935– G1939: Persons born from year 1935 to 1939. Data: Sample from the 2002 national repository, CNAV

Median wages (euros 2000)

Figure 2:

Median Gross Annual Wage Profile by Birth Cohort (Euros ‘000), Women

25 000 20 000 15 000 10 000 5 000 0 16-19

20-24

25-29

30-34

35-39 40-44 Ages

45-49

50-54

G1935-G1939

G1940-G1944

G1945-G1949

G1950-G1954

G1960-G1964

G1965-G1969

G1970-G1974

G1975-G1979

Note: G1935– G1939: Persons born from year 1935 to 1939. Data: Sample from the 2002 national repository, CNAV

55-59

G1955-G1959

A Microsimulation Model of Private Sector Pensions

65

age as a proxy for experience on the labour market. To take account of certain other characteristics intrinsic to wage earners, the regressions include indicators describing gender, receipt of disability or unemployment benefits and receipt of pension contribution credits for non-working parents (AVPF). Concerning the macroeconomic variables, productivity is a variable often used in predicting pensions — and the minimum wage is a key variable used in most French macroeconomic models to take the institutional framework into account. Furthermore, studies on the future of retirement pensions retain macroeconomic hypotheses on productivity and unemployment in order to evaluate the future resources of and spending on the pension schemes. Numerous econometric studies of the relationship between wage levels and unemployment have been conducted over the years (Keynes, 1936; Bils, 1985; Solon et al., 1994). Some studies have examined the wage implications of interaction between individual characteristics and labour market conditions (measured via the unemployment rate). For instance, Arozamena and Centeno (2001) study the way in which seniority and labour market conditions interact and affect the wage fixation process. They show the influence of external labour market conditions on individual wage setting. Our study follows a similar approach by integrating macroeconomic variables into the earnings function. Finally, a person’s wage is made up of two parts as follows: lnðwi;t Þ ¼ f ðagei;t ; age2i;t ; dummiesi Þ þ gðut ; lnðmwt Þ; lnðptyt ÞÞ

ð1Þ

where u is the unemployment rate, ln(mw) the logarithm of the minimum wage and ln(pty) the logarithm of productivity. The f(.) function corresponds to the internal characteristics of the labour market, which depend upon the company’s and the wage-earner’s characteristics, while g(.) describes the labour market’s external characteristics (in other words, the economic situation).8 Thus, the mean real wage is represented, to within one residual term, in the form of a quadratic function of age, with dummy variables for individual characteristics such as receipt of unemployment benefit, receipt of disability benefits and of contribution credits related to family benefits (AVPF), the macroeconomic variables listed above (unemployment, productivity and minimum wage), and a fixed effect representing the difference between the individual career and that predicted by the estimated profiles. Age is used as a proxy of experience on the labour market. The specificity of wages limited 8

There is extensive literature on the introduction of macrodata in microeconometric models (cf Moulton, 1990; Imbens and Lancaster, 1994). In our study, we do not take account of technical problems suggested by this literature.

66

Thierry Debrand et al.

to the Social Security ceiling is taken into account in this equation via a ratio measuring the impact of truncation at the ceiling level. The equation we have estimated and introduced into the model is written as follows: wi;t ¼ expða þ b agei;t þ c age2i;t þ d dumempi þ e ddisabi þ f davpf i þ h pai þ i ddi þ j ut þ k lnðptyÞt þ 1 lnðmwÞt þ millsÞ

ð2Þ

where wi,t is the wage of individual (i) at a given age (age) and a given date (t). b and c can be interpreted as the specific effects of age on wages (such as experience as a first approximation). The dummy variable dunemp indicates the receipt of unemployment benefits at t (1 for receipt of unemployment benefits, and 0 otherwise); ddisab is a dummy variable for the receipt of disability benefits; davpf a dummy variable for the receipt of family benefits; and dsex a dummy variable for sex. pa and dd are two variables to correct the selection bias due to an unbalanced panel dataset: pa, is a dummy variable that indicates the presence of individual i at t–1 in the sample and dd a dummy variable for individuals having experienced a career break. mills is the ‘‘Mills’ ratio’’ that measures the effect of the limitation at the ceiling The parameter signs on age and age squared (age2) are in line with the expected evolution (i.e. wages increase with age but at a progressively slower pace). The results on the parameters presented here match those found in the earlier studies on earnings functions. We also find a negative impact of periods of unemployment, the receipt of disability benefits, and of the presence of children (measured via the receipt of family allowances). The results show that there is a significant difference between men and women (Table 3). We obtain a concave profile with age and a less pronounced effect for women than for men (the coefficient associated with age is higher for men and the one associated with age squared is lower for women, in absolute terms). The effect of disability appears higher for men than for women. The wage of persons receiving a disability benefit represents 72 per cent of a non-disabled wage for women and 61 per cent for men. We also made an estimation on the general population (column 3), by introducing a sex dummy to measure the wage gap between men and women. This type of variable allows us to show that the wage pattern varies according to sex: the gap measured here is 11.4 per cent on average, in favour of men. We lack several explanatory variables (number of children, length of work, job category, etc.) that might help explain this sizeable gap. If we consider the macroeconomic variables, there are also differences between men and women. Wage elasticities to macroeconomic conditions have the expected signs — i.e. the unemployment rate has a negative effect,

A Microsimulation Model of Private Sector Pensions Table 3:

67

Wage Equations of the Model Using Micro and Macro Economic Data Men

Constant Age Age2/100 Dunemp ddisab

Women

1.202 (0.0279) 0.053 (0.0003) 0.069 (0.0003) 0.481 (0.0025) 0.607 (0.0175)

0.319 (0.0392) 0.010 (0.0003) 0.016 (0.0004) 0.562 (0.0038) 0.723 (0.0250) 0.748 (0.0095)

0.816 (0.0027) 0.092 (0.0013) 0.007 (0.0002) 0.633 (0.0047) 0.491 (0.0037) 0.348 (0.0011) 62,831 1,174,296 0.308

0.969 (0.0040) 0.151 (0.0021) 0.005 (0.0003) 0.563 (0.0070) 0.707 (0.0052) 0.319 (0.0012) 47,547 736,146 0.323

davpf dumsex pa dd u ln(pty) ln(mw) mills N Observations Standard deviation

All 0.661 (0.0228) 0.035 (0.0002) 0.049 (0.0003) 0.524 (0.0022) 0.662 (0.0152) 0.677 (0.0083) 0.114 (0.0012) 0.866 (0.0024) 0.110 (0.0012) 0.006 (0.0002) 0.63 (0.0039) 0.566 (0.0030) 0.355 (0.0008) 110,378 1,910,442 0.313

Note: All the coefficients are significant at the 5 per cent level. N is the size of the sample. The estimation method used is the instrumental variables method to deal with the endogenous nature of human capital, combined with a tobit model to measure the effect of the limitation to the ceiling. Data: Sample from the 2002 national repository, CNAV

while the minimum wage and productivity have a positive effect. So, wage fluctuations are procyclical and these three macroeconomic variables appear to measure the impact of external markets on individual wage formation quite well. Women’s earnings are influenced by unemployment in the same way as men’s. However, productivity has a greater impact on men than on women. Male employment is characterised, on average, by a higher level of qualification and more technical tasks. Similarly, technical or technological

Thierry Debrand et al.

68

advances have a greater influence on male than on female employment which, in turn, has a direct impact on wages. On average, women’s wages are lower than men’s. In principle, there should be more women than men whose earnings are equal to or slightly higher than the minimum wage, so that any increase in the minimum wage will have a greater impact on average female earnings than on male earnings. We need to determine what impact these differences will have on our simulations. Will the gaps between men and women cause structural divergences between men and women? It looks as if the effects might offset each other. For example, an increase in productivity will have a greater impact on male wages than on female wages, but an increase in the minimum wage will have a higher impact on the level of female earnings. The wages determined by these equations are then used to determine pension entitlements, in line with current or future scales. In the simulation, we need to make assumptions about the productivity growth rate and the unemployment rate. 3.4 Pension Module The last module incorporates all the rules for determining the amount of the retirement pension paid out to new pensioners and for annual pension indexation for persons already receiving a pension. Calculating the Basic Pension under the General Scheme We present hereafter how the basic pension for private-sector employees is determined in accordance with the legislation introduced by the 1993 reform. The changes introduced by the 2003 reform are indicated in the next section. The basic pension under the general social security scheme for privatesector employees is calculated using the following formula: P ¼ SAMB  T  D=150 The pension (P) depends on three components: (1) The gross mean annual wage (SAMB) earned during the 10 best working years, progressively being raised to the 25 best years; (2) The pension rate (T) applicable to this wage; (3) The duration of contributions to the general scheme (D) expressed in quarters and limited to 150. The gross mean annual wage is determined according to the wages used for calculating pension contributions over the working career, i.e. gross wages limited to the Social Security ceiling fixed each year. These wages are

A Microsimulation Model of Private Sector Pensions

69

indexed by official coefficients close to the price index. The mean annual wage is obtained by averaging the best years of these indexed wages. The 1993 reform has introduced a progressive increase in the number of ‘‘best years’’ in the calculation of the mean annual wage, from the 10 best years to the 25 best years in 2008. The duration of contributions to the general scheme is used to calculate the amount of the pension paid out under this scheme. It includes the quarters validated by compulsory or voluntary pension contributions, periods where contribution credits are awarded on the basis of a fixed rate wage (for contributors receiving the AVPF, for example) and some other periods for which contribution credits are awarded. The duration of contributions is extended for mothers (two quarters per child raised for at least 9 years of its first 16 years), for parental leave to raise young children, and for contributors aged over 65 with fewer than 150 quarters of contributions to the general scheme (2.5 per cent of their contribution years per completed quarter between their 65th birthday and their pension starting point, up to a maximum of 150 quarters). From 1993 to 2003, the duration of contributions required to receive a full pension increased by one quarter per birth cohort, from 150 to 160 quarters, and 40 years of contributions are now required to obtain a full pension. The pension rate is determined according to the beneficiary’s age and duration of contributions (all schemes together). It may vary between 25 and 50 per cent (maximum or ‘‘full’’ rate). Between age 60 and 65, contributors who have a sufficient duration of contributions (under all schemes) are entitled to a full retirement pension (50 per cent) while contributors without the duration of contributions required (all schemes together) are entitled to a reduced rate of between 25 and 50 per cent. From the age of 65, the 50 per cent full rate is systematically awarded, even if the beneficiary does not have the required number of quarters. However, certain special categories of contributors — those with disabilities or incapacities — may obtain a full pension at age 60, whatever the duration of their contribution. For the others, if a person retires with fewer than 160 quarters of contributions, a reduced-rate pension is awarded. In this case, the full rate is adjusted by a coefficient determined either according to the number of missing quarters with respect to the required duration (all schemes and periods being recognised as equivalent), or according to the number of quarters corresponding to the time between the age at which the pension is awarded and the person’s 65th birthday. The smaller of these two numbers is used, to the contributor’s advantage. The reduction rate is 2.5 per cent per missing quarter (i.e., the 50 per cent full rate is reduced by 1.25 points for each missing quarter). The total calculated amount (excluding supplementary benefits) is then compared with the contributory minimum and the maximum pension.

Thierry Debrand et al.

70

A 10-per cent bonus is then awarded for persons who have raised three children or more. The pension amount is adjusted each year according to current scales. People retiring are then classified according to the four different kinds of pensions attributed by the scheme: normal pensions at the full rate or at a reduced rate, pensions for people who previously received a disability allowance, pensions for persons with an incapacity, and minimum pensions. For each cohort, the distribution of retirees by pension category is based on the 1935 cohort distribution (cf. Table 4). The retirement decision is modelled using this distribution, and by assigning people to each pension category on the basis of their individual characteristics. For instance, length of contribution to the scheme, wage level, periods of disability or illness during the working life are used as criteria to assign people to the type of retirement pension they are supposed to receive. Thus, nearly 92 per cent of the 1935 birth cohort and the following ones will receive a pension at the full rate. Two reasons led to this choice of entitlement distribution based on the 1935 birth cohort: the regularity of distribution by category of rights over recent years, and stricter new rules for obtaining a full pension that may encourage certain people to apply for a retirement pension on the basis of disability or incapacity. This means that the observed percentage of 92 per cent for the 1935 cohort is a reasonable assumption for the cohort reaching retirement age during the simulation period. For other types of entitlement (normal full pensions and minimum pensions), the rule of retirement with a full pension is applied. This rule will lead automatically to postponement of retirement age — because of a higher age Table 4:

Distribution of New Retirees by Age and Pension Type for Men Born in 1935

Age

Normal full pension (%)

Normal reduced-rate pension (%)

Disability pension (%)

Incapacity pension (%)

Minimum pension (%)

Total (%)

47.0 2.6 1.7 1.2 1.0 6.0 59.5

6.2 1.0 0.5 0.3 0.4 0.0 8.4

7.5 0.0 0.0 0.0 0.0 0.0 7.5

7.1 0.7 0.5 0.4 0.2 0.0 8.9

7.2 0.6 0.4 0.4 0.5 6.6 15.7

75.0 4.9 3.1 2.3 2.1 12.6 100.0

60 61 62 63 64 65 Total

Source: Pensions awarded from 1995 to 2001, CNAV Note: Retirees at age 65 include all retirees aged 65 and over. Reference population: Retirees with direct pension entitlement.

A Microsimulation Model of Private Sector Pensions

71

at entry into the labour market due to longer time in education and also because of the lengthening of the contribution period required to obtain a full pension.

4. Simulation Results This final part presents some results describing the situation of privatesector pensioners in the years leading up to 2030. In the first section, the baseline scenario uses the conditions of the French pension system as defined by the 1993 reform. And in the second section, the impact of some elements of the new 2003 reforms is assessed. As mentioned earlier, microsimulation has a dual advantage in that it models both individual situations and aggregates (by summation of individual situations). It is this dual set of results that we will illustrate here in terms of number of persons taking retirement, total number of pensioners from 2001 to 2030, distributions of pension amounts, and distribution of changes in retirement ages. 4.1 Baseline Projection The baseline simulation uses the parameters described in the previous section, including the pension scale as defined by the 1993 reform (Blanchet and Legros, 2002). Pensions and reference variables used to calculate pensions (wage indexes, minimum pension, etc.) are price indexed. The macroeconomic assumptions used are those of the latest report by the Pensions Advisory Council (COR, 2001). The mean annual wage rises by 1.1 per cent in 2001 and 2003, 1.3 per cent in 2003, and 1.6 per cent from then on. Pensions are price indexed. Unemployment is assumed to decrease from 8.8 to 5 per cent between 2003 and 2030.9 Indexing of wages, of other benefits (contribution credits for non-working parents (AVTS), contributory minimum, etc.) and of the pension takes place at the same time and at the same pace. Number of Pensioners and Pension Amounts The first noticeable change, illustrated in the following graph (Figure 3), is the rapid and continuous growth in the number of pensioners over the coming years. Altogether, this number will increase from 9.7 million persons 9 The results of the simulation model are very close to the other estimates available (such as the total number of retired people listed by the CNAV for the years 2001 and 2002 in the past, and the results of the macroeconomic model for the simulation period) in terms of both numbers and changes.

Thierry Debrand et al.

72

Number of Pensioners and Number of Retirements (2002–2030)

20.0

900 000

Number of pensioners (in million)

18.0 800 000

16.0 14.0

700 000

12.0 10.0

600 000

8.0 500 000

6.0 4.0

400 000

2.0 0.0

Number of retirement (flow)

Figure 3:

300 000

2002 2004 2006 2008 2010 2012 2014 2016 2018 2020 2022 2024 2026 2028 2030

Years Number of pensioners

Annual number of retirement

Source: ARTEMIS microsimulation model

to 17.7 million — i.e. more than 80 per cent. This result is hardly surprising, as the first generation of baby-boomers reaches retirement age in 2006. At the same time, the mean annual pension will grow by 29 per cent between 2003 and 2030, rising from h5,573 to h7,195. As a result, the total amount of pensions paid more than doubles between 2003 and 2030 (Table 5). This will be the most critical period in terms of retirement system sustainability, as it corresponds to the years when the retired population will be the most numerous — both for demographic reasons (increased life expectancy, numerous baby-boom cohorts) and economic reasons (‘‘noria’’ effect due to the retirement of cohorts with high salaries, at least for the first ones). Distribution of Pension Levels In addition to macroeconomic results, the model gives information on individual situations. At the individual level, the distribution of pension levels is more important and interesting than the average pension alone. With the changes in the pension quartiles shown in Table 6, we can study whether or not the increase in average pensions is distributed uniformly among the different wage levels. Younger cohorts reach retirement age with higher wages than older ones. This leads to an increase in pensions at retirement despite the increasing impact of the 1993 reform (which will reach its full effect in 2008 with the 1948 birth cohort). The increase in pensions seems to be quite equally shared, whatever the level of wages. The distribution changes noticeably over the period under study and, according to the model, becomes less unequal.

A Microsimulation Model of Private Sector Pensions Table 5: jection)

73

Total Amount of Old Age Pensions and Average Pension (Baseline Pro-

Year

2003 2010 2020 2030 Change 2003–2030 (%)

Pensioners (in millions)

Average pension (in euros 2000)

Total expenditures (in billion of euros 2000)

9.7 11.8 15.0 17.7 +82

5,541 5,713 5,982 6,483 +17

53.9 67.4 90.0 114.7 +113

Source: ARTEMIS microsimulation model

Table 6:

Pension Quartiles at the Retirement Year (in 2000 Euros)

Birth cohort 1935–1939 1940–1944 1945–1954 1955–1964 1965–1970

Average

1st Quartile

Median

3rd Quartile

Interquartile Interval

5,609 5,756 6,083 6,680 7,603

1,991 2,114 2,681 3,503 4,008

5,116 5,240 5,761 6,092 6,702

9,130 9,280 9,119 9,463 10,897

4.6 4.4 3.4 2.7 2.7

Source: ARTEMIS microsimulation model

For the youngest generations, the average pension and the pension quartiles are much higher due to a simple selection bias. Indeed, a high percentage of the members of these cohorts are not retired because they have not yet reached the age of 65. Those already retired are therefore those who are entitled to a pension at the full rate before the age of 65, and whose pension levels tend to be above the average. Although the 1993 reform will not be completely effective until the 1948 cohort retires, it starts to modify the rules from the 1934 cohort onwards. It is interesting therefore to analyse the effects of this reform, in terms of retirement age and duration of contributions by birth cohorts. We will see that retirement age is also quite different for men and women. Distribution of Contribution Periods at Age 60 The validated period is a determining variable for retirement, as it has an effect on both the level of pensions and the retirement age. Although it is possible to retire at age 60, retirement at the full rate is only possible if the contribution period reaches at least 150 quarters up to 1993, and between 150 and 160 quarters thereafter. Currently, the basic French retirement

Thierry Debrand et al.

74 Figure 4:

Distribution of Years of Contribution at Age 60 by Birth Cohort

35% 30%

b1935-1939

25%

b1940-1944 b1945-1954

20%

b1955-1964

15%

b1965-1970

10% 5% 0% 0-19

20-39 40-59

60-79

80-99

100119

120139

140159

160179

180199

200+

in quarters

Source: ARTEMIS microsimulation model

system strongly encourages people to retire once they are able to obtain the full pension rate. The penalties for early retirement are relatively high and the benefits in the event of a postponed retirement are low, so the incentives given by the current retirement rules lead to a high proportion of workers retiring at the full rate (Blanchet and Pele´, 1997). To estimate the impact of reforms that lengthen the contribution period required to obtain the full rate, the changes in contributors’ situation at age 60 needs to be analysed. The graph below gives the distribution of the number of years of contributions at age 60 by birth cohort and shows that the distribution seems to become tighter from one birth cohort to the next (Figure 4). The most striking change is the shrinking proportion of long working careers for men: this can be explained in part by the longer time spent in education, which leads to postponed entry into the labour force. Over the generations, the distributions tend to become tighter. For women, the main change is a decrease in the proportion of short careers. The two distributions tend to converge. Changes in Retirement Age Changes in the contribution period are expressed by the mean retirement age. We can see that the mean age increases from one group of birth cohorts to the next (Figure 5): from 61.3 years for the 1940–1944 cohorts to 61.8 for the 1955–1964 ones. The proportion of persons who retire at age 60 and 65 varies quite strongly. The proportion of retirements at age 60 seems to decrease when retirements at age 65 increase. The proportion of retirements between these two ages is quite stable. This change is directly linked to the 1993 pension reform and the increase in the contribution period required to obtain the full pension rate.

A Microsimulation Model of Private Sector Pensions Figure 5:

75

Distribution of Mean Retirement Age by Birth Cohorts

80% 70% 60% 50% 40% 30% 20% 10% 0% 60

61

62

63

64

65

b1935-1939 (mean age : 61,3)

b1940-1944 (mean age : 61,4)

b1945-1954 (mean age : 61,6)

b1955-1964 (mean age : 61,8)

Source: ARTEMIS microsimulation model

The changes are different for men and women. For men, mean age at retirement increases by 0.5 years from 61.1 for the 1935–1939 birth cohorts to 61.6 for the 1955–1964 cohorts (Figure 5). As most men have long careers, more than 70 per cent of them currently retire at age 60. However, for the cohorts born after 1954, long careers tend to disappear. This implies that the proportion of people retiring at the full rate at age 60 will fall dramatically and that retirement will be postponed to age 65, when the full rate is systematically granted. For women, the retirement age distribution becomes more similar to that of men. Nevertheless, the stable proportion of retirements at age 65 is a quite surprising result, as we would have expected a decrease in the propensity to retire at this age as a result of improvements in women’s careers. This phenomenon may be the result of the 1993 pension reform that counterbalances career effects for women. 4.2 Some Effects of the 2003 French Pension Reforms This section deals with the effects of two changes in the rules introduced with the 2003 pension reform — i.e. the lengthening of the contribution period required for entitlement to a full pension and the new rules for calculating the pension rate when a person wishes to retire before 65 without the contribution period required to obtain full rate. For the general pension scheme, the duration required to obtain a full rate is not increased before 2009 because some other pension schemes, such as the civil service pension scheme, have to be aligned with the same contribution period and this will be completed in 2008. The increase after 2008 will be based on the principle of sharing future increases in life expectancy equally between working life

76

Table 7:

Age at Retirement (in Comparison with Baseline Scenario) All

1940–1944 1945–1954 1955–1964 1965–1970a

Total

Persons affected

Total

Women

Persons affected

Total

Persons affected

Average difference

%

Difference (in years)

Average difference

%

Difference (in years)

Average difference

%

Difference (in years)

0.0 0.1 0.4 0.3

0 10 25 23

0.1 0.8 1.5 1.4

0.0 0.1 0.5 0.4

0.8 11 29 26

0.1 0.9 1.6 1.4

0.0 0.1 0.3 0.3

0 8 22 20

0.0 0.8 1.3 1.3

Source: ARTEMIS microsimulation model a Includes only retirements until 2030.

Thierry Debrand et al.

Birth cohorts

Men

A Microsimulation Model of Private Sector Pensions

77

and retirement. The reform will then be continued in such a way as to maintain a constant ratio between length of working life and length of retirement. The results that follow include the changes already programmed up to 2020. From that date, the increases should continue according to the increase in life expectancy at age 60, though the details will be decided later. So for this variant of the simulation, we assume that the contribution period goes up from 160 quarters in 2008 to 164 in 2010, then increases by one quarter per year until 2020 and remains at this level from then on. This lengthening of the contribution period affects both the pension rate (for retirement before age 65) and the pension amount, as the pension is determined according to the pension rate and contribution period. The number of quarters used to calculate the reduced pension rate is itself increased from 150 quarters in 2008 to 167 in 2020, in alignment with the contribution quarters required for a full rate pension. Table 7 illustrates how the different birth cohorts are affected by the 2003 pension reform, in terms of age at retirement compared to the situation without the reform (baseline scenario). The 2003 reform leads to a quite modest increase of five months in mean age at retirement for the 1955–1964 birth cohorts. But for the same birth cohorts, if we focus exclusively on persons affected by the reforms, we observe a postponement of retirement by 1.3 years for women and 1.6 years for men. Table 8 illustrates the effect of the 2003 reform on mean pension at retirement age for different birth cohorts. It will lead to a 7 per cent reduction in the mean pension compared with the baseline scenario for the 1955–1964 birth cohorts. This 2003 reform increases the effect of the 1993 reforms, estimated by the model at 21 per cent. It also shows that the effects of the 2003 reform are similar for men and women.

Table 8: Mean Pension at Retirement Age by Birth Cohorts (by Comparison with the Baseline Scenario) Birth cohorts 1940–1944 1945–1954 1955–1964 1965–1970a

All (%)

Men (%)

Women (%)

0 5 7 10

0 4 7 9

0 5 7 11

Source: ARTEMIS microsimulation model Interpretation: The pension received by men born in 1945– 1954 would have been 4 per cent higher without the reform than with the reform. a Includes only retirements until 2030.

78

Thierry Debrand et al.

At the expenses level, in the baseline scenario (without the 2003 reform), the total cost increases from 54 billion euros in 2002 to 115 billion euros in 2030. The lengthening of the contribution period required to receive a full pension and of the period used to calculate the reduced rate introduced by the 2003 reform leads to total expenses in 2030 of 105 billion euros — a reduction of 10 billion euros and a 9 per cent decrease in total cost compared with the baseline scenario. The estimates made during the preparation of the 2003 reform gave expenses of 4.3 billion euros in 2020 and our model gives similar results, indicating that our assumptions on retirement behaviour, though crude, do not appear to underestimate the effects of the reform.

5. Conclusion According to ARTEMIS, the number of pensioners under the general scheme would have reached 18.1 million under the rules prior to 1993 and this has been reduced to 17.3 million through the reforms of 2003. Because age at retirement changes very little, the increase in the number of pensioners is only slightly reduced by the reforms of 1993 and 2003. Their main impact is on the mean pension amount. On the basis of these hypotheses, the 2003 reform will result in a 5 per cent pension reduction for persons born in 1945–1954, and a 7 and 10 per cent reduction, respectively for the younger cohorts in the year 2030. The microsimulation model of pensions under the general scheme is designed to serve as a complement to the macroeconomic projections currently being performed by CNAV. It is an operational tool to assess the impact of reforms at both individual and global levels. Thanks to this model, it is possible not only to simulate the effects of different planned reforms on the calculation of pensions — but also the impact of reforms on the methods used to award family advantages. The simulations matched to the CNAV statistics can be interpreted in terms of both mass and distribution. Though the ARTEMIS model applies solely to contributors and pensioners under the general scheme, it is original in the sense that it factors in an individual changing from one scheme to another (general scheme/other schemes), and its construction is based on the principle that the institutional rules specific to the general scheme and the wide range of different benefits paid out by this scheme must be represented as accurately as possible. The simultaneous incorporation of microeconomic and macroeconomic components in the determination of wages will enable us to determine the mean wage of an individual for a given birth cohort, to observe the elasticity of wage levels with respect to macroeconomic variables and hence to take more accurate account of the different macroeconomic scenarios presented in the COR report.

A Microsimulation Model of Private Sector Pensions

79

Thanks to the flexibility of our model, it will then be easy to simulate the impact of variants on the different parameters and to introduce new elements in the model specification and outputs.

References Arozamena, L. and Centeno, M. (2001). Tenure, Business Cycle and the Wage-Setting Process. Paper presented at the EEA Conference, 31pp. Becker, G.S. (1962). Investment in Human Capital: A Theoretical Analysis. Journal of Political Economy, 70(5), 9–49. Becker, G.S. (1975). Human Capital: A Theoretical and Empirical Analysis, 2nd edition, Columbia University Press, New York. Bils, M.J. (1985). Real wages over the Business Cycle: Evidence From Panel Data. Journal of Political Economy, 93(44), 666–689. Blanchet, D. and Legros, F. (2002). France: The Difficult Path to Consensual Reforms, in Feldstein, M. and Siebert, H. (eds), Social Security Pension Reform in Europe, NBER, The University of Chicago Press, pp. 109–136. Blanchet, D. and Pele´, L.-P. (1997). Social Security and Retirement in France. NBER Working Paper, no 6214. Brutel, C. (2001). Projections de population a` l’horizon 2050. Insee Premie`re, 762, mai, 4p. Conseil d’orientation des retraites (COR) (2001). Retraites: renouveler le contrat social entre les ge´ne´rations. Orientations et de´bats. La Documentation franc- aise, 400p. Debrand, T. and Privat, A.-G. (2002a). Individual Real Wages over the Business Cycle: The Impact of Macroeconomic Variations on Individual Careers and Implications concerning Retirement Pensions. Document de travail Ined, no 111, 37p. Debrand, T. and Privat, A.-G. (2002b). L’e´volution des carrie`res salariales au cours des cinquante dernie`res anne´es. Retraite et Socie´te´, 36, 188–202. Gle´nat, M. (2003). Table de mortalite´ du re´gime ge´ne´ral 1998–1999. Retraite et Socie´te´, 40, 192–203. Imbens, G.W. and Lancaster, T. (1994). Combining Micro and Macro Data in Microeconometric Models. Review of Economic Studies, 61, 655–680. Jarousse, J.-P. and Mingat, A. (1986). Un re´examen du mode`le de gain de Mincer. Revue e´conomique, 37(6) November, 999–1031. Keynes, J.M. (1936). The General Theory of Employment, Interest and Money. London, Macmillan. Moulton, B.R. (1990). An Illustration of a Pitfall in Estimating the Effects of Aggregate Variables on Micro Units. The Review of Economics and Statistics, 72(2), 334–338. Solon, G., Barsky, R. and Parker, J.A. (1994). Measuring the Cyclicality of Real Wages: How Important is Composition Bias. Quarterly Journal of Economics, 109, 1–26. Toulemon, L. (1997). Cohabitation is here to stay. Population: An English Selection, 93, 11–46. Toulemon, L. and Mazuy, M. (2001). Cinq projections de fe´condite´ fonde´es sur une hypothe`se de stabilite´ des comportements. Population, 4, 647–656.

Thierry Debrand et al.

80

Appendix 1. Events and Variables Used to Model the Various Events Event

Variables

Mortality

Death rates by age and sex Source: French statistical office population projections

Fertility

In three steps: Distribution by total number of children ¼ a priori parity by birth cohort of the woman or the spouse whether the individual under consideration is a man or a woman. Age at 1st birth by a priori parity Following birth by age of mother, a priori parity, parity, number of years since previous birth Source: French survey on family history, 1990, 1999; authors own calculation

Union

Union rate (ever-married before age 50) by birth cohort and sex In the near future, by age and sex Source: French survey on family history, 1990, 1999; authors own calculation

Labour force participation status

Transition probability according to age, sex, former labour force participation status (employed under the general scheme, employed under another scheme, unemployed receiving unemployment benefit, unemployed without benefit or not working), and, for women only, the number of children aged under 3 Source: Survey on labour force participation, 1996– 1998

Wages

Individual characteristics Age, age2, unemployed, disabled or not, receiving contribution credits for non-working parents  Macroeconomic characteristics Unemployment rate, labour productivity, level of minimum wage  Individual effects difference between predicted and observed values Source: The contributors database of the general scheme limited to the 1935 to 1979 birth cohorts; authors own calculations 

Chapter 4

Effects of Demographic Developments, Labour Supply and Pension Reforms on the Future Pension Burden in Norway Dennis Fredriksen and Nils Martin Stølen Statistics Norway

Abstract A much higher old-age dependency ratio, together with more generous pension benefits, will lead to a substantial increase in the future pension burden in Norway. The magnitude of the challenges of financing increasing pension expenditures depend on: the development in demographic characteristics like fertility, mortality and immigration; characteristics affecting the supply of labour, like education, disability, the retirement age, participation rates and part-time work (especially for women); and the design of the pension system. By use of a dynamic microsimulation model, MOSART, this chapter analyses and projects how these factors will affect the expenditures and financing of the Norwegian National Insurance Scheme. We also present some results about the likely distributional effects of possible pension reforms.

1. Introduction Like most Western countries, Norway will face an ageing population in the coming decades. The growth in the supply of labour is slowing down, and the old-age dependency ratio is expected to be significantly higher in the future. Together with more generous pension benefits as a consequence of the present Norwegian social security system, this will lead to a substantial increase in the future pension burden. A growing need for health care and social services also means a higher demand for labour in these sectors and a further increase in public expenditures. Awareness of these problems and a discussion of how they should be met have been present in most countries in the last 15–20 years. Comprehensive studies discussing the situation in several countries may be found in reports International Symposia in Economic Theory and Econometrics, Vol. 15 r 2007 Published by Elsevier B.V. ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15004-8

82

Dennis Fredriksen and Nils Martin Stølen

from the Organisation for Economic Co-operation and Development (OECD) (1996 and 1998), the European Commission (2001) and Kotlikoff and Leibfritz (1999). Among others, Disney (2000) has discussed how to meet the crises in pension programmes in the OECD countries. In the US, Feldstein (1996 and 1998) has advocated funding of public pensions, and America’s demographic dilemma is analysed by Kotlikoff et al. (2001) and by Diamond (2004). In a survey, Lindbeck and Persson (2003) have discussed some of the main principles for pension reforms, regarding the effects on efficiency, distribution and stability. In Norway, the challenges with the increasing pension burden have been discussed in several white papers from the Government — see the LongTerm Programme for 2002–2005 (St. meld. nr. 30, 2000–2001) and the National Budget for 2004 (St. meld. nr. 1, 2003–2004). According to these white papers, expenditure on old age and disability benefits in the National Insurance Scheme is estimated to increase from about 9 per cent of GDP in 2002 to about 20 per cent in 2050. Possible reforms aiming to reduce future pension expenditures have been discussed in NOU (1998:10), NOU (1998:19) and in the report from the recent Norwegian Pension Commission, NOU (2004:1), and this chapter assesses the likely impact of some of the possible reforms under discussion.

The Existing Norwegian National Insurance Scheme Pension benefits from the National Insurance Scheme (NIS) are based on entitlements each person achieves through his or her working years. NIS has its own measuring unit, called the Basic Pension Unit (BPU). The BPU is used to calculate pension entitlements and adjust pension benefits according to inflation and general economic growth. Pension benefits (PB) paid to old age, disability and survivor pensioners consist of a Basic Pension (BP) and the maximum of a Special Supplement (SS) and a Supplementary Pension (SP). That is: PB ¼ BP + Max (SS, SP) A pensioner married to another pensioner receives a Basic Pension of 85 per cent of BPU (from 2006), while single pensioners receive 1 BPU. The Special Supplement for single pensioners was 79.33 per cent of BPU in 2003. The sum of BP and SS is the Minimum Pension Benefit all pensioners are guaranteed. The Supplementary Pension is based on previous labour market earnings. Each year when the person is aged between 17 to 69 years old, the labour market earnings are translated into Pension Points (PP), by using the BPU of the year income (Y) was earned: 8 0 If : YoBPU > > > < ðY  BPUÞ=BPU If : BPU  Yo6 BPU pp ¼ > 5 þ ðY  6 BPUÞ=3 BPU If : 6 BPU  Yo12 BPU > > : 7 If : Y  12 BPU

Future Pension Burden in Norway

83

The main rule for calculating Pension Points is that labour market earnings exceeding BPU are divided by BPU. Labour market earnings exceeding 6 BPU are divided by 3 BPU, and earnings exceeding 12 BPU are neglected. The Final Pension Point (FPP) is calculated as the average of the 20 largest positive PPs, while Pension Point Years (PPY) is the number of years with labour market earnings above BPU. The Supplementary Pension is calculated when using the BPU at the time pension benefit is received: SP ¼ SPR  Min (PPY, 40)/40  BPU  FPP SPR represents a marginal benefit-wage ratio, and its present value is 42 per cent. The second term, PPY/40, represents the earning time percentage. The last two terms, BPU x FPP, represent an income base and, for incomes above 1 BPU and below 6 BPU, it is equal to the former income received as an employee, indexed by the growth in BPU.

The future developments in the old-age dependency ratio and in the future pension burden depend on demographic characteristics like fertility, mortality and immigration, as well as factors important for labour supply, like education, disability, retirement age, labour force participation rates and the prevalence of part-time work (especially for women). One principal objective of this chapter is to provide a systematic quantitative analysis of the sensitivity of future outlays to changes in these characteristics. Another principal objective is to estimate effects on the future pension burden of different pension reforms. Based on detailed register data and use of a dynamic microsimulation model, MOSART (Fredriksen, 1998), it is possible to shed light on the empirical significance of central questions raised in the theoretical literature. Corresponding analyses may be found for countries like Denmark and the Netherlands (Beetsma et al., 2003; Velferdskommissionen, 2004). In these studies, the consequences of ageing related to different assumptions regarding demographic developments, labour supply and the pension system are analysed by using general equilibrium models with overlapping generations. Although a macro approach seems very suitable for analysing the overall effects from different demographic assumptions and labour supply for the old-age dependency ratio and macroeconomic variables, macroeconomic models may lose many potentially important details in analyses of effects of the pension burden. In many countries (and Norway is a good example) the pension system is complicated, including non-linearities regarding the accumulation of pension entitlements. A microsimulation model — including demographic characteristics, labour supply and an accurate description of the pension system — therefore seems to be the most appropriate tool in order to obtain accurate estimates of the direct effects of the future pension burden. This kind of model seems even more relevant when analysing the

84

Dennis Fredriksen and Nils Martin Stølen

direct effects of pension reforms on individual benefits and government expenditures. Possible distributional effects connected to shifts in the pension system may also be included in a consistent way. Macroeconomic effects, including general equilibrium adjustments, of pension reforms are estimated in Fredriksen et al. (2005) and in Chapter 5 in this volume, where the results from the MOSART model are incorporated into a general equilibrium model. Fertility, mortality and migration are the three components that shape the age structure, and it is of interest to analyse how the future pension burden may be influenced by different assumptions about further developments in these areas in the next few decades. In Norway, there has been a significant growth in average life expectancy since the National Insurance system was introduced in 1967 — and this increase is expected to continue in the future. Although fertility has decreased, it is still relatively high in Norway compared to most other Western countries. In recent decades, migration has become a more important factor for population development in Norway, but migration is highly regulated by the authorities. The main components affecting labour supply consist of participation rates and part-time work (especially for women) and the ages when respectively entering the labour force from education and when retiring. Compared to 1967, the expected retirement age has decreased from about 66–67 years to the current level of 59. This development is mainly caused by an increase in the number of disability pensioners, but a reduction in the formal retirement age in 1973 and early retirement arrangements are also of importance. Although there has been a significant increase in female participation rates in the last few decades, women have not reached the male participation level — and especially not among those over age 50. Many women also work part time and thus represent a potential for further increase in the supply of labour. The number of years spent in the educational system may also have an effect on the aggregate supply of labour. Analyses with a microsimulation model like MOSART have been very beneficial in shedding light on the effects from different elements of the pension system and different reforms discussed by the recent Norwegian Pension Commission (NOU, 2004:1). One of the suggestions outlined in a preliminary report (Norwegian Pension Commission, 2002) was to stop the building up of new entitlements for supplementary pensions. The National Insurance Scheme would then move towards a minimum pension system. An alternative way to reduce future pension expenditures may be to index pension entitlements by less than wage growth, also moving towards a minimum pension system in the long run. A more modest shift may be to index entitlements according to wage growth, but letting pensions after retirement only grow according to the increase in consumer prices.

Future Pension Burden in Norway

85

Another direction in accordance with the main suggestion from the Commission in its final report is to alter the pension system towards a more actuarial system. The main effects compared with the existing system are higher proportionality between pension entitlements and former labour income and to make the yearly pension benefits more dependent on retirement age and life expectancy (smaller pensions if you retire early or life expectancy increases). These elements of reform may also stimulate labour supply, by prompting higher retirement age and increased labour supply among people of working age. But, as pointed out by Gruber and Wise (2002) and Lindbeck and Persson (2003), a shift to a more actuarial system may have distributional effects between different groups of the population. A brief presentation of the MOSART model, the main assumptions regarding demographics and supply of labour, as well as projections of the labour force, the number of pensioners and the expenditures in the National Insurance System towards 2080, are presented and discussed in Section 2. The effects of different assumptions regarding demographic developments and labour supply are discussed in Section 3. Some consequences of a shift towards a minimum pension system are discussed in Section 4, while the effects on labour supply, the pension burden and some distributional effects are discussed in Section 5. Section 6 concludes.

2. The MOSART Model and Projections of Pension Expenditures Tax and benefit rules are often detailed and complicated. Different parts of the population may face different rules and there may be substantial problems of aggregation in calculating the total effect on government budgets of changes in taxes or benefits. To meet these problems the use of microsimulation models, as advocated among others by Orcutt et al. (1986), has been more and more commonly used in the last few decades to support governments with analyses regarding the effects of different social and financial policies. The basic idea in microsimulation modelling is to represent a socioeconomic system by a sample of decision units (e.g., persons), and then model the behaviour of these primary units. Contrary to what is possible in a macroeconomic approach, the detailed and complicated tax and benefit rules may be exactly reproduced. Aggregated numbers are obtained by multiplying the variable of interest for each unit with its sample weight and then summing across the sample. When analysing the National Insurance System in Norway, information about the heterogeneity of the population is important, because the building up of pension entitlements depends on former labour incomes in a nonlinear way (cf. the outline of the pension system in Section 4). The National Insurance System in Norway is a pay-as-you-go system, and the pension

86

Dennis Fredriksen and Nils Martin Stølen

burden is also highly dependent on the development of the population by age and the size of the labour force. From a representative sample of the population in a base year, the MOSART model simulates the further life course for each person in this initial population. The life course is simulated by possible transitions from one state to another, decided by transition probabilities depending on each person’s characteristics. Each of these transition probabilities is estimated from observed transitions in a recent period. Events included in the simulation are migration, deaths, births, marriages, divorces, educational activities, retirements and labour-force participation. Public pension benefits are calculated from labour-market earnings and other characteristics included in the simulation. Old-age pensions, disability pensions, survival pensions and early retirement pensions are included in the model. The analyses in this paper are based on a representative sample from 1993 that is mainly calibrated to the situation in 2001 (cf. Table 1). The demographic assumptions are based on Statistics Norway’s demographic projections from December 2002. A total fertility rate of 1.8 and a net immigration of 13,000 persons each year imply that the size of the different cohorts stabilizes towards 2050. The aggregate population may however increase — as a result of a further increase in life expectancy at birth of about 7–8 years in the same period and then a further increase towards 2100. The assumptions about probabilities for entering disability are based on the observations from 2001, which represent an average for the fluctuating probabilities during the 1990s. For early retirement schemes, there has also been a growing probability for those entitled to enter these schemes during the 1990s, and the projections are based on the observed level from 2001. This is also the case for assumptions about participation in the labour force and working hours.

Table 1:

Main Underlying Assumptions in MOSART

Net immigration Life expectancy at birth Total fertility rate Propensities to study Propensities for entering into disability Propensities for entering early retirement schemes Formal retirement age Labour-market participation rates Distribution of labour incomes during life course Wages, prices, basic unit

13,000 persons per year Increases 7–8 years towards 2050 1.8 As in 2001 Observed level from 2001 Observed level from 2001 67 years Observed level from 2001 Observed from the period 1967 to 1993 As in 2001

Future Pension Burden in Norway Figure 1:

87

Estimated Future Number of Pensioners

2250 2000 Early retirees

1000 persons

1750

Disability pensioners

1500 1250 Surviving spouses

1000 750

Old age pensioners

500 250 0 1960

1980

2000

2020

2040

2060

2080

Year

The necessary information about distribution of incomes between individuals over the life cycle is based on observations from a longer period. When pension entitlements are indexed by wage growth in the projections, the choice of base year for wages, prices and the basic unit in the insurance system is of minor importance. For convenience the level from 2001 is chosen. The main implications of these assumptions for the number of pensioners in the next few decades are presented in Figure 1. The figure also shows that there has been a substantial growth in the number of pensioners since 1967, when the National Insurance Scheme was introduced, generating the increasing pension burden of the past decades. Specifically, the number of disability pensioners has increased and, during the 1990s, the use of early retirement schemes became more common. The number of old-age pensioners shifted upwards when the formal retirement age was reduced from 70 to 67 years in 1973. These three components are the main reasons why the expected average retirement age has decreased from about 66–67 years in 1967 to about 59 years in 2001. The increase in the number of old-age pensioners since 1967 is also caused by a growth in life expectancy and, from the assumptions outlined in Table 1, this growth is expected to continue. Together with the decrease in the actual average retirement age, the increase in average life expectancy implies that the today’s young on average will be pensioners for more than 25 years. As a consequence, the number of pensioners is now expected to be three to four times higher in 2040 than the observed level from 1967.

Dennis Fredriksen and Nils Martin Stølen

88 Figure 2:

Estimated Average Old-Age Pension Benefits

Norwegian kroner, 2001 wages

160000

Base line scenario

140000 0.75 per cent less growth in pension entitlements and benefits than in wages

120000 100000

Minimum pension system from 2010

80000 60000 40000 20000 0 1960

1980

2000

2020

2040

2060

2080

Year

Assuming continuing growth in life expectancy towards 2100, the number of pensioners will become almost five times higher. Following small birth cohorts from the 1920s and 1930s, after 2010 the larger birth cohorts from 1945 and onwards reach retirement age, also explaining some part of the growth in the number of old-age pensioners from 2010 to 2040. We are thus moving from a situation with a small number of elderly due to small birth cohorts to a situation with more even (and more normal) composition of the cohorts. Simultaneously with the growth in the number of pensioners, there has also been a significant growth in average pension benefits, as shown in Figure 2. With the present Norwegian pension system (cf. Section 4), this increase may continue until 2030. The main reason is that only labour incomes obtained after 1967 are taken into account when building up entitlements for supplementary benefits. According to the National Insurance Scheme, it is necessary to work for 40 years to obtain maximum benefits — and persons born in 1940 reaching retirement age in 2007 would then be the cohort first able to reach the maximum level. However, the cohort from 1950 is the first one able to include all their potential years in the labour force (17–69 years) in the calculation of their pension entitlements. In addition, growing supply of labour among women since 1967 also contributes to increasing average benefits. As a consequence, the average replacement ratio grows until 2030. It may then stabilize at more than twice the 1967 level. An important assumption behind the further growth in the replacement ratio towards 2030 is that the basic unit in the insurance system is indexed

Future Pension Burden in Norway

89

by wage growth. During the last 20 years, the average increase has been 0.75 percentage points smaller than average growth in wages per man-year. The consequence of continuing this policy, combined with an additional assumption that minimum pensions follow wage growth, is also presented in Figure 2. In this case, average pensions may decrease from 2020 to 2060 but, as a consequence, everyone will become minimum pensioners in the long run. The figure also shows the consequences for average old-age benefits of a shift towards a minimum pension system, by abolishing new entitlements for supplementary pensions from 2010. The growth in the number of pensioners, together with more generous pension benefits, will lead to a substantial increase in the future pension burden. A convenient indicator of the pension burden is the contribution rate, defined as C ¼ P=ðL þ 0:5PÞ where P represents the yearly general pension expenditures in the National Insurance Scheme and L represents aggregate yearly labour incomes, including compensation for the labour effort among the self-employed. This contribution rate is very close to the concept used in the literature and may be considered as a rough estimate of the tax rate on total labour incomes necessary to finance the pension benefits. We have also taken into account that pension benefits are taxed at about half the tax rate on labour incomes. The development in the contribution rate since the foundation of the National Insurance System in 1967 to 2001, and further projected changes towards 2080, is shown in Figure 3. To finance the pension benefits included Figure 3:

Estimated Contribution Rates

60 50 Ageing

Per cent

40

Base line scenario

30

Growth

20 10 0 1960

1980

2000

2020 Year

2040

2060

2080

90

Dennis Fredriksen and Nils Martin Stølen

in the MOSART model, the contribution rate has increased from a level of about 7 per cent in 1967 to about 15 per cent in 2001. Based on the assumptions outlined earlier, the tax burden of financing the sum of old age, disability and survival pension benefits is estimated to increase to 25–30 per cent after 2040. This increase of about 10–15 percentage points is equivalent with the estimated increase in the expenditures in the Insurance Scheme from about 9 per cent of GDP in 2002 to almost 20 per cent in 2050, according to the report from The Pension Commission (NOU, 2004:1). Increasing life expectancy, lower retirement age and higher average pension benefits are the main reasons behind the growth in the contribution rate. A higher number of persons of working age and increased female labour supply have helped to moderate the growth in the financial burden from 1967 to 2001. A drop in fertility below the rate of reproduction has been outweighed by higher immigration in this period and, with the assumptions outlined in Table 1, this may be the case also in the future. For the decades ahead, the magnitude of cohorts of working age is reasonably stable. On the other hand, already high, and even higher, life expectancy means a trend towards an increasing share of elderly. It is of course possible that the different assumptions regarding demographic developments and supply of labour may move in a more favourable or unfavourable direction than is assumed here for the base-line scenario. In the ageing alternative further discussed in Section 3, where several components move in an unfavourable direction, the contribution rate may reach almost 50 per cent in 2060. On the other hand, if several components move in an advantageous direction simultaneously, there may only be a small growth in the contribution rate compared with today’s level.

3. Alternative Assumptions Regarding Demographic Developments and Labour Supply For a period of 50 years ahead, there is a lot of uncertainty about the main demographic components like fertility, mortality, net immigration and labour supply. In projections of the future pension burden it is important to test the sensitivity of results with regard to different assumptions. This is done by analysing the effects from shifts in the different components corresponding to what has been actually observed in the last 10–30 years. A rather detailed survey of the effects from different components on labour supply is presented in Table 2. Effects of less detailed and more simultaneous changes on the contribution rate are presented in Table 3, and for the ageing and growth alternative in Figure 3. As presented in Table 2, higher population growth as a result of a higher fertility rate and/or higher immigration may reduce the contribution rate.

Future Pension Burden in Norway

91

Table 2: Effects on the Labour Force from Alternative Assumptions About Demographic Developments and Labour Supply (‘000 Persons)a

Base-line scenario (level) Total fertility rate 1.5 Total fertility rate 2.1 No net immigration Net immigration of 25,000 persons per year Further increase in life expectancy of 3–4 years Primary school reduced by 1 year No early retirement schemes Propensity to enter disability as in 1993 Propensity to enter disability as in 1999 Retirement age 62 years Propensity to enter disability as in 1993, no early retirement schemes, formal retirement age 70 years Propensity to work over 50 as those 3 years younger Propensity to work as if 3 years younger, propensity for disability as in 1993, no early retirement, retirement age 70 Participation rates for women equal men Total labour supply for women equal men Participation rates as in 1993 a

2001

2020

2060

2,371

2,580 21 13 141 178

2,708 491 312 467 591

3

17

25

31

26 113

35 124

31

28

21 179

16 211

129

150

279

316

99

96

334

334

186

187

The effects presented in this table are compared to a base-line scenario from 1999 based on that year’s population projection.

But eventually newborns and immigrants also get old and, to obtain lasting effects, population growth has to continue. This means an increasing number of births or net immigration when population increases. In a report from the United Nations (2000), it is pointed out that accelerating immigration would be necessary to keep the old-age dependency ratio constant, and such a policy would not be sustainable in the long run. But, as stressed

92 Table 3:

Dennis Fredriksen and Nils Martin Stølen Effects on the Contribution Rate of Shifts in Different Assumptionsa

Base-line scenario (per cent) Effects (percentage point changes) Total fertility rate 1.5 No immigration Total fertility rate 1.5, no immigration Further increase in life expectancy about 2–3 years No increase in life expectancy from 2002 Women equal men in the labour market Formal retirement age 62 years Propensity to enter disability as in 1993 Formal retirement age 70 years, no early retirement schemes, and propensity to enter disability as in 1993 a

2001

2030

2060

15.2

25.2

29.8

0.8 1.4 2.5

4.6 2.0 7.8

0.8

2.0

2.1

6.1

2.7

2.4

2.1 3.8

2.2 3.8

5.4

5.9

Except for the shifts in life expectancy, the effects presented in this table are compared to a baseline scenario from 1999 based on that year’s population projection.

by Espenshade et al. (1982), higher immigration obviously may be a substitute for lower fertility to prevent a population from decreasing. If the relatively high fertility in Norway should decrease towards the level observed in other European countries, higher immigration would then be a relevant means to prevent a further increase in the old-age dependency ratio and the pension burden. From Tables 2 and 3, it is evident that higher immigration may increase labour supply and thus ease the pension burden for several decades. Higher immigration has a more immediate effect on labour supply than higher fertility, because it naturally takes a couple of decades before newborns enter the labour force. Although partially reasonable changes in immigration and fertility have some effects on the contribution rate, a contemporaneous change may induce a strengthened simultaneous effect. From Table 3, we can see that assumptions of a fertility rate of 1.5 and no immigration could increase the contribution rate by about 8 percentage points towards 2060, compared with the base-line scenario. Regarding life expectancy, the base-line scenario is based on an assumption of an increase of about 7–8 years towards 2050. We have also calculated the effects of a further increase of about 2–3 years and the effects of no

Future Pension Burden in Norway

93

further increase in life expectancy from 2001. As most people today live until they become pensioners, the assumptions about life expectancy only have a minor effect on the supply of labour, but they are of significant importance for the number of pensioners and the pension burden. From Table 3, the increase in life expectancy towards 2060 increases the pension burden by about 6 percentage points compared to 2001. Based on the growing life expectancy in Norway in the last decades, it seems strange to assume that there will be no further growth. From the assumptions regarding uncertainty for this component of about 72–3 years according to the Population Projections from Statistics Norway, the effect on the pension burden in 2060 is only 72 percentage points. In Norway, women still undertake paid work less than men. The participation rates are somewhat lower (especially for elderly women), but the main difference is caused by the fact that many women work part time. In alternative calculations we have looked at the effects of women becoming equal with men regarding participation rates, working hours and wage level. From Table 2 it is evident that the effect of equal working hours is about twice as large as the effect of equal participation rates. As an increased supply of labour among women also means larger pension entitlements, the effect on the contribution rate of women becoming equal men in the labour market is limited to a reduction of about 2–3 percentage points. The effect in the long run is smaller than in the medium run but, as the present pension system favours female working patterns, pension expenditures also increase less than labour incomes in the long run as a consequence of this shift. Because of the large amount of early retirement caused by disability pensions and early pension schemes, a change in the formal retirement age, only has minor effects on the supply of labour (cf. Table 2) and the pension burden. A partial reduction in the propensity to enter disability to the low level prevailing in 1993 has far larger effects on the supply of labour and the effect on the contribution rate is about 4 percentage points. Policies affecting the retirement age have twice the effects on the contribution rate, by increasing the supply of labour and reducing pension expenditures. In Table 2 we have also presented the effect of a postponed retirement age by 3 years for everyone. With assumptions of a simultaneous increase in the formal retirement age to 70 years, abolition of early retirement schemes and propensities to enter disability as in 1993, the contribution rate may be about 6 percentage points lower in 2060. If several of the assumptions change simultaneously, the total effect on the contribution rate may be significant. In Figure 3, we have presented the effects of an ageing alternative where the changes in some of the underlying assumptions work in an unfavourable direction regarding financing of the pension system — but where the change in each of them is of reasonable magnitude compared to shifts in the past decades. More precisely, in this

94

Dennis Fredriksen and Nils Martin Stølen

alternative we have looked at a situation with a total fertility rate of 1.5, no immigration, an extra increase in life expectancy of 2–3 years towards 2050, labour-market participation rates at the low level of 1993, propensities to enter into disability from 1999 and a further decrease in retirement age. With these assumptions, the contribution rate may reach as high as almost 50 per cent in 2060 — about three times as high as today’s level. On the other hand, it is possible that there could be more favourable developments than in the base-line scenario regarding fertility, immigration, women’s supply of labour, the propensity to enter disability and retirement age. If all of these components shift by a reasonable magnitude compared to observed shifts, but simultaneously in a favourable direction regarding the pension burden, there may only be a small increase in the contribution rate towards 2060 as illustrated in the growth alternative in Figure 3.

4. Effects of Minimum Pension System Reforms The concern for the growth in the future pension burden has caused much discussion about pension reforms in many countries and several reforms have already been implemented (cf. Lindbeck and Persson, 2003 for a survey). In Norway, a Pension Commission has recently discussed actual reforms. A preliminary report from the Commission was presented in autumn 2002, and the final report was completed in January 2004. One of the main directions for possible pension reform outlined in the preliminary report from the Pension Commission (2002) was to change the public pension system towards an equal minimum pension for everybody. One way to implement such a system is to stop the building up of new entitlements for supplementary pensions from a given year, such as 2010. A minimum pension system is reached after some decades when all present persons with existing entitlements for supplementary pensions either are dead or the current entitlements are too small to exceed the special supplement. As presented in Figure 4, this will cause a drop in the contribution rate compared with the base-line scenario of about 10 percentage points in 2060, and the contribution rate may stabilize at about 5–8 percentage points higher than the level prevailing in 2000. As shown in Figure 5, the share of minimum pensioners will increase and the reform is completely implemented by approximately 2080. With this proposal, future pensioners wishing to maintain their supplementary pensions will have to secure their pension entitlements themselves. Pre-funding in private insurance companies may then be a natural alternative. As pointed out by such researchers as Miles and Timmermann (1999), Brunner (1996) and Feldstein and Liebman (2002), a shift from a payas-you-go system in the direction of more sustainable funding affects the

Future Pension Burden in Norway

95

Figure 4: Effects on the Contribution Rate by Reducing/Removing Supplementary Pension Entitlements 40 35 30 Only pension benefits are indexed less than wages

Per cent

Base line scenario

25 20

Minimum pension system from 2010

Pension entitlements and benefits are indexed less than wages

15 10 5 0 2000

2010

2020

2030

2040

2050

2060

2070

2080

Year

Figure 5: Effects on the Share of Minimum Pensioners in the National Insurance Scheme by Reducing/Removing Supplementary Pension Entitlements 100 90

Minimum pension system from 2010

80

Per cent

70 60 50 40

Pension benefits and entitlements are indexed less than wages

30 20

Only pension benefits are indexed less than wages

10 0 1960

Base line scenario

1980

2000

2020

2040

2060

2080

Year

income distribution between generations. This is illustrated in Figure 6, by showing the effects on the present value of net pension payments for different cohorts when moving towards a public minimum pension system. Net pension wealth is calculated as the present value at age 16 of all contributions to and benefits from the pension system, when assuming a pay-as-you-go

Dennis Fredriksen and Nils Martin Stølen

96 Figure 6:

Net Pension Wealth for Different Cohorts under Different Pension Systems

Per cent of gross life cycle earnings

2.5 0.0 -2.5 -5.0 -7.5

Minimum pension system from 2010

-10.0 -12.5 -15.0

Actuarial pension system

-17.5 -20.0 1950

Base line scenario

1970

1990

2010 2030 Birth year

2050

2070

2090

system and a 2.5 per cent net interest rate (the difference between nominal interest rate and wage growth). In the base-line scenario, reflecting the present pay-as-you-go system, the increasing pension burden causes net pension wealth to fall for every subsequent cohort towards 2090. Only cohorts born before 1950 seem to have benefited from the system. But the negative figures are partly caused by the assumption of a rather high net interest rate in line with the assumptions used by the Pension Commission, giving small weights to future benefits compared with more instant contributions. As cohorts born before 1950 are more than 60 years old in 2010, they are only modestly affected by the reform. Cohorts born between 1950 and 1995 may lose. As the implementation of the reform lasts for more than 60 years, and the expenditures for supplementary pensions benefits in the present system are significantly smaller than the expenditures for minimum pensions, no birth cohorts in present value lose more than 3 percentage points of their gross life earnings. And they are not much worse off than the cohorts born after 2000 gaining from the reform. The connection between labour incomes and benefits as a pensioner has been weakened since the National Insurance Scheme was introduced in 1967. This has been caused by a strengthening of the minimum pension benefit in combination with a lower indexation than wages for benefits exceeding that level. In addition, the payment of supplementary pensions was explicitly tightened in 1992, together with reductions in the possibility of obtaining pension entitlements from high labour incomes. If the policy of lower indexation of supplementary pension benefits compared with wages continues, this also represents a movement towards a

Future Pension Burden in Norway

97

minimum pension system. The effects on the contribution rate of a policy with moderately lower indexation of the basic unit, compared to wage growth of about 0.75 percentage points a year (but keeping up the real value of minimum pensions compared to wages), are also shown in Figure 4. This policy may also be sufficient to prevent the pension burden increasing much compared with today’s level and has rather similar effects on the contribution rate as a shift to a minimum pension system from 2010. Even with a moderate growth in real wages of about 1.5 per cent each year, this still may give a real growth in pension benefits of 0.75 per cent. A lower indexation than wage growth means a decrease in future pension benefits compared with the wage level. And if minimum pension benefits are indexed according to wages, everyone will become minimum pensioners in the long run also in this alternative. However, as shown in Figure 5, the process towards a minimum pension system takes more time in this case than a direct shift towards a minimum pension system from 2010. With a minimum pension system, all individual payments to the pension system may be considered as a tax. A more modest alternative (also considered by the Commission) is to only index pension benefits after retirement by less than wages. As shown in Figure 4, such a reform may reduce the contribution rate by about 5 percentage points in 2060 compared with the baseline scenario. The share of minimum pensioners as a consequence of this reform is, of course, only modestly higher than in the base-line scenario, as this share is mainly dependent on the regulation of entitlements before retirement age.

5. Towards a More Actuarial Pension System An alternative direction for reforms, according to the main suggestions from the majority of the Commission (NOU, 2004:1), is to alter the pension system towards higher proportionality between pension entitlements and earnings. Another important element in the proposed reform is to make the yearly pension benefits dependent on retirement age and life expectancy. If a person retires early, or his life expectancy increases, the pension entitlements are distributed over a higher number of years as a pensioner, resulting in smaller pensions per year. In the literature (cf. Lindbeck and Persson, 2003) such a reform is characterized as a movement towards actuarial fairness. Increased supply of labour caused by higher actual retirement age and a reduction of the tax element in the pension premium are probably positive effects of a shift in this direction. These effects are also discussed in the literature, among others in a paper by Zhang (2003). A change from today’s pay-as-you-go system to full funding may be a possible (but not a necessary) element in such a reform.

98

Dennis Fredriksen and Nils Martin Stølen

It is not simple to give precise estimates of the two positive effects on supply of labour. As presented in several analyses (cf. Gruber and Wise, 1999, 2004 for international comparisons), the loss in total pension payments by postponing the retirement age may be important. They also find that the formal age for (early) retirement is quite decisive. In Norway, Hernæs et al. (2000) and Røed and Haugen (2002) find that the present Norwegian scheme favours early retirement, because there are not any negative consequences for future pension benefits. A tightening of these provisions is then expected to have a positive effect on the participation rates for elderly workers. To make a simple estimate on how the retirement age might be affected by a shift towards a more actuarial pension system, our point of departure has been the average of the observed participation rates among men in the age of 60–66 years from the beginning of the 1980s and the low participation rates observed in 1999 among those in the corresponding group that may enter an early retirement scheme without losing rights for pension. After further corrections (because only 60 per cent of the labour force are included in the early retirement schemes), the average retirement age is estimated to increase by 0.6 years in 2015. With an actuarial system, the average retirement age is expected to increase when life expectancy increases. Average retirement age is thus estimated to increase by 1.6 years in 2030 and 2.6 years in 2050. As shown in Table 4, a reform making the yearly pension benefit dependent on the retirement age may reduce the number of old-age pensioners (including early retirees) by almost 200,000 persons in 2050 — and the total pension expenditures by more than 25 billions Norwegian kroner (NOK) — compared with the present system. A movement towards a more actuarial pension system implies a smaller effective tax rate (Lindbeck and Persson, 2003). By keeping average benefits and tax rates unchanged, this means a positive substitution effect on labour Table 4: Year

2000 2010 2020 2030 2040 2050

Projections of the Number of Old-Age Pensioners and Pension Expenditures Number of old-age pensioners (‘000 persons)

Pension expenditures (2001, billion NOK)

Present

Actuarial

Present

Actuarial

642 678 871 1,075 1,261 1,321

679 784 941 1,077 1,127

71.8 86.6 122.6 156.8 184.7 193.2

85.8 108.9 134.2 156.8 167.6

Note: Early retirements included in pensioner numbers.

Future Pension Burden in Norway

99

Figure 7: The Connection Between Labour Incomes and Pension Entitlements in the Present and an Actuarial Pension System 250000

Actuarial pension system

200000

Pension

Present system

150000 Minimum pension system

100000

50000

0

0

100000

200000

300000

400000

500000

600000

700000

Former labour income

supply without any counteracting income effect. Because the connection between pension contributions and benefits is not evident for an average person employed under the present system, and probably will not be clearly evident in the new system, it is also difficult to obtain a precise estimate of how much the supply of labour among people that have not reached retirement age will be affected. With the assumption of a net interest rate of about 2.5 per cent, a rough estimate of the effect on supply of man-hours of an actuarial reform is about 4 per cent. In the actuarial pension system (compared with the present in Figure 7), the pension entitlements are calculated as 1.25 per cent of total labour incomes during the life cycle, up to a ceiling for annual incomes of eight times the Basic Pension Unit (BPU). A guaranty pension at the same level as the present minimum pension is introduced and financed by ordinary income taxes. But as the guaranty entitlements are means–tested against the level of income-dependent entitlements,1 the total expenditures financed directly by taxes will become rather low. These aspects of the actuarial pension system compared with the present are illustrated in Figure 7. The effects on the contribution rate of a movement towards a more actuarial pay-as-you-go system affecting retirement age and supply of labour 1

This is a simplified description compared with the actual proposal outlined in Fig. 7, where the guaranty pension is reduced by 60 per cent of the increase in income-dependent entitlement in an interval for persons with low incomes.

Dennis Fredriksen and Nils Martin Stølen

100

Figure 8: Effects on the Contribution Rate Caused by a Shift to an Actuarial System, Decomposition in Different Elements 45 40 35

Per cent

30 25 20 15 Present pension system

10

More income dependent entitlements and possibility to retire earlier Including lowerindexation of pension benefits after retirement

5

Actuarial pension system without effects on supply of labour among persons in working age Actuarial pension system

0 2000

2010

2020

2030

2040

2050

2060

2070

2080

Year

before retirement according to what is outlined earlier, is analysed with the MOSART-model and presented in Figure 8. In addition to the actuarial elements mentioned, the annual pensions after retirement are assumed to be indexed by 0.75 percentage points less than wages, to reduce the growth in future pension expenditures. The suggested reform means a drop in the contribution rate of about 5 percentage points in 2060. This corresponds to only one-third of the projected increase in the contribution rate from 15 per cent in 2000 to 30 per cent in 2060 under the present system. The positive effect on the present value of net pension wealth for different cohorts of a shift to an actuarial (but still pay-as-you-go) pension system is illustrated in Figure 6. A movement to an actuarial pay-as-you-go system may increase the net pension wealth for future generations, without having negative effects in a transition period, as the growth in expenses is reduced simultaneously with an increase in labour supply. A further shift to a funded system would obviously be even more beneficial for future generations, but more costly for the generation working in the transition period. As pointed out by Lindbeck and Persson (2003), the degree of actuarial fairness and the degree of funding may be considered as two separate dimensions. Like the shift towards a public minimum pension system, with private funding of supplementary pensions, a movement from an actuarial pay-as-you-go system to an actuarial funded system would introduce a double burden on the labour force in the transition period.

Future Pension Burden in Norway

101

Figure 8 also shows a decomposition of the effects on the contribution rate according to the most important elements of the system. Two elements in the proposed system partially increase pension expenditures. The most important is caused by the increased connection between pension entitlements and former labour incomes in the interval between 6 and 8 BPU — while labour incomes for more than 40 years also give higher entitlements. A more general possibility to retire early (at age 62) in the new system works in the same direction — and is estimated to have an immediate partial effect of 2 percentage points, because 40 per cent of the labour force is not included in the early retirement schemes and will reduce their retirement age if the formal age is reduced from 67 to 62. While the partial effect of this element decreases as time goes by, the effect of closer connection between entitlements and former labour incomes increases. In 2060, the total effect of building up entitlements and lower general retirement, partially, is estimated to increase the contribution rate by almost 4 percentage points. The partial effect from indexing pension benefits after retirement by a lesser amount than wage’s growth is represented by the difference between the upper and the second upper curve. The effect of this element increases as time passes — and is estimated to be about 2 percentage points in 2060. The actuarial element reduces yearly pensions for given total entitlements if a person retires early or life expectancy increases. This is the main factor of the reform regarding reduction of the contribution rate. As life expectancy is assumed to increase during the entire period of projection, the effect of this element increases. The pecked course for the two lower curves is caused by a discrete adjustment of the actuarial element for practical reasons as average life expectancy is assumed to increase by 1 year every 6–8 years. The total partial effect of the actuarial element in 2060 is an estimated reduction of the contribution rate of about 6 percentage points. The difference between the two lower curves represents the partial effect of increased supply of manhours among persons in working age. Although labour supply is estimated to increase as much as 4 per cent as a result of closer connection between contributions to and benefits from the pension system, the partial effect on the contribution rate of this element is only about 1 percentage point. Even with maintaining a pay-as-you-go system, a shift to a higher degree of proportionality with former labour incomes has distributional effects. Based on the main suggestion for a change of the pension system towards actuarial fairness from the Pension Commission (NOU, 2004:1), the MOSART model is used to calculate the distributional effects among pensioners when the new system is almost completely implemented (by about 2050). This is illustrated in Figure 9 for men and women respectively. The figure illustrates the distributional effects of the elements in the reform

Dennis Fredriksen and Nils Martin Stølen

102 Figure 9:

Average Old-Age Pension for Different Percentiles of Old-Age Pensioners

Men

Norwegian kroner, 2004 wages

400000 350000 300000 250000

Income dependent entitlements

200000

Present system Incl. effect on retirement age

150000 100000 50000 0 0

10

20

30

40

50

60

70

80

90

100

Percentiles

Women

Norwegian kroner, 2004 wages

400000 350000 300000 250000 Income dependent entitlements

200000 Present system

150000

Incl. effect on retirement age

100000 50000 0 0

10

20

30

40

50

60

70

80

90

100

Percentiles

causing higher proportionality between pension benefits and former labour incomes over the entire life cycle up to labour incomes eight times the BPU. As the connection between pension benefits and labour incomes in the present pension system is only modest for incomes exceeding 6 BPU, as

Future Pension Burden in Norway

103

outlined in Section 4 and illustrated in Figure 7, such a shift will obviously favour those with higher incomes. When average and minimum old-age pensions are kept at the same level as in today’s system, the benefits for persons with pensions slightly exceeding minimum pension have to decrease. Only taking the components of income-dependent entitlements into account, the two curves for men intersect at about 180,000 NOK (corresponding to about the 40th percentile of male pensioners). When also the actuarial effects regarding lower yearly pensions if you retire early are taken into account, the effects on income distribution are far more evident. This is mainly caused by a higher propensity to retire early among persons with low incomes (and vice versa) — and differences in pension entitlements are strongly strengthened when we look at annual pensions in a system with actuarial elements. Because of more modest incomes among women than among men (partly caused by a greater extent of part-time work), a shift to an actuarial system will be more unfavourable for women. Also, for women the curve for income-dependent entitlements in Figure 9 intersects with the present system at about 180,000 NOK, and this pension level corresponds to more than the 60 percentile of female pensioners. But the loss in pension entitlements compared with the present system for women with middle incomes is only modest. The difference is larger for actual pensions, once we take into account that women with higher labour incomes also tend to retire later. As is evident from this discussion, there may be a conflict between concerns for efficiency and income distribution when reforming the pension system.

6. Conclusions The ageing of the Norwegian population in the next decades together with more generous pension benefits will lead to a substantial increase in the future pension burden if the present pension system is maintained. Norway is in an advantageous position compared to most other Western countries as fertility is relatively high, and lower fertility than the reproduction level is compensated by net immigration. But for political and practical reasons, there are probably limitations for the level of immigration to reduce the future pension burden. Increased supply of labour, especially among women, is beneficial, but as this also means higher pension entitlements in the long run, the effect does not seem to be large. The most effective remedies regarding demographics, or supply of labour, seem to be of the kind contributing to later retirement, thus affecting both pension expenditures and the number of persons in the labour force. But for political reasons it may be quite difficult to reverse the reduction in retirement age that has taken place during the last decades. If several components regarding

104

Dennis Fredriksen and Nils Martin Stølen

demographics and labour supply shifted in the same direction related to the pension burden, it could have a significant joint effect. However, a substantial increase in the pension burden with the present pension system seems to be unavoidable, and the scope for political remedies to affect demographic development and labour supply may be narrow. Reforms of the pension system thus seem to be necessary to reduce the future financial burden. But there has to be a substantial tightening of the system to prevent the taxpayer burden from increasing. Even with a rather comprehensive shift to a public-financed minimum pension system, the expenses for old age, disability and survivor pensions may increase to 20 per cent of the total labour incomes in 2060 versus 15 per cent today. With a minimum pension system each person has to take responsibility for his or hers supplementary pensions, and today’s employees are hurt by a double burden in the transition period. The recent Norwegian Pension Commission has proposed a more moderate tightening, also stimulating labour supply among those at work and exposing retirement age. This may be obtained by a shift towards a more actuarial system, increasing the connection between pension entitlements and former labour incomes and reducing the yearly pensions if a person retires early. In the proposal the pension system is also tightened by decreasing the yearly pensions in accordance with further growth in life expectancy and by indexing pension benefits after retirement less than the growth in wages. The analysis shows that the actuarial element — reducing yearly pension benefits as a consequence of early retirement and higher life expectancy — is the most important one in reducing the future pension burden. Indexation of pension benefits by less than wages also has a tightening effect. Making a closer connection between pension entitlements and former labour incomes may have positive effects on labour supply. The importance of this factor is, however, uncertain, and the effect on the contribution rate seems to be rather small. As elements of redistribution of incomes are included in the present pension system, a closer connection to former labour incomes favours persons with higher incomes. There may thus be a conflict between the concern for income distribution and a movement towards a more actuarial system, which makes a closer connection between pension entitlements and former labour incomes to stimulate the supply of labour.

Acknowledgement The paper is based on analyses for the Ministry of Finance and the Norwegian Pension Commission. Financial support is gratefully acknowledged. Thanks for valuable comments to earlier drafts from Erling Holmøy,

Future Pension Burden in Norway

105

participants at The International Microsimulation Conference on Population Ageing and Health in Canberra, 2003 and The Norwegian Economists’ Research Meeting, 2004.

References Beetsma, R., Bettendorf, L. and Broer, P. (2003). The Budgeting and Economic Consequences of Ageing in the Netherlands. Economic Modelling, 20, 987–1013. Brunner, J.K. (1996). Transition from a Pay-as-you-go to a Fully Funded Pension System. Journal of Public Economics, 60, 131–146. Diamond, P. (2004). Social Security. The American Economic Review, 94(1), 1–24. Disney, R. (2000). Crises in Public Pension Programmes in OECD: What are the Reform Options? The Economic Journal, 110, 1–23. Espenshade, T.J., Bouvier, F.L. and Arthur, W.B. (1982). Immigration and the Stable Population Model. Demography, 1(19), 125–133. European Commission (2001). Reforms of Pension Systems in the EU — An Analysis of the Policy Options. European Economics, 73, 171–222. Feldstein, M. (1996). The Missing Piece in Policy Analysis: Social Security Reform. American Economic Review Papers and Proceedings, 86, 1–14. Feldstein, M. (ed), (1998). Privatizing Social Security. University of Chicago Press, Chicago. Feldstein, M. and Liebman, J. (2002). The Distributional Effects of an Investment based Social Security System, in Feldstein, M. and Liebman, J. (eds), The Distributional Aspects of Social Security and Social Security Reform, University of Chicago Press, Chicago. Fredriksen, D. (1998). Projections of Population, Education, Labour Supply and Public Pension Benefits — Analyses with the Dynamic Microsimulation Model MOSART. Social and Economic Studies 101, Statistics Norway. Fredriksen, D., Heide, K.M., Holmøy, E. and Solli, I.F. (2005). Macroeconomic Effects of Proposed Pension Reforms in Norway. Discussion Papers No. 417, Statistics Norway. Gruber, J. and Wise, D. (eds). (1999). Social Security and Retirement Around the World. University of Chicago Press, Chicago. Gruber, J. and Wise, D. (2002). Different Approaches to Pension Reform from an Economic Point of View, in Feldstein, M. and Siebert, H. (eds), Social Security Pension Reform in Europe, University of Chicago Press, Chicago. Gruber, J. and Wise, D. (2004). Social Security and Retirement Around the World, Micro-Estimation. University of Chicago Press, Chicago. Hernæs, E., Sollie, M. and Strøm, S. (2000). Early Retirements and Economic Incentives. Scandinavian Journal of Economics, 102(3), 481–502. Kotlikoff, L. and Leibfritz, W. (1999). An International Comparison of Generational Accounts, in Auerbach, A., Kotlikoff, L. and Leibfritz, W. (eds), Generational Accounting Around the World, Chicago University Press, Chicago. Kotlikoff, L., Smetters, K. and Walliser, J. (2001). Finding a Way Out of America’s Demographic Dilemma. NBER Working Papers 8258. Lindbeck, A. and Persson, M. (2003). The Gains from Pension Reform. Journal of Economic Literature, 41, 74–112.

106

Dennis Fredriksen and Nils Martin Stølen

Miles, D. and Timmermann, A. (1999). Risk Sharing and Transition Costs in the Reform of Pension Systems in Europe. Economic Policy, 29, 253–286. Norwegian Pension Commission (2002). Ma˚l, prinsipper og veivalg for pensjonssystemet (Aims, Principles and Challenges for the Pension System in Norway). Preliminary Report, September 4, 2002, Oslo. NOU (1998:10). Fondering av Folketrygden (Funding of the Norwegian National Insurance Scheme). The Ministry of Finance, Oslo. NOU (1998:19). Fleksibel pensjonsalder (Flexible Pension Age). The Ministry of Finance, Oslo. NOU (2004:1). Modernisert folketrygd. Bærekraftig pensjon for framtida. (A Modernised National Insurance Scheme. Sustainable Pension Expenditures in the Future). The Ministry of Finance and The Ministry of Social Affairs, Oslo. OECD (1996). Ageing in OECD Countries: A Critical Policy Challenge. Social Policy Studies No. 20. OECD, Paris. OECD (1998). Maintaining Prosperity in an Ageing Society. OECD, Paris. Orcutt, G.H., Merz, J. and Quinke, H. (eds). (1986). Microanalytic Simulation Models to Support Social and Financial Policy. North-Holland, New York. Røed, K. and Haugen, F. (2002). Early Retirement and Economic Incentives — Evidence from a Quasi-Natural Experiment. Mimeo, The Frisch Centre, Oslo. St.meld. nr. 1. (2003–2004). Nasjonalbudsjettet (The National Budget) 2004. Oslo: The Ministry of Finance, Oslo. St.meld. nr. 30 (2000–2001). Langtidsprogrammet (The Long Term Programme) 2002– 2005. The Ministry of Finance, Oslo. United Nations (2000). Replacement Migration: It is a Solution to Declining and Ageing Population, Population Division. Department of Economic and Social Affairs, New York. Velferdskommissionen (2004). Analyserapport — Fremtidens velfærd kommer ikke af sig selv. Electronic version, www.velfaerd.dk, Copenhagen. Zhang, J. (2003). Comparing Social Security Programs with Leisure and Bequests. Economic Letters, 78, 59–66.

Chapter 5

Macroeconomic Effects of Proposed Pension Reforms in Norway$ Dennis Fredriksen, Kim Massey Heide, Erling Holmøy and Ingeborg Foldøy Solli Statistics Norway, Research Department, P.O. Box 8131 Dep., N-0033 Oslo, Norway

Abstract We combine a detailed dynamic microsimulation model of government pension expenditures and a large Computable General Equilibrium (CGE) model to estimate the extent to which two suggested reforms of the Norwegian public pension system improve fiscal sustainability and stimulate employment. The first reform makes the current public pension system more actuarial, whereas the second reform privatises the supplementary (earnings-related) pensions. While maintaining the present system implies an increase in the payroll tax rate from 13 per cent today to 25 per cent in 2050, both reforms make it possible to maintain tax rates below today’s level in all years up to 2050. Most of the fiscal improvement can be attributed to higher employment.

1. Introduction Owing to increased longevity and low fertility rates after 1970, Norway will experience a significant ageing of its population throughout this century. According to population forecasts,1 the ratio of those of working age 20–66 $

The paper has benefited from comments on an earlier version from Asbjørn Rødseth and participants at the Workshop on Fiscal Policy Issues, University of Oslo 28–29 January 2005, Arne Magnus Christensen and Svein Sæterdal Corresponding author. Address: Statistics Norway, Research Department, P.O. Box 8131

Dep., N-0033 Oslo. Phone: +47 21 09 45 80 E-mail address: [email protected] (E. Holmøy). 1

See Statistics Norway (2002).

International Symposia in Economic Theory and Econometrics, Vol. 15 r 2007 Published by Elsevier B.V. ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15005-X

108

Dennis Fredriksen et al.

to those 67 and older will decrease from 4.5 in 2002 to 2.5 in 2050. Although ageing in Norway is expected to be less pronounced than in most other OECD countries, Antolin and Suyker (2001) conclude that the existing welfare state schemes imply that Norway will experience one of the sharpest increases in public expenditures as a share of GDP after 2010. Three forces stand out as being the most important in driving this development. First, the public pension system is still maturing, in the sense that the number of pensioners entitled to supplementary (earnings-related) pensions is still increasing. Second, since there are no actuarial mechanisms in the public pension system, retirees receive their defined annual benefits over more years as they live longer. Third, the nominal value of public pension benefits is indexed to wage growth rather than to some average of wage and price growth.2 The strength of the determinants of government expenditures is a result of policy—especially of the design of the public pension system and other welfare schemes. Accordingly, another fundamental reason for the expected rapid growth in government expenditures is that successive governments have not yet undertaken significant cost-saving reforms of the relatively generous welfare-state schemes. One reason for the lack of policy action may be that the apparently impressive current fiscal situation has not yet forced governments to do so. Internationally, large petroleum revenues make the Norwegian Government an outlier with respect to financial wealth. The value of the Central Government Petroleum Fund (CGPF) amounted to 73.3 per cent of GDP by the end of 2005. Measured as a share of Mainland GDP, it is expected to grow until about 2020.3 On the other hand, most other OECD countries have for several years struggled to limit publicbudget deficits. Even though it is a decade before the baby-boom cohorts become pensioners, several EU countries already have problems with meeting the budget constraints defined by the EU Growth and Stability Pact. The current strong financial position of the Norwegian government gives a very misleading picture of the long-run situation. Long-run projections undertaken by the Pension Commission (NOU, 2004:1), the Ministry of Finance (2001, 2004a, 2004b, 2004c), Aaberge et al. (2004) and Fredriksen et al. (2005) show that Norway faces a serious problem of fiscal sustainability as ageing boosts government expenditures after 2020. Because a

2

Wage indexation is the political intention, and this assumption underlies all Norwegian projections of government pension expenditures. Effectively, however, the historical indexation has been somewhat less generous. 3 The ratio between the return of the fund and trend–GDP for the Mainland economy was 4.5 per cent in 2005. This ratio is expected to peak at about 6.5 per cent around 2030.

Macroeconomic Effects of Proposed Pension Reforms

109

substantial part of the increase can be attributed to the growth in government pension expenditures, pension reforms have been high on the policy agenda, as in other countries. A Pension Commission delivered reform proposals in January 2004. The main proposal from the commission is a More Actuarial Public Pension System (MAS).4 The relationship between income and pensions is strengthened by relating benefits to earnings in all years—and the annual benefit is reduced if the number of years as a pensioner increases because of increased life expectancy or early retirement. The other reform proposal is a Flat Benefit Public Pension System (FBS). Here the supplementary pensions are privatised, and the remaining public system will include only a flat uniform basic benefit with no means testing (equal to the minimum benefit in the current system). Comprehensive quantitative assessments of the effects of the proposed pension reforms have so far been missing. The purpose of this chapter is to provide estimates of the following: 



The need for a pension reform. We do this by projecting how a continuation of the present pension system (and other welfare-state schemes) will affect labour income taxation, represented by the payroll tax rate, given that the government budget deficit follows the current fiscal policy rule over the next 50 years. The long-run macroeconomic effects of two pension reforms proposed by the Pension Commission (NOU, 2004:1). We focus on the scope for tax cuts made possible by the reforms. In particular, we examine to what extent tax rates can be reduced as a result of the expansion of tax bases generated by increased labour supply, rather than reduced average benefits. Stimulating labour supply has been one of the primary purposes of the pension reforms.

To this end, we combine a detailed dynamic microsimulation model, MOSART, with a large-scale dynamic Computable General Equilibrium (CGE) model, MSG6. The MOSART model provides a detailed description of the demographic dynamics, including the development of the labour force and the number of various kinds of pensioners (see Chapter 4 of this volume for more details). Being a microsimulation model, it also provides a complete representation of the relevant heterogeneity of the population and an exact description of the Norwegian social security system. MOSART provides an accurate calculation of individual pension benefits and government 4

This main proposal constituted the basis for the government reform proposal (Ministry of Finance, 2004c), which was discussed and partly adopted by the Parliament during May 2005.

110

Dennis Fredriksen et al.

pension expenditures for given individual work histories. Consequently, it provides precise estimates of what Coile and Gruber (2003) refer to as ‘‘mechanical’’ effects on these variables of pension reforms (i.e., effects for given behaviour and given wage rates and prices). In the following, we will include these effects in what we refer to as ‘‘direct’’ effects (i.e., effects calculated outside the CGE model). The MSG6 model accounts for the equilibrium adjustments to the changes in government expenditures, labour supply incentives and private savings induced by the pension reforms. As the model is rather disaggregated, it captures the equilibrium adjustments of all tax bases and the prices of government consumption. It also provides a relatively rich description of the production structure, including decreasing returns to scale of industry production functions. This property implies a complex determination of the wage rate—and wage adjustments have important feedback effects on the government budget, especially when government pensions are indexed to the wage rate. Quantitative assessments of the macroeconomic consequences of ageing abound in the literature. Chauveau and Loufir (1995), OECD (1998, 2000, 2001), the European Commission (2001) and Visco (2002) provide relevant international comparisons. The literature on numerical simulations of pension reforms has also become large (see Miles, 1999; Kotlikoff et al., 2001; McMorrow and Roeger, 2002; Beetsma et al., 2003; Lindbeck and Persson, 2003; and Bovenberg and Knaap, 2005;). Thøgersen (2001) and Fehr et al. (2003) estimate the effects on macroeconomic aggregates and welfare of a reform of the Norwegian public pension system. All the referred studies utilise CGE models with a rather small number of agents representing overlapping generations. Even a specification of 12 lifetime earning classes in each cohort, as in the model used in Kotlikoff et al. (2001), loses many potentially important aspects of heterogeneity among agents and details of the pension system that are incorporated in a dynamic microsimulation model such as MOSART. Moreover, the MSG6 model provides a rather detailed description of commodity markets, thereby providing a more detailed determination of relative prices and the items in the government budget than what is the case in most OLG equilibrium models. However, accounting for details implies some costs in terms of loss of complete consistency. In our analysis, most, but not all, of the general equilibrium effects computed by the MSG6 model are captured by the MOSART simulations. Our credo is that the shortcomings caused by lack of complete consistency are empirically less important than the details we have been able to account for. By including endogenous retirement behaviour in a dynamic microsimulation model, Coile and Gruber (2003) share some of our ambitions with respect to estimating the fiscal effects of social security reforms in the US. They find that retirement responses have minor effects on the balance of the

Macroeconomic Effects of Proposed Pension Reforms

111

social security system because this system is close to actuarial. However, when other taxes are factored in, delaying retirement raises net government revenue. There are two reasons why we would expect that a pension reform stimulating labour supply at both the intensive and the extensive margin is likely to have a much stronger positive fiscal effect in Norway than in the US. First, since the present public pension system in Norway does not include any actuarial mechanisms linked to life expectancy, delayed retirement has first order budget effects. Second, the effective taxation of labour income is higher in Norway than in the US. This chapter is organised as follows: Section 2 describes briefly the microsimulation model and the CGE model. In Section 3, we present a reference scenario in which the existing pension system is maintained, and we quantify the non-sustainability of the present fiscal policy. Section 4 presents the macroeconomic effects of what we refer to as the MAS, which is the main proposal from the Pension Commission. Section 5 presents the similar effects of another reform proposal, which we refer to as the FBS, since supplementary public pension benefits are phased out in this reform. Section 6 concludes.

2. Modelling Framework To be relevant for estimating the effects of fully specified pension reforms, the model framework should meet some fundamental requirements. First, to be operational the model must include a detailed description of the rules constituting the pension system. Second, a detailed description of the population heterogeneity with respect to age and income is necessary for accurate calculations of individual and aggregate pension entitlements and benefits. Third, a detailed description of all tax bases, as well as of their determinants, is required for a full assessment of likely developments in public finances. The labour supply responses are particularly important in this respect. Fourth, the model should take into account that changes in relative prices affect the prices of government consumption and transfers indexed to wages. Fifth, analyses of fiscal sustainability require a long-run perspective, that captures both the long-run reform effects as well as the capacity effects of investments and productivity growth. 2.1 The Dynamic Microsimulation Model The dynamic microsimulation model, MOSART, simulates the life courses of a representative cross-section of the Norwegian population. Fredriksen (1998) provides a detailed documentation of MOSART and examples of applications. The model captures the following events: migration, deaths,

112

Dennis Fredriksen et al.

births, marriages, divorces, educational activities, retirement and labourforce participation. Transitions between states over the life course depend on individual characteristics, and the transition probabilities have been estimated from observations in a recent period. MOSART is especially designed to analyse the direct effects on individual pension entitlements and government pension expenditures of changes in the pension system. By direct effects we mean effects ignoring behavioural responses and general equilibrium effects. The model includes an accurate description of the pension rules, it captures all relevant details of the population dynamics, as well as the heterogeneity of the pension entitlements accruing to individuals. Labour-market earnings and participation rates depend on individual characteristics, as well as earnings in earlier years. 2.2 The CGE Model The CGE model, MSG6, provides a rather detailed description of the Norwegian economy based on national accounts data. Heide, Holmøy, Lerskau and Solli (2004) provide a detailed description of the model structure and its empirical characteristics. The Norwegian economy is assumed to be too small to affect world prices. The exchange rate is normalised to unity. All agents have access to international capital markets where they face an exogenous interest rate. Goods and factors are perfectly mobile between industries. Supply equals demand in all markets in all periods. In each period, consumers allocate an exogenous time endowment to leisure and labour according to standard consumer theory. The parameters are calibrated so that the uncompensated wage elasticity equals 0.1, consistent with the econometric results in Aaberge et al. (1995). The composition of private consumption is determined in a demand system derived from a separable structure of nested origo-adjusted CES subutility functions. Most imported products are close but imperfect substitutes for the corresponding domestic products. Firms are run by managers with perfect foresight, who maximise present net-of-tax cash flow to owners. Most producers of tradeables allocate their output between the domestic and the foreign market. It is costly to redirect output between these two markets. Whereas world prices of exports are exogenous, firms engage in monopolistic competition in most domestic markets. Industry production functions exhibit decreasing returns to scale. The economy as a whole faces an intertemporal budget constraint, as foreign trade is balanced in present value terms, corrected for initial net foreign wealth. This constraint is met by endogenous determination of a constant growth rate of the wage cost per hour. In our scenarios this growth rate is approximately equal to the equilibrium wage growth rate in the Scandinavian model of inflation, which equals the sum of the growth rates

Macroeconomic Effects of Proposed Pension Reforms

113

of, respectively, world prices and labour productivity in the tradeables sectors.5 Fredriksen et al. (2006) provides a detailed explanation of the equilibrium mechanisms determining the wage rate in MSG6. The model includes comprehensive and detailed accounts of government revenues and expenditures. In real terms all expenditures are exogenous in MSG6, but the projections of these exogenous variables have utilised some specialised models developed at Statistics Norway.6 The projection of government pension expenditures results from a combined use of MOSART and MSG6. All tax bases are endogenous. In particular, the detailed classification of industries, commodities and various types of indirect taxes improves the accuracy of the computations of revenues from indirect taxation. The government budget constraint follows from the fiscal policy rule adopted in 2001. A strict interpretation of this rule is that the non-petroleum deficit should be equal to the real rate of return on the net financial assets, B—that is, ði  pÞBt1 ¼ Dt ; where D is the non-petroleum fiscal deficit, p is international inflation. i– p is the expected real rate of return, denominated in international prices. The rule implies that the fiscal surplus equals Bt  Bt1 ¼ pBt1 þ Pt ; where P is government net cash flow from the petroleum sector. In the relevant long-run perspective, the time path of P is basically an exogenous policy decision. Thus, the time path of B is determined purely by variables that are exogenous for a small open economy such as the Norwegian one. This implies an exogenous time path for the fiscal surplus contingent on the time path of P. In all our scenarios this constraint is satisfied by endogenous pay-as-you-go adjustments of the payroll tax rate, which serves as a representative of a broad tax on labour income.

3. What Happens if No Pension Reform? 3.1 The Existing Public Pension System The National Insurance Scheme (NIS) in Norway was established in 1967, and replaced a general public pension system consisting of a flat pension 5 See Holmøy and Heide (2005) for a discussion of the validity of the Scandinavian model of inflation as a norm for equilibrium wage growth in the Norwegian economy. 6 The projections of government consumption within the sectors of health care and education have utilised a model which decomposes changes in the input of labour and intermediate inputs into (a) changes in the number of persons in different age groups; (b) changes in the service standards; and (c) changes in coverage ratios. Thus, the projections capture the fact that ageing, cet. par., increases public health care expenses.

114

Dennis Fredriksen et al.

benefit. The NIS benefit includes three elements: a basic benefit, a special supplement and a supplementary benefit. The basic benefit and the special supplement constitute the granted minimum benefit. The special supplement is means–tested against the supplementary pension: Pension benefit ¼ basic benefit þ max ðspecial supplement; supplementary benefitÞ The supplementary pension is based on labour market earnings after 1967, and only persons born in 1950 and later will receive supplementary benefits based on their entire working career. Because each new cohort of pensioners will have a larger percentage of their working career included in the computation of their supplementary pension, the average benefit has grown and will continue to grow relative to the wage level until 2030. The growth in the minimum benefit and in female labour force participation also contributes to the growth in average benefit. The income basis for the supplementary benefit is the average labour market earnings over the 20 years with the highest earnings. Full pension is reached after 40 years of labour force participation. Using MOSART to account for all elements in the public pension system for a representative sample of the Norwegian population, we find that increasing labourmarket earnings by 1 NOK raises the average present value of future pension benefits by 0.11 NOK. There is a large variation in the individual increments in benefits. Moreover, the complexity of the rules makes it hard for individuals to compute the impact on pension benefits of increasing their earnings. Given the political intention of wage indexation of both pension entitlements and individual benefits, the NIS benefits imply a pre-tax replacement ratio equal to about 50 per cent for a person with 40 years of labour-market earnings and a steady and normal income level. Special tax rules for pensioners raise the average after-tax replacement ratio of NIS benefits to about 65 per cent. Private pension schemes and special pension schemes for public employees may increase the compensation level further. The formal retirement age in the NIS is 67 years. Both disability pensioners and early retirees obtain entitlements as if they were working until the age of 67. Roughly 40–50 per cent of the population is receiving disability pension when reaching retirement age, and about 60 per cent of the (still) employed are entitled to early retirement from the age of 62. Disability pension and early retirement imply that the present effective retirement age averages 59–60 years in Norway. Note that early retirement through these arrangements does not reduce future pension benefits at any point in time, neither because of a shorter period of labour-market earnings nor through a longer period as pensioner.

Macroeconomic Effects of Proposed Pension Reforms

115

3.2 Key Exogenous Assumptions in All Projections7 













7

Population ageing: We rely on the middle alternative in the population projections presented in Statistics Norway (2002). The ratio of those of working age 20–66 to those 67 and older decreases from 4.5 in 2002 to 2.5 in 2050.8 The labour force: The population aged 20–66, increases by 13.6 per cent, from 2.8 million in 2002 to 3.2 million in 2050. Public pension expenditures: Population ageing more than doubles the number of old-age pensioners from 2002 to 2050. This projection presumes that the age and gender-specific transition rates from work to disability and early retirement observed in 2001, stay constant. The government finances about 40 per cent of the early retirement benefits. We assume that pension entitlements are indexed to wage growth, which is the political intention. Government consumption: We have made the rather cautious assumption that no changes take place in standards and coverage ratios of public services beyond already approved reforms. A plausible interpretation of our scenario is that the growth in private consumption per capita involves privatisation of services traditionally provided by the government sector in Norway, including care for the elderly. Productivity growth: Based on historical trends Total Factor Productivity (TFP) grows by 1.3 per cent annually.9 World prices: Except prices of crude oil and natural gas, measured in NOK, world prices grow by 1.5 per cent annually. The nominal interest rate: Nominal interest rate is assumed to stay constant over the simulation period at 5.5 per cent, which implies a 4.0 per cent real interest rate in terms of foreign goods. This is in line with the assumption in the current fiscal guidelines, and with American interest rates in the second half of the 1990s. In their projections for the EU, McMorrow and

Fredriksen et al. (2006) and Fredriksen et al. (2005) provide some more details on the exogenous assumptions. Further information is available from the authors. 8 An important driving force behind the expected ageing is the increase in life expectancy. In the middle alternative in the projections presented in Statistics Norway (2002), the life expectancy for males increases from 77.0 years in 2003 to 84.2 years in 2050. The corresponding increase for females is from 81.9 to 88.1 years. 9 Private business industries are characterised by decreasing returns to scale in MSG6. Taking this into account, TFP grows by approximately 1 per cent when computed by the standard procedure assuming constant returns. Labour productivity in government sectors is assumed to grow by 0.5 per cent per year (the standard assumption in the Norwegian National Accounts).

116

Dennis Fredriksen et al.

Table 1: Macroeconomic Development in the Reference Scenario (Average Annual Growth Rates in Per Cent)

Private consumption Government consumption GDP Mainland industries Wage cost per hour relative to world prices Payroll tax rate Consumer real wage rate Employment Government sector Private business sector Government financial wealth relative to GDP Net national financial wealth relative to GDP



2002–2025

2026–2050

3.4 0.8 1.9 2.5 2.8 2.5 2.8 0.3 0.5 0.3 3.2 5.0

2.2 1.3 1.6 1.7 2.8 5.4 1.6 0.1 0.9 0.3 1.0 0.4

Roeger (2002) assume that the nominal interest rate falls from 5.5 per cent to 5.25 per cent from 2000 to 2050. Petroleum revenues: In 2004, the export share of petroleum products was 45.8 per cent, and taxes and other revenues from the petroleum sector amounted to 27.1 per cent of total central government income. We have adopted the projections reported in Ministry of Finance (2001). Export of crude oil declines at an annual rate of 4.4 per cent to 2010 in value terms. Thereafter, the percentage of annual decline will be approximately 5.4 per cent. Export of natural gas is projected to increase by an annual rate of 6.8 per cent to 2010 and thereafter to stabilise.

3.3 Implications of Maintaining the Existing Pension System On average, private consumption per capita can grow at about 2.8 per cent, implying a doubling after 25 years, without violating the long-run constraint on foreign debt (see Table 1). The annual GDP growth averages 1.7 until 2050. Most of the difference between the growth in private consumption and GDP reflects our assumption of zero–growth in the quality of government services.10 It is likely that private consumption in such a scenario will include 10

Furthermore, the slowdown of the growth in the petroleum sector contributes to reduce GDP growth, as reflected in the figures for Mainland industries versus overall GDP growth. On the other hand, the savings of the petroleum revenues implied by the fiscal policy rule implies that the time path of consumption does not fluctuate with changes in petroleum revenues.

Macroeconomic Effects of Proposed Pension Reforms

117

an increasing share of services that traditionally have been produced by government sectors. The estimated growth prospects demonstrate that, in the long run, productivity growth is by far the most important source of economic well-being, and that ageing has a much more moderate role in this respect. However, one may question if the no-pension-reform scenario is politically feasible. The reason is that, despite the substantial petroleum wealth and the assumption of constant quality in government services, the present tax rates are not high enough to ensure fiscal sustainability. On the contrary, such a broad tax on labour income as the payroll tax rate must be raised on a pay-as-you-go basis from the present level of 13 per cent to about 25 per cent in 2050—and it follows an increasing trend if the horizon is extended beyond 2050. Growth in public pension expenditures is the main source of the necessary rise in the payroll tax rate. Measured in per cent of GDP, these expenditures grow from 5.3 in 2002 to 15.9 per cent in 2050. Maturing of the existing pension system, as well as increased female labour market earnings, imply a 30 per cent increase in the average public old-age benefit from 2002 to 2050.11 In addition, ageing after 2020 brings about a stronger growth in nominal government consumption than in the tax bases. Ageing alone implies an annual growth in government employment of 0.6 per cent from 2002 to 2020, about 1.0 per cent in 2021–2040 and 0.3 per cent thereafter. Prior to 2020, there is, however, room for substantial reductions in the payroll tax rate without breaking the fiscal policy rule. The necessary increase in the payroll tax rate after 2020 adds to an effective tax on marginal labour income that is already rather high.12 If the continuous increase in the payroll tax rate is politically acceptable, the resulting distortion of labour supply incentives is likely to cause a significant loss in social efficiency of the

11

The scheme for occupational pensions in the government sector guarantees that the sum of all old-age benefits to government-sector employees equals two-thirds of previous earnings. This implies that a reduction in the public pension benefit is exactly compensated by an increase in the occupational benefit. We have assumed that the pension reform does not affect this scheme, but any increase in the occupational benefits is financed by higher premiums. Thus, continuation of this scheme does not imply any additional need for raising taxes. 12 In addition to the payroll tax rate, its most important elements include an average marginal tax on personal labour income approximately equal to 40 per cent, compulsory social security premiums averaging 7 per cent of wages, and net indirect taxation of consumption (including VAT) averaging 19 per cent. In addition, the pension system, especially the early retirement scheme, magnifies the labour supply distortions at the extensive margin.

118

Dennis Fredriksen et al.

allocation of time. Moreover, an increase in rent-seeking activities is likely. Higher international mobility of tax bases exacerbates both these problems. Our estimated continuous increase in the payroll tax rate after 2020 is much stronger than corresponding estimates for other countries. Projections presented in OECD (2001) show that budgetary pressures from ageing populations on average require a 7 per cent increase in the ratio between taxes and GDP. This exceeds the corresponding estimates in Chauveau and Loufir (1995) for the seven major economies. On the other hand, McMorrow and Roeger (2002) find that the ratio between social security contributions and wages in the EU must increase from 16.1 per cent in 2000 to 26.9 per cent in 2050, and this is uniquely because of the rise in the old-age dependency ratio. McMorrow and Roeger explain why their estimate of increase in public pension expenditures as a share of GDP from 2000 to 2050 is about 4 percentage points higher than the corresponding estimate made in the European Commission (2001).13 According to Feldstein (2005), the actuaries of the U.S. Social Security Administration estimate that the payroll tax rate must increase by 48 per cent from today to 2075 to finance the benefits specified in current law (i.e., about half of the percentage increase in the payroll tax rate in our reference scenario). However, this projection neglects the development in other government expenditures as well as general equilibrium effects. Taking these effects into account, Feldstein assesses that the necessary increase in the tax rate must become about 70 per cent. The model-based estimate in Kotlikoff et al. (2001) is somewhat higher; they find that the payroll tax rate must increase by 77 per cent over the next three decades. With respect to welfare state schemes Denmark is more similar to Norway than the U.S. The Danish Welfare Commission (2004)14 projects that government expenditures as a percentage of GDP will increase from 52 per cent in 2001 to 59 per cent in 2050 if the present welfare-state schemes are maintained. Over the same period government revenues as a per cent of GDP will increase from 54 to 55 13

It is somewhat unclear how the projections of McMorrow and Roeger (2002) should be interpreted. The referred estimates are taken from their Table 3 showing deviations between a scenario based on realistic population ageing and a ‘‘technical’’ scenario assuming no ageing. This suggests that the referred figures represent effects of a partial shift in demographic development, not projections as such. For example, the reported 19.0 per cent decrease in GDP per capita by 2050 means that ageing, cet. par, contributes to reduce GDP per capita by 19.0 per cent in 2050 compared to the technical scenario. If the changes in the ratio between pension expenditures and GDP are measured in the same way, i.e., as shift effects, it means that the role of growth over time, because of, for example, productivity growth and capital accumulation, on both pension expenditures (through indexation) and GDP is ruled out. 14 See Andersen et al. (2004) for a review in English.

Macroeconomic Effects of Proposed Pension Reforms

119

per cent. One way of obtaining fiscal sustainability is to increase the base income tax rate permanently from 2011 by 8.7 percentage points, corresponding to an increase in the tax revenue–GDP ratio of 5.1 per cent compared to a scenario based on constant tax rates.15 The reference scenario also demonstrates that productivity growth in the private sector will not contribute to reduce in the fiscal sustainability problems. Thus, policy makers cannot rely on the misconception that economic growth will finance the increase in government expenditure. On the contrary, in our reference scenario, economic growth makes it somewhat harder to finance the Norwegian welfare state. Although productivity growth in the private sector raises most tax bases, government expenditures will increase even more. This is not a general finding, but based on qualifications of which at least two are particularly relevant for Norway. First, government pension benefits are indexed to wages rather than prices (as in e.g. the U.S.). Second, wage-dependent government expenditures exceed wage-dependent government revenues, basically as a result of the fiscal policy rule.16

4. Effects of a More Actuarial Public Pension System (MAS) 4.1 Main Reform Characteristics The MAS is supposed to be gradually phased in over a 15-year period from 2010. It continues to be a pay-as-you-go financed system. We assume that the reform does not affect the fiscal policy rule, which implies that the pension reform does not change government savings.17 The payroll tax rate 15

The estimate presented by The Danish Welfare Commission (2004) of the permanent increase in the base income tax that is necessary to obtain fiscal sustainability, is radically higher than the estimate in Jensen et al. (2001). The latter study concludes: ‘‘the fiscal policy in Denmark is almost sustainable, in the sense that a smooth tax rate, which fulfils the intertemporal budget constraint of the public sector is only 1.1 percentage points higher than the announced base tax rate for 2003’’. 16 The premise for the fiscal effects of productivity growth in the private sector is that real wage growth is basically driven by productivity growth. Furthermore, we assume that productivity growth does not have significant effects on labour supply, an assumption justified by Aaberge et al. (2004). 17 The motivation of the fiscal policy rule is to ensure a fair inter-generational distribution of the petroleum wealth and to ensure that the use of the petroleum wealth is gradually increased. On the other hand, the main intention of the pension system is to help individuals to achieve a rational allocation of consumption possibilities over their life span. In this perspective, there is no reason why a pension reform should change the general long- and short-run considerations underlying the fiscal policy design.

120

Dennis Fredriksen et al.

adjusts annually to meet the same time path of the fiscal surplus as in the reference scenario. Although the reform strengthens the incentives to retire as a disability pensioner, the disability pension scheme is not altered. Moreover, we assume that the reform does not change the rates of transition from work to disability. The most important reform characteristics include: 







The pension benefit continues to include two elements, a granted minimum benefit and an income-based benefit. The minimum benefit is maintained at the same level as the current minimum benefit. Contrary to the basic benefit in the present system, it is means–tested against the income-based pension benefit. The system implies a stronger dependency between earnings and pension benefits. The income-based benefit is basically 1.25 per cent of lifetime labour-market earnings with a few restrictions. The current early retirement arrangements are phased out. They are replaced with a flexible retirement age from the age of 62 years, available to everyone. However, the system becomes more actuarial as the pension benefit is adjusted in accordance with retirement age and current remaining life expectancy, such that the total value of future pension benefits remains roughly constant. However, special rules imply deviations from an exact actuarial adjustment.18 The income-dependent entitlements are indexed by wage growth only until retirement. The new system is calibrated such that those from the 1943 cohort, who retire at the present statutory retirement age of 67 in 2010, will receive the same pension benefit in 2010 as in the existing system. However, over time retirees receive lower annual benefits than in the present system, since the received benefits will be indexed to the average of the growth rates of wages and consumer prices, rather than the wage growth.

4.2 Direct Effects Within our framework, the reform to the MAS may affect the economy through four channels: (1) labour supply at the intensive margin; (2) labour supply at the extensive margin; (3) government pension expenditures; (4) private savings. 18

An important non-actuarial element is the exemption of 30,000 2005NOK, corresponding to 28.5 percent of the present public minimum pension benefit, from the base of entitlements subject to adjustments to early retirement or increased life expectancy. Moreover, the annual benefits and pension premium are independent of gender and other observable characteristics correlated with life expectancy.

Macroeconomic Effects of Proposed Pension Reforms

121

We assume that the reform does not change aggregate private financial savings. Any specific assumption on the private savings response is hard to justify owing to lack of relevant empirical evidence. For example, Carman et al. (2003, p. 4) write: ‘‘Notwithstanding lots of careful estimation, the empirical literature provides little means of knowing precisely how a particular households’ spending will respond to any given policy change.’’ Under our assumptions on retirement behaviour (see later) the average annual public pension benefit will be nearly unaffected by the reform for individuals who work until old-age retirement.19 This is the main rationale for our assumption of no adjustment in private financial savings. Because neither government nor private financial savings change, the time path of net national financial investments and the foreign financial assets will be the same under the MAS as in the reference scenario. However, aggregate savings will change as firms adjust their fixed capital stocks to changes in relative prices. The subsequent sections briefly give the reasons for our estimates concerning the effects working through channels 1–3. Fredriksen et al. (2006) provides some sensitivity analyses of the assumptions on the two labour supply stimuli, and we check the robustness of the effects of the MAS reform with respect to population ageing. Labour Supply at the Intensive Margin Simulations on MOSART reveal that the average increment in the present value of future pension benefits of raising labour-market earnings by 1 NOK, increases from 0.11 to 0.20 NOK when the present system is replaced by the MAS. In addition, the reform makes the individual income dependency more transparent and similar between individuals. All effects contribute to raising the effective marginal wage rate facing workers at the intensive labour supply margin. The aggregate incentive effect will be uncertain because of uncertainty about the effective tax element in the existing system. Moreover, the relevant weights used to compute the increase in the average marginal effective wage rate should take into account that lowincome workers are found to have a more wage elastic labour supply than high-income workers (see e.g., Aaberge et al., 2000). Our preferred estimate, which we regard as cautious, is that the increased income dependency of the benefits translates to an 8 per cent increase in the average effective marginal

19

Individuals who are disabled before they become old-age pensioners will experience a substantial reduction in their annual old-age benefits. However, the majority of this group has low income, which makes an increase in savings implausible.

122

Dennis Fredriksen et al.

wage rate.20 Fredriksen et al. (2006) discuss in more detail to what extent rational individuals will consider the contributions necessary to finance the MAS benefits as taxation or mandatory savings. Labour Supply at the Extensive Margin Several studies find that labour supply is more elastic on the extensive than on the intensive margin, (e.g., Heckman, 1993). However, the recent international empirical literature does not provide clear guidelines for assessing the magnitude of the effects of pension reforms on retirement. From 12 comparable country studies, Gruber and Wise (2004) conclude that the pension system has an ‘‘enormous effect on retirement’’. Chan and Stevens (2003) confirm that forward-looking measures of pension wealth only, as well as broader measures of wealth, are significantly related to individuals’ expectations of continuing work into their 60 s. However, they conclude that existing research, which largely ignores (unobservable) heterogeneity in tastes for retirement, may substantially overstate the responsiveness of individuals to pension-related incentives. Samwick (1998) finds that levels of pension and other wealth are not major determinants of retirement. Norwegian studies on retirement behaviour are surveyed in Hernæs et al. (2002). The MAS reform has both positive and negative effects on the average age of retirement. On balance we assume that the average retirement age increases by 0.6 years when the present early retirement scheme is phased out in 2015.21 Increased life expectancy strengthens the effect over time. Average retirement age is delayed by 1.6 years in 2030 and by 2.6 years in 2050 (see Figure 1). The estimated increase in the retirement age deserves a justification. First, whereas about 60 per cent of the labour force may retire at the age of 62 in the present system, all individuals get this option at the age of 62 in the MAS. This contributes to a reduction in the average retirement age. On the other hand, the reform increases the individual cost of early retirement. Whereas early retirement in the present system does not reduce benefits in subsequent years, the MAS implies that the annual pension benefit is cut, in a close to actuarial way, the earlier one retires. We will refer to this positive effect on the retirement age as the cost effect.

20

If the difference between the interest rate and the wage growth is set to 2.5 instead of 1.1, this estimate falls to 5 per cent. Assuming this growth-adjusted interest rate to be 0, implies an increase in the effective wage rate by 11 per cent. 21 When the reform is implemented in 2010, the immediate increase in the average retirement age is only 0.1 years because it is assumed to take 5 years to phase out the existing early retirement scheme.

Macroeconomic Effects of Proposed Pension Reforms

123

Figure 1: Extent of Postponed Retirement in the MAS and FBS (Deviations from the Reference Scenario, in years) 4.5 4.0 3.5 3.0 2.5 MAS

2.0 1.5

FBS

1.0 0.5 0.0 -0.5 2010

2020

2030

2040

2050

2060

2070

2080

As a starting point to assess the cost effect on the retirement age, we use the observed labour-market participation rates for persons aged 60–69 in Norway in the early 1980s, when no early retirement scheme existed. These participation rates may serve as an upper boundary for what the labourparticipation rates in these age groups will be under a perfectly actuarial system. As a more realistic and cautious estimate, we assume that a perfectly actuarial pension system would raise the present relatively low participation rates of these age groups to the average of the present rates and the rates observed in the early 1980s. Keeping the present life expectancy fixed, this response implies an increase in the average retirement age equal to 2.4 years. Taking into account that only 60 per cent of the labour force has access to the present early retirement scheme, the postponed retirement corresponds to an increase in total labour supply of about 2 per cent. This response is in line with the estimate by Brinch et al. (2001) of abolishing the present early retirement scheme. However, the cost effect of the MAS reform should be modified, since it includes several non-actuarial elements. Most significantly, an amount equal to 30,000 NOK is exempted from an actuarial division of pension entitlements by the expected number of years as pensioner. We also believe that the gravity of 62 years as the norm for the retirement decision will be stronger in the MAS. The reason is that 62 years will be the only statutory retirement age in the MAS, whereas the present system includes several formal age limits—most notably 67 years in the NIS and 62 years in the

124

Dennis Fredriksen et al.

present early retirement scheme. As pointed out by, for example, Gruber and Wise (2004) and Hernæs et al. (2002), statutory retirement ages are likely to have an important effect on the norm for what is considered to be the normal retirement behaviour. The empirical importance of these modifications is highly uncertain. We assume that they reduce the cost effect on the retirement age from 2.4 to 1.2 years. The cost effect is only relevant for the 60 per cent of the labour force that has access to the present early retirement scheme. From this cost effect we must subtract the effect of making early retirement optional for the whole labour force. Provided that the retirement behaviour is not systematically different between the two groups, the ex post reform retirement age will be the same as the one assumed earlier. This implies that the 40 per cent without access to the present early retirement scheme will reduce their retirement age by 0.3 years. As long as we ignore the effect of increased life expectancy, our estimate on the increase in the average retirement age of a more actuarial system becomes 0.6  1.2 years+0.4(–0.3 years) ¼ 0.6 years. However, so far the estimates have been contingent on constant mortality rates. Increased life expectancy is likely to increase the retirement age in an actuarial system (see e.g., Bloom et al., 2004 for a theoretical discussion). One reason is the preference for consumption smoothing: An additional year of consumption can be financed, at least partly, by postponing retirement. In addition, if increased longevity results from improved health, it can be interpreted as an increase in income, taking the form of more leisure time. At a given consumer real wage rate, the optimal response would be to trade some of the leisure increment for consumption in the labour market. Postponing retirement is one way of doing this. We are not aware of information about the empirical magnitude of the effect of life expectancy on the retirement age. Our best guess is that increasing life expectancy by 1 year increases the average retirement age by 0.4 years. This guesstimate takes into account that 40–45 per cent of the population at age 62 either will be disability pensioners or prefer to retire as early as possible. The remaining shares postpone retirement by 2/3 years when life expectancy increases by 1 year. Such a postponement implies that the annual benefit can be kept approximately constant in the MAS. On the other hand, we believe that increased life expectancy will have a negligible effect on the average retirement age if the present system is maintained. The basic reason is that the annual benefit is independent of the number of years as a pensioner under the present system. Thus, if all consumption initially is financed by the benefit, this consumption–leisure combination can be maintained when life expectancy increases. If the initial consumption level is financed out of private funds in addition to the public benefit, the consumption level cannot be maintained when life expectancy

Macroeconomic Effects of Proposed Pension Reforms

125

increases without increasing labour supply. However, at the statutory early retirement age the individual faces a kinked budget constraint when he decides to work or retire. At this age the effective marginal tax rate of labour income jumps to a very high level, because he simply loses the pension benefit that alternatively could be received.22 Government Pension Expenditures MOSART simulations show that the MAS reform implies a 16 per cent direct cut in government old-age pension expenditures in 2050. This reduction can be decomposed into the following sources: First, keeping life expectancy and indexation rules fixed, the average benefits increase by 3–4 per cent when the MAS replaces the present system. Second, in the MAS, annual benefits are indexed to the average of the growth rates of wages and the consumer price index. In the present system the annual benefits are indexed to wage growth. Less generous indexation helps to reduce government pension expenditures by 7–8 per cent in 2050 compared to the reference scenario. The third source is the impact of a more actuarial cut in annual benefits to increased life expectancy. MOSART simulations show that this effect alone contributes to about 13 per cent of the reduction of government pension expenditures in 2050. This reduction works through two channels. The first channel is a reduction in the number of pensioners. As explained above, those working until they become old-age pensioners will on average postpone retirement, so that their annual benefit will be approximately the same as it would have been under the present system. But the increase in the retirement age reduces the number of old-age pensioners in a given year. In 2050, the number of old-age pensioners will be reduced by 11 per cent (145,000) compared to the reference scenario, corresponding to the 2.6 years increase in the average retirement age. The other channel is a nearly actuarial reduction of the annual old-age pension benefit to individuals who do not work prior to old-age retirement. Disability pensioners are the most important example in this category. In 2050, this effect contributes to a 8 per cent reduction of the average annual benefit received by all old-age pensioners. 4.3 General Equilibrium Effects Table 2 shows the macroeconomic effects in 2050 of replacing the present pension system by the MAS when we account for both direct and general 22

Holmøy (2002) and Holtsmark (2002) estimate the effective marginal tax rate on labour income when the early retirement scheme is taken into account.

126

Dennis Fredriksen et al.

Table 2: Macroeconomic Effects of a More Actuarial System (Deviations from Reference Scenarios in 2050, per cent) Total employment GDP Private consumption Wage cost per man hour Payroll tax rate Real consumer wage rate, excluding the pension effect Effective real consumer wage rate, including the pension effect Net national financial wealth/GDP Gross real investment

Figure 2:

10.6 9.7 9.9 8.4 56.1 5.7 13.7 3.2 11.2

Employment (Million Man-Hours)

Note: MAS is the top line in this chart.

equilibrium effects through the iterative use of MOSART and MSG6. By 2050, employment is 10.6 per cent higher than in the reference scenario (see Figure 2). As firms also adjust their stocks of fixed capital, private consumption and GDP can be expanded in almost the same proportion. The slight difference between the growth in inputs and outputs, respectively, reflects decreasing returns to scale in the production function. A 10 per cent increase in private consumption per capita is a large effect compared to what can be expected from most other policy reforms. CGE estimates of the

Macroeconomic Effects of Proposed Pension Reforms Figure 3:

127

The Payroll Tax Rate

consumption effect of tax- and trade-policy reforms are typically close to 1 per cent. However, the effects of a pension reform need a long time to unfold. Figure 3 clearly brings out the point that even a 10 per cent shift becomes rather modest compared to the consumption growth that normal productivity growth is able to generate over 50 years, independent of the pension system. The MAS reform makes it possible to reduce the payroll tax rate substantially in all years compared to the reference scenario (Figure 4). Whereas maintaining the present system requires an increase in the payroll tax rate from the present level of 13.1 to 25 per cent in 2050, only 11 per cent is sufficient in 2050 in the MAS. The tax cut is possible because of reduced government pension expenditures and the expansion of tax bases. Note that the increase in employment expands most tax bases, not only the bases for the personal income tax and the payroll tax. The ratio of government pension expenditures to GDP is 14.1 per cent lower compared to the reference scenario in 2050. The fall in the wage cost deserves an explanation since it demonstrates that MSG6 accounts for mechanisms which make the determination of factor prices significantly different from the textbook model of a Small Open Economy (SOE). In the SOE-model factor prices would, under certain conditions, be unchanged according to the Stolper–Samuelson theorem.

128 Figure 4:

Dennis Fredriksen et al. Private Consumption Per Capita

Contrary to the SOE model, MSG6 captures the econometric findings of decreasing returns to scale, not only in extraction of natural resources such as crude oil, natural gas and hydro power, but also in Norwegian manufacturing industries. Decreasing returns to scale make a decrease in factor prices necessary in order to meet the long-run external balance constraint when the MAS reform expands the economy and thereby the demand for tradeables. If the price of input factors did not fall, firms would not find it profitable to produce the additional exports needed to pay for the import growth. In the domestic markets, lower costs are transmitted into lower prices of Norwegian products, which induce Norwegian firms and households to reduce the import share in their demand. Fredriksen et al. (2006) provides a more thorough explanation of the equilibrium mechanisms determining the wage cost in MSG6. However, the magnitude of the fall in the wage cost, which equals 8.4 per cent in 2050, may appear surprisingly large compared to scale elasticities close to 0.85 and the roughly 10 per cent expansion of the economy. If labour were the only input, and if the expansion reflected a proportional increase in exports of all tradeables, a 1.5 per cent drop in the wage cost would be roughly sufficient. Among all the forces affecting the wage cost in MSG6, most of the relatively large fall in the wage rate can be attributed to two effects. First, exports of crude oil and natural gas, constituting close to half of total exports, do not adjust to changes in the wage rate. Consequently, the relative increase in the adjustable part of total exports must be as large as 21 per cent in 2050. Decreasing returns to scale make the percentage increase in factors allocated to exports even higher—and the

Macroeconomic Effects of Proposed Pension Reforms

129

necessary reduction of the aggregate factor price index becomes more than 3 per cent in 2050. Second, the cost share of wages is less than 50 per cent in the dominating traded goods industries, even when the indirect labour content in domestically produced intermediates and capital goods is accounted for via the input–output structure of the Norwegian economy. Since no other prices of primary inputs change, the necessary reduction of the wage cost must be more than twice as strong as the necessary reduction of the price index of all inputs. Consumers experience an increase in their real wage rate despite the reduction of wage costs because the reduction of the payroll tax rate is shifted over to the consumer wage rate, and because lower wage costs are transmitted into lower consumer prices. Figures 2–4 show that the effects in 2050 are not stationary. The reason is that average life expectancy is projected to increase steadily over the whole century. The effects of replacing the existing system with the MAS will grow over time, as the present public pension system becomes increasingly expensive as more retirees live longer—whereas the actuarial properties of the MAS prevent, to a large extent, increasing life expectancy raising government pension expenditures. Nevertheless, the payroll tax rate follows an increasing trend even in the MAS scenario after 2020, but this is because ageing increases government expenditures on services used by the elderly. Note that prior to 2020, the demographic developments make it possible to reduce the payroll tax rate every year. With the MAS, it is even possible to cut more than the whole payroll tax. It should be stressed, however, that our models do not give a realistic picture of the short-run adjustments to the pension reform.

Table 3: Decomposition of the Effects of a More Actuarial System (Deviations from the Reference Scenario in 2050, per cent)

1 1.1 2

3 4 5

Increased retirement age Direct effect 8 per cent increase in the effective wage rate Reduced average benefits Interaction effects ( ¼ 5-1-2-3) Total effect

Employment (per cent)

Payroll tax rate (percentage points)

Consumer real wage (per cent)

5.6

8.0

3.3

4.1 4.2

4.5

1.2

0.6

3.1

2.1

0.2

1.7

0.9

10.6

13.9

5.7

130

Dennis Fredriksen et al.

Table 3 decomposes the reform effects into contributions from the direct effects. The improved labour supply incentives at both the extensive and the intensive margin dominate the total effect on employment. Postponed retirement enters MSG6 as two exogenous impulses: (i) the number of pensioners falls, reducing the government pension expenditures; and (ii) the workers who postpone retirement represent an increase in the tax bases. Both the effects make it possible to lower the payroll tax rate, which stimulates labour supply at the intensive margin.

5. Effects of a Flat Benefit Public Pension System (FBS) 5.1 Main Reform Characteristics In this reform alternative the public pension benefit is limited to a flat uniform pension benefit for all pensioners, equal to the minimum pension benefit in the present system. The reform implies privatising the supplementary benefits in the NIS; individual benefits beyond the flat public benefit are left to the market, either through private savings or through occupational pension schemes. The flat benefit is assumed to be pay-as-you-go financed by adjusting the payroll tax rate. Feldstein and Samwick (2002) and Feldstein (2005) discuss how such a system could work. In our simulation we assume that the formal retirement age is reduced from 67 to 62. The flat benefit is indexed to wage growth and is not means–tested against any other sources of wealth or income. The reform is phased in from 2010. NIS pension entitlements accrued prior to 2010 are honoured. 5.2 Direct Effects Labour Supply at the Intensive Margin Given our assumptions, MOSART simulations show that removing the income-dependent supplementary pension in the existing system implies, cet. par., a 3 per cent decrease in the average effective wage rate. Fredriksen et al. (2006) discuss in more detail why the tax element in the contributions necessary to finance the public pension benefits will be much higher in the FBS than in the MAS. However, increasing tax cuts are possible as retirees receiving only the flat benefit gradually replace retirees entitled to pre-reform supplementary benefits. The resulting labour supply effect is examined later. Labour Supply at the Extensive Margin The general access to early retirement from 62 years without any cut in the flat pension benefit contributes to a reduction in the retirement age. On the

Macroeconomic Effects of Proposed Pension Reforms

131

other hand, the annual supplement from private savings will be actuarially adjusted to an increase in the expected number of years as a pensioner. However, under our assumptions (see below) the annual benefit that can be financed by private savings on average accounts for less than one-third of total pension benefit. Because the flat benefit is not actuarially adjusted, the effective subsidy of early retirement is greater in the FBS than in the MAS. Our preferred estimate on the average postponement of retirement is therefore reduced compared to the MAS case. Specifically, we assume that employees on average retire 2 months earlier than in the reference scenario in 2010. However, the impact on retirement of increased life expectancy will be about the same as in the MAS case. Compared to the reference scenario, retirement will on average be postponed by 8 months in 2030 and by 1.5 years in 2050, equivalent to a 2.5 per cent increase in labour supply. Government Pension Expenditures As the retirees receiving pre-reform supplementary benefits die, the decrease in government pension expenditures becomes more significant in the FBS than in the MAS. Ex ante indexation, the average public pension benefit will increase slightly from 2010 to 2020, before it declines to about two thirds of the average pension benefit under the present pension system in 2050.23 Government pension expenditures ex ante indexation will be reduced by nearly the same proportion, given our assumptions of postponed retirement. Note that while government expenditures are almost invariant to the retirement age and the life expectancy in the MAS, this is not the case in the FBS (since the retirees receive the granted flat benefit for all the years). Private Savings The removal of the public supplementary pension benefit will stimulate private savings. However, as noted earlier, any specific assumption on the private savings response is hard to justify because of the lack of relevant empirical evidence. An extreme alternative is that the cut in public benefits is fully compensated through private savings. From the literature on savings behaviour, see Mankiw (2000), such a response is unlikely as an average response for several reasons. Our preferred guess is that private savings compensate for 75 per cent of the loss in public benefits.

23

In 2001NOK, ex ante wage indexation of benefits, the average public pension benefit increases from 126,000 in 2010 to 136,000 in 2020. Then it declines to about 100,000 in 2050.

132

Dennis Fredriksen et al.

General Equilibrium Effects Table 4 shows the macroeconomic effects in 2050 of replacing the present pension system by the FBS when we account for both the direct effects and the general equilibrium effects. Compared with maintaining the present pension system, the FBS stimulates labour supply (see Figure 2). However, this stimulus, and thereby the general expansion of the economy, is considerably smaller compared to the MAS reform. Owing to significantly lower government pension expenditures in the long run, the payroll tax cuts are stronger with the FBS than with the MAS after 2025. While the MAS makes a payroll tax rate of 11 per cent sufficient in 2050, the corresponding tax rate can be reduced to 6.3 per cent with the FBS (see Figure 4). There are two main reasons why employment is lower with the FBS than with the MAS. These, as well as other effects, are quantified in Table 5. First, as explained earlier, the average retirement age is lower in the FBS than in the MAS. Compared to the reference scenario in 2050, the direct labour supply effects are, respectively, 2.5 and 4.1 per cent. Second, taking the labour supply incentive effects of the pension system into account, the effective marginal taxation of labour income is lower in the MAS than in the FBS. This is reflected in change rates reported in Table 4 for the effective real consumer wage rate, including the pension effect. The formal gross tax revenue is greater in the MAS than in the FBS. However, in the MAS the reimbursement of a significant share of the tax revenue to retirees makes individuals perceive a significant share of the formal gross tax payments as mandatory savings. As a result, the stronger income dependency in the MAS makes the effective net taxes smaller than in the FBS. Replacing the present pension system with the MAS lowers the effective tax rate by 8 per cent. On the other hand, all of the (remaining) government pension expenditures in

Table 4: Macroeconomic Effects of a Flat Benefit System (FBS) and a More Actuarial System (MAS) (Deviations from the Reference Scenario in 2050. Per Cent)

Total employment GDP Private consumption Wage cost per man hour Payroll tax rate Real consumer wage rate, excluding the pension effect Effective real consumer wage rate, including the pension effect Net national financial wealth/GDP Gross real investment

MAS

FBS

10.6 9.7 9.9 8.4 56.1 5.7 13.7 3.2 11.2

4.8 4.6 5.7 2.3 75.6 14.2 11.2 73.7 7.1

Macroeconomic Effects of Proposed Pension Reforms

133

Table 5: Decomposition of the Effects of a More Actuarial System (MAS) and a Flat Benefit System (FBS) (Deviations from the Reference Scenario in 2050, Per Cent) Employment (per cent)

1 Increased retirement age 1.1 Direct effect 2 Reduced tax rate owing to lower benefits, and changed income dependency 2.1 Changed income dependency 2.2 Lower benefits 3 Accumulation of financial assets 4 Interaction effects ( ¼ 5-1-2-3) 5 Total effect

Payroll tax rate Consumer real wage (per cent) (percentage points)

MAS

FBS

MAS

FBS

MAS

FBS

5.6 4.1 4.8

3.4 2.5 0.2

8.0

5.0

3.3

2.0

7.6

6.9

3.3

5.7

4.2

1.7

4.5

2.1

1.2

0.6

0.6 0

1.9 1.3

3.1 0

9.0 7.7

2.1 0

6.3 6.2

0.2

0.1

1.7

1.1

0.9

0.6

10.6

4.8

13.9

18.5

5.7

14.5

the FBS must be financed by distortionary taxation. Fredriksen et al. (2006) explains the difference in effective taxation between the MAS and the FBS in greater detail. Privatising supplementary pension benefits also affects employment through other mechanisms. An important one is the income effect because of the double burden carried by the working generations under the transition from a (pure) pay-as-you-go pension system to a (more) funded system. Cet. par the transition from the present system to the FBS implies an income loss for the cohorts who must finance pre-reform supplementary benefits through taxes, because they cannot look forward to receiving such benefits themselves. This income loss stimulates labour supply and reduces consumption. The effects are particularly strong in the first couple of decades after the reform is implemented, when the number of retirees with entitlements from the present pensions system is still high. MSG6 captures an important interaction between changes in savings, the real wage rate and employment. Since the government financial investment is unchanged compared to the reference scenario, the increase in private savings is basically matched by accumulation of foreign financial assets. Thus, the net accumulation of private pension funds requires increased net exports. Thus, the inter-temporal reallocation of aggregate consumption

134

Dennis Fredriksen et al.

must be associated with a temporary reallocation of resources from industries producing non-traded consumer goods to the traded goods industries. Such a reallocation requires a reduction of wage costs from the reference path owing to decreasing returns to scale. As the aggregate pension fund converges to its desired level, net exports decrease compared to the reference scenario (because a greater share of imports is financed by interest income). Thus, the wage rate can increase, and by 2050 the percentage increase in consumption exceeds the percentage increase in employment and GDP. The dynamic wage rate adjustments transform the increase in individual supplies of savings into actual saving. Line 3 in Table 5 captures both the income effect experienced by the transition generation and the effect of accumulation of foreign assets, where the former dominates the latter over the entire simulation period. The temporary fall in the wage rate has only modest effects on labour supply. It reduces the initial increase in labour supply and dampens the succeeding fall. The model also captures another important interaction effect: Pension reforms alone affect wage growth, which in turn affects the balance of the public budget and thereby the room for tax cuts. Considering the FBS reform, this effect can be explained as follows: When public pensions are indexed to the wage rate, and because wage costs dominate government spending, an increase in the wage rate yields a close to proportional increase in government expenditures. Whereas most tax bases in the Norwegian Mainland economy are also close to proportional wage rate, the share of petroleum wealth, which the fiscal policy rule allows to be used each year, is independent of the wage rate. Hence, an increase in the wage rate generates cet. par a fiscal deficit, which must be neutralised by raising the payroll tax rate. This interaction effect modifies the cut in the payroll tax rate made possible in the long run by the FBS reform. In addition to the weaker labour supply stimulus, it explains why privatising the supplementary pension benefits does not generate an even greater long-run cut in the formal taxes compared to what is possible in the MAS. Comparison with Related Simulation Studies The simplicity of the FBS system makes it a common reference system in the discussion of pension systems. However, differences between the initial public pension systems entail country-specific effects of a reform to a common FBS, even in the hypothetical case of identical models. Thøgersen (2001) uses an OLG model to simulate the dynamic effects of a FBS reform of the Norwegian system.24 Comparing long-run effects, the reform makes it

24

The FBS reform is called an individualized funding strategy in Thøgersen (2001).

Macroeconomic Effects of Proposed Pension Reforms

135

possible to reduce the labour income tax rate by 13 percentage points from a scenario in which the present system is maintained. Our simulation allows a reduction by 18.5 percentage points in the payroll tax rate in 2050. In addition to model differences, the difference is also likely to reflect more rapid increase in life expectancy in our scenarios than in Thøgersen’s. This reinforces the tax cuts made possible by privatising the supplementary benefit. The steady-state effect on employment is only 1 per cent in Thøgersen’s study. However, his study does not include changes in retirement. Moreover, inter-temporal substitution makes the employment effects stronger in the years 2025–2070 than the steady-state effects. After 40–50 years, employment is 4–5 per cent higher than in his reference scenario, which is not so far from the 5.7 per cent increase we find in 2050. Fehr et al. (2003) have calibrated the OLG model used by Fehr (1999) to Norwegian data and simulate the effects of a FBS reform of the Norwegian pension system. Compared to Thøgersen (2001), this study captures endogenous retirement and income heterogeneity within each cohort. Compared to our results, Fehr et al. (2003) obtain much smaller macroeconomic effects. They estimate the long-run increase in consumption to 2.0 per cent, while private consumption increases by 5.7 per cent in our simulation (Even if one corrects for no growth in government consumption, a significant difference remains). The cut in the consumption tax is 4.0 percentage points in Fehr et al. (2003), while the cut in the payroll tax rate is 19 percentage points in our study. Some of this large difference can be explained by differences in the population projections and the increase in government expenditures. Moreover, the initial equilibrium consumption tax rate of only 15.2 per cent used by Fehr et al. (2003) is very low compared to the VAT rate and other indirect tax rates. It indicates that the base of the consumption tax has been very broadly defined. Kotlikoff et al. (1999) simulate the dynamic effects of a FBS reform of the U.S. social security system.25 The steady-state effects on labour supply and the capital stock are, respectively, 1.2 and 12.4 per cent. These effects are less than one-third of the effects obtained in the case of complete privatisation of social security. The reason is the same as pointed out in the interpretation of our results: Tax financing of the flat benefit implies that a substantial labour supply disincentive remains. Moreover, in Kotlikoff et al. (1999), the FBS reform reduces employment, the capital stock and output from the initial levels during the transition. It takes about 50 years before these variables pass their initial levels.

25

Kotlikoff et al. (1999) refer to the FBS reform as ‘‘Privatisation with a Flat Benefit’’. In the simulation of this reform the income tax is used to finance accrued benefits.

136

Dennis Fredriksen et al.

6. Conclusions We have used a detailed dynamic microsimulation model, together with a large-scale dynamic CGE model, to project the macroeconomic development of the Norwegian economy until 2050 under different public pension systems. The detailed description of population heterogeneity and the pension system in the microsimulation model allows accurate calculations of the direct effects on government pension expenditures of population ageing and pension reforms. The CGE model captures a rich menu of general equilibrium effects caused by changes in these variables. In particular, the CGE model accounts for endogenous adjustments of most tax bases and prices of government consumption. The two models are run iteratively to obtain consistency. Specifically, the equilibrium effects on the wage rate and labour supply have been accounted for in the results produced by the microsimulation model, and the necessary tax rate adjustments rely on the government pension expenditure projections produced by the microsimulation model, as well as the general equilibrium effects captured by the CGE model. The reference path shows that continuation of the present pension system would make the present course of fiscal policy unsustainable after 2020. In 2050, even such a broad-based tax as the payroll tax rate must be raised to 25 per cent, nearly a doubling from the present average of 13 per cent. Moreover, a further steady increase is necessary when the time horizon is extended beyond 2050. The necessary increase in the tax burden after 2020 is much stronger than corresponding estimates for most of the seven major economies in Chauveau and Loufir (1995). Thus, the petroleum wealth enjoyed by Norway is not sufficient to finance the Norwegian welfare state when one looks beyond 2020. The reference scenario provides a compelling case for a pension reform that stimulates labour supply and establishes more actuarial mechanisms that motivates individuals to postpone retirement as they live longer. It is questionable whether the projected increase in the future tax burden will be politically acceptable. Even if it is, the efficiency loss may be severe, since effective tax rates are already relatively high in Norway. Any projection is uncertain. On balance, however, in our opinion the estimate of the necessary increase in future tax rates is likely to be negatively biased, because it rests on the assumption that the standard of government services per user is kept constant over the whole simulation period. Such a development would imply a radical break with historical trends, including a much stronger growth in private than government consumption. It should also be noted that the scope for tax cuts before 2020 rests on these assumptions, as well as on the presumption of a high degree of fiscal discipline. If the room for temporary tax cuts is instead used to improve the standards of services directed towards the elderly, the need to raise tax rates after 2020 will be exacerbated.

Macroeconomic Effects of Proposed Pension Reforms

137

Do the proposed pension reforms have significant macroeconomic effects? Our estimates suggest an ambiguous answer. On one hand, our estimated effects of the proposed pension reforms are very large compared to the effects of most other policy reforms. In sharp contrast to the 10 per cent expansion of the Norwegian economy generated by the MAS reform, CGE studies of tax- and trade-policy reforms typically estimate welfare gains of less than the gains obtained over one year with normal productivity growth. On the other hand, pension reforms are not likely to matter very much for the average individual consumption level three decades or more ahead (see Figure 3). Our projections confirm Paul Krugman’s statement: ‘‘Productivity isn’t everything, but in the long run it is almost everything. A country’s ability to improve its standard of living over time depends almost entirely on its ability to raise its output per worker’’ (Krugman, 1990, p. 9). Compared to several decades of exponential growth, most partial policy reforms will turn out to be rather insignificant, as long as they do not affect the growth rate. This is particularly true for a pension reform, which needs a long time to unfold its long-run effects. Both reforms have a great positive effect on fiscal sustainability, which makes it possible to avoid a dramatic increase in the future tax burden. The effect is strongest in the FBS reform, which privatises the responsibility for supplementary benefits. In 2050, the payroll tax rate can be reduced from 25 per cent in the reference scenario assuming continuation of the present pension system, to 6.3 per cent in the FBS alternative and 10.9 per cent in the MAS alternative. These effects are not stationary; the payroll tax rate follows an increasing trend in both reform scenarios after 2020. However, the necessary growth in the payroll tax rate after 2020 is largely driven by growth in government consumption of services directed to the elderly. One may question whether a pension reform should pay for improvements in these services. Basically, the scope for tax cuts created by the reforms cannot be attributed to ex post reductions in average public pension benefits. Instead most of the tax cuts can be attributed to the growth in tax bases generated by the positive effects on labour supply incentives. The MAS reform implies the strongest labour supply stimulus; in 2050, employment is 10.6 per cent higher than in the reference scenario. The corresponding increase generated by the FBS reform is 4.8 per cent. When calculating the fiscal effects of the reforms it is important that our model framework takes into account that most tax bases are endogenous and highly correlated with employment. The effective marginal tax rate on labour income is lower in the MAS than in the FBS, despite higher formal tax rates in the MAS than in the FBS. This is because a greater share of the total average benefit is actuarially adjusted to early retirement and increased life expectancy in the MAS compared to the FBS. Moreover, because of a higher perceived correlation between labour-market earnings and public pension benefits in the MAS than

138

Dennis Fredriksen et al.

in the FBS and the present system, a substantial share of the formal tax on labour income is regarded as mandatory savings rather than a distortionary tax in the MAS. The difference between the effective marginal tax rates on labour income is the main reason why the increase in employment is stronger in the MAS than in the FBS. A large effective tax rate on the consumers’ return to work implies that the reallocation of time from leisure to market work improves social efficiency. The most important elements of the effective marginal tax rate on labour income include an average marginal tax on personal labour income of approximately 40 per cent, a payroll tax rate averaging 13 per cent and net indirect taxation of consumption, including VAT, averaging 19 per cent. The tax wedge made up by these tax rates makes the ratio between the social and the private marginal rate of transformation of leisure into consumption as large as 2.3.26 The increase in payroll tax from the present 13 per cent to 25 per cent in 2050 in the reference scenario exacerbates the distortion of time allocation. To the extent that employment increases as a result of delayed retirement the efficiency gain will be even larger, because leisure through early retirement is heavily subsidised under the present pension system. Transformation of increased labour supply and individual savings into higher employment and more assets involve equilibrium adjustments of the real wage rate and industry structure that may be hard to realise. In particular, a higher degree of pre-funding of future pension expenditures must mainly take place through net financial investments in foreign assets. Accumulation of foreign assets cannot take place unless real resources are reallocated from consumption to net exports. However, the traded goods sector (e.g., manufacturing industries) will not be willing to employ more labour unless the real wage cost is sufficiently reduced. Thus, cet. par prefunding warrants slower real wage growth. One aspect of the diagnosis of ‘‘Dutch Disease’’ is that re-industrialisation may be much harder to carry through than a process involving real exchange rate appreciation and deindustrialisation. Like many other open economies, Norway has experiences, which make it questionable to what extent actual wage formation follows norms defined in the textbook equilibrium model of a small open economy. The temporary large revenues from the petroleum sector have probably increased the problems of bringing wage growth to a level that is sustainable in a long-run perspective. More pre-funding exacerbates problems caused by rigidities in the wage setting. The welfare state is already under pressure, and ageing will further erode its financial base. In Norway the large cash flows from rapid transformation of petroleum wealth to financial assets make the problems of fiscal

26

See Fæhn and Holmøy (2000) for a derivation of this estimate.

Macroeconomic Effects of Proposed Pension Reforms

139

sustainability less transparent than in other OECD economies. This makes long-run macroeconomic projections even more relevant in Norway than in other countries. In particular these projections should quantify the consequences of different changes in the government provision of subsidised welfare services and income-replacement schemes. Analyses similar to those undertaken by the Danish Welfare Commission (Velfærdskommissionen, 2004) seem highly relevant also in the Norwegian context and are high on our agenda for future research. Such projections should be based on models that cannot be blamed for ignoring available relevant information. This apparently obvious ambition has a more controversial and perhaps non-fashionable implication: Accuracy, gained by including a disaggregated classification of, for example, tax bases, exact descriptions of the tax and pension systems, detailed modelling of population heterogeneity and market structures affecting the real time dynamics of policy reforms, should be given priority over analytical tractability and transparency. Although the model work underlying this paper has gone a long way in accounting for details of potential relevance, there is obvious scope for improvements. Specifically, consistency can be improved by merging the most important aspects of individual life courses and the general equilibrium mechanisms into a CGE model with overlapping generations and income heterogeneity within each cohort. Moreover, since labour supply at both the intensive and the extensive margins—as well as savings behaviour—are crucial for the results, future modelling work should probably give priority to capturing the heterogeneity of behaviour found in micro econometric studies.

References Aaberge, R., Dagsvik, J. and Strøm, S. (1995). Labour Supply Effects of Tax Reforms. Scandinavian Journal of Economics, 97, 635–659. Aaberge, R., Colombino, U. and Strøm, S. (2000). Labour Supply Responses and Welfare Effects from Replacing Current Tax Rules by a Flat Tax. Empirical Evidence from ITALY, Norway and Sweden. Journal of Population Economics, 13, 595–621. Aaberge, R., Colombino, U., Holmøy, E., Strøm, B. and Wennemo, T. (2004). Population Ageing and Fiscal Sustainability. An Integrated Micro-macro Analysis of Required Tax Changes. Discussion Paper 367, Statistics Norway. Andersen, T.M., Jensen, S.E.H. and Pedersen, L.H. (2004). The Welfare State and Strategies Towards Fiscal Sustainability in Denmark. Working Paper 2004.4, The Danish Welfare Commission. Antolin, P. and Suyker, W. (2001). How Should Norway Respond to Ageing? Working Paper 296, OECD Economics Department Series, OECD. Beetsma, R., Bettendorf, L. and Broer, P. (2003). The Budgeting and Economic Consequences of Ageing in the Netherlands. Economic Modelling, 20, 987–1013.

140

Dennis Fredriksen et al.

Bloom, D., Canning, D. and Moore, M. (2004). Health, Longevity and Optimal Retirement. Mimeo, August, Harvard School of Public Health, Boston, MA. Bovenberg, L. and Knaap, T. (2005). Ageing, Funded Pensions and the Dutch Economy. CESIFO Working Paper 1403. Brinch, C., Hernæs, E. and Strøm, S. (2001). Labour Supply Effects of an Early Retirement Program. Memorandum 33/2001, Department of Economics, University of Oslo. Carman, K.G., Gokhale, J. and Kotlikoff, L.J. (2003). The Impact on Consumption and Saving of Current and Future Fiscal Policies. Mimeo, September 2003. Chan, S. and Stevens, A.H. (2003). Do Changes in Pension Incentives Affect Retirement? A Longitudinal Study of Subjective Retirement Expectations. Journal of Public Economics, 88, 1307–1333. Chauveau, T. and Loufir, R. (1995). The Future of Public Pensions in the Seven Major Economies. OFCE Working Paper, Research Department-OFCE, Paris. Coile, C. and Gruber, J. (2003). Fiscal Effects of Social Security Reforms in the United States. Working Paper 2003–2005, Center for Retirement Research at the Boston College. (Downloadable at http://www.bc.edu/crr.) European Commission (2001). Budgetary Challenges Posed by Ageing Populations. The Impact on Public Spending on Pensions, Health Care for the Elderly and Possible Indicators of the Long-Term Sustainability of Public Finances. Report of the Working Group on Ageing Populations to the Economic Policy Committee. Fæhn, T. and Holmøy, E. (2000). Welfare Effects of Trade Liberalisation in Distorted Economies. A Dynamic General Equilibrium Assessment for Norway, in Harrison, G.W., Jensen, S.E.H. and Pedersen, L.H. (eds), Using Dynamic General Equilibrium Models for Policy Analysis, Amsterdam, North-Holland, pp. 171–202. Fehr, H. (1999). Welfare Effects of Dynamic Tax Reforms, Mohr Siebeck, Thu¨ringen. Fehr, H., Sterkeby, W.I. and Thøgersen, Ø. (2003). Social Security Reforms and Early Retirement. Journal of Population Economics, 16, 345–361. Feldstein, M. (2005). Structural Reform of Social Security. NBER Working Paper 11098. Feldstein, M. Samwick, A. (2002). Potential Paths of Social Security Reform, in: Poterba, J. (Ed.), Tax Policy and Economy 2001. MIT Press, Cambridge, MA, 16, 181–224. Fredriksen, D. (1998). Projections of Population, Education, Labour Supply and Public Pension Benefits. Analyses with the Dynamic Simulation Model MOSART. Social and Economic Studies 101, Statistics Norway. Fredriksen, D., Heide, K.M., Holmøy, E. and Solli, I.F. (2005). Makroøkonomiske virkninger av pensjonsreformer, Reports 2005/2. Statistics Norway (In Norwegian). Fredriksen, D., Heide, K.M., Holmøy, E., and Solli, I.F. (2006), Macroeconomic Effects of Proposed Pension Reforms in Norway. Discussion Paper No 417, Research Department, Statistics Norway (available from http://www.ssb.no/english/ publications/) Gruber, J. and Wise, D. (eds). (2004). Social Security and Retirement Around the World. Micro-estimation. University of Chicago Press, Chicago. Heckman, J. (1993). What Has Been Learned About Labour Supply in the Past Twenty Years? American Economic Review, 83, 116–121.

Macroeconomic Effects of Proposed Pension Reforms

141

Heide, K.M., Holmøy, E., Lerskau, L. and Solli, I.F. (2004). Macroeconomic Properties of the Norwegian Applied General Equilibrium Model MSG6. Reports 2004/ 18, Statistics Norway. Hernæs, E., Røed, K. and Strøm, S. (2002). Yrkesdeltakelse, pensjoneringsadferd og økonomiske insentiver. Reports 4/2002. Ragnar Frisch Center of Economic Research, Oslo (In Norwegian). Holmøy, E. (2002). Hva koster tidligpensjonering for samfunnet? Økonomiske analyser 2/2002, Statistics Norway (In Norwegian). Holmøy, E. and Heide, K.M. (2005). Is Norway Immune to Dutch Disease? CGE Estimates of Sustainable Wage Growth and De-industrialisation. Discussion Paper 413, Statistics Norway. Holtsmark, B. (2002). Hva fa˚r AFP-pensjonister igjen for a˚ arbeide? Økonomiske analyser 2/2002, Statistics Norway. Jensen, S.E.H., Nødgaard, U. and Pedersen, L.H. (2001). Fiscal Sustainability and Generational Burden Sharing in Denmark. Nordic Journal of Political Economy, 28(1), 43–60. Kotlikoff, L., K. Smetters and J. Walliser (2001). Finding a Way Out of America’s Demographic Dilemma. Working Paper Series, NBER 8258. Kotlikoff, L., Smetters, K. and Walliser, J. (1999). Distributional Effects in a General Equilibrium Analysis of Social Security. Mimeo, Boston University. Krugman, P. (1990). The Age of Diminished Expectations. The MIT Press, Cambridge, MA. Lindbeck, A. and Persson, M. (2003). The Gains from Pension Reform. Journal of Economic Literature, XLI, 74–112. Mankiw, G. (2000). The Savers-Spenders Theory of Fiscal Policy. American Economic Review, 90, 120–125. McMorrow, K. and Roeger W. (2002). EU Pension Reform — An Overview of the Debate and an Empirical Assessment of the Main Policy Reform Options. DG ECFIN Economic Papers No. 162, European Commission. Miles, D. (1999). Modelling the Impact of Demographic Change Upon the Economy. The Economic Journal, 109, 1–36. Ministry of Finance (2001). Langtidsprogrammet 2002– 2005. St.meld. no. 30 (20002001), Oslo (In Norwegian). Ministry of Finance (2004a). Nasjonalbudsjettet 2005. St.meld. no. 1, Oslo (In Norwegian). Ministry of Finance (2004b). Perspektivmeldingen 2004 — utfordringer og valgmuligheter for norsk økonomi. St.meld. no. 8, Oslo (In Norwegian). Ministry of Finance (2004c). Pensjonsreform — trygghet for pensjonene. St.meld. no. 12, Oslo (In Norwegian). NOU (2004:1). Modernisert folketrygd. Bærekraftig pensjon for framtida. The Ministry of Finance and The Ministry of Social Affairs, Oslo (In Norwegian). OECD (1998). Maintaining Prosperity in an Ageing Society, Paris. OECD (2000). Reforms for an Ageing Society: Social Issues, Paris. OECD (2001). Fiscal Implications of Ageing: Projections of Age-Related Spending. OECD Economic Outlook, 69, Ch. IV, Paris. Samwick, A.A. (1998). New Evidence on Pensions, Social Security, and the Timing of Retirement. Journal of Public Economics, 70(2), 207–236.

142

Dennis Fredriksen et al.

Statistics Norway (2002). Befolkningsframskrivinger. Nasjonale og regionale tall, 2002-2050. Dowloadable at www.ssb.no/emner/02/03/folkfram/ (In Norwegian). The Danish Welfare Commission (2004). Analyserapport — Fremtidens velfærd kommer ikke af sig selv. Downloadable at www.velfaerd.dk. (In Danish). Thøgersen, Ø. (2001). Reforming Social Security: Assessing the Effects of Alternative Funding Strategies. Applied Economics, 33, 1531–1540. Visco, I. (2002). Ageing Populations: Economic Issues and Policy Challenges, in Siebert, H. (ed), Economic Policy for Ageing Societies, Springer, Berlin, pp. 9–47. Velfærdskommissionen (2004). Analyserapport-Fremtidens velfaerd kommer ikke af sig selv, downloadable at www.velfaerd.dk (in Danish).

Chapter 6

Adding Private Pensions to the Canadian DYNACAN Model Richard J. Morrison DYNACAN Team, Social Development Canada, Government of Canada

Abstract Meaningful evaluation of government social security programmes, and projections of the probable impacts of changes to them, should be conducted in the context of the broader retirement income system. Private (employer) pensions constitute a major source of income for Canadian seniors, providing more retirement income than any other single source. Thus, sound policy development for seniors’ programmes — such as the Canada Pension Plan — requires that one also model private pensions, together with other income sources, taxes, and income-tested programmes. However, in Canada there is a scarcity of accessible, comprehensive pension data at both the micro and macro levels, rendering such inclusion challenging. This paper outlines DYNACAN’s development of the initial version of a private pension module. Because ownretirement pensions and survivor pensions have very different characteristics, the module development involves teasing out the mix of these two types of pension. It requires giving the two types of pensions to the right numbers and kinds of people — and at appropriate levels. And it requires projecting pension incidence, average levels, and variation in private pensions well on into the future. The paper also briefly addresses the validation of the pensions against historical data, plus such validation as is feasible for the quality of the projection into the future.

1. Introduction In 1990, Human Resources Development Canada concluded that it required a longitudinal dynamic microsimulation model to perform distributional analysis of possible changes in policy for the Canada Pension Plan (CPP) and to analyse the interactions of CPP with other programmes. International Symposia in Economic Theory and Econometrics, Vol. 15 r 2007 Published by Elsevier B.V. ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15006-1

144

Richard J. Morrison

Construction of DYNACAN began in the mid-1990s and, by the end of the 1990s, the model had become part of the formal process for assessing proposed changes to the CPP. DYNACAN simulates, at the level of the individual and the family, Canada’s population and labour-force dynamics, the operation of the CPP, and the impacts of prospective changes to the CPP. However, for CPP policy development and assessment to be relevant, the impacts of prospective changes must be addressed within the framework of the broader retirement income system. That is, impact analyses must take into account taxes and income-tested benefits, so that the net or ‘‘effective’’ consequences of changes to the CPP can be assessed, rather than just the gross impacts. Treatment of taxes and income-tested benefits are also relevant for those seeking to address issues of equity and incentives. However, adding such breadth to an analysis means that one must also model the several income sources that are taxable, the taxes themselves, and the logic of the relevant income-tested benefit programmes. Morrison (2002a) describes how DYNACAN presently contributes to policy development for the CPP, and outlines possible future contributions. (A fuller summary description of the DYNACAN model is available in Morrison (2006) in the second volume of this series.) This paper summarizes the results of our initial attempts to add to DYNACAN a major income source for seniors, namely private (employer) pensions. It describes the general approach taken, and some of the parameters to be used, as well as the significant data limitations involved. The paper also briefly addresses the validation that has been conducted, or planned, to help ensure that the simulated private pension income is generally realistic.

2. The Decision to Model Private Pension Income Private pensions in Canada are very distinct from governmentally administered plans such as the CPP or Old Age Security programme. Most large employers, and virtually all levels of government, provide such plans as supplements to the government plans. However, there is no legal requirement that employers must provide private pensions for their employees. Most private pension plans are contributory, in that the covered employees pay some of the costs of the pensions that they will receive on retirement. Most plan participants belong to defined-benefit plans — although, in terms of numbers of plans, the fraction of defined-contribution plans is increasing. Private pension coverage is patchy — and very non-random. Part-time workers, those changing jobs frequently, and those in lower-paying jobs are particularly likely not to be covered. Many working Canadians will

Adding Private Pensions to the DYNACAN Model

145

participate in private pension plans for only a part of their working careers, so that their private pensions in retirement will reflect only a portion of their earnings histories. The portability of pension coverage from one job to another is typically limited. Despite such patchiness, private pension income is a particularly important component of Canada’s retirement income system. Specifically, as measured by aggregate amount of income received, it is the largest single source of income for persons 65 or older, despite being received by fewer than half of the tax filers in that age range. For example, for Canadian seniors, income from private pensions exceeds that from the Canada and Quebec Pension Plans combined, from Canada’s Old Age Security programme, from its income-tested Guaranteed Income Supplement, from the total of all forms of earnings, and from all of the individual categories of return to savings. (However, the collective returns to investment — the sum of interest, dividends, annuities, registered savings plans, and capital gains — do exceed the income from private pensions.) Private pension income is consequently a crucial input for the calculation of seniors’ marginal tax rates, their personal income taxes, and their incomes for calculating income-tested benefits. In this sense, it plays a significant role in determining the net impacts of prospective changes to the CPP — i.e., the impacts after the effects of taxes and other programmes, including income-tested programmes, are taken into account. The comparative importance of private pension income makes it an obvious choice as the first income source, beyond earnings and government benefit programmes, to be modelled. As well, we anticipate that the lessons learned in modelling private pensions will be helpful in modelling other income sources in DYNACAN. The private pension module described here is, very intentionally, only a fairly simple, first-cut attempt, with a more sophisticated version already under development. The module described here, itself still under development, models only the benefits flowing from such plans. It does not attempt the more elegant, but considerably more demanding, approach of modelling occupation, pension coverage, benefit formulae, and both employer/ employee contributions longitudinally across the synthetic individuals’ working lives. Quite explicitly, the DYNACAN team expects to learn, from this prototype effort, what appears to work or not to work — and consequently where to put its attention in developing the next generation pension module. That next version will integrate the receipt of pensions more tightly with the labour-force module, including later-life work by recipients of private pensions. (For the reader seeking more background on Canada’s retirement system, Maser and Be´gin (2003) provide a good statistically oriented description that includes very useful chapters about private pensions.)

146

Richard J. Morrison

3. Structure of the Paper Organizationally, the several sections of the paper address: 

 









the major challenges associated with modelling private pensions and the DYNACAN team’s general approach to resolving them in this first-pass pension module: the overall incidence of pension receipt, and its distribution longitudinally; the estimation of point-in-time equations for pension incidence and pension levels; the importance of the distinction between own-retirement pensions and survivor pensions, and the mix of such pension types over time; the addition of private pensions to DYNACAN’s starting database: identifying which individuals will have pensions assigned to them, and the types and sizes of those pensions; a high-level description of the pension module as the team expects it to operate for simulating pensions year by year; and the role of validation in the pension module development.

4. Challenges Any attempt to model private pensions in Canada will necessarily involve several significant challenges. Among these, the most important appear to be as follows. First, private pension receipt in Canada has evolved substantially over time and appears likely to continue to evolve. Modelling pensions in a longitudinal model means having to capture that evolution. Readily available semi-aggregate taxation data over the past few decades show that the incidence of pensions has increased for Canadians, especially for women. Over the same period, the age distribution of such pensions has changed considerably, with more of all private pension income going to Canadians under the age of 65, consistent with observed declines in retirement ages. However, longitudinally, the incidence of pension receipt for men has apparently leveled off and has begun to decline for virtually all age groups. The data suggest that a similar leveling off of pension incidence is underway for women, although it appears to be slowed by the gradual convergence between men’s and women’s labour force participation and earnings levels. Second, any model of pension benefits for individuals must recognize the need for temporal consistency, for both qualitative pension receipt and quantitative pension amounts. Once individuals begin to receive these benefits, they typically continue to receive them for the rest of their lives. Private pension amounts may optionally be indexed, sometimes quite informally or sporadically, for all or part of inflation — but the nominal levels of benefits

Adding Private Pensions to the DYNACAN Model

147

will typically not decrease over time. In one significant exception, however, as a result of the integration between public and private pensions, private pension levels often do decrease at age 65, when individuals become eligible for public pension benefits. At the same time, apart from inflation adjustments, reported pension amounts for individuals may increase as they begin to receive pensions from other previous employers, or become eligible for survivor pensions upon the death of a covered spouse. Achieving temporal consistency becomes more challenging yet when one considers the increasing importance of registered retirement savings plans as an instrument for providing income in retirement. Third, there is an understandable tendency to think of private pensions as effectively consisting of a form of delayed remuneration for work done while employed by an employer who sponsors a pension plan. That is, one tends to think of them as own-retirement pensions. However, under current pension legislation, the death of an own-retirement beneficiary typically generates a survivor pension for a surviving spouse. Thus, individuals may receive own-retirement pensions, survivor pensions, both, or neither. The mix of these pension types naturally varies by gender and age, as well as across time. Fourth, on top of all this, one recognizes that Canadian private pension plans themselves vary considerably. They differ widely in terms of their fundamental characteristics — e.g., benefit formulae, indexation provisions, and survivor benefit levels — as well as the details of their integration with government benefit programmes. Fifth, building models of longitudinal processes is challenging in its own right, but a scarcity of appropriate data makes the challenge significantly more severe. As a result of privacy and confidentiality concerns, there are no publicly available, comprehensive, large-sample longitudinal microdata on private pension receipt. Reasonably comprehensive data for the whole country are freely available only in semi-aggregate form. Even then, they are available only in formats whose age-group boundaries change over time. Further, at the time of constructing this prototype module, we were not aware of any remotely comprehensive individual-level data that distinguished among private plans’ own-retirement pensions, survivor pensions, and dual pensions (though a release of microdata from Canada’s 1999 Survey of Financial Security later included such information and we were able to use the preliminary public-release version of that survey to assist in the validation of our model). Moreover, the available Census and survey microdata are generally based on self-reporting. The data may show significant gaps relative to aggregates compiled on a National Accounts basis and/or a sampling frame that departs from the overall population. And, of course, survey data are necessarily limited to the past. There are no pension data available for the future

148

Richard J. Morrison

projection period that is of greatest interest for policy modelers. Moreover, neither do there exist recognized official projections of future pension incidence and amounts. Projecting pension income in Canada will thus, inevitably, involve a great deal in the way of assumptions and approximations. This paper describes the ones that the DYNACAN team has made in its prototype attempt to include private pensions in the model.

5. Basic Approach It is useful to begin by describing the overall approach and underlying topics at a summary level. Subsequent sections then develop several of these topics in greater detail. Because the DYNACAN prototype pension module is presently a work in progress, this paper will primarily characterize the portions that are already complete and then outline the remaining, planned steps. It will also indicate the nature of validation planned once the module is operational and begins to produce pension projections whose quality one can assess. A major simplification is that the prototype pension module will not attempt to model pension plan participation and pension entitlement accumulation throughout the synthetic individuals’ working careers. Rather, the initial version attempts to ‘‘cut to the chase’’, by modelling just the private pension income that accrues from such participation. This is, at best, a short-term approach. Under it, integration with the retirement decision and the labour-force module, though marginally present via variables in the equations for pension emergence, is much weaker than would be desirable. Within this framework, the fundamental questions are: (1) How many synthetic individuals will be assigned new pensions? (2) Which particular individuals will receive those pensions, and of which types? (3) How much pension income will they receive? (4) How will the levels for pensions already in pay change over time? A key first step, focused on the ‘‘how many?’’ question, is to use publicly available semi-aggregate taxation data to characterize the overall incidence of pension receipt by age and gender over recent history. From this data, with a great deal of smoothing, plus considerable judgment, and a fair component of arbitrariness for projecting into the future, one derives gender-specific, total pension incidence functions for each relevant birth-year cohort. These functions, presented as a function of birth year, age, and gender, show the proportion of the population receiving some private pension income, whether own-retirement, survivor, or a combination. The incidence values will serve primarily as alignment targets for the module’s subsequent selection of the specific synthetic individuals to receive pensions. Since, at any point in simulated time, there are both own-retirement and survivor pensions being paid, one must specify the pension types as new

Adding Private Pensions to the DYNACAN Model

149

pensions emerge to become payable. This specification includes the possibility of individuals receiving both types of pensions — and of acquiring them in either order. The specification thus includes estimation of the likelihood of receiving a survivor pension from a deceased spouse who had not yet begun to collect an own-retirement pension — and of commencing an own-retirement pension while already receiving a survivor pension. Because the pension assignment process is dynamic — i.e., focused on new/emerging pensions — it must take account of persons already receiving pensions. Thus, one must begin by assigning appropriate own-retirement and survivor pensions to individuals in the starting database. This ‘‘pumppriming’’ assignment involves assigning the right numbers of private pensions to the right kinds of people and choosing the right types for these pensions — as well as generating pension amounts that are reasonable as regards the recipients’ personal and family characteristics and consistent with semi-aggregate data. This paper’s description also includes a summary of the overall dynamic pension assignment process, even though some details remain to be developed for various parts of it. Key aspects of this description include the dependence of the process on residuals, the extensive use of alignment, and the kinds of validation that will be used to assess the quality of the pensions thus generated.

6. Overall Incidence: How Many? The first step in the module design is the derivation of functions that describe, by gender and birth-year cohort, the total incidence of pension receipt across age. These functions will play a key role in the algorithm for deciding, each simulation year, how many new individuals should receive pensions. The functions will also serve as a key element in establishing the mix of own-retirement and survivor pensions. (For purposes of this initial module, we ignore the existence of other types of pensions, such as disability pensions and survivor benefits payable to children. In a simplification, pensions payable to persons over 25 years of age are treated as being either own-retirement pensions or survivor pensions.) Although these incidence functions are being used here for a first cut module that models pensions without tracking individuals’ lifetime accumulations of pension entitlements, they will also be useful in developing a subsequent, more sophisticated pension module. The pensions source data used in numerators for the pension incidence derivations come from taxation statistics. These statistics are drawn from a large sample of tax returns known to be reasonably comprehensive. A fairly high coverage of the overall population occurs — in part, because pension plans are obliged to report pension income to the government. In addition,

150

Richard J. Morrison

taxation-based measures of income are used to administer various incometested government benefits, thus encouraging Canadians to file returns. The primary taxation data used here consist of counts of private pension recipients, and aggregate amounts of private pension income received, both of them organized by gender and (generally) by five-year age grouping. Data in this format are available annually for taxation years 1981 through 2002. Data from taxation years prior to 1981 do not provide the same level of detail — and the more recent years’ data are still under development. Web-accessible versions of this information (Canada Revenue Agency) are available for recent years. The population data used for the incidence denominators come from ACTUCAN population projections. These population data are very similar to official Statistics Canada population projections, as well as to DYNACAN’s own population projections. One can combine the taxation data on pension recipients (numerators) with Canadian population data (denominators) to generate incidences of pension receipt for the five-year age groupings appearing in the taxation data. Because the definition of the highest, open-ended age grouping changed over time due to sample size considerations, we made adjustments to extract reasonable, internally consistent, sub-group incidences from the collapsed cells. Plotting the resulting age-groupings’ incidences over time yields curves that are reasonably smooth, though with a few unexpected jumps caused by variation in the numerators. For all of the age groupings, the incidences generally first increase over time, but then flatten out and slowly begin to decrease. Because the incidences for years 1971 through 1980 are also required for the simulations, the incidence values for these earlier years were ‘‘back projected’’ — essentially via linear extrapolation from the early portions of the curves from 1981 through 2000. Simple smoothing was then applied to the curves as a crude adjustment for variation within the samples used for the numerators of the incidences. In this type of data, every five years, each five-adjacent-birth-year cohort ‘‘reappears’’, five years older in the data. (That reappearance is subject, of course, to the population variation that occurs through migration and mortality. The variation affects both the numerators and denominators, though not necessarily proportionately.) One can then plot pension incidences as a function of age for each such five-year cohort. The five-year cohorts chosen for analysis ranged from birth-year groupings 1915–1919 through 1934–1938. We limited ourselves to these groups in order to guarantee several ‘‘appearances’’ for each of the five-year cohorts. Subject to the granularity associated with using the mid-point age for each five-year group (e.g., 62 for those aged 60–64), the resulting curves have similar shapes and are surprisingly smooth and consistent. One can then obtain an approximation for the corresponding curves — but now for individual birth years — via interpolation. In this case, for a first pass, the analysis used simple linear

Adding Private Pensions to the DYNACAN Model

151

interpolation. For example, an approximate incidence for those aged 54 years was taken as being 40 per cent of the difference between the incidence value for 50–54 year olds at one point in time and the incidence value for 55–59-year olds five years later. For a 55-year-old individual, the interpolation fraction would have been 60 per cent. Although the method seems to work fairly well, a more sophisticated curvilinear interpolation method might well be investigated for a later pass. After such interpolation, smoothing based on adjacent point values was applied. The resulting cohort-specific incidences appeared to be fairly smooth across age — and generally consistent in shape across birth-year cohorts. Figure 1 shows the pension incidences for a sampling of female birth-year cohorts three years apart. A major goal for the simulation is the emergence of new pensions as each cohort ages. However, the information just derived, essentially cumulative pension incidences, can be presented in terms of the incremental incidences by age for the several birth years. Smoothed once again, these incremental incidences appear to change fairly smoothly and consistently from one birthyear cohort to the next. Figure 2 demonstrates that the incremental incidences are, very roughly, bell-shaped. The curves look somewhat like the normal density curves one which encounters in statistics. However, the modes (modal year for new pension receipt) and standard deviations Figure 1:

Smoothed Pensions Incidence for Selected female Cohorts 0.40 0.35

Pension Incidence

0.30 0.25 0.20 0.15 0.10 0.05 0.00 45

50

55

60

65

70

75

80

Age 1915

1921

1927

1933

1918

1924

1930

1936

1938

Note: The line for the 1938 cohort is the line at the extreme left of the nine curves shown in the figure.

Richard J. Morrison

152 Figure 2:

Smoothed Incremental Pension Incidence by Age for Selected Male Cohorts 0.06

Incremental Incidence

0.05 0.04 0.03 0.02 0.01 0.00 45

50

55

60

65

70

75

80

Age 1916

1922

1928

1919

1925

1931

1934

(dispersion of that receipt by age) — and in this case the areas underneath them (closely related to cumulative pension incidence for the several birthyear cohorts) — all vary fairly consistently by gender and birth year. For each gender/birth-year combination, we estimated a five-parameter curve for the incremental incidences for ages beyond 45. The parameters were the mode and standard deviation for the normal-like component, a scale factor in recognition that the ultimate pension incidences were well less than unity, together with a component that was linear in age, and whose slope and gender-specific intercept were identical for all birth-year cohorts. For a process involving so much in the way of approximations and smoothing, the estimated parameters proved surprisingly well behaved. The modal age for retirement decreased across cohorts as the birth year increased, and was assumed to level off at 57 years of age. The standard deviation for the ‘‘normal’’ component increased (with birth year) for men, and decreased for women. For both men and women, the scale factor related to ultimate pension incidence increased, leveled off, and then decreased. Figure 3 shows, for men and women, the estimated and projected modal values for beginning to receive some private pension. The higher modal ages for women probably reflect the proportion of women’s pensions that are survivor pensions. Given the intended policy usage for the pension module, it is necessary to project the parameters for birth years beyond those for which estimation is feasible from the taxation data. This includes the choice of the parameters

Adding Private Pensions to the DYNACAN Model Figure 3:

153

Modal Age for Pension Start: Estimated and Projected Parameters 66

Modal Age

64 62 60 58 56 1900

1910

1920

1930

1940

1950

1960

Birth Year Men

Women

for cohorts not even born by the end of the taxation sample period. Obviously, such choice will be very much a matter of judgment/assumption. As Figure 3 suggests, for purposes of this prototype pension module, we assumed constant values for all of the parameters, for years both before and after those for which we could reasonably estimate values from the taxation data. We also slightly smoothed the parameters across birth-year cohorts, and somewhat arbitrarily selected the final values, also controlling the rates at which they would be reached. The major effect of these adjustments was to slow down the rate of decrease in the modal retirement age, to reduce the decline in the scale factor that roughly characterizes ultimate pension incidence, and to increase the assumed ultimate steady-state value for that latter, incidence, parameter. All of these adjustments, together with comparisons to the originally estimated values, are available elsewhere (Morrison, 2002b). Implicit in the estimated and adjusted parameters is an assumption about the likelihood that members of the various birth-year cohorts will eventually receive pensions. Using pension incidence at age 70 as a reasonable representation for such incidence, the effect of the assumptions appears as in Figure 4. Beyond that point, pension incidence should continue to increase, though relatively slowly, due to the emergence of additional survivor pensions. Recall that many recipients of private pensions will have been covered for only a portion of their working lives, so that the pension amounts associated with that receipt may be small. As just described, the characterization of overall pension incidence, broken out by gender, birth-year cohort, and individual year of age, involves a

Richard J. Morrison

154 Figure 4:

Pension Incidence at Age 70 0.6

Pension Incidence

0.5 0.4 0.3 0.2 0.1 0.0 1900

1910

1920

1930

1940

1950

1960

Birth Year Men

Women

Figure 5: Sample Validation — Pension Incidences: DYNACAN vs. Smoothed Taxation Data: Males

Pension Incidence

0.75 0.60 0.45 0.30 0.15 0.00 45

50

55

60

65

70

75

80

Age T: 1919-1923 D: 1919-1923

T: 1924-1928 D: 1924-1928

T: 1929-1933 D: 1929-1933

T: 1934-1938 D: 1934-1938

great deal in the way of approximations, interpolation, and smoothing. Thus, we judged it useful, as a crude validation test, to see if the projected parameters roughly reproduced the five-year age grouping pension incidences derived from the taxation data. Figure 5 shows, for selected five-year

Adding Private Pensions to the DYNACAN Model

155

age groupings for men, that, although the fit is certainly not perfect, it does capture the general patterns for the age groups that might legitimately have significant pension income. Points are plotted at the center of the age ranges, e.g., 62 for the age range 60–64 years of age.

7. Estimations for Own and Survival Pension Incidence and Amounts Knowing how many persons should have positive pension income is a good start, but it is only a start. One would obviously not want simply to give random recipients the average pension amount and declare the job done. Rather, DYNACAN should attempt to assign pensions to those likely to be receiving them — and in amounts consistent with their individual circumstances. To accomplish this, DYNACAN will use survey data to estimate probabilities of pension receipt, as well as pension levels as functions of individuals’ characteristics. The general patterns estimated for one point in time will be assumed to hold over time, subject to exogenous alignment regarding the numbers of persons receiving pensions. Projections based on the resulting equations will then inform the ‘‘which ones?’’ and ‘‘how much?’’ decisions in the pension assignments. In view of the data limitations described below, the projections for both receipt and levels will necessarily be crude. There will be a considerable dependence on alignment to provide some adjustment for the anticipated deficiencies. In the absence of a suitable, available, large-sample longitudinal database, the estimations will use a cross-sectional data source. Specifically, the DYNACAN team will use data from Social Development Canada’s SIMTAB cross-sectional policy model. Advantages include convenience; a relatively large survey base; enhancements such as the attribution of pension income known, by comparison to aggregate data, to be missing in major survey and Census data; and an expectation of greater consistency between cross-sectional and longitudinal results. Disadvantages include an imperfect adjustment for differences between the overall population and the source survey’s sampling frame, particularly for those seniors living in institutions. The equation sets to be estimated will generally be based on gender, age group (o50, 50–64, 65+) and, depending on data density, marital status. The team will estimate gender-specific equations for pension receipt and pension amounts, carrying out the estimations separately for survivor pensions and own-retirement pensions. Because the data do not distinguish between the two types, admittedly crude means will be used to try to separate them. The independent variables must include only those available in both DYNACAN and the SIMTAB database — and that also seem likely to have explanatory power. They will include variables such as specific age,

156

Richard J. Morrison

education, the presence of children in the home, and the receipt of income from other sources such as earnings and public pension benefits. We anticipate that the estimation will be challenging because the SIMTAB source data, like all the others, does not include information about the type of pension — and because marital status at the time of the survey may have changed from the status at the time that the pension emerged. Despite these non-trivial limitations, we think it is important to try to assign pension income to synthetic individuals who generally resemble pension recipients in the real world. Own-retirement incidence. Probit equations will be estimated, separately by gender, only for those 50 years of age or older. The approach will treat pensions for those younger than 50 as necessarily being survivors’ pensions. Within these, there will be separate equations by (current) marital status where the sample size is sufficiently large (and with a dummy variable for marital status where it is not). Note, however, that the results of these equations will then generally be used to guide recipient selection within specific ages, rather than across them. That is, alignment targets will apply to individual years of age. Own-retirement amounts. Equations for own-retirement pension amounts will be estimated, separately by gender, only for those 50 years of age or older, based on individual records that report pension income. As with the equations for probability of receipt, sample sizes permitting, there will be separate equations by marital status. However, the relatively smaller fraction of the population receiving pensions will mean that more categories will likely be merged, and dummy variables used. For the married and widowed categories, smaller pension amounts are relatively likely to be survivor pensions. Consequently, as a rough adjustment, cases in which the pensions are in the lower third of pension levels, by gender, will be excluded from the estimation of the equations for amounts, on the grounds that they are ‘‘probably’’ survivors’ pensions. Survivor incidence. Probit equations will be estimated, separately by gender, for the different age groups. Persons 50 years or older with a marital status of single (never-married or divorced) will not be admitted to the estimation because they are relatively unlikely to be receiving survivor pensions. Pensions received by divorced persons will be interpreted as ownretirement pensions resulting from own-employment or divorce settlements. This expectation is reinforced by DYNACAN’s treatment of common-law unions as married. Within the remaining groups, including both widowed and married individuals, where sample size warrants, there will be separate equations by marital status. Where sample sizes would otherwise be unreasonably small, the estimation will employ a dummy variable for marital status.

Adding Private Pensions to the DYNACAN Model

157

Survivor amounts. Equations for survivor pension amounts will be estimated, separately by gender, for different age groupings, based on individual records that report pension income. As with the equations for probability of receipt, there will be separate equations by marital status where reasonable. However, the relatively smaller fraction of the population receiving pensions will mean that more categories will likely be merged and dummy variables used. For the married and widowed categories, larger pension amounts are relatively likely to be own-retirement pensions. Consequently, for these groups, instances in the upper third of pension levels, by gender, will be excluded from the amounts estimation. The equations for survivor amounts are particularly relevant for the addition of survivor pensions to the initial database. Once a simulation is in progress, survivor pension amounts for which the deceased spouse was already receiving an own-retirement pension will be a fraction of that pension. However, the equations for amounts will still be required when the deceased spouse was not yet receiving an own-retirement pension, but the surviving spouse is selected to receive a survivor pension. These ‘‘adjustments,’’ designed to attempt to isolate own-retirement and survivor pensions, are undoubtedly crude. However, in the absence of any public use data source identifying these two types of private pensions at the level of the individual, the use of rough-cut measures is almost inevitable. The team would like to use final version of the Survey of Financial Security data, when it becomes available, and should it prove suitable, to inform the process of estimating survivor and own-retirement pensions. The current expectation is that the next version of the pension module will utilize that source. Should validation analyses indicate that the procedures are not properly simulating private pensions, then the exclusion parameters (i.e., excluding records with the bottom or top thirds of pension amounts), could be altered. More ambitiously, once a functioning pension module is available, one may be able to use its outputs to help ‘‘bootstrap’’ a superior mechanism for estimating and assigning pension types.

8. Mix of Own-Retirement and Survivor Pension Incidences The characterization of overall pension incidence naturally covers the age spectrum, starting at about 25 years of age, and continuing for all higher ages. In Canada, pension incidence for individuals younger than 25 tends to be very low, and is likely to consist primarily of orphans’ benefits that are payable only while children remain in school. As shown earlier in this paper, new pensions tend to occur disproportionately at ages 50 and higher, primarily as a result of individuals’ retirements and of higher mortality rates

158

Richard J. Morrison

among the elderly. For survivor pensions, particularly those that occur because of the death of a spouse who is not yet receiving an own-retirement pension, one needs to look more closely at younger ages. Under current pension legislation, survivors for whom the deceased spouse was already receiving an own-retirement pension will typically receive a survivor pension if the union pre-dated that retirement. As well, incidence figures for survivor pensions will most naturally be defined with respect to the populations of the ever-widowed rather than the general population. Logically, once one excludes orphans’ benefits payable to those under 25, effectively only spousal survivors can be expected to receive survivor pensions. As indicated, we effectively ignore the portion of pension income that corresponds to disability pensions, treating it as own-retirement or survivor, and tolerating for this round such distortions as this treatment may bring to the analysis. DYNACAN’s derivation of survivor pension incidences again uses agegrouped taxation data as the source for pensions received, with the numbers of persons serving as the numerators for the incidences. In the absence of suitable survey data on ever-widowed individuals to serve as the denominators, the analysis uses, for the same age groups, ever-widowed individuals from its own synthetic life histories. Initially rather less than perfect because of the limited marital history information in the model’s starting data from 1971 Census, the quality of the denominator measure improves as simulated time progresses. The initial plan for the estimation of survivor pensions anticipated that each birth-year-cohort’s survivor incidence would be proportional to widow(er)hood. The team assumed that it would also be roughly proportional to the ultimate pension incidence for the opposite gender of the same cohort. However, as Morrison (2002a, 2002b, 2002c) describes, the resulting five-year group incidences did not exhibit a clear pattern in this respect. Instead, the derived pension incidences for persons less than 50 years of age displayed a number of anomalies that discouraged pursuit of that initial plan, especially for males. Significant among those results were the following: 





there were several year/age-group combinations in which the derived incidences for men exceeded those for women, even though the presence of male survivor pensions should generally reflect the lower female pension coverage; for males, the number of persons with pensions sometimes non-negligibly exceeded the projections for the numbers of men who had been widowered. Clearly, not all of these instances could be survivor pensions; especially for males, there was considerable variation in the derived incidence, both across age groups and from year to year. In some age groups, the incidence of pensions dropped sharply from one year to the

Adding Private Pensions to the DYNACAN Model





159

next, even though about 80 per cent of the individuals would have remained constant. Neither were there any clear longitudinal trends in pension receipt as measured against the widower fraction of the population; the calculated incidences for women, though hardly characterizable as extraordinarily well-behaved, seemed much more reasonable than those for men. They were typically non-negligibly greater than zero, and substantially less than the ultimate pension incidences for the males of the corresponding cohorts; and analysis of the taxation-based numerators, and separately of the DYNACAN-based denominators, suggests strongly that the source of the anomalies lies with the taxation data. Possible causes include inconsistencies in filing rates for younger ages, and the inclusion of disability pensions that disproportionately affect males. However, there is no reason to believe that the causes of problems in the male incidences are somehow completely absent in the data for females.

Despite these findings, in the absence of consistent, large-sample data on survivor pension receipt, the team has elected to use the derived incidences, though in a much simpler manner than that originally envisioned. In the absence of pronounced longitudinal patterns of receipt, the team decided to use, at least as a first pass target for alignment, a constant proportion of widows as recipients for survivor pensions. Given that about one-fifth of widows aged 25 through 49 appear to receive pensions, that value was chosen to apply throughout both the historical and projection periods. The cut-off of 50 years for the calculation of this proportion is based on the expectation that beyond age 50, own-retirement pensions will begin to make up a non-trivial portion of pension receipt. The pension incidence data for both men and women support this interpretation. The one-fifth value, though it has a considerable degree of the arbitrary about it, seems generally reasonable, being somewhat less than the ultimate assumption for men’s combined pension receipt by age 70. Given the anticipated relatively small proportion of survivor benefits generated on the death of spouses who have not yet begun to receive own retirement pensions, we judge this approximation to be acceptable. Validation analyses will assess the quality of the assumption. Recall that the context for this step is the determination of how many new survivor pensions should be generated. Subsequent portions of this paper address the question of which particular survivors should receive those pensions, and the selection will depend heavily on the characteristics of the survivors and the deceased spouses. It remains to specify a corresponding value for survivor pensions to be awarded to widowers. Given the several anomalies encountered in the male

160

Richard J. Morrison

incidence data for those aged 25 through 49, the team decided to base the value for widowers on the value derived for widows. Women’s survivor pensions originate with the deaths of husbands who have pension coverage, and vice versa. Consequently, we decided, for simplicity, to multiply the one-fifth value for women’s survivor benefits by the ratio of the ultimate pension coverage of women relative to that of men. This gives a value of 0.108 as the proportion of widowers who should receive survivor pensions from wives who die while not in receipt of an own-retirement pension. The value seems generally reasonable, but obviously also has a good deal of the arbitrary about it. Nonetheless, in view of the anticipated relatively small number of men aged from 25 to 50 who become widowers when wives not in receipt of an own-retirement pension die, we consider it acceptable, particularly in absence of an obvious alternative. Once again, validation studies will attempt to assess the quality of this assumption. The parameters described above for survivor pension receipt depend only on pension receipt reported by those under 50, because, above that age, pension receipt begins to include significant own-retirement pensions. Nonetheless, we shall also apply them to survivors over the age of 50 whose deceased spouses have not yet begun to receive own-retirement pensions. We do so under the explicit assumption that, by age 50, employment patterns for the spouses, including their employment in firms that provide pensions, are likely to be well established, and unlikely to change a great deal over the remainder of their working careers. Note, too, that these survivor pension parameters address part of the fundamental question of ‘‘how many?’’ The assignment of the specific surviving individuals to receive such pensions and their pension amounts will then be matched to the characteristics of those reporting such income. Later sections of this paper will summarize that implementation. In the absence of exogenous data specifically identifiable as survivor pensions, validation will inevitably prove challenging. Nonetheless, a certain degree of validation of survivor pensions forms part of the planning for the pension module. Much of this analysis will assess whether the survivor pensions exhibit, at least qualitatively, the expected tendencies. Some of the validation will assess the algorithm mechanics — i.e., whether the algorithm, once implemented, does generate survivor pensions at the average 0.200 and 0.108 proportions. Validation will also assess, by gender, the proportions of survivor pensions generated in respect of deceased spouses who did and did not, have own-retirement pensions. Other planned validation tests will assess the fractions of emerging survivor pensions, relative to all pensions, by gender and age at emergence. One would expect to see a U-shaped function, with survivor pensions dominating both the low and high ages. Moreover, of the survivor pensions, those emerging at lower ages should come disproportionately from deceased spouses without own-retirement pensions. Those

Adding Private Pensions to the DYNACAN Model

161

at higher ages should flow disproportionately from deceased spouses with own-retirement pensions.

9. Adding Pensions to the Initial Database DYNACAN begins its simulations with its 1971 database, essentially an enhanced version of the 1-per cent public use sample from Canada’s 1971 Census. Although those Census data include reasonable variables for earnings, there is no variable for pensions for any of the three levels of unit of analysis. Consequently, to provide a starting point for the simulation of new/ emerging pensions, and updates to the amounts for existing pensioners, one needs to synthesize pension recipients and income in the starting database. In the initial pass for that addition, because the breakout of income by type and individual is so crude in the Census data, we carry out the synthesis without attempting to constrain it to the total income values reported in the Census for individuals and families. Thus, for example, one might generate a synthetic pension value that, combined with earnings, would exceed the reported total income for the unit. In addition to making the synthesis considerably simpler, this practice allows one to ensure that the synthesized values conceptually include pension income that was not reported in the Census. The process begins with the estimated target incidences for total pensions in 1971, broken out by age and gender. An earlier section of this paper outlines how these are derived. The combination of the target incidence rates and underlying populations imply target numbers of 1971 pension recipients by age and gender. The next logical issue is that of ‘‘which ones?’’ — i.e., the selection of the specific individuals in the starting database to be identified as receiving pensions in 1971. The selection process for pension receipt is handled by DYNACAN’s standard alignment-by-sorting algorithm (see the text box below). Assuming independence (or any specified level of correlation) between the two types of pension receipt, one can calculate, from the pension incidence equations, the probability of any given individual receiving some pension income, from whatever sources. The alignment-by-sorting algorithm uses the calculated probabilities to select appropriate target numbers of pension recipients (independently of pension type). Then, for each individual selected to receive pension income, Bayes’ rule gives the probabilities of the pension being ownretirement only, survivor only, or both. Next, a random draw contingent on the relative sizes of these three possibilities determines the specific combination of pension types that will be assigned to the individual. Once the Survey of Financial Security data on private pension types become available at the level of the individual, the team can use them to inform the assignment of pension types.

162

Richard J. Morrison

Alignment by sorting The alignment algorithm used throughout DYNACAN is a dual-objective procedure. Not only does it adjust the mean event probability of an alignment group to match a desired target proportion, but it also ensures that the desired numbers of events take place, eliminating the stochastic variation often associated with algorithms using randomnumber draws. The algorithm, based on work by Tom Johnson (CORSIM model), is fairly straightforward. For each unit in the alignment group (for which one has already used the relevant equations to calculate the probability for the occurrence of the event for the unit) one draws a random number between 0 and 1.0. The differences between the numbers drawn and the target probabilities, after the application of certain mathematical transformations that prevent any super-probability effects, are recorded. One then orders the events in the alignment pool numerically, based on those recorded differences. For a target number of events, N, from the pool, one implements the aligned results by drawing the first N units from this sorted list. These are then the units for which the event occurs. (Were one to simply implement the event for all the units for which the difference is less than or equal to zero, one would generate naı¨ve, unaligned results for which the number of implemented events might well depart from the target number of events.) The alignment-by-sorting algorithm generates, at an acceptable computational cost, exactly the desired numbers of events, with results that exhibit a variety of desirable statistical properties.

The next step is the assignment of pension amounts for the own-retirement and survivors pensions. For simplicity, in this prototype module one will calculate them independently from the equations for pension amounts, based on the characteristics of the individuals selected to receive them. This assignment will necessarily include an adjustment for the ‘‘pension inflation’’ between the year of assignment (1971) and the year serving as the basis for estimating the equations for pension amounts. The remainder of the process is reasonably simple. One first uses the equations to calculate the total pension income for each individual selected to receive a pension. Next, one imposes, separately by gender, a transformation to force the calculated pensions to fit a target distribution. The transformation respects the relative ordering of individuals by pension amount, but chooses the new pension amount so that an individual at the nth quantile of the pre-transformation distribution receives the pension associated with that same quantile in the target distribution. If an individual is receiving only one type of pension, the transformation completes the process. If the individual is receiving both

Adding Private Pensions to the DYNACAN Model

163

own-retirement and survivor pensions, then both of the equation-generated values are scaled proportionately so that they yield the new desired value. Once the addition of pension income to the 1971 initial database is complete, it will provide a starting point for the subsequent dynamic assignment of pensions. Given the nature of the algorithm for assigning new pensions, as described in the next section, one must proceed from a foundation that includes knowledge about what pensions are already being paid. The initial pension synthesis as described above is not particularly sophisticated. However, we believe, subject to validation, the assignments will be generally reasonable. Fortunately, even if the initial synthesis is in error, most of those pensions will have ceased, due to the death of the recipient, by the time that the longitudinal simulation completes the simulation of the historical period and begins the projection period. More specifically, given the nature of the algorithm, we expect that the negative impacts of our initial errors will be progressively eroded with the passage of simulated time. Thus, we expect that they will not significantly affect the distributions of recipients and amounts by the time that the model’s projection period begins. However, even though the quality of the initial synthesis of pensions may not prove critical, it will be desirable to assess that quality. For this reason the planning for adding pensions to the starting database includes the usual validation component. Although it would be inappropriate here to try to describe all of the planned validation, one can usefully mention the major highlights. (1) One can examine, mechanically, whether the synthesis creates the desired numbers of male and female pensions by age group, and whether the shapes of the distributions of pension amounts by gender match their targets. (2) One can assess whether the distribution of recipients and pension amounts is consistent with a year shortly afterwards, e.g., as measured against the 1974 Survey of Consumer Finances. The survey’s differing sampling framework, and the known underreporting of pension income, will complicate these comparisons. Even though the 1973 pension amounts associated with the 1974 survey are not precisely those of 1971, relatively little evolution will have occurred in the pensioner population. Thus, a reasonable fit would support the assertion that the synthesis in 1971 is acceptable.

10. Characterization of the Dynamic Algorithm for Pensions The preceding sections have outlined some of the key inputs to the simulation of private pensions in DYNACAN. At this point, even though some of the details and much of the estimation remain incomplete, it is useful to characterize how we expect DYNACAN’s dynamic allocation of pensions to proceed. The summary description that follows is organized in terms of the categories of those individuals who will receive new pensions in any

164

Richard J. Morrison

given simulation year. There is the explicit assumption that once individuals begin to receive private pensions, those pensions will remain payable for the rest of the recipients’ lives. Only the amounts will change over subsequent simulated time. Each sub-section, or category, in this section will cover the key questions noted above: ‘‘How many?’’ ‘‘Which ones?’’ and ‘‘How much?’’ The fourth ‘‘What type?’’ characteristic is already inherent in the pension-type category definitions used for the presentation. The first category involves the assignment of new survivor pensions to partners of deceased spouses who were already receiving own-retirement pensions. The ‘‘How many?’’ and ‘‘Which ones?’’ decisions are particularly simple. If the deceased spouse was receiving an own-retirement pension, then the surviving spouse receives a survivor pension. The only exception to such generation will be for marriages that took place after commencement of the deceased spouse’s own-retirement pension; consistent with the provisions of most private pension plans, those survivors will not receive a survivor pension. For simplicity, the first year pension level will reflect the modal plan provision of 60 per cent of deceased spouse’s own-retirement pension. The second category involves the assignment of new survivor pensions in respect of deceased spouses who were not (yet) receiving an own-retirement pension. For the ‘‘how many?’’ question, based on historical data, 20 per cent of such women, and 10.8 per cent of such men will receive a survivor pension. However, certain individuals for whom receipt would be unreasonable (e.g., because the deceased spouse was too old or had a bad work history), will be excluded. For example, if the deceased spouse had reached 70 years of age without receiving an own-retirement pension, then it appears from the bell-shaped curves of pension emergence that s/he is very unlikely to be eligible for one. By means of such exclusions, either on an absolute or relative basis, the probabilities of receiving a survivor pension will be increased for the remaining survivors. For the ‘‘which ones?’’ question, the actual assignments will be made using the description equations for survivor pension receipt, but aligned to the individual birth-year cohort and gender targets. The alignment-by-sorting algorithm permits one to include ‘‘ineligibles’’, such as the example above in the pool but with a zero probability. The algorithm then naturally permits one to apply the overall target, even while excluding certain individuals from pension assignment. Note that the use of age-specific alignment pools prevents the cumulative nature of pension incidence from distorting the selection of which individuals receive these pensions. A subsequent enhancement may adjust the 20 per cent and 10.8 per cent parameters, making t hem cohort-specific to reflect the anticipated ‘‘ultimate’’ pension receipt proportions for specific cohorts rather than using a cross-cohort measure. Pension amounts will be assigned based on the estimated equations

Adding Private Pensions to the DYNACAN Model

165

for survivor pension levels, but adjusted for the ‘‘pension inflation’’ occurring between the period for the estimation, and the year of the assignment. The third category pertains to the assignment of new own-retirement pensions to persons not receiving either own-retirement or survivor pensions. Such assignments will only be made to individuals who are at least 50 years old. The ‘‘how many?’’ question will be determined as a residual. More specifically, the estimates for total pension receipt by cohort, gender, and specific year of age will indicate what proportion of the cohorts should be receiving some type of pension. Against this, the algorithm will deduct the proportions already receiving either form of pension, including survivor pensions that may have been newly assigned in the two preceding categories. The remainder will give the proportions, and thus the numbers of individuals, not presently receiving pensions, for whom new own-retirement pensions are needed. Technically, it could be the case that the number of new pensions to be generated in this way turns out to be negative: i.e., with the survivor pensions assigned, we already have more pensioners than prescribed by the estimated incidence functions. The ‘quick and dirty’ fix for this situation would be to set to zero the number of new own-retirement pensions to be generated in the group. The implementation will generate validation statistics that will enable the team to tell whether this is, at the practical level, a significant problem. For the ‘‘which ones?’’ question in the third category, the actual assignments will be made using the equations for own-retirement pension receipt. These probabilities will then be aligned to the several age and gender targets using DYNACAN’s standard alignment-by-sorting algorithm. Note that setting targets by individual year of age makes the use of stock-based pension equations more palatable. Within the alignment pool for a given age the equations will indicate the relative probabilities of pension receipt. The presence of age-specific alignment targets is expected to minimize cross-age biases that could otherwise arise from the use of incidence data for simulating emergences. That is, although age will be an input variable to the equations, it will be common across the individuals within any given alignment group — and its impact should thus cancel out within that group. As well, investigations are currently underway regarding ways to estimate equations that would better simulate emergence, rather than incidence, probabilities. They involve an independent variable based on the average incidence for the target individual, less the average incidence for an individual with the same characteristics, apart from age, a year earlier. In addition, during the selection of persons to be assigned pensions, the algorithm will record the expected numbers of persons who would have been assigned pensions if the assignments were to be made using only the raw, equation-generated probabilities, in the absence of the targets. The ratio of

166

Richard J. Morrison

the target to the sum of the equation-generated probabilities will later be used as a scaling factor for the assignment of own-retirement pensions in the fourth category (own-retirement pensions for those already receiving survivor pensions). For the ‘‘how much?’’ question, pension amounts will be assigned based on the estimated equations for own-retirement pension levels, adjusted for ‘‘pension inflation’’ occurring between the period for the estimation, and the simulation year of the assignment. The fourth category consists of own-retirement pensions to be assigned to persons 50 or older who are already receiving survivor pensions. Here, since the individuals are already receiving pensions, the cumulative incidence targets provide no help. With respect to the ‘‘How many?’’ question, the targets will be the sum, over survivor pension recipients who do not presently have own-retirement pensions, of the equation-based probabilities. However, these probabilities will first be adjusted by the ratio derived from the third category’s assignments. In this manner, the assignment of new ownretirement pensions to those already receiving survivor pensions will be rendered consistent with that for new stand-alone own-retirement pensions. Such enforcement is required because the probabilities estimated from crosssectional data in a single year will not necessarily be consistent with the rates assumed for other simulation years. The ‘‘which ones?’’ identification will be carried out using the equationgenerated probabilities, again using DYNACAN’s standard alignmentby-sorting algorithm. However, the probabilities will be adjusted so that they better respect emergence patterns, so that new pensions will be awarded to individuals at rates proportional to which new own-retirement pensions could reasonably be expected. The question of ‘‘how much?’’ ownretirement pension each new recipient will receive will again be based on the own-retirement pension amounts equations. These amounts will, once more, require adjustment for the ‘‘pension inflation’’ between the period for the estimation, and the year of the assignment. Still to be specified in the algorithm is the adjustment of pension amounts for persons who began receiving them prior to the current simulation year. The first pass assignment, with an exception for those turning 65, will be to index the previous year’s amounts by a fraction of inflation (e.g., one-half of it). This fraction could be adjusted if one is able to obtain an exogenous indication of the extent to which pensions are indexed for inflation. It might also be adjusted if, after the alignment adjustments described below, too many individuals receive pensions that decline in nominal terms. We recognize that some pensions are more likely than others to be indexed for inflation. However, in the absence in DYNACAN of occupational data about the individuals or their deceased spouses, it does not appear feasible to make informed judgments about which particular pensions to index either more or less than the central tendency.

Adding Private Pensions to the DYNACAN Model

167

Recall that, in the year over year updating of pension amounts, there is an exception for those turning 65. It occurs because most Canadian private pension plans are integrated in some fashion with Canada’s public pension plans. Such integration typically calls for private pension amounts to decline at age 65 to reflect the eligibility of receipt for public benefits (e.g., CPP benefits, that begin to be payable on an unreduced basis at that age). The presence of these declines is apparent in both survey and taxation data. The algorithm’s first pass implementation of this effect will be to generate a new pension amount from the pension level equations for own-retirement pensions. Different equations will be used for those above and below 65 years of age. This ‘‘regeneration at 65’’ approach should typically result in a decrease in pension level, one that will then be maintained, with partial inflation indexing, over subsequent years of pension receipt. In contrast to the situation with own-retirement pensions, there will be no corresponding recalculation for survivor pension amounts. This treatment corresponds to typical plan provisions, as well as avoiding hitting survivors with the double penalty of reductions to their survivor benefits as well as to their ownretirement benefits, penalties that might easily exceed the new government benefits receivable on attaining age 65. Whatever decisions are made about the updating of pension amounts, there remains the final step of aligning the pension levels to the shape of the distribution seen in the SIMTAB data. The process is very similar to that used for aligning earnings amounts to the target distribution of earnings levels. Che´nard (2001) describes the algorithms involved. In briefest summary, the process aligns, separately by gender, individuals’ total pensions, survivor plus own-retirement, to the shape of a target distribution based on the SIMTAB data. The target distribution is, of course, adjusted for ‘‘pension inflation’’ between the time period for the SIMTAB distribution and the year in which the alignment is being performed. In essence the algorithm largely ignores the absolute pension values assigned, dealing almost exclusively with the information about the relative levels of pensions across individuals, supplemented by the parameters of the target distribution. In other words, the alignment forces the desired shape of the distribution. An individual at the nth quantile in terms of unaligned pension amounts gets assigned the pension amount associated with that same quantile in the target distribution. The transformation maintains the ordering of individuals by amount of total pension, but forces the post-alignment distribution of pension amounts to conform to the shape of the target distribution. Although this alignment adjusts the individuals’ pension amounts, it leaves the types of pensions unchanged. If the individual is receiving both own-retirement and survivor pensions, then the levels of the two are adjusted proportionately. The effect of the alignments on both pension receipt and pension amounts is to ensure that the pension totals are realistic — and

168

Richard J. Morrison

that the distribution of pension amounts conforms, separately by gender, to distributions that reflect the mix of pension types and levels. The scaling has the added advantage that it adjusts for any systemic error in the degree of indexing applied. This paper has made it clear that the development of the pension module is very much a work in progress. Although most of the process is based on algorithms currently used elsewhere in DYNACAN, there is always the possibility that the implementation may reveal obstacles not currently foreseen, so that changes might be needed. For example, several of the steps effectively require intermediate values that would be associated with a pass through the population. Given the computational requirements associated with such passes, it will be desirable to implement the algorithm to capture this information in the fewest passes possible. In the extreme, it may be necessary to derive approximations for information that one would otherwise prefer to calculate by a pass through the population. On the methodological level, there are also some caveats associated with using cross-sectional equations of pension incidence to drive the dynamic process of selecting new pension recipients. Given the limitations in the data available, there may be little alternative to this practice, at least for a first pass. Even with its faults, it appears considerably superior to assigning pension receipt randomly across the eligible population. And the age-specific alignment targets ensure that the increasing pension incidence across age will not bias the selection of individuals to be assigned pensions. However, one can always check to ensure that the stocks of pensioners by type, and their distributions, that result from the dynamic allocation, are generally consistent with those in the survey data. At the same time, as noted, there is some consideration being given to the estimation of emergence rate probabilities for new pension receipt from cross-sectional data. If the team can derive these sufficiently quickly, it may prove feasible to use them for the first pass synthesis algorithm. If not, they could be utilized in the second pass module. With a module involving as much experimentation and approximation as this one, it will be critical to conduct validation studies to assess whether the module is behaving reasonably. A fair bit of such validation has been proposed as an integral part of the design process. Descriptions of proposed validation measures appear in several of the DYNACAN papers cited in the bibliography. Although this characterization of desirable validation is too extensive to replicate here, we can summarize a few of the key elements for the validation. (1) One can compare, over the historical period, recipients, incidences, and aggregates to taxation data, including out of sample aggregates for 1971 through 1980. (2) One can examine, for years including both the historical and projection period, time series of incidences and mix of pension types, informed by ACTUCAN’s mix of CPP own-retirement and

Adding Private Pensions to the DYNACAN Model

169

survivor pensions. (3) One can examine the temporal consistency of individual pension amounts, including the frequency of drops in nominal values. The presence of too many drops at ages other than 65 would suggest a need to adjust the parameters or the algorithm for updating pension benefits each simulation year. (4) One can compare the generated pension incomes to those in Census and survey microdata, including those from the 1999 Survey of Financial Security.

11. Conclusions/Prospects This paper has summarized how the DYNACAN team plans to simulate private pension income for its synthetic individuals, including a progress report on work done to date on a prototype version of the module. Although, as reported here, some parts of the analysis are already complete, a non-negligible portion of the work remains. The uncompleted portion, required before programming can begin, includes the statistical estimations for pension incidence and amounts, and some additional work at the methodological level. Part of this requirement flows from the large gaps in the substance and quality of the data available. Another part reflects the limited attention previously given to the simulation of private pensions by dynamic microsimulation modelers — and especially regarding the critical distinction between own-retirement and survivor pensions. Despite these challenges, we believe that some attempt to simulate private pensions is essential if one wishes to conduct meaningful policy studies of government programmes in the context of the broader retirement income system. Based on discussions with other practitioners, we believe the approach described here is generally reasonable a priori. However, the first casualty of battle is typically the battle plan. Thus, the DYNACAN work plan for the pension module includes a commitment to conduct validation analyses and to revise the module based on the findings. Given the underlying goal of examining government programmes in the context of the broader retirement income system, we believe that despite its obvious, and non-negligible, limitations, several of which are described above, using the approach outlined here will be very much preferable to ignoring private pension income. Despite the many limitations, approximations and caveats involved, we expect the quality of pension synthesis to be sufficient to help get most of the synthetic individuals into the right tax brackets. Moreover, the module outlined here is, quite explicitly, in part a planned learning exercise. It is designed to provide a platform and an experience base that will be useful in designing, implementing, and validating a second, more sophisticated version of the DYNACAN private pension module.

170

Richard J. Morrison

References Canada Revenue Agency, Income Statistics (formerly Taxation Statistics), (annual) (available from http://www.cra-arc.gc.ca/agency/stats/). Che´nard, D. (2001). Earnings in DYNACAN: Distribution Alignment Methodology. DYNACAN Working Paper, Social Development Canada, Ottawa, Ontario. Maser, K. and Be´gin, J. (2003). Canada’s Retirement Income Programs: A Statistical Overview (1990– 2000), Statistics Canada. Income Statistics Division, Catalogue No. 74-507-XPE, Ottawa, Ontario, Canada. Morrison, R. (2002a). Who Ya Gonna Call? Paper Presented at ‘‘Modelling Policy in an Ageing Europe: Diversity and Change in the Life Course’’, International Microsimulation Conference, London, January 10–12. Morrison, R. (2002b). God Bless the Child That’s Got His Own: The Incidence of Private Pensions in DYNACAN. DYNACAN Working Paper, Social Development Canada, Ottawa, Ontario. Morrison, R. (2002c). Estimating the Incidence of Survivor Pensions in DYNACAN. DYNACAN Working Paper, Social Development Canada, Ottawa, Ontario. Morrison, R. (2006). DYNACAN, in Gupta, A. and Harding, A. (eds), Modelling Our Future: Population Ageing, Health and Aged Care, International Symposia in Economic Theory and Econometrics, North-Holland, Amsterdam.

Chapter 7

Post-Secondary Education and Training Participation Rates in Australia in the Next 30 Years: A Microsimulation Approach Sandra Roussel Department of Education, Science and Training, Australia

Abstract This chapter describes the development of an Australian dynamic microsimulation model, designed to project participation rates in seven types of (postschool) education and training from 1997 to 2027. Estimates are made of the strength of the influence on education and training participation rates exerted by population growth, the changing age structure of population, future migration patterns, labour force changes, and changes in family and household formation. In contrast to the results from earlier Australian research (where only demographic changes were considered), findings from this microsimulation model indicate that while the proportion of youth in the population will decline over the 30-year period in question (i.e. 1997–2027), population-wide participation rates in particular types of education and training are projected to increase in the range of 7 to 23 per cent.

1. Introduction A number of earlier papers projecting participation rates in education and training in Australia have found that the ageing of the population will lead to lower engagement in education (e.g. Creedy, 1999; Aungles et al., 2000; Peut, 2001). A standard corollary of this finding is that the proportion of government funding devoted to education is expected to decline, particularly as the government comes under increasing pressure to provide more services for the aged (e.g. pensions and health care — see Commonwealth of Australia, 2002). It is important to note, however, that while the direct effect of ageing is to reduce participation in education and training, there may be countervailing International Symposia in Economic Theory and Econometrics, Vol. 15 Copyright r 2007 Elsevier B.V. All rights reserved ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15007-3

172

Sandra Roussel

effects on educational participation via, for example, technological change, rising levels of prior educational attainment, and changes in the structure of the labour force. To account for some of these additional effects, the microsimulation model detailed in this chapter has been developed to analyse the simultaneous impact of demographic, labour force, and educational attainment changes on participation rates in education and training. Our microsimulation model has been designed to project education and training participation rates in Australia over the medium to long term, the primary objective being to analyse participation rates in education and training in Australia over a 30-year period (1997–2027). These rates are projected on the basis of various assumptions about the future population structure, holding constant character-specific participation rates. Seven forms of education and training are measured: study leading to a postschool educational qualification; external training supported by an employer; in-house training; formal training supported by an employer1; all formal training2; on-the-job or informal training; and any form of education or training undertaken.3 A secondary objective is to examine education and training participation rates over the lifetime of birth cohorts. The microsimulation model that has been developed enables this form of exploration, as it simulates panel data from 1997 to 2027 in five-year intervals. The aims of our analysis are modest — and the structure of our microsimulation model reflects this. The popularity of dynamic microsimulation models has grown in the past 10 years and a number of very large-scale models have emerged internationally in this time (for reviews of the field, see Harding, 2000). In comparison, our microsimulation model is more rudimentary. Using sensitivity analysis, we can also estimate the strength of the influence on education and training participation rates exerted by population growth; the changing age structure of population; future migration patterns; labour force changes; and changes in family and household formation. We expect that a number of factors will influence participation rates in post-secondary

1 Formal training supported by an employer comprises in-house and external training supported by an employer. 2 Formal training comprises in-house training, external training supported by an employer, as well as external training not supported by an employer. Formal training is defined as activities which were undertaken in Australia to obtain, maintain, or improve work-related skills, conducted at a designated time and in a structured format. On-the-job training and study leading to an educational qualification are excluded. 3 The relationships between these seven categories of education and training are depicted in Figure 1.

Training Participation Rates in Australia

173

education and training in Australia over the next three decades. Not all, however, can be captured in our microsimulation model. Thus, an important limitation of our analysis is its focus on the distribution effect and the exclusion of any influence of the participation rate effect (see Section 2.2). As such, our analysis is partial and our projections are by no means comprehensive. Nonetheless, we believe there is merit in our analysis since (i) it moves us away from the somewhat simplistic (and misleading) projections based solely on an ageing population and (ii) helps cast light on the range and magnitude of some of the other factors influencing populationwide rates of participation in education and training in the future. While the partial nature of our analysis precludes presentation of any detailed policy implications, our findings do suggest that the potential for diverting funds from education and training to pensions and health care may be much more constrained than the analyses based solely on an ageing of the population have suggested. Further, we expect that many of the factors not included in our projections would generally lead to higher character-specific participation rates in education and training. For example: 

 

The rise in the ‘‘knowledge economy,’’ and the associated increased requirement for workers to obtain and maintain Information and Communication Technology (ICT) skills. The trend towards an increased need to update skills over the life cycle. The improved levels of foundation skills of the population, with better initial skills associated with a greater propensity to undertake further learning throughout the life cycle.

The following section introduces our microsimulation model, with the descriptive name of the Education and Training MicroSimulation (ETMS) model. The next section discusses results from our microsimulation model, where (as noted above) changes in participation rates over time are driven solely by structural changes in the population, which comprise demographic, educational attainment, and labour force changes. The findings are summarised in the last section.4 4 A fuller paper is available from the author, which contains the following appendices. Appendix A details the general microsimulation modelling framework, specifying the advantages as well as the limitations of this form of modelling for socioeconomic analysis. The focus of Appendix B shifts from outlining the general model to the specifications unique to our microsimulation model. The structural changes in the population underlying the base version of the ETMS model is provided in Appendix C, while projected education and training participation rates for specific birth cohorts are provided in Appendix D.

Sandra Roussel

174

2. Education and Training Microsimulation Model The microsimulation model discussed in this chapter — the ETMS model — is a dynamic cross-sectional model (Harding, 1993, p. 22). It is designed to analyse the long-term impact of demographic and socioeconomic population changes on education and training participation rates and to allow for interaction between micro-units over time. The data inputs into this model include the ABS household Survey of Education and Training, 1997 (Australian Bureau of Statistics, 1998) as the original micro-unit data set and various other sources of ABS micro- and macro-level data. These inputs are used to project the demographic and socioeconomic structure of the population over the 30-year period from 1997 to 2027. 2.1 Objective The objective of the chapter is to provide a general picture of education and training participation rates across population groups and, in particular, over the lifetime of cohorts. The study presents changes in participation rates in seven categories of education and training between 1997 and 2027. These seven categories correspond to the ABS definitions utilised in the publication Education and Training Experience, Australia, 1997 (ABS, 1998) and comprise: 



Formal education — study done in the 1996 calendar year with the intention of obtaining a higher degree, postgraduate diploma, bachelor degree, undergraduate diploma, associate diploma, skilled vocational qualification, basic vocational qualification, or any other qualification. It is not necessary to have completed the course or to have been awarded the qualification. Employer-supported external training — an external formal training course undertaken in the last 12 months prior to the 1997 Survey of Education and Training (SET) survey in which the employer provided at least one of the following types of financial support: – – – – –



paid for fees, paid for study materials, provided paid study leave, paid accommodation or travel expenses, and/or other support.

In-house training — a training course completed while working in the 12 months prior to the survey, which the respondent considered was mainly attended by people working for his or her employer or business. Study for a formal educational qualification (see above) was excluded.

Training Participation Rates in Australia 





Employer-supported formal training — refers to whether a respondent has undertaken an in-house training course and/or an employer-supported external training course in the 12 months prior to the survey. Formal training — a training course undertaken in the 12 months prior to the survey. Training course defines activities that were undertaken in Australia to obtain, maintain, or improve work-related skills, conducted at a designated time and in a structured format. This form of training comprises in-house training, external training courses supported by an employer, and external training courses not supported by an employer. On-the-job training and study for a formal educational qualification are excluded. On-the-job training — relates to respondents who were employed at the time of the survey, or who had worked for a wage or salary in the 12 months prior to the survey, and who indicated they had undertaken at least one of the following activities to improve job skills, while working in any job, in the last 12 months: – – – – –



175

asking questions of co-workers or colleagues, teaching oneself, being shown how to do your job, watching others work, and/or other activities. On-the-job training excludes any training that occurred as a formal training course, or study for a formal educational qualification.

Some education or training undertaken — refers to whether a respondent has undertaken at least one of the following types of education or training: – study for an educational qualification in 1996 calendar year, – a formal training course in the 12 months prior to the 1997 survey, or – on-the-job (informal) training in the 12 months prior to the 1997 survey.

Where none of these categories of education and training was undertaken, the respondent was classified under ‘‘no education or training undertaken.’’ The focus in this chapter is on statistically summarising the changes over time in participation in these forms of education and training for subcategories of the population — in particular, by gender and age groups. The relationships between these seven categories of education and training are depicted in Figure 1. By comparing alternative simulations — i.e. using sensitivity analysis — we can also identify the strength of the growth or decline in education and training participation rates caused by population growth, the changing age structure of population, labour force changes, changes in family and household formation, and the like. The participation rates predicted by the model combine all of these population changes.

Sandra Roussel

176 Figure 1:

Categories of Education and Training Some education or training undertaken

Formal education and training

Formal education

Formal training

Employer-supported formal training

Employer-supported external training

On-the-job training

External training, not supported by employer

In-house training

A supplementary aim of the chapter is to examine the lifetime participation profiles of different types of individuals over time. Much of the literature examining the uneven distribution of education and training participation rates across the population makes use of cross-sectional data. However, point-in-time measures of inequality do not adequately represent inequality over the lifetime of individuals. Since the available evidence (Roussel, 2000, among others) indicates that initial education and continuing training are complements, and not substitutes, cross-sectional analysis is likely to understate inequality over the lifetime. In such cases, analysis of longitudinal data would provide clearer insight into the lifetime distribution of education and training participation rates.5

2.2 Factors Influencing Education and Training Participation Rates Changes in population-wide participation rates in education and training are driven by a combination of two effects. These are 5

Albeit, in this case, a synthetic longitudinal data set.

Training Participation Rates in Australia

177

– a distribution effect — the population-wide participation rate can change due to a change in the distribution or structure of the population. For example, if the proportion of the population that is most likely to participate in formal education (generally those aged 15–24 years) declines, then the population-wide participation rate in formal education will also decline, with all else held constant, and – a participation rate effect — the population-wide participation rate can also change due to a change in character-specific (e.g. age, gender, and the like) participation rates. For example, suppose for males aged 20–25 years the participation rate in formal education changes from 20 to 25 per cent. All else equal, this will increase the population-wide participation rate in formal education. With microsimulation modelling, both of these effects can be captured; alternatively, each effect can also be isolated. The distribution effect can be determined by applying the character-specific participation rates in education and training that held in 1997 to our simulated population structure of the future. Consequently, if the distribution effect is determined in isolation, changes in population-wide participation rates in education and training are driven solely by the changed structure of the population and not by any change in the character-specific participation rates. In contrast, when changes in the character-specific participation rates are applied to the population structure that held in 1997, the participation rate effect is isolated.6 This version of the ETMS model omits changes in character-specific participation rates, and thus only calculates the distribution effect of the impact on population-wide education and training participation rates driven by changes in the population structure. Analysis of the participation rate effect is not undertaken, largely because of the inadequate time series data of character-specific participation rates in education and, particularly, in training. The reader should therefore note that by omitting the participation rate effect our model does not account for a number of factors expected to influence character-specific participation rates in post-secondary education and training in Australia over the next three decades. These include, among others: 

6

The increased requirements for ICT skills (a general increase in the ICT skills required for undertaking one’s job will likely increase the overall need for workers to obtain and maintain their ICT skills).

Note that the distribution and participation rate effect cannot necessarily be added together to achieve the combined impact on education and training participation rates. The total impact can also include a multiplicative component.

178 







Sandra Roussel

The increased onus on individuals to update their skills over the lifecycle (the rate of technological and structural change is making it increasingly necessary for those wishing to maintain their employability to update their skills at strategic stages in the lifecycle). The changing relationship between work and study (in response to increased trends towards combining work and study, institutions are moving towards more flexibly delivered programmes. These demand and supply changes will likely alter the characteristics of those participating in education and training while undertaking work). The second-order impact of the changing age structure of students (a smaller youth population will mean fewer youths entering the labour force in the future. This will place greater demands on existing workers to retrain and/or up-grade their skills and alter the propensities of individuals to participate in education and training according to age. It will also likely have an impact on employer support for training, as employers will need to draw their skill requirements from a proportionately smaller and older workforce). A rising level of educational attainment of the working age population (given the complementarity between initial education and continuing training, a better-educated workforce will, as it ages, likely lead to higher training participation rates among older workers — who will generally be better educated than their parents were at the same age).

Another omission, expected to impact on future education and training participation rates, comes under the umbrella of a distribution effect. In our model we do not update the individual’s industry or occupation of employment, or their wage level. Yet there is an expectation of greater growth of higher-skilled jobs in the future, resulting in a greater proportion of individuals employed as professionals, managers and para-professionals, and in certain high-skill industries. Similarly, a changing distribution of wages across the population can be expected to have an impact on training participation rates. The omissions are important, as earlier research indicates participation rates vary notably by the individual’s industry and occupation of employment, as well as by wage levels (Roussel, 2000). While some of these omissions could be addressed in future iterations of the microsimulation model, it would be difficult to capture unexpected economic, political, and social changes that are likely to occur over such a long horizon — and which could have a significant impact on the population structure and education and training participation rates of the future. It follows from this that the projections of education and training participation rates over the period 1997–2027 provided by our microsimulation model can in no way be considered comprehensive. Nonetheless, we believe there is merit in this exercise and in reporting our findings since (i) it moves

Training Participation Rates in Australia

179

us away from the somewhat simplistic (and misleading) projections based solely on an ageing population and (ii) helps cast light on the range and force of some of the other factors that will influence population-wide rates of participation in education and training in the future. 2.3 ETMS Model Results In this section, we first present participation rates in education and training for our projected populations over the period 1997–2027 using a ‘‘base case’’ scenario. Second, we present the influence of different population assumptions (such as the changing age structure, changes in migration patterns, increasing educational attainment, labour force changes, and the like) on education and training participation rates. This entails running the microsimulation model a number of times, altering one of the assumptions each time. Lastly, using a measure of relative dispersion, we compare the strength of the influence exerted by each factor on population-wide education and training participation rates. Usually, before presenting results, one would seek to validate the ‘‘accuracy’’ of one’s model. In microsimulation modelling, however, there is no test per se for checking the accuracy of the model. What we can do is evaluate the model via two indirect methods. First, as our projected participation rates are based on an Australian population adjusted for demographic and socioeconomic structural changes, we can evaluate the distribution of the simulated populations to see if this is in accordance with our a priori expectations or, equivalently, with the assumptions that we have entered into the model. To summarise this evaluation, there is a close match between our expectations and the changes that occur in the distribution of our simulated populations down to 2027. A second evaluation step is to examine how (a) the actual participation rates for the 1997 SET population differ from (b) the education and training participation rates we predict from the logistic regression model based on the same 1997 SET population. In effect this is an evaluation of how well the logistic regression models (which predict participation rates for each of the seven types of education and training) fit the sampled population. This evaluation is presented in Table 1 and also indicates a close match between the actual participation rates from the 1997 SET-sampled population and those estimated by the logistic regression models. 2.4 Evaluation of 1997 Participation Rates, Actual and Predicted As noted earlier, there are seven participation rates tracked for the purposes of our project. For each of these participation rates, we estimate a logistic regression model. The logistic regression model seeks to quantify the extent

Sandra Roussel

180 Table 1:

Participation Rates in Education and Training, 1997

Participation in: Formal education Employer-supported external training In-house training Employer-supported formal training Formal training On-the-job training (Some Education or Training Undertaken) SETU

Actual (%)

Predicted (%)

Percentage difference (%)

13.7 9.3

13.6 9.3

0.1 0.0

26.7 32.1 40.6 64.4 74.5

26.6 32.0 40.5 64.3 74.5

0.1 0.1 0.1 0.1 0.0

Source: Actual participation rates are based on 1997 ABS Survey of Education and Training. Predicted participation rates are based on the author’s calculations using logistic regression models

of the influence of various (individual-specific) demographic and socioeconomic characteristics on participation rates in these forms of education and training. We evaluate the precision of these regressions by comparing the actual participation rates of the 1997 SET-sampled population with those we predict from applying the logistic regression coefficients to the same 1997 population sample. The results indicate a close match between actual and predicted participation rates, as depicted below. The largest discrepancy is less than onetenth of a per cent, or a 0.7 per cent difference (for formal education). The predicted participation rates also closely follow the age-participation pattern as set by the actual participation rates for the 1997 sample population (Roussel, 2003). The only notable discrepancy between actual and predicted participation rates is in on-the-job training and some education or training undertaken for individuals aged 60–64 years.

2.5 Participation Rates in Education and Training: Base Model The primary objective of the chapter is to project participation rates in education and training out to 2027. With reference to the distribution effect alone, we find that participation rates in each form of education and training are projected to rise between 1997 and 2027 for the population as a whole (Table 2). The greatest growth rate is for employer-supported external training, at 23 per cent (2.2 percentage points); the least is for some education or training undertaken, at 7 per cent (which rounds up to a 5.2 percentage points increase).

Training Participation Rates in Australia Table 2:

181

Participation Rates in Education and Training, 1997 and 2027

Participation in:

Formal education Employersupported external training In-house training Employersupported formal training Formal training On-the-job training SETU

1997 Actual (%)

1997 Predicted (%)

2027 Predicted (%)

Growth rate in predicted participation, 1997–2027 (%)

13.7 9.3

13.6 9.3

15.8 11.5

16 23

26.7 32.1

26.6 32.0

29.4 36.0

11 12

40.6 64.4 74.5

40.5 64.3 74.5

45.7 70.2 79.6

13 9 7

Source: Actual participation rates are based on 1997 ABS Survey of Education and Training unit record file. Predicted participation rates are based on the author’s calculations using logistic regression models as applied to (a) the 1997 SET population and (b) the 2027 population structure simulated by the ETMS model

It is important to recall these participation rate changes have been driven solely by the changing structure of the Australian population (i.e. the distribution effect) and not by changes in character-specific participation rates (i.e. the participation rate effect). As such, the growth in participation rates in Table 2 indicates that potential future structural changes in the population will indeed have an impact on the overall proportions participating in education and training, holding current character-specific participation patterns constant. We also find there are no sudden shifts in either the underlying population structure (Roussel, 2003) or in the predicted education and training participation rates in the future.7 There is a strong gender differential to our projections. We find, for each form of education and training, females are projected to have a higher increase in participation rates — in terms of both growth rate and absolute participation rate levels (see Table 3). This would accord with the increased female labour market participation rate over time present in the Treasury projections, as well as the declining incidence of females with dependent 7

Although we do not present the findings for each five-year period, examination of our findings suggest there has been a steady movement of changes in the underlying population structure from 1997 to 2027.

Sandra Roussel

182 Table 3:

Participation Rates, 1997 and 2027, by Gender

Participation in:

Formal education Employer-supported external In-house training Employer-supported formal Formal training On-the-job training SETU

Males

Females

1997 (%)

2027 (%)

Change, 1997–2027 (%)

1997 (%)

2027 (%)

Change, 1997–2027 (%)

12.7 8.9

14.3 10.2

12 14

14.7 9.8

17.7 13.0

21 33

26.2 31.5

27.6 33.6

5 6

27.0 32.6

31.7 38.9

17 19

40.2 65.8 75.6

43.6 70.1 79.2

9 6 5

41.0 62.6 73.1

48.2 70.4 80.2

18 12 10

Source: Actual participation rates are based on 1997 ABS Survey of Education and Training unit record file. Predicted participation rates are based on the author’s calculations using logistic regression models as applied to (a) the 1997 SET population and (b) the 2027 population structure simulated by the ETMS model

children (using ABS projections).8 Based on past trends, females are also projected to have higher increases in their level of educational attainment — with educational attainment strongly correlated with further participation in education and training. These projected increases in predicted participation rates between 1997 and 2027 also hold over the different age groups. The increase in formal education participation rates is projected to be somewhat greater for youth (see Figure 2). In contrast the increases in employer-supported external training, in-house training, employer-supported formal training, and formal training participation rates are projected to be somewhat greater for middleaged as compared to older workers. Growth in rates of on-the-job training and some education or training undertaken, notably, are greater for older workers (see Figure 3). The differences between the predicted participation rates in 1997 and 2027 for the various age groups confirm expectations. These are (a) participation outcomes differ by age groups and (b) structural changes in the population will impact upon the education and training participation rates for some age groups more than others. Recall that these participation rate changes are

8

Treasury projections refer to projections obtained from the Retirement Income Modelling (RIM) Task Force, Commonwealth Department of the Treasury, Australia. These draw on their Labour Force Status Model (LFSMOD).

Training Participation Rates in Australia

183

Figure 2: Predicted Participation Rates in Post-Secondary Education and Training, by Age, 1997 and 2027 100% Some education or training undertaken

- - -1997

2027

Participation rates

80% 60%

On-the-job training

40% Formal education

20% 0%

15-19 20-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64

Source: Predicted participation rates are based on the author’s calculations using logistic regression models as applied to (a) the 1997 SET population and (b) the 2027 population structure simulated by the ETMS model

Figure 3: and 2027

Predicted Participation Rates in Post-Secondary Training, by Age, 1997

60%

- - -1997

2027

Formal training

Participation rates

50% 40%

Employer-supported formal training

30% In-house training

20% 10% Employer-supported external training

0% 15-19 20-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64

Source: Predicted participation rates are based on the author’s calculations using logistic regression models as applied to (a) the 1997 SET population and (b) the 2027 population structure simulated by the ETMS model

184

Sandra Roussel

driven solely by expected structural changes in the Australian population. As we discuss in the next section, some of the more influential population changes on participation in education and training are employment status, ageing of the population, and level of educational attainment. Over the 30 years examined, changes in these population characteristics are projected to occur at different rates across the population and correspondingly result in different changes in predicted participation rates for particular sub-groups. An additional benefit of dynamic microsimulation modelling is the generation of simulated panel data. Consequently, we are also able to undertake a longitudinal analysis of predicted participation rates of particular cohorts over time. Two trends are evident from our analysis of the simulated panel data. First, and somewhat as expected given we apply the age-specific participation rate patterns that held in 1997, participation rates retain the current ageparticipation rate profile. Specifically, (a) the downward-stepping decline in participation in formal education and (b) the gradual decline in participation in on-the-job training and some education or training undertaken. Second, our projections indicate participation rates increment up with each new cohort: at any particular age, a younger cohort vis-a`-vis an older cohort experiences a higher average participation rate in education and training. Again, this is not unexpected given the increase in education and training participation rates between 1997 and 2027 for the population as a whole (see Table 2). 2.6 Impact of Alternative Scenarios Analysis of only one set of the simulated populations provides limited insights into changes in education and training participation rates over the next 30 years. To address this concern, we also estimate changes in education and training participation rates which derive from a number of other plausible population simulations. Discussion of the education and training participation rates that are generated by these alternative projected population scenarios follows. Demographic Scenarios We draw on three alternative population size projections from the ABS: Series 1, 2, and 3.9 Series 1 represents a population with a relatively high fertility rate of 1.75 births per woman and 110,000 net overseas migrants arriving annually (resulting in the highest estimates of population growth). Series 2 represents a relatively low fertility rate of 1.6 births per woman and 90,000 annual net overseas migrants, while Series 3 represents a population 9

As labelled in the ABS publication Population projections, Australia, 1999– 2101, Cat. No. 3222.0.

Training Participation Rates in Australia

185

with the same relatively low fertility rate, but coupled with an even lower annual net overseas migration of 70,000 persons (resulting in the lowest of the three estimates of population growth). Growth occurs in most of the age groups, with the exception of the slight decline in the absolute number of 15—29-year olds in Series 3. Sizable increases are projected in the number of older Australians, particularly 50–64-year olds. However, in terms of differences between the three series, it is the variance in the projected number of youth (and to a lesser extent, the middle-aged) that is of note. Between Series 1, 2, and 3, relatively marginal differences in the number of older persons are projected. This is not surprising, given that the bulk of our sample in 2027 was alive at the time of the 1997 survey and is thus only affected by assumptions of mortality and migration. The population under the age of 30, however, is additionally affected by current and future fertility rates; the projections, accordingly, reveal significantly greater differences between the three series. It is the notable difference in the number of youth in Series 1 that underlies the differential growth rates in predicted Formal education participation rates (refer to Figure 4). With Series 1 representing a population with a relatively higher proportion of youth vis-a`-vis Series 2 or 3, it is not surprising that the greatest increase in formal education participation rates is projected to occur for this Series (given that formal education is an activity predominantly undertaken by youth). Series 3 represents a population with a relatively larger middle-age population vis-a`-vis Series 2 and Series 1; this

Growth rate in participation rates

Figure 4: Predicted Participation Rates 1997–2027, based on Alternative Population Structure Scenarios (ABS Series 1–3) 25% ABS Series 1 ABS Series 2

20%

ABS Series 3 15%

10%

5%

0% Formal education

Employersupported external training

In-house training

Employersupported formal training

Formal training

On-the-job training

SETU

Source: Growth rates in predicted participation rates are based on the author’s calculations using logistic regression models as applied to (a) the 1997 SET population and (b) the 2027 population structure simulated by the ETMS model

186

Sandra Roussel

drives the slightly higher growth rates in employer-supported external training, in-house training, employer-supported formal training, and formal training participation rates for Series 3, as depicted in Figure 4. Our ageparticipation rate profile for some education or training undertaken (which predominately captures participation in on-the-job training) indicates participation gradually declines with age. While the difference in participation rates in on-the-job training and some education or training undertaken between the three series is marginal, we do find evidence of this trend, with the relatively younger population of Series 1 providing a slightly higher growth rate in predicted participation rates vis-a`-vis Series 2 and, in turn, Series 3. The ABS documents three household and family formation projections: Series A, B, and C.10 In our modified version of Series A, the household and family formation propensities that held in the 1997 SET are held constant from 1998 down to 2027.11 For Series B and C, the growth rate in household and family formation propensities (of being single, forming part of a couple, having a child under 5 years of age present in the home, etc.) is based on the 1986, 1991, and 1996 census data.12 Series B represents a low rate of change, with the full 1986–1996 growth rate applied for projections from 1997 to 2002; half of the 1986–1996 growth rate is applied from 2003 to 2007; and a quarter of the 1986–1996 growth rate from 2008 to 2012.13 From 2013 out to 2027, the propensities that held in 2012 apply. A greater rate of change is captured in Series C, with the full average annual growth rate that held from 1986 to 1996 applied to the entire 1997–2027 period. In 1997, the proportion of the population aged 15–64 years forming part of a couple was 59.5 per cent and 12.5 per cent were single. The remaining 28 per cent of 15–64-year olds were assigned as ‘‘other household.’’ In Series A, with the propensities that existed in the 1997 SET remaining constant over the 1997–2027 projection period, the proportion who form part of a couple is projected to rise to 62 per cent by 2027, while the single proportion is

10

Our labelling of the three series corresponds to that used in the ABS publication Household and Family Projections, Australia, 1996– 2021, Cat. No. 3236.0. 11 For example, the proportion of singles in the projected populations from 1998 to 2027 is kept at the same proportion of singles that existed in the 1997 Australian population. 12 Persons classified as ‘‘other household’’ include non-dependent children or extended family members living in the household; those living in group households; and the like. 13 Note that the growth rate in the household and family formation propensities is based on the 1986, 1991, and 1996 censuses. However, in our modified version of the trend, the base proportion of the population in different household and family formation states is drawn from the 1997 ABS Survey of Education and Training (SET).

Training Participation Rates in Australia

187

projected to rise from 12.5 to 13.0 per cent. The proportion of persons in ‘‘other household’’ in Series A is projected to fall from the 1997 figure of 28 per cent, down to 25 per cent. The 2027 projected outcomes for the full rate of change scenario (or Series C) are for: the proportion forming part of a couple to decline to less than 50 per cent; the proportion expected to be single to rise to 20.7 per cent; and for the remaining 30 per cent to be classified as ‘‘other household.’’ Series B is the mid-level scenario, and projects only a 2 percentage point decline in the proportion forming part of a couple from 1997 and a 3 percentage point rise in the proportion of singles from 1997 to 2027. Series A, B, and C not only provide alternative household formation scenarios (couple, single, and ‘‘other household’’) but also alternative family formation scenarios. Three categories are formed: youngest dependent child aged 0–4 years present in home; youngest dependent child aged 5–14 years present in the home; and no dependent children present in the home. The proportion of persons with a dependent child (aged 0–14 years) in the home declines for all three series. Series A represents a projected population with the highest proportion of individuals with a child aged 0–14 years present in the home, while Series C represents a projected population with the lowest proportion. It is Series C — with the lowest proportion of the 15–64-year-old population forming part of a couple and the lowest proportion of households with a dependent child present — that is projected to produce the highest growth in formal education participation rates (see Figure 5). This also holds for

Growth rate in participation rates

Figure 5: Predicted Participation Rates 1997–2027, based on Alternative Household and Family Formation Scenarios (ABS Series A–C) 25% ABS Series A ABS Series B

20%

ABS Series C

15%

10%

5%

0% Formal education

Employersupported external training

In-house training

Employersupported formal training

Formal training

On-the-job training

SETU

Source: Growth rates in predicted participation rates are based on the author’s calculations using logistic regression models as applied to (a) the 1997 SET population and (b) the 2027 population structure simulated by the ETMS model

188

Sandra Roussel

on-the-job training and some education or training undertaken, although, we find the differential for these latter two forms of training to be much smaller. Conversely, it is Series A — with the highest projected proportion of the 15–64-year-old population forming part of a couple and the highest proportion of households with a dependent child present — that is projected to produce the highest growth in employer-supported external training, in-house training, employer-supported formal training, and formal training participation rates. These results are fairly intuitive. First, undertaking formal education is often more time intensive in nature than undertaking formal training courses. Second, formal training is strongly associated with employment status. Couples and, in particular, couples with children, often are associated with more ‘‘stable’’ full-time employment outcomes and are more likely to be time poor, and are thus also associated with, respectively, a higher likelihood of undertaking formal training and a lower likelihood of undertaking formal education. Educational Attainment Scenarios For the ETMS model, projections of educational attainment are extrapolated from the historical trends collected from the 1986, 1991, and 1996 Census records. The growth rates found in the historical trends are applied to the levels of educational attainment that held in the 1997 SET. We create three alternative projections from the educational attainment propensities that held in these three Censuses: 





Series 1 — the full growth rate that held from 1986 to 1996 is applied to the 1997 SET for the 1998–2027 period (rate of increase ¼ 1); Series 2 — the first period from 1998 to 2002 applies the full growth rate that held from 1986 to 1996, then for each successive five-year period onehalf of the growth rate from the five-year period prior is applied (rate of increase ¼ 1/2); and Series 3 — the first period from 1998 to 2002, again, applies the full growth rate that held from 1986 to 1996, then for each successive five-year period one-third of the growth rate from the five-year period prior is applied (rate of increase ¼ 1/3).

These three series result in three alternative projections of the population, each with different levels of highest educational attainment. For Series 1 (rate of increase ¼ 1), the greatest changes in educational attainment are predicted, with the proportion of the population holding a higher education (HE) qualification projected to rise dramatically from 15 per cent in 1997 to almost 40 per cent in 2027 (see Figure 6). This series sees the proportion of the population with no post-school qualification decline from 51 to 39 per cent over the same period.

Training Participation Rates in Australia

189

Figure 6: Projected Proportion of Population, by Alternative Highest Education Attainment Scenarios, 1997 and 2027 60% Proportion of population

1997 proportion

Rate of increase = 1

Rate of increase = 1/2

Rate of increase = 1/3

50% 40% 30% 20% 10% 0% No post-school qualification

VET qualification Highest educational attainment

HE qualification

Source: 1997 data from the 1997 SET unit record file; 2027 data simulated by the ETMS model

Series 1 (rate of increase ¼ 1) translates into a rise in average years of schooling from 11.9 years in 1997 to 13.5 years in 2027.14 With Series 2 and 3, more moderate increases in educational attainment are projected, with the proportion of the population with a HE qualification projected to rise to approximately 25 per cent in both series, and the proportion without a postschool qualification projected to decline from 51 per cent in 1997 to 45 per cent. Average years of schooling for these two series are projected to increase to approximately 12.9 years. Series 3 has a slightly poorer outcome vis-a`-vis Series 2 in terms of the proportion of the population with post-school qualifications. For all three series, the proportion of the population holding a Vocational Education and Training (VET) qualification as their highest educational qualification is projected to decline from 1997 to 2027. However, caution should be exercised when interpreting these proportions as we are only accounting for highest educational attainment. Thus, individuals holding a HE qualification

14

The years of schooling associated with each level of highest educational attainment are: Never attended high school, 6; did not complete highest level of secondary school and left school at 13/14 years of age, 9; did not complete highest level of secondary school and left school at 15/16 years of age, 10; did not complete highest level of secondary school and left school at 17 years of age or over, 11; completed secondary school, 12; basic vocational qualification, 11; skilled vocational qualification, 12; associate diploma, 14; undergraduate diploma, 15; bachelor degree, 16; postgraduate diploma, 17; higher degree, 18.

Sandra Roussel

190

Growth rate in participation rates

Figure 7: Growth in Predicted Participation Rates 1997–2027, based on Alternative Educational Attainment Scenarios 40%

Rate of increase = 1

Rate of increase = 1/2

Rate of increase = 1/3

35% 30% 25% 20% 15% 10% 5% 0% Formal education

Employersupported external training

In-house training

Employersupported formal training

Formal training

On-the-job training

SETU

Source: Growth rates in predicted participation rates are based on the author’s calculations using logistic regression models as applied to (a) the 1997 SET population and (b) the 2027 population structure simulated by the ETMS model

may have also obtained a VET qualification but will only be counted under the HE qualification category. Consequently, this measure may not accurately record the number of individuals holding a VET qualification. Our logistic regression models and earlier research findings indicate individuals with higher educational attainment are more likely to participate in all forms of education and training.15 It is thus not surprising to find the highest growth in education and training participation rates between 1997 and 2027 is projected for Series 1 (the series projecting the highest growth in educational attainment). This is true for all seven education and training measures (see Figure 7). The differential in growth rates between our three alternative scenarios is of particular note in the case of employer-supported external training, indicating a strong bias on the part of employers to focus their support of training towards those individuals with higher levels of educational attainment. We also vary our proportion of the 15–19-year-old population expected to complete upper secondary school (or Year 12). Again, we examine three alternative scenarios that replicate the principles applied in the educational attainment scenarios. That is Series 1 represents the full rate of change of the

15

Details of the logistic regression models are available from the author on request. Earlier research findings referred to are summarised in Roussel (2000).

Training Participation Rates in Australia

191

historical linear trend, Series 2 uses one-half the rate of change from the fiveyear period prior, and Series 3 uses only one-third the rate of change from the five-year period prior. The historical linear trend in this case is estimated using the 1989–1997 Year 12 and Year 11 retention rates.16 Projecting the proportion of the future 15–64-year-old population that will complete secondary school is not only influenced by changing the proportion of our projected 15–19-year olds of the future, who will complete secondary school, but also by the presence of older cohorts in the sample with a historically lower likelihood of completing secondary school. As these older cohorts age and are removed from the sample, the average likelihood of the in-sample population having completed secondary school will naturally rise. For Series 1, representing the full rate of change scenario, the proportion of the population with completed secondary school rises from 47.6 to 79.0 per cent. Smaller increases in the proportion with completed secondary school are projected for Series 2 and 3, with projected rises of 75.9 and 75.2 per cent, respectively. Of the three alternative scenarios, it is Series 1 (representing the full rate of change) which projects the largest proportion of the population with highest educational attainment of Year 12 and the smallest proportion which did not complete Year 12. Series 3 (representing the smallest rate of increase) results in the poorest outcome, with a smaller proportion of individuals holding a Year 12 qualification as their highest educational attainment and the largest proportion which neither completed higher secondary school nor hold a post-school qualification. Again, given the well-documented evidence that individuals with a higher level of educational attainment are more likely to participate in the various forms of education and training, we are not surprised to find higher growth in participation rates between 1997 and 2027 for Series 1 compared with Series 2, and for Series 2 compared with Series 3 (see Figure 8). One peculiar exception is for employer-supported external training where Series 2, not Series 3, provides the lowest growth in participation rates. A possible explanation is the variability that exists in the microsimulation process itself, this arises from the random selection of individuals chosen for given events. Labour Force Scenarios A number of labour force characteristics are used in our analysis, each extrapolated out to 2027 based on treasury projections.17 We undertook an 16

The historical linear trend is drawn from the ABS Schools, Australia (Cat. No. 4220.0), for 1989–997. 17 Treasury projections refer to projections obtained from the Retirement Income Modelling (RIM) Task Force, Commonwealth Department of the Treasury, Australia. Specifically, drawing on their Labour Force Status Model (LFSMOD).

Sandra Roussel

192

Growth rate in participation rates

Figure 8: Predicted Participation Rates 1997–2027, based on Alternative Secondary School Completion Scenarios (Rate of increase: 1, 1/2, 1/3) 25% Rate of increase = 1

Rate of increase = 1/2

Rate of increase = 1/3

20%

15%

10%

5%

0% Formal education

Employersupported external training

In-house training

Employersupported formal training

Formal training

On-the-job training

SETU

Source: Growth rates in predicted participation rates are based on the author’s calculations using logistic regression models as applied to (a) the 1997 SET population and (b) the 2027 population structure simulated by the ETMS model

extensive sensitivity analysis of the impact made by changing these treasury labour force assumptions. The alternative scenarios included simulating: higher and lower unemployment rates than predicted by the treasury (+ and – 1 percentage point);  higher and lower proportions (+ and – 1 percentage point) of the population aged 15–64 years, employed full time;  higher and lower proportions (+ and – 1 percentage point) of the population aged 15–64 years, employed part time;  higher and lower proportions of the employed population who are employees (again,+and – 1 percentage point); and  for those who are employees, slightly higher and lower proportions working in the private sector (again+and – 1 percentage point of the treasury assumption). Full results are available from the author, but the relative influence of varying these labour force assumptions is summarised below. 

3. Factors Influencing Education and Training Participation Rates In the prior section, we presented the alternative rates of growth in education and training participation rates arising from our base scenario and a wide range of alternative scenarios. In this section, we identify which

Training Participation Rates in Australia

193

scenarios provide the greatest relative impact on participation rates in education and training. However, comparing the scenarios and each associated outcome is complicated by the use of unstandardised scenarios. A method of standardising the variation of each set of alternative scenario statistics is thus required. Our technique is, for each set of scenarios, to calculate the coefficient of variation (CV) for (a) the variation in each of the seven education and training participation rates and (b) the variation in the relevant population scenario statistic.18 We then divide these two measures to adjust the variation in the participation rate by the variation in the relevant population scenario statistic. For example, to calculate the relative variation in formal education participation rates arising from variation in the population age structure scenario, we first calculate the CV for the participation rate.19 Standard deviation f16:17; 15:83; 15:77g Mean f16:17; 15:83; 15:77g ¼ 0:013547 ð1Þ

Participation rate CV ¼

Second, we calculate the CV arising from the variation in the population age structure. We draw on the average age of the population to proxy population structure.20 Population structure scenario CV ¼

Standard deviation f37:27; 37:59; 37:64g Mean f37:27; 37:59; 37:64g

¼ 0:005454

ð2Þ

Lastly, we divide the CV for the population age structure scenarios into the CV for the participation rate to obtain a standardised measure of participation rate dispersion relative to the variation in the relevant scenario statistic. Relative variation ¼

Participation rate CV ¼ 2:48 Population structure scenario CV

(3)

We undertake this analysis for each of the seven measures of participation in education and training and present the results in Table 4 (listed from higher to 18

Refer to Karmel (1957) for more information regarding the coefficient of variation statistic. 19 The numbers in braces in the calculation that follows are the participation rates in formal education in percentage format based on the three population structure scenarios (Series 1, 2, and 3, respectively). 20 The numbers in braces are the average age for the population (in years) for each of the three population structure scenarios (Series 1, 2, and 3, respectively).

Variability of Education and Training Participation Rates to Various Factors (and Direction of Influence — Positive/Negative) 2.48 0.57 0.56 0.38 0.27 0.23 0.19 0.16 0.04

  + +  + +  

2.21 1.86 0.47 0.30 0.16 0.16 0.12 0.11 0.02

 + + + + +   +

1.07 0.77 0.34 0.22 0.21 0.13

– + + + + –

Employer-supported external training Proportion employees Population age structure Years of education Proportion in private sector Proportion full-time employed Proportion part-time employed Proportion completed Year 12 Proportion unemployed Household and family formation Employer-supported formal training Proportion employees Proportion in private sector Population age structure Proportion full-time employed Years of education Proportion part-time employed Proportion completed Year 12 Proportion unemployed Household and family formation On-the-job training Proportion in private sector Proportion employees Proportion full-time employed Years of education Population age structure Proportion unemployed

2.48 0.96 0.49 0.48 0.40 0.39 0.10 0.09 0.02

+ + + ?a +  ?a  +

1.89 1.61 0.44 0.32 0.21 0.15 0.15 0.11 0.02

+  + + +  +  +

0.25 0.20 0.19 0.12 0.12 0.09

– + + +  –

Sandra Roussel

Formal education Population age structure Proportion in private sector Proportion completed Year 12 Proportion employees Proportion full-time employed Years of education Proportion part-time employed Household and family formation Proportion unemployed In-house training Proportion in private sector Proportion employees Population age structure Proportion full-time employed Years of education Proportion completed Year 12 Proportion unemployed Proportion part-time employed Household and family formation Formal training Proportion in private sector Proportion employees Population age structure Years of education Proportion full-time employed Proportion part-time employed

194

Table 4:

0.13 0.04 0.01

+  +

0.24 0.23 0.16 0.11 0.10 0.07 0.06 0.01 0.01

  + + + +   

Proportion completed Year 12 Proportion part-time employed Household and family formation

0.07 0.02 0.00

+  

Source: Coefficient of variation calculations are based on the predicted participation rates in education and training which, in turn, are based on the author’s calculations using logistic regression models and the population structures simulated b y the ETMS model Note: In references to the direction of influence, Population Age structure is measured by mean age of the population; and Family/Household formation is measured by the proportion of the population who are part of a couple and who have a dependent child present in the home. a The direction of influence (positive/negative) on employer-supported external training for (a) the proportion in the private sector and (b) the proportion with completed Year 12, is not clear.

Training Participation Rates in Australia

Proportion completed Year 12 Proportion unemployed Household and family formation Some education or training undertaken Proportion in private sector Population age structure Proportion employees Proportion full-time employed Years of education Proportion completed Year 12 Proportion unemployed Household and family formation Proportion part-time employed

195

196

Sandra Roussel

lower extent of influence for each type of education and training). We find there is no single pattern that is replicated over the seven education and training participation measures, which would have indicated factors that have a consistently stronger or weaker impact on each and every education and training participation rate. Some factors, however, do tend to feature more consistently than others. For example, the proportion of the population comprising employees (consistent positive influence) and the proportion of employees employed in the private sector (generally negative influence) have a consistently important influence on all seven education and training participation rates. The population age structure and the proportion of the population employed full time also appear in the top half of each list of relative variation. At the bottom of the lists, the household and family formation consistently has the weakest or second weakest impact on education and training participation rates. The two scenarios involving the proportion of the population unemployed and the proportion employed part-time also consistently appear in the bottom half of each list of relative variation. In general, the two education factors (proportion of the population with completed Year 12 and years of education) have relatively moderate influence on participation rates in education and training. Taking a closer look at each measure of education and training participation, we find that for formal education a single factor is dominant in its (negative) influence on the likelihood of participation: the age structure of the population.21 The strength and direction of this influence are not surprising given (a) education is an activity predominantly undertaken by youth and (b) the proxy for the population structure scenarios is the median age of the population. We note that the proportion of the population with completed Year 12 is also strongly (positively) correlated with participation rates in Formal education. The other education factor (years of education) has a relatively smaller (and positive) influence on Formal education participation rates. The two least influential factors are the proportion of the population unemployed, and household and family formation. For employer-supported external training the proportion of the population who are employees has, by far, the strongest (positive) influence on participation in this form of training. This is not surprising given that the individual must be an employee in our analysis to receive employer-supported

21

This is a negative influence as we measure the structure of the population by the mean age of the population. In Series A — a population with a relatively younger average age — the likelihood of participation in formal education is highest of the three series. Correspondingly, in Series C — a population with a relatively older average age — the likelihood of participation in formal education is lowest.

Training Participation Rates in Australia

197

external training. Also important, and providing a positive influence, are the structure of the population and years of education. Together, these two trends reflect the higher propensity of employers to invest in their middleaged as well as their better-educated employees. The proportion of employees employed in the private sector has an important influence. However, the direction (i.e. positive or negative) of the influence is not clear. The impact of the other education factor (proportion of the population with completed Year 12) has a negligible influence (but similarly unclear direction of influence), on participation rates in employer-supported external training. Household and family formation and the proportion of the population unemployed also exerted negligible influences on employer-supported external training participation rates. In-house training and employer-supported formal training are both strongly influenced by the proportion of the population in the private sector (negative influence) and the proportion of the population who are employees (positive influence). These findings are not surprising given the higher tendency of employees (than those otherwise employed) to receive employersupported training and the higher training levels for employees in the public sector. Household and family formation, the proportion of the population unemployed, the proportion of the population with completed Year 12, and the proportion of the population who are part-time employed, all exert a negligible influence on participation rates in these two forms of employersupported training. Our alternative scenarios cause less relative variation in formal training participation rates, but are similar in direction to employer-supported external training. The two strongest influences are exerted by the proportion of the population employed in the private sector (negative influence) and the proportion of the population who are employees (positive influence). The proportion of the population unemployed and household and family formation exert the least influence. The two education factors have a relatively negligible influence on participation rates in formal training. For on-the-job training and some education or training undertaken, the five variables providing the strongest influence are: proportion of the population in the private sector (negative influence); proportion of the population who are employees (positive); proportion of the population in full-time employment (positive); the population age structure (negative); and years of education (positive). However, none of these factors have a pronounced impact on participation rates in on-the-job training or some education or training undertaken. The two factors exerting the least influence are the proportion of the population in part-time employment (negative) and household and family formation (positive). The proportion of the population with completed Year 12 has a negligible influence on participation rates in on-the-job training and some education or training undertaken.

198

Sandra Roussel

4. Summary of Findings Based on our microsimulation update of the Australian population and applying the same character-specific participation rates that held in 1997, participation rates in education and training between 1997 and 2027 are expected to increase. It is important, however, to recall that our findings are based on a partial analysis — with focus only on the distribution effect and omission of the participation rate effect. Bearing this qualification in mind and using our base scenario, participation rates are expected to increase between 2 and 6 percentage points (which results in growth rates between 7 and 23 per cent) for our seven forms of education and training. Our projections indicate stronger growth in participation rates in all seven forms of education and training for females compared with males. The projected increase in participation rates between 1997 and 2027 also hold, but vary in magnitude, over the different age groups. Examination of the participation rates disaggregated by age reveals the increase in formal education participation rates is projected to be relatively greater for youth, while the increase in employer-supported external training, in-house training, employer-supported formal training, and formal training participation rates are projected to be somewhat greater for middle-aged to older workers. Growth rates in on-the-job training and some education or training undertaken are also greater for older workers. We also calculated participation rates in education and training based on a number of alternative population scenarios. From this analysis, we find population age structure Series 3 (representing an older population) and household and family formation Series A (representing a population with a higher proportion of couples and dependent children present) exert a positive influence on employer-supported external training, in-house training, employer-supported formal training, and formal training participation rates. In contrast, these same two series exert a negative influence on participation rates in formal education, on-the-job training, and some education or training undertaken. Years of education, the proportion that completed Year 12, and the proportion of the population comprising employees, all have a generally positive impact on all seven measures of participation in education and training. In contrast, factors exerting a generally negative influence are the proportion of the population unemployed and the proportion of employees in the private sector. Two peculiar exceptions exist. Both relate to participation in employer-supported external training: the direction of influence is not clearly discernable from the scenarios of the proportion of the population with completed Year 12 and the proportion of employees in the private sector. A possible explanation for this lack of direction is the random selection

Training Participation Rates in Australia

199

variability that exists in the microsimulation process itself — in the form of the random selection process of individuals for given events. The proportion of the population employed full-time (part-time) has a positive (negative) impact on participation rates in all but one participation measure: a negative (positive) impact is exerted on formal education. Standardising for the influence exerted by these alternative scenarios, we find the proportion of the population, who are employees and the proportion employed in the private sector, have a consistently important influence on all seven education and training participation rates.22 The population age structure and the proportion of the population employed full time also emerge as influential, as they appear in the top half of each list of relative variation. At the bottom of these lists, household and family formation consistently has the weakest or second weakest impact on education and training participation rates. The proportion of the population unemployed and the proportion employed part-time also emerge as weak influences as they consistently appear in the bottom half of each list of relative variation. In general, the two education factors (the proportion of the population with completed Year 12 and years of education) have relatively moderate influence on participation rates in education and training. As noted in the introduction, a number of earlier papers projecting participation rates in education and training in Australia have found that the ageing of the population will lead to lower engagement in education. The overall conclusion from these papers is that the proportion of government funding to education is expected to decline, particularly as governments come under increasing pressure to provide more services for the aged (e.g. pensions and health care). In particular, one study found, when holding agespecific participation rates at 1997 levels, that the direct effect of population ageing would reduce the overall rate of educational participation by between 4 and 6 percentage points over the period to 2030.23 We note that, in contrast to our analysis, this study includes 15–19 year olds attending school and includes secondary school in the definition of formal education. If we expand the scope of our own analysis to include this change in definition and scope, projections from our microsimulation model indicate that overall participation rates in formal education (secondary school plus post-school) still rise, but at the lower rate of 1.2 percentage points between 1997 and

22

Our simulated scenarios do not use a standard deviation of change in the demographic/labour market characteristic of interest. As such, we cannot compare the resulting change in the participation rate in education and training without first standardising the change in the characteristic of interest. We do this using the coefficient of variation. 23 See Peut (2001).

200

Sandra Roussel

2027.24 This is compared with a decline in participation in education of 4–6 percentage points if only the population ageing is accounted for. This example illustrates that focusing solely on the changing age distribution is only one part of the picture. As such, projections focusing on only one aspect of predicted changes in the population can lead to misleading findings. We conclude that the common finding of an expectation of a reduced need for education resources — based on the ageing of the population — may thus be ill founded. Further, the potential for diverting funding from education and training to pensions and health care may be much more constrained than the analysis based solely on ageing of the population has suggested. This argument is based solely on participation rates. However, to obtain a more comprehensive picture of expectations regarding education and training engagement, and funding in the future, the analysis would need to relate to the volume of education and training undertaken. In addition, the analysis would also need to account for other distribution effect changes, as well as the participation rate effect. To reiterate therefore: our model remains only partial in its coverage (mainly due to the omission of the participation rate effect), but does incorporate a number of important (distribution effect) changes — in educational attainment, labour force characteristics, household and family formation, and migration — in addition to the projected changes to the age structure of the population. We find that while age structure plays an important role in influencing our projections in education and training participation rates, other factors, that could potentially exhibit greater variation in the future, have a stronger relative impact on participation rates in education and training. Such factors include the proportion of the population comprising employees (exerting a positive influence on education and training participation rates), the proportion of employees employed in the private sector (negative influence), and the proportion of the population employed full time (a negative influence for formal education, but a positive influence on the other six training participation rates). To increase the level of comprehensiveness of our analysis, a further step would be to incorporate changes in character-specific participation rates (or the participation rate effect) into the model. Access to adequate longitudinal data is required before undertaking this step. As noted above, projections based on our model have, to date, focussed solely on participation rates in education and training. A next step in our

24

We note that this overall increase comprises an initial rise in participation rates in education between 1997 and 2012, followed by a more moderate decline in participation rates between 2012 and 2027.

Training Participation Rates in Australia

201

analysis of education and training in Australia would be to project estimates of the volume and cost of education and training, and on the basis of these projections, projected changes in public versus private funding of education and training in the future. This would help to address the question of how the resourcing — and in particular public resourcing — of education and training can be expected to change in the medium to longer term.

References Aungles, P., Karmel, T. and Wu, T. (2000). Demographic and Social Change: Implications for Education Funding. Department of Education, Training and Youth Affairs Occasional Paper Series 00B, Canberra. Australian Bureau of Statistics. (1998). Education and Training Experience, Australia, 1997. Cat. No. 6278.0, Australian Bureau of Statistics, Canberra. Commonwealth of Australia. (2002). Intergenerational Report: 2002– 03. 2002–03 Budget Paper no. 5, Canberra. Creedy, J. (1999). Population Ageing and the Growth of Social Expenditure. In: Productivity Commission & Melbourne Institute for Applied Economic and Social Research, Policy Implications of the Ageing of Australia’s Population, Conference Proceedings, Canberra, Ausinfo. Harding, A. (1993). Lifetime Income Distribution and Redistribution: Applications of a Microsimulation Model, Contributions to Economic Analysis Series, NorthHolland, Amsterdam. Harding, A. (2000). Dynamic Microsimulation: Recent Trends and Future Prospects, in Gupta, A. and Kapur, V. (eds), Microsimulation in Government Policy and Forecasting, Contributions to Economic Analysis Series, North-Holland, Amsterdam. Karmel, P.H. (1957). Applied Statistics for Economists: A Course in Statistical Methods. Pitman & sons, Melbourne. Peut, A. (2001). Australia’s Ageing Population — Implications for Education and Training. Unpublished manuscript, Department of Education, Science and Training, Canberra. Roussel, S. (2000). Factors Influencing Participation in Post-Secondary Education and Training in Australia: 1989– 1997. Research and Evaluation Branch Report 7, Department of Education, Training and Youth Affairs, Canberra. [http:// www.dest.gov.au/archive/iae/research/docs/Factors_influencing_participation.pdf] Roussel, S. (2003). Post-Secondary Education and Training Participation Rates in Australia in the Next 30 years: A Microsimulation Approach. In: International Microsimulation Conference on Population, Ageing and Health: Modelling Our Future. Conference Proceedings, National Centre for Social and Economic Modelling, Canberra. [http://www.natsem.canberra.edu.au/conference2003/papers/ pdf/roussel_sandra-1.pdf]

This page intentionally left blank

PART II: Taxes, Benefits and Labour Supply

This page intentionally left blank

Chapter 8

Lifetime Redistribution Through Taxes, Transfers and Non-Cash Benefits Thomas Pettersson and Tomas Pettersson Ministry of Finance, Sweden

Abstract Income distribution and redistribution is generally evaluated by examining equivalent disposable income in a cross-sectional context. In this study, the measurement period is extended, using lifetime rather than annual income, and the value of non-cash benefits are added to disposable income. By extending the measurement period, the total redistribution within a birth cohort can be divided into within-person (intrapersonal) redistribution and between-person (interpersonal) redistribution. By including non-cash benefits from, for example, education, health, elderly care and childcare, the income concept is closer to individual utility than the more frequently used concept of disposable income. The dynamic microsimulation model, SESIM, is used to generate individual incomes, taxes, transfers and private consumption of publicly financed goods and services.

1. Introduction The distribution of disposable income within a single year has been analysed in numerous reports and in the academic literature, as well as in more policy-oriented contexts. A common practice is to consider annual disposable income as an approximation of the level of living and sometimes of utility. This is, however, not unproblematic. Firstly, it can be argued that the income level of one single year does not necessarily reveal the true level of living. Longer measurement periods are likely to provide a clearer picture of the actual material standard of the households. It is also known that lifetime income is more evenly distributed than annual income. Secondly, disposable income does not cover all utilities in the form of goods and services available to the households. The household sector receives a large number of publicly subsidized goods and services, while home production and black-market International Symposia in Economic Theory and Econometrics, Vol. 15 Copyright r 2007 Elsevier B.V. All rights reserved ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15008-5

206

Thomas Pettersson and Tomas Pettersson

work generate values that are not captured by income statistics. By extending the measurement period and expanding the concept of income, a distribution closer to the distribution of utility can be achieved and analysed. In this paper, we neither assess home production or black-market values nor do we incorporate the value of leisure. Every year the public sector redistributes large sums between different individuals and households. This redistribution occurs through various taxes and transfers, as well as through publicly financed private consumption. The main motive for a system with redistributive properties may or may not be redistribution of economic resources. Social assistance and housing benefits are systems that are clearly motivated by their redistributive qualities, while transfers such as sickness benefits and unemployment benefits basically redistribute risk. Non-cash benefits are rarely primarily motivated by their equalizing effect on the distribution of economic resources. All of these systems do, however, intentionally or unintentionally, bring about redistribution in a cross-sectional perspective and are thus often analysed from such a perspective (e.g., Harding et al., 2006). For many of the redistributive systems, it is equally relevant to study their effects over longer periods. This paper focuses on the following three main issues: 





To what extent is the income distribution affected if the concept of income is extended to include publicly financed private consumption, and if we study income over the life cycle rather than income for a given year? What are the redistributive effects of various taxes, transfers and non-cash benefits, and do these effects change depending on the length of the measurement period? To what extent does redistribution performed by the public sector result in actual redistribution between individuals and to what extent is the result an intrapersonal redistribution — either in terms of a turnaround within a given year or as an intertemporal redistribution over the life cycle?

In Section 2, the method used to generate lifetime income and the SESIM model is presented. Section 3 describes the data sources. Sections 4, 5 and 6 present the results, and the conclusions are outlined in Section 7.

2. Method To analyse the distribution of lifetime income, a set of complete sequences of annual incomes is required. This can be achieved in several ways, and the possible methods can be broadly divided into historical or hypothetical methods (Blomquist, 1976). The historical method uses an individual’s actual income history. The major advantage of this approach is that the

Lifetime Income Redistribution

207

income patterns studied actually have occurred, and that all observed correlations between variables are true. The disadvantages are that it requires very long high-quality panels where the definition of income is reasonably consistent over time — and that it only allows the study of distribution among deceased persons or incomplete lifetime incomes. When analysing the effects of tax and transfer rules, the historical method will describe properties of systems that vary over time. If the main purpose is to compare lifetime and cross-sectional income distribution, the historical method is well suited. However, if the focus is on evaluating the effects of different systems, the hypothetical method might be preferred. The hypothetical method usually starts with a cross-section of individuals and then, by means of a certain function or model, simulates future incomes. This method avoids the problem of incomplete lifetime incomes — and any system of rules can be analysed in a consistent manner. The disadvantage of using simulated data is that the life patterns are artificial and that correlations over time and between variables might not necessarily correspond to the true relations. For this reason, validation of model results is essential to the credibility of the findings. In this paper we use the hypothetical method: to be more specific, all results are based on incomes simulated by SESIM, a dynamic microsimulation model. 2.1 Earlier Studies Several studies concerning the distribution of lifetime income and the redistributive effects of taxes and transfers have been published. Most of the earliest studies focused on a single system such as income taxes or pensions. Among the earliest studies to construct full lifetime incomes and analyse several taxes and transfers simultaneously in a life cycle perspective were Falkingham and Lessof (1992) and Harding (1993). The results of these two studies were combined in Falkingham and Harding (1996). Based on two dynamic cohort models with a partly common model structure, Falkingham and Harding compared the lifetime redistributive impact of taxes and transfers in the UK and Australia. Net lifetime income was simulated in a constant economic environment and under a fixed set of rules. The redistributive impact of a number of taxes and transfers was assessed both in a lifetime and cross-sectional perspective. A method of dividing lifetime redistribution into intrapersonal and interpersonal redistribution was also developed. This method is described and used in Section 4. Another example is Nelissen (1998), in which the dynamic microsimulation model NEDYMAS was used to analyse the redistributive impact of several Dutch transfer systems. The model was based on a cross-sectional sample of the Dutch population from 1947, which was aged in the model. The lifetime incomes of those born in 1930 and in 1950 respectively were

208

Thomas Pettersson and Tomas Pettersson

compared with the cross-sectional distribution of 1991. A recent example of a dynamic model that comprises a wide range of tax and transfer systems is O0 Donoghue (2001b), in which Irish transfer systems are examined using both simulated lifetime incomes and actual annual incomes. Moving to Sweden, one of the earliest attempts to simulate Swedish lifetime incomes is found in Blomquist (1976). A simple microsimulation model creates a lifetime income consisting of earnings, capital income after tax and bequests. The study is limited to employees who were 25 years of age in 1970. Bjo¨rklund and Fritzell (1992) examined the relation between crosssectional and lifetime income. Using actual data from 1951 to 1989, incomes for 38 years were used as an approximation of lifetime income for cohorts born between 1936 and 1942. The cross-sectional distribution of individual after-tax earnings was compared to the corresponding lifetime distribution. Correlations between annual and lifetime income were also calculated. It was found that the correlations were low or non-existent in the 25–30 age group, but relatively strong in older age groups. A conclusion was that annual income is a good predictor of lifetime income for people well established in the labour market, but a poor indicator for young people. To our knowledge, the only Swedish study assessing intrapersonal vs. interpersonal redistribution over the life cycle is Husse´nius and Sele´n (1994), in which synthetic lifecycle data were generated using methods of statistical matching applied to a two-year panel data source. Husse´nius and Sele´n quantify to what extent redistribution within a cohort is actual redistribution between individuals, and to what extent transfers received during a lifetime are self-financed. In the present paper, we update their results using an alternative income-generating process (dynamic microsimulation) and an expanded income concept comprising relevant taxes and transfers and including some non-cash benefits. 2.2 The SESIM Model The dynamic microsimulation model, SESIM, has been continuously developed at the Swedish Ministry of Finance since 1997. The first version was used for studies of the Swedish student loan system, but the focus soon turned to old-age pensions. Gradually, the model has become a platform for general longitudinal analysis of the interactions between individuals/ households and the public sector. The basic population of SESIM consists of a sample drawn from the 1999 year panel of the Longitudinal Individual Database (LINDA) (Edin and Fredriksson, 2000), and thus includes historical information. The model updates the properties of the sample individuals in a yearly sequence starting from 2000. The properties are updated using a series of statistical models and algorithms. The algorithms are mainly used to describe the tax and benefit systems. The statistical models

Lifetime Income Redistribution

209

are typically stochastic and the updating is based on Monte Carlo simulation methods. SESIM has a recursive structure consisting of a set of modules executed in a predetermined order described in Figure 1. The primary unit of simulation is the individual, but the household also plays a significant role. Because of this, the demography module contains several models that simulate the properties of the households. Many of the simulated processes in SESIM, both general stochastic models and rules of taxes and benefits, refer to household and individual properties. For a more thorough discussion on the SESIM model, see Flood et al. (2003). Figure 1:

Structure of SESIM

Demography − Mortality − Adoption − Migration − Fertility − Children leaving home − Cohabitation − Separation − Disability − Rehabilitation

Model population at time t

Next year (t = t + 1)

Model population at time t + 1

Noncash benefits − Child care − Compulsory education − Upper secondary education − University − Adult education − Labor market activities − Old age care − Health care − Medication

Education − Dropout from upper secondary education − From upper secondary to university − Dropout from university − From labor market to university − From labor market to adult education − From adult education to university

Labor Market − Unemployment − Employment − Miscellaneous status − Labor market sector − Income generation (earnings)

Wealth & Housing − Financial wealth − Real wealth − Income of capital

Taxes & Transfers − Student loans and allowances − Income tax − Real estate tax − Capital income tax − Wealth tax − Maintenance − Child allowance − Housing allowance − Social assistance − Old age pension − Disability pension

210

Thomas Pettersson and Tomas Pettersson

The concept of dynamic microsimulation is not discussed further here, but for a discussion and overview of existing models, see, for example, Rake and Zaidi (2001) or O’Donoghue (2001a). As the purpose here is to analyse the redistributional impact of various forms of public sector activity on the household sector, the effects of the present program rules are isolated by keeping other economic assumptions as neutral as possible. In the simulations, the tax and transfer rules are updated to 2003 and thereafter kept unchanged. No economic growth or inflation is assumed to take place after 2003, but the impact of demographic changes is included.

3. Data Sources 3.1 LINDA The primary data source of this study is the LINDA database (Edin and Fredriksson, 2000), which is used to construct the basic population as well as to estimate most of the statistical models used in the simulations. The database is a sample of 308,000 individuals, corresponding to 3.5 per cent of the Swedish population. The sample is complemented with the sampled individuals’ household members, yielding a total of 786,000 individuals in 1999. LINDA is longitudinal and includes information from 1968 onwards. Every year deceased and emigrating individuals are replaced, so that an accurate cross-sectional representation of the Swedish population in any given year is achieved. The basic population in SESIM is a sample of 104,000 individuals drawn from LINDA 1999, augmented by 8,000 individuals living abroad who are entitled to Swedish pensions. 3.2 HEK In some cases estimation of models is based on the income distribution survey, Husha˚llens Ekonomi (HEK). This is an annual survey conducted by Statistics Sweden. About 30,000 individuals are sampled and the survey information is complemented with information from administrative registers. The survey information allows a more complete definition of the household than LINDA. The HEK household definition is close to that of SESIM — and econometric models in which household information is critical are therefore in some cases estimated using HEK. More information about HEK can be found in Statistics Sweden (2003). 3.3 Non-Cash Benefits The HEK database is also used to estimate models for the simulation of non-cash benefits which are to be included in the income concept.

Lifetime Income Redistribution

211

The database contains information about the actual individual use of various non-cash benefits and the calculated subsidy value of the respective benefits. The value of the subsidy is assumed to equal production costs net of fees. Only benefits that can be attributed to a specific individual are included — that is, subsidies for compulsory education, upper secondary education, university, adult education, child care, old-age care, labour market activities, health care and medication. Logistic regression models are used to predict participation and linear regression models are used to predict subsidy values for those participating. For some of the subsidies, it is not realistic to assume that the value, or utility, is equal to the net production cost. Following Smeeding et al. (1993), imputation of health and old-age care subsidies is based on a risk-related insurance premia approach. That is, old-age care and health care are regarded as an insurance benefit received by all coverees, regardless of their actual usage. Benefit levels (insurance premia) differ by age and gender, according to differences in need (based on actual use within each group). In this way, the insurance premia are considered to be actuarially adjusted to account for differences in need-related values of being covered by the insurance. The size and composition of the imputed subsidies varies with age, as is shown in Figure 2. The impact of non-cash benefits on the crosssectional income distribution has also been studied by the Swedish Ministry of Finance (1999, 2002). The average individual, with a simulated life span of 80 years, receives 74 per cent of total subsidies before the age of 20 and after the age of 64. The corresponding share for an individual with a 95-year life span is 88 per cent. Figure 2:

Average Level and Composition of Non-Cash Benefits, by Age

300000 Health care (incl. medication)

250000

Labor market activities Education

SEK

200000

Old age and child care

150000 100000 50000 0 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 Age

Source: SESIM

212

Thomas Pettersson and Tomas Pettersson

The composition of non-cash benefits differs between age groups. At preschool ages, and from 80 years and on, childcare and old-age care dominate, somewhat more pronounced for children than for the elderly. For the oldest, health care subsidies also comprise a large share of the total. Health care subsidies dominate between the ages of 25 and 78 — and particularly between 50 and 70 years of age. Education subsidies naturally dominate during school age up to age 24. 3.4 Indirect Taxation To estimate the indirect taxation of households, consumption patterns from the household budget survey are imputed to the HEK database, after which household indirect taxation is calculated using current tax rules. The relation between indirect taxation and household equivalent disposable income is calculated for quantiles of equivalent disposable income. This relation is used to calculate indirect taxation of households in SESIM. The share of disposable income used for indirect taxation is higher for low-income households than for high-income households.

4. Distribution of Annual and Lifetime Income When analysing the distribution of income, methodological and definitional choices have to be made. There are no obvious measures and definitions that best reflect the actual utility the households obtain from their income. Usually, one seeks a concept that reflects, as realistically as possible, individual or household consumption possibilities. 4.1 Definitions and Concepts The income concept that is usually used to capture actual consumption possibilities is disposable income, which is defined as the household’s total earnings, capital income and transfers less (direct) taxes. In this paper we start with disposable income, add the value of received non-cash benefits and then subtract indirect taxes. The unit of income is the household. Our aim is to evaluate the distribution among individuals but individual well-being is often dependent on income from other household members. We make the common, but not undisputed, assumption that household income is equally shared between all household members. To compare consumption possibilities between households of different sizes and composition, household income needs to be adjusted or equalized. The presence of economies of scale in household economy makes per capita

Lifetime Income Redistribution

213

income misleading. We use an equivalence scale that takes into account the economies of scale and differences in consumption needs between adults and children. The following parametric form is assumed: Household consumption weight ¼ ðna þ bnc Þ

(1)

where na and nc equal the number of adults and children, respectively. The parameters are estimated from the social assistance norm giving b^ ¼ ^ ¼ 0:7: This scale is used to adjust disposable income and indirect taxation. Smeeding et al. (1993) argue that because non-cash income does not depend on family size or composition but depends only on individual characteristics, there are no economics of scale for non-cash income. We follow this line of argument and aggregate all non-cash income for the household as a whole and express that in per capita terms. In the following, disposable income divided by the equivalence scale is termed equivalent disposable income, and equivalent disposable income less indirect tax adjusted with the equivalence scale plus non-cash benefits per capita is termed total income. The unit of observation is the individual. All incomes are aggregated to household level, but all presentations refer to the individual. Equivalent disposable income or total income of the household is attributed to all individuals in the household. The definitions so far concern annual incomes; lifetime income is defined as the individual average of annual values. Following Harding (1993), averages, instead of sums, are used to account for life span differences. When lifetime income is analysed, only individuals who have lived most of their lives in Sweden are included, those who have lived abroad for more than 9 years are excluded. Averages refer only to years lived in Sweden. In this and the following sections, cross-sectional income refers to the years 2000–2109 and lifetime income refers to birth cohorts 2000–2010. 4.2 Evaluation of Simulated Incomes Before turning to the comparisons between annual and lifetime distribution of different levels of income, we first compare simulated distributions with observed distributions. In Pettersson and Pettersson (2003), the simulated income levels and distributions used were compared with empirical ones using a variety of measures, indices, etc. Only a few of these will be discussed here. The cross-sectional distribution, defined as the Gini coefficients for equivalent disposable income — according to both the income distribution survey and LINDA for the years 1982–2001 — was compared with the corresponding Gini coefficients from SESIM for the years 2000–2023. When viewing a summary index of the distribution as a whole, the result was satisfactory; the simulated coefficients were at the same level as those found

214

Thomas Pettersson and Tomas Pettersson

in actual statistics; and there was no apparent trend. Beneath this summary overview, however, two opposing problems were concealed. First, the share of individuals with very low incomes was underestimated in SESIM, which resulted in an underestimation of inequality. Second, the share of individuals with low (but not very low) incomes was overestimated, causing an overestimation of inequality. The two problems more or less cancelled each other out, so that the summary index showed the ‘‘right’’ level of inequality. Lacking actual data, the simulated distribution of lifetime income was evaluated, comparing incomes over a 20-year period. Incomes from LINDA for the period 1982–2001 was compared to simulated data for the years 2004–2023. In this case, we looked at the earnings distribution for those of active age during the period. Only those receiving earnings in all of the 20 years were included. The simulated distribution of individual earnings coincided surprisingly well with the empirical distribution. The share of individuals with very low earnings was slightly underestimated and the distribution was displaced to the left. The simulated distribution was slightly more uneven than the actual distribution; the Gini coefficient for these 20year incomes was 0.295 and 0.284, respectively. 4.3 Results Cross-sectional data covering the period 2000–2109 was pooled together into one dataset representing annual distributions. The cross-sectional Gini coefficients for equivalent disposable income and total income are 0.217 and 0.189, respectively. The inclusion of non-cash benefits thus reduces the dispersion by approximately 10 per cent. This is a smaller reduction than previous studies have reported: in studies by the Ministry of Finance (1999, 2002) and Fritzell (1994), the reduction has been quantified at approximately 20 per cent. One explanation is that total income is reduced by indirect taxation, which is known to be regressive. If the leap from equivalent disposable income to total income is taken in two steps, and the last step is the inclusion of non-cash benefits, this inclusion brings about a 15 per cent Gini reduction. The next step is to turn from annual to lifetime distribution. The Gini coefficient for total lifetime income (defined as the individual lifetime average of annual total income) is 0.086. Lifetime income is thus almost 60 per cent more evenly distributed than annual income. This result is well in line with Husse´nius and Sele´n (1994). Using actual data, Bjo¨rklund and Fritzell (1992) estimated the reduction in inequality of turning from annual to lifetime income at approximately 40 per cent. One explanation for the smaller impact is that the results were based on net individual earnings, an income concept quite distant from equivalent disposable income or total income. Another reason is that lifetime incomes were approximated with

Lifetime Income Redistribution

215

Table 1: Annual and Lifetime Income, Gini Coefficients (and Quintile Ratios in Brackets) Income concept Market income Equivalent disposable income indirect taxes +non-cash income (i.e., total income)

Annual income 0.490 0.217 0.224 0.189

(15.6) (1.90) (1.95) (1.75)

Lifetime income 0.196 0.102 0.104 0.086

(1.73) (1.35) (1.36) (1.28)

Change (%) 60 53 54 55

Source: SESIM Note: Quintile ratios (in brackets) are the ratio between cut-off points for the first and last quintile.

incomes over a 38-year period. The relative difference between the Gini for annual and lifetime equivalent disposable income is similar to that found by Harding (1993, p. 151) for Australia. Table 1 summarizes the distributional effects of varying timeframes and income concepts. The annual distribution of market income, here consisting of earnings and capital income, is obviously more unequal than income concepts that take redistributive taxes and transfers into account. The income level that separates the fourth and fifth quintile is more than 15 times as high as the income level separating the first and second quintile for annual market income. When the redistribution that occurs through the public sector and through income mobility over the life cycle is taken into account, the gap is reduced to approximately 30 per cent. More than 90 per cent of the initial gap disappears. The Gini coefficient is reduced by more than 80 per cent.

5. Redistributive Impacts of Income Components Within a Year and Over the Life Cycle An income component such as a tax or a transfer can either contribute to, or counteract, income inequality. An income component that is strongly equalizing in a cross-sectional perspective might have other (even opposing) effects when evaluated in a lifetime perspective. To understand what drives inequality, and the possibilities of affecting the dispersion through political measures, we need insight into the distributional impact of various income components. For this purpose, the distribution of total annual income, as well as lifetime income, is decomposed by income components. The aim is to describe the way in which taxes, transfers and non-cash benefits redistribute economic resources between individuals and to determine whether this redistribution is sensitive to variations in the timeframe. Taxes, transfers and non-cash benefits affect inequality both directly and indirectly. The distribution of market income is indirectly affected through individual

Thomas Pettersson and Tomas Pettersson

216

participation decisions, prices and wages. In the absence of taxes and transfers — or with a different tax/transfer structure — participation rates, wage distribution, the required return on investments, etc., would probably all be different than they are today. In the long run, subsidized education affects the distribution of income possibilities and thus the distribution of market income. However, this paper is limited to an assessment of direct, or mechanical, redistribution. 5.1 Method There are two different methods to describe the distributive impact of income components: the decomposition method and the stepwise method. In the decomposition method, the distributional impact of all components is considered simultaneously whereas, in the stepwise method, the components are evaluated one at a time. In the stepwise method, the inequality of one component — for example, market income — is first calculated. Another component is then added and the inequality of this aggregated income is calculated. The new level of inequality is compared with the previous level, and the change is interpreted as the impact of the added component. This procedure continues until all components have been included. The impact of the components, however, will depend on the order in which they are added. This is not the case if a decomposition method is used. The stepwise method describes the distribution of a component in relation to the distribution of the preceding aggregate, while the decomposition method describes the distribution of a component in relation to the distribution of the final aggregate (total income or equivalent disposable income). We have chosen to use a decomposition method — developed by Rao (1969) and Kakwani (1977). Total inequality, measured by the Gini coefficient, is divided into a number of distinct components. Each component is in turn the product of two elements: a weight that defines the relative size of the component and a concentration index. The Gini coefficient can be defined as follows: Gini ¼

n X

wi k i

(2)

i¼1

where wi and ki is the weight and concentration index of component i, and n equals the number of components. The concentration index is a number between 1 and 1 that measures the distribution of a component in relation to total income. The concentration index should be interpreted as the component distribution, given the distribution and size of all other components. The concentration index of a component is not independent of its size. A component that, at a small size, is

Lifetime Income Redistribution

217

considered to be equalizing might, if its size is increased, start to contribute to inequality. Consider a transfer that only goes out to the poorest (before this transfer) individual in the population; this transfer will obviously be equalizing at low levels but, if the transfer is gradually increased, it will at some point push the recipient up into the higher part of the distribution. If the transfer is then further increased, it will then contribute to total inequality. A positive concentration index does not necessarily mean that if the component did not exist, the distribution would be more even. Neither does a positive index by necessity imply that a marginal increase in the component will cause increased inequality. If a component that is more evenly distributed than total income (has a concentration index smaller than the Gini coefficient) is increased, the effect will be a reduction in total inequality as measured by the Gini coefficient. When determining a component’s distributional impact, it is the component’s concentration index compared with the Gini coefficient — not if it is larger or smaller than zero — that is crucial. The weight is defined as the component’s share of total income over all individuals. The weight of taxes assumes negative signs and therefore the interpretation of concentration indices for taxes is opposed. 5.2 Results Concentration indices, weights and Gini contributions of a number of income components are presented in Table 2. The concentration index multiplied by the weight equals the Gini contribution for each income component. The Gini contribution for all components adds up to the Gini coefficient. If the concentration index of a component is smaller than the Gini, the component is progressive — and otherwise regressive. The opposite is true regarding taxes. In the cross-sections, old-age care, market income, indirect taxes and occupational pensions have regressive properties. All other income sources are more or less progressive. In a lifetime perspective, occupational pensions, old-age care, indirect taxes and repaid student loans are regressive. The regressivity of the last component is, however, small and could be considered more or less neutral. In a cross-sectional perspective the income sources with the most progressive profile are social assistance, housing allowance (families with children), parental allowance, subsidized adult education and disability pensions. The corresponding sources in a lifetime perspective are disability pensions, housing supplements for pensioners, social assistance, housing allowance (families with children) and subsidized adult education. In most cases it is relatively easy to understand why an income source is progressive or regressive, but in some cases it is less obvious. Below, we comment on a few of the income sources grouped according to their distributive properties.

218

Contribution to Total Annual and Lifetime Inequality from Different Income Components

Income component

Annual income Index

Market income Occupational pension Old-age pension Disability pension Survivors’ pension Unemployment benefits Sickness allowance Parental allowance Child allowance Housing allowance Housing supplement (pensioners) Received maintenance Student benefits

0.292 0.234 0.060 0.231 0.331 0.034 0.071 0.481 0.096 0.507 0.203 0.162 0.122

Weight

0.875 0.040 0.162 0.039 0.001 0.028 0.033 0.012 0.020 0.003 0.008 0.006 0.011

Lifetime income

Contribution Absolute

Relative (%)

0.256 0.009 0.010 0.009 0.000 0.001 0.002 0.006 0.002 0.002 0.002 0.001 0.001

141 5 5 5 0 1 1 3 1 1 1 1 1

Index

0.131 0.327 0.106 0.319 0.004 0.023 0.053 0.024 0.041 0.226 0.317 0.123 0.032

Weight

0.852 0.044 0.177 0.042 0.001 0.028 0.034 0.011 0.019 0.003 0.010 0.006 0.010

Contribution Absolute

Relative (%)

0.111 0.015 0.019 0.013 0.000 0.001 0.002 0.000 0.001 0.001 0.003 0.001 0.000

132 17 22 16 0 1 2 0 1 1 4 1 0

Thomas Pettersson and Tomas Pettersson

Table 2:

Total income Source: SESIM

0.638 0.108 0.128 0.576 0.136 0.026 0.395 0.065 0.026 0.016 0.140 0.258 0.346 0.183 0.118

0.005 0.064 0.021 0.053 0.011 0.007 0.000 0.011 0.035 0.000 0.005 0.332 0.028 0.010 0.080

0.003 0.007 0.003 0.030 0.002 0.000 0.000 0.001 0.001 0.000 0.001 0.085 0.010 0.002 0.010

2 4 1 17 1 0 0 0 0 0 0 47 5 1 5

1.000

0.182

100

0.291 0.046 0.032 0.199 0.003 0.050 0.134 0.028 0.032 0.004 0.067 0.137 0.145 0.073 0.060

0.005 0.068 0.020 0.056 0.010 0.007 0.000 0.012 0.033 0.001 0.005 0.336 0.023 0.010 0.083

0.001 0.003 0.001 0.011 0.000 0.000 0.000 0.000 0.001 0.000 0.000 0.046 0.003 0.001 0.005

2 4 1 13 0 0 0 0 1 0 0 55 4 1 6

1.000

0.084

100

Lifetime Income Redistribution

Social assistance Health care Child care Old-age care Upper secondary education Labour-market activities Upper secondary education for adults Medication Compulsory education Adult education University Income tax Other direct taxes Repaid student loans Indirect taxes

219

220

Thomas Pettersson and Tomas Pettersson

5.3 Progressive Income Sources Most income sources included have progressive profiles in the cross-section as well as over life. In general, social insurance transfers have such properties. The progressivity of disability pensions increases when the measurement period is extended. During a year, people receiving disability benefits generally have low total incomes, which explains the progressive effect. The probability of being disabled correlates with low education and low income; this and the fact that disability pension is in itself relatively low, leads to a sharpened progressive profile over life. In general, transfers going to families with children, at least in a crosssectional perspective, have progressive effects. This is due to the fact that their consumption needs give them a weaker position in the distribution. Means-tested benefits, housing allowance (families with children and pensioners) and social assistance are redistributive by construction. Pensioners receiving housing supplements have low pensions; these low pensions are in turn an effect of low income during active ages, which explains the stronger progressive effect of these supplements over the lifetime. Student benefits (loans and allowances) work progressively in the crosssection and, to some extent, over life. While they study, students often have small or no other sources of income, placing them at the bottom of the distribution. After graduation, they receive higher earnings, and the expected result is student benefits that are regressive over life. However, the group of students is heterogeneous, as it consists both of university students who can expect high future incomes and adults on supplementary training. Those in the latter group tend to have low-income histories and not particularly high-expected future incomes. Income taxes are progressive within a single year as well as over life. The progressive power is almost the same in the short run as in the long run. Bjo¨rklund and Fritzell (1992) draw the same conclusion using a stepwise method. Other direct taxes include a tax on capital income, wealth and property taxes. The tax bases for these taxes are more unequally distributed than earnings, causing these taxes to be more progressive than income tax. However, not all wealthy individuals are wealthy for life. Some spend their assets and others receive temporary capital gains when selling their house. In a lifetime perspective, certain equalization occurs, causing these taxes to be less progressive over the lifetime. Everyone receives an equal subsidy for compulsory education. Over the lifetime, this subsidy becomes relatively more valuable for those with low lifetime incomes, and the lifetime profile becomes progressive. In the crosssection, the progressivity is caused by the fact that higher consumption needs, on average, put families with children in the lower part of the distribution. The same line of argument is valid for upper secondary education,

Lifetime Income Redistribution

221

which is more or less neutral in a one-year period and marginally progressive over life. As expected, subsidized higher education also has progressive annual effects. As in the case of student benefits, the expected regressive lifetime effect is absent but the size of the progression is greatly reduced. Subsidized adult education has a substantial progressive profile both over life and in the one-year period (see student benefits). As mentioned earlier, health care subsidies have been given an insurancebased profile, where all individuals in each cell, based on gender and age, receive the same amount. Those not receiving hospital care are assumed to derive the same utility from subsidized care as those unfortunate enough to be in actual need of care. One of the drawbacks of this procedure is that the correlation that might actually exist between income and health care needs will be ignored. It would not be unreasonable to argue that low-income earners, relatively speaking, benefit more from subsidized care. On the other hand, they might be less inclined to buy insurance if they received the subsidy in cash and had the choice to participate, suggesting the opposite relation between income and subsidies. Here, however, the risk-related insurance premia is determined only by age and gender. The weak progressivity that nonetheless emerges is because of higher subsidies for women than for men. 5.4 Regressive Income Sources There are four purely regressive income components in the disposition used here: namely market income, occupational pensions, old-age care and indirect taxes. Income mobility results in reduced regressivity in market income over life. Old-age pensioners receiving high occupational pensions also receive a high public pension and thus high total incomes. High public and occupational pensions also suggest high previous earnings, which give even higher regressivity from the lifetime perspective. As in the case of health care subsidies, an insurance premia approach is assumed for old-age care. In spite of this, old-age care is the most regressive income source of all in a cross-sectional context. The powerful regressive pattern is due to the high values assigned to the recipients. Combined with even a low pension, this puts the recipient above the median of the distribution of total income. The effects are also caused by the positive correlation between expected lifetime and lifetime incomes — and the fact that old-age care subsidies increase rapidly with age. 5.5 Income Sources where the Distributional Effect Depends on the Timeframe A vast majority of the income sources have similar distributional properties regardless of the length of the measurement period. The distributive power

222

Thomas Pettersson and Tomas Pettersson

for some components may be sensitive to period length but only a few change sign. One exception is survivors’ pension. This, however, is an effect of the simulated period. Survivors’ pensions consist of children’s and widows’ pensions. In the cross-sections, the regressive widows’ pensions dominate the progressive children’s pensions, making the net effect regressive. The lifetime data are limited to those born between 2000 and 2010 — generations not entitled to a widow’s pension — leaving only progressive children’s pensions. The redistributive effects of old-age pension payments turn from progressive to regressive when the perspective is shifted from annual to lifetime. The average old-age pensioner has below median total income,1 explaining the cross-section progressive effect. In general, a high pension does not lead to a position at the top of the annual distribution but, combined with previous high earnings (that result in the high pension), it might qualify for a top position in the lifetime distribution. It should be noted that these results only consider pension benefits, and not the financing of these benefits. In Pettersson and Pettersson (2003), we show that the lifetime redistributive effects of the old-age pension system as a whole, including financing, are progressive. 5.6 Gini Contribution The focus so far has been solely on the redistributive direction of the income components: in the next step the size of the components is also taken into account. The product of direction and size is the Gini contribution. As the absolute Gini contributions for annual and lifetime income respectively sum up to different totals, a comparison between annual and lifetime effects utilizing these values becomes difficult. Instead a relative Gini contribution, where the values have been normalized by the Gini coefficient, is used. In the distribution of annual incomes, inequality is caused almost entirely by market income and to some extent by subsidized old-age care. Market income is without question the largest component, constituting 88 per cent of total income. The Gini contribution from market income exceeds 100 per cent, suggesting that the joint effect of other components is equalizing. The distributive effect of old-age care has already been discussed. Other sources of income showing a considerable contributing effect are old age and occupational pensions. The pattern remains the same for lifetime income but the effect of pensions is accentuated, and contributions from old age and 1 This is, in part, due to model-specific circumstances; we do not allow part-time retirement, and retirees cannot receive earnings. Combined with the fact that the version of SESIM used here does not model private pensions, the result is an underestimation of the relative position of old-age pensioners.

Lifetime Income Redistribution

223

occupational pensions increase at the expense of market income. Income tax turns out to have the most equalizing effect on annual as well as lifetime distribution.

6. Redistribution between Individuals and over the Life Cycle Redistribution of means between individuals will in the following be termed interpersonal redistribution (e.g., redistribution from the lifetime rich to the lifetime poor). Redistribution of means within individuals will be termed intrapersonal redistribution (e.g., the redistribution of means from the middle of a life cycle to the same individual later in their life cycle). Intrapersonal redistribution can occur within one year (caused by e.g., taxable transfers) and/or between different years. While some components of the tax and transfer system can be classified by their primary redistributive properties on intuitive grounds, this is more difficult for other components. Below, the redistributive property of the public sector is assessed through simulation. In addition to the previously studied income components, the following analyses also include employer’s contributions. Household resources are again assumed to be evenly divided between the members of each household (i.e., this component of the analysis divides all economic resources on a per capita basis whereas, in Sections 4 and 5, our measure of economic resources is based on equivalent disposable income plus per capita non-cash benefits). Since the analyses here focus on the redistribution of means rather than the level of economic well-being of the households, no adjustment is made for consumption needs. 6.1 Method The method used to calculate the components of redistribution presented later has been used by Falkingham and Harding (1996), O’Donoghue (2001b), and Husse´nius and Sele´n (1994). Each birth cohort is assumed to be financially balanced over the lifetime — i.e., the sum of transfers and benefits paid to the cohort exactly equals the sum of taxes paid by the cohort. Because the analysis does not include all components of the public sector, the sum of paid taxes will exceed the sum of transfers and non-cash benefits. The level of taxes (S) is therefore adjusted in order to achieve financial balance within all cohorts — i.e., so that the level of taxes corresponds to the level of transfers (T). Hence, all direct taxes (DT) are used while indirect taxes (IDT) and social insurance contributions (SIC) are proportionally reduced by finding the appropriate factor a in Equation (3). S ¼ DT þ aðIDT þ SICÞ ¼ T

(3)

Thomas Pettersson and Tomas Pettersson

224

The following equations apply to certain specific birth cohorts. Let Sit denote the taxes and social insurance contributions paid by individual i at age t. Let Tit denote the sum of transfers and non-cash benefits received by individual i at age t. The lifetime net2 of the individual to the public sector is given by Ki Ki X X Ni ¼ T it  S it (4) t¼0

t¼0

where Ki denotes the age at which individual i dies. The interpersonal redistribution in the cohort is calculated as the sum of all positive nets N X INTER ¼ ðN i jN i 40Þ (5) i¼1

Note that this equals (the negative of) the sum of all negative individual nets in the cohort. The intrapersonal redistribution consists of two components, redistribution within one year (annual redistribution) and between different years (lifetime redistribution). The annual redistribution for individual i at age t is calculated as the minimum of paid taxes and received transfers/ benefits INTRA_Y it ¼ minðT it ; Sit Þ

(6)

The total annual redistribution within the cohort equals the sum of individual values INTRA_Y ¼

Ki N X X

INTRA_Y it

(7)

i¼1 t¼1

The calculation of the lifetime redistribution of individual i depends on whether the individual net is positive or negative 8 Ki P > > > < t¼i ðS it  INTRA_Y it Þ; if N i 40 (8) INTRA_Li ¼ K i P > > > : ðT it  INTRA_Y it Þ; if N i  0 t¼i

The total lifetime redistribution within the cohort is given by INTRA_L ¼

N X

INTRA_Li

(9)

i¼1

2

Note that no discount factor is used since the simulations assume a stylized economic environment without inflation or growth.

Lifetime Income Redistribution

225

The relation between the total taxes/fees, transfers/benefits and the above calculated components is given by Ki N X X i¼1 t¼0

T it ¼

Ki N X X

S it ¼ INTER þ INTRA_Y þ INTRA_L

(10)

i¼1 t¼0

The following results are based on calculations for birth cohorts 2000–2010 for which complete life cycles have been simulated. All amounts are stated in Swedish kronor (SEK) at 2003 price levels. 6.2 Interpersonal and Intrapersonal Redistribution Table 3 shows the calculated components of redistribution through the public sector. The annual redistribution is the single largest component, accounting for 45 per cent of all redistributed means. It is caused by taxable transfers and also by the assumption about shared resources within households (e.g., children receiving educational subsidies while their parents pay taxes). The interpersonal redistribution is 18 per cent, which means that the share of transfers and benefits that are self-financed amounts to 82 per cent. Net of the annual redistribution, the interpersonal and intrapersonal redistribution is 32 and 68 per cent respectively. The results indicate that the main redistributive effect of the public sector is a transfer of resources across the life cycle. When calculated using another, more regressive scheme for allocating taxes — that uses the full amount of indirect taxes and a certain share of direct taxes and social insurance contributions — these results only show slight changes. In Husse´nius and Sele´n (1994), a somewhat higher estimate of interpersonal redistribution — about 24 per cent — is reported. Net of the annual redistribution, the interpersonal redistribution is 32 per cent, which coincides with the results in Table 3. In international studies, Falkingham and Harding (1996) calculate the interpersonal redistribution as lying between 48 and 63 per cent for Australia and between 29 and 38 per cent for the UK, Table 3:

Redistributive Components: Average Amount per Individual (SEK, ‘000 s)

Component Interpersonal redistribution Annual intrapersonal Lifetime intrapersonal Total Source: SESIM

Amount

Share (%)

Share, net of annual redistribution (%)

1,194 3,024 2,540 6,758

18 45 38 100

32 68 100

226

Thomas Pettersson and Tomas Pettersson

depending on the method used to allocate taxes. In O’Donoghue (2001b), the interpersonal redistribution in Ireland and Italy was estimated at 45 and 24 per cent, respectively. The comparatively low interpersonal redistribution in Sweden (compared to Australia, UK and Ireland) can partly be explained by the fact that these other studies only included the impact of income taxes and cash transfers, and excluded non-cash benefits from their scope. As Harding et al. (2006) show, non-cash benefits in Australia, for example, are less redistributive than cash transfers. In addition, a large share of transfers and benefits in Sweden are social insurances and hence based on the incomes of the policyholder. The old-age pension system alone constitutes about 56 per cent of all transfers. Also, a substantial share of all transfers in Sweden is taxable. Table 4 shows the average flow of taxes, transfers and non-cash benefits between the households and the public sector by quintiles of average equivalent disposable income. Averages are calculated by year of life, thus adjusting for individual differences in life span. The table shows that average taxes increase with increasing life income. Compared to the total population average, individuals in the lowest quintile of life cycle income pay 37 per cent less tax, while individuals in the highest quintile pay 40 per cent more tax. Transfers and non-cash benefits are more evenly distributed across levels of life cycle income. Transfers to the lowest quintile are about 3 per cent higher than the total average, while they are 7 per cent lower for the highest quintile. Non-cash benefits are highest for the second and third quintile (about 6 per cent above the total average), and lowest for the highest quintile (9 per cent below the total average). Table 4: Taxes, Transfers and Non-Cash Benefits: Average Amount per Individual and Quintile of Life Cycle Income (SEK, ‘000 s) Quintile of life cycle income

Taxes % Transfers % Non-cash benefits % Annual redistribution Lifetime redistribution Net

All

1

2

3

4

5

4,264 37 3,924 +3 2,840 3 2,603 1,536 2,500

5,741 15 3,983 +4 3,109 +6 3,057 2,374 1,351

6,597 2 3,915 +2 3,109 +6 3,143 2,784 428

7,691 +14 3,770 2 2,922 0 3,153 3,093 999

9,483 +40 3,543 7 2,679 9 3,162 2,907 3,262

6,758 3,827 2,932 3,024 2,540 0

Source: SESIM Note: Rows labelled as % refer to the relative difference to the total population average (percent).

Lifetime Income Redistribution

227

The average lifetime net outcomes to the public sector are negative for the highest quintiles and positive for the lowest quintiles. This again shows that the public sector also has an equalizing effect on lifetime incomes. The total average lifetime net equals zero (because the cohorts are forced to be financially balanced). The share of self-financed transfers and benefits, defined as the sum of intrapersonal redistribution divided by the sum of transfers and non-cash benefits, increases with increasing life cycle income. In the lowest quintile, 61 per cent of transfers and benefits are self-financed, while the corresponding share is 98 per cent in the highest quintile. 6.3 Taxes, Transfers and Non-Cash Benefits over the Life Cycle The above results indicate that the economic transactions between households and the public sector create a significant redistribution of resources across the life cycle. To understand this process, some information about the timing of these transactions is needed. Average taxes, transfers, non-cash benefits and individual net outcomes calculated by age are shown in Figure 3. Transfers are, primarily through the disability and old-age pension systems, concentrated among individuals aged 60 and above. The calculations show that about 65 per cent of all transfers are allocated to these age groups. During the working ages, taxes develop at about the same rate as incomes. After retirement, incomes Figure 3: Average Taxes, Transfers, Non-Cash Benefits, Individual Nets and Cumulative Net, by Age 400000

8000000 Net Taxes Transfers Noncash benefits Cumulative net (right axis)

300000

6000000

200000

4000000

100000

2000000

0 0

0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95

-100000

-2000000

-200000

-4000000

Source: SESIM

228

Thomas Pettersson and Tomas Pettersson

decrease and social security contributions are no longer paid, bringing about a significant reduction in taxes. The average individual net outcome is negative until retirement, after which the combined effect of lower taxes and increased transfers makes it positive. Because of rapidly increasing non-cash benefits in the highest ages, the average individual net outcome also increases rather rapidly. Note that taxes are incident on children. This is an effect of the assumption of a shared economy within the household; children share their parents’ taxes. In the same way, some of the children’s non-cash benefits from studying are incident on their parents. Figure 3 shows that the average taxpayer is younger than the average recipient of transfers and non-cash benefits. The average individual net outcome is initially negative (except for newborns), but becomes positive after retirement. Thus the obvious question is: When does the average individual attain a permanent positive net to the public sector? The yearly cumulative net — that is, the sum of individual yearly nets from birth until the current age — is shown to be negative and decreasing every year until retirement. The rate of decrease is fastest between the ages of 50 and 60, when the ratio between taxes and transfers/benefits is highest. After retirement the cumulative net result starts to increase and turns (permanently) positive at age 87. As a consequence, it is more likely that an individual with a long life will be a net beneficiary (of the public sector) than an individual with a short life will be. Hence, a redistribution of means takes place from individuals with short lives to individuals with long lives. This is also noted by Harding et al. (2000) in a study of the redistributive properties of the Australian public health outlays. The effect of this redistribution is, however, counteracted by the positive relation between lifetime income and expected life length. 6.4 Gender Redistribution Since average earnings are higher for men than for women, one might suspect that the public sector causes some degree of gender redistribution of lifetime incomes. Table 5 shows the average flow of taxes, transfers and noncash benefits between the households and the public sector for men and women. Compared with the total average, men pay 6 per cent higher taxes and women 6 per cent lower taxes. The transfers differ by about the same relative amount — but with men receiving less than women. The largest difference can be seen in non-cash benefits, where women receive 11 per cent more than the total average and men 11 per cent less. The share of transfers and benefits that are self-financed are about 89 per cent for men and 77 per cent for women. The lifetime net shows that somewhat less than SEK 1 million is transferred from the average man to the average woman over the life cycle.

Lifetime Income Redistribution

229

Table 5: Taxes, Transfers and Non-Cash Benefits: Average Amounts by Gender and Year of Life (SEK ‘000 s) Gender

Taxes % Transfers % Non-cash benefits % Annual redistribution Lifetime redistribution Net effect

All

Men

Women

7,136 +6 3,562 7 2,611 11 2,920 2,547 963

6,389 6 4,085 +7 3,245 +11 3,125 2,534 941

6,758 3,827 2,932 3,024 2,540 0

Source: SESIM Note: Rows labelled as % refer to the relative difference to the total population average (percent).

7. Concluding Remarks This study describes the redistributive process of the public sector from different perspectives. Our aim has not been to suggest any reforms or policy changes. Instead, our aim has been to present an analytical foundation that might serve as input for further discussions concerning the redistributive role of the public sector. Without embarking on such a discussion, it might be interesting to highlight a few conclusions and raise a few questions. Lifetime income is far more evenly distributed than annual income. The redistributive impact of the public sector is stronger in any given year than over life. When analysing redistributive systems and when preparing or evaluating reforms, it is essential that the life cycle perspective is considered. A single focus on annual properties might lead to incorrect conclusions and there is a risk that reforms, in the longer run, might not have the intended effects. The degree of self-financing in public subsidies and transfers is extensive and only a small part of the redistribution conducted by the public sector results in actual reallocation of resources between individuals. In view of the expected future demographic developments, and the anticipated difficulties in the future financing of the public sector, it is important to discuss public sector commitments. Priorities have to be set, and some intertemporal redistribution could perhaps be handled at an individual level. Another essential question concerns the degree to which the individuals themselves could finance the individual consumption that today is financed through public budgets.

230

Thomas Pettersson and Tomas Pettersson

References Bjo¨rklund, A., Fritzell, J. (1992). Inkomstfo¨rdelningens utveckling. Appendix 8 in The 1992 Long Term Survey. Fritzes, Stockholm Blomquist, N.S. (1976). The Distribution of Lifetime Income — A Case Study of Sweden. Dissertation, Princeton University. Edin, P.A. and Fredriksson, P. (2000). LINDA — Longitudinal Individual Data for Sweden. Working Paper 2000:19, Uppsala Universitet, Nationalekonomiska institutionen. Falkingham, J. and Harding, A. (1996). Poverty Alleviation vs. Social Insurance Systems: A Comparison of Lifetime Redistribution, in Harding, A. (ed), Microsimulation and Public Policy, North-Holland, Amsterdam. Falkingham, J. and Lessof, C. (1992). Playing God: The Construction of LIFEMOD, a Dynamic Cohort Microsimulation Model, in Hancock, R. and Sutherland, H. (eds), Microsimulation Models for Public Policy Analysis: New Frontiers, STICRED Occasional Paper No. 17, London School of Economics, London. Flood, L., Jansson, F., Pettersson, T., Pettersson, T., Sundberg, O. and Westerberg, A. (2003). The Handbook of SESIM. www.sesim.org. Fritzell, J. (1994). Fo¨rdelningseffekter av offentliga tja¨nster. Ds 1994: 86, Fritzes, Stockholm. Harding, A. (1993). Lifetime Income Redistribution: Applications of a Dynamic Microsimulation Model. North-Holland, Amsterdam. Harding, A., Lloyd, R. and Warren, N. (2006). The Distribution of Taxes and Government Benefits in Australia, in Papadimitriou, D. (ed), The Distributional Effects of Government Spending and Taxation, Palgrave, New York. Harding, A., Percival, R., Schofield, D. and Walker, A. (2000). The Lifetime Distributional Impact of Government Health Outlays. Discussion Paper no. 47, National Centre for Social and Economic Modelling, University of Canberra. Husse´nius, J., Sele´n, J. (1994). Skatter och socialfo¨rsa¨kringar o¨ver livscykeln – en simuleringsmodell. Ds 1994:135, Fritzes, Stockholm. Kakwani, N.C. (1977). Applications of Lorenz Curves in Economic Analysis. Econometrica, 45(3), 719–728. Ministry of Finance. (1999). Fo¨rdelningspolitisk Redogo¨relse. Appendix 4 to The Govenrment Budget Bill 2000, Regeringens proposition (1999/00:1), Stockholm. Ministry of Finance. (2002). Fo¨rdelningspolitisk Redogo¨relse. Appendix 3 to The Government Spring Fiscal Policy Bill 2002, Regeringens proposition (2001/ 02:100), Stockholm. Nelissen, J.H.M. (1998). Annual Versus Lifetime Income Redistribution by Social Security. Journal of Public Economics, 68, 223–249. O’Donoghue, C. (2001a). Dynamic Microsimulation: A Methodological survey. Brazilian Electronic Journal of Economics, 4(2). O’Donoghue, C. (2001b). Redistribution in the Irish Tax-Benefit System. Dissertation, London School of Economics, London. Pettersson, T. and Pettersson, T. (2003). Fo¨rdelning ur ett livscykelperspektiv. Appendix 9, Long Term Survey. Ministry of Finance. Rake, K. and Zaidi, A. (2001). Dynamic Microsimulation Models: A Review and Some Lessons for SAGE, SAGE Discussion Paper no. 2 (SAGEDP/02).

Lifetime Income Redistribution

231

Rao, V.M. (1969). Two Decompositions of the Concentration Ratio. Journal of the Royal Statistical Society, 132, 418–425. Smeeding, T., Saunders, P., Coder, J., Jenkins, S., Fritzell, J., Hagenaars, A., Hauser, R. and Wolfson, M. (1993). Poverty, Inequality, and Family Living Standards Impacts across Seven Nations: The Effect of Noncash Subsidies for Health, Education and Housing. The Review of Income and Wealth, 39(3), 229–256. Statistics Sweden. (2003). Inkomstfo¨rdelningsunderso¨kningen 2001. HE 21 SM 0301.

This page intentionally left blank

Chapter 9

Income Distribution and Redistribution in a Medium-Term Perspective in Denmark Frederik Hansen Ministry of Finance, Denmark

Abstract Static microsimulation models have been used in Denmark for more than 20 years. The database for the models is very large and contains many pieces of information, mainly from administrative files. Until now the database has covered only one year. Recently another type of database — a panel — was created. We use the panel to evaluate how inequality measures depend on the length of the income period. We show that the Gini-coefficient on 5-year average incomes is around 2/3 of the single-year Gini-coefficient, and that the Gini-coefficient on life span income is around 50 per cent of the singleyear Gini. The fraction of the population in relative poverty is lower when measured on longer-run average incomes, and persistent relative poverty is almost non-existent in Denmark. We also present results of calculations of the distributional effects of an actual and a potential change in the tax legislation over a 6-year period compared to the effects measured over a single-year period. The results — based on a newly developed multi-year tax model which uses the panel — show that there may be fairly large differences in the apparent distributional effects, particularly for individuals aged 20–29 years, depending on whether the effects are measured over the short run or over the medium term.

1. Introduction Any detailed analysis of the economies of households or families requires data with more information than simple averages of the key variables.

International Symposia in Economic Theory and Econometrics, Vol. 15 r 2007 Published by Elsevier B.V. ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15009-7

234

Frederik Hansen

This has motivated the development of microeconomic databases in which there is detailed information about key economic variables (e.g. on income). Such databases contain information about the differences and variations within the population of the key variables, such as income and employment. Static microsimulation models, which employ such databases have been built to provide better answers to questions on household economics, in areas where it is perceived that the variation at the microlevel could have a significant impact on the answers. These databases have traditionally been cross-sectional single-year databases. However, the income distribution found in such a database will overestimate the ‘‘true’’ differences within the population. By this we mean that a cross-sectional sample is a snapshot of the population and the distribution of income, employment, etc. at a single point in time. But incomes vary across the life span — particularly around the time of entering the labour market as well as when leaving it (for retirement). The models would be more complete if they used a database that covered a longer period than a year, because then the database would more accurately reflect the ‘‘true’’ (i.e., longer-run) incomes etc. In other words, the snapshot would be more precise. Analyses of this kind could be performed with the aid of a dynamic microsimulation model, in which the model generates a ‘‘sample’’ of the population or a cohort over a long-time horizon (e.g. see Harding, 1993). But it can also be obtained within a static microsimulation model framework, provided that the analyses are built around panel data. In either case, the first step in building a model to perform a medium-term analysis is to create a database that gives a more complete (and more complicated) picture of the distribution of incomes and economic resources among the population than a cross-sectional sample can give — i.e., to create a panel. In addition to the variation in income etc. among the population within a year, the panel has an extra dimension compared to the cross-sectional sample — time. It contains information about the fluctuations in incomes etc. across the life span (or rather, because of the limitations in the data, across a period of time, a number of years). In this paper we present examples of calculations from various microsimulation models on both single-year data and on panel data for Denmark. In Section 2, we give a very short overview of microsimulation modelling in Denmark and describe the databases. In Section 3, we present results of income distribution in the short run, the medium term and across the life span. In Section 4, we present and compare results of the redistribution effects of a hypothetical and an actual tax change when viewed in the short run and in the medium term.

Income Distribution and Redistribution

235

2. Static Microsimulation Modelling in Denmark1 This section gives a brief description of the Danish Law Model, highlighting some elements that are particularly relevant for the examples mentioned in this paper (see Finansministeriet, 2003, for more details). Static microsimulation models have been used in the Danish central administration since the early 1980s. The models use a common database — the Law Model Population — a 3.3 per cent random sample of the Danish population. The most recent sample contains approximately 180,000 observations and the sample is not stratified because it is used for many different purposes.2 A new random sample is drawn every year. Each law model population contains a large number of pieces of information, mostly taken from administrative files. In Denmark, all persons are given a unique identification number (Central Personal Registration number). This number is used for all contacts with the public authorities and it is used as the identification number in administrative files — e.g. for income tax returns and for receipt of various kinds of transfers. During the 1990s, the need to expand the ‘‘model portfolio’’ became more and more apparent. The first step was taken in 2000, when the first version of the Law Model Panel became available — i.e., a model population in which a basic population is followed across time. A number of restrictions (motivated mainly by confidentiality issues) were laid down on the development of the first version of the law model panel. For this reason the first panel was — very unusually — constructed backwards in time (see Økonomiministeriet, 1993, 2000, for details). The restrictions posed on making the panel data were later relaxed and it became possible to make a new, traditional ‘‘forward-running’’ panel. The law model panel (or multiyear model population) is based on a 3.3 per cent sample of the original 1993 population. Each year the panel is supplemented by a (3.3 per cent) sample of newborns and immigrants (otherwise, the model population in the years after 1993 would not be representative). The type of information in the law model panel (for each individual/year) is exactly the same as in the single-year model populations; generally the

1

Originally the Ministry of Economic Affairs was responsible for the maintenance and development of the model system, which is open for use — and being used — by other ministries. Following a large reform in the Danish central administration in November 2001, the Danish Ministry of Finance has taken over these tasks. 2 A 33.3 per cent sample is available for specific studies focusing on relative minor groups of the population.

236

Frederik Hansen

same variable names and (structure of) names of datasets are used. The sample contains different individuals, however.3 Several static microsimulation models — Law Models — have been developed. The models simulate personal income taxes and various regulatory areas such as social pensions, housing benefits and payments for using day care institutions. From the mid-1990s the Ministries of Economic Affairs and Finance started calculating the number of low-income families, Ginicoefficients and various other measures of income distribution — and since then, results of such calculations have been published more or less annually (see Section 3 for details). The databases include all necessary information about taxable and non-taxable income and income taxes. The calculation of Gini-coefficients (and other distributional measures) is a mechanical calculation, following the (simple) definitions of the concepts — and it is therefore not a simulation in the true sense of the word.

3. Income Distribution 3.1 Income Distribution in a Dynamic Time Perspective Individuals’ disposable incomes are subject to short-run changes — due to changes in employment status, changes in family status or possibly because of lump-sum income, such as in the form of capital gains. In the longer run, changes may occur when a person enters the labour market (e.g. after finishing education) and when a person retires. Measures of the variation in incomes across the population within a year can therefore be said to overestimate the ‘‘actual’’ differences (as conceived over a longer period of time). Gini-Coefficients We use the law model panel to evaluate some of the traditional measures of the income distribution in the short run (1 year) and over a longer period. The methodology is straightforward. For each year in the panel an equivalised disposable income is calculated based on the information in the database on (taxable) incomes, income taxes and most of the non-taxable 3

Another difference is that the single-year law model population is a sample of households — i.e. all persons in a household are included in the sample. The panel is a sample of persons. However, because it may be necessary to have information on (all) household members for some purposes (e.g. to calculate household income), there are — in addition to the ‘‘main’’ dataset (the panel) — datasets with information about these ‘‘auxiliary’’ persons.

Income Distribution and Redistribution

Gini-coefficient

Figure 1:

237

Gini-Coefficients on Incomes for Various Time-Periods. (1993–2001)

25

25

23

23

21

21

19

19

17

17 93

94

95

96

97

98

99

00

01

All population

1-year Gini coefficients

3-year Gini coefficients

5-year Gini coefficients

Source: Calculations based on the Law Model Pane Notes: The curves for the 3-year period and the 5-year period are ‘‘centered’’, indicating that the 3-year-curve begins in 1994, as 1993 is the first year in the basic sample. It ends in 2000, as 2001 is the last year in the sample. Persons for whom income information is missing in one or more years of the relevant period are excluded in the calculations. For that reason around 4 per cent of the sample is excluded.

income. As the basic data are in current prices, disposable income is converted into base-year prices, by use of a price index.4 A 3-year average of disposable incomes is then calculated by a simple arithmetic average and is ‘‘related’’ to the middle year in the graphs. The 5-year averages are calculated in a similar way (though it is not possible to make such calculations for as many years as the 3-year averages). The available panel runs from 1993 to 2001 — i.e., a 9-year period. It would be possible to calculate Gini-coefficients for the entire period — but then it would not be possible to analyse how the 9-year coefficient changed over rolling periods and for that reason this calculation was not done. The single-year Gini-coefficients have increased slowly and continuously during almost a decade (Figure 1), indicating that income inequality has

4

In this way we do not discount for the experienced real rate of growth, so a relatively high income in the last year of the period will weigh (slightly) higher than a relatively high income in the first year. We do this because this is a historic (backward-looking) analysis, and the annual income is an indication of the ‘‘purchasing power’’ that year — and we choose to weigh that power equally in all the previous years.

Frederik Hansen

238

Figure 2: 1-, 3-, and 5-year Gini-Coefficients for Subgroups of the Population (1994–2000) Persons 66 years and older. 23

21

21

19

19

17

17

15

15 93 94 95 96 97 98 99 00 01 All population 3-year Gini-coef.

1-year Gini-coef. 5-year Gini-coef.

Gini-coefficient

Gini-coefficient

Persons aged 25-59 years. 23

23

23

21

21

19

19

17

17 15

15 93 94 95 96 97 98 99 00 01 All population 3-year Gini-coef.

1-year Gini-coef. 5-year Gini-coef.

Note: See Note for Table 1.

increased somewhat. The present level of the Gini-coefficient is still, however, low compared to other countries –(see Burniaux et al., 1998; Fo¨rster and Pearson, 2002; Fo¨rster and d’Ercole, 2005). The levels of the medium-term Gini-coefficients are as expected: The longer the measurement period, the lower the Gini-coefficient. The 3-year average Gini-coefficient is 1.4 percentage points (or around 6 per cent) lower than the ‘‘single-year’’ Gini coefficient, and the 5-year average Ginicoefficient is still lower. The trend in the (single-year) income distribution over the last decade — an increasing Gini-coefficient — is also found in the rolling 3- and 5-year averages. The path of the single-year Gini-coefficients across the years is the least smooth. This is probably an indication of the fact that temporary variations in disposable income have larger effects on income distribution in the short run than in the long run. Temporary variations in income are more common (or have a larger effect) for people in the core working ages (25–59 years) than for pensioners (persons over 66 years) –(see Figure 2). The difference between the singleyear average Gini-coefficient and the 3-year average coefficient for persons over 66 years is rather small, less than 1 percentage point, and even less in the recent years. This is not surprising as pensioners have more stable incomes than people of working age. Low-Income Families The law model population has also been used to evaluate the magnitude of ‘‘relative (economic) poverty’’, a measure that is also referred to as the number of low-income families or rather persons living in low-income families. These numbers have been calculated following the standard methodology — i.e., we create a ‘‘head count’’ of persons living in families with a

Income Distribution and Redistribution

239

disposable income below a relative limit, set at 50 (or sometimes 60) per cent of the median income of the entire population.5 The first calculations of relative poverty in Denmark were published in Økonomiministeriet (1993). The study showed — like subsequent studies — that the low-income group is relatively small in Denmark — 5–7 per cent, depending on the parameters of the equivalence scale and the definition of the family or household (in particular whether or not to include older children). A large fraction of people in the low-income group are students or very young people and self-employed. Students will not remain students for long (although in a 5-year period a non-negligible number of persons are studying most of the time) and the self-employed may have temporary ‘‘bad’’ years or may just have set up business. These considerations suggest that basing the calculation of low-income families on (disposable) income for a longer period than 1 year will result in a lower ‘‘head count’’ of persons in low-income families. Earlier studies — e.g. that used the first version of the panel, such as Økonomiministeriet (2000) and Hansen et al. (2000) — focused on how persistent relative poverty was: i.e., for how long did individuals remain in the low-income group. The study showed that 15 per cent of the persons in the low-income group in 1994 were also in the low-income group in 1998. The outflow from the group is rapid, as more than half of the persons in the group are not in it the following year and only 6 per cent remain after 5 years. This means that only around 0.2 per cent of the population was in the low-income group every year. We supplement these findings with analyses of low-income families based on the new panel data. The number of low-income families is first calculated for a single-year period (2001), and then for longer periods (a 3- and a 5-year period). That is, first we calculate average disposable incomes for the 3- (or 5) year period. A median income for the period is then calculated, and finally we create a head count of the number of individuals with an average disposable income below 50 per cent of this median. On this basis, 4.6 per cent of the population is in the low-income group if incomes are calculated for a single-year period — and only half of that, 2.3 per cent, if 5-year average incomes are used for the basis of comparison (Table 1). The result is hardly surprising, given the results in the previous section and the earlier findings, which indicate that the equality of the income distribution increases with the length of the income period.

5

In some countries a specific low-income limit is set in the legislation, which can be used for such an evaluation, but there is no such limit in Denmark.

Frederik Hansen

240 Table 1:

Percentage of the Population in Low-Income Families in Various Periods

Limit for low-income group

50 per cent of median

60 per cent of median

2001 1999–2001 1997–2001 2001 1999–2001 1997–2001 1 year 3 years 5 years 1 year 3 years 5 years With registered income every year 1997–2001 With registered income in relevant period

4.1

2.9

2.4

8.2

6.9

6.2

4.9

3.2

2.4

9.2

7.3

6.2

Source: Calculations based on the Law Model Panel Note: Equivalence scale 0.6.

3.2 Income Distribution in the Short Run and over the Life Span Disposable income varies over the life span. For most people income will increase up to the age of 40–50 years and, after an age of around 60, disposable incomes will decline gradually. A large part of the income inequality measured in the previous sections is caused by the fact that we compare individuals in different age groups. Using 3- (rolling)or 5-year averages will to some extent reduce the effects of temporary fluctuations. Nevertheless, even a 5-year period is short relative to the life span. In order to analyse lifetime incomes, in particular their distribution and composition, a model that generates incomes across the life span has been developed. A first version of the model has recently been completed in the law model system.6 We have used this model to calculate an average lifespan income as the average of the (equivalised family-) incomes in all years, thereby taking the demographic changes that occur over the lifetime into account. The model is based on the panel data, essentially a sample of actual (8-year) paths of income history; using statistical matching, a large number (50,000) life-paths of income have been constructed. The basic principle is to take a known ‘‘part of a life-path’’ — i.e., a path that is found in the panel — and let that path ‘‘continue’’, by linking it to the life-path (for a different period of life) of another individual in the sample by statistical matching. A summary of the description of the matching technique can be found in the appendix to this chapter. (For a more detailed description and other technical aspects see Finansministeriet, 2004, pp. 84–92.)

6

Following a model development in the Economic Council, cf. Det Økonomiske Ra˚d (2001). This report also contains a number of analyses of various themes that are not addressed in this paper.

Income Distribution and Redistribution Figure 3:

0

30

241

The Distribution of Disposable Income in One Year and over the Life Span

60

90

120 150 180 210 240 270 300 330 360 390 420 450 One-year income

Life span income

Source: Calculations based on the Law Model Panel

In this paper, we primarily focus on how measures of income distribution — based on life span incomes — differ from the measures in the previous sections. Figure 3 compares the income distribution within a single-year and over the life span. The graph shows simple frequencies — i.e., the number of persons with disposable income (single-year or life span) within a fairly narrow range. The number of persons is not indicated on the axis, as this depends on the length of the range, which is not relevant as such. It is very clear that disposable income is more equally distributed over the lifetime than in a given year. A similar pattern for income distribution in Sweden over one year and over the life span has been reported in a report by the Swedish Ministry of Finance (see Finansdepartementet, 2003, for details). A similar conclusion can be drawn from Table 2, which shows average disposable income in all deciles, in both the single-year income distribution and the life span income distribution. Disposable incomes measured over the life span (compared to incomes measured in one year) are higher at the low end of the distribution and lower at the high end of the distribution. The table also shows a summary measure of income distribution — the D90/ D10-fraction. It is the ratio of the highest income in the 9th decile and the highest income of the 1st decile. In general, the higher the ratio, the more unequal is the distribution (but it is a rather crude measure, because it only takes into account the incomes at two points in the distribution).

Frederik Hansen

242

Table 2: Equivalised Disposable Income across Deciles over the Life Span and in a Single Year Decile

Over life span In a single year Difference

All

1

2

3

4

5

6

7

8

9

10

112 66 46

130 100 30

140 115 25

148 130 18

155 144 11

163 159 4

171 175 4

182 195 13

197 222 25

249 327 78

165 163 2

Source: Calculations based on the Law Model Panel (enlarged, 10 per cent of the population) Note: Equivalence scale, 0.6; life span income from 18 to 90 years. Average income is 1,000 DKK.

Table 3:

Lifetime and Annual Gini-Coefficients

1-year incomes All over 18 years 18–70 year old 25–59 year old 50 year old Lifetime income 18–90 year old 18–70 year old 25–59 year old

Annual average income DKK

Annual average income, Gini

Sum of lifetime income, Gini

150,600 153,900 164,100 190,000

25.0 25.0 22.4 21.3

— — — —

153,800 156,800 167,000

12.8 11.2 11.7

16.8 11.2 11.7

Source: Calculations based on the Law Model Panel (enlarged, 10 per cent of the population)

Nonetheless, when measured over a single year the ratio is 2.6, compared with 1.7 when measured over the life span — i.e., around 1/3 lower. Another, more frequently used measure of the income distribution is the Gini-coefficient. The Gini-coefficient for average lifetime incomes is 12.8 per cent — or around half of the single-year value. A part of the difference in the lifetime Gini-coefficient is due to the fact that people do not live the same number of years (see Table 3). Because the income of pensioners is relatively low, persons that live for many years will receive relatively lower average lifetime incomes, because income in the later years of life is (usually) relatively low. The life span Gini-coefficient for average income in the years 18–70 (for those that live that long) is 11.2 per cent. An alternative way to measure the life span Gini-coefficients is to calculate the sum of yearly incomes. This would obviously give persons dying young a very low (relative) income, and — though perhaps not in the same degree — a relatively high lifetime income for those individuals that die at a high age.

Income Distribution and Redistribution

243

This effect can be seen in the last column of Table 3. For the population as a whole the Gini-coefficient based on the sum of life span incomes is 16.4 per cent. This is still lower than the Gini-coefficient based on the most recent 5-year average incomes, which is around 21 per cent (see Figure 2). Similar results on the reduction of the Gini-coefficient as the income period is increased have been found in other studies for other countries, e.g. for Sweden (Finansdepartementet, 1994). A relatively similar gap between lifetime and annual income was also found by Harding (1993, p. 151) for Australia, where the annual income Gini-coefficient was about 54 per cent of the lifetime income Gini. Those studies did not, however, use data from a panel with income data from administrative files for the entire period. Income was either modelled (by a dynamic model) or generated using panel data for a shorter period. 3.3 Low Income-Families in the Short Run and over the Life Span The number of persons in low-income families defined using lifetime incomes (persons with an equivalent income of less than 50 per cent of the median lifetime income) is only a small fraction of the ‘‘traditional’’ number based on one-year incomes. All the results indicate that relative poverty is not persistent in Denmark, although there are around 0.1 per cent of the population in the low-income group more than half of their life span, and around 1/2 per cent below this level for at least 30 per cent of their life span (Table 4). In addition, it can be noted that 12.5 per cent of the population — 1 out of 8 — have a disposable income more than 30 per cent below the median income for more than 30 per cent of their life span. Table 4:

The Fraction of Life Span with Incomes below Low-Income Limit

Fraction of life span with income below low-income limit

50 per cent of median income

70 per cent of median income

Never o10 per cent 10–20 per cent 20–30 per cent 30–40 per cent 40–50 per cent Over 50 per cent Fraction of the population with lifetime income below

39.2 51.3 7.7 1.3 0.4 0.1 0.1 0.1

8.2 33.4 30.1 16.0 7.7 2.6 2.2 3.6

Source: Calculations based on the Law Model Panel (enlarged, 10 per cent of the population) Notes: Equivalence scale, 0.6; life span income from 18 to 90 years, indicating that 10 per cent of life span approximately amounts to 8 years.

244

Frederik Hansen

4. Recent Extensions to the Static Tax Model There have been many tax reforms or minor changes in the income tax systems of OECD countries during the last few decades. In Denmark, there have been three tax reforms — in 1986/1987, 1993/1994 and 1998/1999. All of these reforms were implemented gradually. In 2004 the income tax system was changed, but the change was not of a nature, which implies that it could be labelled ‘‘a reform’’ (and was not called so officially). Details of the latest amendments to the tax legislation are presented below. The tax module in the law model system has been used intensively in the preparation of all the above-mentioned reforms (e.g. see Foxman, 2000, on the 1993/1994 reform). There has been high demand for results of revenue and distributional effects of the various proposals during the preparation phases of the above changes. The results presented are usually: 



Revenue consequences (i.e., budgetary effects or net tax cuts/net tax increase) Distributional effects (the number of ‘‘winners’’ and ‘‘losers’’, e.g. by amount of gain or loss and/or by demographic characteristics).

Recently, there have been increasing efforts to analyse the behavioural effects of changes in the tax legislation (particularly the effects on labour supply). There have also been demands to undertake more detailed analysis of the long-run effects of tax reforms. This demand has led to developments in the tax model (or more precisely the development of additional analytical facilities to ‘‘add on’’ to the model) in two different directions — one of which is to extend the model to a multi-year model, based on the panel data. At the moment this model extension is not fully implemented. In the section below, we present some preliminary results from a prototype of a multi-year tax model. The other development has been to incorporate behavioural effects into the calculations.7

7

The other development is to include behavioural effects, particularly changes in the labour supply, as a response to tax changes. The principal aim of extending the analysis in this direction is to get a more precise estimate of the budgetary effects: Taking into account that e.g. lower tax rates may increase work participation — and hence, since the increased participation will increase the tax base, behavioural effects could partly or fully counterbalance the initial budgetary effects. The effects on work supply enter the model in two ways: By the number of hours worked — for the part of the population that is already employed — and by an increased number of workers. For an outline of a preliminary method see Finansministeriet (2002). The method is under revision, however.

Income Distribution and Redistribution Table 5:

Short-Run and Medium-Term Average Incomes by Age

Age group Number of tax Short-run Medium-term (years) payers (‘000) ( ¼ income 1999) ( ¼ average income (‘000 DKK) 1999–2004) (‘000 DKK) o20 20–24 25–29 30–39 40–49 50–59 60–66 67–75 4 ¼ 76 All

245

386 354 385 784 789 635 292 296 182 4,103

26 124 176 207 220 198 147 119 112 165

64 151 195 216 221 190 140 120 114 173

Difference

38 37 19 9 1 8 7 1 2 8

Source: Calculations based on the Law Model Panel Notes: All incomes are in 1999-prices. Persons that have died or emigrated during the 6-year period are excluded.

4.1 A Multi-Year Tax Model In a static single-year model, the effects of a change in the tax legislation for an individual depend primarily on the income level of the individual and its composition. If the income level in the single-year period differs greatly from the income level the individual will have in the medium term, then the distributional effects will be poorly reflected by calculations based on the singleyear static tax model. The results in Table 5 indicate that such differences could be particularly important for people below 29 years, because their shortrun income on average is considerably lower than their medium-term income. The multi-year tax model resembles the ordinary single-year tax model in many ways — or rather a number of such tax models. The single-year tax model is based on the traditional law model population, i.e., a single-year sample (or a cross-sectional sample). The multi-year tax model is based on the new panel data. It should be underlined that the model population is based on administrative data — and that all citizens in Denmark have to file a tax return. The sample of the tax files is therefore by definition representative (although some statistical noise for a few variables can be detected) — and the problem of ‘‘non-filers’’ or non-respondents, relevant in many other countries and/or types of calculations, is not an issue. Figure 4 describes the main steps of the calculations in the single-year tax model and in the multi-year tax model. As in any microsimulation model, an important part of the model is to ‘‘adjust’’ the registered historic incomes to

Frederik Hansen

246 Figure 4:

The One- and Multi-Year Tax Models The one year model Adjustment (mainly updating of incomes etc.)

Model population, year T

Tax calculation

Calculation of weights

Adjusted database with calculated taxes

Adjusted data base (year T+f)

Adjusted database for analysis

The multi year model Adjustments Updating of incomes etc. + Additional adjustments

Model population, Base year Year (B)

Model population, Year (B+1)

Tax calculation

Calculation of weights

Adjusted database with calculated taxes Year (B+f)

Adjusted database for analysis

Adjusted database Year (B+1+f)

Adjusted database with calculated taxes Year (B+1+f)

Adjusted database for analysis

Adjusted database Year (B+n+f)

Adjusted database with calculated taxes Year (B+n+f)

Adjusted database for analysis

Adjusted database Year (B+f)

Etc.

Model population, Year (B+n)

Notes: f is the number of years from the population base year to the calculation year. N, the number of years between the oldest and the newest in the panel. Note that the first year in the panel is several years before the present year.

the income level in the calculation year. In the single-year tax model all the registered incomes (by type) are uprated from the base year (model population year) to the calculation year (for policy relevant calculations usually the year after the actual year). Usually no other adjustments are made, although it should be added that the calculation of weights, which takes place later in the model, could be considered an adjustment.

Income Distribution and Redistribution

247

In the multi-year tax model, the incomes for each year in the panel are uprated individually and simultaneously. The uprating of each year’s incomes is done in the same manner as in a single-year tax model (based on a law model database from that year). In the multi-year tax model it may be necessary to make additional adjustments (see the appendix for details). The tax calculation is in a way the most straightforward part of the model, although the translation of the tax legislation into programme language results in thousands of lines of programme code. This part of the tax model includes the setting and general handling of tax parameters (rates, schedules, etc.). The final part of the tax model is the calculation of weights to be attributed to the observations in the database. The simplest way of calculating weights in a microsimulation model would be to attribute the same weight to all observations — and set the weight according to simple macroeconomic indicators (e.g. by dividing the total number of ‘‘relevant’’ individuals with the number of (relevant) observations in the sample). In the present version of the single-year tax model a technically more complicated method — based on calibration — is used. The result of the ‘‘calibration’’ is that individual weights are calculated — i.e., we calculate a number, w j, for each individual. The interpretation of the weights is straightforward; observation j in the sample represents w j individuals in the population. At the moment we have developed the model to the point at which the panel is aged from 1995/2000 to 1999/2004. In order to make calculations for the years beyond 2004, detailed assumptions on the projection of income variables are required (cf. Appendix 3). The weight ‘‘follows’’ an individual over the whole period — i.e., the weight that is calculated for individual j — (w j) — is the same in all years. 4.2 Simulations Performed on the Multi-Year Tax Model In order to compare results of the different types of model calculations (in particular the differences in the distributional effects), we first simulate income taxes in the short run — using the single-year model (base: 1995, projected to 1999)8 — and in the medium term — using the multi-year 8

We do not use a traditional single-year tax model for the short-run analysis. Instead, we use a single-year tax model based on the panel data. By using the panel for both calculations, we avoid some statistical noise, because the population base of the two models are — except for migration and death — the same — and the tax change for an individual in the single-year model will by definition be equal to the tax change result for the first of the 6 years in the multi-year tax model.

Frederik Hansen

248 Table 6:

Simplified Tax Rates in the Danish Tax System, 2003

Income range (DKK)

Tax rate (per cent) (including social security)

0–40,000 40,000–210,000 210,000–310,000 4310,000

8.0 43.7 49.2 62.9

Note: Thresholds are rounded to the nearest 10,000 DKK.

model (base: 1995/2000, projected to 1999–2004). In the multi-year tax model, income tax is calculated for all years in the projection period. Next, average real incomes and taxes over the projection period are calculated for each individual (in the 1999 price-level, in order to make them comparable to the results from the single-year model). These are the central results of the multi-year tax model, and when we refer to taxes in the multi-year tax model we mean the average real taxes over the period 1999–2004. Brief Description of the Danish Tax System The Danish income tax system has three rates, plus a zero rate (excluding social security contributions) that applies to very low incomes (see Table 6). Three different income bases are used but in a simplified presentation this fact does not matter. The description is fully correct for people with no capital income and no deductions. For others there may be differences, but the difference between marginal tax rates in the simplified system and in the rates in the ‘‘real world’’ real system is not substantial. Simulation Example The simulation example is equivalent to the tax changes that were adopted by the parliament in June 2003 and which are to be implemented over the period 2004–2007. In the simulation, 2007-rules are used throughout. The changes consist of two elements: an increase in the 2nd threshold by 50,000 DKK (Danish kroner, the Danish currency unit) (1999-prices) and the introduction of a tax allowance to the employed. The after-tax value of the tax allowance is around 2,000 DKK. It should be noted that in this simulation example (and in the adopted changes in the tax legislation) there are no losers. The size of the gain is limited. The revenue effect is a reduction of income taxes by 9,000 Million DKK — approximately 4 per cent of total revenue. In the example, the revenue effect in the one-year model and in the multiyear model is almost the same. The reason that they are not the same is that

Income Distribution and Redistribution Table 7:

249

Average Gain from the Tax Change by Age Group

Age group Number of tax Average change short- Average change medium- Difference (year) payers (1,000) run/1-year model run/multi-year model (DKK) (DKK) (DKK) o20 20–24 25–29 30–39 40–49 50–59 60–66 67–75 4 ¼ 76 All

386 354 385 784 789 635 292 296 182 4,103

184 1,377 2,405 3,081 3,340 2,732 1,108 416 252 2,136

549 1,815 2,649 3,089 3,180 2,350 780 365 247 2,114

365 438 244 8 160 382 328 51 5 (22)

the (real) income level and the income distribution in the short run and in the medium term are different. The difference in the income level is rather small, though, which may be partly explained by the fact that the level of employment and the participation rate have been relatively stable over the projection period (1999–2004) (Table 7). Distributional Effects The main implications of the simulation are that (a) the changes in tax legislation do not produce losers; (b) all persons with income from employment will have a gain, while persons with no income from employment usually will not be affected; and (c) the gain will be higher for persons with higher incomes, but size of the gain is limited. The average gain increases with age until 40–49 years, and decreases for higher age groups — both in a short-run and a medium-term perspective. These results are hardly surprising when the initial conditions about income distribution and work participation rates across the age groups are taken into account. It is also worth noting that the medium-run changes are higher than the short-run changes for people below 30 years and lower for persons aged 40–66 years and around the same for those older than 66 years. The first of these results is explained by the fact that within these age groups work participation rates increase by age, as does income. Note that the fraction of people with no change in taxes is much lower in the medium run than in the short run. In particular, this result holds for persons aged less than 20. It is also clear that persons in the 50 and above age groups have higher tax gains in the short run than in the medium run, because some of them retire, which reduces their labour income to zero (Table 8).

250

Table 8:

Taxpayers by Tax Change and Age Group

Age group (year)

Short-run/1-year model Change +/-100 71 16 18 14 13 25 60 79 85 33 29

Gain 2–5,000

——pct.—— 28 1 63 21 32 47 24 52 20 52 21 41 18 17 11 8 9 6 25 35 25 38

Source: Calculations based on the Law Model Panel

Gain 45,000

Change +/-100

Gain 100–2,000

0 0 3 10 14 13 5 2 1 7 8

24 6 7 6 9 22 58 78 83 23 23

71 55 34 26 23 29 27 14 11 32 28

Gain 2–5,000

All Gain 45,000

——pct.— 4 38 54 57 55 40 13 7 6 38 42

0 1 5 10 13 9 2 1 0 7 7

- 1,000 386 353 385 784 789 634 292 296 182 4,103 3,717

Frederik Hansen

o20 20–24 25–29 30–39 40–49 50–59 60–66 67–75 4 ¼ 76 All All 419

Gain 100–2,000

Medium-run/multi-year model

Income Distribution and Redistribution

251

The result for the individuals aged 40–49 is the most difficult to explain. The medium-run gain is lower than the short-run gain, although there are fewer persons with no tax change in the medium-run. (This would in itself lead to the medium-run average tax gain being the highest). However, the nature of the tax change is such that for people with relatively high income, an (large) increase in income in a subsequent year (or in the medium run) would not imply a higher gain, as this is by definition limited. However, a (large) decrease in income — which could be a result of unemployment, or retirement because of disability — would imply a lower gain. In short, the ‘‘possible’’ changes in the gain, when going from a shortrun perspective to a medium-run perspective, are in a certain sense out of balance. It is important to stress that this result follows primarily from the exact nature of the tax change — in particular the fact that there is a limit to the gain from the tax change as well as at what income level that limit takes effect.

5. Conclusions All the results in this chapter show that it is important to take the ‘‘timehorizon’’ into account. It is well known that the level of the Gini-coefficient holds little information in itself. It should be compared either to coefficients in other countries or coefficients within the country at another point in time. We have observed, not surprisingly, that Gini-coefficients are lower the longer the length of the income period. We have also observed that the trend in income distribution over the last decade does not depend on the length of the income period. The results concerning the size of the low-income group are quite remarkable. It is well established that the low-income group is relatively small in Denmark and that many individuals in the group have characteristics that indicate that they are there only temporarily. The calculations show that in a medium-run perspective, the number of relatively poor is much lower — and, in particular, when viewed over a lifetime, the number of such individuals is very low. Observations on distributional effects in the medium term compared to the short-run effects include that: 

 

results depend on the type of change that is analysed. There is no universal answer to the differences between the two; the differences do not seem to be very large in general; and for some groups differences can be important, in particular for the relatively young (20–29-year olds).

252

Frederik Hansen

It is our conclusion on this subject that it may be important to supplement single-year calculations by multi-year calculations — in particular, if it is relevant to show that the number of losers or non-affected is much higher in the short run than in the medium run.

Acknowledgements A number of my colleagues in the Ministry of Finance have assisted in the preparation of the calculations presented in this paper, in particular Lars Pantmann, Carl-Christian Heiberg and Anita Lindberg. Peter BachMortensen has provided comments to earlier drafts. I would also like to thank an anonymous referee for valuable comments to the paper presented at the conference, which have led me to make some revisions in the paper. The comments and opinions expressed are those of the author and not necessarily shared by the Ministry of Finance.

References Burniaux, J.M., Dang, T.T., Fore, D., Fo¨rster, M., d’Ercole, M. and Oxley, H. (1998). Income Distribution and Poverty in Selected OECD countries, OECD Economic Department Working Paper 189, Paris. Det Økonomiske Ra˚d. (2001). Dansk Økonomi Eftera˚r. Det Økonomiske Ra˚d, Copenhagen. Finansdepartementet. (1994). Skatter och Socialfo¨rsa¨kring o¨ver Livscykeln — En Simuleringsmodel, Rapport til ESO, Ds 1994:135, Finansdepartementet, Stockholm. Finansdepartementet (2003). Fo¨rdeling ur ett Livscykelperspektiv, Appendix to the Swedish Long Term Survey of 2003. Finansdepartementet, Stockholm. Finansministeriet (2002). Fordeling og Incitamenter. Finansministeriet, Copenhagen. Finansministeriet (2003). The Law Model. Finansministeriet, Copenhagen. Finansministeriet (2004). Fordeling og Incitamenter. Finansministeriet, Copenhagen. Fo¨rster, M. and Pearson, M. (2002). Income Distribution and Poverty in the OECD Area: Trends and Driving Forces. OECD Economic Studies no. 24, Paris. Fo¨rster, M. and d’Ercole, M. (2005). Income Distribution and Poverty in OECD Countries in the Second Half of the 1990s, OECD Social, Employment and Migration Working Papers No. 22, Paris. Foxman, P. (2000). The Use of Microsimulation Models for the Danish Tax Reform Act of 1993, in Gupta, A. and Kapur, V. (eds), Microsimulation in Government Policy and Forecasting, Elsevier, Amsterdam, pp. 85–94. Hansen, F., Pantmann, L. and Vestergaard, B. (2000). Income Distribution in a Dynamic View, 6th Nordic Seminar on Microsimulation Models, Copenhagen. Hansen, F. (2001). The Danish Microsimulation Tax Model. Recent Changes in Methodology and Considerations on Future Developments, 5th Conference on Financial Information, the Use of Microsimulation Methods in the Public Sector, Budapest (available from www.itm.bme.hu.).

Income Distribution and Redistribution

253

Harding, A. (1993). Lifetime Income Distribution and Redistribution: Applications of a Microsimulation Model, Contributions to Economic Analysis Series, NorthHolland, Amsterdam. Økonomiministeriet (1993). Lovmodel Oktober. Økonomiministeriet, Copenhagen. Økonomiministeriet (2000). Familier og Indkomster. Økonomiministeriet, Copenhagen, November. The publications by Finansdepartementet (Swedish Ministry of Finance). (2003) are accessible on the website www.sesim.org. The publications by Finansministeriet (Danish Ministry of Finance) are accessible on the web site of the Ministry, www.fm.dk. The publications by Økonomiministeriet (Former Ministry of Economic Affairs) and Hansen, Pantmann and Vestergaard (2000) are accessible on the web via the web site of the former Ministry of Economic Affairs:http://www.statensnet.dk/ pligtarkiv/fremvis.pl?vaerkid=7012&reprid=0&filid=15&iarkiv=1

Appendix 1: Procedures to Generate Life Span Income As mentioned the basic data are organised as a panel, with information about income and other characteristics for individuals from 1995 to 2002. The data are representative for all years. An individual that dies or emigrates ‘‘disappears’’ from the panel. Each year the panel is supplemented by the relevant number of new-borns and immigrants. In addition there is — for each year — information about other persons from the household, including the incomes of these persons. From the basic data it is possible to extract partial life-paths; they contain an income history. Note that some of the life-paths terminate ‘‘before time’’. Such terminating life-paths represent persons that die or immigrate. Note that the number of terminating paths depends on historic mortality and emigration rates. The basic population that is used for the analysis consists only of individuals aged 47–53 years in 2002. Fictitious life span incomes are made by combining life-paths using a statistical match. The basic data (i.e., the panel) include 8 years — which may be expressed as 1+7 years, because one of the years is used to match. The life-path of an individual that is 32 years in 2002 is matched with a life-path of a person that is 32 years old in 1995; from the latter the income data for 1996–2002 are used. That life-path ends in 2002, where the person is 39 years old (unless he has died or emigrated) — and a new match is made with a 39-year-old individual in the 1995 data. The process is repeated until we have fictitious life spans for a period (up to) 18–90 years. Note that the fictitious life spans that are finally constructed include both actual data (for 1995–2002), plus life-paths added on for later years and life-paths added on for earlier years. It should be emphasised that these paths are based on historic developments, (including, e.g. historic exit rates from unemployment to employment) and on earlier legislation, e.g. income taxes in 1996 have not been recalculated using 2002-rules.

254

Frederik Hansen

In order to make as realistic matches as possible, for each age there are (up to) 432 groups defined based on the following criteria: Sex (Male, Female); Number of adults in Family (1, 2); Number of children in family (0, 1, 2); Level of education (6 different, including persons at present in education); Type of dwelling (House owners, Tenants); and Level of capital income (within the family) (3 different groups are specified, which would roughly divide the total population evenly). The final match (of a life-path found in 2002) is made by ‘‘linking’’ the path to a life-path commencing in 1995 with characteristics (in that year) as close as possible and — within the group — a (size-adjusted) disposable income as close as possible to the 2002-individual. In the calculations we, mainly, use average incomes over the life of the individuals. An alternative approach would be to use the sum of income over the (actual) life span, but that would mean rather large variation in the cumulated (disposable) income. The general difficulty in comparing income levels does not, however, disappear completely by using the average approach. Testing the method: There is no statistical information available about life span incomes. And even if there were, it would be difficult to interpret (and of little value), because the income(s) that could be extracted would be a function of (ancient) historic developments — e.g. a function of an economy with much lower labour force participation rates by women, a less-developed technology, higher birth and mortality rates, etc. It would not be possible to use such data for a forward-looking analysis (unless the economy was in an almost complete static state — and expected to remain that way). Because of these limitations, it is also not possible to make a direct test of the life span incomes. A number of indirect tests of the model using Monte Carlo simulations have been carried out –(cf. Finansministeriet, 2004, pp. 84–92). The tests indicate that the life span income model is fairly robust, although in principle the model could underestimate the life span Gini coefficients significantly if there were some variables (factors) with a high significance for the development of the year-to-year value of the disposable income that are not used in the basic matching.

Appendix 2: Calibration Method in the Multi-Year Tax Model The purpose of the calibration is to calculate weights for all individuals in the sample. The interpretation of a weight is straightforward; observation j in the sample represents w j variables in the population. Note that weights can be calculated for different (projection) years, and the weight of a given individual will not be the same in every year.

Income Distribution and Redistribution

255

The single-year model is calibrated against a number of macroeconomic variables, V1,y,Vn. Each of these variables are reflected in the model population — i.e., on the microlevel by the corresponding variables v1,y, vn (an implicit subscript j, denoting individual j, is assumed). The macroeconomic variables used for the calibration are used in macroeconomic forecasting, or are results from such forecasts. They ‘‘originate’’ mainly from the tax statistics; for historic years they are the aggregates of (some of) the (central) variables in the administrative tax files. We have chosen to calibrate against income variables rather than against ‘‘numbers of individuals’’ (in various groups) because, for the present year and the following years, there are estimates of the (macro) variables readily available, as these are projected as an integrated part of a macroeconomic short-term forecast. Also we calibrate the model (and its microvariables) against macrovariables with exactly the same definition. As mentioned, the ultimate aim of the calibration is to calculate the weights of all individuals in the sample; they are calculated so that they meet the criteria: X  w j  vi ¼ V i for all n variables ði ¼ 1 to nÞ The criteria mean that once the sample is ‘‘added up’’ (using the individual weights) in order to represent the entire population, the model’s results ‘‘hit’’ all the target variables, i.e., all the n variables used for the calibration — see (Hansen, 2001), Appendix 3 for a technical outline of the method. There are two variants of the setting of the macrovariables used for the calibration: (1) The database is calibrated against aggregated income variables (and perhaps other relevant variables as well) in all years of the projection period simultaneously. Such version requires that the aggregated variables are available. (2) The database is calibrated against the first year of the projection period only. In the following years the weight of an individual is retained, i.e., individual j represents w j persons in every year. There are three main variants of the length of the projection period, none, medium- and long run. In a model without projection, it is by definition not necessary to uprate and to adjust the income variables. However, the entire projection period would be historic and the first year in the projection period would be several years back in time — in the examples in this paper, almost 10 years back. Such a model variant would not be useful in a forwardlooking analysis, even though the database is the most up-to-date panel. In practise the multi-year tax model must include a projection, and the projection should be for quite a long period: The first year of the panel should at

256

Frederik Hansen

least be projected for long enough that the last year of the panel will be in the future. A part of the development of the model consisted in evaluating the combined effects of the projection of income, the fine-tuning of the additional adjustments and of the calculation of the weights. Note that in a firstyear-calibration variant, the (relevant) income variables are ‘‘tied’’ in the first year of the projection period only. In the following years these variables are in principle ‘‘unrestricted’’. The model calculates it as there is information on the incomes of the individuals and their weights. The income of the individuals in the following years are determined by the income adjustment factors, in particular by the factor(s) that project the microincomes after the first year in the projection period; the additional adjustments (which e.g. ‘move’ individuals between employment and unemployment (cf. Section 3 for details)) also play an important role. An important step in the development of this variant of the model was to test that it was possible to get the ‘‘unrestricted’’ projection of the multi-year model (from 1999 to 2004) to match the known (and projected) key variables from the most recent macroeconomic projections. Such a result should not be too difficult to get after all, because the adjustments and the projection factors help keep the model roughly on track. The tests of the first fine-tuned version of the model were sufficiently acceptable.

Appendix 3: The Interaction between the Projection of Incomes and the Calculation of Weights In the single-year tax model, the uprating of incomes and the calculation of weights are closely connected. Notice that in this model, the ‘‘newest’’ version of the database can be used for the calculation of taxes in several years, beginning in the year of the population itself, and going on as far into the future as it is possible to adjust the income variables and/or calculate weights. In calculations for consecutive years, the weights of the individuals will not be the same. Somewhat simplified, the adjustments of income and the calculation of weights are related in the following way: Once the level of the macrovariables used in the calibration has been set (for consecutive years), the overall growth rate of those variables is also determined, e.g. the growth rate of total wages. This growth rate can — slightly simplified — be decomposed in two parts, a rate of growth in the wage rate ( ¼ hourly wage) and the rate of increase in the number of employed. In the single-year tax model, the uprating of the wages is done directly by the setting of the uprating factors (for hourly wage). An increase in employment will appear as an increase in the weights of the observations

Income Distribution and Redistribution

257

in the sample that have the characteristic being employed. (This increase can be seen only when results of model calculations for two or more consecutive years are compared.) This implies that if an uprating factor is marginally increased, and the corresponding macrovariable used for the calibration is not changed, then the weights (of the individuals with the type of income in question) will be reduced. Usually, the tax model uses a database that is aged 2–3 years. The database is — as noted before — representative in the base year, but if the employment rate and unemployment rate change, then the projected dataset ( ¼ the uprated and projected model population) will not be directly representative in that later year. We use the calibration to correct for such historic changes — i.e., to correct for the fact that the database in the first year in the sample is not representative for the first year in the projection period. In Denmark, the rate of employment increased and the rate of unemployment dropped from 1995 to 2000. For that reason, the weight of the employed in the 1995 dataset will be relatively high when used in a 1999-model (i.e., the weights for the employed will be higher than in 1995, while the reverse will hold for the unemployed). By using higher weights for the employed (and lower weights for the unemployed) we correct for the fact that the (projected) database is not representative in 1999. However, this procedure has a side effect which must be addressed. In the operational variants of the multi-year tax model the weights are calculated on the basis of macrovariables in the first year of the projection period (i.e., on the (adjusted) first year of the panel). Subsequent changes in the individual’s income or employment status (i.e., the level of income and their status in 1996, 1997, etc.) have, by definition, no influence on the weights. The development in employment/unemployment — as it actually happened in the historic years and therefore as reflected in the panel data — will ‘‘automatically’’ be transferred to the adjusted panel. In the grossed-up results in such a model the changes in unemployment rates in the past will trace over to the future. If the (calculated) levels of the (grossed-up) variables in the later years of the panel do not correspond with the historic empirical data (or with projections for future years) then additional adjustments in the multi-year tax model may be warranted to get the database ‘‘on track’’. This will occur if the trend in, for example, unemployment rates changes — which has happened in Denmark, where the rather sharp reduction in unemployment rates stopped around 2000. In the version of the tax model that is operational, the adjustment consists of ‘‘making a number of the employed in 1996/2000 unemployed’’ (as the model population was aged to 2000/2004) — in order to align the path of the calculated number of employed/unemployed with the actual (or expected) development from 2000 to 2004.

258

Frederik Hansen

This is done in a separate module — we change randomly a number of (fully) employed into (fully) unemployed or vice versa. The number of changes is set in a way that reflects the changes over the projection period. It is important to include this adjustment, otherwise the model can get on a track which is not compatible with the latest macroeconomic forecasts. If there are no such forecasts available, the model will in principle follow the track that is determined by the historic fluctuations within the panel.

Chapter 10

Population Ageing and Fiscal Sustainability: Integrating Detailed Labour Supply Models with CGE Models Rolf Aaberge, Ugo Colombino, Erling Holmøy, Birger Strøm and Tom Wennemo Statistics Norway, Norway

Abstract Most studies on the economic consequences of ageing rely on computable general equilibrium (CGE) models that account for feedback mechanisms through changes in relative prices, tax bases and so on. However, since individual labour supply behaviour is considered to be a key element in CGEanalyses of fiscal sustainability problems, the results of these analyses may depend crucially on how labour supply behaviour is modelled. The current practice of combining a simplified representation of the tax and transfer system with the labour supply behaviour of a few representative agents may produce a misleading description of incentives and revenue effects. The purpose of this paper is to demonstrate the importance of using an alternative strategy, by integrating a detailed microeconometric model of labour supply, which is sufficiently flexible to capture a large variety of labour supply responses, with a large-scale CGE model. The integrated micro–macro CGE model is employed to explore how endogenous household labour supply behaviour both affects and interacts with sustainability problems in Norway. The empirical results suggest that the required increase in the future tax burden is less dramatic when the analysis accounts for heterogeneity in labour supply behaviour. Moreover, by replacing the current progressive tax system with a flat tax system, it is found that the pressure on future public finances is significantly reduced.

International Symposia in Economic Theory and Econometrics, Vol. 15 r 2007 Published by Elsevier B.V. ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15010-3

260

Rolf Aaberge et al.

1. Introduction Most industrial countries will experience a substantial change in the age structure of their populations over the next 50 years due to increasing life expectancy and a slowdown in fertility. Ageing of the population is expected to have major consequences for the economy since it may affect labour supply, capital accumulation and growth, the composition of demand, foreign trade and capital accounts as well as public expenditures and incomes.1 The policy debate has focused largely on the effect on public finances. The combination of ageing and welfare state schemes has strong direct effects on government expenditures related to public old-age pensions, health services and care for the elderly — whereas the number of taxpayers may stagnate or even decrease. Current fiscal policy is not sustainable in most OECD countries; governments must sooner or later cut expenditures or raise taxes in order to keep the public budget balanced in the long run. This conclusion also applies to Norway, despite the fact that ageing in Norway is expected to be less rapid and dramatic than in other OECD countries (United Nations, 2001), and even though the Central government is the owner of petroleum wealth estimated to be twice the GDP in 2003.2 Projections made by the OECD (Antolin and Suyker, 2001) demonstrate that Norway actually faces one of the sharpest increases in government expenditures as a share of GDP in the OECD-area. According to projections in the Government’s National Budget 2004, government expenditures related to old-age and disability pension benefits will increase from 9 per cent of GDP in 2002 to about 20 per cent in 2050. The sustainable use of the petroleum wealth can finance less than 25 per cent of these expenditures in 2050. As the expenditures on welfare state-related arrangements are largely financed on a pay-as-you-go basis, maintenance of the existing arrangements may require a larger rise in

1

See Siebert (2002) for an overview. Statistics Norway (2003). This estimate of the petroleum wealth includes the present value of the net cash flow from the petroleum sector and the estimated capital in the Government Petroleum Fund by the end of 2003. The relative importance of the government petroleum revenues can alternatively be expressed in terms of the sustainable use of this wealth. According to the fiscal policy rule adopted from 2002, most of the current petroleum revenues collected by the government are saved as financial assets in the Central Government Petroleum Fund. On average only the expected real return (4 per cent at present) can be used each year. The fiscal policy rule implies that the current use of petroleum wealth increases gradually to the permanent income level as the petroleum reserves are depleted. Measured as a share of GDP the annual use of the petroleum wealth is then projected to reach a maximum of not more than 5 per cent around 2030 and less than 25 per cent of government expenditures related to old-age and disability pensions in 2050. 2

Population Ageing and Fiscal Sustainability

261

the tax burden than will be politically acceptable. Thus, an important challenge is to provide a good projection of the future tax burden and examine the extent to which the tax burden can be reduced by introducing appropriate income tax reforms. Evaluation of the long-run economic consequences of ageing and policy adjustments requires development and use of computable general equilibrium (CGE) models, capturing resource constraints, incentives and feedback mechanisms through changes in relative prices, tax bases, etc.3 However, to be tractable the labour supply behaviour in CGE models normally relies on a few representative agents and rough and simple approximations of the tax and transfer system. Simplifications are of course inevitable in economic modelling. However, the aggregate ‘‘representative agent’’ style of modelling labour supply implies obvious drawbacks in studies concerning fiscal sustainability, since responses to tax and social security reforms are key mechanisms in the public budget effects. Heterogeneity in behavioural responses may seriously violate the autonomy of aggregate labour supply functions. The lack of accuracy in the formal description of the tax and transfer systems also makes it hard to evaluate proposed tax reforms. Moreover, aggregate models obviously limit the scope for distributional evaluations. Empirical research on labour supply behaviour has been dominated by microeconometric work. This research has identified substantial heterogeneity in both supplied man-hours and the wage- and income elasticities. Thus, microeconometric models used to study labour supply responses of tax- and transfer reforms typically provide a very detailed description of heterogeneous individuals constituting representative samples of the population. Heckman (1974) is probably the first exercise to perform an explicit structural modelling of preferences and budget constraints in order to simulate the effects of a reform of family-related benefits. Heckman took full account of the non-linearity of the budget constraint in the estimation and simulation of the microeconometric model. The problem addressed was an evaluation of a child-related welfare policy that made significant complications in the budget set. Shortly after, Burtless and Hausmann (1978) proposed a method specifically addressed to piece-wise linear budget constraints. Both Heckman (1974) and Burtless and Hausmann (1978) work through the implications of the Kuhn–Tucker conditions. These contributions account for complex budget sets but at the cost of using very

3 Examples of recent projection studies of ageing and fiscal sustainability based on CGE models include Kotlikoff et al. (2001) for the US, Pedersen and Trier (2000) for Denmark, and Beetsma et al. (2003) for the Netherlands. Recent Norwegian studies are Ministry of Finance (2001) and Fredriksen, Heide, Holmøy and Solli (2005).

262

Rolf Aaberge et al.

simple and restrictive functional forms for utility functions or behavioural functions. Alternatively, very flexible and complex preference or behavioural functions are adopted (e.g. Blundell and Meghir, 1986) but at the cost of replacing the exact budget sets by simplified approximations — or in some cases even by linear subsets. Computational feasibility is the main reason for these choices. Most of these models are based on the ‘‘marginalist’’ version of the standard neo-classical model of leisure-income choice, where the derivation of a behavioural function (labour supply) involves comparisons between marginal utilities. The extension of the basic framework with a constant marginal wage to more complex cases with non-linear and kinked budget sets is theoretically straightforward but computationally burdensome. The focus on these problems has first and foremost generated econometric research on estimation and computation problems and this may explain why most behavioural models have been complemented with mainly illustrative policy simulation exercises based on some representative typology of households of welfare change such as average compensating or equivalent variation. During the last decade a novel approach has emerged, making use of models that allow for a representation of different degrees of availability of the opportunities in the choice set. For example, in a certain country, part-time jobs might be relatively difficult to find (maybe with differences from household to household). More generally, jobs of any kind might be more or less difficult to find (relative to non-market opportunities) in different countries, or regions and/or for different individuals. These new models belong to the multinomial logit family or some extension of it (see e.g. Ben-Akiva and Watanatada, 1981). They differ from the standard labour supply models by characterizing behaviour in terms of a comparison between utility levels, rather than between marginal utilities. Examples of applications to labour supply are Dickens and Lundberg (1993), Aaberge et al. (1995, 1999), van Soest (1995). These studies do not model choice by means of a more or less simple formula representing the reduced-form behavioral function (e.g. an equation expressing chosen hours of work as a function of wage, other income and socio-demographic characteristics). Instead, in a way, they mimic the actual process of choice by checking the alternative with the highest utility. Without entering into the technical details that can be found in the references above, among others, the benefits of this novel approach are notable: the resulting models are fully structural and specify a flexible direct utility function to describe the preferences of single individuals or married couples. Moreover, they allow for any type of complexity in the choice set, including fixed costs of labour market participation, quantity constraints and different availability of

Population Ageing and Fiscal Sustainability

263

different types of jobs and piecewise linearity of the budget constraints. Although the labour supply models that we use are strictly speaking supply models, and do not contain an explicit representation of the (labour) demand side, they nonetheless permit employment/unemployment effects to be taken into account. However, when the multinomial logit type of models are used as a basis for simulating the effects of policy changes, such as tax reforms, the partial equilibrium nature of these behavioural models is a drawback. They ignore possible feedback effects from changed behaviour due to changes in constraints, since prices, wages and quantity constraints are kept fixed. Since the CGE studies lack what the microeconometric models highlight, and vice versa, these strands of research are complementary. Thus, the integration of the two approaches is certainly important, given that the purpose of empirical studies is to provide as realistic results as possible by exploiting all relevant available information. This paper meets this challenge. It demonstrates how a CGE model and a detailed microeconometric labour supply model can be integrated and used to provide new insight on the fiscal sustainability problems in Norway. More specifically, we discuss the following question: What is the required tax increase in a future situation characterised by a much older population, provided that (1) the current welfare state arrangements are maintained; and (2) the current fiscal policy rule for using petroleum wealth is maintained? Given our long-run perspective, we find it relevant to consider adjustments in broad-based tax rates and estimate the required change in the payroll tax when the current income tax system is maintained. Since labour supply responses are the driving forces behind the growth in the tax bases, we will get useful information from an evaluation of how the fiscal sustainability assessment is affected by a hypothetical flat tax reform. In this alternative scenario, we assess the necessary pay-as-you-go adjustments in a flat tax rate levied on all types of personal income. The effects of the flat tax reform become clear when we compare them with the results of the reform based on pay-as-you-go adjustments of the payroll tax rate. The paper is organised as follows. Section 2 provides a brief description of the modelling framework, i.e. the macroeconomic CGE model, the microeconometric labour supply model and how these models have been integrated in a joint simulation framework. Section 3 discusses three different long-run scenarios: (i) a base line scenario in which individual labour supply is fixed; (ii) a scenario based on the same assumptions as in (i) except that individual labour supply responds according to the microeconometric model and (iii) a scenario where the existing income tax system is replaced by an endogenous flat tax rate. The two former scenarios use the payroll tax as the tax instrument (i.e. the payroll tax is changed in order to keep the time path

264

Rolf Aaberge et al.

of the public budget surplus consistent with the fiscal policy rule). The third scenario examines to what extent our results depend on the design of the income tax system. We focus on the equilibrium adjustments of labour supply, tax bases, public expenditures and the endogenous tax instruments. The interpretation emphasises what can be learned from taking account of a detailed model of labour supply rather than an aggregate representation, the relative importance of the various general equilibrium effects, as well as the contribution to the labour supply effect from changes in real wages and nonlabour income. Section 4 summarises the conclusions and briefly discusses further research projects.

2. The Integrated Micro–Macro Model Framework 2.1 The CGE Model The CGE model, MSG6, of the Norwegian economy has been developed with a focus both on long-run projections and analyses of tax policy and other structural policies.4 It includes a detailed account of government expenditures and revenues. Specifically, the model determines equilibrium adjustments in the determinants of individual labour supply (i.e. consumer real wage rates, real non-labour income and tax rates). The following exposition focuses on the aspects of the model that are considered to be the most relevant for the present study. The most important equilibrium mechanisms are explained in Section 2.3 and in Section 3, where the simulation results are reported and discussed. Heide, Holmøy, Lerskau and Solli (2004) describes formally the macroeconomic structure of the model, and provides a more thorough explanation of the mechanisms behind the results in this paper. The model assumes that the Norwegian economy is too small to affect world prices in NOK5 and interest rates. All agents have access to 4

Statistics Norway has been engaged in CGE modelling since the late 1960s, and this work has resulted in several generations of MSG models. MSG6 should be regarded as a family of models, which differ with respect to closure rules etc. All of these versions differ radically from older MSG-generations. Depending on the issue, different model versions have been applied in Holmøy and Vennemo (1995), Fæhn and Holmøy (2000, 2003), and Bye (2002). The Norwegian Ministry of Finance has regularly used different versions of the model to generate long run projections. 5 The exchange rate is fixed. This is an innocent assumption in a CGE model like MSG6, since the pass-through of exchange rate changes to all nominal prices is immediate and complete, leaving relative prices unchanged. The exchange rate can therefore be interpreted as a numeraire in the model.

Population Ageing and Fiscal Sustainability

265

international markets for financial capital. Supply equals demand in all markets in all periods, which implies no unemployment. The resource constraints on the economy as a whole include the time endowment of the labour force, the technology of firms and an intertemporal budget constraint, which ensures that the net foreign debt does not explode.6 In practice, macroeconomic growth is dominated by exogenous assumptions on growth in, respectively, total factor productivity (TFP) in private business industries, and labour productivity in government sectors. Aggregate labour supply is exogenous in the CGE model as the microeconometric labour supply model determines this variable. Goods and services, including those from labour and capital, are perfectly mobile across industries, and fixed capital is malleable. By specifying 60 commodities, 32 private business industries and 7 government sectors, MSG6 provides a detailed description of indirect taxes, taxes of private companies in different industries and various industry subsidies. Compared to a more aggregate model, this helps to make the calculations of government budget effects more accurate. Government employment, the government purchases of goods and services measured in fixed prices, and transfers before indexation, are all exogenous. In the simulations presented in this paper, the public budget constraint is satisfied by endogenous adjustments of alternatively: (a) the payroll tax rate, which works like a broad flat tax on labour income; and (b) a hypothetical flat tax rate on wages, capital income and transfers. Most imported products are considered as close but imperfect substitutes for the corresponding domestic products. Thus, the import shares of these tradables fall as the import price increases relative to the price of the corresponding domestic product.7 Output and input in an industry can change — because of changes at the firm level and as a result of entry or exit of firms, which are heterogeneous with respect to productivity. Managers of private firms have model consistent expectations, and maximise present after- tax value of the cash flow to owners. Producers allocate their output between the domestic and the foreign market. In most industries, it is costly to change the composition of deliveries to these two markets. Whereas world prices of exports are exogenous, domestic producers of manufactures and services engage in monopolistic competition in their

6 This intertemporal national budget constraint reflects that households and the government obey their intertemporal budget constraints. The corporate sector is assumed to distribute all after tax profits to the owners of the companies, which include households, the government and foreigners. 7 The price elasticities of import shares rely on Naug (1994).

Rolf Aaberge et al.

266

home markets.8 The production functions at the firm level between output and a composite variable input factor exhibit decreasing returns to scale. The scale elasticities range from 0.85 to 1.00.9 2.2 The Microeconometric Labour Supply Model The labour supply model used in this study can be considered as an extension of the standard multinomial logit model, and is designed to allow for a detailed description of the labour market.10 The modelling approach for labour supply used in this study differs from the traditional marginal criteria models of labour supply in several respects. First, it accounts for observed as well as unobserved heterogeneity in tastes and choice constraints, which means that it is able to take into account the presence of quantity constraints in the market. Second, it includes both single-person households and married or cohabiting couples making joint labour supply decisions. A proper model of the interaction between spouses in their labour supply decisions is important as most of the individuals are married or cohabiting. Third, by taking all details in the tax system into account the budget sets become complex and non-convex in certain intervals. For expository simplicity, in what follows, we consider only the behaviour of a single-person household. In the model, agents choose among jobs characterised by the wage rate w, hours of work h and other characteristics j. The problem solved by the agent looks like the following: max Uðy; h; jÞ

ðw; h; jÞ2B

8

(1)

This aspect of the technology is captured by assuming that output is a Constant Elasticity of Transformation (CET) function of deliveries to the export market and deliveries to the domestic market. Aukrust (1970) and Bowitz and Cappelen (1994) find empirical evidence supporting the view that Norwegian firms behave more like price takers on the export market than they do in the domestic markets. The mark-up ratios between the price of domestic deliveries and marginal costs are consistent with the econometric evidence in Klette (1999). 9 Klette (1999) estimates decreasing returns to scale at the firm level in Norwegian manufacturing industries. 10 Examples of previous applications of this approach are found in Aaberge et al. (1995), and Aaberge et al. (1999, 2000). The modelling approach used in these studies differs from the standard labour supply models by characterising behaviour in terms of a comparison between utility levels rather than between marginal variations of utility. These models are close to other recent contributions adopting a discrete choice approach such as Dickens and Lundberg (1993), van Soest (1995) and Euwals and van Soest (1999).

Population Ageing and Fiscal Sustainability

267

subject to the budget constraint y ¼ f(wh,m), where h denotes the hours of work, w the pre-tax wage rate, j the other job and/or household characteristics, m the pre-tax non-labour income (exogenous), y the disposable income, f (.,.) the tax rule that transforms pre-tax incomes (wh,m) into net income y and B the set of all opportunities available to the household (including non-market opportunities, i.e. a ‘‘job’’ with w ¼ 0 and h ¼ 0). Agents can differ not only in their preferences and in their wage (as in the traditional model) but also in the number of available jobs of different type. Note that for the same agent, wage rates (unlike in the traditional model) can differ from job to job. As analysts, we observe the chosen h and w, but we do not know exactly what opportunities are contained in B. Therefore we use a probability density function to represent B. Let pðh; wÞ denote the density of jobs of type ðh; wÞ: By specifying a probability density function on B we can, for example, allow for the fact that jobs with hours of work in a certain range are more or less likely to be found, possibly depending on agents’ characteristics — or for the fact that for different agents, the relative number of market opportunities may differ. We assume that the utility function can be factorised as Uðf ðwh; mÞ; h; jÞ ¼ V ðf ðwh; mÞ; hÞ  ðh; w; jÞ,

(2)

where V and e are the systematic and the stochastic component, respectively. Moreover, we assume that e is i.i.d. according to: Prð  uÞ ¼ expðu1 Þ

(3)

The term e is a random taste-shifter that accounts for the effect on utility of all the characteristics of the household-job match observed by the household but not by us. It can be shown that under the assumptions (1)–(3) we can write the probability density function of a choice (h,w) as11 jðh; wÞ ¼ RR

V ð f ð wh ; m Þ; h Þ p ðh ; w Þ V ð f ðzq ; m Þ; q Þ p ð q ; z Þ d q d z

(4)

qz

which is analogous to the continuous multinomial logit model. The intuition behind expression (4) is that the probability of a choice (h, w) can be expressed as the relative attractiveness — weighted by a measure of ‘‘availability’’ p (h, w) — of jobs of type (h, w). The tax rule, however complex, enters the expression as it is, and there is no need to simplify it in order to make it differentiable or manageable as in the traditional approach. While the traditional approach derives the functions representing household behaviour on the basis of a comparison of marginal variations 11

See Dagsvik (1994) and Aaberge et al. (1999), who provide two alternative methods for deriving (4).

268

Rolf Aaberge et al.

of utility, our approach is based on comparison of discrete levels of utility. To account for the fact that single individuals and married couples may face different choice sets and exhibit different preferences over income and leisure, separate models for single females and males and married/ cohabitating couples have been introduced. The parameters of the models have been estimated on the basis of data for individuals between 25 and 62 years old from the 1995 Survey of Level of Living,12 by the method of maximum likelihood after choosing convenient but still flexible parametric forms for V and p(h,w). We have restricted the ages of the individuals to be between 25 and 62 to minimise the inclusion in the sample of individuals who in principle are eligible for retirement, since analysis of retirement decisions is beyond the scope of this study. For a more detailed description of the data and definition of variables, see Aaberge (2006). The empirical specifications of V and p(h,w) and the estimation results are given in Aaberge et al. (2006). The behavioral implications of empirical labour supply models are normally displayed in terms of wage and income elasticities. The elasticities of the random utility models given by (2)–(4) are computed by means of stochastic simulations of the model since — as alluded to above — we (as analysts) do not observe all variables affecting preferences and opportunity sets. Draws are made from the distributions related to preferences U and opportunities B. Given the responses of each individual we then aggregate over the individuals to get the aggregate elasticities. Tables 1 and 2 report these elasticities.13 Since many individuals in this model of discrete choice will not react to small exogenous changes, the elasticities in Tables 1 and 2 have been computed as an average of the percentage changes in labour supply from a 10 per cent increase in the wage rates. The third and sixth rows of Table 1 give the unconditional elasticities of labour supply, which means that both the impact on participation and hours supplied is accounted for. Table 1 demonstrates that all own wage elasticities of married females and married males are clearly positive — whereas single females and males, on average, will only respond slightly positively to a wage increase. Second, we observe that all aggregate cross wage elasticities are negative due to the income effect. Thus, an increase in, say, the wage rate for a male implies that the labour supply of his spouse goes down. The negative cross wage elasticities means that an overall wage increase has a far

12

For further details on the microdata we refer to Aaberge and Colombino (2006). Detailed income-dependent elasticities are reported in Aaberge and Colombino (2006). These elasticities show a sharp decline with respect to income.

13

Population Ageing and Fiscal Sustainability

269

Table 1: Aggregate Labour Supply Elasticities with Respect to Wages, for Single and Married Individuals, Norway 1994 Family status

Type of elasticity

Female elasticities

Male elasticities

Own wage Cross Own wage Cross elasticities elasticities elasticities elasticities Single Elasticity of the females probability of and males participation

0.12

0.04

Elasticity of the conditional expectation of total supply of hours

0.09

0.02

Elasticity of the unconditional expectation of total supply of hours

0.02

0.02

0.21

0.19

0.23

0.11

Elasticity of the conditional expectation of total supply of hours

0.31

0.23

0.16

0.13

Elasticity of the unconditional expectation of total supply of hours

0.52

0.42

0.39

0.23

Elasticity of the Married probability of females participation and males

weaker impact on labour supply, both for males and females, than a partial wage increase for the two genders. Note also that hours supplied (given participation) for married/cohabitating females are far more responsive than is participation. This result is a reflection of the flexibility of the Norwegian labour market, where jobs with part-time working hours are common. Moreover, generous maternity leave arrangements and subsidised kindergartens make it attractive for women to combine the raising of children with participation in labour market activities. In contrast, for single females we find that participation increases when wages increase, whereas hours supplied (given participation) decrease. A similar, but weaker, effect is found for single males.

Aggregate Labour Supply Elasticities with Respect to Non-Labour Income, for Single and Married Individuals, Norway 1994

Family status

Single females and males

Female elasticities

Male elasticities

Non-labour income (capital income+cash transfers)

Capital income

Cash transfers

Non-labour income (capital income+cash transfers)

Capital income

Cash transfers

Elasticity of the probability of participation

0.79

0.20

0.71

0.19

0

0.08

Elasticity of the conditional expectation of total supply of hours

0.09

0.03

0.06

0.05

0.15

0.02

Elasticity of the unconditional expectation of total supply of hours

0.89

0.23

0.77

0.23

0.16

0.09

Elasticity of the probability of participation

0.20

0.11

0.09

0.23

0.12

0.10

Elasticity of the conditional expectation of total supply of hours

0.09

0.04

0.02

0.10

0.04

0.05

Elasticity of the unconditional expectation of total supply of hours

0.30

0.15

0.11

0.32

0.16

0.15

Rolf Aaberge et al.

Married females and males

Type of elasticity

270

Table 2:

Population Ageing and Fiscal Sustainability

271

The simulated income elasticities are reported in Table 2. Since the income elasticities are household-specific, the aggregate labour supply response to a shift that involves changes in non-labour income is the result of a complex calculation. Our simulations on capital income and cash transfers are unevenly affected by the general economic growth and the tax rate adjustments. Table 2 shows how the elasticity of labour supply with respect to changes in these income categories is estimated to depend on gender and household type. The major feature of the estimated labour supply elasticities can be summarised as follows:  

Labour supply of married women is far more elastic than for married men Individuals belonging to low-income households are much more responsive than individuals belonging to high-income households.

As demonstrated by the review of Røed and Strøm (2002) these findings are consistent with findings in many recent studies14. Aaberge and Colombino (2006) have demonstrated that the estimated random utility models used in this study reproduce the observed 1994 distributions of hours of work and disposable income almost precisely. To evaluate the prediction performance of the models estimated on Norwegian 1994 data, Aaberge and Colombino (2006) used these models to predict the distributions of disposable income for couples, single females and single males in 1979 and 2001, accounting for changes in taxes as well as in the composition of the population. The results of the out-of-sample predictions reported show that the models perform rather well.15 2.3 Integrating the CGE Model and the Partial Labour Supply Model The micro–macro modelling framework works as follows: for given values on the after-tax real wage rate and non-labour income, the microeconometric model simulates the households’ labour supply for a representative sample of households. The assessed percentage change in the supply of manhours is inserted into the CGE model, in which labour supply is exogenous. 14

Of particular interest when it comes to the responsiveness of the low-wage workers is a randomised experiment in Canada, the Canadian Self-Sufficiency Project. Blank et al. (1998) report an almost doubling of employment rates for persons offered inwork benefits compared to a control group. 15 See also Aaberge et al. (2006), who provide an evaluation of how the representation of choice sets in models of labour supply affect the prediction performance of the models.

Rolf Aaberge et al.

272 Figure 1: Model

The Interaction between the CGE Model and the Partial Labour Supply

Wage rate Cash transfers Capital income

Micro model

MSG 6

Labour supply

The CGE model then computes the equilibrium adjustments in the real wage rate, the revenue neutral tax rate and non-labour income. Next, the changes in these variables are used as a basis for changing the associated variables in the microeconometric model, which then produces new values for households’ labour supply. The process continues until equilibrium values in the labour market are reached. Figure 1 illustrates the exchange of information in the iteration process. The iteration approach faces the problem of exchanging the comparative statics results derived from the microeconometric labour supply model with the time paths derived from the dynamic CGE model. In practice, it is not feasible to carry out the iteration process every year within the time horizon (out to 2050). Our solution to this problem has been to interpret our results as stationary long-run effects. To this end, we compute what we call a stationary equilibrium associated with the projected situation characterising the year in focus, say 2050. This is achieved by letting all exogenous variables be constant at their 2050-levels. The CGE model then computes a transition path, where the stocks of real and financial assets converge to their stationary solutions, while resources used to produce the capital goods are gradually reallocated to consumption goods industries. It is in this computation of stationary 2050-equilibria that we use both the CGE-model and the partial labour supply model iteratively. We discuss the problems related to iteration between a static and a dynamic model further in Aaberge, Colombino, Holmøy, Strøm and Wennemo (2004). Figure 2 illustrates the equilibrium adjustment of the real wage cost per hour to a given increase in labour supply generated by the micro model. The LL- and the BB-locus describe reduced-form long-run equilibrium relationships between the producer wage rate and private consumption that are consistent with, respectively, labour market equilibrium and the budget constraint for the total economy implied by the external balance requirement. Since the loci capture reduced-form equilibrium relationships, they account for the equilibrium adjustments of all other endogenous variables as

Population Ageing and Fiscal Sustainability

273

Figure 2: Equilibrium Adjustments of the Consumption Level and Wage Rate Caused by an Exogenous Increase in Labour Supply in MSG6 L0 B0

Pre-tax wage rate

A B

L0

B0 L1

Consumption

well. The point where the two loci intersect represents the stationary general equilibrium. The LL-locus is upward sloping for the following main reasons: A partial increase in private consumption implies excess demand for labour. Increasing the wage rate restores labour market equilibrium because (1) firms substitute labour for other factors of production, and (2) changes in the industry structure reinforce the fall in aggregate labour demand. The latter effect can be explained as follows. The surge in the unit cost functions depends positively on the cost shares of labour. Higher costs reduce the international competitiveness of Norwegian producers. In particular, export supplies are sensitive to higher costs. The result is a negative scale effect on labour demand. In addition, households will face an increase in the relative price of the most labour-intensive products, and substitution effects contribute to a reallocation of resources from the most labour-intensive to less labour-intensive industries. The main reason why BB-locus is downward sloping is that a partial increase in private consumption raises imports. The wage rate must fall in order to boost exports and reduce import shares so that the external balance is restored. An exogenous increase in labour supply shifts the LL-locus from LL0 to LL1, since the wage rate must fall in order to raise labour demand (for a given level of private consumption). The BB-locus is unaffected by changes in labour supply. Thus, the new equilibrium (B) is characterised by higher private consumption and a lower wage cost per hour compared to the initial

274

Rolf Aaberge et al.

one (A). Rybczynski effects are at work in MSG6, but they are modified by the changes in large labour-intensive non-traded goods sectors and by decreasing returns to scale. As a result, the labour supply expansion is not completely absorbed through a reallocation of resources in favour of the most labour-intensive industries. Decreasing returns to scale necessitates a reduction of the wage costs, which induces firms to choose more labourintensive input combinations. Note that although the real producer wage rate must fall, the consumer real wage may increase if direct taxes on labour income or indirect taxes on consumption are used to endogenously restore the government budget constraint. The reason is that the surge in employment, other inputs, output and demand expands most direct and indirect tax bases. The net budget effect of the changes in the wage rate is less important since both tax bases and government consumption are negatively affected. Assessments of the empirical importance of the various budget effects require a relatively detailed CGE model. From the equilibrium adjustments in the wage rate and private consumption it is rather straightforward to explain the general equilibrium effects on other variables — see Heide et al. (2004). Focusing on labour supply decisions, the changes in households’ non-labour income deserve special interest. Capital income is increasing in labour supply since profits and output are positively related. Indexation is the only endogenous element in the cash transfers. From the discussion above, the nominal consumer wage rate may increase if the decrease in the taxes levied on labour income is sufficiently strong.

3. Long-Run Macroeconomic Scenarios 3.1 A Reference Scenario with Fixed Individual Labour Supply Our starting point is a reference projection of the Norwegian economy to 2050 in which behavioural effects on labour supply are neglected. This projection is based on the same assumptions as used in Fredriksen et al. (2005), which in turn relies heavily on the Norwegian Ministry of Finance (2001). The subsequent overview of exogenous assumptions is confined to the most important determinants of the endogenous payroll tax rate and of the individual labour supply responses simulated in the subsequent sections. Key Exogenous Assumptions Demography and resources. We rely on the projections in the middle alternative of the population projections in Statistics Norway (2002). Thereafter the labour force stays roughly constant throughout the scenario. Owing to demographic changes, measured man-hours have increased by 12.8 per cent

Population Ageing and Fiscal Sustainability

275

from 1995 to 2050. The old-age dependency ratio, defined as those 67 and older relative to those of working age 20–66, rises from 22 per cent in 2002 to 40 per cent in 2050. Over the same period, the ratio of people 80 years or older to those of working age will rise from 7.2 to 13.8 per cent, and the number of old-age pensioners will grow by 78.7 per cent. TFP grows by 1.3 per cent per year in private business sectors. Taking decreasing returns to scale into account, this is in line with long-run historical trends. In 2002, the export share of petroleum products was 42 per cent, and taxes and petroleum revenues amounted to approximately 27 per cent of total Central Government income. Our assumptions are the same as those used in the Governments’ Long Term Programme 2002–2005, see Norwegian Ministry of Finance (2001). In our scenarios, the net cash flow measured in current prices declines from 170 billion NOK in 2002 to 128 billion NOK in 2010 and to 110 billion NOK in 2050. As Norway, especially the government, accumulates financial assets, the international interest rate is important for the national and government income. The nominal interest rate is 5.5 per cent throughout the scenario, whereas all world prices, except petroleum prices, measured in NOK grow annually by 1.5 per cent. A 4-per cent real interest rate, measured in international purchasing power, is also the assumption underlying the present fiscal policy rule adopted in Norway in 2001. Government expenditures. The time path of the government budget surplus is determined by the fiscal policy rule explained in Section 1. It is realised by endogenous adjustment of the payroll tax rate. The exogenous projections of government consumption, in particular employment in government sectors, is based on various models developed in Statistics Norway. First, government consumption within the sectors health care and education has utilised a model that decomposes changes in the input of labour and intermediate inputs into (a) changes in the size of different age groups who differ in their use of public services; (b) changes in the service standards; and (c) changes in coverage ratios. We have made the rather cautious assumption that no changes take place in standards and coverage ratios beyond already approved reforms. A plausible interpretation is then that a scenario characterised by further growth in private consumption per capita involves privatisation of services traditionally provided by the government sector. Ageing alone implies an annual growth in government employment of 0.6 per cent from 2002 to 2020, 1.1 per cent in 2021–2030 and 0.8 per cent in 2031–2040. Thereafter, government employment grows by 0.3 per cent per year. The government expenditures related to public pension benefits have been projected by simulations on a detailed dynamic microsimulation model,16

16

See Fredriksen (1998) for a detailed description of this model (called MOSART) and of some applications.

276

Rolf Aaberge et al.

designed for this purpose. This model simulates entry into public pension schemes based on old age, disability, widow(er)hood and early retirement. The relevant transition rates have been estimated based on historical data. The total number of pensioners in 2050 is projected to be 57.8 per cent higher than in 2002. The model also includes a detailed description of how the public pension schemes determine the individual pension entitlements. Government pension expenditures will also grow for the following reasons: 







According to political intention, public pension benefits are indexed to wages rather than, say, inflation. The public pension system, implemented in 1967, is still maturing as the number of pensioners entitled to supplementary pensions is still increasing. Measured in terms of unindexed benefits, the average old-age public pension benefit is projected to increase by about 20 per cent from 1999 to 2050. An important reason is the growth in female labour income. The scheme for occupational pensions guarantees employees in the government sector two-thirds of previous earnings. The number of early retirees will grow during the next decades as it did in the 1990s. The early retirement benefits are partly financed by the government.17

Macroeconomic Growth Table 3 shows how the macroeconomic key variables grow from 1995 to 2050. The combined effect of exogenous growth in employment and TFP, as well as endogenous capital deepening, expands GDP by 1.7 per cent per year as an annual average over the period 1995–2050. On average, private consumption per capita can grow by an annual rate equal to about 2.5 per cent without breaking the intertemporal constraint on net foreign debt. This implies a doubling of private consumption per capita in about 28 years. The deviation between the growth rates of private consumption and GDP is partly due to the moderate growth in government consumption, which may be interpreted as a higher degree of privatisation of services traditionally provided by the government sector. Also, it reflects that a part of the present value of the private consumption is financed by the initial petroleum wealth. 17

A main reason for this has been the pension arrangement referred to as AFP (an abbreviation for the Norwegian term Avtalefestet pensjonsordning). Currently, AFP covers the entire public sector, employing about one-third of all employees, and about 43 per cent of private-sector employees. This arrangement provides strong incentives to retire at the age of 62 years.

Population Ageing and Fiscal Sustainability

277

Table 3: Long-Run Macroeconomic Development with Fixed Individual Labour Supply and Endogenous Payroll Tax Rate Simulated 1995levels, Billions NOK Private consumption Government consumption Gross fixed capital formation Exports Imports GDP Real average consumer after-tax wage rate Real cash transfers received by households, net of old-age pensions Real capital income received by households Employment, million man-hours Payroll tax rate, per cent

The ratio between 2050- and 1995levels

418.6 195.4 209.2 383.3 358.4 848.1 145.2 92.9

5.3 1.6 1.4 1.5 2.4 3.0 3.1 3.1

294.2

2.8

2,975.3 13.1

1.1 2.0

Note: Variables are measured in fixed prices, except where indicated.

A conclusion that turns out to be robust with respect to individual labour supply behaviour is that the expected ageing in Norway will not represent any strong drag on aggregate economic growth over the period 1995–2050. As mentioned in Section 2, the sustainable growth in consumption possibilities is determined almost solely by productivity growth, which is driven by the exogenous TFP-growth set in line with historical trends. The unique role of productivity as the fundamental growth determinant is not an artefact of our particular CGE model, but reflects common knowledge about economic growth. In particular, compared to variations in the TFP-growth rate, plausible variations in the age structure of the population are of minor importance.18 Determinants of labour supply The average annual growth in the nominal pre- and after-tax wage rate is found to be 4.2 per cent when individual labour supply responses are

18

The importance of productivity growth for the long-run living standards is pointedly discussed in Krugman (1992, p. 9), who declares: ‘‘Productivity isn’t everything, but in the long run it is almost everything. A country’s ability to improve its standard of living over time depends almost entirely on its ability to raise its output per worker.’’

278

Rolf Aaberge et al.

ignored. It reflects the growth in the producer value of the marginal product of labour in the traded goods sector,19 which is primarily a result of the TFPgrowth, reduced labour intensity in the input composition and a growth in world prices of 1.5 per cent per year. The resulting annual growth in the pretax consumer real wage rate averages 2.2 per cent. All tax rates but the payroll tax rate are fixed along the base line scenario. This implies that the after-tax consumer real wage rate in 2050 is simulated to become 3.1 times the 1995-level. Non-labour income includes cash transfers from the government and capital income. Deflated by the consumer price index, capital income and total transfers (net of old-age pension benefits which are not received by the labour force) would in 2050 be, respectively, 2.8 and 3.1 times the 1995-level. Future Tax Burden Ageing in Norway causes a substantial increase in the future tax burden. The payroll tax rate has to be increased from 13 per cent today to nearly 26 per cent in 2050 to meet the public budget constraint determined by the fiscal policy rule. This result is first and foremost due to the fact that public old-age pension benefits are projected to increase from the current 7–20 per cent of GDP in 2050. In 2050, old-age pension benefits grow to become five times as high as in 1995. The increase in the payroll tax is almost completely shifted from firms to labour. This incidence works through two channels. First, as pointed out in Section 2.3, the wage cost per hour is basically determined by the producer value of the marginal labour productivity in a sufficiently large traded goods sector. Thus, in the new general equilibrium the increase in the payroll tax rate results in a reduction of the wage rate of roughly the same order of magnitude. The second channel of incidence is the pass-through of wage costs to the prices of non-traded goods. The simulated figures illustrate a general insight: productivity growth in the private sector will not ease the pressure on public finances, provided that (i) public pension benefits and other public transfers are indexed by the wage growth; and employment is unaffected by productivity growth. In Norway, the opposite is true. Those believing that productivity growth will ensure fiscal sustainability without increases in taxes or cuts in government expenditures should recognise the following mechanisms: The general equilibrium

19

Note that it follows from the general equilibrium conditions that the traded goods sector is large enough to ensure that foreign trade is balanced in the long run. The size of the traded goods sector affects the value of the marginal productivity of the input factors because the model assumes decreasing returns to scale in most industries.

Population Ageing and Fiscal Sustainability

279

effect of a positive shift in TFP will be an increase in the wage rate in all sectors of approximately the same order as the endogenous increase in labour productivity. The resulting surge in household incomes and consumption will expand most of the tax bases by about the same proportion. An important exception in Norway is the petroleum revenue collected by the government. However, cet. par., government expenditure also increases approximately proportionally to the wage rate because government consumption is dominated by wages, and because transfers are indexed to wages. Since the Norwegian government runs a fiscal deficit when petroleum net revenues are excluded, this non-oil fiscal deficit will increase as a result of TFP growth and reinforce the pressure on public finances. 3.2 Projections Accounting for Endogenous Labour Supply Table 4 displays the predicted change in the key variables when we take into account endogenous individual labour through our integrated model framework. In the new stationary equilibrium in 2050, employment is 4.6 per cent Table 4: Comparison of Long-Run Macroeconomic Development with Fixed and Endogenous Individual Labour Supply (L) and Endogenous Payroll Tax Rate Ratio between 2050- and 1995-levels

Private consumption Government consumption Gross fixed capital formation Exports Imports GDP Real average consumer after-tax wage rate Real cash transfers per capita received by households, net of old-age pensions Real capital income per capita received by households Employment, million man-hours Payroll tax rate, per cent

Exogenous L

Endogenous L

Effect of endogenous L in 2050. Percentage deviations from base line

5.3 1.6

5.6 1.6

4.2 0.8

1.4

1.5

3.6

1.5 2.4 3.0 3.1

1.5 2.4 3.2 3.1

3.8 2.3 4.0 0.8

3.1

3.1

1.9

2.8

2.9

0.9

1.1

1.2

4.6

2.0

1.6

19.2

Note: Variables are measured in fixed prices, except where indicated.

Rolf Aaberge et al.

280

higher compared to the 2050-equilibrium base case, where labour supply was supposed to change exclusively due to demographic changes. The behavioural effects come on top of the 12.8 per cent employment growth from 1995 to 2050 caused by demographic changes. The isolated effects of endogenous labour supply on the estimates of GDP and private consumption in 2050 are somewhat smaller than the percentage increase in employment — 4.0 and 4.2 per cent, respectively. This is due to decreasing returns to scale in the private business industries and to a small reduction in the overall capital intensity in production. The reduction of the average capital–labour ratio is due to a decrease in the hourly wage cost relative to other factor prices. Section 2.3 explained why increased labour supply must lead to lower real wage costs per hour (in order to keep the economy within the intertemporal foreign debt constraint when the private sector production functions exhibit decreasing returns scale). However, the endogenous reduction of the payroll tax is strong enough to give room for an increase in the wage rate facing workers. When individual labour supply is endogenous rather than fixed, the endogenous payroll tax in 2050 is reduced from 26 to 21 per cent due to expansion of the tax bases. However, an increase in the broadly defined payroll tax rate from the present 13 to 21 per cent still suggests that Norway has a severe fiscal sustainability problem. Given that the consumer real wage rate would be 209.3 per cent higher in 2050 than in 1995, the increase in labour supply of 4.6 per cent may at first glance seem surprisingly small. A weighted average of the individual wage elasticities of labour supply, reported in Table 5, equals 0.12. Note, however, that this is a local measure of the aggregate wage elasticity. A first-order approximation of the wage effect on labour supply from 1995 to 2050 would be to multiply this elasticity by the 209.3 per cent growth in the consumer real wage rate. Such an approximation suggests that the wage growth contributes to a growth in labour supply by 25.1 per cent. However, the growth in nonlabour income must also be accounted for. A rough first-order approximation of the non-labour income effect on labour supply from 1995 to 2050 would multiply the aggregate labour supply elasticity with respect to non-labour income, which equals –0.17, by the 190 per cent growth in total non-labour income from 1995 to 2050. Such an approximation suggests that growth in non-labour income contributes to a reduction of labour supply by 32.3 per cent. Table 5:

Labour Supply Elasticities in 1995 and 2050 (1995 Tax System)

Labour supply elasticity w.r.t.

1995

2050

Wage Non-labour income Wage and non-labour income

0.12 0.17 0.07

0.14 0.12 0.04

Population Ageing and Fiscal Sustainability

281

The net effect of the two approximations is a reduction in aggregate labour supply of roughly 7.1 per cent. By contrast, the simulated effect is a 4.6 per cent increase. Thus, a first-order approximation based on the local properties of the microeconometric labour supply model produces a very misleading impression of the effects of the large changes projected over the period 1995–2050. In order to explain the deviation between the simulated effects and the firstorder approximation based on the local elasticities, it is necessary to take the global properties of the microeconometric model into account. Two global properties emerge as most important in this respect. First, the wage elasticity rises by 17 per cent when it is computed at the income levels projected in 2050 compared to the one computed in 1995. Second, the labour supply elasticity with respect to non-labour income is reduced (in absolute value) by income growth; in the 2050 situation it is about two-thirds of the corresponding 1995level. Taking into account the gradual changes in the relevant elasticities, the 4.6 per cent increase in labour supply is well within what might be expected from a back-of-the-envelope check of the simulated results. The aggregate labour supply elasticities may vary for two reasons: (i) The micro-elasticities are not fixed structural parameters but sensitive to changes in labour and non-labour income; and (ii) changes in the composition of aggregate labour supply between individuals with different elasticities. The latter effect clearly helps to explain the increase in the average wage elasticity. The most wage-elastic individuals increase their shares in aggregate labour supply. This positive correlation between changes in weights and wage elasticities raises the average wage elasticity. Although the outcome of the microeconometric model simulations is sensitive to the change in the income level, the pattern of negative correlation between labour supply elasticities and income is maintained (i.e. low-income families respond more strongly to changes in economic incentives than high-income families). 3.3 The Effect of Replacing the 1994 Tax System by Flat Taxation Is it possible to ease the future tax burden through tax reforms? This important question can only be given a tentative answer within the scope of this paper. We restrict our study to simulating the elimination of the progressivity of the tax structure. More precisely, we simulate the equilibrium in 2050, when all personal income tax rates in the 1995 system are replaced by a flat tax rate levied on all labour market earnings, as well as on cash transfers and capital income. Under this hypothetical system, the government budget constraint is met through endogenous adjustments of the flat tax rate, instead of the payroll tax rate. In 1995, the revenue collected from direct income taxes amounted to 24 per cent of labour income, capital income and cash transfers. The difference between the equilibrium in 2050 and the initial equilibrium in 1995 is now a result of (a) growth effects between 1995 and 2050

282

Rolf Aaberge et al.

Table 6: Long-Run Macroeconomic Developments with Endogenous Individual Labour Supply (L), under the 1995 Tax System and a Flat Tax System 1995 tax system (endogenous Flat tax system payroll tax rate) (endogenous flat tax rate) Exogenous L L Endogenous Private consumption Government consumption Gross fixed capital formation Exports Imports GDP Real consumer pre-tax wage rate Real consumer after-tax wage rate Real cash transfers per capita received by households, net of old-age pensions Real capital income per capita received by households Employment, million man-hours Payroll tax rate Flat tax rate

Endogenous L

5.3 1.6 1.4 1.5 2.4 3.0 3.1 3.1 3.1

5.6 1.6 1.5 1.5 2.4 3.2 3.1 3.1 3.1

6.1 1.6 1.6 1.7 2.6 3.5 3.1 4.2 3.2

2.8

2.9

3.0

1.1 2.0 —

1.2 1.6 —

1.3 1.0 1.0

Note: Ratios between 2050-Levels and Base Case 1995-Levels. Variables are measured in fixed prices, except where indicated.

accounted for in Sections 3.1 and 3.2, and (b) the flat tax reform. Here, we confine the discussion to examining the sensitivity of our projections in 2050 for those variables most relevant for the future tax burden. Tables 6 and 7 show how the effects of allowing endogenous individual labour behaviour are affected by the tax system. Table 6 compares three different equilibria in 2050 with the corresponding 1995-levels. The first and second columns are identical to the second columns in, respectively, Table 3 and Table 4. In the third column, the 2050 projections in the scenario with endogenous individual labour supply and endogenous adjustments in the flat tax rate are compared with the corresponding observed 1995-levels. In Table 7, the first two columns report the percentage deviations, measured in 2050, between the two scenarios with endogenous individual labour supply and the base line scenario. The third column shows the equilibrium effects of implementing the flat direct income tax system in 1995, in terms of percentage deviations between the new equilibrium and the base line 1995-levels. Thus, the difference between the results of the second and the third columns captures how the pure growth effects work under a flat tax system.

Population Ageing and Fiscal Sustainability

283

Table 7: Macroeconomic Changes Caused by Endogenous Individual Labour Supply Responses in 2050

Private consumption Government consumption Gross fixed capital formation Exports Imports GDP Real consumer pretax wage rate Real consumer aftertax wage rate Real cash transfers per capita received by households, net of old-age pensions Real capital income per capita received by households Employment, million man-hours Payroll tax rate Flat tax rate

2050 (with 1995 tax system) (1)

2050 (with flat tax system) (2)

1995 (pure effect of flat tax reform) (3)

4.2 0.8

14.3 2.9

12.4 1.7

3.6

10.7

6.3

3.8 2.3 4.0 0.8

14.3 8.5 13.7 10.3

8.7 9.4 7.8 8.6

0.8

1.7

1.1

1.3

7.0

5.6

0.4

15.7

17.4

4.5

16.7

10.4

19.2 —

— 28.4

— 23.8

Note: Deviations in percentages from Base Line. Variables are measured in Billions NOK in fixed 1995-prices, except where indicated.

The overall impression from comparing the results in Tables 4, 6 and 7 is that a flat tax reform would boost labour supply and cause a substantial increase in government net revenue. Endogenous labour supply behaviour under a flat tax regime generates a 16.7 per cent increase in employment in 2050, compared to the projection based on fixed individual labour supply. The flat tax rate would have to increase from 24 per cent in 1995 to 32 per cent in 2050 in the case where individual labour supply was assumed to be fixed and not responsive to incentives. Relaxing this assumption and allowing for endogenous labour supply behaviour implies that the flat tax rate can be set equal to 22.9 per cent in 2050. A flat tax rate of 18.3 per cent provided sufficient tax revenue in 1995. Thus, the combined effects of population ageing and economic growth require an increase in the flat tax rate of approximately 5 percentage points from 1995 to 2050.

284

Rolf Aaberge et al.

Table 7 demonstrates that about two-thirds of the employment expansion can be attributed to the flat tax reform. The second column in Table 7 displays the percentage deviation between the 2050 case, given a flat tax system and endogenous labour supply, and the 2050 case given the 1995 tax system and fixed labour supply (no individual labour supply responses to growth in wages and non-labour income). This deviation is decomposed into (A) the pure effect of a flat tax reform, measured by the effect in 1995, and (B) the effect of growth from 1995 to 2050, given the flat tax system. The last column measures the effect of A in this decomposition. Effect A is (approximately) the difference between the total effect (second column) and effect B (third column). Thus, the pure employment effect of the growth from 1995 to 2050 is 6.3 per cent under the flat tax regime. Recall that the corresponding employment effect was 4.6 per cent when the 1995 tax system was maintained. Thus, the pure growth effect on labour supply is slightly stronger under the flat tax system than when the payroll tax rate is increased under the 1995 tax system. The main reason for this difference is that workers exclusively pay the increase in the payroll tax, whereas the increase in the flat tax rate is shared between workers, capital owners and transfer recipients. Thus, the negative impact on the consumer real wage rate is reduced, and heavier taxation of non-labour income has a positive effect on labour supply.

4. Conclusions We have developed an integrated micro–macro CGE model and employed it to explore how endogenous individual labour supply behaviour both affects and interacts with fiscal sustainability problems in Norway caused by ageing, combined with the maintenance of a generous welfare state policy. In particular, the existing pay-as-you-go financed pension system will bring about a sharper increase in the ratio of government expenditures to GDP in Norway than in most other countries up to 2050. The results of the simulation exercise discussed in this paper present a rather differentiated picture, depending on the methodology employed and the hypothesis made upon the tax system. The standard procedure underlying long-run CGE-studies of ageing is to let a few representative agents determine the aggregate labour supply responses. Specifically, previous projections of the Norwegian economy have been based on exogenous assumptions of labour supply, and no responses to changes in the wage rate, non-labour income and taxes. In our first projection, we repeat this approach by running the CGE model simulations without accounting for the behavioural responses coming from the microeconometric model of household labour supply, and keeping the tax system unchanged. The resulting perspectives are indeed worrying: fiscal sustainability would require doubling the payroll tax rate (from 13.0 to 26.0 per cent). However, if we take

Population Ageing and Fiscal Sustainability Table 8:

285

Estimated Equilibrium Tax Rates Tax system Current system (instrument: payroll tax rate)

Alternative flat tax System (instrument: flat tax rate)

2050

1995

2050

Exogenous

26.0

24.0

32.0

Endogenous

21.0

18.3

22.9

Labour supply

Note: The average payroll tax rate was 13 per cent in 1995.

endogenous labour supply into account, the picture starts to look better, the required payroll tax rate in 2050 being 21.0 per cent instead of 26.0 per cent. Since labour supply seems so important, it makes sense to hypothesise a reformed tax system that gives better incentives to work. The simplest idea consists of introducing a Flat Tax. In this case we use the flat tax rate directly — instead of the payroll tax rate — as the instrument to balance the public budget. For the sake of comparison, let us start by keeping labour supply exogenous. Then, in 1995 we would need a 24.0 per cent flat tax rate in order to generate the same total tax revenue as obtained with the actual tax system; in 2050, the rate would be 32.0 per cent. Now, allow endogenous labour supply: the required flat tax rates are then 18.3 per cent in 1995 and 22.9 per cent in 2050. The results are summarised in Table 8. As a tentative conclusion, it appears that the fiscal sustainability problems expected in the future decades can be reduced to manageable dimensions provided the tax system is reformed in order to improve the incentives for labour supply. Two qualifications are in order at this point. First, there are of course many other ways to stimulate labour supply that might be alternative or complementary to reforming the tax system. The results in Fredriksen et al. in Chapter 3 of this volume suggest that pension reform is perhaps the most important candidate in this respect. Other policies might also work, although they might be more expensive (such as improving public substitutes for parental childcare, etc.). Second, even confining ourselves to tax reforms, the flat tax rule itself — besides its advantages in term of simplicity — is not necessarily the best one in order to get the desired effects: for example, it is likely to increase income inequality. Alternative tax rules might produce similar efficiency effects without increasing income inequality.20 More

20

For example, a flat tax coupled with a Negative Income tax or a workfare mechanism may produce a favourable result (Aaberge et al., 2001).

286

Rolf Aaberge et al.

generally, the pattern of labour supply elasticities (Table 3) reveals that what matters in order to bring more people into the labour market is increasing the net wage for individuals living in low- and average-income households. This would suggest — rather than a pure Flat Tax — a reduction of progressivity focussed on low and average income brackets. From the methodological point of view, the exercise shows very clearly the importance of both general equilibrium effects and the effects captured by a detailed microeconometric model of labour supply behaviour in heterogeneous households, as well as of their interactions. Moreover, it would be hard to decide which one is less harmful to ignore as a simplifying strategy — if we had to. Their relative importance seems to vary depending on the point in time and the policy environment considered by the simulation exercise. A criticism often raised against large and complex empirical models is that they are black boxes, leaving outsiders with limited opportunities to check the logic and the driving forces behind the results. Integrating two large models make us vulnerable to such a criticism. However, we want to emphasise that we give priority to realistic assessments rather than to numerical illustration of particular effects. It is then inefficient to neglect available information about mechanisms of potential empirical significance because they complicate the analysis. In this respect, it is interesting to note that recent research in the labour supply, human capital and social policy evaluation literature has augmented CGE models with a more detailed description of the heterogeneity in behaviour.21 A ‘‘cost–benefit’’ evaluation of those effects that should be given priority in empirical assessments should be based on experiences with rich models, rather than on ex ante conjectures. In particular, such evaluations should — in economics as in other quantitative disciplines — take advantage of the dramatic improvement in computational methods, rather than cling to the same constraints facing to Ricardo and Marshal.22

21

Heckman et al. (1998) includes a parametric distribution of heterogeneity in abilities in policy analyses of human capital accumulation; CGE studies in the international trade and public economics literature have been complemented with microsimulation modules to allow detailed distributional analyses, see e.g. Bourguignon et al. (2001); OLG models addressing ageing issues and effects of tax- and social security reforms have recently been expanded by including a larger number of representative individuals in order to capture both more details of the tax- and social security systems and distributional effects, see Kotlikoff et al. (1998, 2001), Fehr (1999), Broer (2001), Fehr and Steigum (2002) and Fehr et al. (2003). 22 See Judd (2001) for an expert discussion of the usefulness of computational methods in economics.

Population Ageing and Fiscal Sustainability

287

Acknowledgment We would like to thank Torstein Bye and Vishnu Kapur for helpful comments, and the Norwegian Council of Research for financial support.

References Aaberge, R. and Colombino, U. (2006). Estimation and Prediction Performance of a Structural Random Utility Model for Labor Supply, Statistics Norway, Mimeo. Aaberge, R., Colombino, U., Holmøy, E., Strøm, B. and Wennemo, T. (2004). Population aging and fiscal sustainability: An integrated micro-macro analysis of required tax changes, Discussion Paper No. 366, Statistics Norway. Aaberge, R., Colombino, U. and Strøm, S. (1999). Labour Supply in Italy: An Empirical Analysis of Joint Household Decisions with Taxes and Quantity Constraints. Journal of Applied Econometrics, 14, 403–422. Aaberge, R., Colombino, U. and Strøm, S. (2000). Labour Supply Responses and Welfare Effects from Replacing Current Tax Rules by a Flat Tax: Empirical Evidence from Italy, Norway and Sweden. Journal of Population Economics, 13, 595–621. Aaberge, R., Colombino, U. and Strøm, S. (2001). Do More Equal Slices Shrink the Cake? An Empirical Evaluation of Tax-Transfer Reform Proposals in Italy. CHILD Working Paper 19/2001 (http://www.child-centre.it/). Aaberge, R., Colombino, U. and Wennemo, T. (2006). Evaluating Alternative Representations of the Choice Sets in Models of Labour Supply. IZA Discussion Paper No. 1986. Aaberge, R., Dagsvik, J.K. and Strøm, S. (1995). Labour Supply Responses and Welfare Effects of Tax Reforms. Scandinavian Journal of Economics, 97(4), 635–659. Antolin, P. and Suyker, W. (2001). How Should Norway Respond to Ageing? Working Paper 296, OECD Economics Department Series, OECD. Aukrust, O. (1970). PRIM 1. A Model of the Price and Income Distribution Mechanisms of an Open Economy. Review of Income and Wealth, 16, 51–78. Beetsma, R., Bettendorf, L. and Broer, P. (2003). The Budgeting and Economic Consequences of Ageing in the Netherlands. Economic Modelling, 20, 987–1013. Ben-Akiva, M. and Watanatada, T. (1981). Application of a Continuous Spacial Choice Logit Model, in Manski, C.F. and McFadden, D. (eds), Structural Analysis of Discrete Data with Econometric Applications, MIT Press, Cambridge, MA. Blundell, R. and Meghir, C. (1986). Selection Criteria for a Microeconometric Model of Labour Supply. Journal of Applied Econometrics, 1, 55–80. Bourguignon, F., Robilliard, A.S. and Robinson, F. (2001). Crisis and Income Distribution: A Micro–Macro Model for IndonesiaMimeo. Bowitz, E. and Cappelen, A˚ (1994). Price Formation and Factor Demand in Norwegian Industries. Social and Economic Studies 85, Statistics Norway, Oslo (in Norwegian). Broer, D.P. (2001). Growth and Welfare Distribution in an Ageing Society: An Applied General Equilibrium Analysis for the Netherlands. De Economist, 149, 81–114.

288

Rolf Aaberge et al.

Burtless, G. and Hausmann, J. (1978). The Effects of Taxation on Labour Supply. Journal of Political Economy, 86, 1103–1130. Bye, B. (2002). Taxation, Unemployment and growth: Dynamic Welfare Effects of ‘‘Green’’ Policies. Journal of Environmental Economics and Management, 43, 1–19. Dagsvik, J.K. (1994). Discrete and Continuous Choice, Max-Stable Processes and Independence from Irrelevant Attributes. Econometrica, 62, 1179–1205. Dickens, W. and Lundberg, S. (1993). Hours Restrictions and Labour Supply. International Econonomic Review, 34, 169–191. Euwals, R. and van Soest, A. (1999). Desired and Actual Labour Supply of Unmarried Men and Women in the Netherlands. Labour Economics, 6, 95–118. Fæhn, T. and Holmøy, E. (2000). Welfare Effects of Trade Liberalization in Distorted Economies: A Dynamic General Equilibrium Assessment for Norway, in Harrison, G.W., Hougaard Jensen, S.E. and Rutherford, T. (eds), Using Dynamic General Equilibrium Models for Policy Analysis, North-Holland, Amsterdam. Fæhn, T. and Holmøy, E. (2003). Trade Liberalisation and Effects on Pollutive Emissions to Air and Deposits of Solid Waste. A General Equilibrium Assessment for Norway. Economic Modelling, 20, 703–727. Fehr, H. (1999). Welfare Effects of Dynamic Tax Reforms. Mohr Siebeck, Tubingen. Fehr, H. and Steigum, E. (2002). Pension Funding Reforms in a Small Open Welfare State, in Fossati, A. and Wiegard, W. (eds), Policy Evaluation with Computable General Equilibrium Models, Routledge Applied Economics, Florence, Kentucky, pp. 232–248. Fehr, H.W., Sterkeby, I. and Thøgersen, Ø. (2003). Social Security Reforms and Early Retirement. Journal of Population Economics, 16, 345–361. Fredriksen, D. (1998). Projections of Population, Education, Labour Supply and Public Pension Benefits. Analyses with the Dynamic Simulation Model MOSART. Social and Economic Studies 101, Statistics Norway, Oslo. Fredriksen, D., Heide, K.M., Holmøy, E. and Solli, I.F. (2005). Macroeconomic Effects of Proposed Pension Reforms in Norway. Discussion Paper 417, Statistics Norway. Heckman, J. (1974). Effects of Child-Care Programs on Women’s Work Effort. Journal of Political Economy, 82(2-II), 136–163. Heckman, J., Lochner, L. and Taber, C. (1998). General Equilibrium Treatment Effects: A Study of Tuition Policy. American Economic Review, 88(2), 381–386. Heide, K. M., Holmøy, E., Lerskau, L. and Solli, I. F. (2004). Macroeconomic properties of the Norwegian applied general equilibrium model MSG6, Reports 2004/18, Statistics Norway. Holmøy, E. and Vennemo, H. (1995). A General Equilibrium Assessment of a Suggested Reform in Capital Income Taxation. Journal of Policy Modelling, 17(6), 531–556. Judd, K. (2001). Computation and Economic Theory: Introduction. Economic Theory, 18(1), 1–6. Klette, T.J. (1999). Market Power, Scale Economies and Productivity: Estimates from a Panel of Establishment Data. Journal of Industrial Economics, 47, 451–476. Kotlikoff, L., Smetters, K. and Walliser, J. (1998). The Economic Impact of Privatising Social Security, in Siebert, H. (ed), Redesigning Social Security, Mohr Siebeck, Tubingen.

Population Ageing and Fiscal Sustainability

289

Kotlikoff, L., Smetters, K. and Walliser, J. (2001). Finding a Way Out of America’s Demographic Dilemma. NBER Working Paper 8258. Krugman, P. (1992). The Age of Diminished Expectations. MIT Press, Cambridge, MA. Naug, B. (1994). En økonometrisk analyse av utviklingen i importandelene for industrivarer 1968–1990, Social and Economic Studies, Statistics Norway, Oslo, p. 84. Norwegian Ministry of Finance (2001). Long Term Programme 2002– 2005. St. melding no.30 (2000/2001), Norwegian Ministry of Finance, Oslo (in Norwegian). Pedersen, L.H. and Trier, P. (2000). Har vi ra˚d til velferdsstaten. DREAM Working Paper Series 2000:4. Siebert, H. (ed). (2002). Economic Policy for Ageing Societies. Springer, Berlin. Statistics Norway (2002). Befolkningsframskrivinger. Nasjonale og regionale tall, 2002–2050. www.ssb.no/emner/02/03/folkfram/ 05.12.2002. United Nations (2001). World Population Prospects: The 2000 Revision. (Database). van Soest, A. (1995). Structural Models of Family Labour Supply: A Discrete Choice Approach. Journal of Human Resources, 30, 63–88.

This page intentionally left blank

Chapter 11

Canadian Retirement Behaviour: A Microeconomic Examination$ Greg Maloney, Mashood Mirza and Franc- ois Paris Canada Customs and Revenue Agency, Canada

Abstract Over the next few decades, Canada will experience significant demographic and economic changes as the baby-boom generation retires. Economic literature has examined the stress that this phenomenon will place on the economy — specifically, the publicly administered Canadian income security programmes, composed of CPP/QPP, OAS, GIS and the ASA. Recent literature, however, has begun to examine the longer-term consequences of these programmes, finding evidence that the programmes may provide labour supply disincentives. In this paper, we investigate the robustness of sample and surveybased findings that the Canadian income security programmes provide labour supply disincentives. Our econometric analyses draw upon population microdata available from the CCRA T1 Mini-Universe database. This database allows a more comprehensive study on retirement behaviour — by providing complete information on income security programmes, including entitlements and payouts, as well as complete information relating to private pension plans and other sources of non-labour income. Our findings show that income security wealth has a minimal impact at best on the retirement decision of Canadians. Instead, our findings suggest the decision to retire is influenced by more prevalent factors such as non-labour income and employer pension plans.

$ This research paper represents strictly the views and opinions of the authors. It should not be taken as a reflection of Canada Customs and Revenue Agency or the Federal Government. CCRA has since been renamed the Canada Revenue Agency.

International Symposia in Economic Theory and Econometrics, Vol. 15 Copyright r 2007 Elsevier B.V. All rights reserved ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15011-5

292

Greg Maloney et al.

1. Introduction Income security (IS) is an important consideration for individuals. People need sufficient levels of income to sustain themselves through periods of unemployment, sickness or disability and, especially, retirement. In Canada, IS programmes provide assistance to eligible individuals in each of these circumstances, with retirees being the main recipients. A recent study on population ageing in Canada projected that cohorts aged 65 and more will grow from 11 per cent of the population in 1998 to over 20 per cent by 2020 (McGregor, 2001). The results of this population ageing will have significant impacts on IS programmes in Canada, as the number of cohorts expected to receive retirement pensions will rise some 40 per cent over a similar time horizon (Human Resources Development Canada, 2001). Canadian Census figures also show that the working-age population between 45 and 64 and cohorts aged 80 and more grew by 36 per cent and 42 per cent, respectively, over the 1991–2001 period. Much has been written on the possible impacts that an ageing population will have on the Canadian economy. Among the largest impacts brought on by the retirement of these individuals from the labour force will be substantially lower revenue for governments — because of large decreases in incomes associated with higher levels of retirement, higher pension payments to retirees and changed administrative costs. Recent literature, however, has added a new dimension to this discussion, focusing not on the impacts of retirees on the economy per se, but rather focusing on the impacts that the IS system may have on the retirement decision for retirees. Several of these studies suggest that Canadian IS programmes significantly influence individuals’ retirement decisions by providing labour supply disincentives. The retirement decision is commonly considered as a choice between income and leisure, where leisure is a normal good and income-generating employment is an inferior good. Consequently, as wealth increases, an individual will spend more time in leisure and less time in employment. Since income from security programmes increases the pool of wealth for individuals, it is logical to assume, then, that this wealth may also influence the retirement decision. However, actual retirement, the basis for the retirement decision and the many factors which lead to this actual outcome, are difficult to define and precisely measure. Various limitations arise, such as the accurate identification of those that have retired, as individuals who might have worked for an employer may become self-employed in later years, or may exit the workforce temporarily only to return years later. The results of these identification matters often lead to erroneously labelling individuals as retired when they are in fact working. Furthermore, findings that show Canadian IS programmes provide labour supply disincentives that may be misleading — since IS is not a main

Canadian Retirement Behaviour

293

factor in the retirement decision but, rather, a component of a combination of factors that includes household characteristics and personal wealth. The main focus of this study is to examine the Canadian IS system and its possible impacts on the retirement decision. The interrelation of other personal wealth factors and demographic factors will also be canvassed. The models and results from this study provide an extension to the models and results of other studies — most notably studies by Gruber (1997), Baker and Benjamin (1999a, 1999b), Baker et al. (2001) and Compton (2001). The paper is organized into six sections. Section 2 surveys the literature on the possible correlations between benefit take-up and work disincentives. Section 3 provides background information on the pension programmes in Canada. Section 4 describes our data set in detail, and provides various descriptive statistics. Section 5 focuses on the empirical framework used in this study and presents the results of our regression analyses. Section 6 provides concluding remarks and discusses potential future research directions.

2. Literature Review The concept of an individual’s decision to retire, known in the retirement literature as the ‘retirement decision’, has received a great deal of attention in Canada. Recent studies, however, are examining a new dimension of the retirement decision, focusing on the impacts that the Canadian IS system may have on the labour force participation of older individuals. Studies range from examining early retirement incentives that, collectively, the pension system may provide to examining the temporal link between retirement from the workforce and the take-up of particular benefits within the pension system. 2.1 All Components of the IS System Baker et al. (2001) examine the impact that the Canadian IS system has on the retirement decision of older individuals. Using an amalgam of surveys, census and other sources, the authors obtain a large sample of older workers with detailed information on their earnings histories, marital status, spousal and job characteristics and labour supply choices. These data sources also provide them with the necessary information to construct individuals’ entitlements to the federal IS system, which are then used to calculate the present discounted value of the stream of benefit entitlements for each possible retirement age. By constructing these entitlements, the authors are able to measure labour-market disincentives determined by the IS system through accrual effects that capture the change in labour behaviour with

294

Greg Maloney et al.

additional years of work. Using probit analysis as the basis for their empirical framework, the main finding by Baker et al., is that the IS system affects the retirement decision for individuals, by providing work disincentives beyond the age of 60 for both males and females, with disincentives being larger and more robust for males than for females. Gruber (1997) investigates the interaction between IS and the labour force behaviour of older individuals in Canada. Using data obtained from the Survey of Consumer Finances and a simulation model of benefits determination, which includes the present discounted value of the entire stream of benefit entitlements, Gruber finds that labour supply disincentives are inherent within the IS system for the current cohort of retirees. Within this finding, Gruber also shows that the net effect of the IS system on labour incentives is sensitive to whether the family is in the range of means-tested benefits. Furthermore, an older worker with low lifetime earnings and no outside income will face considerable work disincentives to continue working past 60 years. 2.2 Particular Components of the IS System Compton (2001) examines the relationship between labour force behaviour of older workers and take-up of Canada Pension Plan (CPP) and Quebec Pension Plan (QPP) benefits. Using the internal files of the Survey of Labour Income Dynamics (SLID), she finds that CPP/QPP benefits and retirement are not temporally connected for most Canadians. Furthermore, she finds no evidence that wealth affects the retirement decision. Instead, demographic factors such as disability status — and spousal labour force characteristics such as class of worker, union status and the presence of an employer pension plan — dominate her ordered probit regressions. Compton also questions the efficacy of minor adjustments to benefit levels as a means of changing retirement behaviour. Her analysis suggests that small changes to the CPP/QPP programme may not have an observable impact on labour force participation of older workers in the short term. Baker (2002) examines the effects that the Spousal Allowance (SPA) has on the retirement behaviour of couples. Based on data obtained from the Survey of Consumer Finances, Baker separately compares changes in the retirement behaviour of males aged 65–75 and females aged 60–64 who became eligible for the SPA to that of their counterparts of the same age who, due to the age of their spouse, did not qualify for SPA. Using the control groups as a counterfactual, Baker’s findings suggest that the availability of the SPA is associated with a relative increase in the proportion not in the labour force. Relative to each control group, eligible men experienced an increase in their non-labour force rate and a decrease in their employment rate with the introduction of the programme. Furthermore, Baker observes reductions in the

Canadian Retirement Behaviour

295

labour-market activity among eligible females and a correlated increase in joint absence from the labour market among eligible couples. Baker and Benjamin (1999a) examine the effects of the CPP and QPP early retirement provisions introduced in the 1980s on the labour force behaviour of males. Using the individual files obtained from the Survey of Consumer Finances, the main finding from their models suggests that the introduction of early retirement provisions led to significant increases in the CPP/QPP take-up among 60–64 year olds. However, these provisions had no effect on their labour force behaviour. The authors conclude that the lack of strong labour effects results from the fact that new pension beneficiaries were not men induced to retire early, but rather men who would not have been employed anyway. Baker and Benjamin (1999b) analyze the impacts of the retirement behaviour of older men by removing the CPP/QPP retirement/earnings tests established in the 1970s. Using a microdata set obtained from the Census Family files of the Survey of Consumer Finances, the authors’ OLS results suggest that the removal of earnings tests from Canada’s pension plans results in individuals delaying pension take-up beyond the age of 65. Thus, the removal of earnings tests results in a relative movement of older workers from part-year full-time employment to full-year full-time employment. To summarize, there appears to be mixed evidence that the Canadian IS system provides work disincentives beyond the age of 60. Among those that have found evidence of labour supply disincentives, their findings focus on examining the IS system, either in part or in whole, using varying degrees of empirical frameworks. However, the survey sources that each of these studies rely upon provide limitations within themselves. For instance, previous studies are not able to fully capture the effect of private pension plans; nonlabour income from non-labour market activities, such as investment income and rental income and CPP disability effects. Furthermore, because complete earnings histories and pension entitlements are not always reflected in these surveys, it is difficult to precisely measure the true impact that estimated wealth and substitution effects will have on the retirement decision for individuals. While the data set used in this study, the Canada Customs and Revenue Agency (CCRA) T1 MiniUniverse, does not provide complete earnings histories either, it does provide us with the unique opportunity to observe actual amounts of pensions in the case of CPP and Old-Age Security (OAS). Having these actual amounts may help to minimize any potential bias in our analyses. The T1 data set also allows a more robust identification of those individuals defined as self-employed. This will help us to eliminate any bias in our results that could come from mislabelling those who are self-employed as retired. Finally, the T1 data set allows the inclusion of more demographic factors that may influence the retirement decision.

296

Greg Maloney et al.

3. Canada’s Income Security Programmes In the context of IS programmes administered by the Federal government, Canada’s IS system comprises of the following four components: OAS, Guaranteed Income Supplement (GIS), the Allowance and Survivor’s Allowance (ASA) and CPP and QPP. A detailed description of each component of the IS system as well as a description of employer pension plans follows.

3.1 Old-Age Security The government-supported tier of Canada’s income security system distributes pensions through a combination of three programs: OAS, GIS and the ASA. Old-Age Security The OAS, established in 1952, is Canada’s largest public pension plan, providing monthly pension payments to most individuals from the age of 65 who meet certain residency requirements.1 The amount of OAS benefits that individuals may be eligible to receive is dependent upon the extent to which residency requirements are met.2 From its inception in 1952 through 1988, the OAS was a nonincome–tested pension plan. During this time, beneficiaries would receive the full amount of the OAS benefit regardless of their income or wealth status, having met all other requirements. This changed in 1989 with the implementation of a special tax or ‘clawback’, imposing a 15 per cent tax on a taxpayer’s net income in excess of $50,000 ($56,968 in 2002).3 In this way, 1 Eligibility for OAS involves the following requirements. Canadian citizenship or legal resident of Canada on the day preceding the application’s approval; if no longer living in Canada, must have been a Canadian citizen or a legal resident of Canada on the day preceding the day he or she stopped living in Canada and a minimum of 10 years of residence in Canada after reaching age 18. 2 A person who has lived in Canada, after reaching the age of 18, for periods that total at least 40 years, may qualify for a full OAS pension. A proportionate pension is payable for those with 10–40 years of residence after the age of 18. For example, an individual with 14 years of residence would receive 14/40ths of the maximum amount. Full pension can also be received by individuals who meet the following criteria: (1) born on or before July 1, 1952; (2) resided in Canada for some period of time between turning 18 years and July 1, 1977 and (3) was living in Canada 10 years immediately before application was approved. 3 For 1989, a 15 per cent tax was levied on net income in excess of $50,000.

Canadian Retirement Behaviour

297

the total OAS benefit amount is taxed away for high-income earners (earning approximately $93,000 in 2002). OAS benefits are also indexed to increases in the Consumer Price Index, with adjustments made on a quarterly basis to adjust for changes in the cost of living. This indexation practice has been in place since 1972. Guaranteed Income Supplement The GIS is the second of three benefits established under the governmentsupported tier of Canada’s IS system. It is a non-taxable supplement paid to low-income seniors living in Canada who already receive all or part of the OAS. The GIS is also subject to the OAS eligibility requirements and to an additional income test, whereby the supplement is reduced by $1 for every $2 earned of monthly income over and above the income received from the OAS pension.4 The supplement is determined by the previous year’s income and therefore must be re-applied for every year. Similar to the provisions of the OAS, the GIS is indexed to increases in the Consumer Price Index. Rates for the GIS differ between those who are living as couples and those who are not. The maximum amount of GIS distributed to individuals is the same for a single person and for those living as couples where a spouse or common-law partner does not receive either the OAS or the ASA. GIS amounts for a single person, or a person whose spouse or common-law partner does not receive either an OAS pension or an allowance, are also greater than the GIS received by individuals whose spouse or common-law partner receives either the OAS or the ASA. Allowance and Survivor’s Allowance The ASA, formerly known as the Spouse’s Allowance and the Widowed Spouse’s Allowance, respectively, is the third of three benefits under the government-supported tier of Canada’s IS system. The Allowance was established in 1975 with the first payment for beneficiaries of the Survivor’s Allowance occurring in 1979. To be eligible to receive this non-taxable benefit, individuals must either have a spouse or common-law partner who is already an OAS recipient, or be a survivor of a deceased pensioner. In addition, applicants must be

4

Income from other sources is defined by employment earnings, RRSP income, CPP/ QPP pension benefits (other than death benefits), private pension plans, foreign pension income, employment insurance benefits, interest on savings, capital gains or dividends, rental property income and income from other sources, among which include worker’s compensation and alimony. Pension payments and allowances under the OAS Act, and payments under the Family Allowance Act are not included.

298

Greg Maloney et al.

between the ages of 60 and 64, have lived in Canada for 10 years since the age of 18, and be a Canadian citizen or legal resident of Canada on the day preceding the application.5 There are also separate benefit entitlements for single individuals and those living as couples. As the ASA is applicant based, individuals must re-apply annually to receive this benefit. ASA benefits are subject to an income test, whereby the Allowance is reduced by $3 for every $4 of the couple’s income from other sources,6 not including other OAS benefits, until the amount of reduction is equal to the OAS pension. Above that point, it is reduced by $1 for every $4 of income. The Allowance is paid until either the date of death or until the month proceeding the date of reaching the age of 65, whichever is earlier. Payment is also stopped in the event of separation or the death of the pensioner or if the recipient is absent from Canada for more than 6 months. The Allowance for the survivor is paid until either the month of death, the month preceding the date of reaching the age of 65 or if that person remarries or lives in a common-law relationship for more than 12 months, whichever occurs first. Payment is also stopped if the recipient is absent from Canada for more than 6 months. 3.2 Canada and Quebec Pension Plans The CPP, established in 1966, is a jointly controlled federal-provincial contributory plan designed to provide monthly retirement pensions to qualified individuals.7 It is fully funded from the contributions of Canadian employees and their employers, self-employed individuals, as well as from the investments of the CPP Investment Fund. The CPP also acts as an insurance plan by providing disability benefits, a lump sum death benefit, benefits for children of disabled and/or deceased contributors and benefits for the surviving spouse/common-law partner of a deceased contributor. Under federal and provincial income-tax laws, the CPP is fully taxable and is fully indexed to the Consumer Price Index every

5 For those individuals who do not meet the 10-year residency requirement, the Allowance is prorated for each year the applicant has resided in Canada at the rate of one-tenth of the benefit. 6 Income from other sources is defined by employment earnings, RRSP income, CPP/ QPP pension benefits, private pension plans, foreign pension income, employment insurance benefits, interest on savings, capital gains or dividends, rental property income and income from other sources, among which include worker’s compensation and alimony. 7 In 2000–2001, 2.7 million Canadian retirees received $13.5 billion in CPP retirement benefits (Human Resources Development Canada, 2001, p. 1).

Canadian Retirement Behaviour

299

January. It applies to all provinces except Quebec. Individuals who work in Quebec contribute to the QPP, which provides similar benefits and is managed by the Quebec provincial government. All Canadian employees, employers and self-employed individuals, with some exceptions, are required by law to contribute to the CPP/QPP. Contributions are based on the annual salary of employees and net business income for those who are self-employed. Participants of the plan only pay contributions on annual earnings between a minimum level of the Year’s Basic Exemption (YBE) and the maximum level of the Year’s Maximum Pensionable Earnings (YMPE). In 2002, participants paid contributions on earnings between $3,500 and $39,100. The YBE (frozen at $3,500, since 1996) and the YMPE are adjusted each January based on increases in the average wage. Employees and their employers pay an equal portion of the contribution, while self-employed individuals pay both the employee and employer’s portion. In 2002, the maximum employee and employer’s contribution was $1,673.2 while the maximum self-employed contribution was $3,346.4.8 To qualify for CPP/QPP benefits, individuals must have contributed to the plan for at least one year, have reached age 60 and have stopped working or have earnings below a specified level of income for a period of time between the ages of 60 and 64. While full benefits are paid to participants at age 65, early or late retirements are allowed between ages 60 and 70. If individuals choose to retire between ages 60 and 64, the amount of the benefits will be reduced by 0.5 per cent for each month under the age of 65, up to a maximum reduction of 30 per cent for those claiming benefits at age 60. If individuals do not choose to retire between 60 and 65, the amount of their pension will increase by 0.5 per cent for each month over 65, up to a maximum increase of 30 per cent until age 70. The adjustment is permanent, as there is no re-calculation of benefits during the life of the plan. To compensate individuals for periods of unemployment, low earning years, time spent rearing children, sickness and disability, provisions of the plan allow certain periods to be ‘dropped out’ or ignored in the calculation of average earnings. These dropout provisions include periods receiving CPP disability benefits and periods of time spent rearing children under the age of seven. Additional dropout provisions involve the following: up to 15 per cent of the contributor’s lowest earning years prior to age 65, provided a minimum of 120 months are left in the contributory period; months included in a period of indemnity (QPP only); and periods after age 65 while contributing to CPP.9

8

For employees and employers, the contribution rate is set at 4.7 per cent. For selfemployed individuals, the contribution rate is set at 9.4 per cent. 9 For more information, see CCH Canadian Limited (2002).

300

Greg Maloney et al.

Once low earning periods have been removed, average adjusted earnings are then calculated. In general, the pension replaces about 25 per cent of the earnings which were paid into the plan, with the exact amount depending upon the amount and the length of the contribution and the age the individual commences benefit take-up.10 3.3 Employer Pension Programs In addition to pension income from the IS system, another important element of retirement income for individuals is income from employer pension plans (EPP). Approximately 40 per cent of workers in Canada are covered by an employer pension plan (Human Resources Development Canada, 2001). Under these plans, an employer deducts an employee’s required contribution and reports these totals on their T4 tax returns. Contributions are also tax deductible for employees. In Canada, EPP are classified under two types: defined benefit plans and defined contribution plans. Under a defined benefit plan, an individual can receive a monthly pension that is determined by a formula that may be a combination of earnings, job classification and/or length of service. Under a defined contribution plan, both the employer and employee contribute a set amount to the plan, usually based on a percentage of earnings (Human Resources Development Canada, 2001). The employer, on behalf of the employee, then invests these contributions, with the accumulated funds serving as the basis of an individual’s pension.11

4. Data and Descriptive Statistics 4.1 Data Source Experts have observed a lack of adequate panel data for retirement research in Canada. This study has taken a novel approach in developing a data set that addresses this deficiency by utilizing one of the most comprehensive databases available — the CCRA ‘T1 Mini-Universe’. This database draws upon the T1 tax returns of individuals in Canada, from 1995 to 2001, using selected fields for tax analysis purposes. The estimated total filing population of Canadians, over the age of 18 for each tax year that is represented by

10

In 2002, the average monthly benefit was $440.39 and the maximum was $788.75. For more information on EPP, see Human Resources Development Canada (2001).

11

Canadian Retirement Behaviour

301

the T1 Mini-Universe, is approximately 20 million individuals. In this study, we use an abridged version of this database. Although Canada’s taxation system relies upon voluntary self-compliance, individuals have an inherent incentive to file a tax return each year. Information, such as earnings or demographic information, is used in the calculation of tax credits and benefit payments — so accuracy and precision, to a high degree, are voluntarily achieved. However, there are instances where tax information is simply not captured. In our abridged data set, only a few instances of omissions occur and these are directly related to the marital status field. In minor instances where individuals do not state their marital status, and their status cannot be verified, their marital status is categorized as unstated. As a result of the potential bias in attempting to fit these individuals into a model, these observations are removed from the data set. The removal of these individuals represents less than 1 per cent of the data. 4.2 Description of Data Set The data set for this study is in the form of a panel or longitudinal data set based on the period of 1995–2000. Individuals are selected in 1995, and are followed over the remainder of the period. The selection of these individuals is based on the following two main criteria: individuals must be aged 60–70, and must also be employed. Individuals’ income from all sources is noted in Canadian dollars. The employment constraint implies that an individual must have positive T4 earnings. If an individual has zero T4 earnings in 1995, then that individual would be excluded from the data set. The objective in building this data set is to simply follow an individual’s observations to the point where the decision to retire is made, after which subsequent observations are dropped. Those individuals that have positive employment earnings through the duration of our data set are not included, because they do not retire within the observed timeframe. Individuals who are considered retired within the data set are identified on the basis of a year of positive employment earnings followed by a year of zero employment earnings. Thus, our definition of retirement is based on observing positive T4 earnings in a given year, followed by zero T4 earnings in the next year. Our definition of work is then based on observing two consecutive years of employment earnings. In the case of self-employed individuals, defining retirement based solely on T4 earnings would be an inappropriate measure of retirement. This is because many self-employed individuals do not work for an employer and, hence, do not receive a T4 slip. In this manner, individuals would have a zero recorded in their T4 earnings field. As a result of the obvious shortcomings that would result from the inclusion of this group, we then remove those individuals

302

Greg Maloney et al.

identified as unincorporated self-employed.12 Identification of this group can be determined by observing specific fields obtained from the T1 MiniUniverse.13 However, the limitation which this approach presents is that we cannot always determine whether an individual is strictly self-employed, having no employer earnings. To solve this limitation, self-employment is identified on the basis of the following two methods: primary income generated from self-employment, or T4 earnings equal to zero and self-employment income greater than the amount of public pension income they receive. Identifying individuals who are strictly self-employed enables us to minimize the attenuation bias that Baker et al. (2001) encounter. This bias involved mistakenly classifying individuals who may have had positive T4 earnings in one year and zero T4 earnings in the next year as retired when in fact they might have entered into self-employment.14 4.3 Data Methodology While the intent of this study is to examine the possible work disincentives that the IS system in Canada may provide to older individuals, there are other factors, among them EPP, that may affect the retirement decision. A discussion of these and other influencing factors follows. IS Incentive Variable Similar to the methodology used in Baker et al. (2001) and Gruber (1997), our analysis of possible work disincentives stemming from the IS programme requires constructing each individual’s IS incentives. To construct this, we must first estimate each individual’s IS entitlement. The IS entitlement, which factors in all of the public pension components of the federal IS incentive system, is calculated at the family level for those 12

Unincorporated self-employed pertains to those individuals who do not incorporate their business (i.e., conduct their business as a corporation) to engage in business activities that enable them to take advantage of various federal and provincial tax considerations, such as anticipated earnings, tax deferrals and savings, scientific research incentives, provincial incentives, and income splitting and estate planning (Anderson and Mallin, 2002). 13 Specific T1 fields that can provide self-employment information are as follows: field 222 on the T1 form (i.e., deductions for CPP or QPP contributions on selfemployment and other earnings), and fields 162–170 and fields 135–143 on the T1 form (i.e., self-employment gross and net income for various professions). 14 Baker et al. state that the effects of this bias are such that they will ‘underestimate the marginal effects of the characteristics of individuals and their jobs, as well as their IS incentives, on the probability of retirement’ (Baker et al., 2001, p. 16).

Canadian Retirement Behaviour

303

living as couples and at the individual level for those who are unmarried. This means that for individuals living in a married or common-law relationship, IS entitlement will also have to be calculated for their spouse or common-law partner (Baker et al., 2001, p. 17). These calculations, inherent in determining the amount of IS incentives, involve the take-up of CPP/ QPP, OAS, GIS and the ASA. A detailed description on the calculation of entitlements for each of these programmes follows. IS Entitlement Canada/Quebec Pension Plan. Our microdata set affords us the unique opportunity to observe the actual CPP/QPP benefits that individuals receive. As a result, we are able to bypass the investigative process of determining CPP/QPP entitlement using T4 earnings and projected earnings histories that previous studies have had to rely upon. Additionally, we are also able to factor in CPP disability benefits that previous studies have not been able to capture. Our microdata set also allows us to bypass the following programme intricacies: past and recent effects of reforms to the programme, drop-out provision for low earning months, children’s benefits and time off for childcare matters. The most obvious advantage of having observed CPP/QPP benefits is that individuals will continue to receive about the same level of benefits for each year thereafter. This is because the benefit contributory periods, as well as the intricacies of the program, have already been factored into the benefit payout individuals receive. The drawback of not having historical data on individuals, however, is that we cannot precisely determine CPP/QPP entitlement at each possible retirement age for which we do not observe a benefit amount. To solve this limitation, a proxy method is constructed to calculate past entitlements (see Appendix A for formula). By observing the first instance of benefit take-up, this method calculates the proportion of the amount an individual received in the first year of observation of the total maximum possible benefit that could have been received during that year.15 The proportion is then multiplied by the maximum benefit entitlement for that year and by the age adjustment factor for those before the age of 65. This method thus allows for an approximate measure of what an individual’s CPP entitlement would have been if he or she had retired before we could observe the first instance of their actual benefit take-up.16

15

The maximum amount of CPP/QPP benefits provided in a particular year is based on individuals participating in the programme from its inception and contributing the maximum yearly amount from that point on until retirement. 16 Those individuals who do not have an observed CPP/QPP amount during the time period in the study are thus removed from the data set.

304

Greg Maloney et al.

Old-Age Security. A similar situation exists with respect to OAS entitlements. Again, our microdata set provides us with the actual benefit amounts that individuals receive. The most obvious advantage this yields is that we do not need to concern ourselves with residency requirements. Unlike the provisions in the CPP/QPP that allow benefit take-up before the age of 65, the OAS benefit is provided only to individuals who reach the age of 65 and meet certain residency requirements. Consequently, there is no need to calculate OAS entitlements at each possible retirement age before 65 years. Guaranteed Income Supplement and the Allowance and Survivor’s Allowance. The remaining two federal pension programmes, the GIS and the ASA, are eligibility-based programmes associated with take-up of the OAS program. All components of these programmes, including age requirements, benefit clawbacks and/or spousal requirements, are fully implemented within the data set. ISW Entitlement As a precursor to constructing an ISW variable for individuals, an assumption pertaining to the amount of income individuals receive during their retirement years is necessary. The assumption concerning retirement income is such that retirement income received by individuals is relatively constant for all post-retirement years. This assumption is based on two factors: (i) we only observe individual observations until they retire or reach the age of 70, whichever event comes first; and (ii) pension benefit take-up, specifically the ASA, GIS and the OAS are fully or partly means tested. As such, it is important to introduce this assumption — otherwise, we would not be able to construct an appropriate gauge with which to measure pension benefit take-up during retirement years. The final stage in calculating IS entitlement for individuals involves calculating the net present value (NPV) of their ISW for each possible retirement year (see Appendix A for formula). Similar to both Baker et al. (2001) and Gruber (1997), NPV is calculated on the basis of marital status. For single workers, NPV is simply the present worth of their future benefits discounted by the number of years of life expectancy at each possible retirement age, and at a real discount rate of 3 per cent.17,18 For married or common-law workers, NPV is discounted by the number of years of jointlife expectancy and at the same real discount rate of 3 per cent.19 17

Life expectancy figures are taken from the Statistics Canada (1984). In this study, we adopt the discount rate of 3 per cent used in Baker et al. (2001). 19 Joint-life expectancy figures are taken from the Statistics Canada, ‘Marriage, Divorce and Mortality: A Life Table Analysis for Canada and Regions’ catalogue. 18

Canadian Retirement Behaviour

305

The NPV calculations for all individuals results in what Baker et al. call the present discount value (PDV) of all individuals at each possible retirement date (Baker et al., 2001, p. 19).20 This in turn allows us to construct two incentive variables that allow for varying degrees of inference on each individual’s ISW. The first of these two incentive variables, the one-year accrual variable, simply examines the difference between adjacent retirement ages. The second of these two incentive variables, introduced by Coile and Gruber (2000) as the peak value accrual, is simply the difference between the ISW at a given retirement age and the ISW at the most financially opportune time for retirement.21 Employer Pension Plans An important input into the retirement decision for individuals is the amount of an EPP that they may be entitled to. Although Baker et al. factor this input into their empirical framework, they do so relying upon two survey-based cross-section samples of males and females that allows them to only impute the probability of individuals having a private pension. This probability is based upon the industry that workers are employed in (Baker et al., 2001, p. 15). While the entire stream of IS wealth is certainly an input into the decision to retire, focusing more or less on the work disincentives that may or may not be found within the IS system provides limited inference. This is because income security is not a main factor in the decision to retire, but rather a component of a combination of factors that includes, among other factors, EPP. In the same manner in which the ISW is constructed, EPP for each individual in this study will also reflect the entire stream of private pension wealth at each possible retirement age. The inclusion of this input may provide more inference as to why individuals may retire earlier. In much the same way that we observe actual benefit amounts for the previous pensions, our data set also provides actual amounts of EPP for

20

Although not all pensions are available to individuals before the age of 65, it is assumed that individuals are forward-looking in nature when making the decision to retire. The PDV is based on this assumption and reflects the pensions available to individuals at each possible retirement age. For instance, if an individual retires at the age of 60, and is expected to live until the age of 77, that PDV of the ISW might only reflect CPP/QPP income for the first five periods, with the additional pensions, including CPP/QPP, reflected in the remaining periods. 21 Baker et al. describe the peak value accrual approach as being able to accommodate non-linearities in the PDV of ISW profiles (Baker et al., 2001, p. 20). The idea behind this approach is that working additional years may in fact yield a higher ISW for individuals since CPP provisions allow for contribution until the age of 70.

306

Greg Maloney et al.

individuals. Assuming that EPP amounts are relatively constant each year after the last observed EPP amount for each worker in our data set allows us to then calculate the NPV of the entire stream of pension benefits at each possible retirement age. Other Retirement Inputs Additional inputs in the retirement decision that we include in our empirical framework can be broken down into household characteristics and personal wealth characteristics. In terms of household characteristics, we include the following inputs: non-labour income, such as rental income and investment income in any given year; earnings and spousal earnings; and average and spousal pensionable earnings. In terms of personal characteristics, we include marital status, disability status, and the number of dependents that an individual has. Descriptive Statistics Descriptive statistics are presented in Table A1 in the Appendix. The total number of observations in our data set is 333,626, comprising 107,362 individuals. Our data set contains more females (61,333) than males (46,029) — and on average each individual contributes 2.89 observations to the data set. Furthermore, 42.83 per cent of individuals are found to be either married or common-law, while 57.13 per cent are considered single. Single status includes those who are widowed, divorced, separated and single. Finally, amongst those who are considered married or common-law, the average age differences between couples are found to be 2.48 years for females and 1.3 years for males.

5. Empirical Framework and Findings The retirement model used in our study focuses on a given two-year period for individuals. In this period, individuals are working in the first year (t), and will make their decisions during that year on whether to continue working or retire in the next year (t+1). Implicit in the retirement decision of individuals is that they are forward-looking. That is, they take into account the entire stream of their Income Security Wealth (ISW), as well as economic and demographic factors similar to those discussed in both Baker et al. (2001) and Gruber (1997). If we assume that leisure is a normal good, an individual’s entitlement to their ISW, in the context of a retirement decision, might very well be considered an income effect. This is because more leisure will be consumed with an increase in wealth. In the same manner, employer pension wealth might

Canadian Retirement Behaviour

307

also be considered as an income effect. On the other hand, the change in wealth brought about by an additional year of work, as captured by the economic and demographic variables, might then be considered as a substitution effect. The empirical framework used to examine the retirement decision that individuals are faced with involves a probit analysis of the following equation: Rit ¼ b0 þ b1 ISW it þ b2 ACC it þ b3 NLI it þ b4 PPW it þ b5 X it þ git

ð1Þ

In this equation, the dichotomous variable Rit takes on a value of 0 if an individual is working and a value of 1 if an individual is retired. ISWit represents the present discount value of federal ISW individuals may receive, comprised of CPP, OAS, GIS and ASA benefits, in the actual years of retirement.22 ACCit, identified in the literature as an accrual variable, allows for the construction of two incentive variables that provide for varying degrees of inference on each individual’s ISW. The first of these two incentive variables, the one-year accrual variable, simply examines the difference of ISW between adjacent retirement ages. The second of these two incentive variables, the peak value accrual, is simply the difference between the ISW at a given retirement age and the ISW at the most financially opportune time for retirement. NLIit represents the amount of non-labour income attained from nonlabour market activities, such as rental, investment and dividend income; PPWit, or private pension wealth, is the stream of earnings received from EPP that individuals receive during the years of retirement; and Xit represents a set of control variables for marital status, number of dependents, disability status, province of employment and other descriptive elements.

5.1 Empirical Results The results of our preliminary estimates are presented in Table 1. Results are broken down into two main components to examine possible work disincentives that may be inherent within the Canadian IS system. Specifically, these components involve the one-year accrual and the peak-value accrual. 22

Estimation of the model does not account for the potential endogeneity of the ISW to the retirement decision. A possible cause for the endogeneity could be the correlation between the ISW with some unobserved component of the retirement decision such as health or the self-assessment of life expectancy. See discussion in Compton (2001).

Greg Maloney et al.

308 Table 1:

Estimation Results Accrual — one year 1

2

3

4

NLI

1.29E-06 (1.99E-08) [4.43E-07] 6.47E-06 (6.53E-08) [2.23E-06] N/A

PPW

N/A

Male

N/A

1.17E-06 (2.02E-08) [4.34E-07] 6.49E-06 (6.51E-08) [2.18E-06] 1.28E-05 (6.18E-07) [4.3E-06] 1.44E-06 (1.6E-08) [4.86E-07] N/A

9.8E-07 (2.8E-08) [3.3E-07] 6.55E-06 (6.42E-08) [2.21E-06] 1.33E-05 (6.28E-07) [4.49E-06] 1.43E-06 (1.61E-08) [4.83E-07] N/A

Female

N/A

N/A

Married

N/A

N/A

0.0265 (0.00437) [0.00893] 0.0627 (0.0061) [2.11E-02]

9.8E-07 (2.8E-08) [3.3E-07]] 6.55E-06 (6.43E-08) [2.21E-06] 1.33E-05 (6.28E-07) [4.49E-06] 1.43E-06 (1.61E-08 [4.83E-07] 0.0265 (0.00437) [0.00893] N/A

ISW

ACC

0.0627 (0.0061) [2.11E-02]

Accrual — peak value 1

2

3

4

NLI

1.89E-06 (1.87E-08) [4.54E-07] 2.94E-06 (3.87E-08) [2.28E-06] N/A

PPW

N/A

Male

N/A

1.8E-06 (1.89E-08) [6.21E-07] 2.81E-06 (3.85E-08) [9.72E-07] 1.07E-05 (5.93E-07) [3.7E-06] 1.39E-06 (1.6E-08) [4.82E-07] N/A

1.22E-06 (2.81E-08) [4.21E-07] 3.28E-06 (4.25E-08) [4.21E-07] 1.2E-05 (6.1E-07) [4.13E-06] 1.39E-06 (1.61E-08) [4.79E-07] N/A

Female

N/A

N/A

Married

N/A

N/A

0.00436 (0.00436) [0.0015] 0.187 (0.00653) [0.0643]

1.22E-06 (2.81E-08) [4.21E-07] 3.28E-06 (4.25E-08) [2.21E-06] 1.2E-05 (6.1E-07) [4.13E-06] 1.39E-06 (1.61E-08) [4.83E-07] 0.00436 (0.00436) [0.00893] N/A

ISW

ACC

0.1866 (0.00653) [0.0211]

Notes: ( ) ¼ standard errors; [ ] ¼ marginal effects. All coefficients are significant at the 1 per cent level.

Canadian Retirement Behaviour

309

Within these components, we introduce various determinants and control factors to examine the change in the ISW. Accrual: One Year The first column of the model specification for the one-year accrual shows several interesting results that are each statistically significant. First, the coefficient is positive, suggesting that an increase in ISW will influence an individual’s decision to retire. In this regard, leisure is a normal good. However, the coefficient for the accrual variable is negative, having a relatively stronger impact on the decision to retire over the ISW. This suggests that the likelihood of retirement is lowered, since an individual may be financially better off to wait an additional year before retiring in order to accumulate more ISW. In this manner, individuals might be observing their opportunity cost of retiring early, favouring instead waiting until a more financially optimal time. Despite the statistical significance of each of these factors, their marginal effects are extremely low. Thus, their true impacts on the decision to retire may be minimal at best. In the second column, more determinants are introduced that further weaken the significance of the ISW in the decision to retire. While the signs of the coefficients for the accrual and the ISW variable remain the same, the significance of the accrual variable is even stronger compared to before. Furthermore, employer pension wealth and non-labour income, represented by PPW and NLI, respectively, have stronger marginal effects on the decision to retire than the ISW. The PPW and the NLI variables may very well undermine the significance that Baker et al. (2001) and Gruber (1997) place on the ISW in influencing the decision to retire. For PPW, the results seem to imply that employer pensions may exert a stronger influence on the decision to retire than ISW. Similarly, the results for NLI seem to imply that non-labour income such as rental income, investment income and/or dividend income may also have more of an impact on the retirement decision than ISW. The results found in both the third and fourth columns show that as more determinants and control variables are added to the probit model, the weaker the ISW becomes. This result is further evidenced by the decline in the marginal effect of the ISW on the decision to retire. Here, factors such as sex and marital status dominate the regressions. For females, the results indicate that their likelihood of retirement is lower than males. Although this result may appear contradictory, this result may be potentially biased by the fact that our data set, and for that matter the Canadian tax filing population over the age of 18, is comprises more females than males. As well, the average life expectancy of females is much higher than males. The results for marital status seem to indicate that the likelihood of retirement over the age of 60 for a married couple increases.

310

Greg Maloney et al.

Accrual: Peak Value The corresponding set of results for the model that incorporates the peakvalue accrual approach also provides several statistically significant findings. The main similarities between both accrual methods are that the signs of the coefficients for the accrual variables themselves remain the same (i.e., negative). In column 1, the marginal effect of the ISW is slightly stronger under the peak-value method than under the one-year accrual method. This implies that the likelihood of retirement increases under the peak-value approach. However, the coefficient for the peak-value accrual variable is larger than the one-year accrual variable presented earlier. This peak-value method suggests that individuals may delay the decision to retire, preferring instead to retire at the most financially optimal time. This may involve postponing retirement until after the age of 65, as CPP is increased by the age adjustment factor for each year after the age of 65 that an individual does not collect CPP. Column 2 provides further evidence that employer pension wealth and non-labour income, represented by PPW and NLI, respectively, have a stronger impact on the decision to retire than the ISW as evidenced by comparing their marginal effects. These results are quite similar to the oneyear accrual method. The results found in both columns 3 and 4 are also similar to the corresponding results found under the one-year accrual method, with the exception of the sex variables. Under the peak-value accrual method, these variables are not statistically significant. To summarize, while the ISW is found to provide labour supply disincentives, the observed marginal effects show that these disincentives are minimal at best. Instead, factors such as employer pensions and non-labour income dominate the regressions, providing stronger marginal effects on the decision to retire. Furthermore, the opportunity cost of retiring early as seen through the accrual variables suggests that delaying retirement until the most financially optimal time may be the preferred choice among Canadians.

6. Conclusion and Potential Future Research Preliminary findings show that ISW has a minimal impact at best on the retirement decision of Canadians. Instead, our findings suggest that the decision to retire is influenced by more prevalent factors such as non-labour income and EPP. In the context of the Canadian IS system, these preliminary findings contradict the notion by previous studies that the IS system provides

Canadian Retirement Behaviour

311

strong retirement incentives. Under the assumption that individuals are by nature forward-looking in their decision to retire, our results suggest that individuals will delay the decision to retire in order to maximize their ISW. As time and resources permit, further avenues of analysis of our more comprehensive model and findings will be explored in the future.

Acknowledgement This paper would not have been possible without the excellent work from Monica Yeung and Denis Milette. We are also grateful to Robert Behrend, Nicole Bowie, Lyndsy Chowns, Diana Goldin, Janine Hum and Lynn Smith. Lastly, we would like to thank Christopher McCann for his support in this project.

References Anderson, A. and Mallin, M.G. (2002). Preparing Your Income Tax Returns: 2002 Edition for 2001 Returns, CCH Canadian Limited, Canada. Baker, M.A. (2002). The Retirement Behaviour of Married Couples — Evidence from the Spouses Allowance. Journal of Human Resources, 37(1), 1–34. Baker, M. and Benjamin, D. (1999a). Early Retirement Provisions and the Labor Force Behavior of Older Men: Evidence from Canada. Journal of Labor Economics, 17(4), 724–756. Baker, M. and Benjamin, D. (1999b). How Do Retirement Tests Affect the Labor Supply of Older Men? Journal of Public Economics, 71, 27–51. Baker, M., Jonathan, G. and Kevin, M. (2001). The Retirement Incentive Effects of Canada’s Income Security Programs. Working Paper no. 8658, National Bureau of Economic Research, December, United States. CCH Canadian Limited (2002). The Handbook of Canada Pension and Benefit Plans12th edition. CCH Canadian Limited, North York, ON. Coile, C. and Gruber, J. (2000). Social Security and Retirement. Working Paper no. 2000-11, Center for Retirement Research at Boston College, Boston, December. Compton, J. (2001). Determinants of Retirement: Does Money Really Matter? Working Paper no. 2001-02, Department of Finance, February, Canada. Gruber, J. (1997). Social Security and Retirement in Canada. Working Paper no. 6308, National Bureau of Economic Research, December, United States. Human Resources Development Canada (2001). Annual Report of the Canada Pension Plan: 2000–2001. Human Resources Development Canada, Ottawa. McGregor, G. (2001). A Fact Sheet on the Economics of Ageing in Canada. Law Commission of Canada, Ottawa. Statistics Canada (1984). Life Tables, Canada and Provinces, 1980–1982. Minister of Industry, Ottawa, (Catalogue 84-532).

Greg Maloney et al.

312

Appendix A. Proxy Method Calculation of the proportion of the maximum benefit: proportion of maximum CPP benefit ¼

Actual received amount Maximum CPP benefit

Age adjustment factor: CPP pension is reduced by 0.5 per cent for each month, for a total of 6 per cent for each year, that the take-up of benefits precedes an individual’s 65th birthday. Proxy formula: Expected CPP ¼ proportion of maximum CPP benefit  maximum CPP benefit  age adjustment factor Net present value: NPV calculations — if individuals decide to retire at a certain age, the NPV is calculated starting from that period until life expectancy. Formula used to calculate the NPV of the ISW: x x x x x ISW it ¼ x þ þ þ þ þ 1 2 3 4 ð1 þ rÞ ð1 þ rÞ ð1 þ rÞ ð1 þ rÞ ð1 þ rÞn1 Based on this formula, if an individual decides at age 64 to retire and expects to live until the age of 70, then the NPV will be based on 6 periods (n ¼ number of periods).

Table A1: Summary Statistics

Mean

Median

SD

Max

Min

Females ISW PPW1 NLI2 Age Age difference

187,839.85 163,026.50 110,871.56 145,168.42 103,634 139,019.79 5,618.22 1,793 10,595.25 64.25 64 2.36 2.48 2 3.15 Males

1,428,258 1,858,807 148,928 70 14

2.83 14.75 1 60 9

ISW PPWa NLIb Age Age difference

186,906 157,254.80 100,085.13 176,710.11 136,069.50 155,739.20 7,245.47 2,030 13,648.18 63.99 64 2.29 1.3 1 2.87

1,384,234 143,870 148,928 70 9

0.92 11.64 1 60 12

Canadian Retirement Behaviour

313

Table A1 (Continued )

Males Females Married Common-law Singlec Number of individuals Total observation Average number of observations a

Counts

Percentage

46,029 61,333 45,361 625 61,376 107,362

42.87 57.13 42.25 0.58 57.17

333,626 2.89

25.78 per cent of females and 22.66 per cent of males receive pensions from other sources. 5.53 per cent of females and 2.52 per cent of males receive non-labour income. c Widows, divorced, separated and single individuals are all grouped as ‘single’ status. b

This page intentionally left blank

Chapter 12

STINMOD: Use of a Static Microsimulation Model in the Policy Process in Australia Rachel Lloyd National Centre for Social and Economic Modelling (NATSEM), University of Canberra, Australia

Abstract STINMOD is NATSEM’s static microsimulation model of Australia’s tax and transfer system. In the 10 years since it was first developed, STINMOD has been used as a tool providing input to Australian policy development and for research into the tax and transfer system. This paper gives an overview of STINMOD and recent enhancements made to its structure. It also discusses recent examples of the use of STINMOD in policy advising and research, including the costing of a paid maternity leave scheme and analysis of the distribution of effective marginal tax rates for the working population.

1. Introduction STINMOD is the National Centre for Social and Economic Modelling’s (NATSEM) static microsimulation model of Australian income taxes and cash transfers (STINMOD stands for the STatic INcome MODel). It is publicly available, runs on a personal computer and can be accessed via a user-friendly interface. STINMOD can be used to analyse the distributional impact of current tax-transfer policy or to estimate both the fiscal and distributional impacts of policy reform (see Lambert et al., 1994, for more information on STINMOD). It is now the standard model used by Australian federal government departments for their analyses of possible budget policy options in this area. The first version of STINMOD was released in 1994. Since then, the modules that simulate the tax and transfer systems have been largely rewritten, to take account of major changes in the rules of the programmes being modelled — including those associated with the introduction of the International Symposia in Economic Theory and Econometrics, Vol. 15 r 2007 Published by Elsevier B.V. ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15012-7

316

Rachel Lloyd

Goods and Services Tax reform package in July 2000. The progressive release of more recent survey data has allowed the base data to be regularly updated and other enhancements — such as the introduction of an effective tax rates model and the ability to project into future years — have been introduced. The purpose of this paper is to provide an overview of STINMOD and discuss ways in which it has been used in the last few years in policy and research. Section 2 gives an overview of STINMOD. Section 3 looks at some of the ways in which STINMOD has been used as the basis for other models, as an input to policy processes and for research. Section 4 concludes.

2. Overview of STINMOD STINMOD is NATSEM’s publicly available static microsimulation model that simulates the payment of personal income taxes and the receipt of social security and family payments cash transfers. STINMOD is used to estimate the impact of these systems on Australian families and on the government budget. In essence, STINMOD applies the rules of the income tax and government cash transfer programmes to a database of income units1 representing the Australian population (Bremner et al., 2002). From a modelling perspective, STINMOD can be conceptualised as having two major components (Figure 1). The first is the suite of ‘entitlement’ modules. The algorithms in these modules simulate the policy rules of the major federal tax and transfer programmes, including eligibility, entitlement and interaction. These rules are translated into computer code using SAS software. The second component is the basefile. The policy rules are applied to a population database comprising income units, the individuals within which make up a representative sample of the Australian population (Bremner et al., 2002). This database is referred to as the basefile, and is constructed from the representative population samples interviewed in national Australian Bureau of Statistics (ABS) surveys. The standard base data for STINMOD is the unit record data from the latest ABS Survey of Incomes and Housing Costs (SIHC) — though it can also run on the data from the Household Expenditure Survey (HES). Each record in the basefile, representing an income unit, contains information about the income unit as a whole and information on each person in 1 The ABS defines an income unit as ‘one person or a group of related persons within a household, whose command over income is assumed to be shared. Income sharing is assumed to take place within married (registered or de facto) couples, and between parents and dependent children’ (ABS, 2001).

Use of STINMOD in Australia Figure 1:

317

Overview of STINMOD

Veteran’s Affairs

Income Uprating Factors

Hypothetical families

Family & Community Services Family Tax Benefit Youth Allowance

Program Parameters

Tax Medicare Levy SIHC & Institutionalised Population

Reweighting Factors (Labour Force Survey)

Current Base Population

Child Care Benefit HECS

Base Output File (pre- policy change)

User Interface

New Output File (post-policy change)

OUTPUT Analysis of policy change (compare new world to base world)

the income unit. These basefiles include a wide range of demographic and economic indicators — as well as income unit, family and household structure. Each person, income unit, family and household in the HES or SIHC data sets has a unique identifier attached to it. These identifiers allow users to link people in the same income unit, family or household (Bremner et al., 2002). In this way, the impact of policy changes can be investigated with respect to not only narrowly defined groups of individuals but also by types of families. Where possible, the data from two surveys are combined, to provide a greater sample size. In addition, in some versions of STINMOD, records representing a synthetic sample of the institutionalized population are included (see Percival and Lim, 1999, for more detail). In these cases, the complete basefile contains about 20,000 income unit records. There is a lag between when surveys are taken and when they are made available in unit record format from the ABS so, in creating the basefiles, the microdata are statically aged to better reflect the current world.

318

Rachel Lloyd

There are three important elements in generating a STINMOD base population file. These are:   

calculating the weights attached to each income unit; uprating private incomes; and imputing family links (along with other miscellaneous imputations required to model programmes).

Each of the individuals in an income unit has to be assigned a weight. This weight represents the likelihood of finding persons with a similar set of characteristics in the Australian population. Weights generated by the ABS are provided in the ABS microdata files, but these apply to the time of collection of the surveys — and therefore need to be adjusted to better match up-dated administrative programme numbers and compositional changes in the population. Unit records are reweighted using ratio reweighting and the CALMAR software (Bremner et al., 2002; Deville and Sarndal, 1992) to match targets from such sources as ABS population numbers and labour force data, and Department of Family and Community Services (FaCS) (and other) administrative data. The aim is to match STINMOD output as closely as possible to available administrative, survey or census data. Private incomes — such as earnings and investment incomes — are uprated from the year of the survey to the year of interest, using changes in average weekly earnings and the consumer price index. Similarly, housing costs such as mortgages and rent are uprated, using changes in the parts of the consumer price index related to housing. The value of transfer payments recorded in the surveys is not uprated, since STINMOD calculates these amounts for the time period under examination. As the ABS microdata do not contain all the information needed to be able to accurately apply the rules of the government cash transfer and income tax systems modelled in STINMOD, a number of imputations need to be performed as well. For example, the SIHC does not contain any information about parental income and family structure for single income units (young adults) living away from home and still considered dependants for cash transfer purposes. This information is vital for accurately determining Youth Allowance and family assistance outcomes. Similarly, the information needed to determine workforce independence for single income units (which changes entitlement to Youth Allowance) is not collected in the surveys, so it must be imputed in STINMOD (Bremner et al., 2002). In addition to the distributional model, STINMOD also includes a hypothetical or effective tax rates model, which provides outcomes for a hypothetical family. A number of family types are pre-set within the model or users can create their own type of family. Users can analyse effective tax

Use of STINMOD in Australia

319

rates under current and alternative policy scenarios (see Section 3.5 for more details). Each year, a version of STINMOD with a user-friendly interface is released. This provides a convenient method for quantifying the immediate distributional and revenue implications of changes to payment rates and means test arrangements in a wide range of government programmes. Users do not require knowledge of SAS (the computer language in which STINMOD is constructed). Figures 2–5 illustrate the STINMOD interface front screen, the tax scales parameter screen and examples of the fiscal and distributional output screens — in this case showing the results arising from implementing a flat tax (that is, a proportional income tax) of 25 per cent for all incomes over $7,000. To analyse the impacts of such a policy, the user enters these parameters into the tax scales parameter screen (Figure 3) and then uses the point-and-click menus to run the distributional version of the model. Figure 4 shows that the impact on the government budget of this policy change is a decrease in revenue of over $13 billion — all impacting on the Tax Office (with no impact on the Departments of Family and Community Services or Veterans’ Affairs). Figure 5 shows that about one-third of families gain from this policy change, one-third lose, and one-third face no change. The winners, however, gain more on average than the losers lose. The interface version is publicly available, in that government departments, community groups, university researchers and students, as well as Figure 2:

STINMOD’s Front Screen

320

Rachel Lloyd

Figure 3: Tax

STINMOD’s Tax Scales Interface Screen: Simulating the Impact of a Flat

Figure 4: Change

Example of STINMOD Output: Summary Revenue Impact of the Flat Tax

members of the general public, are able to purchase the model. This is done under a licence agreement with the Commonwealth of Australia, which owns the intellectual property, whereby NATSEM can use the model both for research and commercial exploitation including the right to sub-licence.

Use of STINMOD in Australia

321

Figure 5: Example of STINMOD Output: Summary Distributional Impact of the Flat Tax Change

A small charge is imposed for a licence, some of which goes to the ABS as a contribution towards the survey data on which the model is based and some as a fee for the ABS approval process that ensures that confidentiality is maintained. Some of the fee goes to SAS for a cut-down version of the software for users that do not have a full SAS licence. As part of the licence agreement, users are required to acknowledge that they have used STINMOD in any output. Users with the interface version have the ability to change parameters and access output, but are not able to manipulate the underlying database. Using the interface version of the model does not require special expertise beyond an ability to use Windows and drop-down menus and some knowledge of the welfare and tax system of STINMOD models. Installation is done by users through a set of sequential steps, like many other types of software. A substantial user guide that explains installation, how to use the model (with examples), and information about the government programmes it models, is provided with STINMOD. If users have further queries or problems, they are able to access user support from NATSEM. Benchmark data is available in the model and some of the limitations of the data are explained in the user guide. Regular user groups are held and these issues are explored more fully at those sessions. For users who want to model entirely new programmes or make complex changes to existing programmes, a source code version of STINMOD is available (subject to Commonwealth approval). This consists of the many thousands of lines of SAS computer code lying behind the user-friendly interface and users require sophisticated programming skills and a good

322

Rachel Lloyd

understanding of the government programmes and their interactions to effectively use the source code. 2.1 Outyears Facility In response to requests from users in federal government departments — who needed to provide ‘outyear’ (that is, projected future year) costings as part of the budget process — STINMOD is now an outyears model, with users able to choose the financial year they wish to analyse2 (Bremner et al., 2002). As a result, STINMOD has a series of basefiles, some of which are based on historical data and some of which are projected base data sets. Basefiles are produced for two points in the current year and for five years into the future. Early STINMOD basefiles were based on known benchmark data. The move to projected basefiles required the development of a projection methodology. Private incomes are uprated based on expected changes to average earnings and the consumer price index. The projected basefiles are reweighted against projected population characteristics and customer numbers. In addition to the base data set, early versions of STINMOD included a single set of SAS data sets that contained values for the parameters of the tax and transfer programmes modelled in STINMOD. These data sets are referred to as parameter files. As a result of the outyears functionality, one set of parameter files for each financial year covered is produced and provided with STINMOD. The parameters are projected in line with indexation provisions and associated projections in the consumer price index and male total average weekly earnings. Prior to the introduction of the outyears facility, most users of STINMOD used the ‘current year’ basefiles and parameter files constructed by NATSEM. In theory, users could generate their own basefiles using the NATSEM basefile creation programmes and their own assumptions. However, this required advanced programming skills and a considerable investment of time. The way in which the outyears functionality has been developed has made it easier for users to include their own assumptions about future labour force, economic and demographic outcomes. This is done by editing a spreadsheet containing those assumptions — and then using custom menu options to create new STINMOD basefiles and parameter files that reflect those assumptions (Figure 6). The process is thus automated and much more user friendly. 2

The development of the outyears capacity in STINMOD stemmed from an outyears version of the model created by the federal treasury.

Use of STINMOD in Australia Figure 6:

323

Outyears Basefile Creation Interface

2.2 Regionalising STINMOD Regional issues have recently assumed much greater importance in Australia. There is a growing realisation that the gains from economic growth have not been equally distributed amongst different regions in Australia. For example, the overall stability in national poverty rates since the early 1980s appears to have disguised increasing poverty and inequality in many areas of regional and rural Australia (Vinson, 1999; Gregory and Hunter, 1995; Harding et al., 2001; Lloyd et al., 2001). During the past three years, NATSEM has developed regional microsimulation techniques to assist research and policy development directed at these issues. The techniques and the ways they have been used in conjunction with STINMOD are described in Section 3.7. With support from a number of state and territory governments, the ABS, and the Australian Research Council, NATSEM is currently developing a regional version of STINMOD (Harding et al., 2004a; Chin et al., 2005; Chin and Harding, 2006 — and see the chapter on spatial microsimulation in the accompanying volume to this one). This will allow extensive analysis of the regional distributional effects of the tax and transfer systems.

324

Rachel Lloyd

3. STINMOD and Policy 3.1 Returns to Education The Returns to Education Model (REDMOD) is an example of a model that uses STINMOD as an input. REDMOD estimates the returns to individuals and to the government of people’s decisions to pursue different levels of education and different fields of study. For example, the model can be used to estimate the net lifetime benefit to an individual and to the government if the individual completes a particular university degree. These benefits can be compared with the net public and private benefits for an individual who does not complete a university degree, or completes a different university degree. The model calculates net private and government returns to education for hypothetical individuals and couples as well as groups of people with the same characteristics. The hypothetical part of the model allows the user to specify the characteristics of hypothetical individuals or couples. The model also comes with sets of ‘typical’ or basic singles and couples cases and a set of default earnings profiles. In the group part of the model, groups are defined to represent the entire cohort of (initially) 18-year-olds, rather than just examining ‘typical’ cases. In the hypothetical and the group components of the model, the costs of education for individuals and couples are calculated as the tuition costs, Higher Education Contributions Scheme (HECS)3payments and taxes (including income tax, indirect taxes and tax on superannuation). The benefits are calculated as earnings from employment, government transfer payments (unemployment benefits, student assistance, age pension, etc.) and superannuation. Costs to the government include the transfer payments, education subsidies and tax subsidies on superannuation, while the benefits include taxes and HECS contributions. The model does not include the less tangible benefits of education, such as improved employee productivity, enhanced innovation or a more robust democracy. By examining rates of return to education at the individual level, the model has a high degree of flexibility that enables changes to virtually all parameters. These include personal details, labour force participation, government policy options and general economic factors. The model has three distinct phases: data assembly, calculation and presentation of results. The data assembly is conducted in Excel. This involves specifying the characteristics of the individuals that the user wishes to examine as well as specifying general parameters of the model, such as the level

3

HECS is Australia’s income-contingent student fees scheme.

Use of STINMOD in Australia

325

of HECS charges associated with courses. Calculations of the private and public returns to education are performed in SAS using STINMOD. REDMOD-specific modules in STINMOD calculate superannuation and HECS payments. The results are output in Excel to allow the presentation to be adapted to suit the needs of the user. For more information, see Johnson et al., 2002. 3.2 Direct and Indirect Taxes In the late 1990s, STINMOD was joined with Professor Neil Warren’s STATAX model of indirect taxes. STATAX estimates indirect taxes by applying tax rates to expenditure categories in the HES — and to income from wages and salaries and dividends. When these rates are applied to a HES-based version of STINMOD, the resulting STINMOD–STATAX model can be used to estimate the distributional impact of direct and indirect taxes and government benefits. STINMOD–STATAX model was used to assess the likely distributional impact of the government’s GST tax reform package for the Senate Committee on a New Tax System (Warren et al., 1999). Results from the model were one of the factors leading to the government delivering more generous compensation to social security recipients and reducing the proposed income tax cuts to high income earners. After all of the changes, NATSEM found that the final tax reform package provided the greatest benefits to single income couples with children and sole parents. As Figure 7 indicates, single income families with children were estimated to fare particularly well under the final tax reform package. Dual income families, particularly those without children, and single people on low to average incomes, did less well (Harding et al., 2000a). 3.3 ADMOD: Using Administrative Data As noted earlier, STINMOD generally runs on basefiles developed from survey data as administrative data are not widely available (mainly for privacy reasons). An exception to this is ADMOD, a model developed by NATSEM with FaCS. ADMOD is a microsimulation model of the department’s payments, based on departmental administrative data. The strength of this model lies in the database, in terms of both the population and the range of data items it contains. Sub-populations are captured and complex business rules can be modelled. NATSEM has developed a user interface to enable non-programmers to use the model. When used in conjunction with STINMOD, ADMOD is another important tool in the development of social and economic policy. Because of confidentiality issues, the model can

Rachel Lloyd

326

Figure 7: Percentage Gain in Disposable Income from July 2000 Tax Reform Package for Selected Household Types 12

Percentage gain

10 8 6 4 2

0 5, 00 10 0 ,0 0 15 0 ,0 0 20 0 ,0 0 25 0 ,0 0 30 0 ,0 0 35 0 ,0 0 40 0 ,0 0 45 0 ,0 0 50 0 ,0 00 55 ,0 0 60 0 ,0 0 65 0 ,0 0 70 0 ,0 00 75 ,0 0 80 0 ,0 00 90 ,0 10 0 0 0, 0 1 2 00 5, 0 15 0 0 0, 00 0

0

Private income $ pw Single taxpayer Dual income couple without children

Single income couple with two children

Source: Based on graphs in Harding et al., 2000a

only be used internally within FaCS. However, outcomes from ADMOD have been used to analyse and improve STINMOD outcomes. 3.4 Paid Maternity Leave In 2002, the Human Rights and Equal Opportunity Commission (HREOC) asked NATSEM to estimate the cost to the government of a paid maternity leave scheme (see Lloyd et al., 2002b, for more information). The issue received considerable interest, as the Australian government considered ways in which to help families at a time when the fertility rate continues to fall, childcare places are in short supply and women want to participate in the labour market. NATSEM used STINMOD as the basis of a group model to estimate the costs of the proposed scheme. Using the STINMOD base data, we developed 200 hypothetical family types (based on marital status and incomes of the family) and estimated the number of women in families such as these that would be eligible for paid maternity leave in a year. The STINMOD entitlement modules were then used to estimate each family’s entitlement to government benefits and tax liability under the current system and under the

Use of STINMOD in Australia

327

proposed scheme. Under the proposed scheme, a woman will no longer be eligible for some benefits and, because of her higher income, would have reduced entitlement to other benefits, as well as higher tax liability. The net government costs for the proposed scheme outlays would be reduced by this clawback. NATSEM estimated the gross cost to the government — the total cost of paid maternity leave for all women — to be $460 million in 2003–2004. The net cost of the scheme was estimated to be $213 million in 2003–2004. In other words, $247 million, or over half of the gross cost, would be saved by reductions in other government outlays and increased income tax. Other estimates of the cost of the scheme were not able to include the impact of savings to the government from decreases to other benefits and increased tax. By using STINMOD, NATSEM was in a unique position to provide a reliable estimate of the net costs of the scheme. 3.5 Financial Incentives to Work STINMOD has been used over a number of years to estimate effective tax rates — to analyse their distribution and the impact they have on the financial incentives to work. An Effective Marginal Tax Rate (EMTR) is the proportion of a dollar increase in private income, which is lost to income tax and reductions in Australia’s income-tested social security and family assistance payments. Effective Average Tax Rates (EATRs) are the weighted sum of EMTRs over a range of private income (not just $1). Australia has a tightly targeted system of social security and family payments, with most payments subject to income tests. These income tests can create high effective tax rates and may reduce work incentives for those in receipt of government benefits. The challenge for government is to balance the targeting of transfer payments and the incentives for those receiving these payments to increase the amount that they work. Beer (2003) used STINMOD to examine the distribution of EMTRs across the Australian labour force in 2002. She also looked at how the number of people facing high EMTRs had changed between 1997 and 2002. This period was of particular interest, because the changes to the tax and social security system that accompanied the introduction of the Goods and Services Tax in 2000 were aimed at reducing effective tax rates and boosting work incentives. Toohey and Beer (2004) used STINMOD to examine the financial incentives for mothers to increase their workforce participation by looking at the EATRs for families with different income levels and numbers of children. Once a mother on the minimum wage married to a low-income father began working eight hours a week, the family’s EATRs generally remained above 60 per cent (Figure 8). The main factors contributing to the EATR were the

Rachel Lloyd

328

Figure 8: Effective Average Tax Rates for a Low-Income Father and a Mother Working Increasing Hours. 140 120 100

EATR (%)

80 60 40 20 0 0

5

Couple - 1 dependent child

10

15 20 25 Hours worked by mother per week Couple - 2 dependent children

30

35

Couple - 3 dependent children

Source: Toohey and Beer, 2004 Note: The father is earning $515 per week and the mother is earning $11.70 per hour, with one, two or three children — one child in childcare at $4.30 per hour.

income tests of family payments, income tax, the Medicare levy and the net costs of childcare. The highest EATR of 119.1 per cent occurred when a mother in the family with three children went from 11 to 12 hours of work per week. 3.6 Regional Microsimulation As noted earlier, NATSEM has developed regional microsimulation modelling techniques with a view to building spatial microsimulation models. The techniques blend the population census and sample survey data together to create a synthetic unit record file for every small area — see Melhuish et al., 2002, for more information). The census data has detailed regional information but limited information on incomes and expenditures. On the other hand, the ABS sample surveys such as the HES contain exceptionally detailed expenditure and income data at the individual and household level, but lack any detailed geographic information. The first model to be constructed by NATSEM using these new techniques was the Marketinfo model, which provided detailed regional expenditure and income estimates. The model first recoded the HES and census variables to be comparable, and then reweighted the HES, utilising detailed socio-demographic profiles from the census. This was done for each Collectors District separately, and a reweighted HES unit record file was generated for each district. Initially, the output from this model was used by

Use of STINMOD in Australia Figure 9:

329

Estimated Dollar Tax Cut per Week per Household in 2005–2006, Sydney

$3.50 - $9.50 $9.51 - $13.70 $13.71 - $19.30 $19.31 - $34.10

Source: Chin et al., 2005

private sector clients — to determine where to put new shopping centres, to examine what percentage of total spending in an area is received by their shops; to maximise the efficiency of direct marketing efforts, or to examine the estimated incomes and assets of consumers living within each Collectors District. The modelling techniques have now been extended to address the concerns of public policy makers. For example, we have estimated poverty rates and characteristics of those in poverty by small area (Harding et al., 2000b; Lloyd et al., 2002a). The STINMOD uprating methodology and entitlement algorithms are used to estimate current disposable incomes and poverty rates and characteristics by postal area. By using STINMOD, we are also able to estimate the impact of a policy change on different regions (Harding et al., 2004a). Recent output, for example, has simulated housing affordability at the small area level (Taylor et al., 2004) and the impact of the 2005 Federal Budget tax cuts by region (Chin et al., 2005 and Figure 9). 3.7 Service Delivery STINMOD has also been used as the basis for a small area model of the characteristics and access channel usage of Centrelink clients, both now and in five years time. Centrelink is the agency that has responsibility for the

330

Rachel Lloyd

delivery of a range of Australian government services including income support payments and associated services. The model assists Centrelink with its property management strategies, as well as providing forecasts of the likely demand for each of the various methods of accessing Centrelink services. The purpose of the Customer Service Projection (CuSP) Model, is to provide a tool that will assist decision makers through short- to mediumterm projection of customers and channel use demands at the small area level and under alternative scenarios of customer numbers, customer characteristics, access preferences and opportunities (see King et al., 2002, for more information, and the chapter by King, 2006, in the accompanying volume to this one). The core of the CuSP Model is a combination of a standard static microsimulation model — STINMOD, regional microsimulation techniques and benchmarking to administrative data. Application of small area weights to a base population file derived using STINMOD gives the estimate of the numbers and characteristics of Centrelink customers in each small area. Application of channel use propensities to the customer estimates then gives the estimate of channel use. The model has a ‘what-if’ capability allowing various scenarios to be modelled. 3.8 Other Applications NATSEM has regularly used STINMOD to estimate the distributional impact of various increases in award wages for the Department of Employment and Workplace Relations. The results were used in the Commonwealth Government’s submissions to the Industrial Relations Commission Safety Net Reviews in 2000, 2001 and 2002 (DEWR, 2002). STINMOD has also been used as the basis for the model of the Pharmaceutical Benefits Scheme built for Medicines Australia (Harding et al., 2004b; Brown et al., 2004) and in undertaking fiscal incidence studies which examine the distributional impact of taxes and both cash and non-cash benefits (Harding et al., 2002, 2006).

4. Conclusion STINMOD provides reliable estimates of the first-round distributional impact of changes in tax and transfer policy. For example, when the government is considering changing the income tax schedules or liberalising the age pension income test, it can use STINMOD to provide an estimate of how much the proposed changes would cost and what types of families would be affected. STINMOD provides estimates of the number of winners and

Use of STINMOD in Australia

331

losers, the net impact on government outlays or taxation revenue and the magnitude of the gains or losses for families of different composition and at different income levels. The STINMOD model has now been used for more than five years by federal government departments — such as FaCS and The Treasury — to look at the impact of policy change. A user-friendly interface provides a convenient method for quantifying the immediate distributional and revenue implications of changes to payment and taper rates in a wide range of government programmes. Users of the user-friendly version of the model do not require knowledge of SAS. NATSEM continues to enhance and develop STINMOD in response to requests from client departments and users, to meet internal needs and to improve the quality of the model. In response to requests from federal government departments who needed to provide ‘outyear’ costings as part of the budget process, NATSEM developed an outyears version of STINMOD, with users able to choose the financial year they wish to analyse for up to five years ahead. As part of the development of an outyears model, we have developed a spreadsheet interface that allows users to create basefiles and parameter files based on their own assumptions. STINMOD has been used as the basis for, or as an input to other models including the Returns to Education Model, STINMOD–STATAX, Centrelink’s CuSP Model and ADMOD. In addition, it has been used for a wide range of policy input — including analysing the major changes to the tax and transfer system introduced in Australia in July 2000 and costing a paid maternity leave scheme. STINMOD has also been used to research effective tax rates. The development of regional modelling techniques means that the next goal is a regional version of STINMOD that allows policy makers to assess the impact of a policy change spatially as well as distributionally and fiscally.

Acknowledgement The author would like to acknowledge the work of other NATSEM staff, some of which she presents in this paper, and the comments provided by Ann Harding and Matthew Toohey on a draft of this paper. This chapter was completed while the author was employed at NATSEM.

References Beer, G. (2003). Work Incentives under a New Tax System: The Distribution of Effective Marginal Tax Rates in 2002. Economic Record, 79, S14–S25. Bremner, K., Beer, G., Lloyd, R. and Lambert, S. (2002). Creating a Basefile for STINMOD. Technical Paper no. 27, National Centre for Social and Economic Modelling, University of Canberra, Canberra.

332

Rachel Lloyd

Brown, L., Abello, A., Phillips, B. and Harding, A. (2004). Moving Towards an Improved Micro-Simulation Model of the Australian Pharmaceutical Benefits Scheme. Australian Economic Review, 37(1), 41–61. Chin, S.F., Harding, A., Lloyd, R., McNamara, J., Phillips, B. and Vu, Q. (2005). Spatial Microsimulation Using Synthetic Small Area Estimates of Income, Tax and Social Security Benefits. Australasian Journal of Regional Studies, 11(3), 303–335. Chin, S.F. and Harding, A. (2006). Regional Dimensions: Creating Synthetic Smallarea Microdata and Spatial Microsimulation Models. Technical Paper no. 33, National Centre for Social and Economic Modelling, University of Canberra, Canberra. Deville, J. and Sarndal, C. (1992). Calibration Estimators in Survey Sampling. Journal of the American Statistical Association, 87, 376–382. DEWR (Department of Employment and Workplace Relations). (2002). Safety Net Review — Wages 2001– 2002: Commonwealth Submission. DEWR, Canberra. Gregory, R.G. and Hunter, B. (1995). The Macroeconomy and the Growth of Ghettos and Urban Poverty in Australia. Centre for Economic Policy Research Discussion Paper no. 325, Canberra. Harding, A., Warren, N., Robinson, M. and Lambert, S. (2000a). The Distributional Impact of the Year 2000 Tax Reforms in Australia. Agenda, 7(1), 17–31. Harding, A., Lloyd, R., Hellwig, O. and Bailey, G. (2000b). Building the Profile: Report of the Population Research Phase of the ACT Poverty Project. Poverty Task Group Paper no. 3, Canberra. Harding, A., Lloyd, R. and Greenwell, H. (2001). Financial Disadvantage in Australia 1990 to 2000: The Persistence of Poverty in a Decade of Growth. The Smith Family, Camperdown, NSW (available from www.smithfamily.org.au). Harding, A., Warren, N., Beer, G., Phillips, B. and Osei, K. (2002). The Distributional Impact of Selected Commonwealth Outlays and Taxes and Alternative Commonwealth Grant Allocation Mechanisms. Australian Economic Review, 35(3), 325–334. Harding, A., Lloyd, R., Bill, A. and King, A. (2004a). Assessing Poverty and Inequality at a Detailed Regional Level — New Advances in Microsimulation. Research Paper no. 2004/26 of the United Nations University World Institute for Development Economics Research (WIDER), Helsinki. Harding, A., Abello, A., Brown, L. and Phillips, B. (2004b). The Distributional Impact of Government Outlays on the Australian Pharmaceutical Benefits Scheme in 2001–02. Economic Record, 80 (Special Issue), S83–S96. Harding, A., Lloyd, R. and Warren, N. (2006). The Distribution of Taxes and Government Benefits in Australia, in Papadimitriou, D. (ed), The Distributional Effects of Government Spending and Taxation, Palgrave, New York. King, A., McLellan, J. and Lloyd, R. (2002). Regional Microsimulation for Improved Service Delivery in Australia: Centrelink’s CuSP Model. Paper prepared for the 27th General Conference, International Association for Research in Income and Wealth, Djurhamn, Sweden, 18–24 August. King, A. (2006). Providing Income Support Services to a Changing Aged Population in Australia: Centrelink’s Regional Microsimulation Model, in Gupta, A. and Harding, A. (eds), Modelling Our Future: Population Ageing, Health and Aged

Use of STINMOD in Australia

333

Care, International Symposia in Economic Theory and Econometrics, NorthHolland, Amsterdam. Johnson, P., Beer, G. and Lloyd, R. (2002). Does Higher Education Pay? Results from the Returns to Education Model. Paper Prepared for the Department of Education, Science and Training Conference on Education and the Labour Market, Canberra, Australia, 20–21 June. Lambert, S., Percival, R., Schofield, D. and Paul, S. (1994). An Introduction to STINMOD: A Static Microsimulation Model. STINMOD Technical Paper no. 1, National Centre for Social and Economic Modelling, University of Canberra, Canberra. Lloyd, R., Harding, A. and Hellwig, O. (2001). Regional Divide? A Study of Incomes in Regional Australia. Australasian Journal of Regional Studies, 6(3), 271–292. Lloyd, R., Harding, A. and Greenwell, H. (2002a). Worlds Apart: Postcodes with the Highest and Lowest Poverty Rates in Australia, in Eardley, T. and Bradbury, B. (eds), Competing Visions: Refereed Proceedings of the National Social Policy Conference 2001, SPRC Report 1/02, Social Policy Research Centre, University of New South Wales, Sydney, pp. 279–297. Lloyd, R., Phillips, B., Beer, G. and Harding, A. (2002b). Appendix: Costing a Paid Maternity Leave Scheme, in HREOC (Human Rights and Equal Opportunity Commission), A Time to Value: Proposal for a National Paid Maternity Leave Scheme, HREOC, Sydney. [http://www.hreoc.gov.au/sex_discrimination/pml2/ index.html, accessed 1 April 2003] Melhuish, A., Blake, M. and Day, S. (2002). An Evaluation of Synthetic Household Populations for Census Collection Districts Created Using Spatial Microsimulation Techniques. Australasian Journal of Regional Studies, 8(3), 369–387. Percival, R. and Lim, P. (1999). Simulating Australia’s Institutionalised Population. Technical Paper no. 17, National Centre for Social and Economic Modelling, University of Canberra, Canberra. Taylor, E., Harding, A., Lloyd, R. and Blake, M. (2004). Housing Unaffordability at the Statistical Local Area Level: New Estimates Using Spatial Microsimulation. Australasian Journal of Regional Studies, 10(3), 279–300. Toohey, M. and Beer, G. (2004). Financial Incentives for Working Mothers under A New Tax System. Australian Journal of Labour Economics, 7(1), 53–69. Vinson, T. (1999). Unequal in Life: The Distribution of Social Disadvantage in Victoria and NSW, The Ignatius Centre, Melbourne. Warren, N., Harding, A., Robinson, M., Lambert, S. and Beer, G. (1999). Distributional Impact of Possible Tax Reform Packages. Report prepared for the Senate Select Committee for a New Tax System, 1 April, Canberra.

This page intentionally left blank

Chapter 13

The Impact of Canadian Population Ageing on Federal Personal Income Tax: Microsimulation Results from 2000 to 2026 Weng-Fong Lu, Wei Li and Earl Bailey Statistics Division, Corporate Strategies and Business Development Branch, Canada Revenue Agency (CRA)

Abstract It is envisioned that Canadian population ageing will have implications for both the government’s expenditure and revenue sides. This chapter focuses on the revenue implications. It profiles the ageing of the Canadian population and its effect on the patterns of personal income and taxes at the federal level. The research is based on the Statistics Canada medium population growth projection over the period 2000–2026 and the income tax microsimulation model developed by the Canada Revenue Agency (CRA). Making reasonable assumptions about likely demographic and income change, the results suggest that income tax revenue will not actually decline in the future but that the growth rate will slow considerably after 2011.

1. Introduction In recent decades, the proportion of senior1 people in many countries has increased considerably compared to the total population. Canada is among this group of countries. According to Statistics Canada, the proportion of seniors in the year 2000 is more than 12 per cent and it will continue to rise to the middle of this century. Population ageing is expected to have an effect on the labour force, federal taxation, federal health care expenditures and 1

Throughout this paper, the terms senior and elderly pertain to the population of 65 or more years of age. International Symposia in Economic Theory and Econometrics, Vol. 15 Copyright r 2007 Elsevier B.V. All rights reserved ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15013-9

336

Weng-Fong Lu et al.

other socio-economic characteristics — and will also place financial pressure on Canada’s social system. Since research on the effect of population ageing on government expenditure has already been undertaken by Health Canada, Human Resources Development Canada and Finance Canada, this paper will focus on the government revenue side. The major impact of population ageing on taxation is expected to be on the revenue from personal income tax, which accounts for about one-half of total federal tax revenue. Personal income tax is directly associated with people — and could be significantly affected by factors associated with the socio-economic characteristics of people. Since the retired population usually earns less income than employed people, average taxable income might be expected to decrease with population ageing. However, total income tax revenue is affected not only by population ageing but also by other factors — the number of taxpayers, their income level and the effective income tax rate. Based on the population census and data from personal income tax returns — and by using the T1 Tax Analysis Model — this chapter examines the impact of Canadian population ageing on federal income tax revenues. In Section 2, the literature on population ageing and taxation is reviewed. Section 3 provides a profile of the Canadian ageing population. Section 4 illustrates income and tax by age group. Section 5 introduces the CRA’s income tax microsimulation model and parameters used in the model. Section 6 presents the results of our microsimulation model runs. Finally, Section 7 provides a brief conclusion of the empirical analysis.

2. Literature Review In recent years, there has been significant research into the potential impact of population ageing on industrialized countries’ socio-economic development (Corak, 1998; Denton and Spencer, 1999; Health Canada, 2002; Merette, 2002). However, relatively little research has been done into the taxation implications of population ageing. A U.S. study suggested that population ageing would cause slow growth in the number of people working and paying taxes but rapid growth in the cost of health and social security programmes (United States Government Printing Office, 1997). Robson (2003) predicated that Canadian taxpayers in the future would pay more for the entire package of public programmes than their predecessors (providing the current age/ sex distribution of the public expenditure in these programmes remained the same). A study commissioned by the Group of Ten (1998) countries concluded that government revenues would be adversely affected as the baby boom generation moved from its high-income-generating years to retirement — and that countries whose revenues depended heavily on income or payroll taxes would face deterioration in revenues.

Impact of Canadian Population Ageing

337

An Australian study indicated that even though population ageing would create downward pressure on total personal income tax revenue (because of the declines in the average tax paid per person in the older age groups), several other factors such as GDP growth, labour participation rate, wage rates, etc. made it uncertain which direction total personal income tax revenue would take (Department of the Treasury of Australia, 2002). Some studies that were done by Finance Canada on the fiscal implications of population ageing show that a less than severe impact is expected to occur over the next half century. King and Jackson (2000) stated that, although ageing will have some impact on public finances, it will play a minor role in the interaction of all the factors that are expected to cause fiscal pressures. Jackson and Matier (2002) analysed the long-term impact of population ageing on important revenue and expenditure categories. Under their definition of existing federal, provincial and territorial fiscal structures (all assumed to remain constant) and their criterion for long-term fiscal sustainability, their projection suggested that most governments will be in a fiscally sustainable position over the long run. While the existing literature thus gives us both pessimistic and optimistic expectations for the future change in government revenue, it does not give specific figures on the change in income tax caused by population ageing. In contrast, by using the CRA’s microsimulation model, this paper will provide detailed quantitative information on the impact of future population ageing on personal income tax.

3. Canadian Population Ageing Profile 3.1 Population Projection In 2001, Statistics Canada published its population projections for 2000–2026. The projections use the 2000 preliminary population estimates that were based on the 1996 Census data. For the projections of the overall Canadian population, there are three growth scenarios — high, medium and low growth. Table 1 summarizes the assumptions that were made in the scenarios. These assumptions reflect the following components of population growth: the total fertility rate (TFR), life expectancy at birth, immigration level, emigration rates and the number of non-permanent residents. The TFR in Canada was 1.49 in 2000. According to Statistics Canada (2001), the TFR in the medium assumption, which is the average of the low and high assumptions, will be 1.48 by 2001 and will remain constant thereafter. Since the fertility level has been quite stable in recent years, the continuation of the current level appears reasonable. The medium life expectancy at birth is based on current trends in the age-specific mortality

Weng-Fong Lu et al.

338 Table 1:

Component Assumptions for Population Projections 2000–2026

Component of population change

High growth

Medium growth

Low growth

Total fertility rate Age expectancy at birth (male/female) Immigration (persons) Total emigration

1.8 81.5/85.0

1.48 80.0/84.0

1.3 78.5/83.0

Non-permanent residents (persons)

270,000 225,000 180,000 This is based on the 2-year average of the age–sex specific emigration rate from 1997–1998 to 1998–1999 240,000, assumed constant over the projection period

Source: Statistics Canada (2001)

Figure 1:

Canadian Population Size by Age Group, 2000–2026

40,000

Population (000's)

35,000 30,000 25,000 Aged 65+

20,000

Aged 15-64

15,000

Aged 0-14

10,000 5,000 0 2000

2006

2011

2016

2021

2026

Year

Source: Based on Statistics Canada, 2001

rate — and it is reasonable to assume that the rate of mortality improvement achieved recently will continue through the project period. The medium immigration assumption is based on the current government’s target of 225,000. The government has no plan to change the quota in the short term. To set future migration equal to the current level is the most popular method in international migration projection. Based on the above arguments, Statistics Canada’s medium-growth quinquennial projections have been chosen for this study. Figure 1 shows the population estimates for 2000 and the population projections from 2006 to 2026. The Canadian population will increase over the next 20–30 years, with the total population expected to reach

Impact of Canadian Population Ageing Figure 2:

339

Canadian Population Age Structure, 2000–2026

80 70 60

%

50 0-14

40 30

15-64

20

65 and Over

10 0 2000

2006

2011

2016

2021

2026

Year

Source: Based on Statistics Canada, 2001

33.4 million in 2011 (an 8 per cent increase over year 2000) and 36.2 million in 2026 (an 18 per cent increase over year 2000). Also, the distribution of population among the age groups will vary. During the 2000–2026 period, the population of the age groups under 15 will decrease slightly, from 5.9 to 5.4 million. The population aged 15–64 will increase moderately from 21.0 to 23.5 million by 2016, and then decline slightly to 23.1 million by 2026. The number of persons aged 65 and over will be 4.8 million by 2011 and 7.8 million by 2026, or 1.3 and 2.0 times, respectively compared to the 3.9 million of year 2000. 3.2 Population Ageing The Canadian population will age more rapidly after 2011, when the ‘‘baby boomers’’ (1946–1960) reach their retired age. The senior population, which was 12.6 per cent of the total population in 2000, is expected to rise to 14.5 per cent in 2011 and 21.4 per cent in 2026 (see Figure 2). There will be one senior in every five people in year 2026. The aged dependency ratio, which is defined as the ratio of the number of persons 65 years and over to the 15–64 population, will increase from 0.18 in 2000 to 0.21 in 2011 and to 0.34 in 2026 (see Figure 3).2 The child dependency ratio (the number of persons 2 Although the aged dependency rate could continue increasing beyond 2026, the discussion of ageing implications in this paper stops at the year 2026 because of the constraint on availability of population age specific forecast data after 2026 from Statistics Canada.

Weng-Fong Lu et al.

340 Figure 3:

Canadian Population Dependency Ratios, 2000–2026

0.6 0.5 Child Dependancy Ratio Aged Dependancy Ratio Total Dependancy Ratio

Ratio

0.4 0.3 0.2 0.1 0 2000

2006

2011

2016

2021

2026

Year

Source: Based on Statistics Canada, 2001

under 15 years to the 15–64 population) will decrease from 0.28 to 0.23 during the same period. The total dependency ratio (the number of persons under 15 and 65 years and over to the 15–64 age group population) will initially decrease from 0.46 to 0.44 by year 2011 because of the decreasing proportion under 15 and then increase to 0.57 by year 2026 (which is an increase of 0.11 over 2000). By 2026, it is implied that, on average, for every working age person there will be 0.23 children and 0.34 seniors. According to the United Nations’ estimates shown in Table 2, 14.4 per cent of the Canadian population will be aged 65 or over in 2010, rising to 21.3 per cent in 2025. The corresponding values are 12.8 per cent and 17.8 per cent, respectively in the U.S. and 6 and 9.5 per cent, respectively in Mexico. Although the total dependency ratio in Canada by year 2025 will be lower than the U.S. (0.56 versus 0.60), the aged dependency ratio, as a consequence of Canada’s higher proportion of elderly people, will be higher in Canada (0.33) than in the United States (0.29) and, most noticeably, higher than in Mexico (0.14).3 Compared to our North American neighbours, the Canadian population will be much older. The median age of Canadians will be 40.6 in 2010 and 43.9 in 2025, which is 4.3 and 6.3 years older than the respective American

3 Since the population under 15 would pay little or no income tax and the majority of the ageing population would still pay their income tax, the aged dependency ratio is more relevant to this research than the total dependency ratio from a revenue standpoint.

Impact of Canadian Population Ageing Table 2: Indicator

341

Major Population Indicators for North American Countries, 2000–2025 2000

2005

2010

2015

2020

2025

Canada Population (thousands) 30,769 31,972 33,069 34,133 35,166 36,128 Percentage aged 0–14 (%) 19.0 17.3 15.6 14.8 14.7 14.8 Percentage aged 15–64 (%) 68.4 69.5 70.0 68.8 66.6 63.9 Percentage aged 65+ (%) 12.6 13.2 14.4 16.4 18.7 21.3 Aged dependency ratio 0.18 0.19 0.21 0.24 0.28 0.33 Median age (years) 36.9 38.9 40.6 41.9 43.0 43.9 U.S. Population (thousands) 285,003 300,038 314,921 329,669 344,270 358,030 Percentage aged 0–14 (%) 21.9 21.3 20.5 20.3 20.1 19.8 Percentage aged 15–64 (%) 65.8 66.4 66.7 65.5 64.0 62.4 Percentage aged 65+ (%) 12.3 12.3 12.8 14.2 15.9 17.8 Aged dependency ratio 0.19 0.19 0.19 0.22 0.25 0.29 Median age (years) 35.2 35.9 36.3 36.6 37.0 37.6 Mexico Population (thousands) 98,933 106,385 113,320 119,618 125,176 129,866 Percentage aged 0–14 (%) 33.8 31.3 28.7 26.4 24.3 22.5 Percentage aged 15–64 (%) 61.4 63.4 65.3 66.8 67.7 68.0 Percentage aged 65+ (%) 4.8 5.3 6.0 6.8 8.0 9.5 Aged dependency ratio 0.08 0.08 0.09 0.10 0.12 0.14 Median age (years) 22.9 24.7 26.6 28.5 30.6 32.7 Source: Population Division of the Department of Economic and Social Affairs of the United Nations Secretariat (2003)

median ages and 14.0 and 11.2 years older than the respective Mexican median ages. In Canada, the population will continue to age at a rapid rate into the future. The elderly proportion of the Canadian population is projected to double from 10 to 20 per cent over the 40-year period from 1984 to 2024. In the U.S., the proportion of the elderly population is expected to double over a longer period of about 60 years.4 It is worth noting that Fougere and Merette (1998) suggested that the aged dependency ratio in some OECD countries (Japan, Italy, Sweden, United Kingdom and France) would be greater than that of Canada. This shows that although Canada’s population is ageing faster than U.S. and Mexico, it will be better off than Japan and many industrialized European countries.

4

Derived from the United Nations’ publication at http://www.esa.un.org/unpp.

342

Weng-Fong Lu et al.

4. Incomes and Taxes by Age 4.1 Proportion of Taxable Returns The major indicators of personal income and tax — the proportion of taxable returns, distribution of total income, average taxable income, average non-refundable tax credits and average net federal income tax — are all related to age. Based on the most recent available data (year 2000), the proportion of taxable returns will be analysed in relation to age first. Since the Canadian personal income tax system is closely associated with the administration of means-tested social development programmes, a large proportion (90 per cent of the total population aged 20 and over) submitted their tax returns for the year 2000. The percentage was higher in the older age groups than in the younger age groups. For example, about 96 per cent of the 65 and over age groups filed a tax return, compared with about 90 per cent in the 40–64 groups (a result that may be due to a greater incidence of means tested benefits within the senior population). However, the higher percentage in older age groups does not necessarily equate to more taxes, because personal income tax is mainly determined by one’s taxable income level. Furthermore, not all tax returns are taxable.5 Figure 4 shows that the proportion of the number of taxable returns to that of total returns varies in different age groups. For example, persons aged under 20, who have little or no tax to pay because they have relatively less income, show only 22.7 per cent of their total tax returns to be taxable. The proportion increases with age and reaches a maximum of 81.3 per cent for persons aged 45–49. Then, the higher age groups show lower percentages. Persons that were aged 65 and over had an average of 59.5 per cent of their total tax returns being taxable. 4.2 Income Sources Seniors usually have less income than workers with differences in their sources of income. Figure 5 shows the sources of income6 for both seniors and non-seniors in year 2000. For the non-seniors, 78 per cent of their total income came from employment, 9 per cent came from other sources (e.g., EI benefits, dividends, investment, rents, annuities, capital gains and RRSP), 5 A taxable return is defined as a return in which the net combined federal and provincial tax payable is at least one dollar. 6 The reference is to the total income assessed in line 150 of the return. For more information on the total income and income items, see Canada Customs and Revenue Agency (2002).

Impact of Canadian Population Ageing Figure 4: 2000

343

Taxable Returns by Age Group (as a Share of Total Returns), Tax Year

90 80 70 60

%

50 40 30 20 10 0 Under 20-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 70-74 20

Age Group

75 and over

Source: Based on Canada Customs and Revenue Agency (2002)

6 per cent was from self-employment and only 3 per cent came from pensions. Tax-exempt income — including workers’ compensation payments, social assistance payments and net federal supplements — made up 2 per cent of the total income. The remaining portion of income, referred to as other income, accounted for another 2 per cent. The sources of income for seniors are quite different from those for nonseniors (see Figure 5). Pension income was the largest portion, accounting for 58 per cent of seniors’ income. Income from other sources was 25 per cent, while employment income was only 7 per cent. Tax-exempt income accounted for 6 per cent. Self-employment income and other income represented 2 per cent each. The proportions for pension income, income from other sources and tax-exempt income were significantly more for seniors than for the non-seniors. 4.3 Average Taxable Income Average taxable income is defined as the total taxable income assessed,7 divided by the total number of tax filers who reported a taxable income. Figure 6 illustrates average taxable income by age group. For persons aged 7

Based on line 260 of the year 2000 tax return, and this is the amount on which we calculate income tax.

Weng-Fong Lu et al.

344 Figure 5:

Sources of Income, Tax Year 2000

Non-seniors

seniors

2% 6% 2%

6% 2% 2%

9%

7%

3% 25%

58%

78%

Employment

Self-employment

Pension

Tax-exempt

Other Sources

Other Income

Source: Based on Canada Customs and Revenue Agency (2002)

under 20, average taxable income was $5,950 per person. The average increased to a maximum of $39,146 for persons aged between 50 and 54, then decreased to the $22,000–$26,000 range for persons aged 65 and over, with the average being about 65 per cent of the highest taxable income. Senior people had a lower average taxable income than working people aged 30–64, but were better off than other younger age groups. There is no big difference in average taxable income level in the three senior age groups, since the pensions are not related to seniors’ age.

4.4 Average Non-Refundable Tax Credits Non-refundable tax credits usually have predetermined common values for all Canadians, regardless of their income levels. These credits reduce their federal income tax payable. However, the excess over tax payable is not refunded. The total tax credits are 17 per cent of the total credit amounts including basic personal amount, age amount, spousal amount, contributions to the Canada and Quebec Pension Plans, employment insurance premiums, etc. It must be emphasized that senior tax filers whose income is less than $49,824 may be allowed to claim an age amount up to the maximum of $3,531. The total non-refundable tax credits are the sum of total tax credits and tax credits on donations.

Impact of Canadian Population Ageing Figure 6:

345

Average Taxable Income by Age Group, Tax Year 2000

45,000 40,000 35,000 30,000

$

25,000 20,000 15,000 10,000 5,000 0 Under 20-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 70-74 20 Age Group

75 and over

Source: Based on Canada Customs and Revenue Agency (2002)

As shown in Figure 7, the average non-refundable tax credits8 for the age groups between 20 and 64 were about $1,700 per person. Note that the nonsenior age groups do not show large differences, because the basic personal amount is the major component of non-refundable tax credits in which the same amount is applied. However, the amount jumps to about $2,100 for the 65–69 age group, an increase of about $400 per person. This jump is a result of seniors using the age amount. Average non-refundable tax credits for seniors in different age groups showed little variation, ranging from $2,127 to $2,280. 4.5 Average Net Federal Tax Federal income tax rates for the tax year 2000 were 17 per cent for taxable income of $30,004 or less, an additional 25 per cent on income under $60,009 in excess of $30,004 and an additional 29 per cent on income over $60,009. The average net federal tax for the under 20 age group was just $881 per taxable return.9 It increased rapidly with increasing age, reaching a maximum of $7,843 per taxable return in the 50–54 age group (see Figure 8). The 65–69 age group showed an average of $4,976 per return and this remained 8

Average non-refundable tax credit per return refers to the credit claims reported on line 350 divided by the number of returns with a non-zero claim. 9 Based on line 420 of the year 2000 tax return. The average of net federal tax is equal to the total amount of net federal tax divided by the number of taxable returns.

Weng-Fong Lu et al.

346 Figure 7:

Average Non-refundable Tax Credits by Age Group, Tax Year 2000

2,500 2,000

$

1,500 1,000 500 0 Under 20-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 70-74 20

Age Group

75 and over

Source: Based on Canada Customs and Revenue Agency (2002)

Figure 8:

Average Net Federal Tax by Age and Sex, Tax Year 2000

12,000 Male

Female

Total

10,000

$

8,000

6,000

4,000

2,000

0

Under 20-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 70-74 75 and 20 over

Age Group

Source: Based on Canada Customs and Revenue Agency (2002)

stable for the older age groups. The taxes paid by the seniors were only 63 per cent of the highest tax payable of the age group 50–54. The difference of average net federal tax between male and female was quite noticeable, with the highest net tax payable for males being $10,325 (at 50–54 age group) compared with only about $5,000 for females (at 45–49 age group).

Impact of Canadian Population Ageing

347

Table 3 puts the observations about income and personal income tax by age into perspective. The average net federal tax for seniors (about $4,800) was considerably less than that of persons aged under 65 (about $6,100). This observation is interpreted as seniors having, on average, higher nonrefundable tax credits and lower taxable income that is mainly sourced as pension income. As the senior population grows more quickly than the rest of the population, it is expected that total individual income tax will be impacted. However, whether population ageing will reduce total federal income tax revenue depends on not only the increase in the senior population but also the change in the population under 30 (whose average income tax was less than seniors’) and the change of population aged 30–64 (whose average income tax was more than seniors’).

5. Income Tax Microsimulation Model 5.1 Microsimulation Model Microsimulation models are computer-based models that operate at the microlevel, such as a person, family or firm. Such models simulate how a socio-economic programme could operate under proposed changes and how participants would be affected, based on a large representative micro database of individual records and variables. The difference between micro and macrosimulation models is the explanatory variables. In the former case, they represent individual characteristics but, in the latter, they represent collective properties. From the results of microsimulation, conclusions that apply to macro levels of aggregation such as an entire country can also be drawn. The model applies the socio-economic programme rules to individual records, selects eligible records and computes the simulated results. Each record could represent an individual person or a group of individuals, depending on the ratio of the total population to the sample size. Microsimulation models can be static or dynamic. The static model operates on a cross-sectional database at a time point and it typically simulates the direct effects of policy changes. It can also simulate behavioural responses to programme changes and can be used to produce forecasts. The dynamic model generates a longitudinal database by applying transition probabilities to individual records and then uses the micro longitudinal data to simulate changes under proposed policy scenarios. In this way, the effects of demographic and economic processes as well as proposed policy changes can be traced.10 10

For more information on microsimulation modelling, see Citro and Hanushek (1991), Harding (1996) and Anderson (2001).

348

Table 3: Age group

Proportion of taxable returns

Average incomea

Employment income

Pension income

Average taxable income

Average net federal tax

($)

Average nonrefundable tax credits ($)

(%)

($)

(%)

(%)

54.4 78.3 72.1 59.5

16,748 38,234 32,632 27,406

87.8 76.7 78.1 6.9

0.2 4.0 3.5 58.5

15,825 34,946 29,926 24,629

1,532 1,738 1,684 2,210

2,793 6,862 6,060 4,809

69.9

31,714

67.3

11.8

28,969

1,776

5,872

($)

Source: Based on Canada Customs and Revenue Agency (2002) a Average income per return refers to the aggregate of all income reported on line 150 of the year 2000 tax return divided by the number of returns with a nonzero income.

Weng-Fong Lu et al.

Under 30 30–64 Under 65 65 and over All

Selected Individual Income Tax Statistics by Age Group, Tax Year 2000

Impact of Canadian Population Ageing

349

5.2 The T1 Tax Analysis Model The T1 Tax Analysis Model11 has been designed to simulate the assessment of taxes for individuals who file T1 income tax returns. The model can define what components of income to include in the calculation of total income, what exemptions and deductions to include in the determination of both net and taxable income and what items to include in the assessment of taxes and in the calculation of tax credits. It is a static microsimulation model, which does not self-adjust its inputs by using the model-simulated outputs with respect to various economic, demographic and other socio-economic variables. However, by changing population weights, it can project income tax revenues forward or backward. The main quantitative definitions in the model that relate to this study include: 











Individual total income ¼ sum of individual income items (such as employment income, pension, net business income, etc.). Individual taxable income ¼ individual total incomeindividual total deductions (such as RRSP deduction, union and professional dues, child care expenses, etc.). Individual non-refundable tax credits ¼ sum of different individual tax credit amounts (such as basic personal amount, age amount, CPP or QPP contributions, etc.)  17%. Individual net federal tax payable ¼ individual taxable income  Federal tax rateIndividual non-refundable tax creditsother credits. Sampled total net federal tax payable ¼ sum of individual net federal tax payable (about a half million population in T1 sample dataset). Estimates of total net federal tax payable ¼ sum of (population weights by age group and sex  sampled total net federal tax payable by age group and sex).

The tax model itself consists of three components — the T1 Tax Analysis Model Interface (TMI) software, the manipulation system (MA) and the base file. TMI menu and dialog screens contain the elements that make up a T1 tax return and are used to organize the tax model study parameters, such as the inclusion or exclusion of a subpopulation of tax filers, deciding upon which items to create or modify, and describing how the calculations are to be done. The MA system is compiled and stored as load modules on a

11

The model is designed and run by the staff of the Statistics Division, CRA.

350

Weng-Fong Lu et al.

mainframe computer. It reads in the parameters from the file created by the TMI and applies the specified study conditions. The base file consists of all records from the T1 Statistical Sample file12 for a particular tax year, plus a selection of fields from the Child Tax Benefit data file, spousal data, T1 Assessing Master file, historical data and computed tax model fields. The model study run output is a matrix file that includes the accumulated counts and amounts of selected variables.13 5.3 Implicit Assumptions in the Model Study Runs The major objective of this study is to identify the impact of population ageing on federal income tax in the future. However, the relationship between the future population and income tax is not linear and cannot be identified precisely by a simple linear regression model. Considering the huge diversity of the tax filer population, the T1 Tax Analysis Model (2000 tax year) will be used in this study. The base file in this model consists of personal income tax data for 2000. Personal income tax in Canada is mainly based on individuals’ income and deductions and the tax rate applies to individual income instead of household income. The childcare deduction and spousal credit information are in the personal income tax form and their amounts are relatively small. We assume the tax relationship between individual and household and family structure in the future will be the same as that in 2000. Tax parameters such as tax rates, income tax brackets, capital gains inclusion rate, dividend gross-up rate, tax credits, etc. are set according to 2000 T1 tax system. To simplify the quantitative analysis and emphasize the effect of population ageing, we assume that the tax parameters will remain constant over the period of 2001–2026. The tax projections of the model were not adjusted for the federal Five-Year Tax Reduction Plan announced in budget year 2000 (Department of Finance Canada, 2000). Furthermore, all of the growth in income and tax comes through real value growth at constant year 2000 dollars. Table 4 gives the population parameters for the T1 model. The parameters are age–sex specific population ratios for future years over the base year 2000 (base year ¼ 100). By using these population parameters (or weights) and other tax parameters in year 2000 tax system, the T1 model can project forward. 12

A sample of 503,180 returns was selected to form T1 sample data for 2000 tax year and represents a total of 22,237,000 returns that were filed in 2001. 13 For more information on the model, see ‘‘T1 Tax Analysis Model Overview’’ and ‘‘T1 Tax Analysis Model Interface (TMI)-User Document’’ by the Personal Taxation Modelling Section, Statistics Division, CRA.

Impact of Canadian Population Ageing Table 4: Age

o20 2024 2529 3034 3539 4044 4549 5054 5559 6064 6569 7074 75+

351

Population Parameters for T1 Model, 2000–2026 2000

2006

2011

2016

2021

2026

M

F

M

F

M

F

M

F

M

F

M

F

100 100 100 100 100 100 100 100 100 100 100 100 100

100 100 100 100 100 100 100 100 100 100 100 100 100

97 104 105 97 86 103 115 115 133 125 107 106 120

97 104 104 96 86 102 114 116 133 126 108 101 118

94 108 108 101 85 91 117 129 149 160 131 114 134

93 108 107 100 84 90 114 129 149 160 131 109 127

91 109 111 104 88 90 104 131 167 179 168 141 149

90 109 111 103 88 88 102 129 166 179 167 133 136

91 98 112 107 91 93 102 117 170 202 189 181 175

91 98 112 106 90 91 100 115 166 199 187 169 153

92 94 102 108 93 96 106 115 152 205 213 205 219

92 95 101 107 92 94 103 113 148 199 208 190 183

Source: Calculated from the medium-growth projection by Statistics Canada (2001) Abbreviations: M, male and F, Female.

Personal income level almost certainly has an effect on income tax obligations. Figure 9 shows the trend of income items14 in T1 returns in the most recent 10 years (at year 2000 constant dollar). The values in the figure represent the average of the item, based on the returns that reported the item.15 The major average annual income increases during the period of 1991–2000 are16: $349 for employment income, $673 for commissions, $310 for other pensions or superannuation, $687 for net professional income, $649 for net commission income and $409 for net fishing income. Other income items had a relatively small per capita amount or low increase rate and we assume that they will remain constant at the year 2000 level into the future. Productivity is a major factor in causing income change and affecting income tax payable (Department of the Treasury of Australia, 2002). From 1991 to 2000, the average annual growth rate in labour productivity was 1.86 per cent (Statistics Canada, 2002). The Conference Board of Canada (2003) has forecast that Canada’s economy will grow at 2.6 per cent on an annual average basis between 2001

14

Total income in T1 tax returns includes 20 income items or categories such as employment income, commission, OAS, CPP, EI, RRSP, etc. 15 Average in this case means the total amount of a specified type of income in a given year divided by the total number of T1 returns reporting that specific type of income. 16 We use the slopes of regression lines to indicate the increases.

Weng-Fong Lu et al.

352 Figure 9:

Trend of Selected Income Items (average), Tax Years 1991–2000

60,000 50,000

$

40,000 30,000 20,000 10,000 0 1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

Year Employment

Net professional

Commissions

Net commission

Other Pen.

Net fishing

Sources: Canada Customs and Revenue Agency (2000– 2002) and Revenue Canada (1993– 1999) Table 5: Percent Growth Rate of Income Items per Return Used as Input to the T1 Model Runsa, 2006–2026 (%) Income items in T1 return

2006

2011

2016

2021

2026

Employment income Commissions Other pensions or superannuation Net professional income Net business commission income Net fishing income Others

5.01 15.24 13.88 3.91 25.43 15.72 0.0

10.54 29.79 24.81 10.81 45.24 29.93 0.0

16.07 44.33 35.75 17.71 65.05 44.14 0.0

21.60 58.88 46.68 24.60 84.86 58.35 0.0

27.13 73.42 57.62 31.50 104.67 72.56 0.0

Note: Based on the slopes of the regression lines over the past 10 years. a The percentage growth rates in this table were calculated for the 6, 11, 16, 21 and 26-year intervals from 2000 as a base year. The growth rates for each item are changing over the years and can be different from the historical growth rates. Although each individual had his or her own set of income items in 2000, the average rate of increase for each income item for each interval indicated in the table was applied to each individual.

Impact of Canadian Population Ageing

353

and 2020. Although it is quite possible for Canada’s economy to experience a slow down during some years in the future 20–30 years, it is expected that the average growth rate will be positive, because the total population and productivity will keep increasing. The personal income growth rate should be lower than the economic growth rate, and it should be more stable. In this research, extrapolation is used to predict the various income items (average) based on their historical trends or regression lines.17 Although there are other approaches to project income growth, for the specific income items used in our model, projecting from trends over 1991 to 2000 is considered as the most appropriate method with available data source. For the purpose of this study, we consider two different approaches to projecting income tax revenues into the future. The first approach is a demographic approach, to estimate the sole effect of population ageing. In this approach, we assume that other socio-economic conditions related to personal income tax (such as income level, employment rate, employment structure, population geographic distribution, etc.) will remain constant at the year 2000 level over the period 2000–2026 and that only the structure of the population will change. The second approach is a demographic-income approach, to estimate the combined effect of demographic change and real income level change. In this approach, we assume that not only the structure of the population will change but also the average of the selected major income items mentioned previously will change by following the historical trend indicated in Figure 9. The future growth rates of each income item, which are needed as an input to the T1 model runs, are determined by the slopes of the historical regression lines and the number of years from the base year 2000 (see Table 5).

17

The relevant regression lines are as follows:

Employment income ¼ 667,283 + 349.169 * year R2 ¼ 0.909, F ¼ 79.5 (8.538) (8.915) Commissions ¼ 1322,823 + 672.718 * year R2 ¼ 0.857, F ¼ 47.8 (6.816) (6.917) Other pensions or superannuation ¼ 607,172+310.744 * year R2 ¼ 0.974, F ¼ 303.0 (17.044) (17.407) Net professional income ¼ 1,326,160 + 686.896 * Year R2 ¼ 0.554, F ¼ 10.0 (3.052) (3.155) Net business commission income ¼ 1,280,932 + 648.789 * year R2 ¼ 0.765, F ¼ 29.0 (5.049) (5.103) Net fishing income ¼ 804,315 + 409.261 * year R2 ¼ 0.365, F ¼ 5.0 (2.111) (2.143) t is in parenthesis. Other income items are considered as constant during the projection period, and the total personal income is the sum of all income items.

354

Weng-Fong Lu et al.

6. Results of the Income Tax Simulations 6.1 Demographic Approach Table 6 summarizes the results of the income tax simulations under the demographic approach. The steady growth in the Canadian total population in the next 26 years will cause the number of total returns to increase, as well as the number of taxable returns. However, the growth rate in the number of taxable returns will slow down significantly between 2021 and 2026. The proportion of taxable returns among the total returns will decrease from 69.2 per cent in 2000 to 68.4 per cent in 2026. Total income will continue to grow over the projection period, from $696 billion in 2000 to $796 billion in 2011 and to $883 billion in 2026. The reasons for these changes stem from the increase in the workforce population and the contribution from seniors’ population. The increase will become much smaller during 2021–2026. The proportion of employment income will drop from 68 per cent in 2000 to 65 per cent in 2011 and to 59 per cent in 2026. The proportion of pension income will increase from 11 per cent in 2000 to 13 per cent in 2011 and to 17 per cent in 2026. Total taxable income will increase from $619 billion in 2000 to $708 billion in 2011 and to $770 billion in 2021 — but from 2021 to 2026 it will only increase by $9 billion. Average taxable income will rise from $29,914 in 2000 to $30,189 in 2006 — and then decline to $30,178 in 2011 and $29,791 in 2026. The average taxable income per return in 2026 will be $123 less than it was in 2000. Total non-refundable tax credits will climb from $39.5 billion in 2000 to $45.1 billion in 2011 and to $51.6 billion in 2026, which represent increases of $5.6 billion and $12.1 billion, respectively over year 2000. Of the increments, $2.9 and $11.3 billion will be contributed by the age amount tax credits that accounted for 52 and 93 per cent of the total increments in 2011 and 2026 over 2000. Average non-refundable tax credits will increase from $1,778 in 2000 to $1,791 in 2011 and to $1,836 in 2026. Population ageing is the critical factor that causes the increase in the reported tax credit amounts. The net federal tax will rise from $90.9 billion in 2000 to $104.2 billion in 2011 and to $113.9 billion in 2026. The average net federal tax per taxable return will increase from $5,906 in 2000 to $5,974 in 2006 and to $5,976 in 2011 — and then it will go down to $5,949 in 2016 and to $5,922 in 2026. This result can be explained as a consequence of population ageing. Population structure change during 2000–2011 will not have a negative effect on federal income tax revenue, but after 2011, when

Table 6:

Results of T1 Tax Model, Demographic Approach, 2000–2026 (constant 2000 dollars)

a

2011

2016

2021

2026

22,237 15,394 69.2 695.6 67.5 10.9 12.0 5.3 2.6 1.7 100.0 32,332 618.7 29,914 39.5 11.3 28.2 1,778 90.9 5,906 0.89 0.06 0.15

23,868 16,555 69.4 754.3 66.3 11.6 12.4 5.3 2.6 1.7 100.0 32,631 670.8 30,189 42.6 12.6 30.0 1,785 98.9 5,974 0.89 0.06 0.15

25,171 17,440 69.3 796.1 64.8 12.6 12.8 5.3 2.7 1.8 100.0 32,628 708.0 30,178 45.1 14.2 30.9 1,791 104.2 5,976 0.89 0.06 0.15

26,389 18,228 69.1 832.1 62.8 14.0 13.4 5.2 2.8 1.8 100.0 32,500 740.2 30,043 47.5 16.6 30.9 1,803 108.4 5,949 0.89 0.06 0.15

27,669 19,068 68.9 873.1 61.2 15.1 13.8 5.1 2.9 1.8 100.0 32,500 777.0 30,022 50.3 19.4 30.8 1,818 113.4 5,948 0.89 0.06 0.15

28,119 19,226 68.4 882.5 58.9 16.8 14.6 5.0 3.0 1.8 100.0 32,281 785.6 29,791 51.6 22.6 29.0 1,836 113.9 5,922 0.89 0.07 0.14

355

The values for year 2000 are based on the T1 model runs.

2006

Impact of Canadian Population Ageing

Number of total returns (thousands) Number of taxable returns (thousands) Proportion of taxable returns (%) Total income (billion $) Employment income (%) Pension income (%) Other sources (%) Self-employment income (%) Tax exempt (%) Other income (%) Total (%) Average total income per return ($) Total taxable income (billion $) Average taxable income per return ($) Total non-refundable credits (billion $) Age amount credits (billion $) Other credits (billion $) Average non-refundable credits per return ($) Net federal tax (billions $) Average net federal tax per taxable return ($) Taxable income /total income Non-refundable credits/taxable income Net federal tax/Taxable income

2000a

356

Weng-Fong Lu et al.

population ageing becomes more and more severe, it will show some negative effects.18 Tax yield is indicated by the three ratios shown in the last three rows of Table 6. In the case of the demographic approach, all three ratios show very little change over the period 2000–2021 and it shows the effect of population ageing on tax yield is also limited. But by year 2026, the increase in the ratio of non-refundable tax credits over taxable income (0.06 to 0.07) associated with an increasing senior population leads to a lower ratio for the net federal tax over taxable income, from 0.15 to 0.14. Therefore, if other things remain equal, population ageing will have some effect on the federal income tax system in 2026 when the number of seniors reaches the maximum and higher income working age population (aged 30–64) starts to decline.

6.2 Demographic-Income Approach Table 7 summarizes the results under the demographic-income approach. Compared with the demographic approach, there will be more taxable returns in the demographic-income approach. The proportion of number of taxable returns will be 71.7 per cent in 2026, 3.3 per cent higher than that in the demographic approach in the same year (68.4 per cent) and 2.5 per cent higher than that in the demographicincome approach in 2000 (69.2 per cent). Total income will reach $867 billion in 2011 and $1,073 billion in 2026, with these increases being 25 and 54 per cent respectively over the year 2000 income. The introduction of the factor of wage rate change accounts for an additional $71 and $191 billion respectively in total income over the demographic approach. Similarly to the demographic approach, the proportion of employment income will fall — and the proportion of pension income will rise — during the period of 2000–2026. With the income items growing over the 2001–2026 projection period, total taxable income will rise to $778 billion in 2011 and $975 billion in 2026, an increase of $70 and $190 billion, respectively over the demographic approach. The average taxable income per return will increase to $33,117 in 2011 and to $36,918 in 2026, which represents an average annual increase of $269 per return from 2000 to 2026. The total non-refundable tax credits show little difference from those in the demographic approach: $45.2 billion in 2011 and $51.7 billion in 2026. 18

Here, year 2011 should be around 2011 (i.e. it could be a year before 2011 or after 2011), and the specific year cannot be identified here since we only have five-year population and income tax projection data from 2000 to 2026.

Table 7:

Results of T1 Tax Model, Demographic-Income Approach, 2000–2026 (constant 2000 dollars) 2006

2011

2016

2021

2026

22,237 15,394 69.2 695.6 67.5 10.9 12.0 5.3 2.6 1.7 100.0 32,332 618.7 29,914 39.5 11.3 28.2 1,778 90.9 5,906 0.89 0.06 0.15

23,868 16,868 70.7 787.0 66.8 11.8 11.9 5.3 2.5 1.7 100.0 34,045 703.4 31,610 42.6 12.4 30.3 1,788 106.8 6,334 0.89 0.06 0.15

25,171 17,933 71.2 866.6 65.8 12.9 11.8 5.3 2.5 1.7 100.0 35,515 778.2 33,117 45.2 13.7 31.5 1,796 121.5 6,777 0.90 0.06 0.16

26,389 18,888 71.6 942.4 64.5 14.4 11.8 5.2 2.5 1.6 100.0 36,809 849.9 34,443 47.7 15.9 31.8 1,809 135.8 7,189 0.90 0.06 0.16

27,669 19,881 71.9 1,026.6 63.5 15.6 11.8 5.2 2.4 1.5 100.0 38,212 929.5 35,856 50.4 18.3 32.1 1,824 151.8 7,635 0.91 0.05 0.16

28,119 20,167 71.7 1,073.4 61.6 17.3 12.0 5.1 2.5 1.5 100.0 39,267 975.2 36,918 51.7 21.1 30.7 1,841 162.0 8,032 0.91 0.05 0.17

Impact of Canadian Population Ageing

Number of total returns (thousands) Number of taxable returns (thousands) Proportion of taxable returns (%) Total income (billion $) Employment income (%) Pension income (%) Other sources (%) Self-employment income (%) Tax exempt (%) Other income (%) Total (%) Average total income per return ($) Total taxable income (billion $) Average taxable income per return ($) Total non-refundable credits (billion $) Age amount credits (billion $) Other credits (billion $) Average non-refundable credits per return ($) Net federal tax (billion $) Average net federal tax per taxable return ($) Taxable income/total income Non-refundable credits/taxable income Net federal tax/taxable income

2000

357

358

Weng-Fong Lu et al.

The age amount credit remains as the major contributor to the total nonrefundable tax credits in the projection period. An important observation is that the total net federal tax will continue to increase over the next 26 years, from $91 billion in year 2000 to $122 billion in 2011 and to $162 billion in 2026. In terms of growth rates, the average annual compound growth rate over the period 2000–2011 will be 2.8 per cent, which exceeds the average growth rate of 1.9 per cent for the period 2011–2026. This indicates that the growth of total net federal tax will slow down over the latter part of the projection period. The average tax per taxable return will also continue to increase at its own average annual compound growth rate of 1.3 per cent to $6,777 in 2011 — and at a rate of 1.1 per cent over 2011–2026 to a level of $8,032. Under the demographic-income approach, which allows for changes in income, the three tax yield ratios show some changes. By year 2026 the ratio of taxable income to total income will increase from 0.89 to 0.91 and the ratio of deduction to total income will reduce by 2 percentage points. The ratio of non-refundable tax credits will decrease by 0.01–0.05 and the ratio of the net federal tax will increase over from 0.15 in year 2000 to 0.17 in year 2026.

7. Conclusion Canadian population ageing will be a significant factor in the next 20–30 years. The proportion of the population aged 65 and over of the total population will reach about 15 per cent in 2011 and 21 per cent in 2026. The Canadian population will age faster than other North American countries. According to CRA 2000 personal income tax information, personal income tax is related to age and changes by age. On average, the seniors’ population has historically had pension income as their major source of income and has had less taxable income and paid less tax than persons aged 30–64, who make up the vast majority of the workforce. Therefore, population ageing will have some impacts on personal income tax revenue. The microsimulation tax projections carried out under the demographic approach — which assume no increases in income and simply model demographic change — suggest that population ageing will cause average taxable income to decrease and non-refundable tax credits to rise significantly from 2011. Although the total personal income tax will not decrease over the whole projection period, 2000 to 2026, growth is projected to be very slow. Furthermore, the average federal net tax per taxable return will start to decline within the period of 2011–2026. Projections under the probably more realistic demographic-income approach suggest that the total and average taxable income and federal income tax collections will continue to increase

Impact of Canadian Population Ageing

359

over 2001–2026. However, the average annual compound growth rate will be slower over the latter part of the projection period — i.e., from 2011 to 2026. The overall effect of population ageing on personal income tax relates not only to the proportion of the senior population but also to the proportion of the working age population. During the years 2000–2011, the proportion of the population represented by both seniors and working age people increases, with the result that average federal income tax per return will actually rise slightly. From year 2011 when the proportion of the senior population starts to increase rapidly and the proportion of the working age population starts to drop, Canadian population ageing will have a negative effect on federal personal income tax revenue. However, population ageing may not have a severe impact on personal income tax because the expected personal income increases, if they match past growth rates, will have a positive effect on income tax revenue in the future. Overall, the growth of the concentration of seniors in our population over time could lead to a slower growth rate in federal income tax revenues over the next 26 years, especially after 2011. However, based on the projections mentioned previously, it will not reduce the number of taxable returns and total personal income tax revenue, because there will be growth in the total population and an increase in total personal income in real terms.

Acknowledgement and Disclaimer Although the authors are employees of the CRA (formerly CCRA), the views presented in this paper do not necessarily reflect those of the Agency. The authors appreciate the comments on drafts given by Margot Greenberg, Allan McGrath, Andrew Buis, Phillip King and Brain Murphy.

References Anderson, J.M. (2001). Models for Retirement Policy Analysis, Report to the Society of Actuaries, http://www.soa.org/research/macrodemographic/Macrodemographic. html. Canada Customs and Revenue Agency (2000–2002). Income Statistics 1998–2000. CCRA, Ottawa. Citro, C.F. and Hanushek, E.A. (eds). (1991). The Uses of Microsimulation Modelling, Vol. 1, Review and Recommendations. National Academy Press, Washington, DC. Corak, M. (ed). (1998). Government Finances and Generational Equity (Cat. No. 68-513-XIB), Statistics Canada, Ottawa. Denton, F.T. and Spencer, B.G. (1999). Population Aging and Its Economic Costs: A Survey of the Issues and Evidence, Social and Economic Dimensions of an Aging Population Research Papers, McMaster University. Department of Finance Canada (2000). Budget 2000: Five-Year Tax Reduction Plan. Finance Canada, Ottawa.

360

Weng-Fong Lu et al.

Department of the Treasury of Australia (2002). Intergenerational Report 2002– 03, 2002–03 Budget Paper no. 5, Canberra. Fougere, M., and Merette, M. (1998). Population Ageing and Economic Growth in Seven OECD Countries. Department of Finance Working Paper, Department of Finance, Ottawa. Group of Ten (1998). The macroeconomic and financial implications of ageing populations, Bank For International Settlements, http://www.bis.org/publ/ gten04.htm. Harding, A. (ed). (1996). Microsimulation and Public Policy, Contributions to Economic Analysis 232, North Holland, Amsterdam. Health Canada (2002). Canada’s Aging Population. PWGSC, Ottawa. Jackson, H. and Matier, C. (2002). Public Finance Implications of Population Ageing: An Update, Department of Finance Working Paper, Department of Finance, Ottawa. King, P. and Jackson, H. (2000). Public Finance Implications of Population Ageing. Department of Finance Working Paper, Department of Finance, Ottawa. Merette, M. (2002). The Bright Side: A Positive View on the Economics of Ageing. Choices Economic Growth, 8, (1), 1–28, IRPP, Montreal. Population Division of the Department of Economic and Social Affairs of the United Nations Secretariat (2003). World Population Prospects: The 2002 Revision and World Urbanization Prospects: The 2001 Revision, http://www.esa.un.org/unpp. Revenue Canada (1993–1999). Taxation Statistics 1991–1997. Revenue Canada, Ottawa. Robson, W.B.P. (2003). Time and Money: The Fiscal Impact of Demographic Change in Canada. Commentary, No.185. C.D. Howe Institute, Toronto. Statistics Canada (2001). Population Projects for Canada, Provinces, and Territories 2000– 2026, (Catalogue no. 91-520). Statistics Canada, Ottawa. Statistics Canada (2002). CANSIM II, Tables 383-0001 and 383-0005. Statistics Canada, Ottawa. The Conference Board of Canada (2003). Canadian Outlook Long-Term Economic Forecast: 2003. Conference Board of Canada, Ottawa. United States Government Printing Office. (1997). Economic Report of the President 1997, Washington, DC.

Chapter 14

Basic Income or Vital Minimum? A Note on the Distributive Effects of Possible Reforms of the Spanish Income Tax Xisco Olivera and Amedeo Spadarob a

Universitat de les Illes Balears, Palma de Mallorca, Spain PSE-Paris-Jourdan Sciences Economiques, Paris, FEDEA, Madrid, and Universitat de les Illes Balears, Palma de Mallorca

b

Abstract During the last 20 years, the Spanish redistribution system has undergone wide-scale changes. Since 1979, the year of the creation of income tax in Spain, there have been several reforms. The political and academic debate on their effects on equity and efficiency is still open. In this chapter, we contribute to this debate by using a microsimulation model that allows us to analyze the redistribution effect of different tax policy scenarios. Taking as a basis for the comparisons, the 1999 system, we first analyze the redistributive performance of two scenarios based on a flat tax: one with a basic income and another one with a vital minimum. Finally, we concentrate our attention on some possible reforms contained in the Socialist Party proposal (PSOE, 2002). As expected, the results of our work suggest that a basic income-flat tax system has a strong redistributive impact, when compared with a vital minimum-flat tax mechanism. One interesting finding is that the cost, in terms of fiscal pressure, of such a reform is not too high when compared with the current fiscal system. The flat tax depends strictly on the amount of basic income given to each citizen, but with a flat tax around 25–30 percent, it is possible to achieve a strong redistributive impact.

1. Introduction During the last 20 years, the Spanish redistribution system has undergone wide-scale changes. Since 1979, the year of the creation of income tax in Spain, there have been several reforms. In 1996, after 14 years in International Symposia in Economic Theory and Econometrics, Vol. 15 Copyright r 2007 Elsevier B.V. All rights reserved ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15014-0

362

Xisco Oliver and Amedeo Spadaro

government, the Socialist Party lost the elections and the new government (headed by the Popular Party) decided to reform the redistribution system, primarily by changing the personal income tax (PIT) system. The new tax law was introduced in 1999, and political and academic debate on its effects on equity and efficiency is still open. In 2002 (PSOE, 2002), the Socialist Party proposed an alternative tax reform, consisting of the replacement of the current PIT system with a vital minimum — flat tax scheme. After the general elections of 2004, the new government (headed by Zapatero) started an internal debate about possible reforms of the PIT to reduce complexity and to improve the redistributive performance by minimizing the costs in terms of taxes collected. In this paper, we contribute to this debate by using a microsimulation model that allows us to analyze the redistributive effects of different tax policy scenarios. Taking as a basis for the comparisons, the 1999 system (defined by the Ley 40/1998), we first analyze the redistributive performance of two scenarios based on a flat tax: one with a basic income (BIFT), and another one with a vital minimum (VMFT). We then focus our analysis on some possible reforms contained in the Socialist Party proposal (PSOE, 2002). Of course, all these results must be considered with care, because of the simplifying assumptions made. The most important is the lack of behavioral reactions. The reforms analyzed in this paper are structural and they could have a strong impact on labor supply and consumption. This work must be interpreted as a further attempt to shed light on issues that are key elements in the political and economic debate. As mentioned before, since 1998 — the year in which the Popular Party’s government proposed the reform — various authors have analyzed the effects of the change. Most of these analyses have been based on microsimulation techniques. Castan˜er et al. (2001) use the panel data of the Spanish Institute of Fiscal Studies to look at the implications of the reform in terms of redistribution and welfare, showing that the 1999 scheme reduces total redistribution, mainly as a result of the reduction of tax receipts. Moreno et al. (1999) use tax office statistics and completed tax returns to measure progressivity, with similar results. Levi and Mercader-Prats (2002) focus on the analysis of the withholding mechanism and the effects on efficiency of the new income tax system, showing that the 1999 reform fails to reduce the compliance costs of taxpayers. Using another database, the Encuesta de Presupuestos Familiares, Sanchı´ s and Sanchı´ s (2000) simulate the new PIT system, taking into account the effects on household consumption of a VAT increase introduced to compensate for the fall in income-tax revenue that the reforms involved. Their results show that mean taxable income decreased in real terms between 1998 and 1999 (by 16 percent) and that the overall simulated new system increased disposable income inequality.

Basic Income or Vital Minimum?

363

Another strand of literature about the Spanish fiscal system put the accent on BIFT, reflecting the propositions of such authors as Atkinson (1995) and Hall and Rabushka (1995). Gonzalez-Paramo (1986) and Fuentes Quintana (1987) present the first serious discussion of a possible implementation of a BIFT system in Spain. Another interesting paper analyzing the implications of implementing alternative BIFT schemes on households of the Spanish Household Budget Survey 1990–1991 is by Dura´n-Cabre´ (2001). Their findings show that BIFT systems have a big impact in the reduction of the income inequality. The problem with the simulations carried out in the Dura´n-Cabre´ work is that the proportional tax rates proposed to maintain the neutrality of revenue and to reduce significantly the inequality of pre-tax income are around 60 percent, and so are too high to be politically feasible. More recently, Castan˜er and Sanz (2002) have focused on the analysis of redistribution and welfare implications of a VMFT reform. In this work, they simulate the redistributive effects of replacing the 2001 Spanish income tax by a VMFT scheme on a sample of taxpayers coming from the 1995 wave of the Panel of Tax Returns of the IEF. First, they look at the mix of vital minimum and flat tax rates maintaining the same aggregate fiscal pressure as the 2001 system showing that there is a non-linear (exponential) relationship between the level of tax and the value of the vital minimum. Second, they analyze the winner–loser effects of a particular combination of vital minimum and flat tax that gives the same redistributive impact as the 2001 system — showing that, with the equivalence scale they use, households with more taxpayer units are the winners from the reform. Prieto et al. (2002) also analyze a VMFT reform, additionally focusing on the polarization effects. They show that a VMFT scheme reduces both inequality and polarization. This chapter is organized as follows. In the following section, we describe the microsimulation model and the database used. In the Section 3, we describe the 1999 structure of the Spanish tax-benefit system. Section 4 is devoted to the BIFT and the VMFT simulations. In Section 5, we present the results of the simulations carried out following the proposition contained in the PSOE report (2002). Finally, in the last section, the main conclusions are summarized.

2. The Data and the Microsimulation Model Following the Bourguignon et al. (1998) experience with the Eur3 model, we have built a microsimulation model called GLADHISPANIA to simulate changes in the Spanish tax-benefit system, starting from microdata. The model uses a dataset containing economic and socio-demographic information on households, and simulates the impact that different tax-benefit policy scenarios have on the income distribution of the population.

364

Xisco Oliver and Amedeo Spadaro

The database used is the Spanish part of the 1995 European Community Household Panel (ECHP), published by EUROSTAT, which includes sociodemographic characteristics, income characteristics, and labor status. Our dataset contains information at both the individual and household levels. After filtering the sample for records without information on the head of the household, we obtained a sub-sample of 6,420 households out of 6,522. The original dataset was then updated, using a correction factor including inflation and growth rates from 1995 to 1999. Because we have disposable income sources in our database, we have used our microsimulation model to compute gross values.1 No changes in the socio-demographic structure have been taken into account. Obviously, many simplifying assumptions have been made owing to data deficiencies in order for us to build the model. For this reason, a validation and calibration exercise has been carried out to check the behavior of the microsimulation model (for further details see Oliver and Spadaro, 2004).

3. The 1999 Spanish Tax-Benefit System In this section, we describe the main aspects of the 1999 Spanish tax-benefit system. 3.1 Social Contributions Social security contributions can be divided into the social security contributions paid by the employee and those paid by the employer. Social security contributions depend on several factors: a person’s gross earned income, type of employment contract (temporary or permanent), employment hours (part-time or full time), work status (graduate workers, engineers, unqualified assistants, white-collar workers, etc.), employment sector, occupational status (self-employed, dependent worker, civil servant, etc.), and their previous status before being employed in their current job. There are various categories of ‘‘social affiliation status’’, each with its own system of regulation.2 1

We define gross income as disposable income plus personal income taxes and social contributions paid by the employee. The hypothesis made to treat monthly withdrawals is that they correspond exactly to the effective average income tax paid by each individual at the end of the year. 2 In some cases (maritime workers and certain types of government employees), the information needed to compute the social affiliation status was not available on the individual). The details of what was done in such cases can be found in Oliver and Spadaro (2004).

Basic Income or Vital Minimum? Table 1:

Social Security Contribution Rates

Item contribution Common contingencies% Mean of industrial accidents and professional illnesses% Unemployment Full-time worker (permanent worker)% Full-time worker (temporal worker)% Part time worker% Social guaranty fund% Professional training%

Table 2:

365

Firm

Worker

Total

1999 23.60(%) 4.00(%)

1999 4.70(%) 0.00(%)

1999 28.30(%) 4.00(%)

6.20(%) 6.70(%) 7.70(%) 0.40(%) 0.60(%)

1.60(%) 1.60(%) 1.60(%) 0.00(%) 0.10(%)

7.80(%) 8.30(%) 9.30(%) 0.40(%) 0.70(%)

Monthly Minimum and Maximum Bases in ptas., 1999

Minimum base Maximum base

80,815 ( ¼ minimum wage/12) 399,780

Two elements must be considered if we wish to compute social security contributions. There is a base rate for contributions, closely related to the worker’s earned income, between an upper and a lower limit, and there is also a contribution rate that is split into two: the employer’s contribution rate and the employee’s contribution rate. Table 1 details the contribution rates of the general social affiliation status system. Meanwhile, in Table 2, the maximum and minimum contribution base rates are shown under the 1999 Spanish system of regulation. In Table 1, we can see that the social security contributions paid by the employer amount to about 35 percent of the total, while the social security contributions paid by the employee only represent 6.4 percent. This is not usual in other European countries, where the social security contributions paid by the employer are quite low. Although the employer’s social security rates are high and are clearly specified, many contracts involve reductions in rates, depending on the employee’s conditions, prior to starting, or his or her current conditions. For example, there are rate reductions if the worker was previously unemployed, if the worker is over 45 years of age, or if the worker is disabled. Previous conditions are impossible to model due to the lack of information available. 3.2 Personal Income Tax (PIT) The Spanish PIT system is a yearly income tax system. During the year, income tax is paid — and withheld at the source — when people receive wages, capital income, or other income sources. At the end of the tax year, however, they must fill in an income-tax return and compute whether they

Xisco Oliver and Amedeo Spadaro

366 Table 3:

Tax Rates Scheme, 1999

Individual and joint income-tax return Bracket (in ptas.) 0–600,000 600,000–2,100,000 2,100,000–4,100,000 4,100,000–6,600,000 6,600,000–11,000,000 411,000,000

Marginal tax rate 0.18 0.24 0.283 0.372 0.45 0.48

have to pay additional sums of money or whether they are entitled to get money back from the treasury department. A very small number of people, those with the lowest incomes, are not required to fill one in, although they can do so if it is in their interest. The Spanish PIT system has undergone a dramatic change with the major reforms of 1999.3 The system moved from a structure in which people’s specific conditions were taken into account mainly by means of tax deductions to one where they are taken into account by means of tax allowances. Let us take the case of a dependent child, as an example. Before the 1999 reforms took place, there was a tax reduction of 25,000 ptas. for the first child, 35,000 ptas. for the second one, and 50,000 for the third and any subsequent children. Under the 1999 system, there is a deduction of 200,000 ptas. for each of the first two children and 300,000 ptas. for the third child and any of the following ones, but this amount reduces taxable income rather than tax itself. The 1999 reforms followed the German philosophy of a subsistence-level minimum income: the income that is taxable must only represent the surplus income, once basic needs have been covered. These reforms also conform to the government’s announcement that it would lower the tax rate schedule and reduce the number of tax brackets from 8 to 6, as can be seen in Table 3. We can see that there is only one schedule for single persons and family income-tax returns. We can also observe that the maximum and minimum marginal taxes have fallen. The maximum amount of tax has gone down from 56 to 48 percent, while minimum marginal tax rates have been reduced from 20 to 18 percent. The main characteristics of the 1999 PIT system are described in Table 4. The income that is subject to PIT includes earned income (gross wages and income of self-employed), income from property, capital income, and

3

The 1999 reforms were introduced by virtue of ‘‘Ley 40/1998 de 9 de diciembre, del impuesto sobre la renta de las personas fı´ sicas y otras normas tributarias.’’

Basic Income or Vital Minimum? Table 4:

367

Main Characteristics of the Spanish PIT in 1999

Gross wages (includes wages, retirement pension, unemployment benefits, etc.) + Self-employment income + Property income (Owner occupied dwelling is not an income source) + Capital income Paid dividends must be increasing in 40%, but there exist a tax credit in the same amount. This has been made to avoid double taxation of firm profits ¼ Taxable income before vital minimum -Tax allowances Personal minimum: 550,000 ptas. (+100,000 for people older than 65) Family minimum Ascendants: 100,000 each one if their rents are lower than the minimum wage Dependent children: 200,000 each of the first two and 300,000 for the rest. These amounts are increased in 50,000 ptas. per child under 3 and 25,000 per child between 3 and 16 (dependent children are children under 25 and with rents under 1 million) ¼ Taxable income before tax allowances - Tax allowances Pension plan: with a maximum of 1,100,000 ptas. or 20% of earned income ¼ Taxable income ) Tax before tax credits - Tax credits ¼ PIT Notes: Tax credits in 1999 are cultural items investment (15 percent); donations (10–25 percent); paid dividends (40 percent) in the general case; and house investment. For house investment, there is a tax credit of 15 percent for the amount invested in special housing bank accounts with a limit of 1,500,000 ptas. per year, during a maximum of 4 years; Mortgage payments ¼ mortgage interests+mortgage repayments; Mortgage payments without loans yields a tax credit of 15 percent and with loans there are two categories: (1) First two years: 25 percent for the first 750,000 ptas. and 15 percent until 1,500,000 ptas.; and (2) Third year and following: 20 percent for the first 750,000 ptas. and 15 percent until 1,500,000 ptas.

changes in wealth. These last two are all classified as capital income in our model, due to the lack of information and the complexity of the taxation of these sources of income. The 5 percent deduction on gross earned income, with an upper limit of 250,000 ptas., is eliminated by the reform. Instead, new deductions on earned income are introduced, depending on the level of earned income in question. Earned income can be reduced by between 375,000 and 500,000 ptas., depending on the amount earned (i.e., by 500,000 ptas. if a person’s earned income is lower than 1,350,000 ptas. and 375,000 if it is greater than 2,000,000 ptas.). The deduction on gross income for mortgage interest

368

Xisco Oliver and Amedeo Spadaro

payments on the purchase of a house (the earner’s main residence) is also eliminated and, instead, a new tax credit is introduced. As for capital income, the reform eliminates the supposed income from owner-occupied dwellings (2 percent of the registered council value of the property). In addition, the tax deduction on returns on capital income (the ‘‘minimum income exemption’’ of 29,000 ptas.) is also eliminated. Before the reforms, there was no minimum personal exemption or minimum family exemption, but there were personal and family tax deductions. Under the 1999 system, once taxable income has been calculated (before the subsistence-level minimum income), we have to apply the personal and family minima, which then give us the taxable income before allowances. The minimum personal exemption is 550,000 or 1,100,000 ptas. in the case of a couple who fill in a joint-family income-tax return. This personal minimum exemption amounts to 650,000 ptas. when the earner is over the age of 65 and 850,000 or 1,000,000 ptas. in the case of a disabled person. The minimum family exemption involves two tax deductions. The first is a deduction of 100,000 ptas. for each dependent relative over 65 years of age with an income below the minimum wage. The second is a deduction per dependent child: 200,000 for each of the first two and 300,000 ptas. per child after the second child. In both cases, these quantities are increased by 25,000 ptas. per child for children aged between 3 and 16 and by 50,000 if the children are under 3 years of age. Tax allowances changed relatively little with the reforms. Mortgage interest is grouped together with mortgage repayments and become a tax credit under the new tax system. Pension plans remain unchanged, except for a modification of the upper limit for deductions, which changes for people over 53 years of age — the maximum deduction rises from 1,100,000 to 2,200,000 ptas. When tax allowances are subtracted, we get taxable income and we are ready to compute the tax before tax deductions. Then, tax deductions must be taken into account (see notes to Table 4). In 1998, there were a lot of tax deductions but, in 1999, some of them were included in the subsistence-level minimum income (i.e., personal and family tax deductions). Others became tax deductions on different kinds of expenditure (i.e., tax deductions on employees wages) and some of them were eliminated (i.e., expenditure because of illness and house rentals). With the new PIT system, earnings allowances and increases in personal or family minima replace deductions for personal disabilities. After the application of tax deductions, we obtain the amount of tax that must be paid (the cuota ı´ntegra) but, as mentioned before, tax is withheld at source every month. So, at the end of the tax year, people must calculate whether they have to pay additionally (a cuota lı´quida) or whether they are entitled to get money back. In the microsimulation model, we do not take into account monthly withholdings. Instead, we make a direct calculation of the net tax due at the end of the year.

Basic Income or Vital Minimum?

369

4. Vital Minimum-Flat Tax vs. Basic Income-Flat Tax To explore the implications on welfare and redistribution of the introduction of a flat tax, we have run, as a first stage, two kinds of simulations: the BIFT and the VMFT reform. The VMFT reform replaces the 1999 PIT with a vital minimum, which consists in a tax allowance per equivalent adult;4 and a proportional tax on the rest of the income. The BIFT reform consists in an universal lump-sum transfer, called the ‘‘basic income’’ (i.e., an amount of money that the government allocates to each household, independent of income and status) plus a flat tax on any remaining income. As in the VMFT option, we take into account the number of members of the household, giving a basic income per equivalent adult. The advantages or disadvantages of a VMFT or BIFT scheme are well known in the literature.5 They can be summarized as follows: Advantages

Disadvantages

Eliminating all the current allowances and deductions would broaden the tax base. Then, all sources of income would be treated equally (horizontal equity).

These schemes can affect the labor supply of more productive people if the flat tax is too high.

Simplicity for taxpayers, and consequently, more transparency, since all income is taxed at the same rate.

High rates can cause capital flows toward other countries with better capital fiscal treatment.

Simplicity for the treasury department, and then, minor collection costs and less tax evasion.

Lower flat taxes can generate redistribution toward the rich people. In the Spanish case, the incentive to save disappears.6

We have run four simulations for different flat tax rates. To facilitate the analysis of the redistributive performance of the various alternatives, the 4

The equivalence scale used is the square root of the number of household members. For more details of the general properties of a basic income-flat tax scheme, see Atkinson (1995). 6 It is important to stress that, in Spain, the main tax deductions (tax allowances and tax credits) are pension plans and house investment. 5

Xisco Oliver and Amedeo Spadaro

370 Table 5:

BIFT and VMFT Simulated Scenarios

Flat tax rate (%) 46 38 30 25

BIFT Basic income 770,650 586,750 402,850 287,900

ptas. (column 4) ptas. (column 6) ptas. (column 8) ptas.(column 10)

VMFT Vital minimum 2,328,900 1,996,900 1,595,400 1,287,400

ptas. ptas. ptas. ptas.

(column (column (column (column

5) 6) 7) 8)

Note: The column references refer the reader to the correct column of results in Table 6.

basic income or vital minimum has been chosen to respect the government’s budget constraint (with respect to our year of reference, 1999). In Table 5, we show the four simulated scenarios. We start from the maximum marginal tax rate of the 1999 system (46 percent), which allows 770,650 ptas. of annual basic income per equivalent adult (and 2,328,900 ptas. as vital minimum), and we reduce the flat tax rate to 38, 30, and 25 percent. Obviously, reducing the flat tax implies reducing the basic income or vital minimum simulated.

4.1 Overall Results Table 6 show the generalized Lorenz curve results of the 1999 scenario, as well as the percentage variations for each of the simulations carried out. The table is useful to identify the winners and losers from each policy, relative to the reference scenario (1999) as well as the redistributive effects of each. The results are presented by deciles of 1999 gross income per equivalent adult. The second and third columns show the values of 1999 disposable and gross income respectively, while the others represent the percentage variation in the disposable income of each decile, with respect to the 1999 reference values. From the fourth column to the eleventh, we show all the BIFT and VMFT simulated reforms. All of the various scenarios result in the collection of the same total tax revenue — namely, the same as that actually collected in 1999. Columns (4) and (5) show the simulations for a flat tax of 46 percent, a basic income of 770,650 ptas. and a vital minimum of 2,328,900 ptas., respectively. With the BIFT, the first 6 deciles win, while with the VMFT the first 8 deciles win. But there are big differences in gains. For example, in the first decile, with BIFT there is a disposable income increase of 122 percent (disposable mean income moves from 443,130 ptas. to 984,561 ptas.), while there is an insignificant gain of income with the VMFT scenario. These simulations clearly reveal that BIFT reform is more redistributive than VMFT; and both are more redistributive than the 1999 scenario.

Table 6:

Differences in Disposable Income by Decile from the BIFT and VMFT Reform Scenarios

Decile

46%

38%

30%

25%

(2) Disposable income

(3) Gross income

(4) BIFT (%)

(5) VMFT (%)

(6) BIFT (%)

(7) VMFT (%)

(8) BIFT (%)

(9) VMFT (%)

(10) BIFT (%)

(11) VMFT (%)

443,130 849,291 1,020,399 1,219,141 1,433,091 1,678,794 1,965,069 2,367,400 2,921,164 4,736,055 1,863,702

483,126 881,266 1,069,590 1,308,247 1,574,453 1,906,255 2,294,961 2,849,655 3,633,073 6,517,792 2,252,176

122.18 42.56 28.11 15.99 7.86 1.06 4.23 9.29 13.33 14.75 0

0.02 0.17 0.64 2.15 4.47 7.38 8.88 3.40 3.03 8.40 0

89.84 29.36 18.43 9.50 3.80 0.82 4.31 7.40 9.69 7.63 0

0.02 0.17 0.64 2.15 4.47 6.82 4.41 0.13 3.80 4.00 0

57.41 16.14 8.74 3.00 0.29 2.70 4.35 5.54 6.07 0.55 0

0.02 0.17 0.64 2.13 3.59 1.82 0.50 2.34 3.48 1.05 0

37.19 7.87 2.71 1.05 2.83 3.87 4.38 4.39 3.80 3.93 0

0.02 0.17 0.57 0.99 0.46 1.84 2.65 2.95 2.64 4.65 0

Basic Income or Vital Minimum?

1 2 3 4 5 6 7 8 9 10 Overall mean

1999

Note: Generalized Lorenz curves differences relative to the reference scenario 1999. Disposable income is given in ptas. per equivalent adult of the household. The percentages are differences relative to the reference scenario (1999 disposable income). BIFT and VMFT reforms collect the same tax revenue as in 1999.

371

Xisco Oliver and Amedeo Spadaro

372 Table 7:

Effects of the Reforms on Inequality Indices

Measure

Gini Atk e ¼ 0.5 Atk e ¼ 0.9 Atk e ¼ 1.5 Atk e ¼ 2 Entr c ¼ 0.1 Entr c ¼ 0.5 Entr c ¼ 0.9 Entr c ¼ 2

1999

46%

38%

30%

25%

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

0.330 0.102 0.229 0.296 0.525 0.285 0.209 0.202 0.246

0.374 0.129 0.268 0.349 0.575 0.341 0.267 0.267 0.374

0.223 0.046 0.077 0.117 0.145 0.089 0.093 0.100 0.140

0.313 0.093 0.218 0.285 0.519 0.270 0.191 0.181 0.210

0.259 0.061 0.102 0.155 0.193 0.119 0.124 0.132 0.188

0.318 0.097 0.222 0.289 0.521 0.276 0.199 0.191 0.235

0.295 0.078 0.132 0.201 0.252 0.156 0.160 0.169 0.243

0.326 0.102 0.228 0.294 0.523 0.284 0.210 0.205 0.268

0.318 0.091 0.153 0.236 0.299 0.183 0.186 0.195 0.281

0.334 0.107 0.234 0.300 0.526 0.293 0.219 0.217 0.295

If the tax rate is 38 percent, 30 percent, or 25 percent, we get similar conclusions (columns 6–11). The smaller the flat tax, the smaller is the increase in income of poorer people and the smaller is the decrease in the disposable income of richer deciles. An interesting result is that, under the BIFT scheme and for the simulation with a relative small tax rate (i.e., the 25 percent scenario), we observe an increase in the disposable income of the richest deciles. This happens because a small flat tax will reduce the tax burden of rich households with respect to the 1999 scheme. Under the VMFT scheme, the first deciles always observe no significant gains. The middle-class deciles are the winners of the reform while the richest deciles lose. As in the BIFT case, when the flat tax is small (30 percent or 25 percent), the last decile wins with the reform. Moreover, if tax rate is 25 percent, the VMFT scheme is less redistributive than the 1999 scenario. Table 7 shows the most relevant inequality indexes: Gini index, Atkinson index, and Entropy index for each scenario. We get a Gini index of 0.374 and 0.330, respectively7 for 1999 gross and disposable income. If the tax rate is 46 percent, we get a lower Gini index (0.223 in BIFT scenario and 0.313 in VMFT scenario), which implies less inequality. Moreover, the Gini index in BIFT is lower than the Gini index in VMFT scenario for every tax rate. If the tax rate is small, 25 percent, the Gini index is similar to the reference scenario (0.318 in BIFT scenario and 0.334 in VMFT scenario).

7 These values are very close to the ones found by Castan˜er et al. (2001) using the ‘‘Panel de Declarantes por IRPF del Instituto de Estudios Fiscales.’’ They are also very similar to values founded for Portugal by Farinha and Gouveia (1999), but they are relatively high if we compare them with other European countries.

Basic Income or Vital Minimum?

373

Atkinson and Entropy indexes, reported in Table 7, drive us to the same conclusions. If the flat tax is 46 percent, inequality decreases relative to the 1999 scenario. As expected, with a lower flat tax the redistributive power of the BIFT and VMFT scheme is lower. Under BIFT-38 percent and BIFT-30 percent, there is a big reduction in inequality (always with respect to 1999). Under VMFT-38 percent, there is some reduction of inequality, which disappears under VMFT-30 percent, where the inequality is the same as in the 1999 scenario (some indexes increase and some decrease, but the differences are very small). As observed before, under VMFT-25 percent, there is a slight increase in inequality when compared with the 1999 scenario.8 4.2 Taking Account of Household Size In this sub-section, we present the results of the analysis per household size, without using equivalent scales. To analyze the results on homogenous families, we have classified the households into five types: singles, couples,9 couples plus a dependent child,10 couples plus two dependent children, and couples with three or more dependent children (and we have excluded all households who do not fit into this classification). The results obtained for singles are presented in Table 8. The comparison between the two scenarios for any flat tax shows that the BIFT scheme increases the income of the poorest households, while the VMFT has no relevant effects with respect to the 1999 scenario. For example, with the BIFT-38 percent, for the first decile the increase in net income is 174.5 percent while for the second decile net income increases by 39.4 percent. The gains are positive and decreasing progressively until the eighth decile. For the last two deciles, net income is lower than under the 1999 scheme. On the other hand, with the VMFT-38 percent, there are no relevant changes in the deciles 1–6, a small increase in net income in deciles 7 and 9, 8

We have also tested change in tax progressivity and redistributive effect, using the Kakwani and Reynolds–Smolensky indexes. The figures are available from the authors, but reach the same conclusions as shown in Table 7 — namely that redistribution and progressivity under the 1999 scheme is always lower than in the BIFT and VMFT scenarios (except with VMFT-25 percent) and that the differences are reduced when the flat tax is lower, as expected. If we compare the BIFT and the VMFT we still observe that the BIFT scheme is much more progressive and redistributive than the VMFT. 9 For us, couples means two adults living in the household, whether they are married or not, because we are interested in seeing how many people can earn income. 10 Following the economic classification given by the ECHP, we consider as dependent children all the children below 16 years or between 16 and 25 years, if they live with his/her mother/father and they are inactive or unemployed.

374

Table 8:

Differences in Disposable Income by Decile from the BIFT and VMFT Reform Scenarios for Single Persons

Decile

46%

38%

30%

25%

(2) Disposable income

(3) Gross income

(4) BIFT (%)

(5) VMFT (%)

(6) BIFT (%)

(7) VMFT (%)

(8) BIFT (%)

(9) VMFT (%)

(10) BIFT (%)

(11) VMFT (%)

269,172 755,353 922,220 963,082 992,923 1,112,631 1,314,832 1,758,263 2,479,992 4,096,476 1,469,694

300,863 771,604 926,393 964,401 997,234 1,123,196 1,373,027 2,002,043 3,062,402 5,452,572 1,701,583

233.1 55.6 37.3 33.9 31.5 23.1 13.6 1.6 8.7 14.5 11.3

0.0 0.1 0.0 0.0 0.1 0.4 3.0 9.8 3.4 7.2 0.0

174.5 39.4 25.4 22.8 21.0 14.6 7.8 0.2 6.2 8.3 8.1

0.0 0.1 0.0 0.0 0.1 0.4 3.0 7.8 0.7 4.1 0.1

115.2 23.0 13.6 11.8 10.5 6.2 2.2 1.2 3.7 2.1 4.8

0.0 0.1 0.0 0.0 0.1 0.4 2.9 3.1 0.6 0.3 0.4

77.6 12.7 6.1 4.8 3.9 0.9 1.3 2.0 2.1 1.7 2.8

0.0 0.1 0.0 0.0 0.1 0.3 1.1 0.1 0.7 2.6 0.6

Note: Generalized Lorenz curves differences relative to the reference scenario. 1999 disposable income is given in ptas. per equivalent adult of the household. The percentages are differences relative to the reference scenario (1999 disposable income). BIFT and VMFT reforms collect the same tax revenue as in 1999.

Xisco Oliver and Amedeo Spadaro

1 2 3 4 5 6 7 8 9 10 Overall mean

1999

Basic Income or Vital Minimum?

375

and a net decrease of 4 percent in the last decile — while the eighth decile is the great winner with increase of 7.8 percent in their income. As shown in the previous cases, the BIFT-38 percent is more redistributive than the VMFT-38 percent. This conclusion can be extended to any flat tax simulated and, as we will see later, to any type of family. When the flat tax is 46 percent, the results are similar to those mentioned previously. Winner and loser deciles are the same, but with the BIFT-46 percent increases and decreases in net income become bigger. If we compare the VMFT-46 percent and the VMFT-38 percent, differences do not appear until the eighth decile (in the first seven deciles we are under the vital minimum and for this reason we do not observe changes). Yet, the change observed for the eighth decile is the most important with an increase of 9.8 percent — while in the top decile, the reduction in income goes from 4.1 to 7.2 percent. If the flat tax rate is smaller, 30 or 25 percent, we observe that the pattern of the changes remains equal, but the system is less redistributive. With a flat tax of 25 percent we discover two facts. First, it is notable that in the top decile there are no losers. This is due to the fact that the marginal tax rate for higher incomes decreases with respect to the 1999 system. Second, there is a little difference between the 1999 system and the VMFT-25 percent (the overall mean income just increases by 0.6 percent with the VMFT). The results of the simulations for couples without children are presented in Table 9. As we found before, the VMFT has no effect on low income households. We find also that, with a flat tax of 25 percent, the highest deciles increase their income under both the BIFT and the VMFT, due to a fall in marginal tax rates with respect to the 1999 scenario. Similar results are observed for ‘couple with one child’ households. As expected the BIFT scheme improves the situation of the poorest deciles, while there are no changes under the VMFT schemes. Other tests conducted by us showed that inequality indexes do not change too much. When we compare couples with two children with couples with three children (see Table 10), there are two striking facts. First, in the first four deciles, mean income is greater for couples with two children under the 1999 system. Second, in the last decile, income of couples with three children is greater than mean income for couples with two children. This is certainly because of the economic and socio-demographic structure of the ‘‘couples with three children’’ population: looking at the dataset we observe that there are many very rich families in this group (gross income rises from 8,735,110 in ‘‘couples with two children’’ to 13,424,778 in ‘‘couples with three children’’). An interesting general result observed by looking at the redistributive performance of the different schemes simulated on each type of household separately is that singles are better treated by the BIFT schemes than other types of households. The higher the number of people in the household, the

Differences in Disposable Income by Decile from the BIFT and VMFT Reform Scenarios for Couples with No or One Child

Decile

1999

46% (3) Gross income

1 2 3 4 5 6 7 8 9 10 Overall mean

788,788 1,159,970 1,439,602 1,749,165 1,978,025 2,205,613 2,574,716 3,026,521 3,759,331 6,367,603 2,508,715

817,042 1,173,705 1,480,095 1,806,140 2,052,039 2,368,584 2,878,647 3,517,235 4,538,190 8,529,217 2,918,466

1 2 3 4 5 6 7 8 9 10 Overall mean

811,055 1,522,312 1,875,458 2,238,250 2,641,054 3,139,915 3,679,078 4,409,611 5,305,897 8,222,401 3,388,817

895,331 1,650,575 2,085,779 2,519,532 3,085,819 3,665,457 4,468,354 5,473,628 6,630,307 1,131,9536 4,188,919

(4) BIFT (%)

(5) VMFT (%)

(6) BIFT (%)

Couples without children 89.0 0.0 64.9 47.4 0.0 33.0 28.9 0.2 19.0 15.7 1.4 9.3 9.4 2.3 4.5 4.6 4.6 1.4 1.1 8.1 2.3 6.2 7.1 -5.5 11.8 0.6 9.0 16.1 9.5 9.4 2.1 0.2 1.0 Couples with one child 111.1 0.0 81.5 37.6 0.0 25.4 21.9 2.0 13.7 12.5 4.2 7.1 5.4 8.5 2.7 2.0 9.2 2.7 7.0 6.6 6.2 11.8 0.0 9.2 14.7 4.9 -10.7 15.5 9.2 8.4 1.7 0.3 1.3

30%

25%

(7) VMFT (%)

(8) BIFT (%)

(9) VMFT (%)

(10) BIFT (%)

(11) VMFT (%)

0.0 0.0 0.2 1.4 2.3 4.5 6.0 2.5 2.6 5.7 0.3

40.5 18.6 9.1 2.8 0.3 1.8 3.4 4.8 6.2 2.8 0.0

0.0 0.0 0.2 1.3 2.0 2.7 0.7 1.3 3.4 1.2 0.4

25.4 9.7 3.0 1.3 3.2 3.8 4.1 4.3 4.5 1.3 0.6

0.0 0.0 0.2 0.6 0.9 1.7 2.3 2.8 3.2 2.0 0.7

0.0 0.0 2.0 4.2 8.5 6.6 1.8 2.5 5.1 4.8 0.4

51.2 13.2 5.7 1.9 0.2 3.4 5.1 6.3 6.5 1.1 0.8

0.0 0.0 2.0 4.2 4.7 0.7 1.5 3.3 4.1 0.5 0.3

32.4 5.5 0.7 1.5 1.8 4.0 4.6 4.6 4.0 3.4 0.5

0.0 0.2 1.6 1.1 0.4 2.1 3.0 3.3 2.9 4.1 0.2

Note: Generalized Lorenz curves differences relative to the reference scenario. 1999 disposable income is given in ptas. per equivalent adult of the household. The percentages are differences relative to the reference scenario (1999 disposable income). BIFT and VMFT reforms collect the same tax revenue as in 1999.

Xisco Oliver and Amedeo Spadaro

(2) Disposable income

38%

376

Table 9:

Table 10: Differences in Disposable Income by Decile from the BIFT and VMFT Reform Scenarios for Couples with Two and Three Children Decile

1999

46% (3) Gross income

1 2 3 4 5 6 7 8 9 10 Overall mean

864,306 1,691,397 2,122,422 2,552,593 2,922,703 3,459,094 4,042,138 4,911,492 5,957,639 8,735,110 3,730,593

971,003 1,833,280 2,320,688 2,884,803 3,448,228 4,104,464 4,906,724 6,073,781 7,675,951 12,188,569 4,649,590

1 2 3 4 5 6 7 8 9 10 Overall mean

697,657 1,555,095 2,033,793 2,493,147 3,053,019 3,676,934 4,333,177 5,677,032 7,284,143 13,424,778 4,432,462

811,022 1,707,360 2,185,559 2,754,342 3,439,653 4,226,644 5,203,907 6,974,084 9,488,940 20,806,905 5,795,302

(4) BIFT (%)

(5) VMFT (%)

(6) BIFT (%)

Couples with two children 125.5 0.0 91.9 40.9 0.4 27.9 23.9 1.2 15.3 12.3 4.4 7.1 6.8 8.4 3.5 0.5 10.3 0.8 4.5 9.9 3.8 9.7 2.5 7.2 12.4 2.4 8.3 12.8 6.0 5.9 0.8 1.7 0.9 Couples with three children 194.3 0.0 144.4 61.4 0.0 43.9 36.1 0.2 24.5 21.6 1.5 13.5 10.6 4.9 6.0 2.3 8.2 0.3 2.8 10.3 2.8 10.0 2.2 7.6 12.6 2.9 8.0 7.8 12.7 1.4 3.4 1.5 4.5

30%

25%

(7) VMFT (%)

(8) BIFT (%)

(9) VMFT (%)

(10) BIFT (%)

(11) VMFT (%)

0.0 0.4 1.2 4.4 8.4 8.4 4.7 0.3 2.5 2.0 1.5

59.2 15.0 6.6 1.9 0.4 2.0 3.2 4.8 4.1 1.0 1.1

0.0 0.4 1.2 4.2 5.4 2.4 0.5 1.7 1.6 2.7 1.3

38.2 6.9 1.2 1.4 1.5 2.6 2.8 3.3 1.5 5.3 1.1

0.0 0.4 1.2 1.2 0.8 0.7 1.2 1.9 0.4 6.1 1.2

0.0 0.0 0.2 1.5 4.9 8.0 5.8 0.7 2.7 4.3 3.0

94.0 26.0 12.6 5.6 1.4 1.7 2.9 5.1 3.5 10.6 5.5

0.0 0.0 0.2 1.5 4.2 2.9 1.0 2.1 1.1 11.9 4.5

62.5 14.5 4.9 0.6 1.4 3.1 3.0 3.5 0.6 16.3 6.2

0.0 0.0 0.2 1.4 0.8 1.0 1.2 2.2 0.4 16.9 5.5

377

Note: Generalized Lorenz curves differences relative to the reference scenario. 1999 disposable income is given in ptas. per equivalent adult of the household. The percentages are differences relative to the reference scenario (1999 disposable income). BIFT and VMFT reforms collect the same tax revenue as in 1999.

Basic Income or Vital Minimum?

(2) Disposable income

38%

378

Xisco Oliver and Amedeo Spadaro

lower is the redistribution obtained by the BIFT. This is certainly due to the equivalence scale used to assign the basic income that gives, proportionally, more weight to households of smaller size. On the contrary, the VMFT scheme benefits large households. This is because of the lower average tax rate that richer households (which are the ones with more components) face under the VMFT scheme (with respect to the 1999 one). This second result is in-line with the one obtained by Castan˜er and Sanz (2002).

5. Other Simulations 5.1 Description of Different Scenarios As it has been mentioned before, debate continues regarding the appropriateness of a new reform of the Spanish system of redistribution. In fact, in 2003, the Popular Party implemented marginal changes of the PIT. The reform goes in the same direction as the 1999 one. It basically reduces the minimum and maximum marginal tax rate, increases the disabled people expenditure allowance, and raises the vital minimum (family and personal). On the other hand, a number of politicians from the Socialist Party, with the support of some economists, recently proposed the introduction of a scheme similar to our VMFT. The underlying idea of simplifying the tax structure and introducing a sort of vital minimum gave rise to a great deal of debate on the eventual effects in terms of equity and efficiency. This proposal is currently under evaluation at the Spanish Ministry of Finances. In this section, we simulate five possible reforms. The description of each reform is summarized in the following Table 11. As before, the reforms only replace the PIT, leaving social contributions unaltered. The new elements introduced in the scenarios simulated are (1) the consideration of a marginal tax scheme with two rates (the second tax rate is introduced for incomes up to 5,750,000 ptas. and tries to avoid those on higher incomes paying less income tax because of a low flat tax); (2) the introduction of tax credits that will replace the present tax allowance due to the employee social contributions and vital minimum, personal, and family minimum (several economists have pointed out that a tax allowance benefits people with higher marginal taxes, while a tax credit is independent of the marginal tax faced by the individual); (3) keeping the tax credit due to house investment. Though some economists believe that this tax credit causes a distortion in the housing market (pushing up house prices and creating disincentives for house renters), the house investment tax credit has strong social support. Moreover, an immediate suppression of this deduction is very difficult, because a lot of tax payers have bought their house taking into account the existence of this deduction in the future.

Basic Income or Vital Minimum?

379

Table 11: Summary of Five Additional Tax Reform Options

Tax allowances Vital minimum Tax credits Employee social contributions House investment Vital minimum: Ptas. per person Ptas.per household Marginal Tax Rates Over 5,750,000 Basic Income

Reform 1

Reform 2

Reform 3

Reform 4

Reform 5

No

No

No

563,400a

No

Yes Yes

Yes No

Yes Yes

Yes Yes

No No

No

No

38%

25% +15% 194,300a

75,000 225,000 26.3% +14% No

75,000 No 225,000 25.55% 38% +15% No 455,450a

No

a Amounts are per equivalent adult, where the scale of equivalence used is the squared root of the number of the household members. In every case, the reforms collect the same tax revenue as in 1999.

Reforms 1 and 2 are very similar to two of the three reforms that the Socialist Party suggested in their report.11 We have only modified the first marginal tax rate in order to obtain the same tax revenue as in our reference scenario, the 1999 system. They both consist of: a 100 percent tax credit for employee social contributions, a vital minimum tax credit of 75,000 ptas. per person and 225,000 ptas. per family, and two marginal tax rates. They differ because reform 1 maintains the house investment tax credit, as in the 1999 system (see Table 4). Reform 3 and 4 are similar to our BIFT and VMFT reforms, with a flat tax of 38 percent — but including house investment and social contributions tax credits. The basic income and vital minimum that guarantees the 1999 tax revenue are 455,450 and 563,400 ptas., respectively. Reform 5 is a BIFT reform, with a basic income of 194,300 ptas. per year and per adult equivalent and a flat tax of 25 percent; it includes an extra marginal tax rate, of 40 percent, for incomes over 5,750,000 ptas. 5.1 Results The results of the simulations of the five scenarios described above are presented in Table 12. The first panel in Table 12 reports the Generalized Lorenz curve expressed in percentage changes from the reference situation 11

‘‘Una alternativa fiscal para Espan˜a.’’ Report of an independent committee realized for the PSOE, May 2002.

380

Table 12: Impact of the Five Possible Reform Scenarios Decile

1999 Disposable income

Gini Atk e ¼ 0.5 Atk e ¼ 0.9 Atk e ¼ 1.5 Atk e ¼ 2 Entr c ¼ 0.1 Entr c ¼ 0.5 Entr c ¼ 0.9 Entr c ¼ 2 Kakwani Reynolds–Smolensky T t/(1t)

Gross income

1

2

Generalized Lorenz curves (differences relative to the reference scenario) 631,866 666,754 0.02% 0.02% 1,135,848 1,173,391 0.17% 0.17% 1,583,190 1,683,786 0.56% 0.61% 1,991,181 2,153,832 0.57% 0.87% 2,402,931 2,666,514 0.14% 0.56% 2,868,149 3,298,190 0.83% 0.99% 3,421,074 4,010,357 0.97% 1.16% 4,148,550 4,989,131 0.93% 0.96% 5,261,016 6,491,807 0.19% 0.29% 8,570,309 11,724,108 0.88% 1.10% 3,201,954 3,886,975 0.14% 0.15% 0.330 0.102 0.229 0.296 0.525 0.285 0.209 0.202 0.246 0.220 0.046 0.172 0.208

Progressivity and redistributive measures 0.374 0.329 0.129 0.101 0.268 0.228 0.349 0.295 0.575 0.524 0.341 0.284 0.267 0.208 0.267 0.201 0.374 0.248 0.221 0.046 0.171 0.207

0.329 0.101 0.228 0.295 0.524 0.283 0.207 0.200 0.247 0.225 0.046 0.171 0.207

3

4

5

70.42% 18.37% 10.10% 4.55% 0.59% 1.68% 3.73% 5.33% 6.46% 4.47% 0.04%

0.02% 0.17% 0.64% 2.15% 4.05% 3.70% 1.59% 0.77% 2.76% 2.19% 0.09%

42.84% 7.41% 4.78% 2.54% 0.41% 0.44% 1.42% 2.55% 4.40% 4.67% 0.57%

0.283 0.071 0.120 0.183 0.229 0.141 0.145 0.153 0.212 0.439 0.091 0.172 0.208

0.322 0.099 0.225 0.291 0.522 0.279 0.203 0.196 0.244 0.252 0.053 0.173 0.210

0.300 0.080 0.137 0.218 0.282 0.163 0.163 0.168 0.221 0.350 0.075 0.177 0.215

Xisco Oliver and Amedeo Spadaro

1 2 3 4 5 6 7 8 9 10 Overall mean

Reform number

Basic Income or Vital Minimum?

381

(the 1999 scheme) by deciles of disposable income. The lower panel reports the inequality, redistribution and progressivity index. In general, we observe that reforms 1, 2, and 4 have a marginal effect on the reduction of inequality. On the contrary, the reforms 3 and 5 (both inspired by a basic income scheme) seem to improve substantially the equality of the income distribution. Under schemes 1 and 2, the disposable income of the first eight deciles increases and only the last two deciles lose a small amount (less than 1 percent). In any of these two scenarios, the percentage variation in disposable income (with respect to the 1999 scenario outcomes) is bigger than 1 percent. Looking at the Gini and Reynold– Smolensky indexes, we observe that inequality is reduced by an insignificant amount (see Table 12). Reform 4 results in an increase in net income for the first seven deciles and a decrease for the last three deciles. The reduction in inequality is bigger than under scenarios 1 and 2, but it is still marginal. In general, we observe that the replacement of the 1999 PIT with a linear (2 brackets) marginal tax rate (such as the one we used in simulations 1 and 2), or with a vital minimum and a flat tax, as in simulation 4, does not produce any relevant effect on the net income distribution. The story is completely different when we look at reforms 3 and 5. In these two cases, the redistributive power of the two systems is much greater than under the 1999 system. This effect depends strongly on the introduction of the basic income. Under scenario 3, the first half of the population observes an increase in their disposable income. The first decile’s net income rises 70 percent compared with the 1999 system. The deciles from 5 to 10 receive less net income than under the initial 1999 situation. An interesting outcome is that decile 9 loses more (4.47 percent) than decile 10 (6.46 percent). This is due to the fact that, under the 1999 system, the marginal tax rate for the 10th decile of the distribution is higher than the flat tax used for this simulation. If we look at the inequality index, we observe that the Gini goes from 0.33 in 1999 to 0.283 under reform 3. This is certainly an important reduction in the inequality of disposable income. We observe also that there is an important increase in the progressivity of the system: the Kakwani index goes from 0.22 to 0.439. The pattern of the results is the same when we look at reform 5. As in the previous case, the redistributive impact of this basic income flat tax plus tax credits system is important. The reduction in inequality favors the first five deciles and, in particular, the first one (which experiences an increase in disposable income of 42 percent with respect to the 1999 system). Looking at the Gini values, we observe a decrease in its value from 0.33 to 0.3. This means that reform 5 is more redistributive than the 1999 system but less than scheme 3. This is totally because of the size of the basic income transfer (which is 455.450 ptas. in reform 3 and only 194.300 ptas. in reform 5).

382

Xisco Oliver and Amedeo Spadaro

As in Section 3, the results of all the simulations presented in this section show that higher redistributive effects are obtained by introducing a basic income. This mechanism represents a strong income injection for lowincome households.

6. Conclusions Using the microsimulation model GLADHISPANIA, we have simulated the redistributive effects on a sample of Spanish households coming from the 1995 ECHP panel, of various alternative scenarios (using the 1999 Spanish income tax system as reference framework). The scenarios simulated are a basic income-flat tax structure and a vital minimum-flat tax structure. Different variants have been analyzed and the main results indicate that both structures perform better than the 1999 system in reducing initial (market) income inequality. Results also show clearly that, if the objective of the fiscal authority is to reduce inequality, the instrument that achieves higher levels of redistribution is a basic income. The main reason for this is that the 1999 redistribution system is basically structured around the progressivity of the income tax. No subsidies (means-tested or not) are implemented in order to guarantee a minimum level of income to poor households. The introduction of a basic income improves substantially the welfare of the lowest deciles of the income distribution. The cost of financing it depends on the amount of basic income given. The simulations show that a flat tax around 25–30 percent can be enough to finance a good redistributive performance without incrementing the tax burden. The results presented in this work are limited to the first-order effects. No behavioral reactions are considered. Labor supply reactions can affect in an important way the final disposable income of the households and should be taken into account if the objective is to perform a robust welfare analysis of alternative redistributive schemes. Nonetheless, the use of arithmetical microsimulation models still remains a powerful instrument to assess the effects of alternative redistributive policies.

Acknowledgements The authors acknowledge the useful comments of He´ctor Calvo, Jose´ Manuel Gonza´lez-Pa´ramo, Javier Ruiz-Castillo, and the participants of several seminars and conferences in which they presented this paper. They also acknowledge the financial support of the Spanish Government — MCYT (SEC2002-02606) and FBBVA. Usual disclaimers apply.

Basic Income or Vital Minimum?

383

References Atkinson, A.B. (1995). Public Economics in Action: Basic Income-Flat Tax Proposal. Clarendon Press, Oxford. Bourguignon, F., O’Donoghue, C., Sastre J., Spadaro, A. and Utili, F. (1998). Technical Description of Eur3: A Prototype European Tax-Benefits Model. DAE Research Note N.9801 Micro-Simulation Unit, Cambridge University, Cambridge. Castan˜er, J. M., Onrubia, J. and Paredes, R. (2001). Efectos de la reforma del IRPF sobre la renta disponible, su distribucio´n y sobre el bienestar social. Economistas, n1 87, Extraordinario an˜o 2000. Castan˜er, J. M. and Sanz, J. F. (2002). Un ana´lisis del impuesto lineal sobre la renta a trave´s de un ejercicio de microsimulacio´n. Presented at IX Encuentro de Economı´ a Pu´blica, 2002. Text available at http://www.uvigo.es/9ecopub/programa.html. Dura´n-Cabre´, J. M. (2001). Un estudio del impuesto dual sobre la renta aplicado al caso espan˜ol. Hacienda Pu´blica Espan˜ola, Monogra´fico: tendencias de reforma fiscal Farinha, C. and Gouveia, M. (1999). The impact of a Minimum Guaranteed Income Program in Portugal. ISEG — Departamento de Economia. Working Paper No. 3. Lisbon. Fuentes Quintana, E. (1987). El impuesto lineal: una opcio´n diferente. Papeles de Economı´a Espan˜ola, 30–31, 175–192. Gonzalez-Paramo, J.M. (1986). El impuesto lineal sobre la renta. Papeles de Economı´a Espan˜ola, 27, 297–302. Hall, R. and Rabushka, A. (1995). The Flat Tax. 2nd edition. Hoover Institution Press. Stanford. Levi, H. and Mercader-Prats, M. (2002). Simplifying the personal income tax system: Lessons from the 1998 Spanish reform. Fiscal Studies, 23, 419–443. Moreno, C., Paredes, R. and Utrilla, A. (1999). Efectos de la reforma del IRPF sobre la renta disponible y su distribucio´n y sobre el bienestar social: un ejercicio de simulacio´n con microdatos. Papeles de Trabajo n. 13/99 del Instituto de Estudios Fiscales. Madrid. Oliver, X. and Spadaro, A. (2004). A Technical Description of GLADHISPANIA: A Spanish Micro-Simulation Tax-Benefit Model. DEA Working Paper 7, Universitat de les Illes Balears. Available at http://www.uib.es/depart/deaweb/deawp/ Prieto, J., Rodriguez, J.G. and Salas, R. (2002). Linear Tax Reforms: Polarization, Redistribution and Horizontal Inequity. Theory and Empirical Simulations for the Spanish Case. Presented at TMR Network Living Standards, Inequality and Taxation, Fourth Annual Meeting, Lu¨beck, Alemania, September 2002. PSOE. (2002). Una alternativa fiscal para Espan˜a, Informe de la Comisio´n nombrada por el PSOE para elaborar una propuesta sobre la reforma del IRPF. Sanchı´ s, J.A. and Sanchı´ s, A.S. (2000). A Micro-Simulation Analysis of the Distributive and Incentive Effects of the Spanish 1999 Tax Reform: A Special Focus on Children Benefits. Presented at Workshop Fighting Poverty and Inequality through Tax Benefit Reform: Empirical Approaches, Barcelona. Text available at http://selene.uab.es/mmercader/workshop/index.html.

This page intentionally left blank

Part III: Wealth and Services

This page intentionally left blank

Chapter 15

Self Provision in Retirement? Forecasting Future Household Wealth in Australia Simon Kelly National Centre for Social and Economic Modelling (NATSEM), University of Canberra, Australia

Abstract The costs associated with an ageing population in Australia are putting increasing pressure on the fiscal resources of the government. This pressure will intensify as the proportion of elderly doubles over the next few decades. Under this increasing fiscal pressure, the government will be increasingly looking to individuals to provide for themselves. This chapter considers what capacity older Australians have to provide financially for themselves. It finds that if the traditional measure of economic well-being (income) is used then they will have very little capacity to contribute to the cost of their retirement. Almost all of those currently aged 65 and over rely almost entirely on the governmentfunded Age Pension. Despite the modest level of the pension (indexed at onequarter of average weekly earnings), only 17 per cent of current retirees have private income that matches or exceeds the pension. Taking a broader economic view — including wealth — provides a very different perspective. Those aged 65 and over currently have an estimated 22 per cent share of total household wealth and this proportion is likely to increase to 47 per cent by 2031. By including this wealth in the evaluation of capacity to contribute to the costs of retirement, it appears there is considerable scope for self-provision.

1. Introduction Australia has an ageing population. The Australian Bureau of Statistics (ABS, 2003) has recently revised upwards the projections for the median age. The ABS Series B projections now suggest that the median age will increase from 35.9 years in 2002 to 41.2 in 2021 and to 46.8 in 2051. This ageing population is causing a fall in the ratio of working age people (15–64 years) to those of retirement age (people aged 65 and over). There International Symposia in Economic Theory and Econometrics, Vol. 15 Copyright r 2007 Elsevier B.V. All rights reserved ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15015-2

388

Simon Kelly

were 7.3 working age people for each older Australian in 1960 while, 40 years later, at the start of the 21st century, there were 5.6 people. In 2040, there will only be 2.3 people of working age to support each person aged 65 and over (ABS, 2003). Unless changes occur, the government’s Intergenerational Report concludes that Commonwealth spending will exceed revenue from 2013 onwards (Commonwealth Treasurer, 2002). Almost all western countries face the same economic problem — a stagnant or shrinking workforce paying taxes and an expanding older population with high health, aged care and pension costs. Looking just at pension financing issues, the OECD (Organisation for Economic Co-operation and Development) concluded that future governments in most countries would be ‘‘hard put to finance [retirement pensions] out of pay-as-you-go contributions from people still in employment’’ (OECD, 2000). In addition, Disney and Johnson (2001) claim that most OECD pension systems are in a state of flux, as governments struggle to meet the costs of their current retirement income systems and grapple with the issue of future costs. Australia recognised the economic aspects of an ageing population in 1989 and developed a retirement income policy based on three pillars — a public pension scheme, compulsory private savings and voluntary private savings (see King et al., 2001, for a more complete description of the public and private schemes in Australia). A major emphasis of these pillars was to change the culture of public provision of retirement incomes to selfprovision. The government intention was that compulsory and voluntary retirement saving (called superannuation in Australia) would become the major source of retirement income — with the means-tested public pension providing a social protection function (i.e., a safety net). Despite this early response and accolades from the observers such as Khan (1999), Australia’s solution still faces problems. As will be shown later in this chapter, the vast majority of those currently aged 65 and over have very little private income and rely almost entirely on the public pension to provide their retirement income. On the face of it, it would seem that the implemented policy to achieve the goal of private retirement saving and its associated self-provision have not been effective. However, there are a range of factors that are influencing these current findings. First, the three-pillar approach has only been in place for 10 years. It takes many years of contributions to accumulate enough funds for superannuation to become the principal source of retirement income. Most current retirees will have contributed for a maximum of 10–15 years — not long enough to make a significant difference. Second, while the public pension was designed as a social protection system and not as an income replacement mechanism, many Australians believe they have earned an entitlement to this pension and an entire industry has developed around finding ways to

Self Provision in Retirement?

389

circumvent the means-testing aspects of the pension. Third, while the baby boomers have enjoyed high incomes during their working lives, they have consumed rather than voluntarily saved. Saving for retirement has always been a low priority and encouragement to voluntarily save for retirement has been required. The complex and constantly changing nature of the superannuation taxation laws has not been helpful in this regard. Fourth, redundancies and early retirement have resulted in the consumption of accumulated superannuation before age 65 is reached, with some Australians retiring early and then spending their ‘lump sums’ with the expectation of going onto the age pension when the official retirement age is reached (Kelly et al., 2004). The outcome is that, at present, 79 per cent of those over 65 receive the public pension (FaCs, 2001). Of these, 68 per cent receive the full public pension. In summary, the funding of public pensions and other costs associated with an ageing population is a global issue. In Australia, the government has taken a three-pillar approach to limit future costs — an approach that is based on a means-tested public pension supplemented by compulsory and voluntary private savings. The success of this approach and the capacity of people to provide for themselves are crucial to the health of future Federal Budgets (Kelly and Harding, 2004). Despite the importance of this issue, little is known of the assets owned by individuals or of their ability to contribute to the costs associated with ageing. This chapter attempts to address this issue by (a) providing estimates of the income and assets of older Australians in 2001 and (b) using the National Centre for Social and Economic Modelling (NATSEM) dynamic microsimulation model and its integrated wealth module to provide estimates of wealth in the future.

2. Modelling Income and Wealth 2.1 Current Income and Wealth The ABS collected information on personal income as part of the 2001 Australian Census. While the information was collected in income bands and only for gross (i.e. total) income, it does provide the most complete picture of the distribution of income in Australia. In this chapter these income data have been divided into those of working age (15–64) and those aged 65 and over. Additional information on the sources and levels of income can be obtained from the ABS 2000/01 Survey of Income and Housing Costs (SIHC), which is available at the unit record level. It is not possible to obtain data on all of the assets owned by an individual. For some assets, such as art works, it is difficult to obtain

390

Simon Kelly

information on the ownership of the asset let alone assign it a valuation. In this chapter the definition of wealth is restricted to significant long-term assets to which a valuation can be attributed with some degree of confidence. The assets include owner-occupied home, shares, cash deposits, superannuation and the residential rental properties. To obtain net worth, the liabilities associated with a home mortgage and loans for residential properties have been subtracted from the value of the assets. Despite limiting the definition to this small group of assets, information on wealth holdings at the individual or household level is not readily available. However, current housing values, current mortgage, the ownership of rental properties and a variety of investment income levels are available on the SIHC. As the most significant asset of most Australians is the equity in the family home, and as this information is available on the SIHC, it is the logical starting place for estimating wealth. Using the investment capitalisation methodology (Kelly, 2001) and the investment income data available on the SIHC, valuations for cash deposits and shareholdings can be imputed for each person. In addition rental property values and loans can be imputed using other variables available on the SIHC and a regression equation derived from data on the ABS 1997 Survey of Rental Investors (Kelly, 2001). The final asset required is superannuation. Using age, gender, labour force and income characteristics data on the SIHC, combined with a probability distribution derived from DYNAMOD superannuation simulation outcomes for 2001, reasonable estimates can be imputed for individuals. Superannuation modelling is discussed in more detail in Kelly et al. (2002). 2.2 Future Wealth There are a number of methods of projecting future personal wealth. The DYNAMOD model developed by NATSEM — a dynamic microsimulation model — is a complex, general-use tool that can answer a myriad of questions and dynamically handle changes to influences over time. This model looks at the circumstances of every individual, every month, and makes decisions based on their circumstances at that time — whether they will become pregnant or not; whether they will continue to live or not; and whether they will emigrate or not. A major advantage of dynamic microsimulation modelling is that because the unit of measurement is an individual it can provide great detail at this level and yet the individuals can be aggregated to provide a macro view or can be grouped to provide collective or group outcomes. There are currently about a dozen dynamic microsimulation models being used internationally. They include models currently being used in government policy making in the US (the CORSIM, POLISIM and MINT

Self Provision in Retirement?

391

models), Canada (DYNACAN and Lifepaths), France (Destinie), Sweden (FASIT, SESIM), the UK (PENSIM and the new SAGE model) and Norway (MOSART). These models are described and compared in Zaidi and Rake (2001). Most of the models have been constructed within the social security departments or the central statistical agencies of these countries — and they are being actively used to help guide policy reforms in pension systems and to provide labour force and retirement income projections. The NATSEM model, DYNAMOD, begins with a confidentialised unit record file taken from the 1986 Australian Census (a 1 per cent sample file). Onto the individuals in this file are imputed additional characteristics (such as more detail on the level and type of education being undertaken, the state in which the person lives, the value of the family home, etc.). The result is a base population of 150,000 synthetic individuals with over 80 characteristics each. The model takes this base population and simulates the events that occur in each individual’s life, from 1986 onwards, by stepping monthby-month through time until the middle of the 21st century. As the individuals progress through life, they experience a range of life events, in line with Australian data about the probabilities of those events happening to real Australians. The life events include death; fertility; couple formation and dissolution; emigration and immigration; primary, secondary and tertiary education; labour force changes (full and part-time employment, unemployment and NILF); disability onset and recovery; earning income; accumulating or diminishing wealth and superannuation; taxation and government cash benefits. The construction of the model is described in King et al. (1999a, 1999b, 2002); Bækgaard (2002a, 2002b); Robinson and Bækgaard (2002) and Abello et al. (2002). The development of the wealth module is described in detail in Kelly (2003). Within DYNAMOD there are five assets that have been modelled to represent the wealth of Australian families. These assets are cash deposits, owner-occupied housing, equities, rental investment properties and superannuation. Kelly (2002) describes in detail how DYNAMOD assigns ownership and values to each of these assets over time.

3. Older Australians in 2001 3.1 Income This analysis of the distribution of income in 2001 is based on a set of tables specially produced by ABS. The tables are from the 2001 Census of Population and Housing and classified by age and income. The available income ranges have been compressed to 13 groups in the analysis below and persons with incomes of not stated or overseas visitor have been excluded. It is worth

Simon Kelly

392 Figure 1:

Weekly Income of Individuals Aged 15 Years and Over, Australia, 2001

30 15-64 year olds

Proportion (%)

65+ years old 20

10

0 <= Zero 1-79 80-159 160199

200299

300399

400499

500599

600699

700799

800999

1000- 1500+ 1499

Total weekly income ($)

Data Source: ABS Census 2001.

noting that the income ranges are not of equal size. Age is divided into two groups: those aged 15–64 years and those aged 65 years and over. The incomes of working age and older Australians are presented in Figure 1. The graph groups all Australians into the two age groups and shows the proportion of each group in each total weekly income range (e.g., approximately 8 per cent of those aged 15–64 have an income of zero or less per week). There is a wide distribution of income for working age Australians. As can be seen in Figure 1, there is a reasonably even spread of working age people across each of the income ranges. The minimum proportion of this age group in any income range is 4.7 per cent ($1–79 per week and $1500+ per week) and the maximum is 9.9 per cent ($200–299 per week). The even distribution of incomes for those of working age contrasts strongly with the income distribution for those aged over 65. The incomes of those aged 65 and over are highly concentrated — with the greatest proportion of older Australians in the $200–299 per week range (34 per cent) and with each of the income ranges above $699 having less than a 2 per cent share. Almost six in ten (58.5 per cent) older Australians have an income of between $160 and $299 per week. Only 17 per cent of Australians of working age have incomes in this range. It is worth observing here that the Government-funded Age Pension in 2001 was $201.00 per week for a single person and $167.75 for a member of a couple. It would seem that the majority of those aged 65 and over live on little more than an income equivalent to the pension.

Self Provision in Retirement? Figure 2: 200l

Private Weekly Income of Individuals Aged 65 Years and Over, Australia, 39.0

40

Proportion of 65+ population (%)

393

35.0

30

20

10 6.9 4.8 1.9

3.7

2.8

1.9

0.9

0

Zero $1-$79 $80$159

0.8

1.0

0.8

0.3

$160- $200- $300- $400- $500- $600- $700- $500- $1000- $1500+ 199 299 399 499 599 699 799 999 1499

Private weekly income ($)

Note: Private income is defined as total weekly income from all sources less current weekly income from government cash pensions and allowances. Data source: ABS 2000/01 Survey of Income and Housing Costs unit record file.

The Census data do not provide information on the source of income, but the level of private income can be obtained from the SIHC. This private income should provide an indication of the capacity of older Australians to provide for them in retirement. In Figure 2, the current weekly income from government cash benefits and allowances has been subtracted from total weekly income from all sources, to give the current weekly private income. Three-quarters of older Australians (74 per cent) have private incomes of less than $80 per week and one-in-three have no private income at all. As noted above, Age Pension in 2001 was $201.00 per week for a single person and $167.75 for a member of a couple. The single rate represents 25 per cent of average weekly earnings. We can assume that this level, say $200 per week, is the minimum adequate income required in retirement. On the basis of this assumption, it can be seen in Figure 2 that only 17.0 per cent of older Australians in 2001 would have been able to provide an adequate income for themselves from their own resources. The self-provision capacity of the current generation of older Australians is severely limited. Only one in six older Australians currently has sufficient private income to provide him or her with an income equivalent to the pension — and recall that the pension is only one quarter of average weekly earnings.

Simon Kelly

394

3.2 Wealth in 2001 Using the techniques described earlier to impute wealth holdings in 2001 onto individuals, estimates can be made of the income and wealth of all Australian adults. The estimated average 2001 income and wealth by age group is shown in Table 1. The table also includes proportion of wealth that is in the form of equity in the family home, as well as the average income and wealth of working age and older people. It can be seen that those aged 45–54 had the highest incomes, with an average of $33,300 — and that this peak was 125 per cent of the overall average of $26,500. The average income of those aged 65–74 was $13,400 (51 per cent of the average), while those aged 75 and over had an income of $14,000 (53 per cent of the overall average). Comparing the average incomes of those aged 15–64 with those aged 65 and over, the average older person has an income less than half that of those of working age. These incomes reinforce the view that older Australians have very low incomes and have relatively little capacity to contribute to their own retirement. Despite the low income of older Australians, they are estimated to have significant wealth. Wealth, as shown in Table 1, grows with age until retirement (age 65–74), where it peaks at an average of $217,400 per person. This peak is 165 per cent of the overall average wealth ($131,600). The average wealth for those aged 75 and over is slightly below the peak at $209,100. Understanding the reasons for this pattern of wealth levels can be quite complicated, because a number of effects — ‘age’, ‘period’ and ‘cohort’ — have combined to produce the current levels of wealth by age. The ‘age’ effect is the growth described above — as age increases, wealth increases until retirement, when some running down of wealth occurs. ‘Period’ effects are those that reflect the conditions prevailing at different times. For example, the returns on investments, taxation, and accessibility of Table 1: Age group 15–24 25–34 35–44 45–54 55–64 65–74 75+ 15–64 65+ Overall

Estimated Average Income and Wealth Per Adult by Age, 2001 Total income ($p.a)

Net wealth ($)

Home equity as % of wealth

19,100 30,300 32,700 33,300 22,500 13,400 14,000 28,900 13,600 26,500

10,800 50,800 104,900 199,000 210,500 217,400 209,100 116,200 213,800 131,600

9 38 55 53 55 73 86 45 79 50

Source: Income — ABS 2000/01 SIHC; Wealth — see text

Self Provision in Retirement?

395

home-ownership vary over time. People’s wealth accumulation will clearly be affected by the conditions prevailing in different stages of their life. Finally, ‘cohort’ effects refer to the way in which cohorts of the population behave differently. Thus, people aged in their 50s in the 1990s may well have behaved differently to people aged in their 50s in the 1950s. And we know that there are many such differences in family formation, labour force activity and so forth. The 2001 levels of wealth appear to clearly show the effect of age but it is very difficult to separate age effects from cohort and period effects in a cross-sectional snapshot like the one shown. Despite not being able to isolate the exact effects on wealth accumulation, we are able to estimate that, based on SIHC-derived data, in 2001 older Australians have almost double the wealth of working age people on average ($213,800 and $116,200, respectively). Examination of the final column in Table 1 highlights the importance of the family home in the average wealth portfolio. On average, when a person is young they have not purchased a house or if they have then it has been financed by a large mortgage, and they are also unlikely to have many other assets. This is reflected in the overall wealth of those aged 15–24 being $10,800 — and only 9 per cent of this being equity in the home. At the other end of the lifecycle, those aged 75 and over have been in the housing property for considerable time and now predominately outright own their home. This exposure to the housing price boom and a negligible mortgage produces the high levels of net wealth shown in Table 1. It is also clear that they have limited wealth outside of their home, as 86 per cent of their wealth is tied up in the family home. It could be argued that older people have simply benefited from a period effect — a housing boom — and apart from this form of unintentional saving they have accumulated very little. The counter argument is that they have very wisely put all of their efforts into paying off the mortgage and now own an asset that is growing at a strong rate. In summary, according to NATSEM estimates, the average older Australian has almost double the wealth of the average working age Australian, but the vast majority of this wealth is in the family home. 3.3 Distribution of Wealth The discussion of wealth above is concerned with the average person: we now turn our attention to the distribution of wealth among these people. In Figure 3 various levels of wealth are shown for those of working age and those aged 65 years and over. The lines represent the proportion of each age group with the various levels of wealth. Examination of the distribution for those aged 15–64 (the shaded line) shows a bimodal distribution — one group of people has very little wealth

Simon Kelly

396 Figure 3:

Estimated Wealth of Individuals Aged 15 Years and Over, Australia, 2001

30

Wealth per person- aged 15-64

Proportion (%)

Wealth per person- aged 65+ 20

10

0 <0

Zero

1-

25-

50-

75-

100- 200- 300- 400- 500- 600- 800- 1000- 1200- 1400- 1600+

24

49

74

99

199

299

399

499

599

799

999 1199 1399 1599

Wealth per person ($'000s)

Note: Wealth is defined as the sum of the values of housing, shares, cash deposits, rental properties and superannuation less mortgages and rental investment loans. Data source: See text.

($1,000–24,000) while another has wealth in the range of $100,000–299,000. This distribution, combined with the information on wealth and home equity by age in the previous table, seems to suggest that the first group consists of both young people and non-homeowners. The second group would be predominately older homeowners without a mortgage or those with only a small mortgage. The 65+ line (the dark solid line) reinforces some of the findings from the table above. As most of the wealth of this age group is represented by home equity and most are outright home owners, it is not surprising that there is a concentration of older people around the $100,000–299,000 level. The higher proportion of those age 65+ in this wealth band than those aged 15–64 and the alignment of the spikes seem to confirm that the spike is due to home equity. Comparison of the 15–64 line with the 65+ line shows some interesting differences. One difference is the higher proportion of people aged 65+ with zero assets compared with 15–64 year olds. Initially this is surprising, given the strong relationship between age and wealth accumulation — but it does seem reasonable when the superannuation circumstances of each are considered. Almost no young people will have zero wealth because of the Superannuation Guarantee (SG). The SG was introduced in 1992 for all employees (with some very limited exceptions) and legislates that an

Self Provision in Retirement?

397

employer must contribute a given proportion of a person’s earnings to superannuation. The accumulated superannuation cannot be withdrawn until at least age 55. The impact of the SG on wealth is that for those aged less than 55, if they have ever been employed since 1992, they will have a level of wealth above zero — albeit wealth they cannot access until their later years. People aged 65 and over may either have not worked under the SG or may have drawn down their superannuation since retirement. Either of these circumstances could lead to older people having zero wealth. The other significant difference between the two distributions is the higher proportion of older people with considerable levels of wealth. Most working age people are still in the accumulation phase of the wealth cycle. This is evident in the higher proportions of working age people than older people in the $1,000–99,000 wealth ranges. For older people it seems greater numbers have already accumulated a high level of wealth and this produces higher proportions in the $200,000–799,000 wealth range. In 2001, the dominant form of wealth for older people is home equity. This dominance produces a large proportion having wealth in the $100,000–299,000 range. The distribution of wealth is not as concentrated as income is for this age group. There are larger proportions with zero wealth and considerable wealth, when compared with the wealth of working age Australians.

4. Comparison of Wealth Estimates The previous section provided estimates of the distribution of income and wealth in Australia in 2001. In the next section, the distribution of wealth 30 years later will be estimated. These 2031 estimates are produced using NATSEM’s DYNAMOD dynamic microsimulation model, discussed earlier in the chapter. Before looking at the findings from this modelling, it is interesting to compare the simulated wealth distribution from DYNAMOD in 2001 with that produced from the SIHC data. The simulated distribution of wealth in 2001 is compared with the SIHC-based estimates in Figures 4 and 5. As both the simulation and the SIHC-based estimates are for the same point in time, there should be good alignment between the two distributions. The alignment is not expected to be perfect for a number of reasons — the benchmark SIHC data has limitations in regards to underreporting and omissions; the SIHC imputation techniques make a number of assumptions which impact on estimates; the behaviour in the simulation cannot completely capture actual behaviour; and a number of the simulation assumptions impact on the estimates. Examination of the graphs of the 15–64 wealth distributions (Figure 4) show there are differences but there is general agreement between the

Simon Kelly

398

Figure 4: Simulated and SIHC-Based Estimates of the Wealth of Individuals Aged 15–64 Years, 2001 30 SIHC-based wealth per person aged 15-64

Proportion (%)

Simulated wealth per person aged 15-64 20

10

0 <0

Zero

1-

25-

50-

24

49

74

75- 100- 200- 300- 400- 500- 600- 800- 1000- 1200- 1400- 1600+ 99

199

299 399 499 599 799 999

1199 1399 1599

Estimated wealth per person ($'000s)

Figure 5: Simulated and SIHC-Based Estimates of the Wealth of Individuals Aged 65 and Over, 2001 30 SIHC-based wealth per person aged 65+

Proportion (%)

Simulated wealth per person aged 65+ 20

10

0 <0

Zero

1-

25-

50-

75-

100-

200- 300- 400- 500- 600- 800- 1000- 1200- 1400- 1600+

24

49

74

99

199

299

399

499

599

799

999 1199

Estimated wealth per person ($'000s)

Data Source: For Figures. 4 and 5 see text.

1399 1599

Self Provision in Retirement?

399

approaches. The bimodal nature of the 15–64 SIHC-based wealth distribution has been repeated in the simulation estimates, with the second peak being repeated at exactly the same wealth range and to the same height. In overall terms the simulated data has more people with negative and zero wealth, fewer with moderate wealth ($1,000–99,000), and slightly more with wealth in the $200,000–499,000 range than the estimates based on SIHC. The distributions of wealth for those aged 65 and over are shown in Figure 5. Alignment between the SIHC and the simulated data is very good above $75,000 but, below this wealth level, it seems that older people reporting low wealth in the SIHC (the shaded line) are being simulated as having zero wealth in DYNAMOD (the solid line). This discrepancy is not considered significant in regards to qualification for a public pension as the pension means test allows people to have around $100,000 of assets before the rate of pension is tapered. In other words, both zero and low wealth older Australians qualify for the maximum rate public pension. Alignment by age is generally good, except for those aged 75 and over. As shown in Table 2, the simulated average wealth in DYNAMOD for people aged 75 and over is estimated at $138,900, while the SIHC-based estimate is significantly higher at $209,100. The reason the simulated data is less than SIHC appears to be related to the higher proportion of older Australians that have zero wealth, as discussed earlier. The reason for the higher proportion with zero wealth could relate to incorrect imputation of housing ownership levels in the simulation; a wider distribution of superannuation assets than that assigned by the SIHC; or the simulation drawing down wealth faster in retirement than the SIHC data suggests is the actual case. Further research into the exact reason for the discrepancy will be undertaken in the future. At this stage it is simply worth noting here that the simulated data appears to be underestimating the wealth held by older Australians. Table 2:

Estimated Average Wealth Per Person by Age, 2001

Age group 15–24 25–34 35–44 45–54 55–64 65–74 75+ Overall Source: See text

SIHC-based estimated wealth ($)

DYNAMOD simulated estimated wealth ($)

4,600 38,400 117,300 209,600 245,600 200,100 138,900 131,200

10,800 50,800 104,900 199,000 210,500 217,400 209,100 131,600

400

Simon Kelly

There is general concordance between the SIHC-based estimates of the distribution of wealth and the simulated wealth distribution, with the exception of those aged 75 and over. At these older ages the simulation is producing lower estimates of average wealth. Generally, however, the level of agreement provides us with sufficient credibility to say that if our assumptions about future behaviour are correct, then the wealth projections are likely to be reasonably accurate in terms of distribution and quantum.

5. Older Australians in 2031 The methodology for simulating the accumulation and distribution of wealth was briefly discussed earlier in the chapter. We now use the simulation to make some projections of wealth distribution in 2031. Clearly the projected asset values are limited to the underlying assumptions being reasonably accurate. The importance of the underlying assumptions and the complexity of alignment are a significant issue in dynamic modelling and their impact cannot be overstated. Some of the issues surrounding alignment of DYNAMOD and the innovative approaches used to ensure that the model responds correctly to the alignment are discussed in Kelly and King (2001). 5.1 Income Unless changes have been made to the eligibility age for the pension, or the income thresholds have been increased in real terms, it seems probable that the primary sources of income for older Australians will remain the public pension and private investment income. The most significant difference will be the contribution of private investment income to total income of older people. The dominant source of this private investment income is expected to be the income coming from the increased levels of accumulated superannuation, relative to Australia today. 5.2 Wealth Using the microsimulation model developed at NATSEM, estimates of wealth in 2031 have been projected. The estimates are based on the latest version of DYNAMOD — version 3.5 — and are slightly different to estimates provided using older versions of the model. The new version incorporates actual investment data and performance up to June 2003, whereas older versions only contained actual data up to June 1999. Enhancements have also been made to the handling of superannuation and living costs in retirement. Table 3 shows the estimated wealth by age group in 2001 and

Self Provision in Retirement?

401

Table 3: Estimated Average Personal Wealth and Share of Total Wealth by Age Group, 2001 and 2031 ($2001) Age group

15–24 25–34 35–44 45–54 55–64 65–74 75+ 15–64 65+ Overall

2001

2031

Average wealth ($)

Share of total wealth (%)

Average wealth ($)

Share of total wealth (%)

4,600 38,400 117,300 209,600 245,600 200,100 138,900 122,900 172,200 131,200

0.4 5.9 18.0 29.5 24.1 13.9 8.1 78.0 22.0

6,700 51,500 175,100 363,000 597,900 610,100 626,000 262,100 618,000 359,700

0.2 2.2 7.8 16.1 26.7 23.6 23.5 53.0 47.0

Source: NATSEM simulation

2031. The estimates for both 2001 and 2031 are expressed in 2001 dollars. The table also shows the proportion of the total ‘wealth pie’ held by each age group. The wealth of the average Australian will increase during the 30 years from 2001 to 2031. Average personal wealth (after taking out inflation) will increase from an estimated $131,200 in 2001 to $359,700 in 2031. Despite this strong average growth, some age groups are estimated to do much better than others. For example, the wealth of those aged 75 and over is projected to increase more than 3.5 times from $172,200 to $618,000, while the average wealth of those aged 25–34 will only increase by $13,100 or 34 per cent. The differential growth rates, combined with an ageing population will result in a significant change in the proportion of wealth held by older Australians. In Table 3 the share of the ‘wealth pie’ held by various age groups is shown. The share held by those aged 65 and over increases from 22 per cent in 2001 to 47 per cent in 2031. Remembering that these estimates are based on simulated results that tend to underestimate the wealth held by those aged 75 and over, this implies that the share held by older Australians could be greater than half by 2031. 5.3 Distribution of Wealth Among Older Australians in 2031 In Figure 6 the 2031 distribution of wealth is plotted for older Australians. The 2001 distribution (shaded line) is also included for comparative

Simon Kelly

402

Figure 6: Simulated and SIHC-Based Estimates of the Wealth of Individuals Aged 65 and Over, 2001 30 Simulated wealth per person aged 65+ in 2001

Proportion (%)

Simulated wealth per person aged 65+ in 2031 20

10

0

<0

Zero 124

25-

50-

75-

100- 200- 300- 400- 500- 600- 800- 1000- 1200- 1400- 1600+

49

74

99

199

299 399 499 599 799 999 1199 1399 1599

Estimated wealth per person ($'000s)

Data Source: NATSEM simulation.

purposes. Examination of the 2031 distribution for those aged 15–64 still shows that a large proportion will have zero wealth, but that proportion is lower than it was in 2001. By 2031, it can be expected that few people will not have accumulated some superannuation before age 65 and so this high proportion with zero would suggest that they have drawn down their superannuation to zero after retirement. While there is an improvement in the situation of most people over the 30 years, some are still in a very poor economic state. The second feature of this distribution is the lower concentration of people in one specific wealth range. In 2001 27 per cent of older people had estimated wealth of between $100,000 and $199,000 but in 2031 there are estimated to be no wealth ranges with more than 12 per cent of the group within them (other than zero, with 16 per cent). The clear bimodal distribution has been replaced with a more even distribution of wealth across the ranges. This seems to indicate that the home will be a less dominant source of wealth for older Australians and that superannuation, cash deposits, shares and rental properties will be found more often in the wealth portfolios of those aged 65 and over. Another significant feature of the 2031 distribution is the increased proportions of older people with considerable levels of wealth. There are very significant forecast increases in the proportions of people having $300,000 or more in wealth. The increased levels of wealth and the form in which it is

Self Provision in Retirement?

403

held (that it is not entirely in the family home) should provide these people with some capacity to provide for themselves in retirement.

6. Conclusions In the introduction to this chapter, the impact of the ageing population on future fiscal budgets and their ability to financially provide for themselves during retirement were noted. Information on income and wealth are required to properly examine their capacity but information on wealth is not readily available. This chapter attempts to address this by providing estimates of wealth in 2001 and 2031. NATSEM has assigned 2001 valuations to assets based on a variety of sources. The primary source of data for estimating and imputing 2001 wealth is the 2000-01 ABS SIHC. Information on the current value of the family home and the level of mortgage are self-reported on the SIHC. The SIHC also contains information on the level of current investment income being received by individuals. This investment income information allows us to impute the accumulated value of cash deposits and equities (from which those income streams are being generated). Rental property values and equity are imputed using a range of income and ownership variables on the SIHC and a regression equation derived from information on the 1997 ABS Rental Investors Survey. The final remaining asset that is included in our definition of wealth is the 2001 value of funds accumulated in pension plans (called superannuation in Australia). Dynamic microsimulation modelling is used to project the level and distribution of wealth in 2031. The NATSEM model, DYNAMOD, provides estimates of the level of assets held by individuals at all points in their life, including future points. Examination of this wealth provides an estimate of their private savings and allows estimation of private income in retirement. It is not generally possible to get this level of detail with other modelling techniques. In general, economic well-being is often measured by the observation of income alone. Based on this narrow measure, the current generation of older Australians appear to have very little capacity to provide for themselves in retirement or even to contribute to the cost of the retirement. Only 13 per cent have private income equal to or greater than the $201 per week provided by the public pension. One-third have no private income at all and three-quarters of those aged 65 and over in 2001 have private income of less than $80 per week. A wider definition of economic well-being, one that considers both wealth and income, provides a somewhat different view. Using wealth estimates derived from the SIHC and other sources, NATSEM has estimated that in 2001 older Australians have almost double the average wealth of working

404

Simon Kelly

age people ($213,800 and $116,200, respectively). However, the majority of this wealth appears to be home equity and hence is not producing income to supplement or replace the public pension. Using DYNAMOD, the simulated distribution of wealth of older Australians in 2031 has been projected. Average wealth for all age groups is projected to increase but the share of wealth held by those aged 65 and over increases significantly. According to the simulation, in 2001, 22 per cent of wealth was held by those aged 65 and over. By 2031, this is projected to have increased to 47 per cent. Almost half of all personal wealth is in the hands of older people. The other feature of this estimated 2031 wealth is the diminished dominance of home equity. The inclusion of other assets in the 2031 wealth portfolio, such as superannuation and cash deposits, should provide greater scope for self-provision in retirement. Currently, older people have little capacity to contribute to the costs associated with ageing — most have little other than their home and entitlement to the Australian public age pension. In 30 years time the picture is quite different, with average wealth projected to have increased and to be spread across a more diverse range of asset classes. These other assets should be able to provide income to supplement and, in some cases replace, the public pension. In 2031, older Australians will be better placed for selfprovision and may be able to ease the burden on the public purse.

References Abello, A., Kelly, S. and King, A. (2002). Demographic Projections with DYNAMOD-2. Technical Paper no. 21, National Centre for Social and Economic Modelling, University of Canberra (available from www.natsem.Canberra.edu.au). ABS (Australian Bureau of Statistics) (2003). Population by Age and Sex, Australian States and Territories. ABS Catalogue no. 3201.0, Canberra (various issues). ABS (Australian Bureau of Statistics) (2003). Population Projections Australia 2002 to 2101. ABS Catalogue no. 3222.0, Canberra, September. Bækgaard, H. (2002a). Modelling the Dynamics of the Distribution of Earned Income. Technical Paper no. 24, National Centre for Social and Economic Modelling, University of Canberra (available from www.natsem.Canberra.edu.au). Bækgaard, H. (2002b). Micro– Macro Linkage and the Alignment of Transition Processes: Some Issues, Techniques and Examples. Technical Paper no. 25, National Centre for Social and Economic Modelling, University of Canberra (available from www.natsem.Canberra.edu.au). Commonwealth Treasurer (2002). Intergenerational Report. 2002/03 Budget Paper no. 5, Commonwealth Treasury of Australia, Ausinfo, Canberra, April. Disney, R. and Johnson, P. (2001). Pension Systems and Retirement Incomes across OECD Countries. Edward Elgar, Cheltenham, UK.

Self Provision in Retirement?

405

FaCS (Department of Family and Community Services) (2001). Income Support and Related Statistics: A 10-year Compendium, 1989– 1999. Occasional Paper Number 1, FaCS, Canberra, p. 135. Kelly, S. (2001). Trends in Australian Wealth — New Estimates for the 1990s. Paper presented to the 30th Annual Conference of Economists, University of Western Australia, September (available from www.natsem.canberra.edu.au). Kelly, S. (2002). Simulating Future Trends in Wealth Inequality. Paper presented to the 2002 Annual Conference of Economists, Adelaide, South Australia, October (available from www.natsem.Canberra.edu.au). Kelly, S. (2003). Estimating the Wealth of Australians: A New Approach Using Microsimulation. Ph.D. thesis, University of Canberra. Kelly, S., Farbotko, C. and Harding, A. (2004). The Lump Sum: Here Today, Gone Tomorrow. AMP. NATSEM Income and Wealth Report, Issue no. 7, March (available from www.amp.com.au/ampnatsemreports). Kelly, S. and Harding, A. (2004). Funding the Retirement of the Baby Boomers. Agenda, 11(2), 99–112. Kelly, S. and King, A. (2001). Australians Over the Coming 50 Years: Providing Useful Projections. Brazilian Electronic Journal of Economics, 4(2), 1–23. Kelly, S., Percival, R. and Harding, A. (2002). Women and Superannuation in the 21st Century: Poverty or Plenty? Competing Visions: Proceedings of the National Social Policy Conference, SPRC Report 1/02, University of New South Wales, pp. 223–249. Khan, Q. (1999). Australia’s Retirement Income System: An Example of Sustainable Cost-effective Coverage in Productivity Commission and Melbourne Institute of Applied Economic and Social Research. Policy Implications of the Ageing of Australia’s Population, AusInfo, Canberra, 141–153. King, A., Baekgaard, H. and Robinson, M. (1999a). DYNAMOD-2: An Overview. Technical Paper no. 19, National Centre for Social and Economic Modelling, University of Canberra (available from www.natsem.Canberra.edu.au). King, A., Baekgaard, H. and Robinson, M. (1999b). The Base Data for DYNAMOD2. Technical Paper no. 20, National Centre for Social and Economic Modelling, University of Canberra (available from www.natsem.Canberra.edu.au). King, A., Baekgaard, H. and Harding, A. (2001). Australian Retirement Incomes, in Disney, R. and Johnson, P. (eds), Pension Systems and Retirement Incomes across OECD countries, Edward Elgar, Cheltenham, UK, pp. 48–91. King, A, Walker, A. and Bækgaard, H. (2002). Modelling Overseas Migration in DYNAMOD-2. Technical Paper no. 22, National Centre for Social and Economic Modelling, University of Canberra (available from www.natsem.Canberra. edu.au). OECD (2000). The Costs of an Ageing Society. OECD in Washington no. 22, OECD Washington Center, Washington DC, November/December. Robinson, M. and Bækgaard, H. (2002); Modelling Students in DYNAMOD-2. Technical Paper no. 23, National Centre for Social and Economic Modelling, University of Canberra (available from www.natsem.Canberra.edu.au). Zaidi, A. and Rake, K. (2001). Dynamic Microsimulation Models: A Review and Some Lessons for SAGE. SAGE Discussion Paper no. 2, ESRC SAGE Research Group, London School of Economics, United Kingdom, March.

This page intentionally left blank

Chapter 16

Household Net Wealth in New Zealand: Implications for Retirement Incomes Grant M. Scobiea, and John Gibsonb a b

New Zealand Treasury, New Zealand The University of Waikato, New Zealand

Abstract This chapter uses a new survey of household net wealth in New Zealand to analyse the determinants of net wealth, especially the net wealth of older age groups. Implications are drawn regarding the adequacy of household retirement incomes. The savings rate needed to sustain pre-retirement consumption levels is modelled and compared to actual saving rates from the Household Economic Survey. The results suggest that low income New Zealanders do not currently need to save, because the public pension scheme, New Zealand Superannuation, provides sufficient income in retirement to maintain preretirement levels of consumption.

1. Introduction There are few countries, especially among the OECD, where retirement issues are not at the forefront of public debate. Is the public system for the provision of pensions sustainable? Are individuals saving enough for retirement? Will they have choices to remain in the labour market? There are thus a range of public policy issues that might be better formulated and evaluated if information were available on the assets and liabilities of households. Until now, no such information has been available in

Corresponding author.

E-mail address: [email protected] (G.M. Scobie). International Symposia in Economic Theory and Econometrics, Vol. 15 Copyright r 2007 Elsevier B.V. All rights reserved ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15016-4

408

Grant M. Scobie and John Gibson

New Zealand.1 However, in 2001 Statistics New Zealand undertook the Household Savings Survey (HSS) (Statistics New Zealand, 2002a, 2002b). This paper has two goals. The first is to report the patterns of asset accumulation for retirement found in the survey. The second is to analyse the implications for retirement income and savings, based on the projected levels of retirement wealth. An ancillary question is whether New Zealanders are saving adequately for their retirement. There are limitations to the available data, so we reach only a tentative conclusion related to adequacy. The chapter is organised as follows. In Section 2, we give a very brief synopsis of the HSS (the microdata used in our analysis). In Section 3, we summarise the analytical framework. Section 4, discusses the results and ends with a discussion of the adequacy of retirement savings. Section 5 presents the conclusions, and outlines unfinished business.

2. The Data The HSS survey covered those over 18 years old living in private dwellings and usually resident in New Zealand. For the core sample a total of 6,600 households were approached. One person from those qualifying in the household was chosen at random — and information was collected from and about that individual. In the case where that person had a partner, information was collected for the couple (i.e., where the respondent and his/ her partner were living in the same household, the couple were interviewed as a single unit). It should be stressed that because of the special sampling procedure, the results cannot be interpreted as pertaining to households. For example, a 20-year-old student living at home with two parents might have been the unpartnered individual drawn in the sample and for whom the information was collected. No information about the remainder of the household in this case would have been included. To improve the accuracy of estimates for Ma¯ ori, a booster sample was used. In total the response rate was 74 per cent and the final number in the sample was 5,374 households. There were 2,392 individual interviews and 2,982 for couples. It is important to stress that the term household refers to the unit of selection. The results are for individuals (living as individuals or partnered) and not for households or families. The survey covers a population of 930,900 unpartnered individuals and 1,711,800 individuals in couples — or a total of 2,642,700 people aged 18 and more. 1

Estimates of household wealth at the aggregate level have been made (Claus and Scobie 2001; Thorp and Ung, 2000, 2001) but no data at the unit record level had ever been assembled in a national survey.

Household Net Wealth in New Zealand

409

3. The Model Any attempt to assess how adequately the New Zealanders are preparing for retirement through saving must immediately confront the question: How is ‘‘adequacy’’ to be measured? By what criteria would we assess savings and the associated level of wealth accumulation for retirement to be adequate? We have taken as a measure of adequacy the capacity to sustain an equal level of consumption before and after retirement — that is, we invoke consumption smoothing as the aim of retirement saving. We estimate the saving rates that would be required to build present wealth (observed in the survey) up to the level that would generate a post-retirement income adequate to sustain pre-retirement consumption. In addition, we derive estimates of the income in retirement that is implied by the estimated replacement rates. We then analyse the distribution of the predicted retirement incomes and calculate how many people would have incomes in retirement below 60 per cent of the median income of that cohort (i.e., a relative poverty-line approach). The basic model we adopt follows Moore and Mitchell (1997). This approach computes jointly the prescribed saving and income replacement rates for each household in the survey. A graphical illustration is given in Figure 1. At the current time, a household has a net worth (depicted as Wa) measured in the HSS. This is projected to grow to an amount denoted Wp by the time they reach a pre-determined retirement age (here we assume 62, 65 or 68). To have a given level of income in retirement they would need to have accumulated retirement wealth depicted in Figure 1 as the stock, Wr. Part of their retirement income is provided by the universal public pension New Zealand Superannuation (NZS)2 and the stock or wealth at retirement equivalent to that flow of income is incorporated in Wr and Wp. The difference between the required wealth Wr and the projected wealth Wp is labelled as the shortfall — and is the amount that would need to be accumulated between now and retirement to add to the projected stock, and hence support an income in retirement of level Yr.3 This additional amount, in the absence of inheritances or unanticipated windfall gains or losses in 2

New Zealand Superannuation (NZS) is a universal public pension payable without income or asset testing subject only to a residency requirement. It is paid at a rate of 66 per cent of the average wage for a married couple and is taxable. 3 Projections of retirement wealth were made on the basis of assumed growth rates defined as after-tax real rates of return for different classes of assets. Net housing wealth was assumed to remain constant in real terms (i.e., no capital gains). Business and financial assets were assumed to grow at 2 per cent per annum, while other classes such as farms, property and vehicles were assumed to maintain constant real value. The projected levels of wealth make no allowance for any additional saving that might occur between now and retirement age.

Grant M. Scobie and John Gibson

410

Figure 1: Stylised View of Stocks and Flows of Income, Savings and Retirement Wealth in a Model of the Joint Determination of Saving and Replacement Rates (a) Stocks

$ Shortfall to be met from savings

wr Convert to stock of retirement wealth

Project growth of assets

wp

wa

From private saving

Yr NZS Retirement Age

Current Age Working Life

Life Expectancy at Retirement

(b) Flows Tp = Taxes Pre-retirement income (Yp)

S = Savings Yp - Tp - S = Pre-retirement Consumption

Yp Tp Yr Tr

Age

= Pre-retirement income = Pre-retirement taxes = Pre-retirement income = Post-retirement taxes

Yr - Tr = Post-retirement Consumption

Wa = Wealth at current age Wp = Projected wealth at retirement Wr = Wealth at retirement needed to supply a Yr S = Savings

asset values, would need to be accumulated through savings. These flows are depicted in Figure 1(b). The approach assumes that some fixed share s of pre-retirement income will be saved (s ¼ S/Yp) and the replacement rate (R) is given by the ratio of gross income in retirement to gross income pre-retirement (i.e., R ¼ Yr/Yp). Under the New Zealand taxation system of TTE,4 post-retirement taxes (denoted as Tr) are assumed to be zero, so real after-tax consumption is equal to total pre-retirement income. Clearly, some values of retirement

4 TTE refers to a system where the savings are made from after-tax income, the returns are taxed and the withdrawals are exempt. It differs from those systems which exempt savings or earnings from taxation and tax withdrawals (TET, ETT or EET).

Household Net Wealth in New Zealand

411

income could imply a substantial shortfall in retirement wealth, which might in turn require unrealistic or unfeasible levels of saving pre-retirement. It is for this reason that the saving and replacement rates are jointly determined. A number of additional factors arise which are not depicted in Figure 1. Instead of a constant pre-retirement income, we assume that income grows from its actual level (as observed in the survey) by a fixed annual growth rate of 1 per cent, chosen to approximate the average annual rate of labour productivity and real wage growth in the economy. The gross income at retirement (Yp) is then based on the observed actual earnings plus a compound growth of 1 per cent annually.5 Pre-retirement tax rates are based on pre-retirement real income (Yp). NZS payments are assumed to be constant in real terms. A formal statement of the model is given in the appendix. From a theoretical perspective, the model of joint determination of saving and replacement in Section 2 has a potential shortcoming: it specifically is not built on a behavioural model of individual utility maximisation. Models based on utility maximisation, while theoretically more appealing, can become extremely complex in their implementation. Furthermore, they are reliant on estimates of key parameters that describe consumption and saving behaviour, and these are seldom, if ever, known with certainty. In this study, we have relied on the admittedly more pragmatic approach of a model that is more akin to a financial planning approach, rather than one having a theoretical utility-maximising foundation. However, to provide a comparison, and in part to test the difference with our basic model, we estimated a simple version of a utility-maximising model, albeit without uncertainty. We obtained results which were very comparable to those reported here, suggesting that the approach taken here is a reasonable approximation to that based on a utility-maximisation framework (which would equate the marginal utility of consumption across periods rather than the level of consumption as done here.6

5

An alternative approach would have been to estimate age earnings profiles for the survey. However, with a single cross section as in the HSS one cannot isolate cohort effects, as these would have been compounded into earnings estimates. There are a number of individuals in the sample who report negative or very low incomes. These reported incomes could include a significant transitory component, such as a temporarily low income due to redundancy or losses in an unincorporated business. Some estimate of consumption is often used in such cases as a better proxy for permanent income. In this study we use the unemployment benefit rate as an estimate of a minimum consumption level for those reporting negative incomes or income below the benefit rate. 6 Details are given in Scobie and Gibson (2003).

Grant M. Scobie and John Gibson

412

4. Results 4.1 Retirement Wealth We define four categories of wealth: housing, financial, pension and NZS. Housing wealth is the current net worth of the primary residence. Financial wealth includes farms, businesses, other property (holiday homes, rental property, commercial and overseas property) together with life insurance, bank deposits, positive credit-card balances, shares and managed funds, money owed, motor vehicles, cash, collectibles and other assets, less any financial liabilities (credit card or hire purchase debt, student loans, etc.). Pension wealth is based on the holdings in personal superannuation schemes, defined contribution schemes and defined benefit schemes. In the first two cases, the value was that which the respondent would receive if the scheme were cashed up on the day of the interview. In the third case, the estimate of value was provided by the government actuary. Assets in Ma¯ ori trusts and incorporations were excluded from the analysis, as it is not clear that any individual reporting net Maori assets could annuitise those for retirement income. The wealth represented by NZS is computed as the present value of the stream of payments, accounting for differences in life expectancies. The composition of wealth for unpartnered individuals is shown in Figure 2. For the lowest four deciles, NZS represents almost the entire Figure 2: Level and Composition of Mean Current Wealth-by-Wealth Decile, Unpartnered Respondents Aged 45–55 ($NZ) 1,000,000 900,000 800,000 700,000 600,000 500,000 400,000 300,000 200,000 100,000 0 -100,000

1

2

3

4

5

6

7

8

9

10

Wealth Deciles Housing Wealth

Financial Wealth

Pension Wealth

NZ Superannuation

Household Net Wealth in New Zealand

413

amount of retirement wealth, dropping to 15 per cent for the wealthiest decile. For the median 10 per cent of individuals, it represents about one-half of retirement wealth. NZS wealth increases with total wealth (i.e., the absolute value of NZS is greatest for the wealthier individuals, as their life expectancies are typically greater). Similarly, the current value of NZS is greater for women than men. For Ma¯ ori and Pacific Islanders, NZS comprises 58 per cent of retirement wealth, compared to 40 per cent for all Pakeha respondents. In the case of couples aged 45–55, NZS represents over 90 per cent of total retirement wealth in the lowest wealth decile. This falls to under 15 per cent for those in the highest wealth decile. For the whole cohort, the median share of NZS in total retirement wealth is 45 per cent. The contribution of NZS to reducing the inequality of retirement wealth amongst couples is illustrated in Figure 3. The lowest 30 per cent of couples have 4 per cent of the total wealth excluding NZS, but their share rises to 13 per cent when NZS is included. The lowest 70 per cent have 30 per cent of the total retirement wealth, rising to 44 per cent when NZS is included. The top 10 per cent have 40 per cent of the total wealth excluding NZS, but their share falls to 29 per cent when NZS is included.

Figure 3: Distribution of Retirement Wealth for Couples Aged 45–55 with and without New Zealand Superannuation 100% 90%

Share of Wealth

80% 70% 60% 50% 40% 30% 20% 10% 0% 0%

10%

Mean Total Wealth

20%

30%

40% 50% 60% 70% Share of the Population

80%

Mean Total Wealth Excluding NZS

90% 100%

Series 3

Grant M. Scobie and John Gibson

414

Figure 4: Prescribed Median Saving Rates for Unpartnered Individuals Retiring at Age 65 with no Consumption of Housing, by Wealth and Income Deciles 20.0% 10.0% 0.0%

1

2

3

4

5

6

7

8

9

10

-10.0% -20.0% -30.0% -40.0% Wealth

Income

4.2 Prescribed Saving Rates This section presents the results of applying the model of joint determination of replacement rates and saving rates. The prescribed saving rates for unpartnered individuals are summarised in Figure 4. The wealthiest two deciles have negative-prescribed saving rates.7 The implication is that they have projected levels of retirement wealth that are more than adequate to smooth consumption, and require no further savings. A similar pattern of negative savings is found among low-income individuals. In this case, this is due to the fact that those with very low current incomes face an increase in income at retirement, once they become eligible for NZS. As such they have no incentive to reduce current consumption further and accumulate savings for a period when they expect to have a higher level of consumption. Bernheim et al. (2000) report similar findings for the USA: ‘‘The fact that the recommended saving rate is close to zero for the low income group and that the rate rises with income is not surprising. Most of the low-income households will receive the majority of their 7

F. Scott Fitzgerald to Ernest Hemingway: ‘‘The very rich are different from you and me’’, to which Hemingway replied: ‘‘Yes. They have more money’’. Cited in Bartlett (1980)

Household Net Wealth in New Zealand

415

Table 1: Mean and Median Prescribed Saving and Replacement Rates by Income Decile with no Consumption of Housing: Couples Aged 45–55 Retiring at Age 65 Income decile

1 2 3 4 5 6 7 8 9 10 Total sample Median 10% Ethnic sub-groups Pakeha Ma¯ ori-Pacific

Mean projected income at retirement

20,589 36,634 47,066 57,766 65,996 76,445 88,341 98,955 120,610 253,674

Saving rate

Replacement rate

Mean (%)

Median (%)

Mean (%)

Median (%)

62.0 6.6 14.0 18.9 17.6 21.8 12.9 14.6 14.5 17.5 6.3

13.0 10.9 18.8 23.8 24.9 22.1 16.9 17.6 17.6 19.5 18.1 12.8

141.3 85.9 65.3 60.4 61.7 50.6 53.0 51.2 48.6 42.3 66.1

92.3 68.4 60.5 56.3 54.4 49.2 49.3 48.2 44.4 40.2 54.5 58.7

4.8 11.3

18.0 17.4

66.7 65.3

53.9 60.4

Note: All values are weighted to population averages and are in 2001 dollars. The values for the median 10 per cent refer to the means for those respondents between the 45th and 55th percentiles of the (weighted) distribution of total wealth.

post-retirement incomes from Social Security. And the higher the level of income, the smaller the fraction of pre-retirement income being replaced by Social Security’’. The results by income decile for couples are given in Table 1, based on the assumption that none of the equity in the primary residence would be drawn down to provide retirement income. Rather, all remaining equity at the time of death of the surviving partner would be bequeathed. It will be noted in Table 1 that a significant share of those on low incomes has negative savings rates. This should be interpreted as not requiring further savings to smooth their lifetime consumption, rather than implying they should run down their existing stocks of wealth. To explore this further — and to see the extent of these negative prescribed saving rates — we analysed the next age group, aged 56–64. These results are shown in Table 2. Nearly one-half of couples and virtually all non-partnered individuals in the lower 40 per cent of the income distribution would not need to save further to sustain their level of consumption in retirement. To the extent that we know that some do have positive saving rates, then we can presume

416

Grant M. Scobie and John Gibson

Table 2: Percentage of Population Aged 56–64 for whom the Prescribed Saving Rate is Negative Income quintile 1 (lowest) 2 3 4 5 (highest) Total

Couples

Non-partnered individuals

65 22 8 17 16 32

100 85 20 31 40 66

that this reflects some combination of precautionary saving for contingencies and the desire to make bequests beyond the present equity in their house. 4.3 Distribution of Retirement Incomes While the means and medians give an overall picture, they do not reveal the full extent of the underlying heterogeneity across individuals. We therefore now consider the distribution of predicted retirement income.8 In doing this, we use one possible indicator of the adequacy of predicted retirement incomes. We have chosen the cut-off as below 60 per cent of the median predicted retirement income. This proportion of the median is sometimes used as an indicator of a poverty level. In all, about one-seventh of unpartnered individuals have predicted retirement incomes based on their projected retirement wealth that fall below 60 per cent of the median for all those aged 45–55. This is illustrated in Figure 5. The results in Table 3 show the characteristics of those above and below 60 per cent of the median predicted retirement income. Given that this group is from a narrow age band then, perhaps unsurprisingly, there is no difference in age between those below and those at or above 60 per cent of median predicted income. Likewise, there is no difference in gender, and there is only a modest difference in years of post-primary education. However, there are large differences in ethnicity, income and the probability of having ever received an inheritance. While illuminating, the results in Table 3 do not allow for measuring the effect of any one variable while holding the others constant. To do this, we have estimated a probit model, in which the dependent variable is the 8

Retirement income Yr is predicted from the estimated replacement rate and preretirement income (Yr ¼ Yp  R).

Household Net Wealth in New Zealand

417

Figure 5: Distribution of Predicted Retirement Incomes for Unpartnered Individuals Aged 45–55, Retiring at Age 65 with no Consumption of Housing Wealth

14.6% fall below 60% of the median

NZS $12,420

60% Median Median $13,061 $21,768

Mean $25,412

Predicted Retirement Income

Table 3: Characteristics of Unpartnered Respondents by Predicted Retirement Income (Retirement at Age 65 with no Consumption of Housing) Characteristic

Age Years of educationa Income Net worth Total wealth Proportion who are Male Pakeha Ma¯ ori or Pacific Islanders Those inheriting 4$10,000 Homeowners a

Predicted retirement incomes o60% of median

Z60% of median

All respondents aged 45–55

49.5 4.8 10,412 7,365 129,712

49.7 5.7 37,500 210,710 342,722

49.7 5.5 33,840 183,235 313,941

0.41 0.54 0.46 0.10 0.17

0.41 0.77 0.18 0.22 0.57

0.41 0.74 0.22 0.20 0.52

Post-primary school.

probability of falling below 60 per cent of the median retirement income. The aim is to determine which characteristics are associated with a greater likelihood that an individual’s predicted retirement income will fall below this retirement income ‘‘poverty line’’. The probit model of the probability of low predicted retirement income has as explanatory variables: income, years of education, gender, ethnicity, inheritances and home ownership. The results are summarised in Table 4.

Grant M. Scobie and John Gibson

418

Table 4: Probability of having a Predicted Retirement Income below 60 per cent of the Median (Un-partnered Individuals Retiring at Age 65 and with no Consumption of Housing) Dependent variable: probability of low retirement income (o60% median) Income ($’000) Years of education Male Pakeha Ever inherited Owns home

Coefficients

t-Statistic

p-Value

0.003 0.001 0.027 2.150 0.003 0.017

5.68 1.14 1.54 1.77 0.64 1.67

0.00 0.25 0.12 0.08 0.52 0.10

Notes: n ¼ 361; pseudo R2 ¼ 0.60. The coefficients show the change in the probability of having predicted retirement income below 60 per cent of the median for a one-unit change in the independent variable. All coefficients are multiplied by 1012.

Figure 6: Prescribed Saving Rates for Unpartnered Individuals Aged 45–55 (Retirement at Age 65 and Consumption of 50 per cent of Housing Wealth) 0.40 0.35

Frequency

0.30 0.25 0.20 0.15 0.10 0.05 0.00 <0

0-0.05

0.05-0.10

0.10-0.15

0.15-0.20

0.20-0.25

>0.25

Saving Rates: Share of Pre-retirement Income

The probit analysis suggests that the main characteristics of individuals with low predicted retirement income is that they have low current incomes, that they are non-Pakeha and that they do not own their own home. These results reinforce the cross-tabulations reported in Table 3. To further explore the variability across individuals, Figure 6 plots the frequency distribution of the prescribed saving rates. Over one-third of respondents aged 45–55 have negative savings rates (implying no further

Household Net Wealth in New Zealand

419

Table 5: Key Variables Associated with the Prescribed Saving Rate and Projected Retirement Wealth (For Individuals Aged 45–55 Retiring at Age 65 with no consumption of Housing Wealth) Explanatory variables

Dependent variable Prescribed saving rate

Projected retirement wealth

Income Net worth Years of education Male Pakeha Ma¯ ori-Pacific Islander Ever inherited Own home

++ –

++ NA

  

++



++

 

– – ++ –

Notes: ++ and — indicate the variable is significant at the 1 per cent level; , not significant; na, not applicable.

saving would be required to meet their replacement rate, given their current levels of retirement wealth). In total, 57 per cent have a rate below 10 per cent of their pre-retirement (before tax) income. A further 35 per cent would require between 10 and 20 per cent, and 8 per cent require a saving rate of over 20 per cent (recall that the prescribed saving rate is predicated on the assumption that individuals seek to smooth their consumption over the life cycle). In Table 5 we present the results of descriptive regressions that test for the importance of some key selected explanatory variables for the prescribed saving rate and for the level of projected retirement wealth. What is associated with a higher prescribed saving rate? Higher incomes are positively related, as a consequence of the fact we have found that the prescribed saving rates are low (or even negative) for the lowest income groups. Those with higher net worth predictably have a lower prescribed saving rate, holding income and other factors constant. People from the Maori and Pacific Islander ethnic groups have higher prescribed saving rates. Education and gender are not however significant in explaining the prescribed saving rate. Inheriting is negatively associated with the saving rate but not significantly so — and owning a house is associated with a higher prescribed saving rate, in the case where none of the wealth in housing is to be made available for retirement income. Projected retirement wealth is (unsurprisingly) higher for those with greater incomes. It is lower for both Pakeha and Ma¯ ori-Pacific Islanders than for Asians and other (the reference group in this case). Again, gender and education have no effect. Inheritances boost projected wealth, but home ownership reduces it. This result for home ownership reflects the assumption

420

Grant M. Scobie and John Gibson

Table 6: Median Prescribed Saving and Replacement Rates by Conversion of Housing Wealth Conversion of housing wealth to retirement income (%) None 50 100

Prescribed saving rate (%)

Replacement rate (%)

12.8 9.1 5.5

58.7 62.4 66.1

Note: The values in the table are the averages for respondents between the 45th and 55th percentiles of the (weighted) distribution of total wealth.

that housing wealth cannot be consumed, and so does not get counted in the projected retirement wealth. If full consumption of housing wealth is allowed, there is no significant effect of home ownership in the regression for projected retirement wealth. 4.4 Converting Housing Equity The extent of conversion of housing equity to retirement income clearly has a marked effect on the prescribed saving rates for couples as well as unpartnered individuals. The median saving and replacement rates for couples are shown in Table 6 for three assumed levels of conversion of housing. In the case with no conversion (i.e., the entire value of housing wealth is retained and bequeathed), the prescribed median saving rate is 12.8 per cent of pre-retirement (before tax) income. This falls to under half (5.5 per cent) if the entire value of housing wealth is converted. Obviously, in this case, housing rental would become an expense to be met from the higher retirement income. 4.5 Is there Under-Saving? It is widely believed that people typically do not save adequately for retirement. From one perspective this seems improbable. After all, saving decisions are made by individuals who arguably are the best judge of their own welfare. If they could increase their own well-being by foregoing some consumption today, and use those savings to enjoy a higher level of consumption tomorrow, then arguably that is exactly what they would do. However, there may be other elements to the optimal saving story not captured by this view of the rational consumer. Three possible candidates for the under-saving hypothesis are externalities, transaction costs and procrastination. The first requires that there is some distortion that precludes people from making the ‘‘right’’ choices. The second argument is that

Household Net Wealth in New Zealand

421

because of the complexity of the problem, and the uncertainties that are inherent, people simply find the task too difficult and are discouraged from investing time and effort in retirement planning. Finally there is an emerging behavioural economics literature offering richer insights into how people make decisions. The standard model used widely in economics and finance assumes that the rate at which individuals discount the value of funds in the future is constant, regardless of the length of the planning horizon. However, as Ackerlof, 2002 argues: ‘‘y individuals use high discount rates to evaluate options that require an immediate sacrifice for a future reward and lower discount rates when the same sacrifice is deferred’’. People thus appear to procrastinate about saving (and taking more exercise, dieting, quitting smoking, etc.). They are much more willing to undertake the investment in the future than they are to start immediately.9 4.6 Comparing Actual with Prescribed Rates of Saving In this paper we have estimated the rates of saving that would be needed were people aiming at smoothing consumption over their lifetimes. A possible test of this underlying theory is whether the predicted saving rates are matched by the rates at which we observe people actually saving. To conduct this test we have used estimates of actual savings behaviour derived from the Household Economic Survey (HES), which is the best source available for this information at the individual household level.10 From that survey we were able to derive estimates of the ratio of household saving to disposable income by decile of disposable income. We then adjusted these to ratios of savings to pre-tax (gross) income, using the relevant tax rates.11 We recognise that, at best, this test can only give indicative results — as the two surveys are quite separate and do not cover the same individuals, which would make a more ideal comparison. Rather than compute the actual savings rates from the HES for a single year, a more valid comparison with the predicted rates would be to compare them with the estimated actual saving rates over the remainder of the 9

Ackerlof, (2002) provides a concise synthesis of the theory and cites a growing body of empirical evidence that suggest that insights from behavioural economics can help explain the observed pattern of saving. 10 See Gibson and Scobie (2001). 11 From the HES we obtained S/Yd, where Yd is disposable income. Now S/Yp ¼ (S/Yd) (Yd/ Yp). However, as (Yd/Yp) ¼ (Yp Tp)/Yp ¼ 1tp, where Tp is total preretirement taxes and tp the rate of personal income tax applicable to the particular income level, then S/Yp ¼ (S/Yd)) (1–tp).

422

Grant M. Scobie and John Gibson

working life. This was accomplished for two selected cohorts by forecasting the savings rates from regressions by age cohorts based on the pattern of saving by age, reported in Gibson and Scobie (2001). The HES data were based on a sample that excluded single person households, so that the results were more comparable to the HSS definition of couples. A set of conditioning variables was included in the regressions to allow for the effect of house tenure, income, gender, ethnicity and employment type. The mean values of these variables for the respective cohorts were assigned when making forecasts. For example, the cohort born between 1930 and 1939 were on average 56 years old in the years covered by the HES. We used the corresponding regression equation for the actual saving rate to forecast saving rates for each year of age from 56 to the specified retirement age of 65. To construct the results we used two types of regression equations. The first is the standard OLS, which passes through the mean of the variables and is denoted ‘mean regression’. The second is the case where the regression equation passes through the median. The mean regression results are appropriate for considering the savings position for society as a whole. The median results can be viewed as those for a ‘typical’ household. We use both regression models to forecast the savings rates over the years remaining to retirement and we summarise these by taking either the means or the medians. The results are summarised in Table 7 for the cohorts born in 1930–1939 and 1940–1949. Consider first the results for the older cohort. The third column in Table 7 gives the prescribed rates of saving needed for attaining consumption smoothing. In the case of the regression through the mean, we find that the actual saving rates are well above the prescribed. The conclusion is that for Table 7: Income)

Comparison of Prescribed and Actual Saving Rates (Percentages of Gross Prescribed saving rate

Actual saving rate

Cohort born 1930– 1939 OLS Mean Median Median regression Mean Median

1.7 15.9 1.7 15.9

18.6 18.3 9.6 9.9

Cohort born 1940– 1949 OLS Mean Median Median regression Mean Median

7.2 16.0 7.2 16.0

21.4 22.7 13.5 14.6

Household Net Wealth in New Zealand

423

this group taken as a whole, the level of actual saving appears to be adequate to sustain retirement consumption levels. However, when we turn to the regression through the medians, we find that the lower bound estimate of the actual saving rates falls below the required rate. A similar pattern of results is presented for the cohort born 1940–1949. In this case, while the mean results for both methods lie above the prescribed rates, using the median forecasts, we find the actual rate falls marginally below the required saving rates for this cohort. Given any reasonable judgement about implicit standard errors, it is unlikely this difference is significant. These results suggest that, given the requirement of consumption smoothing, we cannot make unequivocal statements about the extent of adequacy of retirement savings. Clearly the results are dependent on the method that is used, together with the criterion that is adopted to judge ‘‘adequacy’’. Whether or not the mean or median savings rates of a group can be viewed as ‘‘adequate’’ abstracts from arguably a more critical question — namely, what is the distribution of rates? While the median rate might indicate that a ‘‘typical’’ couple could be close to or above the prescribed rate, it tells us nothing about the distribution. However, based on our projections of retirement incomes there are a significant number of people in the lower income brackets for whom no additional saving would be required. NZS provides a floor under their retirement income, which when viewed together with free medical care and long-term rest home care (subject to an asset test), would make it irrational for people to postpone consumption today from income levels that are low (and may well be higher in the future when receiving NZS). It should be stressed that we have deliberately adopted a conservative position in estimating future retirement incomes: the net effect of these conservative assumptions is to make the required saving rates we estimate higher than they would otherwise be. We allow for no further capital appreciation on housing, farms, commercial or rental properties, time shares and all other property. We exclude the primary residence from the calculation of the retirement annuity — that is, no equity in the house is used to support retirement incomes. We assume a modest rate of return on other assets. Furthermore, we assume constant real consumption throughout retirement. In fact consumption requirements typically fall with age. Based on an analysis of the HES, Gibson and Scobie (2001) show that the median level of consumption expenditure falls by some 30 per cent as the age of the household head increases from 65 to 75. Arguably this could arise due to income being constrained, forcing a decline in real consumption. In fact savings rates (both mean and median) rise, not fall over this period. One would not expect to observe rising saving rates if consumption were to be constrained

424

Grant M. Scobie and John Gibson

by inadequate income. We conclude that our assumption of requiring a constant level of retirement consumption is conservative, given the evidence that people in fact appear to have reduced consumption needs as they grow older. The effect is to make the prescribed savings rates higher than they would otherwise be.

5. Conclusions In this study we have based simulations of retirement income on the projected stock of retirement wealth that a person might be expected to have. This stock of wealth is based on the accumulation of net worth observed in the HSS, and includes the expected value of NZS. The stock of retirement wealth is converted to an annuity. We have considered the effect of both including and excluding the primary residence as part of the stock of retirement wealth. We have been able to develop a profile of retirement income for various classes of individuals. We are able to explain some, but by no means all of the variation in net worth across individuals in the pre-retirement age group aged 45–55 years. What really explains retirement wealth — i.e., the amount an individual say 50 years old has accumulated and could be used as a source of retirement income. The answer is a complex set of economic and social conditions combined with individual preferences and circumstances. Only with much more detailed information about personal life histories and preferences could we expect to improve our ability to explain variations in retirement wealth accumulation. Furthermore, cohort effects are likely to be a significant factor in them. The assumptions underlying the modelling must be stressed. The fundamental behavioural assumption is that people would select a rate of saving that would allow them to smooth their consumption over the life cycle. In other words, we determine the level of saving and the implied income replacement rates assuming that this is how people behave. In no way do the results represent what people actually choose. Nor in any sense do they imply what they ‘should do’. There is no unique answer to the question: are New Zealanders saving adequately for retirement? There would be a different answer to that question for each definition of what is viewed as ‘‘adequate’’. Our approach has been to make estimates of potential retirement income based on the observed levels of retirement accumulation at the time of the survey. From that we have derived the saving rate that would be implied to attain the objective of consumption smoothing. As a test of whether the underlying model is capable of predicting actual saving behaviour, we compared the prescribed saving rates with the actual

Household Net Wealth in New Zealand

425

rates from the HES. The results are mixed. Even if the means and medians of the prescribed and actual distributions of saving rates are reasonably close, this should not be taken as evidence that all individuals within, say, an income decile, have an actual saving rate that is close to the prescribed rate. The only way to make this comparison, individual by individual, is to have panel data at repeated intervals. However, even if we were to have such a data set, there would be a further obstacle to concluding that there was evidence of under-saving. Under-saving is detected in this context by comparing the actual saving rates that people choose to adopt with a rate that would have been chosen had they been following some norm proposed by someone else. Suppose we were to find that, say, 20 per cent of people in a particular income decile were saving at rates below the prescribed level. The possibility that these individuals are operating with a different concept of what is the best rate for them cannot be ruled out. They may, for example, have different expectations about future outcomes (their health, life expectancy, retirement income policies, etc.), or they simply may have different preferences for consumption now versus tomorrow. Had we captured adequately those characteristics of this group, it is entirely possible we would have found that their observed rates of saving were compatible with the rate that was optimal for them. Of course this still leaves open the possibility that they had incomplete information — and that the provision of better retirement planning information may have lead them to alter their current rate of saving. The effect of NZS is to make the distribution of retirement wealth much less unequal. It provides a significant floor under the retirement income of the lower income and wealth deciles. Thus, only between 10 and 15 per cent of people could be expected to fall below 60 per cent of median predicted retirement income for this cohort. The actual proportion would vary depending on the proportion of the housing wealth that was converted to retirement income. For the lowest income groups NZS represents over 90 per cent of total retirement wealth. One consequence of this is that the prescribed savings rates are low or even negative for the lowest income deciles. In short, the evidence suggests that among the lowest 40 per cent of the income distribution, lifetime welfare would be reduced if these groups were to increase current savings and thereby reduce their pre-retirement consumption. In general, our results provide some comfort but fall short of absolute assurance that New Zealanders are saving adequately. We have erred on the conservative side in our assumptions and NZS plays a critical role in supporting the retirement incomes of the lower income deciles. It represents the majority of their retirement wealth, and contains a significant insurance component in that it is a lifetime annuity. However, when combined with

426

Grant M. Scobie and John Gibson

other social policies in health, education, long-term care and hardship benefits, it demonstrably reduces the incentives of those groups to forego current consumption to save for their retirement. Our findings are consistent with two recent studies from the United States. A study by Scholz et al. (2004) concluded, in summary, that Americans were generally behaving quite rationally and that ‘‘y [f]ewer than 20 per cent of households have less wealth than their optimal targets, and that the wealth deficit of those who are undersaving is generally small.’’ Furthermore work by Engen et al. (2004) concluded: ‘‘y that households at the median of the wealth-to-lifetime earnings distribution are saving as much as, or more than, what the underlying model suggests is optimal, and households at the high end of the wealth distribution are saving significantly more than what the model indicates. But the results also show undersaving among the lowest 25 percent of the population.’’ In other words, if there is a saving problem in the US, it is just focused on the bottom quartile of household incomes. Arguably, this group is least well placed to do anything about that problem for themselves. But it is precisely this group in New Zealand for whom the universal (i.e., not means-tested) state pension (NZS) provides a significant floor under retirement incomes. Unfinished Business Clearly the rate and nature of asset decumulation will depend on the expectations a person holds about his or her life expectancy, the rates of inflation and interest that might prevail, and the probability of poor health leading to unanticipated expenses. The assumptions made in this study about asset accumulation for retirement have been somewhat restrictive. They have largely ignored differences in the composition of the stock of retirement wealth and with the exception of housing have not allowed for any further bequests. An important aspect warranting further work is the relation between wealth accumulation and the timing of retirement decisions. Ideally allowance would be made for the fact that the retirement age which we have taken as fixed may vary systematically with the level of accumulated wealth. Arguably the two would be simultaneously determined, although Compton (2001) based on evidence for Canada, finds no evidence that wealth explains the decision to retire. Type of occupation, the labour force status of the partner, and health and disability are the dominant factors influencing the decision to retire.

Household Net Wealth in New Zealand

427

Some limited research has shown that many people (especially at younger ages) do not expect to receive the current level of NZS when they reach retirement.12 To what extent do those who expect significantly lower NZS payments hold greater levels of private retirement wealth? Has the creation of the NZS fund altered these expectations and hence the rate of accumulation of retirement wealth? These are all testable questions, the answers to which would greatly enhance our understanding of the determinant of retirement wealth among New Zealand householders. We recognise that the consumption smoothing model, with or without uncertainty, is a stylised representation of how decisions are made. In fact, the evidence is that some if not all individuals do not actually behave this way, leading to the so-called ‘‘retirement puzzle.’’ Banks et al. (1998) find that for British data, well-educated males with occupational pensions are able to largely smooth consumption across retirement. However they find a considerable number take a drop in living standards. In this study we have found that due to the presence of NZS, the consumption levels of a large segment of the lower income earners could be sustained with their current stock of retirement wealth. Because of the limitations of the HSS it is not possible to state whether households in the survey are in fact saving ‘‘too little’’ or ‘‘too much’’. The survey has provided a valuable snapshot of the assets and liabilities of New Zealanders. The current net worth observed in the survey reflects the past decisions that people have made about how much to accumulate. But by its very nature, it cannot reveal their present rate of saving. Repeated surveys at periodic intervals in future would add further critical insights into the pattern of accumulation across households.

Acknowledgement and Disclaimer The authors acknowledge comments and suggestions from Richard Disney, John Creedy, Geoff Lewis and an anonymous referee. The staff at Statistics New Zealand played a supportive role in providing access to and advice QJ;on the data in a secure environment designed to give effect to the 12

In a survey conducted in September 1999 as part of the Super 2000 Taskforce, 27 per cent of respondents reported that they expected to have a higher standard of living in retirement while 38 per cent felt these standards of living would decline. When asked whether they were concerned that many New Zealanders would not have a comfortable retirement, 77 per cent agreed with the statement. Of all 30year-old respondents, 50 per cent expected to retire at or before age 60, while only 4 per cent of all respondents believed the current level of NZS payments would be sustained over the next 20–30 years (UMR Insight Ltd, 1999).

428

Grant M. Scobie and John Gibson

confidentiality provisions of the Statistics Act (1975). The results in this study and any errors contained therein are those of the authors, not Statistics New Zealand. The views, opinions, findings and conclusions or recommendations in this chapter are strictly those of the authors and should not be reported as those of the New Zealand Treasury. The chapter is presented not as policy, but with a view to inform and stimulate wider debate. The New Zealand Treasury takes no responsibility for any errors, omissions or for the correctness of the information in this chapter. For more detailed results, see Scobie, Gibson and Le (2005).

References Ackerlof, G.A. (2002). Behavioral Economics and Macroeconomic Behavior. American Economic Review, 92(3), 411–433. Banks, J., Blundell, R. and Tanner, S. (1998). Is There a Retirement-Savings Puzzle? American Economic Review, 88(4), 769–787. Bartlett, J. (1980). Familiar Quotations, Little Brown. Boston. Bernheim, D., Lorenzo, F., Gokhale, J. and Kotlikoff, L. (2000). How Much should Americans be Saving for Retirement?. American Economic Review, 90(2), 288–292. Claus, I. and Scobie, G.M. (2001). Household Net Wealth: An International Comparison. Wellington, New Zealand Treasury, Working Paper no. 01/19. Compton, J. (2001). Determinants of retirement: Does money really matter? Ottawa, Department of Finance, Working Paper no. 2001-02. Engen, E.M., Gale, W.G. and Uccello, C. (2004). Lifetime Earnings, Social Security Benefits and the Adequacy of Retirement Wealth Accumulation. Chestnut Hill, MA, Centre for Retirement Research, Boston College, Working Paper No 200410. http://www.bc.edu/centers/crr/papers/wp_2004-10.pdf. Gibson, J. and Scobie, G. (2001). A Cohort Analysis of Household Income, Consumption and Saving. New Zealand Economic Papers, 35(2), 196–217. Moore, J.F and Mitchell, O.S. (1997). Projected Retirement Wealth and Savings Adequacy in the Health and Retirement Study, National Bureau of Economic Research, Cambridge, MA. http://www.nber.org/papers/w6240. Scholz, J.K., Seshadri, A. and Khitatrakun, S. (2004). Are Americans Saving ‘‘Optimally’’ for Retirement? Cambridge, MA, National Bureau of Economic Research, Working paper 10260. ohttp://papers.nber.org/papers/W102604 Scobie, G.M. and Gibson, J.K. (2003). Retirement Wealth of New Zealand Households: An Initial Analysis Based on the Household Savings Survey. Paper presented to a Symposium on Wealth and Retirement sponsored by the Office of the Retirement Commissioner (Wellington). Scobie, G.M., Gibson, J.K. and Le, T. (2005). Household Wealth in New Zealand, Institute for Policy Studies, Victoria University of Wellington, Wellington. Statistics New Zealand. (2002a). The Net Worth of New Zealanders: A Report on Their Assets and Debts, Statistics New Zealand, Wellington.

Household Net Wealth in New Zealand

429

Statistics New Zealand. (2002b). The Net Worth of New Zealanders: Data From the 2001 Household Savings Survey: Standard Tables and Technical Notes, Statistics New Zealand, Wellington. ohttp://sorted.org.nz/survey_index.php4. Thorp, C. and Ung, B. (2000). Trends in Household Assets and Liabilities Since 1978. Reserve Bank of New Zealand Bulletin, 63(2), 17–37. Thorp, C. and Ung, B. (2001). Recent Trends in Household Financial Assets and Liabilities. Reserve Bank of New Zealand Bulletin, 64(2), 14–24. UMR Insight Ltd. (1999). New Zealand Retirement Income — A Tracking Study, Wellington. Prepared for the Super 2000 Taskforce.

Appendix The starting point for the model is the condition that real consumption (i.e., income net of taxes and saving) be equal before and after retirement, as given by ðY p  T p  SÞ ¼ Y r  T r

(1)

where Yp ¼ pre-retirement gross income Tp ¼ pre-retirement taxes S ¼ savings Yr ¼ post-retirement gross income Tr ¼ post-retirement taxes Next define s ¼ pre-retirement savings ratio ¼ (S/Yp) and R ¼ replacement ratio ¼ (Yr/Yp) so that substituting these definitions in (1) and dividing the Y p gives 1  ðT p =Y p Þ  s ¼ R  ðT r =Y p Þ

(2)

Now let T p ¼ tp Y p and T r ¼ tr Y r ; where tp and tr are the pre- and postretirement proportional tax rates, so that s ¼ ð1  tp Þ  ð1  tr ÞR

(2a)

Equation (2a) defines a set of combinations of s and R which satisfy the condition specified in (1). By first finding a value for R; we can then solve for the corresponding value of s that satisfies (2a). The post-retirement income flow ðY r Þ can be converted to a lump sum at retirement by applying an annuity factor (a). This expresses the stream of post-retirement income in terms of a stock in wealth at the time of retirement. In other words, were a person to have accumulated this amount he or she would be able to receive a lifetime annuity of Y r :

Grant M. Scobie and John Gibson

430

Denoting the ‘‘required’’ wealth needed to generate Y r as W r then   W r ¼ aY r ¼ a ð1  sÞY p  T p þ T r

(3)

The amount of saving needed to reach this required level of retirement income ðwr Þwill depend on the existing stock of net wealth, the expected returns on investment, future income and tax rates. We define ðW p Þ as the projected level of wealth so that the shortfall is   W r  W p ¼ a ð1  sÞY p  T p þ T r  W p (4) We are now in a position to derive the rate of saving needed to reach the required level of wealth. This rate is the share of pre-tax income the household would need to save to have the level of income Y r in retirement. The amount accumulated by retirement would then be Wr  Wp ¼

T X t¼1

¼ sY a

Y a ð1 þ gÞt ð1 þ rÞTt s "

T X

# t

ð1 þ gÞ ð1 þ rÞ

Tt

¼ sY a Z

ð5Þ

t¼1

where Ya ¼ actual income in year t ¼ 1 y T g ¼ annual growth rate of income r ¼ after-tax real rate of return on saving Using (4) and (5) we can solve for the saving rate, s13 s¼

aðY p  T p þ T r Þ  W p   Y p a þ Z=ð1 þ gÞT

where Y p ¼ Y a ð1 þ gÞT : Now dividing by Y p gives  atr R þ að1  tp Þ  ðW p Y p Þ    s¼ a þ Z ð1 þ gÞT

(6)

where T is the number of years from the person’s current age until the predetermined age of retirement. Note again that s is a linear function of R. Now equating (2’) and (6) we have  atr R þ að1  tp Þ  ðW p Y p Þ  ¼ ð1  tp Þ  ð1  tr ÞR, a þ ðZ ð1 þ gÞT Þ

13

Note the saving rate is defined as a constant share of pre-retirement income.

Household Net Wealth in New Zealand

which after collecting terms in R gives   atr þ ð1  t Þ ¼ ð1  tp Þ R r a þ Z=ð1 þ gÞT     að1  tp Þ  W p Y p  a þ Z=ð1 þ gÞT

431

ð7Þ

Now consider the coefficient of R on the LHS of (7) ¼

  atr tr Z þ ð1  t Þ ¼ 1  r a þ Z=ð1 þ gÞT að1 þ gÞT þ Z

while the RHS of (7) can be written as: "    # að1  tp Þ  W p Y p  ¼ ð1  tp Þ   að1 þ gÞT þ Z ð1 þ gÞT    Zð1  tp Þ þ W p Y p ð1 þ gÞT ¼ að1 þ gÞT þ Z Now substitute (8) and (9) in (7) and solve for R:   Zð1  tp Þ þ ðW p Y p Þð1 þ gÞT að1 þ gÞT þ Z  R¼ að1 þ gÞT þ Z  tr Z að1 þ gÞT þ Z

so that R ¼

 ð1  tp ÞZ þ ðW p Y p Þð1 þ gÞT ð1  tr ÞZ þ að1 þ gÞT

(8)

ð9Þ

(10)

(11)

We return now to the question of tr : It is argued that in the context of the New Zealand system of taxation, private retirement saving is made from after-tax pre-retirement income ðY p  T p Þ and the earnings on the investments are taxed. However, once those accumulated funds are withdrawn (in this case to purchase an annuity) then there is no further taxation on the income received in retirement. Furthermore, NZS payments are received net of tax. Hence under this system (denoted TTE) and unlike in the USA, tr ¼ 0: With this simplification we can solve for R for each individual having T years remaining until retirement, such that R¼

 ð1  tp ÞZ þ ðW p Y p Þð1 þ gÞT Z þ að1 þ gÞT

(12)

Grant M. Scobie and John Gibson

432

Figure A.1: Joint Determination of the Saving Rate (s) and the Replacement Rate (R) s 1-tp

From Equation (12)

A

s0 s1

B From Equation (13) R0

R

R1

Figure A.2: Effect of a Reduction in the Pre-Retirement Tax Rate (tp) on the Rate of Saving (s) and the Replacement Rate (R) s 1-tp´ Locus of (s, R) 1-tp

B s1 C s2 s0 A

R0

R1

R

and then substitute this value to find s from s ¼ ð1  tp Þ  R

(13)

The linear relation between s and R specified in (13) is shown in Figure A.1 as a negatively sloped line with slope equal to –1. It is derived from the

Household Net Wealth in New Zealand

433

condition that pre- and post-retirement consumption levels are equal (refer to (1)). The maximum value s can take is (1tp), which would imply (unrealistically) that all after-tax income was saved. For a given individual, (11) specifies a replacement rate (R0) and the intersection at point A determines the corresponding saving rate (s0). A higher value of initial wealth (Wa) and hence projected wealth (Wp) would lead to a greater replacement rate (R14R0) and all else being equal, this would allow a lower rate of saving (s1os0). For example, to the extent that NZS is viewed as part of retirement wealth (Wp), given an increase (decrease) in the level of payment, the model would predict a decrease in the level of the saving rate. Consider the effect of a reduction in the pre-retirement tax rate from tp to tp0 . This will raise the intercept on the s axis from (1tp) to (1–tp0 ) as, tp0 otp This is shown by an outward shift of the s–R line in Figure A.2. Initially this would raise the saving rate to s1 corresponding to the intersection at B. However as is evident from (11), R is negatively related to tp, so a fall in tp will result in an increase in R. This will now lead to a new saving rate of s2, corresponding to the point C; i.e., the new (s,R) combination lies in the north-east quadrant from A and the lower tax rate would induce a higher saving rate (s24s0) and a higher replacement rate (R14R0). Only in the case that the marginal propensity to consume exceeded unity could the saving rate fall below its initial level of s0. By making successive changes to the tax rate the locus of equilibrium combinations of s and R is traced out as shown in Figure A.2. The key theoretical proposition to emerge from this comparative statics exercise is that lower pre-retirement tax rates can lead to higher saving rates and simultaneously, higher replacement rates.

This page intentionally left blank

Chapter 17

Population Ageing, Housing Demand and House Prices in Australia: A Microsimulation Analysis Ross S. Guest Griffith University, Australia

Abstract This paper presents a simulation analysis of the impact of population ageing on housing demand and house prices in Australia for the period 2002–2052. The demand for owner-occupied housing is derived from a life-cycle model of housing tenure choice, which is then combined with a simple model of housing supply to analyse the interaction of housing demand, supply and house prices. The model predicts that from the year 2002 to 2052, population ageing will cause a drop in real house prices by up to 10 per cent in the absence of confounding factors.

1. Introduction Mankiw and Weil (1989) painted a somewhat extreme scenario for the effect of population ageing on house prices in the U.S., predicting that the decline in housing demand due to an older population would see real house prices fall by up to 40 per cent. A decade later, population ageing is underway in Australia, the U.S. and other developed economies — but real house prices have not yet declined. However, this may be due to confounding factors, not least of which are the much lower interest rates that exist now compared with the 1980s when Mankiw and Weil made their prediction. Although the Mankiw and Weil story was widely criticised as being too alarmist, it is accepted that the gathering pace of demographic change over the coming decades will have at least some effect on housing demand. The question is: how much and what effect will this have on house prices? This is of more than academic interest to the future cohorts of homebuyers in Australia, given that housing affordability is currently at a low record. The purpose of this chapter is to address this question. The method is to apply a microsimulation model of housing tenure choice by households. International Symposia in Economic Theory and Econometrics, Vol. 15 Copyright r 2007 Elsevier B.V. All rights reserved ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15017-6

436

Ross S. Guest

The key premise of the model is that the decisions of households about how much to spend on the purchase of a home, when to buy it, how much to save and how much to spend on non-housing consumption, are joint decisions made in a life-cycle context. This presumption has both a theoretical and an empirical basis. The desire to pay off a home provides an incentive to sacrifice current consumption and therefore to save. The Mercantile MutualMelbourne Institute Household Savings Survey (Loundes, 2001) reports ‘‘paying off home’’ as the most common and significant form of household saving in Australia. The survey reports that 25–44 year olds and households with an income between $41,000 and $100,000 are most likely to save in the form of paying off the mortgage (Loundes, 2001).

2. Previous Research A number of studies have analysed the optimal purchase of housing services by households in the context of a life-cycle model of consumption. The focus of these studies has been on the effect of one or more of the following: liquidity and/or wealth constraints (Artle and Variaya, 1978; Slemrod, 1982; Miles, 1992); tax policy (Slemrod, 1982; Hayashi et al., 1988 ); and mortgage instruments and inflation (Alm and Follain, 1982). The effect of liquidity or wealth constraints at the time of purchase plays a critical role in the optimal planning of housing, saving and consumption decisions. A binding liquidity constraint on the purchase of a house acts to increase saving and reduce consumption early in the life cycle and increase consumption later in the cycle, compared with the life-cycle consumption profile of a household that rents throughout the life cycle. The incentive to distort consumption in order to buy a house is due to the tax incentives for owner-occupied housing. Hence there is a trade-off between the utility cost of distorting the life-cycle consumption plan to meet the housing down payment and the utility gain available through the tax incentives and other less tangible benefits of owner-occupied housing. Some of the above authors — Alm and Follain, Slemrod and Hayashi et al. — note the computational difficulties in modelling tenure choice. In particular, in order to find the optimal tenure pattern over the life cycle, it is necessary to calculate the utility maximising consumption choice for all possible rent-own combinations over periods of the life cycle. This implies significant increase in computational difficulties with increases in the number of periods in the life cycle. Slemrod (1982), for example, considers only one lifetime tenure pattern, Alm and Follain (1982) consider four possible lifetime tenure patterns, and Hayashi et al. consider 10 possible tenure patterns. However, in the time since these studies, increases in processing speed of standard PCs have reduced computational costs considerably.

Population Ageing, Housing Demand and House Prices

437

The present study searches over a much greater number of possible tenure patterns to find the one that maximises lifetime utility (discussed below). An alternative empirical approach to investigating housing demand and tenure choice is by econometric estimation using time series and/or crosssection data. With respect to tenure choice, some studies have found that older homeowners do not reduce their housing consumption at all until retirement — and thereafter by very little, implying a low incidence and degree of trading down in retirement (VanderHart, 1998; Ermisch and Jenkins, 1999; Skaburskis, 1999). With respect to borrowing constraints and housing affordability, an Australian study by Bourassa (1995a) shows that borrowing constraints are a significant determinant of the probability of home ownership. His conclusion is that government programs to encourage first-home buyers should focus on reducing deposit requirements. On the impact of demographic change, a number of econometric studies have found that demographic factors are important in determining likelihood of home ownership. These include Bourassa (1994, 1995b) for Australia and Ermisch (1996) for U.K. data. The latter study shows how age can affect the demand for housing through three channels. First, depending on the relationship between the return to saving and the household’s rate of time preference, the demand for housing services can rise or fall over time as the household ages. Second, if incomes are growing over time, younger cohorts will have a greater demand for housing than older cohorts, other things being equal. Third, borrowing constraints in the presence of rising incomes with age will be more likely to bind on younger households. However, by far the most controversial and widely cited study of the impact of demographic change on housing demand was that by Mankiw and Weil (1989) — hereafter M-W. They estimated an age-specific demand function for housing. Their controversial finding was that just as the baby boomers were responsible for the increase in housing prices in the 1970s, so they will be responsible for a decline in house prices in the first decade or two of the new millennium. This is because supply is found to be relatively inelastic in the long run and hence changes in housing demand are felt through changes in price (rather than changes in the housing stock). The M-W study spawned a large number of papers where their method was extended and their findings were, in most but not all cases, criticised. Ohtake and Shintani (1996) find the opposite to M-W for Japan — that in the long run the price elasticity of housing supply is large. The same result was found for Canada by Engelhardt and Poterba (1991). Both Swan (1995) and Green and Hendershott (1996) argue that M-W failed to control for other demand variables correlated with age. Hamilton (1991) refuted the M-W finding by replacing the asset price with the rental price, which he argued is the variable more directly influenced by demand than the asset price. He also took issue, along with Hendershott (1991), with the interpretation of the time trend variable in the M-W study. Holland (1991), in the

438

Ross S. Guest

same journal issue devoted to the M-W study, argued that the M-W result was due to the failure to take account of non-stationarity. Alperovitch (1995) argued, like Hamilton (1991), that the M-W equation was incorrectly specified but he argued that Hamilton’s was also incorrectly specified and that there was no correct structural model that could generate the reduced form equation used by M-W. DiPasquale and Wheaton (1994) found that the long-run supply schedule for housing is ‘‘quite price elastic’’ but rising — and therefore the negative shock to housing demand arising from population ageing will not be sufficient to result in any significant and sustained decline in prices. Poterba (1991) found that the population distribution is not a significant variable in regression of house prices for cities in the U.S. The study in McFadden (1994) found rare support for the findings of M-W who in their 1991 reply to their critics maintained, in the face of all of this criticism, that there is a relationship between their demographicdriven demand variable and house prices.

3. The Model The model of tenure choice used here is a slightly modified form of the model of Hayashi et al. (1988), which is in turn based on that in Slemrod (1982).1 Given demographic data, aggregate housing demand is derived from the tenure choice model and then combined with a simple model of housing supply in order to determine the effect of demographic change on house prices. The housing supply model applied here is the stock-flow model in M-W (1989), which is a variation of that in Poterba (1984). The description of the modelling in this section is heuristic. Technical details, including functional forms and parameter values, are given in Appendix A. The tenure choice model is characterised by a representative single-person household that plans its optimal tenure choice — rent or own — over its lifetime. The lifecycle consists of 13 periods, each period representing 5 years of a person’s adult lifetime, which starts at age 20 and finishes at age 85. The household chooses the consumption of a composite commodity and housing services for each period over the lifetime, given perfect foresight about its future income and age of death. 1

The main differences between the model here and that in Hayashi et al. (1988) model are: (i) the lifetime here is represented by 13 periods of 5 years whereas Hayashi et al. used 6 periods of 10 years — he recommended a model with more periods but did not pursue this due to computational cost; and (ii) parameters are chosen, where possible, to reflect realities of the Australian housing market and the tax system (e.g., unlike in the U.S., mortgage interest is not tax deductible in Australia).

Population Ageing, Housing Demand and House Prices

439

Housing services may be obtained either by purchasing a house or by renting housing. It is assumed that housing purchases and sales take place at the end of a period, which implies that the household must wait at least until the end of the first period (at age 25) to buy a house. It is also assumed that the household must sell the house at the end of the second last period (at age 80) and move into rented accommodation. Therefore, the household has a choice of owning a house for any duration between periods 2 and 13 — but can buy only once. That is, there are no opportunities for trading up or down to a new house, nor are there opportunities to improve the house through renovations. And, finally, the house does not depreciate. The household chooses the own/rent lifetime pattern that maximises the discounted sum of lifetime utility — subject to the lifetime budget constraint, a wealth constraint and the down payment constraint. The household’s objective is to maximise the discounted value of its consumption index, consisting of the housing and non-housing good, subject to its budget constraints and the additional constraint that it leaves a bequest equal to the equity it owned in the house at the time the house was disposed. The household’s choice variables are: consumption of the non-housing good in all periods, consumption of rented housing services in periods when the household rents, consumption of owner-occupied services which remain constant for the period of ownership and the timing of the purchase and sale subject to the constraints given above. The consumption of housing services is assumed to be proportional to the size of the house, so that a decision to consume more housing services implies occupying a bigger house. Owning a house is more attractive than renting for two reasons: first, the tax advantages arising from the absence of a tax on either the implicit rent from owner occupation or the capital gain on the sale of the house and, second, what Hayashi et al. (1988) call the ‘‘pride-of-ownership’’ effect. The latter includes the psychological and real advantages of home ownership, including security of occupation and freedom to make alterations. The tax advantage applying to the implicit rent from owner occupation is captured in the model, because the repayments of principal on mortgages that renters would otherwise have made must instead be accumulated in financial assets, the interest from which is taxed at the rate t2 — whereas, for owners, the repayment of principal allows the accumulation of housing equity, the implicit income from which is untaxed.

2

It is possible to simulate the effect of the tax break on implicit rental income from home ownership by running a counterfactual simulation in which the implicit income from housing equity, Pr  Heq, where Heq is housing equity and Pr, the rate of income earned on rental property, is taxed at the rate t: The model does not, however, capture the advantage of home ownership that arises from the absence of a capital gains tax on owner-occupied housing.

440

Ross S. Guest

The pride-of-ownership effect is captured by weighting the housing services from owner occupation more highly than the services from renting (see the parameter, g, in equation (3) in Appendix A). Given these advantages of owning, households will prefer to own rather than rent a house of equivalent size in any period. However, households may be constrained by the need to have saved the deposit before the house can be purchased. This constraint is stronger, the greater is the preference for consumption smoothing over the life cycle. The desire to smooth consumption can make it optimal to have little or no saving early in the life cycle when income is relatively low and to accumulate saving as income rises. The desire for little or no saving early when households are young can delay the accumulation of sufficient saving for the deposit on the house. The exogenous variables in the model are labour income per worker by age, the labour force participation rate by age and population shares by age. The population estimates up to the year 2002, are calculated from the Australian Bureau of Statistics (ABS) Catalogue 320109.1 and the projections beyond 2002 are calculated from ABS Catalogue 3222.0. The age-specific labour income levels are the mean weekly earnings of full-time employees from ABS Catalogue 6310.0; and the age-specific labour force participation rates are in person units, from ABS Catalogue 6291.0. Both of these variables are held constant from 2002 onwards. Some qualifications are in order before turning to the results of the simulations. At this point in the analysis, housing supply is assumed to be infinitely elastic and hence increases in housing demand do not increase the relative price of housing. Later on we introduce a simple model of housing supply, which allows us to specify a finite supply elasticity. Also, the asset demand for land is not modelled; only the consumption demand for housing services is modelled. This is likely to bias downwards the level of housing purchases compared with actual housing purchases, which include the demand for land as an asset separate from buildings.3 Another limitation is the absence of uncertainty, which implies that the saving and housing plans that households make at age 20 for the next 65 years of their lives are exactly realised — there are no errors of judgement or revisions of plans. Also the assumption of single person households means that the number of households in each age group is simply the number of persons in that age group. Other assumptions described above such as the absence of opportunities to

3 The expected real appreciation of land reduces the imputed rental price of housing services which equals the real cost of capital minus the expected real appreciation of the house and land combined (minus any depreciation of buildings which is also assumed to be zero in this model).

Population Ageing, Housing Demand and House Prices

441

trade up or down and no home improvements or depreciation are further abstractions in order to make the model tractable. These simplifications mean that, as predictions of actual responses to shocks, the quantitative outcomes can only be interpreted as orders of magnitude. On the other hand, parsimony and simplicity has its advantages. In this context, it allows us to draw insights from the life-cycle approach to analysing the joint decisions of how much to save, how much to spend on a house, and when to buy and sell it.

4. Results 4.1 The Base Case Simulation In the base case the supply elasticity is infinite. The income and (nonhousing) consumption profile are illustrated in Figure 1, for a life-cycle plan made by a household aged 20 in the year 2002. People over the age of 70 are assumed to earn no labour income and the income for people aged from 20 to 70 exhibits a hump-shaped profile, with workers in middle age earning higher labour incomes than both younger and older workers. The consumption profile, on the other hand, is relatively flat — reflecting the optimal smoothing of consumption subject to the constraints imposed by the need to save for a housing deposit and the non-negativity restriction on wealth. The liquidity constraint is clearly binding in the period in which the house is purchased. Figure 1 also illustrates that the period of high saving is in middle age and the period of dis-saving is from ages 60 to 65 onwards. We now consider how well the simulation results for the base case fit the observed characteristics of the Australian households with respect to their Figure 1:

Lifetime Income and Consumption

2.5 cons

2

income

1.5 1 0.5 0 25

30

35

40

45

50

55

60

65

70

75

80

85

442

Ross S. Guest

housing and saving decisions. Appendix B discusses the sensitivity of the results to alternative parameter values. In the base case the optimal age at which to purchase a house is at age 30, which is at the end of the second five-year period in the household’s lifetime planning period. This fits the observation that the most common age group in which households buy a house is the 25–34 age group (ABS, 2003). The optimal private saving rate in 2002 is 18.2 per cent. The actual gross private saving to private income ratio in Australia in 2002 was 11 per cent, although the national saving rate was 19 per cent (ABS Cat. 5206.027). Hence the model over predicts the actual private saving rate unless households reduce their private saving in response to positive public sector saving.4 The mortgage interest to household income ratio in the model is 5.4 per cent in 2002, which is close to the actual rate of 6.5 per cent; and the total debt service (principal plus interest) of 11.6 per cent of income in the model is also close to the actual ratio of 10 per cent (MacFarlane, 2003). The optimal housing demand to private income ratio is 13.1 per cent. The actual expenditure on homes by ‘‘recent homebuyers’’ from 1997 to 1999 (ABS Cat. 4182.0) averages 13.3 per cent of total private income.5 This is only an approximation to expenditure by first homebuyers because it includes purchases of homes for rent. It also includes expenditure on land, which is not accounted for in the model. Therefore, although the optimal and actual numbers are close, the model in fact is overstating the demand for owner-occupied housing services to the extent that it does not include the asset demand for land. The optimal household debt to income ratio, at 25 per cent, is well below the actual level — which has risen from about 60–120 per cent (MacFarlane, 2003). This difference could be explained by the fact that in the model households cannot incur debt for the purchase of land in expectation of a real appreciation in its price. This also implies that there cannot be any housing bubble phenomenon explaining demand for housing credit. Nor does the model allow for borrowing for the purchase of homes for rental. The model yields housing equity of 43 per cent of household wealth which, given the above comments regarding land, is probably compatible with the actual figure between 60 and 70 per cent (Caplin and Joye, 2002). A figure of 57 per cent was estimated by the Australian Treasury as the ratio of dwelling assets to the total private sector network (Littrell, 1999).

4 By equating private saving with household saving in the model we are recognising that households save partly through private firms. 5 The actual new dwelling expenditure, which excludes demand for existing dwellings, is approximately 7.5 per cent of private income in 2002 (ABS Cats. 5206.03, 5206.027).

Population Ageing, Housing Demand and House Prices Table 1:

443

Recent First HomeBuyers, 1999

Age of reference person (years) 15–24 25–34 35–44 45–54 55–64 65 or over

% 12.7 57.9 17.8 7.6 3.0 1.1

Source: ABS (2003) Note: The results are for households that purchased their first home in the three years to 1999.

4.2 The Effect of Population Ageing Infinite Supply Elasticity The M-W thesis suggests that we should find a relationship between the age distribution of the population and housing demand. Table 1 shows the age distribution of recent first homebuyers according to the 1999 Australian Housing Survey. The table indicates that 58 per cent of first homebuyers were in the 25–34 age group. The next most common age group was the 35–44 age group, which accounted for 18 per cent of recent first homebuyers. The M-W thesis would suggest, therefore, that we should look to the population growth rate of the 25–34 year age group for information about the behaviour of housing demand. The annual projected growth rate of 25–34 year olds is illustrated in Figure 2 from 1952 to 2052, using the ABS Series B projection.6 Figure 2 shows that the annual growth rate of the 25–34 age group peaked at over 4 per cent in 1972 and declined steadily to about zero in 2000. Between 2000 and 2052, the growth rate follows a cyclical pattern, as the children of the baby-boomers enter this age group in the first two decades of this century, followed 30 years later by another wave as the grandchildren of the baby-boomers enter the 25–34 age group. These ‘‘echo effects’’ of the baby boom diminish with each cycle. Population ageing does not disturb the optimal life-cycle consumptionsaving plan for a particular household. But the changing population distribution alters the proportion of households of house-buying age and the proportion of households in the high- and low-earning and saving periods of their life cycles. It therefore alters aggregate saving and aggregate housing demand.

6

The Series B projection is based on assumptions about fertility, migration and life expectancy that yield population growth rates that are in the middle of those of the three series.

Ross S. Guest

444 Figure 2: p.a.)

Housing Demand and Population of 25–34 Year Olds (Growth Rate, %

5

4

25-34 yr olds H.demand

3 2 1

52 20

42 20

20 32

20 22

12 20

20

02

19 92

19 82

72

-1

19

19 52 19 62

0

-2 -3

Figure 3:

Aggregate Housing Demand and Private Saving 0.19 saving H.demand

ratio to aggregate income

0.18 0.17 0.16 0.15 0.14 0.13 0.12 0.11

47 20

37 20

27 20

17 20

07 20

97 19

87

77

19

19

67 19

57 19

19

47

0.1

Figure 3 illustrates the simulated effects of population ageing on housing demand and optimal private saving, each as a proportion of income. The saving rate peaks in the period 2002–2007 and the period after 1992 exhibits a hump-shape in the saving rate. This reflects the movement of the babyboomers through their high-income period (where they save a high

Population Ageing, Housing Demand and House Prices

445

proportion of their income), to their retirement (when they draw down their savings). From the peak, the saving rate declines monotonically as the babyboomers begin to retire and draw down their savings. The decline in the saving rate from 2007 until 2052 is 6 per cent. Housing demand, as a proportion of income, also reveals a hump-shaped pattern but the peak is earlier, between 1977 and 1987, which coincided with the peak in the share of population of homebuyers who are the 25–34 year olds. Between 2002 and 2052 housing demand falls by 2.6 per cent of income. This is not to say that optimal housing demand falls after its peak in 1987 but that it falls as a proportion of income. During this period housing demand still grows,7 but not as fast as income. The fact that housing demand growth lags income growth is explained by the projected growth of the population of housebuyers lagging behind the growth rate of medium and high income-earning age groups. A decline in the ‘housing demand to income’ ratio of 2.6 percentage points — from 13.3 per cent of income to 10.7 per cent of income — is a reduction in housing demand relative to income of 20 per cent over the whole period from 2002 to 2052. This requires some elaboration. In the absence of demographic change, in 50 years from 2002, both housing demand and income would be 210 per cent of their levels in 2002, given the assumed annual growth of labour productivity of 1.5 per cent. With demographic change, however, in 2052 housing demand and income will only be 174 and 140 per cent, respectively, of their 2002 levels. Hence, the increase in housing demand has lagged the increase in income by 20 per cent ( ¼ (174140)/190). On an average annual basis over 50 years, this lag is small — less than half of 1 per cent. Finite Supply Elasticity In the long run the supply of housing structures is highly elastic — in which case the assumption of infinite supply elasticity is reasonable for housing structures. However, the supply of land and structures taken together, as they are in this model, is not infinitely elastic even in the very long run, because the supply of land for housing is not infinitely elastic. With a finite supply, elasticity changes in housing demand cause changes in house prices which feedback to the quantity of housing demanded — resulting in a smaller change in quantity demanded than in the case of an infinite supply elasticity (see Figure 4).

7

In the absence of demographic change the demand for all consumption goods would grow at the growth rate of income that is equal to the growth rate of labour productivity (assumed to be 1.5 per cent per annum — see Appendix A).

Ross S. Guest

446 Figure 4:

Housing Demand with Finite Supply Elasticity 0.16 elast=inf

ratio to aggregate income

0.15

elast=2 elast=0.5

0.14 0.13 0.12 0.11

Table 2:

47 20

37 20

27 20

17 20

07 20

97 19

87 19

77 19

67 19

57 19

19

47

0.1

Private Saving, Housing Demand and House Price Ratio to private income Tenure choice Private saving Housing purchase House prices % change from 2002 to 2052 Buy Sell 2002 2052 2002 2052

No demog. Change Base: s. elast. ¼ inf. s. elast. ¼ 2.0 s. elast. ¼ 0.5 a

30 30 30 30

80a 80 80 80

0.148 0.182 0.182 0.182

0.148 0.119 0.119 0.119

0.135 0.133 0.130 0.128

0.135 0.107 0.112 0.116

0 0 5.1 9.4

Denotes buy at age 30, sell at 80.

Simulations are run for supply elasticities of 2.0 and 0.5, with 0.5 being at the low end of elasticities found in the literature and 2.0 being an intermediate case. With a supply elasticity of 2, a one per cent increase in house prices induces a 2 per cent increase in quantity of housing supplied. This implies that in equilibrium, where quantity demanded equals quantity supplied, a two per cent increase in house prices is associated with a one per cent increase in the quantity of housing demanded and supplied. As Table 2 indicates, an elasticity of 2 dampens the drop in the quantity of housing demanded and implies a drop in house prices of 5.1 per cent. In the extreme case of an elasticity of 0.5, the drop in house prices is 9.4 per cent

Population Ageing, Housing Demand and House Prices Figure 5:

447

Real House Prices

1.15 elast=2

1.1

elast=inf elast=0.5

1.05 1 0.95 0.9

47 20

37 20

27 20

17 20

07 20

97 19

87 19

77 19

67 19

57 19

19

47

0.85

(see Figure 5). A sensitivity analysis under alternative parameter values, discussed in Appendix B, revealed the maximum drop in house prices (based on an elasticity of 0.5) to vary by 1 per cent plus or minus the 9.4 per cent reported in Table 2 (see Table B2). Let us say, then, that a drop in house prices of 10 per cent would be the maximum demographic-induced change in house prices that we could expect between 2002 and 2052.

5. Conclusion How does this square with the prediction by M-W (1989) that house prices would fall 40 per cent in the U.S. as a result of population ageing? Apart from the fact that the analysis here refers to Australia rather than the U.S., a valid comparison requires us to calculate the demographic-induced fall in house prices from their peak in 1987 rather than from 2002 (see Figure 5). The fall in house prices from peak in 1987 to trough in 2042 is 27 per cent. This is a good deal closer to the M-W prediction and, on this basis, one might conclude that the M-W prediction seems extreme but perhaps not ridiculous. It is important to reiterate that these changes in housing demand and prices are occurring gradually over a period of 50 years, which amounts to a virtually imperceptible change in any one year. It is also important to emphasise that housing demand does not fall in absolute terms in these simulations — due to the assumed growth in labour productivity, which ensures that demand for and supply of all consumption goods, including housing services, continues to grow. The falls in housing

448

Ross S. Guest

demand reported here are falls relative to income and therefore relative to other consumption goods. It is also worth repeating that a model as simple as this is always subject to important qualifications because by its very nature it abstracts from a number of real world phenomena affecting the housing market. In particular, the asset demand for land is not modelled; there is no uncertainty in the model and no myopia on the part of households; households cannot trade up to a bigger house as their incomes increase in their 40 s and 50 s, nor can they trade down to a smaller house as they move into their 60 s and 70 s. Despite these simplifications the simulations reported here demonstrate some insights from treating housing demand and saving simultaneously in a life-cycle framework.

Acknowledgement The author wishes to thank Ian McDonald from the University of Melbourne for his helpful comments on a draft of this paper.

References Alm, J. and Follain, J.R. (1982). Alternative Mortgage Instruments: Their Effects on Consumer Housing Choices in an Inflationary Environment. Public Finance Quarterly, 10(2), 134–157. Alperovitch, G. (1995). The Baby Boom, the Baby Bust and the Housing Market: A Further Look at the Debate. The Annals of Regional Science, 29, 111–116. Artle, R. and Variaya, P. (1978). Life Cycle Consumption and Homeownership. Journal of Economic Theory, 18, 38–58. Australian Bureau of Statistics (ABS). (2003). Australian Social Trends, ABS, Canberra. Bourassa, S.C. (1994). Immigration and Housing Tenure Choice in Australia. Journal of Housing Research, 5(1), 117–137. Bourassa, S.C. (1995a). The Impacts of Borrowing Constraints on Home-ownership in Australia. Urban Studies, 32(7), 1163–1173. Bourassa, S.C. (1995b). A Model of Housing Tenure Choice in Australia. Journal of Urban Economics, 37, 161–175. Caplin, A. and Joye, C. (2002). A Primer on a Proposal for Global Housing Finance Reform. Menzies Research Centre (www.mrcltd.org.au). DiPasquale, D. and Wheaton, W.C. (1994). Housing Market Dynamics and the Future of Housing Prices. Journal of Urban Economics, 35, 1–27. Ermisch, J. (1996). The Demand for Housing in Britain and Population Ageing: Microeconometric Evidence. Economica, 63, 383–404. Ermisch, J. and Jenkins, S. (1999). Retirement and Housing Adjustment in Later Life: Evidence from the British Household Panel Survey. Labour Economics, 6, 311–333.

Population Ageing, Housing Demand and House Prices

449

Green, R. and Hendershott, P.H. (1996). Age, Housing Demand, and Real House Prices. Regional Science and Urban Economics, 26, 465–480. Hamilton, B.W. (1991). The Baby Boom, the Baby Bust, and the Housing Market: A Second Look. Regional Science and Urban Economics, 21, 547–552. Hayashi, F., Ito, T. and Slemrod, J. (1988). Housing Finance Imperfections, Taxation, and Private Saving: A Comparative Simulation Analysis of the United States and Japan. Journal of the Japanese and International Economies, 2, 215–238. Hendershott, P.H. (1991). Are Real House Prices Likely to Decline by 47 Per cent? Regional Science and Urban Economics, 21, 553–563. Holland, A.S. (1991). The Baby Boom and the Housing Market. Another Look at the Evidence. Regional Science and Urban Economics, 21, 565–571. Littrell, C. (1999). Australian Investment Income Taxation: The Worst of All Worlds. Economic Papers, 18(2), 49–60. Loundes, J. (2001). Household Saving Behaviour in Australia. Economic Papers, 20(1), 67–84. MacFarlane, I. (2003). Do Australian Households Borrow Too Much? Reserve Bank of Australia Bulletin, (April), 7–16. Mankiw, N.G. and Weil, D.N. (1989). The Baby Boom, the Baby Bust, and the Housing Market. Regional Science and Urban Economics, 21, 573–579. McFadden, D. (1994). Demographics, the Housing Market, and the Welfare of the Elderly, in Wise, D. (ed), Studies in the Economics of Ageing, The University of Chicago Press, Chicago. Miles, D. (1992). Housing Markets, Consumption and Financial Liberalisation in the Major Economies. European Economic Review, 36, 1093–1136. Ohtake, F. and Shintani, M. (1996). The Effect of Demographics on the Japanese Housing Market. Regional Science and Urban Economics, 26, 189–201. Poterba, J.M. (1984). Tax Subsidies to Owner-Occupied Housing: An Asset Market Approach. Quarterly Journal of Economics, 99, 729–752. Poterba, J.M. (1991). House Price Dynamics: The Role of Tax Policy and Demography. The Brookings Papers on Economic Activity, 2, 143–203. Skaburskis, A. (1999). Modelling the Choice of Tenure and Building Type. Urban Studies, 13, 2199–2215. Slemrod, J. (1982). Down-Payment Constraints: Tax Policy Effects in a Growing Economy with Rental and Owner-Occupied Housing. Public Finance Quarterly, 10(2), 193–217. Swan, C. (1995). Demography and the Demand for Housing. A Reinterpretation of the Mankiw-Weil Demand Variable. Regional Science and Urban Economics, 25, 41–58. VanderHart, P.G. (1998). The Housing Decisions of Older Households: A Dynamic Analysis 7. Journal of Housing Economics, 21, 21–48.

Ross S. Guest

450

Appendix A This Appendix describes the technical aspects of the model, which is a variation on the model in Hayashi et al. (1988). The household’s problem is to maximise with respect to t(b), t(s), {c(t), t ¼ 1,y,13}, {h(t), t ¼ 1,y,13}, H, tðbÞ X

UðcðtÞ; hðtÞÞ þ

t¼1

tðsÞ X

ZðcðtÞ; HðtÞÞ

t¼tðbþ1Þ

þ

13 X

UðcðtÞ; hðtÞÞ þ F ðBÞ

ð1Þ

t¼tðsþ1Þ

subject to Að0Þ ¼ 0 AðtÞ ¼ ð1 þ Rð1  tÞÞAðt  1Þ þ yðtÞ  cðtÞ  Pr Ph hðtÞ þ INHðtÞ  0

t

¼ 1; . . . ; tðb  1Þ; s þ 1; . . . ; 13 AðtÞ ¼ ð1 þ Rð1  tÞÞ Aðt  1Þ þ yðtÞ  cðtÞ  Pr Ph hðtÞ  dPh H þ INHðtÞ  0

t

¼ tðbÞ AðtÞ ¼ ð1 þ Rð1  tÞÞAðt  1Þ þ yðtÞ  cðtÞ  V ð1  dÞPh H þ INHðtÞ  JðtÞ  Ph H

t

¼ tðbÞ þ 1; . . . ; tðsÞ  1 AðtÞ ¼ ð1 þ Rð1  tÞÞAðt  1Þ þ yðtÞ  cðtÞ  V ð1  dÞPh H þ Ph H  JðtÞ þ INHðtÞ  0 t ¼ tðsÞ

where, A(t) is the end-of-the-period financial asset value; y(t) and c(t) are labour income (net of taxes and transfers) and consumption in period t, respectively; h(t), the size of a rental unit (which could vary every period); H, the size of an owner-occupied unit (which remains constant once purchased) and the units of H are defined such that the price of H is normalised to one. The flow of housing services is assumed to be proportional to the size of H and h; B, the bequest;

Population Ageing, Housing Demand and House Prices

451

INH(t), the inheritance which is assumed to be received in a lump sum at age 55 on the assumption that heirs are 30 years younger than their parents;8 V represents the equal payments of interest and principal such that the mortgage is paid off at maturity of the mortgage; Pr is the price of a rental unit relative to the price of a house; R, the before tax interest rate; Ph, the price of a house and is normalised to one by defining the unit of H accordingly; J(t), the amount of outstanding mortgage debt; t and d represent the constant tax rate on return from saving and the required down-payment ratio, respectively. When a house is purchased with a down payment, d, of the house value, the down-payment expenditure is deducted from income of the period of house purchase. The mortgage debt (1d) becomes (1+R)(1d) at the beginning of the next period. An equal payment of V for m periods amortises the mortgage debt. When a house is sold, the value of the house, less remaining mortgage, is used for consumption after the period of the sale. The inequality restrictions on A(t) imply that the sum of financial and real wealth cannot be negative. In other words, households cannot hold net financial liabilities in excess of the equity in their homes which is equal to J(t)PhH.9 The lifetime plan at age 20 is made on the basis of perfect foresight regarding future income, parameter values and death at age 85. The plan consists of values for non-housing consumption, c, housing services, h, in each five-year period, and the size, H, of owner-occupied housing and the timing of its purchase and sale, t(b) and t(s), respectively. Aggregate variables at time t are a cross section of the per-household variables summed over i generations. For example, aggregate income at time t is the sum of yi,t, i ¼ 1,y,12. The first year of interest for calculating aggregate variables is 2002. The plan for the cohort aged 80–85 years in 2002 was made in 1937

8 Given that population growth, the inheritance, INH, per person is less than the Qt bequest, B, per person. Hence, INHt ¼ B k¼1 ð1 þ nk Þ where k is the growth rate of population in period k. 9 It has become more common for households to borrow against housing equity through financial products such as home equity loans and mortgages with redraw facilities. A reverse mortgage is a similar product in that a bank advances a loan against security on a property; the difference is that both accrued interest and principal are repaid in one lump sum on either the sale of the property or death of the borrower.

Ross S. Guest

452

when that cohort was 20 years of age. Hence five-year plans must be calculated from 1937. Following Hayashi et al. (1988), Pr ¼ R due to the implicit arbitrage condition between financial asset investment and rental property investment.

Functional Forms, Data and Parameter Values The following functional forms are adopted:10 Uð Þ ¼ bt1 ½ðcðtÞÞac ðhðtÞÞah 

(2)

Zð Þ ¼ bt1 ½ðcðtÞÞac ðgH Þah 

(3)

F ðBÞ ¼ b12 Bu

(4)

where g represents the pride-of-ownership coefficient, and ac þ ah o1; implying decreasing marginal utility with respect to total consumption. The values of ac and ah are determined as follows. Intratemporal optimisation implies that expenditure on rent and non-housing consumption are linearly related. This is derived as follows: @U=@cðtÞ ac hðtÞ 1 ¼ ¼ @U=@hðtÞ ah cðtÞ Pr where Pr ¼ R due to arbitrage between the financial and the real asset given constant housing prices which rule out capital gains (the latter condition is relaxed in Section 5.2). Therefore, Pr h ¼

ah c ac

(5)

Hence expenditure on rent, Prh, is a linear function of non-housing consumption expenditure. If we assume that expenditure on rent is 25% of nonhousing consumption expenditure — that is, Prh ¼ 0.25c — then ah ¼ 0:25ac

10

(6)

Hayashi et al. (1988) also adopt a Cobb–Douglas form, but they set (implicitly) ac ¼ 1 and ah ¼ 0.15. The justification for our chosen values is given in the text.

Population Ageing, Housing Demand and House Prices

453

Further, assuming that the elasticity of marginal utility with respect to c is 0.5,11 then ac þ ah ¼ 0:5

(7)

Equations (6) and (7) imply that ac ¼ 0.4 and ah ¼ 0.1. These are approximately equal to the values adopted by Slemrod (1982). He adopts ac ¼ 0.417 and ah ¼ 0.083. Hayashi et al. (1988), on the other hand, assume that ah ¼ 0.15 and ac ¼ 1. These values do not seem as plausible as those above. The values in Hayashi et al (1988) imply, given their value of R ¼ 0.5 and the first order conditions below, that h ¼ 0.3c and that expenditure on rent is equal to 0.15 of expenditure on the other consumption good, c. This seems too low, at least for Australia. Their values, because they sum to greater than one, also imply increasing marginal utility with respect to consumption. The rate of time preference is given by b which, in conjunction with the rate of interest, affects the rate at which consumption grows over time. The Euler equation that describes the optimal consumption path implies that, between any two periods in which a house is owned, consumption growth is given by12 cðtÞ ¼ ½ð1 þ Ra Þb1=ð1ac Þ cðt  1Þ

(8)

If consumption growth were equal to the growth rate of labour income per capita, the implied annual rate of time preference would be about 1.7 per cent. This assumption is not, however, required to achieve a stable aggregate consumption to income ratio because there are overlapping generations. We adopt the slightly higher rate of 3.0%, which is also the rate adopted by Hayashi et al. (1988), on the basis that it gives a rate of private saving that is not too high by current and historical standards. The value for u is chosen arbitrarily with the goal of achieving a target value of the bequest, B, using the first order condition in the final period (where t ¼ 13). In the base case, the target value of the bequest equal to 100% of the value of the house when it was owned by the household. The full list of parameter values is given in Table A1. 11

This elasticity can be found by substituting (5) into (2). Equation (7) applies only for any two periods in which a house is owned. The corresponding equation is slightly different for periods in which housing is rented or for periods in which the household switches from renting to owning or vice versa. However, our simulations (and those of Hayashi et al.) show that households generally will choose to own for most of their adult lives. This justifies our focus on the period of home-ownership for the purpose of determining the value of b. The value of b is constant throughout the lifetime.

12

Ross S. Guest

454

Table A1: Parameters and their Values Parameter

Value

Annual growth rate of labour income per capita, g Annual interest rate (based on the five year interest rate, R) Annual rate of time preference (implied by the five year rate, b)a Required deposit ratiob, d Size of bequest, B, as a proportion of equity in the house at disposal Elasticity of marginal utility with respect to housing consumption, ah Elasticity of marginal utility with respect to non-housing consumption, ac Tax rate on income from financial assets, t Pride-of-ownership coefficientc, g Maturity of the mortgaged, m

0.015 0.04 0.03 20% 100% 0.1 0.4 0.3 1.1 4 (20 years)

a

Hayashi et al. (1988) implicitly assume an annual rate of time preference of 0.029. The typical deposit ratio for Australia is between 10 and 20 per cent (ABS Cat. 4182.0). Hayashi et al. used 25 per cent. c Hayashi et al. (1988) used 1.4. d Hayashi et al. (1988) assume that mortgages are for 30 years, but households in Australia typically pay off their mortgages in about 20 years (based on author’s calculations using data on monthly repayments obtained from the Housing Industry Association of Australia and data on average borrowing size from ABS Cat. 5609.0). b

The aggregate, cross-section, values of endogenous variables of interest – consumption, saving and owner-occupied housing purchases are defined for period t as follows, where i refers to generation i alive in period t: Housing consumption: 13 X ðPr hði; tÞ þ RJði  1; t  1ÞÞN ði; tÞ

(9)

i¼1

Non-housing consumption: 13 X

cði; tÞNði; tÞ

(10)

i¼1

Private saving:13 13 X

ðyði; tÞ þ RAði  1; t  1Þ  cði; tÞ  Pr hði; tÞ

i¼1

þ RJði  1; t  1ÞÞNði; tÞ

13

ð11Þ

There is no distinction in the model between private and household saving, because the amount that households choose to save through corporate entities is not modelled.

Population Ageing, Housing Demand and House Prices

455

where y(t), private income, is 13 X ð yði; tÞ þ RAði  1; t  1ÞÞNði; tÞ

(12)

i¼1

Housing demand14: 13 X

Hði; tÞNði; tÞ

(13)

i¼1

A Finite Supply Elasticity In this model housing supply, HS(t), is equal to the flow of housing investment and is described by the following equation:15 H S ðtÞ ¼ bðtÞPch ðtÞ

where bðtÞ; c40;

(14)

Equation (14) says that gross investment in housing is an increasing function of both the price of housing and a shift variable, b(t), which grows at the rate of growth of aggregate labour income; the latter is the sum of the exogenous rate of growth of labour productivity and the growth of age-weighted employment. Households are assumed to be fully forward looking and therefore equate the implicit rental rate on owner-occupied housing, Pr, with the user cost of capital, R, minus the capital gain:   Ph ðt þ 1Þ  Ph ðtÞ Pr ðtÞ ¼ R  (15) Ph ðtÞ where [ ] is the capital gain at time t (since the model assumes perfect foresight, there is no need to model expectations). Ph(t) is determined by equating aggregate housing demand (13) with aggregate housing supply (14). The model implies that households correctly anticipate the effect of future demographic change on Ph(t) and the rental rate, Pr(t).16 14

The term ‘‘housing demand’’ refers to the purchases of owner-occupied housing which, in period t, is the sum of housing purchases, H, planned to take place in period t by cohorts aged 20 years and over in period t. 15 Unlike the M-W model, there is no depreciation, so gross investment equals net investment. 16 M-W compare the forward-looking model with a naive expectations model in which market participants do not expect any effect of demographic change on house prices. The naive expectations model results in bigger increases in house prices. Here we apply only the forward-looking model to ensure time-consistent planning by households with respect to their lifetime consumption and tenure choice plans as described in Section 3.

Ross S. Guest

456

A range of values of the housing supply elasticity parameter, c, have been adopted in the literature. A low value is 0.5 and a typical value is 2.0. We conduct simulations with these values and compare the results with those reported above for the infinite elasticity case.

Appendix B This appendix discusses the sensitivity of the results to alternative parameter values. Sensitivity with respect to the supply elasticity parameters was discussed in the text. Further sensitivity analysis was conducted by varying, one at a time, each of the following parameters in the model: the deposit on the housing purchase, the maturity of the mortgage, the size of the bequest, the ‘‘prideof-ownership’’ coefficient, the elasticities of marginal utility of c and h, and the tax rate on saving and financial assets. The effects are reported in Table B1 by showing the optimal saving and housing purchases in 2002 and 2052. The optimal timing of the sale of the house is unaffected by the choice of parameters. That is, it is optimal in every case to retain owner–occupier status for as long as possible after the purchase. The timing of the housing purchase differs for some alternative parameter values. However, except for the very extreme case of a 35 per cent deposit, the optimal purchase occurs within one period of the optimal period of purchase in the base case. This suggests that the tenure pattern is quite robust to alternative parameter values. Table B1: Optimal Tenure Choice, Private Saving and Housing Demand Purchase Sell house house at age at age

Ratio to private income Private saving

b

Base case Mortgage maturity ¼ 25 years Deposits ¼ 10% Deposits ¼ 35% Bequest ¼ 200% of house Bequest ¼ zero Pride-of-ownership ¼ 1.5 Rent ¼ 15% of other con. Rent ¼ 40% of other con. Elasticity of con. ¼ 0.2 Tax rate ¼ 30% a

30 30 25 35 30 30 25 30 30 35 30

80 80 80 80 80 80 80 80 80 80 80

Housing purchasesa

2002

2052

2002

2052

0.182 0.179 0.172 0.185 0.195 0.172 0.175 0.190 0.181 0.171 0.175

0.119 0.117 0.117 0.119 0.135 0.097 0.113 0.120 0.120 0.117 0.115

0.133 0.130 0.140 0.110 0.113 0.159 0.127 0.136 0.130 0.120 0.127

0.107 0.106 0.112 0.088 0.091 0.130 0.101 0.109 0.105 0.098 0.103

Housing purchases are in real terms; hence when the supply elasticity is infinite, real ¼ nominal purchases. b In the base case simulation the supply elasticity is infinite.

Population Ageing, Housing Demand and House Prices

457

Altering the required deposit has the expected effects on saving and housing demand. That is, a lower deposit increases housing demand and reduces the saving rate and a higher deposit has the opposite effects. For the 10 per cent deposit simulation, population ageing reduces the saving rate by only 0.8 per cent between 2002 and 2052, compared with 1.5 per cent in the base case (20 per cent deposit). The effect on housing demand for the 10 per cent deposit simulation is very slightly larger (by 0.3 per cent) than in the base case. The simulation for a larger deposit (35 per cent) has the opposite effects — it increases the effect on the saving rate and lowers the effect on housing demand compared with the base case. Sensitivity to the bequest is not as great as might be expected. The simulated variations to the bequest were quite large — an increase from 100 per cent of the value of the house to 200 per cent in one simulation and a reduction to zero per cent in another simulation. The size of the bequest has two offsetting effects on saving, as pointed out by Hayashi et al. (1988, p. 230). A higher bequest implies a higher inheritance of the next generation. Therefore the need to raise savings to accumulate the bequest is reduced by the boost to wealth from a higher inheritance. Recall from above that the size of the bequest is linked to the size of the house by setting the parameter, m, in the utility function accordingly. Hence a forward-looking household can effectively adjust the size of the bequest by adjusting the size of the house to buy; this in turn affects the lifetime consumption and saving pattern. The outcome is that a bequest equal to 200 per cent of the value of the house lowers the level housing demand and the zero bequest raises the level housing demand at all stages of the projection period compared with the base case. That is to say, there is no difference in the impact of ageing on housing demand under the alternative bequests. However the effect of ageing on the saving rate differs for alternative bequests. The higher bequest reduces the impact of ageing on private saving and the lower bequest increases the impact of ageing on private saving. A larger ‘‘pride-of-ownership’’ parameter brings forward the housing purchase by one period (5 years), which reduces housing demand and the saving rate. The interpretation of this result is that because ownership is more highly valued it is optimal to purchase the house earlier, but the liquidity constraint means that the size of the purchase must be somewhat smaller. The effect of ageing on housing demand and private saving is virtually unaffected by the larger pride-of-ownership parameter. Recall from Section 3 that, the relative size of the parameters ac and ah imply a certain ratio of rent expenditure to other expenditure in household budgets. In the base case this was assumed to be 25%. Table A1 indicates that varying this ratio to either 15 per cent or 40 per cent has a very small effect on housing demand and private saving — less than 1 per cent of income in 2002. Again, there is no difference in the effect of ageing on

458

Ross S. Guest

housing demand and private saving. Finally, lowering the tax rate on the return from accumulating financial assets from 30 per cent to 20 per cent results in a reduction in the private saving rate by 0.8 per cent and a reduction in the demand for housing by 0.6 per cent, in 2002. There is very little difference, compared with the base case, of the effect of ageing on housing demand and private saving for the lower tax rate. The sensitivity analysis reported here is limited in that it considers perturbations in parameters and assumptions one at a time. A stochastic simulation analysis would be able to account for interaction of various perturbations of parameters with each other and with the policy simulations and demographic changes, and assign probabilities to the various outcomes for housing demand and the saving rate. This is left as a possibility for future work. Table B2: Effect on House Prices (Supply Elasticity ¼ 0.5)

% change from 2002 to 2052 Base case Mortgage maturity ¼ 25 years Deposits ¼ 10% Deposits ¼ 35% Bequest ¼ 200% of house Bequest ¼ zero Pride-of-ownership ¼ 1.5 Rent ¼ 15% of other con. Rent ¼ 40% of other con. Elasticity of con. ¼ 0.2 Tax rate ¼ 30%

9.4 9.5 10.2 10.0 9.0 10.5 10.8 9.9 9.8 9.9 9.6

Chapter 18

Population Ageing and State Concessions Policy Karen Piper and Don Siemon Victorian Department of Human Services, Australia

Abstract Victoria is one of six states and two territories within Australia and has its own elected government. There are many areas in which the state has powers and responsibilities that differentiate it from the national Australian Government. The Victorian Government provides a range of concessions to support the spending of lower-income households on state government taxes, charges, and household essentials. Spending on these concessions is biased somewhat to age pensioners, though less so than in some other states. Over the past two decades a number of efforts have been made to develop a better understanding of how low-income households use essential services — such as electricity, gas, and water — and the impact of changes to prices and concessions. Based on a stratified survey of 2,000 households, a Victorian Concessions Model was developed in 1997 to allow modelling of policy alternatives. This survey was repeated in 2001 and a new version of the model was produced. Long-term issues for state concessions policy are targeting; dealing with pressures for extension of entitlement; and demographic changes leading to higher levels of take-up (in particular, population ageing). This paper reports the likely impacts of population ageing on spending on a group of state concessions, contrasting these with the impacts of policy decisions that might influence prices or change entitlement. It concludes that while population ageing will result in higher spending in the longer term than would otherwise be the case, policy changes to broaden entitlement could have a similar long-term — and much greater short-term— impact.

1. Introduction The ageing of the population is commonly projected to lead to higher Government spending, as much larger numbers of older people need income support and other benefits and/or require more expensive health or personal International Symposia in Economic Theory and Econometrics, Vol. 15 r 2007 Published by Elsevier B.V. ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15018-8

460

Karen Piper and Don Siemon

supports. While these impacts have been examined at a Commonwealth level (Treasurer, 2002), there is less attention to the flow on to state and territory governments. This paper explores the use of a static microsimulation model to examine the impacts of an ageing population on one particular category of state spending — a category that might be thought to be particularly prone to ‘blow out’ as the population ages. State concessions are subsidies targeted to lower-income households, to assist with the payment of essential services or state taxes or charges. Out of its total budget of some $25 billion, the Victorian Government contributes over $700 million, either through outlays or through foregone revenue, on these concessional subsidies. Because most of this money goes to older people — age and service pensioners — spending on these subsidies might grow disproportionately as the population ages. This paper examines the likely impacts of population ageing on state concession spending over the medium to long term — and contrasts these with the impacts of possible pricing changes and policy decisions that narrow or broaden entitlement. Before this, however, the paper provides some details of state concessions in Victoria; examines their role as a tool of social and economic policy; outlines efforts in Victoria to develop a stronger evidence base for concessions design; and introduces the model developed to assist decision-making.

2. State Concessions in Victoria The objective of Victorian state concessions is to improve the affordability of key services — those upon which people rely to maintain their housing, health, mobility, and their participation in education and social life. Around 700,000 households benefit from Victorian Government concessions each year. Pensioners, unemployed families and single people, incapacitated veterans and war widows, and other people living on low- and fixed-incomes gain through lower taxes, charges, or bills. Concessions are generally directed to people who have been assessed by the Commonwealth Government as having a sufficiently low income to qualify for a pensioner concession card (PCC) or a health care card (HCC). They are also provided to holders of some Department of Veterans’ Affairs (DVA) and repatriation health cards (gold cards). It is estimated that some 1.26 million Victorians hold a concession card. Table 1 lists the range of state concessions funded either as outlays or as revenue foregone by the Victorian Government. The table also indicates to whom concessions are targeted, the number of households or individuals benefiting, and the cost to Government. Victoria has a greater degree of centralized budget control and administration for state concessions than some other states. In Victoria, most

Population Ageing and State Concessions Policy Table 1:

461

Major Victorian State Concessions, 2001–2002

Applying to Water and sewerage Municipal rates Electricity Gas LPG and other domestic fuels Public transport fares Vehicle registration TAC (insurance) Stamp duty Education maintenance allowance Preschool concessional subsidy Ambulance subscriptions Health (dental and allied)

Eligible cardholders

Number of beneficiaries

Expenditure ($/million) 60.3 53.2 37.2 31.6 1.8

PCC PCC PCC PCC PCC

and HCC only and HCC and HCC and HCC

516,218 households 396,300 households 685,231 households 531,430 households 22,560 households

PCC PCC PCC PCC PCC

and HCC and HCC only and HCC and HCC

— 727,209 vehicles 503,466 vehicles 2,536 recipients 146,777 children

PCC and HCC

17,146 children

PCC and HCC PCC and HCC

Unknown Unknown

61.5 100 (est.) 75.9 3.2 40.1 4.2 169.0 113.4

concessions involving actual outlays — transfers of funds to a supplier or an individual —are funded through the Department of Human Services budget and administered by the Concessions Unit, which publishes details annually (Victorian Department of Human Services, 2002, Melbourne). About 60 per cent of Victoria’s concessions contribution is handled in this way. The Victorian Minister for Community Services has responsibility under the State Concessions Act 1996 for administration of these concessions and their targeting. However, where departments or agencies directly forego revenue so as to offer concessional fees or taxes — such as discounted motor registration fees — administration generally remains with the relevant bodies. Details of concessions classed as tax expenditures are also provided in a statement within the budget papers (Victoria, 2003, Budget Paper no. 1). Together, concessions provided as outlays and revenue foregone has accounted for around 2.5 per cent of Victorian spending over the past decade. Centralization of budget and administrative control over concessions has brought significant benefits. Eligibility criteria have gradually become increasingly modernized and consistent. Specialized policy advice is available to government. Administration of concessions has become streamlined and responsive to changing priorities — and administration of concessions has been linked to provision of other forms of specialized assistance (utility relief grants, financial counselling services).

462

Karen Piper and Don Siemon

3. State Concessions as a Social and Economic Policy Tool Administrative centralization brings a risk that the role of concessions can be misunderstood. Instead of being recognized as particular solutions to particular problems, they can be viewed as a single homogenous program — equivalent to funding for preschools or mental health services. The purpose of such an undifferentiated program is then understood to be simply redistributive; concessions are construed as a way in which state governments seek to alleviate poverty or reduce inequality. Given that state governments lack both the taxing and income support powers of the Commonwealth, can concessions really make that much difference? The debate over the value of concessions, in terms of poverty alleviation or redistribution, is important. Concessions are clearly a powerful redistributive tool, because they do target benefits solely to the low-income part of the population.1 However, their contribution to reducing poverty, inequality, and disadvantage is found not only in the overall pattern of their use but also in their particular areas of application. Their flaws, like their histories, are particular, rather than general. Poor design and lack of adaptation to changed circumstances are the real problems that limit their effectiveness. Viewing state concessions as a single homogenous program is not helpful to understanding their real value, nor their limitations, as a social

1 Concessions are directly intended to be helpful to those who have little money. However concessions are not designed to redress income poverty as such (i.e., to direct benefits to those people whose relative incomes are lowest, perhaps due to the inadequacies of the social security system). This has been acknowledged in past reviews of concessions in Victoria. Rather they are a way in which governments respond to other dimensions of poverty: unfair hardship, too great a degree of inequality and a lack of opportunity. Underpinning public concern about poverty is the expectation that people should have a reasonable standard of living — that consumption by lower-income households is not too detached from that of the rest of the population. In Australia, creating this situation has traditionally been seen as a matter of decent wages, housing, services and expenditure support, as much as of income support. An ‘antipoverty’ ethos is embedded in the whole matrix of social spending undertaken by governments, and poverty researchers have long appreciated the importance of accessible generalist universal services in protecting people from the worst hardship associated from poverty (such as untreated illness, illiteracy or discrimination). In general, there are independent policy rationales for many of these services — public health or the need for a high skill base for economic success — but they nevertheless have an important place in ameliorating or preventing poverty, and this is acknowledged at least tacitly in policy development.

Population Ageing and State Concessions Policy

463

and economic policy tool. Concessions are better understood as a technique of government than as a service. They are more like a tax than a hospital. Concessions are a type of expenditure support2 provided by governments that are targeted just to low-income households — as opposed to other forms of assistance, which lie elsewhere on the spectrum between universal (such as public hospitals) to residual (emergency grants for the destitute). Moreover, concessions employ a simple means of targeting compared with either discretionary grants or the more graduated scales used in determining entitlement to expenditure support in public housing, legal aid, or childcare. Concessions are a tool employed by all levels of government. Indeed, it is the Commonwealth that issues the cards used to determine eligibility for concessions, as a way of targeting higher levels of expenditure support for pharmaceuticals and some other health benefits. However, it is not hard to see why state governments have found them particularly useful. Generally, state governments: 







are directly engaged in delivery or direct funding of a wider range of services to households or individuals: expenditure support decisions are common; have less access to detailed information on the income or on other circumstances of individuals or households; have a smaller population and are attracted to administratively simpler forms of targeting, since these demand less capacity to monitor or test entitlement; and are not able to directly target offsetting benefits to individuals or households by way of income support or income tax relief when raising taxes or charges.

Concessions are an effective tool for providing benefits of relatively low-dollar value to households or individuals, because they are offered to an easily identified group, generally through a discount on a bill. They are

2

On this analysis, state concessions can be seen generally as just one form of expenditure support, whereby Government subsidies that which would otherwise be paid for by individuals or households. Expenditure support is distinct from: 



Broader social provision of services, which are available for use but where there is no visible transfer of a financial benefit to the household and Income support, in that benefits are conditional or focused on households undertaking a particular type of spending rather than being available for any use.

464

Karen Piper and Don Siemon

readily understood and widely accepted.3 Design and administrative simplicity allow transaction costs to be minimized — in Victoria delivery of state concessions costs less than three per cent of the benefit transferred.4

4. Domains of Application As a tool, concessional subsidies can be put to a range of different purposes. In Victoria, there seem to be four distinct purposes: As a tool of tax design, concessions are used to offset the inherent regressivity of many state taxes and charges. For example, concessions are used to reduce the impact of municipal rates, some stamp duties, and motor vehicle registration fees. As a tool for service design, concessions help provide more equal access to important public services. Health services required to recoup some of their costs usually charge a much lower fee to concession cardholders. As a tool to support participation in compulsory or central social institutions, a concessional payment — the education maintenance allowance — is provided to assist families with costs associated with schooling. As a tool to make privately purchased essential goods or services more affordable for lower-income households, so as to ensure that their consumption is able to remain at community norms. Examples here are rebates on gas, electricity, water, and, at the Commonwealth level, the concessional rate of the pharmaceutical benefits scheme. As governments have moved the provision of public goods out of the public domain, and market or cost-reflective pricing has replaced more uniform tax-like pricing5 so these purposes have become more distinct. 3

A 2001 telephone survey of concession cardholders by Roy Morgan Research (2001) and commissioned by the Department of Human Services found that cardholders are overwhelmingly positive about the value to them of state concessions. Almost all recipients valued highly each concession received. Energy and water concessions were considered as the most important; just over half rated the vehicle registration concession as one of the three most valuable. Respondents generally felt that it was the whole package of concessions that made a difference to them. 4 Where delivery is via a private supplier (as with electricity and gas) under a community service obligation (CSO), the Victorian Government refunds the supplier’s costs of delivery. Where delivery is by a public sector agency, these costs are often absorbed by the agency. 5 Water is illustrative. Almost all revenue to fund collection and provision of water to households and the removal of sewerage and maintenance of storm water drainage was formerly collected from a rate paid by property owners — a simple asset tax. Today revenue for almost all these are raised from residents, including tenants, through fixed standing and volumetric charges.

Population Ageing and State Concessions Policy

465

Each different rationale has implications for concession design. For example, a tax concession is one of a number of possible elements or features available to policymakers in designing a particular tax — others are tax thresholds, exclusions from a tax base, differential rates, and so on. Where specific consumption or household-based taxes are employed, the use of a concession can effectively moderate their regressivity. Cardholders as a group have about half the equivalent disposable incomes of noncardholders, so a 50 per cent discount provides a crude approximation to proportionality. By contrast, concessions on household essentials are a tool to moderate general pricing, so that the commodity is more affordable for lower-income households so that: 



use of the commodity by low-income households is not kept below what would be reasonably expected more generally in the community; and there is little risk that households will impoverish themselves in other ways in order to maintain the level of necessary consumption.

Because concessions are a pricing tool, government pricing policies and regulatory frameworks form the starting point for consideration of concessions design. Concessions should moderate but not undercut socially desirable price signals, for example. In this way long-term equity and efficiency goals can be aligned in pricing.6

5. Creating an Evidence Base for Policy Major tax, pricing, and service delivery changes over the 1980s and 1990s created demands for concessions policy options to be better understood and more clearly articulated. Some areas of major public debate included: 

6

electricity and gas pricing, and the need for concessions to protect low-income households from the introduction of new or higher charges prior to privatization;

Concessions have often been introduced in the context of significant real price shocks — for example in the early 1980s, when electricity prices rose sharply in Victoria. They are often used to moderate the impact of new taxes or charges in areas previously provided for no charge. It is important that concessions have a longerterm policy rationale. However, over time the price shock may erode while entitlement to a concessional discount will persist. For this reason, the Victorian Government has a clear policy framework on extension of any concession.

466  

Karen Piper and Don Siemon

the erosion in value of the municipal rate concession; and adaptation of the water concession to reflect major changes in water and sewerage pricing.7

The consolidation of policy and delivery responsibility for the majority of state concessions over this period — within what became the Department of Human Services — gave rise to a need for a greater independent capacity to understand and forecast how policy changes would impact on target households: 





How would different households be affected by pricing changes being proposed? How and why were concessional subsidies affecting recipient households? How well could changes to concession design be used to protect recipients from the distributional impacts of price restructuring?

In order to develop a more secure evidence base for concessions policy, therefore, the department contracted Reark Research in 1996 to undertake a stratified survey of some 2,000 households across Melbourne and major regional centres. For the first time, the department could obtain a clear view of how electricity, gas, water, and rates bills varied within and between age/service pensioner households, other concession cardholding households, and non-concession households. Additional information was also collected on household size, appliance mix, and housing stock — known from previous work to be associated with consumption — and on conservation attitudes and behaviour. Findings from the survey were published and the data made available to interested consumer and welfare organizations. Findings from the survey were influential in policy discussions around the restructuring of the water concession, then being undertaken with input 7

Reflecting the origins of water supply as a responsibility of local government, until the 1980s, water was provided to households with revenue collected by way of property rate linked to property value. Although water use was metered in Melbourne, only very small bills were levied for ‘excess’ water use. A shift to ‘user pays’ took place over the next few years which eventually led, for most households, to the replacement of the water rate with uniform service charges, a flat volumetric charge for water and a higher volumetric charge for sewerage. Reflecting these changes, the water rate concession was broadened to include components for water and sewerage use. These four components are subject to a range of different caps; the complexity of the design makes it very difficult for households to understand.

Population Ageing and State Concessions Policy

467

from non-government organizations.8 However it was clear that the survey could also be adapted to allow different price and concession alternatives to be modelled directly. In 1997, the National Centre for Social and Economic Modelling (NATSEM) at the University of Canberra was contracted to produce a modelling tool for future concessions policy — initially around gas, electricity, and water, and then subsequently around municipal rates. Five years later, in the context of a new Government, and at the end of a long period of falling prices, the department repeated the 1996 survey (Roy Morgan Research for the Department of Human Services, Melbourne, 2001). The new data provided the basis for the development by NATSEM of a new version of the Victorian Concessions Model (VCM).

6. The Victorian Concessions Model The VCM is a static microsimulation model of the Victorian population’s use of household essential services, the annual bills incurred, and Victorian Government concessions expenditure. It is designed for policy analysis — in particular, allowing users to simulate the effects on household bills or the concessions budget of changes to: 

 

prices or price structures in electricity, gas, water, or rates in different parts of Victoria; concession rates or structures applying to these bills; and entitlement to concessions by major card group (PCC, HCC) and housing tenure.

The current version of the model is based on the 2001 survey data, of around 2,000 households. This sample was stratified both by cardholding status (roughly one-third age and service pensioners; one-third other cardholders; and one-third non-cardholders); and geographically (Melbourne and regional towns). Weights assigned to individual households in the sample were adjusted to reflect the known incidence of cardholding across Victoria and populations within particular supply areas. Household Expenditure Survey data were used as a basis for comparison with the 2001 survey data — and outcomes were reconciled with actual Victorian concessions spending estimates. 8

In the event, however, the Victorian Government chose to couple the restructuring of water prices with a major capital injection of $1.6 billion into the water industry in such a way that not only did the bills of homeowners fall substantially, but relatively few households were actually worse off. Consequently, the drive for reform of the water concession diminished.

468

Karen Piper and Don Siemon

The VCM runs under SAS with a user interface allowing manipulation of prices and concessions. The model then estimates the aggregate expenditure by households and the cost of concessions. It then enables comparisons to be made with the base case scenario. Estimates of the number of households who gain or lose through the change and average net costs and benefits can be generated for various combinations of households (by housing tenure, cardholder group, metropolitan, or non-metropolitan location). The VCM does have some limitations of note 











The original survey data excludes households in rural areas and smaller regional towns; some of these households may have quite different consumption patterns.9 Data used to derive population weights, particularly around concession card type and household characteristics are not particularly reliable — and while a reconciliation at a statewide level with administrative data on households claiming concessions provides an aggregate check, the volatility in claim rates between years, particularly for individual suppliers, suggests that data here are of variable quality. Although the user interface allows for a wide range of alternative price and concession rates and structures, it has limits: some alternatives cannot be directly modelled.10 The introduction of private ‘market contracts’ means that a growing, if still small, proportion of households are no longer charged at the supplier’s standard price. The model allows estimates of spending to be projected ahead of the current year, making provision for price inflation and growth in the cardholder population — but projection is limited to five years ahead. The model cannot take account of behavioural changes in consumption of utilities brought about by seasonal variation, significant unknown price increases or improvements in approaches to conservation.

Despite these potential sources of inaccuracy and limits in flexibility, the VCM has proved to be a valuable tool, allowing the Department of Human Services much greater confidence in estimating costs or impacts of 9

The survey data also represent a snapshot of a particular year; they are not based on regularly repeated surveys. In the case of water, household consumption can vary significantly from year to year in response to weather and shortages. Administrative data can also be surprisingly variable. 10 The VCM interface assumes a fixed set of electricity and gas retailers with assigned geographic areas; the same prices are assumed to apply to all households in these areas; variable pricing elements tend to be based on current practice; for example, ‘time of day’ prices are not readily modelled.

Population Ageing and State Concessions Policy

469

alternatives and thereby informing policy. In practise, however, the model is primarily used to indicate the way in which costs or benefits from a changed policy are likely to be distributed. These are then applied against externally derived aggregates.

7. Modelling the Longer-Term The main focus of this paper is on examining the long-term impacts of population ageing on concessions spending. A hybrid approach, using data from various sources, was needed to undertake this work. The revised model was tested to determine its value in exploring longerterm policy issues, as well as to contribute to the ongoing debate on ways for governments to address the challenges of demographic change. It was also used to model the impact of concessions on electricity, gas, water, and rates. Concessions on these items involve outlays of around $180 million a year, representing about one-third of the value of all concessions. In general, government spending on a concessional subsidy is driven by the level of prices of the subsidised good, service, or tax; the growth in the recipient population; and the tightness of targeting. The following sections consider the following in turn: 





the impact of changing electricity, gas, and water prices and municipal rates; the impact of population ageing and the growth in the proportion of concession cardholders; and the impact of policy changes to the targeting of Victorian concessions.

8. The Impact of Changing Prices and Rates Beyond the short term, estimating the future prices of household fuels and water or the level of any individual tax is hazardous. The history of price and tax changes in Victoria over the past 25 years suggests that levels have been driven by exogenous factors and policy change, as much as by underlying cost pressures or productivity improvements. For example, the real price of residential electricity fell steadily in Victoria across the 1960s and 1970s until the emergence of high real-interest rates — a secular change which affected electricity utilities in many parts of the world and occurred at a time of high-capital investment and ‘resource boom’ construction wage rises — forced the electricity authority into sudden and large real price rises at the start of the 1980s. After this price shock, however, residential electricity prices stabilized and then fell in real terms over the rest of the decade.

470

Karen Piper and Don Siemon

In the early 1990s, there was a second step increase in residential electricity prices. While part of this was associated with the depth of the Victorian recession, the rise also reflected government policy of achieving a higher dividend requirement. Costs were redistributed away from business customers and towards households, as part of a move to more cost-reflective pricing associated with disaggregation of the electricity industry and its sale. While residential prices rose by over 10 per cent prior to privatization, real falls were mandated for the coming years by regulations that capped prices — so that, by the end of the 1990s, real prices had fallen to the 1990 level. The following year there was another price shock. This reflected the introduction of the Goods and Services Tax. Following this, Victorian households began to experience some divergence in prices between different parts of the state, as the regulated price paths came to an end and price competition for households began to come into play. Prices for metropolitan households fell once more and a significant gap emerged between prices in the city and the country — a gap that has widened since, despite Government efforts to moderate it. Associated with each of these price shocks was a Government response to reduce the impacts on lower-income households via concessions: electricity concessions were first introduced in the early 1980s; several new elements were put in place in the early 1990s to moderate the impacts of price restructuring; and a concession on what are termed ‘off-peak electricity charges’ was introduced in 2002. This was a pricing tariff previously introduced to encourage use of electricity. Gas and water prices have followed somewhat different paths — but here too there have been quite large price step increases (and sometimes decreases), produced by Government policy decisions and sometimes by exogenous factors. Concessions again changed in response. The recent history of municipal rates is rather the inverse. The underlying real trend has been upwards — reflecting higher real property values and higher incomes — rather than downwards. In the 1990s, however, a real rate cut of around 20 per cent was imposed on local governments as municipalities were consolidated and a program of compulsory competitive tendering of council services was introduced. Rate capping ended with the change of Government in 1999 — and more recently rates have been rising in real terms. One observation from these histories is that while price levels have bounced around, for a variety of reasons they have tended to later return to much the same real level. For this reason a base case scenario for future prices and rates is simply that they move with CPI (in this study assumed to be 3 per cent per annum). However, history also indicates that step changes in price levels are also very likely — and it is useful to indicate the impact of such changes on the

Population Ageing and State Concessions Policy Table 2:

Price Scenarios Modelled Base

Electricity Gas Water Rates

471

Prices Prices Prices Prices

rise rise rise rise

with with with with

CPI CPI CPI CPI

Base plus shock (3% (3% (3% (3%

p.a.) p.a.) p.a.) p.a.)

+30% +20% +10% Rise at

increase in final block increase in overall price increase in volumetric charges CPI+0.5% each year

concessions budget. These are modelled with a number of scenarios. The scenarios chosen certainly do not reflect any Government pricing policy, but are nevertheless plausible in the medium to long term. Table 2 summarizes these scenarios. The growth in summer air conditioning demand may require investment in more expensive peak generating plants which could be manifested in a much higher increase in the electricity price for electricity consumption exceeding a certain level — modelled here as a 30 per cent rise in this block of a ‘standard’ price. Similarly, declining natural gas reserves might over the next 40 years lead to oil and natural gas being significantly more expensive. This is modelled here as a 20 per cent rise in the price. Continuing water shortages may lead to a perception that household water is under priced, and an increase to the marginal price would be a useful signal to consumers, complementing other water conservation measures. This is modelled here as a 10 per cent increase in charges across the state based on the volume of water consumed by households. In terms of municipal rates, the alternative scenario is not so much a shock as a different path. Rates remain the dominant source of local government revenue, and it is a clear growth tax, being linked to increases in land values. It is quite imaginable that rates might rise slightly above inflation over the long term; this is modelled here as 0.5 per cent real annual increase. The impacts of these price changes across 2003–2050 were modelled and are shown in Table 3. The modelling was based on the assumption of no change in the Victorian population (i.e., with no underlying growth in the eligible population), so that changes are the results of inflation and price effects demonstrated in the scenarios alone. An annual increase in the CPI of 3 per cent was assumed. Spending in 2003 was predicted from the VCM model. The table also shows the real value of future spending on each concession (in $2003) and this spending as a proportion of total government spending on residential bills for each service or tax. For electricity and gas, simulation of the price shock was undertaken using the VCM but the task of projecting the inflation in electricity and concessions spending could be done directly, since key concessions are proportional to

472 Table 3:

Karen Piper and Don Siemon Impacts of Price Inflation and Price Shocks

Electricity Concessions spending (current dollars) Concessions spending ($2003) (m) Concession per cent of all bills Gas Concessions spending (current dollars) Concessions spending ($2003) (m) Concession per cent of all bills Water Concessions spending (current dollars) Concessions spending ($2003) (m) Concession per cent of all bills Municipal rates Concessions spending (current dollars) Concessions spending ($2003) (m) Concession per cent of all bills Total spending Total (current dollars) (m) Total ($2003) (m)

2003 VCM

2050 Base

2050 Base+shock

(m)

$40.1 $40.1 3.1

$160.9 $40.1 3.1

$168.5 $42.0 3.1

(m)

$25.2 $25.2 2.7

$101.1 $25.2 2.7

$114.2 $28.5 2.7

(m)

$68.4 $68.4 8.4

$71.2 $17.4 2.4

$71.2 $17.4 2.2

(m)

$36.6 $36.6 5.1

$36.8 $9.2 1.3

$36.8 $9.2 1.1

$170.3 $170.3

$369.2 $91.9

$390.7 $97.1

use. No allowance is made for price elasticity effects. A number of other drivers of consumption are also ignored: these include changes to appliance mix; improved appliance efficiency due to labelling and regulatory schemes; an increasing infloor space per capita; and a fall in people per dwelling. Any of these might have significant effects over 50 years. The VCM was necessary to predict future spending where price shock is a factor in the cases of water and rates, because the concession designs for these items are more complex — with eligibility varying by tenure and type of card and the level of concession being ‘capped’ to a total of $135 per annum (but is less for some households). Since the great majority of eligible households already receive the maximum concession in 2003, expenditure on these concessions is only slightly higher in 2050 than today — a real decline of 75 per cent. The concession is not subject to indexation. The effect of above inflation prices or rate rises is evident in higher spending on electricity and gas than would otherwise be needed. The effect of the caps on water and rates concessions, however, means that by 2050 price rises do not generate any further spending in these areas, and real spending per capita is about one quarter of that today. In summary, total spending on these four major state concessions falls in real terms by more than 40 per cent, even with real price shocks.

Population Ageing and State Concessions Policy

473

9. The Impact of Population Ageing The second driver of concessions spending is growth in the concession cardholder population. Although clearly targeted to people on low incomes, concessions nevertheless flow to a substantial number of people. Demographic changes, persistent high unemployment, and some past policy changes broadening access to concession cards have meant that a significant minority of the population — 25 per cent of people, 30 per cent of adults — hold cards. In 2002—2003, some 37 per cent of Victorian households claimed a concession on their electricity bill. Older Victorians — full or part-rate age pensioners and a smaller number of DVA cardholders — are the cardholders in the majority of these households. Over 80 per cent of older people hold a concession card.11 The Department of Human Services has developed short-term estimates of the number of cardholders from Victorian population estimates (Victorian Department of Infrastructure, 2000), using data on types of concession cards and short-term forecasts produced by the Commonwealth where available. These estimates suggest that, in the short term, the number of:   

DVA cardholders will continue to decline by 2–3 per cent a year; other age pensioners will rise by more than 2 per cent a year; and non-aged cardholders will rise less than 1 per cent a year.

Overall, the number of cardholders is likely to rise by around 1 per cent a year and the number of cardholding households by slightly more; this is slightly slower than trend growth over the 1990s.12 Other things being equal, population ageing will increase the proportion of the population likely to have entitlement to a PCC. The Intergenerational Report (Treasurer, 2002) suggests that an increased proportion of retired people with higher superannuation incomes may offset this, but the extent is unclear. To produce assessments of the possible impact of population

11

These high numbers reflect the fact that while fewer people rely only upon the Age Pension in retirement, eased income tests do not exclude many older people from access to the PCC. 12 The number of households claiming electricity concessions, the most widely claimed household Victorian concession, actually fell from 718,000 in 2000–2001 to 685,000 in 2001–2002. Annual data on concession receipt is of variable quality. However the trend from the mid-1990s was for annual growth to be around 1.5–1.8 per cent.

474

Karen Piper and Don Siemon

ageing, this study used the population forecasts of the Intergenerational Report without trying to include countervailing factors.13 Specifically, our forecasts assume, 

  

no changes in access to or extent of reliance on the Age Pension by older people; no further change in household size; no substantial redistribution of population between states; and no offsetting reduction in concession entitlement in the non-aged population; that is, no decrease in unemployment and joblessness. Again, this may lead to an overestimation of the number of cardholding households.

These assumptions imply that the estimates generated are probably towards the upper end of the effects of population ageing in Victoria. The estimates forecast that from 2003 to 2050,  

the Victorian population will increase by 35 per cent; and the number of concession cards per head of population is likely to increase by a further 20 per cent.

These estimates were used to build on the previous estimates of concessions spending given CPI-based increases in prices and rates (the base scenario) — and the results are shown in Table 4. Again, the table also shows the real value of concessions spending (in $2003) and concession spending as a proportion of total spending on bills. Although, the VCM has the capacity to model the effects of changing concession take-up rates — the fraction of those entitled who do so — the user interface does not allow direct modelling of the effects of ageing. This was done independently. Population ageing leads to spending on all concessions being higher than would be expected based on the current population distribution. Spending on the rates concession particularly rises, since this is limited to pensioners. But this increased demand does little to reverse the erosion of real spending on water and rates — and, again, this also means that total per capita spending on these four major state concessions falls in real terms by 33 per cent, despite population ageing.

13

While some aspects of the Intergenerational Report are contested, with debate focused on its policy implications, the population forecasts provide a good basis for consideration of possible impacts on state budgets.

Population Ageing and State Concessions Policy Table 4:

475

Impacts of Ageing

Electricity Concessions spending (current dollars) Concessions spending ($2003) (m) Concession per cent of all bills Gas Concessions spending (current dollars) Concessions spending ($2003) (m) Concession per cent of all bills Water Concessions spending (current dollars) Concessions spending ($2003) (m) Concession per cent of all bills Municipal rates Concessions spending (current dollars) Concessions spending ($2003) (m) Concession per cent of all bills Total spending Total (current dollars) (m) Total ($2003) (m)

2003 VCM

2050 Base

2050 Base+ageing

(m)

$40.1 $40.1 3.1

$160.9 $40.1 3.1

$193.0 $48.1 3.7

(m)

$25.2 $25.2 2.7

$101.1 $25.2 2.7

$121.3 $30.2 3.2

(m)

$68.4 $68.4 8.4

$71.2 $17.4 2.4

$85.4 $21.3 2.9

(m)

$36.6 $36.6 5.1

$36.8 $9.2 1.3

$51.6 $12.9 1.8

$170.3 $170.3

$369.2 $91.9

$451.3 $112.5

10. The Impact of Targeting The final driver of concession spending considered in this paper is that of policy change to the targeting of concessions. Victorian concessions are paid to people whose circumstances, as assessed by the Commonwealth, are such that they require direct support in order to sustain themselves and their dependents. Possession of a PCC or HCC provides a simple and very low-cost proof of this. Access to a card generally ceases if the person is not entitled to a pension, allowance, or maximum rate of Family Tax Benefit.14

14

There are three minor exceptions. A small number of DVA Gold Cards (for War Widows and totally and permanently incapacitated veterans) are not income-tested; Victoria has however long regarded these as deserving of concessions. Second, people who have just moved off income support payments hold a small proportion of cards — some three per cent; allowing cardholding to continue on for some time is thought to reduce disincentives to work. Third, a similar proportion of cards are held by people whose incomes are close to or below the social security threshold but who are not recipients. This ‘working poor’ category may overlap with the previous group.

476

Karen Piper and Don Siemon

Access to the PCC and HCC matches the traditional categories in the social security system of pensioners (age, sole parent, and disability) and beneficiaries (unemployed, students, sick); with families or single people with very low market incomes also able to gain an HCC. Since the 1980s, Victorian Government policy has extended concessions equally to both PCC and HCC holders, although some anomalies remain.15 There is no rationale for the concession system giving preference to one of these groups. 





In terms of disposable income, HCC cardholders as a group tend to have less immediate capacity to pay for essentials than do PCC cardholders. Income tests are markedly more generous for pensioners than beneficiaries — in 2001, a single person could hold a PCC with an income of up to $30,000 per annum compared with $16,600 for an HCC — and around one-third of pensioners now have significant private incomes. Age pensioners in particular are also more likely to be asset holders, with lower housing costs. However, holders of an HCC have some greater prospect of improving their income through employment; their incomes are less likely to be fixed. They may be on payments for only a limited time and able to defer payment or borrow. The distinction between some the PCC and HCC households is becoming less clear in other ways. Work expectations of sole mothers (holding the PCC) and partnered mothers (holding the HCC) are becoming more similar — and holders of a Disability Support Pension are also being encouraged to enter or re-enter the work force.

While living standards vary within it, the PCC/HCC group of households taken as a whole forms a fairly distinct body in the overall income distribution, with their average equivalent disposable incomes being less than half those of employed, self-employed, or superannuated households. Their capacity to afford taxes and charges is qualitatively lower. Targeting expenditure support to this group via concessions provides a simple and robust basis for entitlement that enjoys widespread community support and is regarded as authoritative by governments. There are two challenges to the consensus over targeting of concessions, however. The first suggests that the targeting is too broad. Around 37 per cent of households benefit from one or more concessions. By contrast, poverty incidence estimates suggest that less than half of this number could be expected to be ‘poor’ (Harding et al., 2001). 15

The exceptions are the concessions on municipal rates, Transport Accident Commission (compulsory third party vehicle) insurance levies and hunting, fishing, wildlife, and firewood license fees. These are available only to pensioners.

Population Ageing and State Concessions Policy

477

It is difficult, however, to establish an appropriate way to target ‘poverty’. There is no poverty measure agreed upon, between, or within governments. Poverty incidence — the size of the target group — is very sensitive to the poverty line drawn. The poverty lines tend to be close to base social security payments, so that the lines tend to highlight the gaps or inadequacies in the income security system.16 An alternative would be to target only households with incomes equivalent to base social security payments. The major group of households that receives concessions, despite having a significantly higher income, is partpensioners. About one-quarter of the PCC cardholders are part-pensioners, and a single person can hold the PCC with an income of nearly three times the base pension. Part-pensioners are mainly age pensioners; around 35 per cent receive less than the full pension. In principle, it would be possible to try to wind concessions back to ‘full-rate’ recipients. In fact, however, the Commonwealth provides funding for most of Victoria’s spending on ‘core concessions’ for part-pensioners as a result of compensation arrangements entered into in 1992–1993. This special purpose payment provides some $46 million a year to the state for provision of energy, water, rates, and transport concessions. Net of this compensation, Victoria’s own spending on concessions remains fairly tightly focused on households with low incomes. After the increase associated with the inclusion of part-pensioners, the proportion of Victorian households benefiting from concessions has increased only slightly since 1995. The second challenge is from the other direction — that targeting is too narrow. In particular, ‘pensioner concessions’ are sometimes seen as an automatic entitlement for all older people — a reward for a lifetime of work — and pressures to extend benefits to a more affluent group of older people have been present for many years.17

16

As a result, narrower targeting runs the risk of the state seeking to directly compensate for specific unfair social security or taxation policy decisions of the Commonwealth. Concessions are properly a form of expenditure support, not income support. Just as it would be inappropriate, for example, for the Victorian Government to try to use concessions to counteract regressive effects of national tax reform, so concessions should not be targeted to gaps or weaknesses in the social security system — for example, spending should not be targeted solely towards tenants because Rent Assistance is inadequate to meet housing costs. 17 Victoria has a seniors card available on a non-means tested basis to all older people. This offers access to discount off-peak public transport fares and some other benefits, mainly discounts from private retailers. The seniors card and related initiatives can be seen as offering support for ‘healthy ageing’, by encouraging mobility and social participation.

478

Karen Piper and Don Siemon

In May 2001, the Commonwealth announced that it would encourage states to offer concessions to non-pensioner older people holding a Commonwealth Seniors Health Card (CHSC). This card is available to single older people with annual incomes of up to $50,000 and couples with incomes of up to $80,000.18 Most states, including Victoria, have not as yet accepted the Commonwealth’s offer of compensation in exchange for extending entitlement to this group.19 The proposal to extend entitlement to holders of the CSHC is an example of a policy change to the targeting of concessions with significant effects over the long term: as the number of older people grows, the impact of the policy change will also be magnified. This proposal will also be modelled. A forecast of CSHC cardholding to 2050 was prepared; it suggests that the number of CSHC holders per head of population is likely to nearly double. This is arguably an underestimate; it is possible that recent growth in the availability of superannuation might lead to this group growing even faster. For modelling purposes, however, and in keeping with previous assumptions, it was assumed here that the proportions of older people who are pensioner and non-pensioner CSHC holders are fixed. Once again, these forecasts were built onto the earlier estimates of concessions spending given CPI increases in prices and rates (the base scenario) — and the results are shown in Table 5. The table also shows the real value of concessions spending (in $2003) and concession spending as a proportion of total Victorian bills. The effect of the policy change being modelled to extend entitlement to CSHC holders is similar to that of population ageing, although about threequarters of the size. Overall, the increase is once more outweighed by the

18

The Commonwealth Seniors Health Card (CSHC) was originally introduced in 1994 for a small group of older people who met the pension income test but were excluded by the assets test or because they had not been resident in Australia for long enough (10 years is usually required to qualify). The aim of the CSHC was to ensure that these groups were not exposed to hardship or underused medicines. Its introduction also had the benefit of easing the effect of the asset test by providing a card with comparable health benefits to people whose means might be only slightly greater than those of part-pensioners. Over 1997—2001, there was progressive widening of entitlement to the CSHC. 19 There may be some policy basis for the extension of Commonwealth health and pharmaceutical benefits to this more affluent group of retired people, but it is much less clear that there is any basis for the extension of state concessions — other than ‘rewarding retirement’. Policies to promote health and encourage economic and social participation may be of far more value to affluent older people than discounts on household bills.

Population Ageing and State Concessions Policy Table 5:

479

Impacts of Policy Change (Extension to Holders of CSHC)

Electricity Concessions spending (current dollars) Concessions spending ($2003) (m) Concession per cent of all bills Gas Concessions spending (current dollars) Concessions spending ($2003) (m) Concession per cent of all bills Water Concessions spending (current dollars) Concessions spending ($2003) (m) Concession per cent of all bills Municipal rates Concessions spending (current dollars) Concessions spending ($2003) (m) Concession per cent of all bills Total spending Total (current dollars) (m) Total ($2003) (m)

2003 VCM

2050 Base

2050 Base+CSHC

(m)

$40.1 $40.1 3.1

$160.9 $40.1 3.1

$181.3 $45.2 3.5

(m)

$25.2 $25.2 2.7

$101.1 $25.2 2.7

$113.7 $28.3 3.0

(m)

$68.4 $68.4 8.4

$71.2 $17.4 2.4

$82.9 $20.7 2.8

(m)

$36.6 $36.6 5.1

$36.8 $9.2 1.3

$47.8 $11.9 1.7

$170.3 $170.3

$370.0 $91.9

$425.7 $106.1

effect of the caps on water and rates concessions, so that despite this, real spending is much lower in 2050 than today.

11. Conclusions A steadily ageing population is forecast by the Commonwealth to have major impacts on government revenue and expenditure — of the order of five per cent of GDP (Treasurer, 2002). The impacts are likely to be mostly felt in the long term, beyond 2015. This paper reports on efforts within a state government context to consider the implications of such long-term changes. State concessions, linked as they are to the Commonwealth income support system, are clearly one area of Victorian policy which demographic change might impact. High levels of chronic unemployment, for example, have clearly had such an impact over the past two decades. The long-term impacts of population ageing on four major Victorian concessions administered by the Department of Human Services have therefore been assessed. The concessions chosen were those for which Victoria has detailed consumption and household data obtained from two household

480

Karen Piper and Don Siemon

surveys. Policy options for these concessions can be investigated directly using the VCM, although its value is limited in modelling long-term impacts. The work undertaken suggests that the expected effect of population ageing by 2050 might lead to an increase in the number of households entitled to state concessions of some 22 per cent above what would otherwise be expected. Given Victoria provides concessions by way of outlays and revenue foregone worth around $700 million a year, population ageing is likely to cost $140 million per annum. However, population ageing is not an overnight phenomenon. The increase will be felt over at least 30 years. More immediate pricing or policy changes may well emerge with fiscal impacts of a similar magnitude to those of population ageing, at least in the medium term. Extension of entitlement to CHSC holders, as summarized in Table 5, would produce a somewhat smaller but broadly similar impact on spending in 2050 to that arising from population ageing — and much more immediately. More importantly, the previous tables indicate that the dominant longterm direction of real spending on this group of concessions taken as a whole is downwards. The driver for this is to be found in the unindexed caps on water and rates concessions. Even with population ageing, this is forecast to result in a one-third fall in real per capita spending on these four concessions, as inflation takes its toll.20 Of course, just as inflation takes its toll, so economic growth may continue to compound. Presumably, increased GDP will make some contribution to increased budget flexibility. Interest in concessions as an instrument is likely to re-emerge as state governments respond to these challenges, as it has in other times when tax innovation, revenue raising, and pricing policy have become a priority — the early 1980s and early 1990s are examples. The challenge for public policy will be to ensure that the best use is made of this policy instrument — that the concession is part of the design of the tax; is well targeted; can be effectively implemented; and is in keeping with Victoria’s overall framework for the improvement of state concessions.

Acknowledgments This paper was written with the assistance of staff in the Concessions Unit, Victorian Department of Human Services. In particular the authors thank Iwona Lider and Andrew Muscat for their assistance in the use of VCM and forecasting concession cardholding. Views expressed are those of the authors. 20

In this context, it should be noted that in the 2004–2005 Victorian state budget the cap was increased and indexation of this cap was introduced.

Population Ageing and State Concessions Policy

481

References Harding, A., Lloyd, R. and Greenwell, H. (2001). Financial Disadvantage in Australia 1900 to 2000: The Persistence of Poverty in a Decade of Growth. The Smith Family, Camperdown, NSW, November (available from www.smithfamily. org.au). Roy Morgan Research. (2001). Victorian Utility Consumption Survey 2001. Final Report. Department of Human Services, Melbourne, June. Treasurer. (2002). Intergenerational Report 2002– 03. Commonwealth Budget Paper no. 5, Canberra. Victoria. (2003). Budget Statement 2003– 04. Budget Paper no. 1. Appendix F, 319-323, Melbourne. Victorian Department of Human Services. (2002). State Concessions 2001– 2002. Helping Lower Income Victorians Afford Essential Services. Department of Human Services, Melbourne. Victorian Department of Infrastructure. (2000). Victoria in Future 1996– 2021. Department of Infrastructure, Melbourne.

This page intentionally left blank

Chapter 19

Future Service Provision at the Local Level: Demographic Modelling and Forecasting in the City of Bayside Simon Byth Social Research, City of Bayside, Australia

Abstract To plan for service provision and development in the City of Bayside, the Bayside City Council has developed a model of population and housing, known as ‘‘100 Streets’’. Based upon monitoring of population and housing developments in the city — and upon policy for future development — the model produces estimates of future population and housing characteristics for each Census Collection District within the City. The ‘‘100 Streets’’ model utilises a broad range of data from the Census of Population and Housing, as well as many administrative datasets arising from the operations of the City Council. The model operates through three components: population cohort component modelling, household formation modelling, and residential housing modelling. Local administrative data is used to update and provide further detail to external data sets. Into the future, it is envisaged that trends identified through the development of the ‘‘100 Streets’’ model will be used to develop further our local knowledge of our community and our suburban region. We also propose the development of a specialised microsimulation model of population and housing developments, which will be able to add other dimensions to this work, using the other Census variables available through the Confidentialised Unit Record File (CURF) from the 2001 Census of Population and Housing.

1. Introduction The way that government agencies and service providers respond to ageing in a local community depends upon a range of community characteristics International Symposia in Economic Theory and Econometrics, Vol. 15 Copyright r 2007 Elsevier B.V. All rights reserved ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15019-X

484

Simon Byth

which are most apparent at a very local level. In established suburban communities, the concentration of older residents in particular locations has important consequences for the quality of those residents’ lives — as well as for the agencies that provide services to those populations. At a Local Government level, this requires a detailed understanding of the dynamics of ageing in local neighbourhoods, with a view to ensuring that community facilities and services are in place to meet the needs of local population cohorts as they age. During 2003, the Bayside City Council developed a model of local demographic monitoring and forecasting under the project name ‘‘100 Streets’’. The main objective of the project was to identify local concentrations of population according to their age, in order to match the available housing and service provision to the requirements of the local population.

2. Background The City of Bayside is one of 31 Local Government Areas in the metropolitan area of Melbourne in Victoria, Australia. It covers an area of 37 square kilometres in the inner southern suburbs of Melbourne, including the suburbs of Brighton, Sandringham, and Beaumaris. In June 2002, the city had an Estimated Resident Population of 89,359 (Australian Bureau of Statistics, 2003). In the Australian Standard Geographical Classification (ASGC), the City of Bayside lies within the Statistical Division (SD) of Melbourne. It contains 2 Statistical Local Areas (SLAs) and 149 Collection Districts (Figure 1). In comparison to the Statistical Division of Melbourne, the City of Bayside has several differentiating demographic characteristics, including: 



a comparatively high proportion of older residents (aged 70 years and over) and a comparatively low proportion of young adults (aged 18–25 years); and comparatively high household incomes.

The Bayside City Council is primarily responsible for the control of building activity within its area. The Victorian Local Government Act (1989) requires it to regularly update its Planning Scheme and its Municipal Strategy Statement, describing its policies for permitting certain building activity in particular areas of the city. Between December 1999 and October 2002, the Government of Victoria developed an overall strategy for the development of the suburban area of Melbourne through to the year 2030, under the title of Melbourne 2030.

Future Service Provision at the Local Level

485

Figure 1: Map of the City of Bayside and its Neighbouring Local Government Areas, Showing the Collection Districts within the City of Bayside

Yarra (C) Hobsons Bay (C)

Maroondah (C)

Boroondara (C)

Melbourne (C)

Whitehorse (C)

Stonnington (C) Port Phillip (C)

Glen Eira (C)

Monash (C)

Knox (C)

Bayside (C)

Kingston (C)

Greater Dandenong (C)

0

5

10

kilometers

The Melbourne 2030 strategies divide the metropolitan area of Melbourne into five regions and detail the extent of household and residential dwelling development which is anticipated in each region — although they do not (as yet) specify the extent of the development within each Local Government Area. The Victorian Government’s Melbourne 2030 strategies included forward estimates of population and household numbers by Local Government Area through to 2030. One of the objectives of the ‘‘100 Streets’’ project in the Bayside Local Government Area was to produce forward estimates of these measures for Collection Districts within the municipal area, to show what the State Government’s objectives would mean in this area. The City also provides a range of local services, such as maternal and child health services, kindergartens and pre-school centres, and aged care services. Planning for the provision of these services requires reliable forward estimates of population and housing on a specifically local basis, to be able to allocate resources appropriately to the various local communities within the City according to designated policy objectives. Planning for service provision within the City of Bayside requires detailed projections of the local population over a 10-year period (through to 2011), with an emphasis upon the household formation and housing patterns associated with the local population.

Simon Byth

486

3. The ‘‘100 Streets’’ Model: Methods and Data Sources Owing to resource constraints, the development of the ‘‘100 Streets’’ model has been based upon the interaction of three simple modelling components (Figure 2):   

a population component (or ‘‘cohort survival’’) model; a model of household formation; and a residential housing development model.

In many respects, it is (in its current form) a ‘‘naı¨ ve’’ model. It makes no attempt to consider the probability distribution of its forward estimates — and thus does not enjoy the strengths of conventional stochastic forecasts (Tuljapurkar and Lee, 1999). Similarly, it cannot (in its current form) produce forecasts on the basis of the patterns of behaviour of individual people or households, so as to produce spatial microsimulation forecasts (Ballas et al., 1999; Taylor et al., 2004; Chin et al., 2005, 2006a, 2006b). It does, however, enjoy some strengths, which arise from its ‘‘grass roots’’ development at a Local Government level. Principally, it is based upon local area data at a finer level of spatial definition than would otherwise be possible, most of which are verifiable with a degree of precision which would not otherwise be possible. Because of its links to land use and building control, and to service provision at a very local level, the model is required to operate with small, local groupings (termed Collection Districts and containing about 240 households or 500 people).

Figure 2:

Structure of the ‘‘100 Streets’’ Model Population: Measured by Age, Sex, Location, and Time. Driven by: Births, Deaths, and Migration.

Housing: Measured by Dwelling Type, Location, and Time. Driven by Constructions, Renovations, and Demolitions.

Households: Measured by Household/Family Type, Location and Time. Driven by Population Propensities and Housing Propensities.

Future Service Provision at the Local Level

487

3.1 Population Component Model The ‘‘population component’’ (or ‘‘cohort survival’’) model is a conventional demographic model — which takes a base population by age and sex, and updates it using assumptions about future levels of births, deaths and migration. Our base population data are the Australian Bureau of Statistics (ABS) Estimated Resident Population at 30 June 2001, by 5-year age groups and sex, for Collection Districts in the City of Bayside. Trends in local age-specific birth rates are used to estimate future births. Similarly, trends in local age- and sex-specific death rates are used to estimate future deaths by age and sex. Forecasts of local migration are based upon the Census trends (in reporting the residents’ ‘‘place of usual residence’’), and upon the migration component of the recorded change in the ABS Estimated Resident Population series. Since 1996, we have encountered some divergence in the relevant local data on births in the City of Bayside. In Victoria, Local Government Authorities are required to provide Maternal and Child Health services to newly born babies and their mothers, and special administrative mechanisms exist to allow for this. In summary, the system for recording births is: 1. When a child is born, a registered midwife or medical practitioner who is present is required to report the birth to the Victorian Department of Human Services (DHS). This report includes details such as the mother’s age and usual address. These data are collected by the Victorian Perinatal Data Collection Unit within DHS. 2. DHS staff then send out a notice of the birth to the Maternal and Child Health nurses in the Local Government Authority responsible for the mother’s usual address. 3. After the child is born, the mother is given a form which she (or someone acting on her behalf) can use to register the birth with the Victorian Registrar of Births Deaths and Marriages. These data are subsequently collected and reported by the ABS. As a result, there are three sets of local birth data available to Local Government Authorities. Most of the apparent anomalies between the data from the DHS Perinatal Data Collection Unit and the enrolments on the Local Government Authorities’ Maternal and Child Health services relate to inconsistent or inaccurate reporting of the mother’s usual address. Anomalies can arise between these data sets and the births data available from the Registrar of Births Deaths and Marriages (and subsequently from the ABS) because of the time delay which can occur between the birth and its registration through this mechanism.

488

Simon Byth

For administrative reasons, the number of births reported by the Victorian Registrar of Births Deaths and Marriages (and subsequently by the ABS) has been consistently less than the number of births recorded by the Victorian Perinatal Data Collection Unit within the Victorian DHS. Similarly, the number of births reported by the Victorian Perinatal Data Collection Unit has been consistently less than the number recorded by our Maternal and Child Health nurses. As our nurses actually visit each newborn child, confirming the details of the child’s birth and the mother’s usual address within the City of Bayside, we are very confident that our own administrative data provide the most accurate number of births in these circumstances. However, as our Maternal and Child Health records have not consistently recorded the mother’s age, we were forced to apply the maternal age distribution and the geographical distribution from the ABS data to the total number of births from our Maternal and Child Health records. Consequently, by using the ABS Estimated Resident Population series (by age group and sex) as the denominator, we calculated an approximate agespecific birth rate, by 5-year age groups, by SLA with the City of Bayside, from 1998 to 2001 — and then were able to extrapolate that trend forward to the forecast horizon. Some special attention was devoted to the local birth rates, as the City of Bayside has experienced an increasing ‘‘crude birth rate’’ (number of births/ number of females) since 1996. This is very unusual, as the ‘‘crude birth rate’’ for the Melbourne Statistical Division and for the State of Victoria have both declined significantly during this period. The increasing crude birth rate in our City, and especially the unusually high birth rate among mothers over 30 years of age, are distinctive, differentiating characteristics of our local area. The numbers of deaths, fortunately, have not provided the same difficulties. Our age- and sex-specific death rates were calculated — with the deaths data from the Registrar of Births Deaths and Marriages and subsequently reported by the ABS providing the numerator, and ABS Estimated Resident Population series (by age group and sex) providing the denominator. A time series from 1996 to 2001 was projected forward to our forecast horizon (2011). Data on local migration, of course, caused some concern, as the scale of inward and outward migration from our Local Government Area is considerably greater than that of local births and deaths. To assess local area migration, we used two tables from the 2001 Census of Population and Housing: 1. Usual Residence on Census Night, by Usual Residence One Year Ago (for persons aged 1 year and over), by 5-year age group and sex, for the SLAs in the City of Bayside, the rest of Australia and overseas; and (similarly)

Future Service Provision at the Local Level

489

2. Usual Residence on Census Night, by Usual Residence Five Years Ago (for persons aged 4 years and over), by 5-year age group and sex, for the SLAs in the City of Bayside, the rest of Australia and overseas. From these tables, we were able to estimate, at the SLA level, the inward migration and outward migration, by 5-year age group and sex, during 1 year prior to the Census (which was taken on 8 August 2001) and during the 5 years prior to the Census, allowing for an estimated outward migration overseas. From these we derived the net migration to each SLA, by 5-year age group and sex, during these two time periods. From these data, we estimated the net migration to each SLA in each of the 5 years prior to the Census, and extrapolated this trend forward to the forecast horizon. Given the difficulties associated with the ‘‘not stated’’ counts for these Census tables, we sought local administrative data which could further uncover the patterns of local migration during this period. One relevant data source was the registration of new library memberships in our Local Government Area. Like all Victorian Local Government Authorities, the City of Bayside administers free public libraries in our area. Membership and borrowing rights in these libraries are free, and library registrations currently include about 59 per cent of our resident population (52,855 registrations in June 2003). Although people who move out of the City rarely cancel their library registration, most people who move into the city will register. As a result, the numbers of new registrations are an indicator of migration into the city. The trend in new library registrations was found to be consistent with the identified trend in inward migration to the city. From this point, the Population Component model operated as a conventional ‘‘cohort survival’’ model: calculating and adding births, calculating and subtracting deaths, calculating and adding net migration, and ageing the population in place. As the ABS Estimated Resident Population series (by 5-year age groups and sex, for Collection Districts) is updated each year, it can also be included into this model. As point forecasts were required only for 5-year intervals (2001, 2006, 2011, etc.), the estimates were all produced using 5-year age groups for the local population. As a parallel to the project at this point, we also examined the local origin of our inward migration, and the local destination of our outward migration. For this purpose we used two Census tables: 1. Usual Residence One Year Ago, by Usual Residence on Census Night, for SLAs in the Melbourne SD, the remainder of Victoria and the remainder of Australia, for all persons, and for persons with a different residence in 2000, and (similarly)

490

Simon Byth

2. Usual Residence Five Years Ago, by Usual Residence on Census Night, for SLAs in the Melbourne SD, the remainder of Victoria and the remainder of Australia, for all persons, and for persons with a different residence in 1996. Our knowledge of the inward and outward migration with our surrounding Local Government Areas has added to our knowledge of our local community. However, in addition, it has enabled us to assess the ‘‘overflow effect’’ on local migration arising from an undersupply of accommodation in particular areas. 3.2 Household Model Resource constraints required that we produce a simple method for estimating families and households in our area. The method used was a simplified, local area adaptation of the ‘‘household size propensity’’ method used by the ABS to estimate households in its publications,1 which in turn was based upon earlier work by McDonald and Kippen (1998). Whereas the Population Component model deals with forecasts of the Estimated Resident Population, the Household model (and the Housing model) deals with the ‘‘household population’’ (i.e., the resident population living in Occupied Private Dwellings). The difference between the Estimated Resident Population and the ‘‘household population’’ is the resident population in ‘‘non-private dwellings’’ — and we have had to project the development of this sector of the resident population separately. In the City of Bayside, there were 1,582 people enumerated in 59 non-private dwellings in the 2001 Census of Population and Housing (1.8 per cent of the Estimated Resident Population). About 75 per cent of this population lived in nursing homes and cared accommodation for the retired or aged. Using our local administrative information, we were able to list out these establishments and to forecast their development individually. We used a summary classification of Household/Family Types (Table 1) and Census table of Persons Enumerated in Occupied Private Dwellings by Household/Family Type, for Collection Districts in the City of Bayside, with those persons cross-classified by 5-year age groups and sex. From this we calculated propensities for usual residents (in each Collection District) to live in particular Household/Family Types, by their 5-year age groups and sex. Within our forecast horizon (to 2011), these local propensities were assumed to remain constant. 1

Australian Bureau of Statistics. Household Estimates, Australia: 1986, 1991–1994. (Cat. No. 3229.0). This ABS publication was followed by another, 1998, ‘‘Household and Family Projections, Australia: 1996–2021’’ (Cat. No. 3236.0).

Future Service Provision at the Local Level Table 1:

491

Listing of Categories of Household/Family Type

Household/family type One family household: couple family with children One family household: couple family without children One family household: one parent family One family household: other family Multi-family household: total Lone person household Group household Non-classifiable household

A table of Occupied Private Dwellings and Persons in Occupied Private Dwellings by Household/Family Type, for Collection Districts in the City of Bayside, was used to estimate the mean number of persons per Occupied Private Dwelling, for each Household/family Type and for each Collection District in the City of Bayside. As forward estimates of local population characteristics (by 5-year age group, sex and Collection District) were output from the Population Component Model, the propensities were applied to calculate the numbers of usual residents (by 5-year age group, sex and Collection District) expected to live in each Household/Family Type. The mean number of persons per Occupied Private Dwelling, for each Household/Family Type, for each Collection District was then applied to calculate the number of Households/ Families of each type (in each Collection District). We have not modelled the change in these propensities at a Collection District level. We have assumed that changes in these propensities at a SLA level are consistent across the SLA, and have projected a 5-year time series forward on that basis. 3.3 Housing Model The local Housing Model was of particular interest to us because of the comprehensive administrative data available on housing in our area — both historically from Council records and prospectively in terms of Council policy, especially the Bayside Planning Scheme. As with the Household Model, we modelled the changes to non-private dwellings separately. In the 2001 Census of Population and Housing, the City of Bayside recorded only 59 ‘‘non-private dwellings’’ and we were able to make forward estimates of these dwellings separately. By considering the demand for housing arising from the formation of households, we were implicitly considering the number of occupied private

Simon Byth

492 Table 2:

Listing of Variable, Dwelling Type

Dwelling type Separate house Semi-detached, row/terrace, townhouse, etc. Flat, unit or apartment Other dwelling Not stated

dwellings. As a result, we needed also to consider local vacancy (or ‘‘nonoccupancy’’) rates, in order to come back to a total number of dwellings. We used a summary classification of Dwelling Types (Table 2) which summarised the Census variable, ‘‘Dwelling Structure’’, while also being comparable to the variable ‘‘Land Classification Code’’, which is used in Local Government property valuations in Victoria (and which is required in statutory reporting to the Victorian Valuer-General). A Census table of Dwelling Structure by Household/Family Type (by SLA)2 was used to calculate the propensity of each Household/Family Type (in each SLA) to live in each Dwelling Type. Within our forecast horizon (to 2011), these propensities were assumed to remain constant. As forward estimates of households/families (by Household/Family Type and Collection District) were output from the Household Model, these propensities were used to estimate the demand for occupied private dwellings (by Dwelling Type and Collection Districts). In addition to this linkage, we also linked the Housing model back both to the Household Model and the Population Component model, through data on the Number of Bedrooms. From the 2001 Census of Population and Housing, we used a table of Private Dwellings by Dwelling Location, by Dwelling Structure and by Number of Bedrooms, for Collection Districts in the City of Bayside. The Victorian Local Government Act (1989) requires Local Government Authorities, such as the City of Bayside, to collect revenues (commonly known as ‘‘rates’’) based upon the value of properties in their area — and to manage a system of controls over the construction and demolition of buildings in their area. Consequently, Council maintains detailed records of the ownership of property, and of the characteristics which would determine the ‘‘Capital Improved Value’’ of the property. Council also develops and

2

Table 47 from the ‘‘Extended Community Profile’’ series, published by Australian Bureau of Statistics.

Future Service Provision at the Local Level

493

manages detailed policies on the types of buildings which will be permitted in particular areas of the municipality. From our Valuation records, which are updated every two years, we produced a similar table of Residential Dwellings, by Dwelling Type and by Number of Bedrooms, for Collection Districts in the City of Bayside. Data on the Number of Bedrooms enabled us to check, in our forward estimates, that the anticipated numbers of dwellings (by Dwelling Type and Collection District) provided appropriate and sufficient accommodation for the anticipated number of households (by Household/Family Type and Collection District) and the anticipated population (by 5-year age group, sex and Collection District). In each area, the effective ‘‘bedroom utilisation rate’’ provided a measure of ‘‘crowding’’ in that area, along with the local population density. Council records were summarised to produce a table of Private Dwellings by Dwelling Type, for Collection Districts in the City of Bayside. For this purpose, we developed a concordance of Dwelling Types to the Land Classification Code (LCC) which is used in Property Valuations. We could then compare this with the count of Private Dwellings by Dwelling Location and Dwelling Structure, for Collection Districts in the City of Bayside (arising from the ABS Census of Population and Housing). ABS Census data on dwellings are collected in a slightly different form than Local Government data on rateable properties. In particular, there are many instances (e.g. a separate house with a bungalow ‘‘granny flat’’ in the backyard) which the ABS would classify as two dwellings (and two households), even though Local Government Valuers would assess it as one rateable property. Where the count of dwellings from the Council records was higher than the Census count, we used the Council count in our modelling. Where the Census count was higher than the count from Council records, we reviewed our Council records and the operation of our concordance thoroughly, and made necessary adjustments. In some instances, we referred to aerial photographs of the area to count the relevant dwellings. The local vacancy rate (or ‘‘unoccupancy’’) rate, by Dwelling Type and Collection District, was applied to the number of Occupied Private Dwellings in each area, to provide an estimate of the total number of Private Dwellings. 3.4 Linkages between the Models Up to this stage in the development of the model, we have assumed that local population development is driving the development and utilisation of domestic housing in the area. In practice, of course, these two developments are inter-dependent, with each influencing the other. The City of Bayside is a mature residential area, the oldest sections of which were originally developed in the 1840s and the youngest sections of which were originally

494

Simon Byth

developed in the 1950s. Broadly speaking, it is an expensive and desirable place to live, and the demand for residential housing in the City almost always exceeds its supply. In these circumstances, the supply of residential housing (as measured by the count of Private Dwellings, by Dwelling Type and Collection District) is strongly influenced by the Council’s policies on Urban Land Use and Building. In particular, the Council’s policies on the sub-division of current residential land, and the re-development of current residential housing to accommodate more dwellings and/or more residents, are significant local issues which affect both the supply of residential housing in the City and (consequently) the households living in the City and the City’s population. In addition to our Council having specific policies in this regard, the Victorian Government also has its own policies and plans for the development of our region (especially those set out in the Melbourne 2030 strategies) — so there is some opportunity for policy difference between the two levels of government. In this context, it has been useful for us to be able to estimate the social effect of particular policies on local housing development by reversing the linkages between the models. Beginning from proposed changes to particular allotments of land, we have been able to work up to a change in the local housing stock (Private Dwellings, by Dwelling Type and Collection District). From this point, we have reversed the propensities (looking at the propensity of particular types of dwellings to accommodate particular types of households), to estimate the effect of that change on the local households (Households, by Household/Family Type and Collection District) and ultimately upon the local population (Persons, by 5-year age group and sex, and Collection Districts). In many instances, local service provision — such as pre-schools and aged care — is based upon local population characteristics, so the facility to anticipate changes in population characteristics has offered the opportunity to make appropriate longer term plans. For the development of specifically local policy, this process has helped to demonstrate the linkages between Urban Planning decisions taken at a Local Government level and Community Services planning at a Local Government level, as well as providing a solid basis upon which we, as a Local Government Authority, can act as an advocate for our local community in our dealings with State and Federal Government agencies. In practice, we have been able to fine-tune our forward estimates of housing and household formation by combining our estimates of the accommodation provided by ‘‘new’’ dwellings with our anticipated levels of ‘‘ageing in place’’ and migration. Because we are able to produce these estimates at the finest level of spatial detail, we have effectively used our detailed monitoring of local housing to refine our population estimates.

Future Service Provision at the Local Level

495

4. Project Outputs to Date Several of the interim outputs have been significant for the provision of, and planning for, Council services. To date, the major outputs of the project have been: 













A detailed reconciliation of numbers of births in the City of Bayside, from 1996 to the present, accounting for the differences between the three available data sources (showing an unusual increase in the number of births in the City during this period). A comparison of age-specific birth rates in the City of Bayside since 1996, to the comparable rates for the Melbourne SD and the state of Victoria (showing that Bayside women in their 30 s have an exceptionally high ‘‘crude birth rate’’). Graphing and projecting the patterns of inward and outward migration to the City, by 5-year age group and sex. Mapping the origin of people who move into the City of Bayside, and the destination of people who move out of the City. Forward estimates of the local population, by 5-year age group and sex, for Collection Districts in the City. Forward estimates of households/families, by Household/Family Type, for Collection Districts in the City. Forward estimates of residential dwellings, by Dwelling Type, for Collection Districts in the City.

As the project has proceeded, the interim results have been presented to a Staff Working Group made up of Council officers from various sections of the administration that use the results of the project — and also to a Council Steering Committee comprised of elected Councillors and Council officers actively working on the project. Where our interim results have dealt with local issues which span our municipal boundaries, we have involved our colleagues from the neighbouring Local Government Areas in the project. We have presented our interim results to the relevant State Government officials, and had informal discussions with them regarding the further development of the project.

5. Strengths and Weaknesses of the Project As noted earlier, the ‘‘100 Streets’’ project does not offer the benefits of a conventional stochastic forecasting model, although it could be developed into one. In particular, stochastic models allow the forecast user to compute probabilities for quantities of interest. Given proposed policy objectives, they allow forecast users to estimate the risk that a particular policy objective will not be achieved (McDonald and Kippen, 1998). The benefits of

496

Simon Byth

stochastic forecasting techniques are well documented, and it is a weakness of this project that it cannot provide those benefits. Similarly, it does not offer the benefits of modern microsimulation models, although it could contribute to the development of one. Further, this form of modelling faces several difficulties which arise from the inter-relationship of its underlying assumptions. First, a changing migration pattern (measured by the age and sex of the population) can be expected to influence the propensities to form particular types of households — because some people will migrate as individuals while others will migrate as family or household units. While the Population Component Model deals with the population as a count of individuals, the Household Model attempts to ‘‘bundle’’ them into groups as if they were all independent individuals, whereas (of course) most of them are already ‘‘bundled’’ into established family or household groups. The fundamental difficulty here is that local migration can be associated with a change in household structure, and the available data on local migration (from the Census of Population and Housing) cannot distinguish between individuals who relocate on their own and family or household groups who relocate together. Second, both individuals and family or household units will exercise some freedom of choice when forming household or family groups and when choosing a dwelling in which to locate themselves, to the extent that they have this freedom. In areas in which many residents are well resourced (such as the City of Bayside), the residents who can afford to do so will deliberately buy spare housing capacity (such as extra bedrooms and bathrooms, a larger backyard, etc.) which they do not immediately require — or else they will maintain the spare housing capacity after their real need for it has passed. In these neighbourhoods, the selection of housing is related more closely to the residents’ values and wealth (Foran and Poldy, 2002) than it is to the population characteristics by age and sex or to household formation patterns. An economist would refer to the selection of a dwelling as an act of consumption (in the case of rental) or of consumption and investment (in the case of a purchase). A person’s house provides accommodation for their household group, but its size, location, appearance, etc. will also reflect the values and resources of its owner. The influence of people’s values and resources on their patterns of household formation and their selection of a dwelling is currently beyond the scope of the ‘‘100 Streets’’ model, although the model could be enhanced by further development in this area.

6. Further Development of the Project Into the future, with the endorsement of the Steering Committee for the project, we propose to further develop the project in several directions, some

Future Service Provision at the Local Level

497

of which relate to its internal operation, while others relate to its application and its extension. 6.1 To Utilise Economic Data Access to the economic resources of income and wealth has a major influence upon the selection of a dwelling and the choice to remain in that dwelling. In an established suburban area such as the City of Bayside, these factors influence which people will take up residence in the City, as well as how and when current residents are likely to move house (e.g. as they age). In its current form, the project offers some scope to explore the availability of income and assets as a factor which influences migration to and from particular areas at particular times. 6.2 To Enhance Local Area Information Council officers frequently need information on small local areas within the city. Information might be required with regard to the use of a small local park, enrolments in a local pre-school, or a drainage issue affecting several adjacent dwellings. Some local area information is available through the operation of this model. In other instances, the information from this model can be used to enhance the processes of subsequent information collection using survey questionnaires. As our knowledge of the local community is improved by the output from this project, we have also been able to collect some specific local area information which enhances the model. For example, we recently collected information from our older residents aged 65 years and over regarding their housing choices, levels of social interaction, and transportation — and this has enabled us to specify the anticipated demand for housing by our older residents. 6.3 To Deal with Sub-Regional Issues between Adjacent Local Government Areas Many of the service demands on Local Government Authorities actually span their boundaries. Residents can and do cross municipal boundaries to use Council services — such as pre-schools and libraries — so it is important for Local Government Authorities to understand and communicate with their immediate neighbours. The movement across municipal boundaries is also an important component of the local economy and is closely monitored by the Economic Development sections of Local Government administrations. The permanent movement of people across Local Government boundaries is local area migration. Only a small percentage of people moving house in our area will move more than 10 km, so the movement of

498

Simon Byth

population from one Local Government Area will almost always be a movement to another adjacent Local Government Area. As a result, neighbouring Local Governments have much to gain from an agreed understanding of local migration. Also, some State Government services are administered on a regional basis, effectively requiring Local Government Authorities to work together in groups. The opportunity to share the methodology underlying the project with the Victorian Government staff preparing population projections for the state by Local Government Area ensured that the forward estimates in the City of Bayside were consistent with their larger scale forward estimates and policy settings for the region. 6.4 To Develop a Microsimulation Model of Population and Housing We are currently investigating the manner in which it might be possible to use the Confidentialised Unit Record File (CURF) arising from the 2001 Census of Population and Housing, along with the population, household and housing propensities calculated in this project, to develop a microsimulation model of population and housing in our City and/or our region. The objective here would be to adapt the propensities which were calculated in this project so far, and adapt them as patterns of change or ‘‘rules’’ in a microsimulation. The main benefit of this development should be the opportunity to use the other census variables which are included in the CURF. In particular, the opportunity to study patterns of income and wealth in our region would be especially welcomed. As mentioned earlier, in the City of Bayside, the selection of a particular dwelling is often a reflection of a person’s values and resources, in addition to meeting the accommodation requirements for their family or household group. The integration of Local Government housing information (including commercial valuations, etc.) into such a project would create a special opportunity to examine both current lifestyle issues and intergenerational issues affecting wealth and its relationship to population and housing change. Our immediate region includes some of Victoria’s wealthiest residential suburbs (such as Brighton in the City of Bayside, and Toorak and South Yarra in the nearby City of Stonnington) — but it also includes significant areas of lower income and higher density housing (such as St. Kilda in the adjacent City of Port Phillip). It includes some areas currently experiencing rapid urban re-development (such as Port Melbourne, in the City of Port Phillip) and other areas where there is considerable local resistance to urban re-development. The opportunity to integrate the Census variables relating to occupation, place of work, journey to work, etc. would offer a new insight into the operation of the regional suburban economy.

Future Service Provision at the Local Level

499

Further to this, we expect that the development of a spatial microsimulation model at this level would provide some clarity for the assumptions underlying the population modelling which was described earlier. The ability to simulate life events, as they occur to individual people, offers the opportunity to build the processes of household change from its component events for individual people. For example, when young adults gain their first substantial employment and move out of home, they are altering the structure of two households (the one they have moved out of and the one they are moving into), establishing a new income unit, and making a local migration. Similarly, a divorcee could make these three changes, as well as probably splitting some structure of assets. The opportunity to base a model clearly upon one basic unit — the individual person — and to simulate life events as they occur to those units should explicitly articulate the assumptions which are driving the modelling. In addition, access to the other Census variables, especially those relating to employment and income, offer an opportunity to examine the effect of income and wealth on changes to population and housing.

7. Conclusions Initiating the ‘‘100 Streets’’ project at the City of Bayside has been a challenging and rewarding activity for us. It has been challenging in that it has required us to: (a) develop an integrated approach to the official demographic data for our area (primarily from the ABS); and (b) improve our understanding of the relationship between the official data for our area and our own administrative records. In many instances, it has required us to re-assess our administrative data — with the result that we have realised that we actually have very thorough and reliable local information available to us as a Local Government Authority which is not available to other levels of government or to private organisations. This realisation has prompted us to develop the work further. In addition, the project has also required us to: (c) look in detail at the characteristics of small areas within our City; and (d) look forward into the future, so as to anticipate and plan for change. Within our area, this has prompted us to look at the differing resources and service requirements across the various neighbourhoods within our City,

500

Simon Byth

and to direct our Community Building activities towards the relevant local characteristics. The project has also required us to do all of the above in a manner which is readily communicable to the relevant State and Commonwealth Government bodies. Looking forward in time to anticipate changing characteristics, we have been able to proactively assess the specific local impact of Council policy, especially with regard to residential building re-development. In doing so, we have also examined the impact of broad social and economic changes which will affect our community. However, most importantly, by maintaining the focus of the project on Collection Districts containing approximately 240 households (and 500 people), we have been able to deliver verifiable results to local communities within our City, informing both the elected Council and the local residents about the processes of change in which we live.

References Australian Bureau of Statistics (2003). Regional Population Growth Australia and New Zealand, 2001– 02. (Cat. No. 3218.0). Ballas, D., Clarke, G. and Turton, I. (1999). Exploring Microsimulation Methodologies for the Estimate of Household Attributes. Paper Presented to the 4th International Conference on GeoComputation, Mary Washington College, VA, USA. Chin, S.F. and Harding, A. (2006a). Regional Dimensions: Creating Synthetic Smallarea Microdata and Spatial Microsimulation Models. Technical Paper no. 33, National Centre for Social and Economic Modelling, University of Canberra, May (available from http://www.natsem.canberra.edu.au). Chin, S.F., Harding, A. and Bill, A. (2006b). Regional Dimensions: Preparation of 1998– 99 HES for Reweighting to Small-area Benchmark’s. Technical Paper no. 34, National Centre for Social and Economic Modelling, University of Canberra, April. Chin, S.F., Harding, A., Lloyd, R., McNamara, J., Phillips, B. and Vu, Q. (2005). Spatial Microsimulation Using Synthetic Small Area Estimates of Income, Tax and Social Security Benefits. Australasian Journal of Regional Studies, 11(3), 303–336. Foran, B. and Poldy, F. (2002). Dilemmas Distilled: A Summary Guide to the CSIRO Technical Report. Future Dilemmas: Options to 2050 for Australia’s Population, Technology, Resources and Environment, CSIRO, Australia, p. 21. McDonald, P. and Kippen, R. (1998). Household Trends and Projections: Victoria, 1986–2011. Victorian Department of Infrastructure, Melbourne. Taylor, E., Harding, A., Lloyd, R. and Blake, M. (2004). Housing Unaffordability at the Statistical Local Area Level: New Estimates Using Spatial Microsimulation. Australasian Journal of Regional Studies, 10(3), 279–300. Tuljapurkar, S. and Lee, R. (1999). Population Forecasts, Public Policy and Risk. Paper presented to the conference of the International Statistical Institute, Finland.

Chapter 20

Characteristics and Travel Patterns of Older Australians: Impact of Population Ageing on Tourism Afzal Hossaina,, Geoff Baileyb and Milly Lubulwac a

The Bureau of Transport and Regional Economics, Department of Transport and Regional Services, Canberra, Australia b Tourism Division, Department of Industry, Tourism and Resources, Canberra, Australia c Tourism Research Australia, Australia

Abstract Senior travellers (55 years and over) constitute a significant and growing segment of Australia’s domestic tourism market. In 2002, more than 21 per cent of domestic tourism expenditure was due to travel by senior Australians. This chapter presents an analysis of the travel patterns of senior Australians in 2002, covering domestic day visitors, domestic overnight visitors and outbound visitors. The chapter also provides 10 and 20-year projections of the Australian travel sector, based on the changing profile of the Australian population.

1. Introduction Senior travellers are making a significant contribution to Australia’s domestic tourism market, in terms of number of trips and expenditure. In 2002 Australian seniors represented 22 per cent of all domestic overnight travellers and 26 per cent of all domestic day travellers — and these senior travellers spent an estimated $10.8 billion on their trips, accounting for

Corresponding author.

International Symposia in Economic Theory and Econometrics, Vol. 15 Copyright r 2007 Elsevier B.V. All rights reserved ISSN: 1571-0386 DOI: 10.1016/S1571-0386(06)15020-6

502

Afzal Hossain et al.

21 per cent of total expenditure on domestic tourism in that year.1 The significant impact of senior travellers on the domestic travel market is attributable to seniors having more time to travel and spending a large proportion of their discretionary income on travel and leisure activities (Teaff and Turpin, 1996; Horneman et al., 2002; Harding et al., 2003, p. 24). The ageing of Australia’s population, the increased mobility and longevity of senior Australians and the financial boost of pay-outs from compulsory superannuation all combine to make the nation’s retirees an important travel sector, set to play an important role in shaping tourism activity over the next 20 years. To date there has been little analysis of the characteristics and travel behaviours of Australian seniors. The few studies include a study of the contribution of senior travellers to the domestic tourism and demographic profile of day and overnight senior travellers (Hossain, 2003), a profile of older adult ecotourists (Lawton, 2001) and a profile of seniors based on demographic and psychographic characteristics (Horneman et al., 2002). Other studies have focussed on the perceived needs of travelling seniors with respect to accommodation facilities (Ruys and Wei, 1998), use of travel information sources (Cleaver, 2000), motivation for holiday travel (Cleaver and Muller, 1998) and life cycle travel patterns of outbound Australian travellers (Collins and Tisdell, 2002). This chapter presents the likely impact of population ageing on the tourism sector. The objectives of the analysis in the chapter are (i) to examine the significance of senior travellers to the Australian domestic travel market using 2002 data from the National Visitor Survey (NVS), (ii) to compare travel patterns between senior and non-senior travellers, and (iii) to examine the likely impact of population ageing on the Australian tourism sector in terms of the size and composition of the travel market over the next 20 years. The term ‘senior’ traveller refers to those who are 55 years and older, while ‘non-senior’ traveller refers to those less than 55 years of age, but more than 15 years of age. The senior traveller category is further segmented into two groups: ‘younger senior’ (55–64 years) and ‘older senior’ (65 years and

1

Day visitors are those visitors who travel for a round trip distance of at least 50 km, are away from home for at least 4 hours, and who do not spend a night away from home as part of their travel. Same day travel as part of overnight travel is excluded, as is routine travel such as commuting between work/school and home. Overnight visitors are defined as visitors who take a trip involving a stay away from home for at least one night, at a place at least 40 km from home. Only those trips where the respondent is away from home for less than 12 months are in scope.

Characteristics and Travel Patterns of Older Australians

503

over). Younger seniors have a high propensity to travel and are at the peak of their earning and spending power (Smith and Jenner, 1997), while people in the older age group are usually retired from the workforce and may be entitled to the age pension (Tomljenovic and Faulkner, 2000). The younger senior and older senior age groups were chosen because they conform to age groupings often reported in other studies (Javalgi et al., 1992; Backman et al., 1999; TIA, 2000). Data used in this study were extracted from the Bureau of Tourism Research’s (BTR)2 NVS. The survey is conducted using computerassisted telephone interviewing of around 80,000 Australians aged 15 years and over and is a major survey to identify the characteristics and travel patterns of Australian travellers. This study analyses data from the NVS for 2002.

2 Expenditure by Senior Travellers, 2000 In 2002, senior domestic overnight and day visitors spent $10.8 billion on their travels in Australia, accounting for 21 per cent of national domestic tourism expenditure. On the other hand seniors spent $4.0 billion on outbound travel (Figure 1). This high proportion of expenditure by domestic senior visitors was partly due to their willingness to allocate a significant amount of their discretionary expenditure to travel. Senior travellers tend to spend more because their home-related debts are paid-off and their children no longer depend on them (Anderson and Langmeyer, 1982). Of the $10.8 billion total expenditure in Australia by domestic day and overnight senior travellers, $8.5 billion was by overnight senior travellers and $2.3 billion by daytrip senior travellers. Around $6.3 billion was spent by younger senior travellers and $4.5 billion by older senior travellers. The higher expenditure by younger senior travellers compared to that of older senior travellers was because many younger seniors had not retired and were at the peak of their earnings. Similar results were also observed for other countries. For example, using the 1995 US Consumer Expenditure Survey data, Hong et al. (1999) reported that American younger seniors (55–64 years) were likely to spend more on domestic trips than older seniors (65 years and over). 2

The Bureau of Tourism Research changed to Tourism Research Australia (TRA) from 01 July 2004. TRA is a Division of Tourism Australia (TA). TA is a statutory body established 01 July 2004, incorporating four former organisations: Australian Tourist Commission (ATC), Bureau of Tourism Research (BTR), Tourism Forecasting Council (TFC) and See Australia.

504 Figure 1:

Afzal Hossain et al. Total Expenditure by Domestic Senior Travellers, 2000

Outbound $4.0 billion (27 per cent) ($1.6 billion from older seniors)

Domestic day $2.3 billion (16 per cent) ($1.0 billion from older seniors)

Domestic overnight $8.5 billion (57 per cent) ($3.5 billion from older seniors)

3. Profile of Senior Overnight Travellers Domestic senior travellers (55 years and over) are making a significant contribution to the domestic tourism market. In 2002, around 16.7 million domestic overnight trips were undertaken by senior travellers, accounting for 22 per cent of all domestic overnight trips in Australia (Table 1). These senior overnight travellers spent a total of 91.5 million nights, accounting for 31 per cent of total nights spent by all domestic overnight visitors. This equates to an average length of stay of 5.5 nights, higher than the average length of stay of 3.5 nights spent by non-senior (15–54 years) overnight travellers. Other studies in Australia and overseas have also shown that senior travellers tend to stay away longer than travellers in any other age cohort (Shoemaker, 1989; Golik, 1999; Fleischer and Pizam, 2002; Hossain, 2003). Of the total overnight trips undertaken by senior travellers in 2002, 9.2 million were by younger senior travellers (55–64 years) and 7.5 millions were by older senior travellers (65 years and over), corresponding to 55 and 45 per cent of the total senior overnight trips, respectively. Among the senior overnight travellers, older seniors tended to take longer trips (6.2 nights) compared to 4.9 nights on average by younger seniors. Senior overnight travellers on average spent less money, both on a daily basis and per overnight trip, than non-senior travellers. Each senior traveller

Characteristics and Travel Patterns of Older Australians Table 1:

505

Domestic Overnight Trips, Visitor Nights and Expenditure,a 2002 Non-senior (15–54 years)

Overnight trips (million) Visitor nights (million) Average length of stay (nights) Total expenditure ($ billion) Average expenditure per trip ($) Average expenditure per night ($)

58.6 207.1 3.5 31.4 536 152

Senior Total (55 years and over)

Younger (55–64 years)

Older (65 years and over)

16.7 91.5 5.5 8.6 509 93

9.2 44.7 4.9 5.1 551 113

7.5 46.8 6.2 3.5 458 74

Source: BTR (2002) a Excluding expenditure on capital items and the purchase of motor vehicles.

on average spent $509 per overnight trip and $93 per night, while each nonsenior traveller spent $536 and $152, respectively. Younger senior travellers spent more money on their trips than older senior travellers. Of the $8.5 billion expenditure by senior travellers in 2002 on overnight trips, $5.1 billion was spent by younger seniors — much more than the amount spent by older seniors ($3.5 billion). These younger senior travellers spent more on average per trip ($551) and per night ($113) than older senior travellers, who on average spent $458 per trip and $74 per night. These patterns of expenditure are similar to those of younger and older senior travellers, who travelled in 2001 (Hossain, 2003). 3.1 Demographic Characteristics There are marked differences in demographic characteristics (gender, marital status, employment status and annual household income) between senior and non-senior as well as between younger and older senior overnight travellers (Table 2). About three-quarters of senior travellers were part of a couple, while about six in every 10 were retired. 3.2 Purpose of Overnight Trips Holiday and visiting friends and relatives (VFR) remain the most favoured trip purposes for both senior and non-senior overnight travellers (Table 3). This study supports the findings of Anderson and Langmeyer (1982), who compared those under and over the age of 50 years, indicating that both groups travelled for reasons of holiday or visiting friends/relatives. However, there were differences in average length of stay by senior and non-senior

Afzal Hossain et al.

506 Table 2:

Demographic Characteristics of Overnight Travellers, 2002 Non-senior (per cent)

Gender Male Female Marital status Singlea Part of a coupleb Refused Employment status Working full time or part time Retired or on a pension Otherc Annual household income (per annum) $1–25,999 $26,000–51 999 $52,000–103 999 $104,000 and more Refused/do not know Total overnight visitors (million)

Senior (per cent) Total

Younger

Older

54.1 45.9

50.0 50.0

53.4 46.6

45.8 54.2

37.0 62.8 0.2

23.2 76.1 0.6

17.3 82.1 0.7

30.5 68.9 0.6

77.6 1.9 20.5

35.4 58.3 6.3

55.9 35.3 8.7

10.3 86.3 3.3

6.0 21.1 36.8 19.0 17.1 58.6

30.9 22.8 15.8 8.5 22.0 16.7

18.2 24.6 22.1 13.3 21.7 9.2

46.4 20.6 8.1 2.6 22.3 7.5

Source: BTR (2002) Note: Owing to rounding, figures may not add to total. a Never married, divorced, separated, widowed or not part of a couple. b Married, de facto or living together. c All other categories, studying, refused or do not know.

travellers. For example, in regard to holiday travel, senior travellers stayed significantly longer (6.8 nights) than non-senior travellers (4 nights). Further analysis revealed that senior women took more overnight trips to visit friends and relatives in 2002 and stayed longer than their counterpart male travellers, while the opposite pattern was observed for the business category (data not shown). However, there was no difference between senior male and female travellers who travelled for holidays or other reasons. 3.3 Accommodation Used Many of the senior travellers stayed at a friend’s or relative’s property (46 per cent) and spent there 45 per cent of the total visitor nights, with an average length of stay of 5.4 nights. However, older senior travellers tended to stay more in a friend’s or relative’s property and stayed longer than younger senior travellers.

Characteristics and Travel Patterns of Older Australians

507

Table 3: Overnight Visitors, Visitor Nights and Average Length of Stay by Trip Purpose, Domestic Travellers in Australia, 2002 Non-senior

Senior Total

Overnight visitors (per cent) Holiday VFR Business Other Visitor nights (per cent) Holiday VFR Business Other Average length of stay (nights) Holiday VFR Business Other

Younger

Older

42 31 22 4

41 42 11 6

41 37 17 5

40 49 4 6

47 28 20 5

50 39 7 4

54 31 12 3

47 45 2 5

4.0 3.1 3.2 4.1

6.8 5.0 3.3 4.1

6.4 4.1 3.3 3.2

7.2 5.8 3.3 4.9

Source: BTR (2002) Note: VFR, Visiting friends and relatives.

The second most popular form of accommodation for senior travellers was hotel, resort, motel and motor inn (27 per cent). On average, senior visitors spent 3.6 nights in these establishments. Seven per cent of senior travellers stayed in a rented house/apartment/flat/unit and spent 9 per cent of the total nights in this type of accommodation, with an average duration of stay of 7.6 nights — while 9 per cent of senior travellers stayed in caravan parks or in commercial camping grounds or camping near a road or on private properties and tended to stay for more nights than non-senior travellers, with an average length of stay of 8.5 nights. 3.4 Activities Undertaken In 2002, 87 per cent of senior overnight travellers participated in social or other activities, or undertook art/heritage activities. The proportion of senior overnight travellers who undertook an active outdoor or sport activity was lower than the proportion of non-senior overnight travellers (16 and 26 per cent, respectively), although bushwalking/rainforest walking was popular among senior overnight travellers (9 per cent). This is due to the relative physical well-being (i.e. health status — objective and self-reported) and mobility concerns of older seniors (McGuire, 1984). Only 13 per cent of

Afzal Hossain et al.

508 Table 4:

Overnight Trip Expenditure by Purpose, 2002 Holiday

VFR

Average expenditure per overnight trip ($) Non-senior 625 322 Senior 703 314 Younger senior 730 330 Older senior 670 300 Average expenditure per night ($) Non-Senior 158 102 Senior 104 63 Younger senior 114 80 Older senior 93 52

Business

Other

All visitors

702 605 639 433

383 383 420 349

536 509 551 458

221 183 193 129

94 94 132 71

152 93 113 74

Source: BTR (2002)

older overnight travellers undertook active outdoor/sport activities, much less than the proportion of younger senior overnight travellers (18 per cent). This result supports findings from other studies (Golik, 1999; Fleischer and Pizam, 2002) that senior travellers tend to undertake different activities compared to non-senior travellers. 3.5 Expenditure by Purpose of Overnight Trips Data for expenditure by overnight trip purpose show some differences between senior and non-senior overnight travellers. In terms of overnight trip expenditure, VFR and business senior overnight travellers spent comparatively less than non-senior travellers, but senior holiday visitors spent more than the non-senior holiday travellers (Table 4). Similar results were also reported for USA mature travellers (TIA, 2000). Average expenditure per night was higher for non-senior overnight travellers than senior overnight travellers, irrespective of trip purpose. Similarly, younger senior travellers spent more each night than older seniors. Further analysis revealed that expenditure per visitor and per night increased with increase in average annual household income, irrespective of trip purpose. 3.6 Expenditure by Item The itemised expenditure for senior and non-senior overnight travellers is shown in Table 5. Of the $8.5 billion spent by senior overnight travellers in 2002, the largest expenditure item was accommodation (23.1 per cent), followed by takeaway and restaurant meals (15.2 per cent), fuel (11.6 per cent) and shopping, gifts or souvenirs (10.5 per cent), a pattern similar to non-senior overnight travellers. However, senior overnight travellers spent

Characteristics and Travel Patterns of Older Australians Table 5:

509

Itemised Expenditure by Overnight Travellers, 2002

Expenditure items

Non-senior (per cent)

Senior (per cent) Total Younger Older

Accommodation (can include food) Takeaways and restaurant meals Fuel Shopping, gifts, souvenirs Airfares Package tours Groceries for self-catering Alcohol, drinks (not already reported) Othera

21.7 14.8 11.2 11.5 12.4 3.0 5.3 6.2 13.8

23.1 15.2 11.6 10.5 9.4 6.7 6.3 3.2 14

23.8 15.3 12.3 10.2 10.4 4.0 6.4 3.8 13.9

22.1 15.0 10.7 11.1 7.9 10.7 6.1 2.3 14.1

Total expenditure ($billion)

31.4

8.5

5.1

3.5

Source: BTR (2002) a Entertainment, museums, movies, other long distance transport, car hire, organised tours, vehicle maintenance/repairs, gambling, taxis, conference fees, other local transport and education or course fees.

Table 6:

Domestic Daytrip Visitors and Expenditure,a 2002 Non-senior

Daytrip visitors (million) Daytrip expenditure ($billion) Average expenditure per daytrip ($)

105.8 9.6 90

Senior Total

Younger

Older

36.4 2.3 64

17.4 1.3 74

19.0 1.0 54

Source: BTR (2002) a Excluding expenditure on capital items and the purchase of motor vehicles.

9.4 per cent of their total expenditure on airfares, less than the proportion of total expenditure by non-senior overnight travellers on this category (12.4 per cent).

4. Profile of Senior Day Travellers In 2002, domestic senior day travellers took an estimated 36.4 million daytrips in Australia, accounting for 26 per cent of all domestic daytrips and spent $2.3 billion, accounting for 20 per cent of total daytrip expenditure in Australia (Table 6). Among the senior day travellers, older seniors took more day trips but spent less than younger seniors.

510 Table 7:

Afzal Hossain et al. Demographic Characteristics of Domestic Day Travellers, 2002 Non-senior (per cent)

Gender Male Female Marital status Singlea Part of a coupleb Refused Employment status Working full time or part time Retired or on a pension Otherc Annual household income (per annum) $1–$15,599 $15,600–51,999 $52,000–103,999 $104,000 or more Refused or do not know Total day visitors (million)

Senior (per cent) Total

Younger

Older

52.1 47.9

48.5 51.5

50.1 49.9

47.1 52.9

40.1 59.6 0.4

25.3 74.4 0.3

18.0 81.7 0.3

32.0 67.7 0.3

69.9 3.3 26.8

25.9 67.6 6.5

47.7 40.7 11.5

5.9 92.3 1.8

8.6 24.4 34.2 14.0 18.8 105.8

38.3 22.4 14.0 4.6 20.7 36.4

23.0 25.2 23.8 8.2 19.8 17.4

52.4 19.8 5.0 1.3 21.5 19.0

Source: BTR (2002) Note: Due to rounding, figures may not add to total. a Married, divorced, separated, widowed or not part of a couple. b Married, de-facto or living together. c All other categories, studying, refused or do not know.

4.1 Demographic Characteristics There are marked differences in travel patterns for different demographic groups (Table 7). Of the 36.4 million senior day travellers in 2002, nearly 49 per cent were male and around 51 per cent were female, opposite to the gender mix of non-senior day travellers. In the younger senior group, slightly more males than females undertook day trips, while the reverse situation was observed for the older senior group. This result is likely due to the greater longevity of females. In 2002, the majority of senior day travellers were part of a couple (74 per cent), a higher percentage than for non-senior travellers, of whom 60 per cent travelled as a couple. Two-thirds (68 per cent) of senior day travellers were either retired or on a pension and 26 per cent were working either full time or part time. In 2002, average annual household income for senior day travellers was lower than that of non-senior day travellers. However, 21 per cent of senior

Characteristics and Travel Patterns of Older Australians

511

day travellers refused to report or did not know their annual income. Although seniors have less income, they were willing to spend a significant amount on travel. This is probably due to their savings and lower expenditures than younger senior travellers on items such as mortgage repayments. 4.2 Purpose of Day Trips There were no notable differences between senior and non-senior day travellers’ purpose of travel. Taking a holiday and VFR were the two main reasons for taking day trips by both senior and non-senior day travellers. There was no marked difference between younger and older seniors who took day trips for either holiday reasons or to visits friends or relatives. As might be expected, older seniors took less day trips for business purposes compared to younger seniors, because a large proportion of older seniors are no longer in paid employment. 4.3 Leisure Activities In 2002 the proportion of senior day travellers who undertook outdoor or nature activities and active outdoor or sport activities was lower than the corresponding proportion of non-senior day travellers undertaking these activities. The proportion of senior day travellers who undertook arts/ heritage or festival activities, local attractions or activities and social/other activities was similar to the proportion of non-senior day travellers. As noted for overnight senior travellers, older senior day travellers tended to take less active activities compared to younger senior day travellers. 4.4 Expenditure by Purpose of Day Trips In terms of expenditure per day trip, senior day travellers spent less than non-senior day travellers, irrespective of trip purpose. When comparing younger and older seniors, younger senior day travellers tended to spend more per trip than older seniors, particularly for holiday and business travel. 4.5 Day Trip Expenditure by Item The major expenditure items for domestic senior and non-senior day travellers were shopping/gifts/souvenirs, fuel (petrol and diesel) and food (takeaways and restaurant meals and groceries etc. for self-catering), which comprised 82 per cent of total expenditure by senior day travellers and 78 per cent by non-senior day travellers (Table 8). In 2002, senior day travellers spent a relatively higher share of their travel budget on each of these items compared to non-senior day travellers.

512 Table 8:

Afzal Hossain et al. Major Expenditure Items by Day Travellers, 2002 Non-senior (per cent)

Shopping, gifts, souvenirs Fuel (petrol, diesel) Takeaways and restaurant meals Groceries etc. for self-catering Total expenditure ($million)a

30.8 21.8 18.9 6.0 9,562

Senior (per cent) Total

Younger

Older

32.3 23.5 19.5 6.5 2,317

34.2 24.4 18.5 5.0 1,293

29.9 22.4 20.9 8.3 1,024

Source: BTR (2002) a Expenditure on capital items and on the purchase of motor vehicles is not included in total expenditure.

Table 9: 2002

Outbound Trips, Total Nights and Expenditure by Australian Residents, Non-senior (per cent)

Senior Total

Outbound trip (million) Total nights (million) Average stay (nights) Expenditurea ($billion) Expenditure per outbound trip ($)

2.41 52.6 22 12.12 5,024

Younger Older

0.66 0.41 0.25 16.8 8.7 8.0 25 21 32 4.02 2.45 1.57 6,057 5,964 6,207

Source: BTR (2002) a Expenditure excludes motor vehicles and other major equipment.

5. Senior Outbound Travellers In 2002 Australian residents made nearly 3.1 million overseas trips, of which senior travellers made an estimated 663,000 outbound trips (or 22 per cent of the total). Senior travellers spent around 16.8 million nights away from home, with an average trip length of 25 nights, slightly longer than for nonsenior outbound travellers (22 nights) (see Table 9). Senior outbound travellers spent just over $4.0 billion on overseas trips in 2002, with an average expenditure of nearly $6,100 per trip. This average expenditure per trip for senior outbound travellers was around 21 per cent higher than the average trip expenditure by non-senior outbound travellers. Although the number of older senior outbound travellers was less than the number of younger senior outbound travellers, they tended to stay away longer and spend more on trips than younger senior outbound travellers.

Characteristics and Travel Patterns of Older Australians

513

6. Future Prospects for the Senior Travel Sector As shown earlier, travel by people over 55 years currently represents a significant market segment of the Australian domestic travel sector. Over the next 20 years, as the senior cohort is expected to increase, further resources and infrastructure requirements will be required to meet this growing demand from senior travellers. This section provides one way to value the potential growth of the senior travel sector in the next 20 years, by linking the scenarios for the Australian population, as published by the Australian Bureau of Statistics (ABS)3 to activity and expenditure data for senior travel in the 2002 edition of the NVS. The projections presented in this paper do not compete with current forecasts published by the Tourism Forecasting Council (TFC) (2003). Instead, this work should be viewed as isolating the likely potential demographic impacts from other potential influences, such as economic growth and relative cost of travel, which are incorporated into the TFC forecasts. This section comprises two components. The first component links the ABS’ population projections with the current expenditure information from the NVS. The second component provides a qualitative assessment of the likely sensitivities surrounding the projections. A slight change in the definition of the older senior traveller is made in this section. Currently, the cohort aged 85 years and over represents only 1.4 per cent of Australia’s population, but its share is likely to increase over the projection period to 2022. Correspondingly, this cohort is also less likely to travel over the projection period, reflecting decreased ‘average’ health and mobility and is not likely to conduct a significant amount of travel. On this basis, the older senior cohort has been truncated to now represent the 65– 84-year cohort, with forecasts to be benchmarked against the NVS data for the ‘65 years and over’ traveller profile. 6.1 Population Growth Scenarios The Australian Bureau of Statistics (2002a) describes three population projection scenarios representing the likely range of outcomes based on changes to three main components: changes to the net overseas migration (NOM), fertility rates and life expectancy (at birth). Assumptions used to develop our scenarios are that the total fertility rate is unlikely to be substantially 3 The ABS describes these scenarios as ‘not intended as projections or forecasts but as illustrations of growth and change in the population which would occur if certain assumptions about future levels of fertility, mortality and net overseas migration (NOM) were to prevail over the projection period’ (ABS, 2002b).

514

Afzal Hossain et al.

Table 10: Population Projections and Average Annual Growth Rates 2002

Population, 15– 54 years (million) Share 15–54 years (per cent) Average annual growtha (per cent) Population, 455 years (million) Share 455 years (per cent) Average annual growtha (per cent) Australian population (million) Average annual growtha (per cent) Senior travellers Population, 55– 64 years (million) Share 55–64 years (per cent) Average annual growtha (per cent) Population, 65– 84 years (million) Share 65–84 years (per cent) Average annual growtha (per cent)

Population projections 2002–2022 High

Baseline

Low

11.3 57.3 0.9 4.4 22.5 3.1 19.7 1.2

12.7 51.2 0.6 7.7 31.3 2.7 24.7 1.1

12.2 51.9 0.4 7.6 32.2 2.6 23.5 0.9

11.7 52.3 0.2 7.5 33.4 2.5 22.4 0.6

1.9 9.8 4.4 2.2 11.2 1.7

3.0 12.3 2.2 4.0 16.3 2.9

3.0 12.7 2.1 4.0 16.9 2.8

2.9 13.2 2.0 3.9 17.6 2.8

Source: ABS (2002b) a For 2002 is the average growth rate between 1998 and 2002.

different to the current level of 1.7 births per woman. Similarly, the NOM is forecast to be near to the 10-year average of 100,000 persons under the base scenario, with the low growth scenario assuming 80,000 persons and 125,000 assumed for the high-growth scenario. Using these assumptions, under the baseline scenario, Australia’s population is projected to increase by 3.8–23.5 million persons in 2022, representing an average annual growth rate of 0.9 per cent. Using the assumptions listed above, Australia’s population could range between 22.4 and 24.7 million people by 2022 (Table 10). The population of people aged 55 years and over is forecast to increase by 3.2 million people to 7.6 million people by 2022. However, the number of senior persons is unlikely to vary significantly over the projection period. The range presented in the high and low growth scenarios indicate that the number of seniors may range from 7.5 to 7.7 million persons, reflecting the fact that the assumptions used in this assessment impact more on the ‘replacement rate’ (or the size of the non-senior cohorts). The average growth rates disguise the variation that is likely to occur for each cohort (Figure 2). While the growth rate for the younger 15–54 age cohort is projected to remain low, younger seniors are projected to fall considerably from nearly 5 to 0.3 per cent. At the same time, the older senior

Characteristics and Travel Patterns of Older Australians Figure 2:

515

Population Projections by Cohorts, Percentage Change, Baseline Scenario Per cent

Per cent 15-54 years 5

5

55-64 years 65-84 years

4

4

3

3

2

2

1

1 0

0 2002

2007

2012

2017

2022

Source: ABS (2002b)

cohort is projected to increase from around 2 to 3 per cent for much of the projection period,4 resulting in an increase of nearly 1.8 million persons. Under the baseline growth assumption, the senior cohort is projected to increase, with the share of seniors relative to the total Australian population expected to increase to 32 per cent, up by nearly 10 percentage points (Figure 3a). Of that cohort, the share of younger seniors increases faster in the period to 2010, rising from 10 to 12 per cent. However, as growth in this cohort begins to taper, growth in the older senior cohort increases, reflecting the baby boomer generation shifting into the older cohort (Figure 3b). Older seniors are projected to increase from 11 to 17 per cent of the projected population, or an increase of nearly 1.8 million persons in the projection period to 2022. The older senior cohort is projected to increase to 4.0 million persons, representing an average annual increase of 2.8 per cent over the period to 2022, slightly faster than the younger senior cohort, which is forecast to increase at a rate of 2.1 per cent to 3.0 million by the year 2022, still higher than the projected total population growth rate of 0.9 per cent. 6.2 Forecasts for Seniors Travel Activity and Expenditure Changes to the Australian demographic structure are likely to be a leading influence on the Australian domestic travel sector. As there is inadequate time series data available to conduct a full econometric analysis, this potential influence can be examined by calculating the propensity to travel 4

As indicated in Figure 2, the significant increase in the older senior cohort in 2012 (matched by a decline in the younger senior cohort) reflects the movement between cohorts of persons born in 1947.

Afzal Hossain et al.

516

Figure 3: Cohort Share of Australia’s Population, Baseline Scenario. (a) Non-seniors and seniors and (b) Younger and Older Seniors a) Percentage of Aus population

Percentage of Aus population

60

60

50 40

50

55 years and over Share in 2002: 22% Share in 2022: 32%

15-54 years Share in 2002: 57% Share in 2022: 52%

40

30

30 20

20 2002

2007

2012

2017

2022

b) Percentage of Aus population

Percentage of Aus population

20

20 65-84 years Share in 2002: 11% Share in 2022: 17%

16

16

12

55-64 years Share in 2002: 10% Share in 2022: 13%

8 2002

2007

2012

2017

12

8 2022

Source: ABS (2002b)

(defined as the quantity of travel for each cohort divided by the population cohort) for each type of travel — day trip, overnight trip and overseas trip. Expenditure projections can also be calculated by multiplying average trip expenditure by trip activity. The assumptions used in this paper for travel activity propensities for the 2002 calendar year for trip type and gender are presented in Table 11. For example, non-senior males on average took 9.8-day trips per year and 5.6-overnight trips. Around one-quarter of non-senior males took an outbound trip, around the same rate as younger seniors. Using these assumptions, senior travel is projected to increase by around 65–70 per cent (i.e. index ¼ 165 to 170) for each type of trip in the period to

Characteristics and Travel Patterns of Older Australians

517

Table 11: Propensity Assumptions, by Type of Trip, Cohort and Sex, 2002 Activity propensities Day trips

15–54 years 55–64 years 65–84 years

Overnight trips

Overseas trips

Male

Female

Male

Female

Male

Female

9.8 8.9 8.8

9.0 9.1 8.4

5.6 5.0 3.4

4.8 4.5 3.4

0.25 0.25 0.14

0.18 0.18 0.09

2002

2012

2022

Change

Indexa

Indexa

Indexa

Per centb

100 100 100 100 100

106 132 137 129 113

108 168 155 180 124

0.4 2.6 2.2 3.0 1.1

100 100 100 100 100

107 134 133 134 112

109 167 152 188 121

0.4 2.6 2.1 3.2 1.0

100 100 100 100 100

106 134 136 129 112

109 165 155 181 121

0.4 2.5 2.2 3.0 0.9

Source: BTR (2002); ABS (2002b)

Table 12: Activity Projections

Day trips 15–54 years (non-senior) 55 years and over (senior) 55–64 years (younger senior) 65–84 years (older senior) All travel Overnight trips 15–54 years (non-senior) 55 years and over (senior) 55–64 years (younger senior) 65–84 years (older senior) All travel Outbound trips 15–54 years (non-senior) 55 years and over (senior) 55–64 years (younger senior) 65–84 years (older senior) All travel

Source: BTR (2002) a The percentage change from the 2002 estimate. That is, values for 2002 equal to 100. b Average annual.

2022, substantially higher than the non-senior traveller cohort increase of 8–9 per cent (Table 12). As a cohort, older senior travel is forecast to increase fastest, increasing by around 80–88 per cent in the period from 2002 to 2022, at an average annual growth rate of around 3.0–3.2 per cent depending on the type of travel.

Afzal Hossain et al.

518

Table 13: Average Expenditure per Trip, by Type of Travel, Cohort and Sex Day trip

Overnight trip

Overseas trip

Male

Female

Male

Female

Male

Female

90.2 77.8 61.4

90.6 70.8 47.4

597 640 578

464 448 357

5,238 6,581 6,924

4,728 5,084 5,319

15–54 years 55–64 years 65–84 years Source: BTR (2002)

Table 14: Expenditure Projections for Years 2002, 2012 and 2022

Type of trips Day trips Overnight trips Outbound trips Age cohort 15–54 years (non-senior) 55 years and over (senior) 55–64 years (younger senior) 65–84 years (older senior) All travel expenditure Cohort share (per cent) 15–54 years (non-senior) 55 years and over (senior) 55–64 years (younger senior) 65–84 years (older senior)

2002

2012

2022

Growth

11.9 39.9 16.1

$ billion 13.3 44.8 18.3

14.2 48.2 19.8

per centa 0.9 0.9 1.0

53.1 14.8 8.8 6.1 67.9

56.5 19.8 11.9 7.9 76.3

57.7 24.6 13.6 11.0 82.2

0.4 2.6 2.2 3.0 1.0

78.2 21.8 12.9 8.9

74.0 26.0 15.7 10.3

70.1 29.9 16.5 13.4

Source: BTR (2002) a Average annual.

Total expenditure projections are calculated by multiplying average trip expenditure by total activity for each cohort. It is also assumed average per trip travel expenditure will remain constant for the projection period to 2022. Using these assumptions, younger seniors will spend $78 for every day trip and $640 for every domestic overnight trip each year to 2022 (Table 13). Interestingly, for outbound travel, on average, seniors will spend considerably more when compared to younger travellers, possibly reflecting a decision to undertake travel on retirement. Also, senior females for all cohorts are assumed to spend significantly less on each trip when compared to their male counterparts. Total expenditure by age cohort and travel type is presented in Table 14. If demographic change is the main factor, tourism expenditure is

Characteristics and Travel Patterns of Older Australians

519

Figure 4: Expenditure Projections (percentage change). (a) Non-seniors and seniors and (b) Younger and Older Seniors a)

Per cent

Per cent 4.0

55 years and up Growth: 2.6% per year

3.0

3.0 2.0

4.0

15-54 years Growth: 0.4%per year

2.0

1.0

1.0

0.0 2002

2007

2012

2017

0.0 2022

b) Per cent

Per cent 65 to 84 years Growth: 3.0% per year

5.0

5.0

4.0

4.0

3.0

3.0 2.0

2.0 55-64 years Growth: 2.2% per year

1.0

1.0

0.0 2002

2007

2012

2017

0.0 2022

projected to grow by $14 billion (or 1.0 per cent per year to 2022) to $82 billion — marginally higher than the rate of population growth over the same period. Senior travel expenditure is projected to contribute around 68 per cent of the total growth in expenditure (around $9.7 billion) over the same period, with older seniors assumed to spend more than younger seniors. The share of senior travel expenditure as a proportion of total tourism expenditure is expected to increase from 22 per cent in 2002 to nearly 30 per cent in 2022. Also, growth in the senior traveller expenditure segment is expected to steadily decrease over the projection period (Figure 4a). Younger seniors are forecast to provide most of the growth in the period to 2010, with older seniors expected to maintain higher growth over the remainder of the projection period to 2022 (Figure 4b).

520

Afzal Hossain et al.

7. Other Influences By extrapolating the likely changes to Australia’s demographic mix, it can be seen that senior travellers will remain as a growing segment for the Australian travel sector. As identified in overseas studies5 of travel segments, higher disposable incomes/assets and an increase in average education attainment are positively correlated to increases in travel. Increases in these two factors are likely to boost current propensities for travel by senior Australians. Other social and economic factors could also contribute to senior Australian travel activity and expenditure, but many of these are difficult to estimate in a quantitative demographic-response framework. These factors include: 







early retirement and associated increases in leisure time, contributing to likely higher travel activity; variability (but likely increase) in average private retirement incomes (superannuation and other income producing assets) related to the increase in self-funded retirees, with the study by Harding et al. (2002), pointing to an increasing share of wealth held by older Australians; the eligibility of older persons for transfer payments (including travel cards for aged pensions); and health and mobility of older persons: increases to life expectancy and mobility are also likely to boost travel activity for older persons.

The risks involved in forecasting each of these factors are considerable (and heavily dependent on government policy over time), but are likely to provide ‘more upside’ to senior travel and expenditure propensities. Another possible positive effect for the travel sector of additional increases in senior travel is a likely reduction in the seasonality of domestic travel (particularly for holiday and VFR travel where seniors are not tied down to school holidays for holiday travel). However, on the downside, the Australian travel sector may become more sensitive to negative economic or political shocks (or terrorism acts), reflecting the higher risk aversion of older persons. For example, seniors may decide not to travel (or take a longer time to regain confidence to travel) if threats of, or terrorism acts occur in one or more favoured outbound or domestic destinations.

5

See TIA (2000) for these and other potential factors relevant to the North American senior travel market.

Characteristics and Travel Patterns of Older Australians

521

8. Conclusion In 2002 Australian seniors spent an estimated $10.8 billion on their domestic trips, accounting for 21 per cent of total expenditure on domestic tourism in that year. Of this expenditure, senior travellers spent $8.5 billion on overnight travel and $2.3 billion on day trips. Twenty two per cent of all domestic overnight trips and 26 per cent of all domestic day trips in 2002 were by senior travellers. The likely change to the demographic mix of Australia’s population reveals the potential for the senior travel sector. By adopting current population forecasts and current propensities to travel and spend by Australians, travel expenditure is projected to increase by $14.3–$82.2 billion (in 2002 dollars) by 2022. The senior cohort is projected to provide around two-thirds of this increase in tourism expenditure. However, as the changes between the non-senior, younger and older seniors indicate, the younger senior (baby-boomer) cohort is likely to provide the largest boost in spending by any cohort in the next 10 years. After this period, as the baby boomers enter the older cohort, changing demographics suggest that tourism growth will decline as the baby boomers enter the older senior cohort. However, this conclusion abstracts from other factors that influence travel decision making, such as economic growth and travel prices.

References Anderson, B. and Langmeyer, L. (1982). The Under 50 and Over 50 Travellers: A Profile of Similarities and Differences. Journal of Travel Research, 20, 20–24. Australian Bureau of Statistics (2002a). Australian Social Trends 2002. Cat. No. 4102.0, ABS, Canberra. Australian Bureau of Statistics (2002b). Population Projections, Australia, 2002– 2101. Cat. No. 3222.0, ABS, Canberra. Backman, K.F., Backman, S.J. and Silverberg, K.E. (1999). An Investigation into the Psychographics of Senior Nature-Based Travellers. Tourism Recreation Research, 24, 13–22. Bureau of Tourism Research (BTR) (2002). National Visitor Survey, 2002, Bureau of Tourism Research, Canberra, Australia. Cleaver, M. (2000). Australian Seniors’ Use of Travel Information Sources: Perceived Usefulness of Word-of-Mouth, Professional-Advice, Marketer-Dominated and General-Media Information. Proceedings of Australia New Zealand Marketing Academy Conference (ANZMAC), Griffith University, Brisbane, Australia, pp. 182–187. Cleaver, M. and Muller, T.E. (1998). Testing the Theoretical Link between Cognitive Age and Travel Motives: A New Way to Segment the Australian Senior Tourism Market. Proceedings of Australia New Zealand Marketing Academy Conference (ANZMAC), University of Otago, Otago, New Zealand, pp. 1634–1648.

522

Afzal Hossain et al.

Collins, D. and Tisdell, C. (2002). Gender and Differences in Travel Life Cycles. Journal of Travel Research, 41, 133–143. Fleischer, A. and Pizam, A. (2002). Tourism Constraints Among Israeli Seniors. Annals of Tourism Research, 29, 106–123. Golik, B. (1999). Not Over the Hill, Just Enjoying the View, Queensland Department of Families, Youth and Community Care, Brisbane, Australia. Harding, A., King, A. and Kelly, S. (2002). Incomes and Assets of Older Australians: Trends and Policy Implications. Agenda, 9(1), 3–18. Harding, A., Lloyd, R. and Greenwell, H. (2003). The spending patterns and other characteristics of low income households, in Zapalla, G. (ed), Barriers to Participation: Financial, Educational and Technological, Smith Family, Sydney (available from www.smithfamily.org.au). Hong, G.S., Kim, S.Y. and Lee, J. (1999). Travel Expenditure Pattern of Elderly Households in the US. Tourism Recreation Research, 24, 43–52. Horneman, L., Carter, R.W., Wei, S. and Ruys, H. (2002). Profiling the Senior Traveller: An Australian Perspective. Journal of Travel Research, 41, 23–37. Hossain, A. (2003). Senior Travellers: Their Contribution to the Domestic Tourism Market in Australia. Tourism Research Report, 5, 1–18. Javalgi, R.G., Thomas, E.G. and Rao, S.R. (1992). Consumer Behaviour in the US Travel Marketplace: An Analysis of Senior and Nonsenior Travellers. Journal of Travel Research, 31, 14–19. Lawton, L. (2001). A Profile of Older Adult Ecotourists in Australia. Journal of Hospitality & Leisure Management, 9, 113–132. McGuire, F.A. (1984). A Factor Analytical Study of Leisure Constraints in Advanced Adulthood. Leisure Sciences, 6, 313–326. Ruys, H. and Wei, S. (1998). Accommodation Needs of Mature Australian Travellers. Australian Journal of Hospitality Management, 4, 51–59. Shoemaker, S. (1989). Segmentation of the Senior Pleasure Travel Market. Journal of Travel Research, 27, 14–21. Smith, C. and Jenner, P. (1997). The Seniors’ Travel Market. Travel & Tourism Analyst, 5, 43–62. Teaff, J.D. and Turpin, T. (1996). Travel and the Elderly. Parks & Recreation, 31, 16–19. Tomljenovic, R. and Faulkner, B. (2000). Tourism and Older Residents in a Sunbelt Resort. Annals of Tourism Research, 27, 93–114. Tourism Forecasting Council (TFC) (2003). Latest Forecasts. TFC Secretariat, Department of Industry, Tourism and Resources, Canberra, Australia. Travel Industry Association of America (TIA) (2000). The Mature Traveller, TIA, Washington, DC.

Subject Index

ACTUCAN, 150, 168 ASTER, 8 Behavioural response, 6, 10, 112, 261, 284, 347 CALMAR, 318 Canada Pension Plan (CPP), 16, 17, 20, 143–145, 167, 291, 294–299, 302–305, 307, 310, 312, 349, 351 Cash transfers, 3, 5, 226, 270, 271, 274, 278, 281, 315, 316, 318 Child care, 18, 209, 211, 219, 317, 349 Conference themes, 3, 9, 10 Consumption, 6, 22, 48, 50, 110–113, 115–117, 119, 124, 126–128, 133–138, 206, 212, 213, 220, 272–277, 279, 280, 362, 389, 409–411, 414, 415, 417–427, 429, 433, 436–441, 443, 450–455, 464–466, 468, 471, 472, 496 CORSIM model, 9, 162, 390 Customer service projection (CuSP) model, 330, 331 Danish law model system, 235, 236, 238, 240, 244 Danish tax reform, 244 DESTINIE, 9, 391 DYNACAN, 9, 16, 17, 143–169, 391 Dynamic cohort models, 7, 9, 59, 207 Dynamic microsimulation alignment, 17, 149, 155, 156, 159, 161, 162, 164–168 future trends, 8 recent trends, 9 variance reduction, 17 Dynamic population models, 6, 13

DYNAMOD, 9, 21, 390, 391, 397, 399, 400, 403, 404 DYNASIM, 9 Economic growth, 2, 3, 82, 119, 210, 265, 271, 276, 277, 283, 323, 353, 480, 513 EUR3, 363 EUROMOD, 8 Firms, simulation of, 4, 347 GLADHISPANIA, 8, 21, 363, 382 Goods and services tax (GST), 316, 325, 327, 470 HARDING model, 9 Hypothetical models, 11 Imputation, 35, 58, 60, 211, 318, 397, 399 Income distribution lifetime, 7, 207, 242 and redistribution, 7, 233–258 total lifetime, 240–243 Income inequality, 215, 237, 240, 285, 362, 363 Income taxes annual impact of, 220, 247 lifetime impact of, 220, 247 Indirect taxes, 113, 135, 212–215, 217, 219, 221, 223, 225, 265, 274, 324, 325 Integrated macro–micro modelling, 19, 264–274 Labour supply, 17, 90, 121, 122, 130, 259, 266, 271, 274, 277, 279 LAW model, 8

524

Subject Index

LifePaths, 9, 391 LOTTE, 8 Macro–micro linkages, 19, 264 Maine conference, 9 MICROHUS model, 9 Microsimulation advantages and disadvantages, 56 future developments, 17, 496 historical developments, 253, 254 modelling infrastructure, 3 overview of models, 210, 234, 349 programming language, 316, 318 recent developments, 7, 8, 235, 331 static vs dynamic models, 4, 6, 7, 272, 347 tax policy analysis, 264, 362, 436 Microsimulation models dynamic cohort models, 4, 6, 7, 9, 207 dynamic population models, 6, 13 history of, 4 of firms, 4, 347 projection of data/estimates, 85–90 reliability of estimates, 36, 327 static models, 4, 5, 235 types of base data, 4, 316, 322, 326, 363 validation, 364 MITTS model, 6, 10 MOSART, 9, 15, 16, 81, 83–86, 90, 100, 101, 109–114, 121, 125, 126, 130, 275, 391 Non-cash benefits, 18, 36, 206, 208, 211–231, 215, 223, 224, 227–229 annual impact, 211, 215 Orcutt’s dream, 4, 8, 85 Panel data, 7, 19, 172, 184, 208, 234, 235, 239, 240, 243–245, 247, 257, 300, 362, 425

PHARMASIM, 8 Policy analysis, 23, 286, 467 Policy process, 20, 315–331 Poverty estimates, 329 Progressivity, tax, 281, 286, 373 PSSRU and NCCSU aged care model, 9 Replacement rates, 14, 22, 34, 48–52, 409–411, 414, 415, 419, 420, 424, 432, 433, 514 Retirement incomes, 7, 12, 17, 22, 48, 144, 145, 300, 304, 388, 391, 407–412, 416–420, 423–426, 429–431, 520 Revenue Canada – taxation statistics, 149 Reweighting, 4, 318 SAGE, 9, 391 SESIM model, 9, 14, 18, 34–36, 41, 44, 208–210, 212, 391 SFB3 model, 9 SIMTAB, 155, 156, 167 Social services, 17, 81 Spatial microsimulation, 10, 323, 328, 486, 499 SPSD/M, 8 Statistical matching, 208, 240 STATAX model, 32, 331 Static ageing, 4, 6, 24 Static models, 4–7, 59, 235, 244, 315, 347 STINMOD, 8, 20, 315–331 SVERIGE, 9 T1 statistical sample, 350 T1 tax analysis model, 21, 336, 349–350 Tax policy, 8, 264, 362, 436 TRIM3, 5, 8 TTSIM, 8 Typical taxpayer model, 11 Uprating, 4, 247, 256, 257, 318, 329

Subject Index Validation, 144, 146–149, 154, 157, 159, 160, 163, 165, 168, 169, 207, 364

525

Wage rates, 6, 19, 110, 113, 121, 122, 124, 128–130, 132–134, 256, 264, 266–268, 271–274, 277–280, 284, 337, 356

This page intentionally left blank