DYNAMIC MODELING AND APPLICATIONS FOR GLOBAL ECONOMIC ANALYSIS
A sequel to Global Trade Analysis: Modeling and Applicat...
98 downloads
1081 Views
6MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
DYNAMIC MODELING AND APPLICATIONS FOR GLOBAL ECONOMIC ANALYSIS
A sequel to Global Trade Analysis: Modeling and Applications (Cambridge University Press, 1997, edited by Thomas W. Hertel), this book presents the technical aspects of the Global Trade Analysis Program’s global dynamic framework (GDyn) and its applications within important global policy issues. The book covers a diverse set of topics including trade reform, growth, investment, technology, demographic change, and the environment. Environmental issues are particularly well suited for analysis with GDyn, and this book covers its uses with climate change, resource use, and technological progress in agriculture. Other applications presented in the book focus on integration issues such as rules governing foreign investment, e-commerce regulations, trade in services, harmonization of technical standards, sanitary and photo-sanitary regulations, streamlining of customs procedures, and demographic change and migration. Elena I. Ianchovichina is Lead Economist for the Middle East and North Africa region of The World Bank. Since joining The World Bank in 2000, she has served in its Research Department, East Asia and Pacific Region, and managed the Economic Policy and Debt Department’s program on inclusive growth. Her work has focused on country-specific analyses of economic growth, emerging Asia, and fiscal and trade reform. Dr. Ianchovichina has published more than twenty articles in a variety of journals, including the Canadian Journal of Economics, Contemporary Economic Policy, Review of International Economics, World Bank Economic Review, and Ecological Economics. She received Purdue University’s 2008 Apex award for outstanding contributions to quantitative trade analysis. Terrie L. Walmsley is an Associate Professor at Purdue University and a Principal Fellow and Associate Professor at the University of Melbourne, Australia. Dr. Walmsley is also the Director of the Center for Global Trade Analysis, the Purdue home of the Global Trade Analysis Project, a global network of 8,500 researchers from 150 countries (www.gtap.org). Dr. Walmsley leads the construction of the GTAP Data Base, a global database used worldwide to examine the impact of international trade and environmental policies. Her research has focused on international trade in goods and services and the movement of capital and labor across national boundaries; it has been used extensively by The World Bank.
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
Dynamic Modeling and Applications for Global Economic Analysis Edited by ELENA I. IANCHOVICHINA The World Bank
TERRIE L. WALMSLEY Purdue University and University of Melbourne, Australia
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
cambridge university press Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, S˜ao Paulo, Delhi, Mexico City Cambridge University Press 32 Avenue of the Americas, New York, NY 10013-2473, USA www.cambridge.org Information on this title: www.cambridge.org/9781107002432 C Cambridge University Press 2012
This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2012 Printed in the United States of America A catalog record for this publication is available from the British Library. Library of Congress Cataloging in Publication data Dynamic modeling and applications for global economic analysis / [edited by] Elena Ianchovichina, Terrie Walmsley. p. cm. Includes bibliographical references and index. ISBN 978-1-107-01169-4 (hardback) – ISBN 978-1-107-00243-2 (paperback) 1. International trade – Mathematical models. 2. International economic relations – Mathematical models. 3. International trade. 4. International economic relations. I. Ianchovichina, Elena. II. Walmsley, Terrie Louise. III. Title. HF1379.D957 2011 2011025080 337.01 5195–dc23 ISBN 978-1-107-01169-4 Hardback ISBN 978-1-107-00243-2 Paperback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party Internet Web sites referred to in this publication and does not guarantee that any content on such Web sites is, or will remain, accurate or appropriate.
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
Contents
Contributors
page vii
Acknowledgments
ix PART I INTRODUCTION AND OVERVIEW
1 Introduction Elena I. Ianchovichina
3
PART II STRUCTURE OF THE DYNAMIC GTAP FRAMEWORK
2 Theoretical Structure of Dynamic GTAP Elena I. Ianchovichina and Robert A. McDougall 3 Behavioral and Entropy Parameters in the Dynamic GTAP Model Alla Golub and Robert A. McDougall 4 An Overview of the Dynamic GTAP Data Base: The Data Base Construction and Aggregation Programs Robert A. McDougall, Terrie L. Walmsley, Alla Golub, Elena I. Ianchovichina, and Ken Itakura
13
71
120
5 A Baseline Scenario for the Dynamic GTAP Model Terrie L. Walmsley, Betina V. Dimaranan, and Robert A. McDougall
136
6 Welfare Analysis in the Dynamic GTAP Model Terrie L. Walmsley, Robert A. McDougall, and Elena I. Ianchovichina
158
7 Implementing the Dynamic GTAP Model in the RunDynam Software Ken Itakura, Elena I. Ianchovichina, Csilla Lakatos, and Terrie L. Walmsley v
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
173
vi
Contents PART III APPLICATIONS OF DYNAMIC GTAP
8 Assessing the Impact of China’s WTO Accession on Investment Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina 9 Dynamic Effects of the “New-Age” Free Trade Agreement between Japan and Singapore Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura
205
235
10 Resource Use and Technological Progress in Agriculture Elena I. Ianchovichina, Roy Darwin, and Robin Shoemaker
269
11 Global Economic Integration and Land-Use Change Alla Golub and Thomas W. Hertel
290
12 The Contribution of Productivity Linkages to the General Equilibrium Analysis of Free Trade Agreements Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer
312
13 Global Demographic Change, Labor Force Growth, and Economic Performance Rod Tyers and Qun Shi
342
PART IV EVALUATION OF THE DYNAMIC GTAP FRAMEWORK
14 Household Saving Behavior in the Dynamic GTAP Model: Evaluation and Revision Alla Golub and Robert A. McDougall 15 Implications for Global Economic Analysis Elena I. Ianchovichina and Terrie L. Walmsley Appendix: Negative Investment: Incorporating a Complementarity into the Dynamic GTAP Model Terrie L. Walmsley and Robert A. McDougall Glossary of GDyn Notation Terrie L. Walmsley Index
379 406
413
421
429
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
Contributors
Elena I. Ianchovichina (The World Bank, USA) Robert A. McDougall (Purdue University, USA) Terrie L. Walmsley (Purdue University, USA, and University of Melbourne, Australia, Australia) Roy Darwin (U.S. Department of Agriculture, USA) Betina V. Dimaranan (IFPRI, USA) Alla Golub (Purdue University, USA) Thomas W. Hertel (Purdue University, USA) Ken Itakura (Purdue University, USA) Csilla Lakatos (Purdue University, USA) Jeffrey J. Reimer (University of Wisconsin, USA) Qun Shi (Australian National University, Australia) Robin Shoemaker (U.S. Department of Agriculture, USA) Rod Tyers (University of Western Australia, Australia)
vii
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
Acknowledgments
We would like to thank a number of individuals for their help in making this book a reality. The dynamic GTAP work builds on the intellectual accomplishments of the GTAP project, and we are indebted to the many individuals who have contributed to GTAP over the years. We also recognize that the dynamic GTAP project would not have become a reality without the unwavering support of Professor Thomas W. Hertel – the founder of the GTAP project – who encouraged and steered this work in the right direction from the very beginning. We are grateful to several individuals for their advice and guidance in the early stages of the dynamic GTAP project. We were fortunate to have Robert A. McDougall on the dynamic GTAP team. His leadership and vast technical expertise were invaluable. Philippa Dee encouraged us to embark on this work because she foresaw the appeal of a dynamic GTAP version to analysts of a wide range of global economic policy issues. As in the case of GTAP, Alan Powell was generous with his time and support, whereas Ken Pearson and Mark Horridge helped us operationalize the dynamic model and address GEMPACK and other software issues. Over the years a number of researchers contributed to this project in different ways. Ken Itakura worked on updating the tab file to match the GTAP v6.2 tab file, whereas Csilla Lakatos developed postsimulation processing programs. We are grateful to the instructors of and participants in the dynamic GTAP short courses held at Purdue University in 2000, 2006, 2008, and 2010. These courses provided an opportunity for fruitful exchanges on important research questions concerning the theory behind the model and many of the applications discussed in this book. We would like to recognize in particular Robert A. McDougall, Anna Strutt, Thomas W. Hertel, Alla Golub, Csilla Lakatos, Ken Itakura, Amer Ahmed, Angel Aguiar, Peter Minor, Peter Dixon, and Kevin Hanslow. Finally, we are grateful to the three anonymous reviewers for their useful insights and comments. ix
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
PART I
INTRODUCTION AND OVERVIEW
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
ONE
Introduction Elena I. Ianchovichina
The objective of the Global Trade Analysis Project (GTAP), launched in 1992, was to lower the cost of entry into the world of applied computable general equilibrium (CGE or AGE) modeling using a global, economy-wide framework. The birth of the GTAP project and the subsequent publication of the GTAP book (Hertel 1997), which documented the model structure, data, and software, were timely because there was an increasing demand for quantitative analyses of trade policy issues on a global basis. Most notably, the Uruguay Round negotiations under the auspices of the General Agreement on Tariffs and Trade (GATT) were a catalyst in moving forward the GTAP database and model, as were the heated debates over the North American Free Trade Agreement (NAFTA) and subsequently the World Trade Organization’s (WTO) Doha Development Agenda. In response to this demand, the GTAP project grew from a few people in a handful of countries in 1992 to more than 8,500 people from 140 countries in 2010; the GTAP book has been widely cited; and the GTAP model and data have been actively used by a large number of public institutions around the world and in analyses on various topics published in numerous refereed journals, books, and reports. The dynamic GTAP model (GDyn) is a follow-up to the GTAP model (Hertel 1997). It is a recursively dynamic applied computable general equilibrium framework of the world economy that extends standard GTAP to include features that improve the treatment of the long-run in the model, but retains all its other features. Part II of this book documents these extensions to the GTAP model structure and data, the construction of a baseline, and the software. Part III consists of six applications of the model that highlight the versatility of the modeling framework.
3
4
Elena I. Ianchovichina
1. Motivation for the GDyn The main objectives of GDyn are to provide a better treatment of the longrun within the GTAP framework and a way of tracing the evolution of the global economy through time. A good treatment of the long-run and issues of timing is essential when analyzing the economics of some of today’s most prominent global issues, such as climate change, natural resource management, globalization, and demographic change. For a good long-run treatment, we need international capital mobility that will allow us to capture how policy shocks and other developments diversely affect incentives to invest in different regions. We also need to determine regional capital stocks, which is most satisfactorily done in a dynamic model. GDyn aims to facilitate analysis of the economic implications of climate change, economic growth, and other issues affecting the global economy in a dynamic context.
2. Data for GDyn With capital mobile between regions, the database for GDyn needs to extend beyond the standard GTAP Data Base. It needs to allow for foreign and domestic ownership of regional capital stocks, as well as international income payments and receipts. This is necessary because the assets owned by a region need no longer be the assets located in that region and the income generated by the assets in a region need no longer accrue to that region’s residents. To limit the burden of data construction for GDyn, and because data on foreign assets and liabilities are limited and inconsistent globally, we prefer a treatment of foreign assets that is parsimonious in its data requirements. This treatment is discussed in Chapters 2 and 4. New pieces of data are also needed to accommodate the new lagged adjustment, adaptive expectation theory of investment. In GDyn investors act so as to eliminate disparities in expected rates of return, not instantaneously, but progressively over time. The parameters determining the speed of convergence in rates of return are presented in Chapter 3. These parameters have either been estimated econometrically or have been informed by econometric or empirical evidence. Finally, macroeconomic and policy projections data are needed for the construction of a baseline scenario. On the macroeconomic side these data include projections of gross domestic product, gross domestic investment, population, and skilled and unskilled labor. On the policy side, these data include policies that are important elements of the baseline scenario, and their inclusion will depend primarily on the issue being examined. For
Introduction
5
example, if one is interested in free trade agreements among the Association of South-East Asian Nations (ASEAN) countries, it would be important to incorporate those agreements that have already been ratified. However, if one is interested in agreements between the EU and South Africa, then agreements between ASEAN countries may be of limited concern. Chapter 5 discusses the construction of the macroeconomic and policy projections. The data for GDyn adhere to the same principles as the GTAP Data Base – public availability at cost, upgrades coordinated with the release of the standard GTAP Data Base, and broad participation. The network of GDyn users includes those who would identify areas for improvement or extension of the database and who are encouraged to work with GTAP staff to incorporate their ideas into future database releases. The operational concept that has worked for years for the GTAP Data Base continues to apply in the case of GDyn: “If you do not like it, help fix it!”
3. Model and Software The investment theory and the treatment of financial assets and associated income flows in GDyn are discussed in Chapter 2. The main features include the treatment of time; the distinctions between physical and financial assets, and between domestic and foreign financial assets; and the treatment of capital and asset accumulation, assets and liabilities of firms and households, income from financial assets, and the investment theory of adaptive expectations. The discussion focuses on those areas in which the new treatment of the long-run required us to make changes to the standard GTAP model. All distinguishing features of GTAP, apart from those discussed in Chapter 2, remain unchanged. These include the treatment of private household behavior, international trade, and transport activity. Auxiliary variables in GTAP that facilitate the construction of alternative closures, including partial equilibrium specification, are preserved in GDyn. The GDyn model is implemented using the GEMPACK software suite (Harrison and Pearson 1998) developed at the Centre of Policy Studies and IMPACT Project, Monash University, under the direction of Kenneth Pearson. Other general equilibrium models solved using the GEMPACK software suite include the standard GTAP model and the Monash model of Australia. The software that makes it easy to run GDyn is RunDynam, which is a program created by Ken Pearson and specially tailored to the needs of GDyn. It allows users to examine the data, construct and modify experiments, produce solutions, and examine results. Users who wish to alter the underlying theory of the model will need to acquire GEMPACK
6
Elena I. Ianchovichina
and RunDynam from the Centre of Policy Studies at Monash University, Australia. Those who wish to make their own data aggregations will need to purchase the GTAP Data Base and GDyn extensions from the GTAP Center, Purdue University, United States. At the time of publication of this book, a number of applications of GDyn have been published in refereed journals, professional books, and magazines, and a half dozen are currently underway worldwide. These applications address a variety of issues, including trade policy reform, regional integration, equilibrium real exchange rate analysis, technical change, natural resource management, global climate change, and demographic change. Six of these applications were selected for inclusion in this book. They are representative of the work being undertaken currently with GDyn.
4. Short Course in Dynamic Global Economic Analysis Although its dynamic nature makes GDyn somewhat more complex than standard GTAP, the use of GDyn has spread around the world. Since 2000, when the Center for Global Trade Analysis held the first short course in Dynamic Global Economic Analysis, the model has been used by economists in universities and public research institutions in more than 20 countries on five continents. The dynamic GTAP course was offered again in 2006, 2008, and 2010, and there are plans to have this course offered at regular intervals in different parts of the world.
5. Overview of the Book This book is divided into four parts, of which this chapter is the first. Part II presents the technical aspects of the GDyn framework. In six chapters it covers data and new theoretical extensions that enable improved treatment of the long-run in GDyn, the construction of a baseline, welfare analysis in a dynamic model, and the software used to run the model. Chapter 2 presents an in-depth exposition of the investment theory and the treatment of financial assets and associated income flows. Chapter 3 discusses the techniques used to determine the magnitude of behavioral and entropy parameters used in the theory presented in Chapter 2. Chapter 4 discusses the data construction and aggregation programs. Chapter 5 documents the steps involved in building a baseline for GDyn. Chapter 6 develops a method for decomposing welfare in GDyn. Chapter 7 presents the software for running the model and analyzing model solutions.
Introduction
7
Part III of the book is a collection of six applications of GDyn. These are grouped by topic and examine a diverse set of issues: trade reform, growth and investment, climate change, natural resources, technology, and demographic change. The first of these applications, Chapter 8, is authored by Walmsley, Hertel, and Ianchovichina. This application formally explores the linkage between China’s WTO accession and investment in China in the period between 1995 and 2020. The application is similar to the one presented in Walmsley and Hertel (2001), who also used GDyn, but makes a number of enhancements, including the depiction of the duty drawback regime in China and the liberalization of trade and investment in services. Walmsley, Hertel, and Ianchovichina find that investment in China has increased substantially as a result of China’s accession. Accession doubles the extent of foreign ownership of Chinese assets relative to the no-accession baseline by 2020. Central to this increase in foreign ownership is the expected catch-up in the productivity of the services sectors driven by the opening of these sectors to foreign investment. The resulting impact on GDP is also large – 22.5% higher than the baseline by 2020. China’s welfare gain (15% by 2020) is dampened because a substantial share of additional investment comes from overseas. These estimates are far larger than those predicted by earlier studies, which ignored the reforms affecting the services sectors of China and which also abstracted from capital accumulation and international capital mobility. In Chapter 9 Hertel, Itakura, and Walmsley turn their attention to issues that are the focus of “new-age” free trade agreements (FTAs), such as rules governing foreign investment, e-commerce regulations, trade in services, harmonization of technical standards, sanitary and phyto-sanitary regulations, and the streamlining of customs procedures. They use GDyn, which is well suited to capturing the long-run effects of new-age FTAs, to assess the dynamic gains from the FTA between Japan and Singapore. They find that the impacts of this new-age FTA on bilateral trade and investment flows are significant – with customs automatization playing the most important role in driving increases in merchandise trade. The FTA also boosts rates of return in the two economies, thereby increasing both foreign and domestic investment as well as GDP. Chapters 10 and 11 address issues related to climate change and the natural resource implications of technical change in agriculture. In Chapter 10, Ianchovichina, Darwin, and Shoemaker incorporate different types of land in GDyn to analyze the impact of a slowdown in agricultural total factor productivity (TFP) growth on agriculture and forest resources. They find that a slowdown in agricultural TFP growth might lead to higher crop
8
Elena I. Ianchovichina
prices in all regions, with South East Asia facing the steepest increases. A slowdown in agricultural TFP growth might also be accompanied by increases in conversion rates of forestland to farmland as well as by a worsening of environmental or ecological damages on the remaining forestland. In Chapter 11, Golub and Hertel investigate the role of globalization and growth in determining long-run patterns of land-use change. They are able to isolate the impact on land markets of the following elements of globalization: population growth, real income growth, access to forestland, and international trade. Of the two demand-side factors, real income growth is more important than population growth. The potential for accessing new forestland plays a small role in dampening the growth in global land rent, whereas international trade plays a very substantial role in mediating between the land-abundant, slower growing economies of the Americas and Australia/New Zealand, and the land-scarce, rapidly growing economies of Asia. When combined, the forces of globalization are expected to play a large role in determining the pattern of land-use change. Chapter 12, authored by Itakura, Hertel, and Reimer, incorporates into GDyn existing econometric evidence of the strong correlations that exist between firm productivity, on the one hand, and investment and trade on the other, to study the economic effects of a recently proposed East Asian FTA. They find that although conventional applied computable general equilibrium (AGE) modeling effects predominate and are reinforced by the productivity effects, in some cases, the latter actually reverse the changes predicted by the conventional effects. In the final application, Chapter 13, Tyers and Shi introduce a global demographic submodel into a version of GDyn in which regional households differ by gender and age group. Their goal is to study the global implications of demographic change. They find that increased longevity of the global population slows down growth in real per capita incomes, lowers saving rates, and alters the distributions of global economic activity in favor of those regions with high aged labor force participation. Part IV of the book offers an evaluation of the GDyn framework. In Chapter 14, Golub and McDougall evaluate the evolution of foreign assets and liabilities in GDyn by comparing the model-determined and actual trends in foreign assets, foreign liabilities, and net foreign assets – all as a share of wealth. Their choice of the first two indicators is driven by their interest in regional wealth allocation and the composition of capital in GDyn. The third indicator is of more general interest and is often used in discussions in the financial press and among policy makers in reference
Introduction
9
to the sustainability of net foreign liabilities. How large can gross and net foreign assets and liabilities get in GDyn simulations? Have such changes been observed in past data? These are the types of questions Golub and McDougall ask in Chapter 14. For this comparison they use a country portfolio database constructed by Kraay et al. (2000). The database covers 68 countries, including all industrial countries and many developing countries, for the years ranging from 1966 to 1997. Golub and McDougall find that, unlike in real data, net foreign positions in GDyn grow without bounds in the very long-run. As economies with high saving rates in the initial year (e.g., China) grow, there is a glut of savings in GDyn, and rates of return to capital fall without bound in the very long-run. The reason for this is the assumption of fixed propensity to save in each region in GDyn. To ensure that, as in reality, gross foreign assets and liabilities do not diverge without bound in the very long-run, Golub and McDougall endogenize the saving rate in GDyn and make it a function of the share of wealth in income. The new theoretical structure supports balanced growth scenarios, stabilizes global rates of return to capital, and prevents net external assets or liabilities from growing implausibly large. Chapter 15 by Ianchovichina and Walmsley provides an overall evaluation of the effort to date, as well as some observations about the future course of global economic analysis with GDyn.
6. Reader’s Guide For those who wish to master the material presented in the book, the most efficient approach is to read the book chronologically – beginning with the model structure, reading about the parameters and data, accessing the software from the Web site, mastering the construction of the baseline, and only then turning their attention to the applications and the evaluation of the model. However, many readers may be interested in a particular topic. Those interested in climate change, natural resources, and demographics might want to start with Chapters 10, 11, and 13, respectively. Having read these chapters and understanding the essence of the changes introduced to address these topics, you might want to study the features of the main dynamic model. For this you need to backtrack to Chapter 2 and then go to Chapters 3–7. Those interested in trade policy and investment might want to look at Chapters 8, 9, and 12 first, before going back to the details of the dynamic model presented in Chapters 2–7.
10
Elena I. Ianchovichina
References Harrison, W. J. and K. R. Pearson. 1998. An Introduction to GEMPACK. GEMPACK Document No. GPD-1 (4th ed.). Melbourne, Australia: Centre of Policy Studies and Impact Project, Monash University. Hertel, T. W. (ed.). 1997. Global Trade Analysis Modeling and Applications. Cambridge: Cambridge University Press. Kraay, A., N. Loayza, L. Serven, and J. Ventura. 2000, July. Country Portfolios. National Bureau of Economic Research Working Paper Series No. 7795. Washington, DC: NBER. Walmsley, T. L. and T. W. Hertel. 2001. “China’s Accession to the WTO: Timing is Everything.” World Economy 24(8), 1019–49.
PART II
STRUCTURE OF THE DYNAMIC GTAP FRAMEWORK
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
TWO
Theoretical Structure of Dynamic GTAP Elena I. Ianchovichina and Robert A. McDougall
1. Introduction GDyn is a recursively dynamic AGE model of the world economy. It extends the standard GTAP model (Hertel 1997) to include international capital mobility, capital accumulation, and an adaptive expectations theory of investment. This chapter discusses the rationale behind the design decisions affecting GDyn and presents its technical features in detail. The main objective of GDyn is to provide a better treatment of the long-run within the GTAP framework. In standard GTAP, capital can move between industries within a region, but not between regions. This inability of capital to move between regions impedes the analysis of policy shocks and other developments that affect incentives to invest in different regions. For a good long-run treatment, then, we need international capital mobility. With capital mobile between regions, we need to expand the national accounts to allow for international income payments. Policies that attract capital to a region may have a strong impact on the gross domestic product, but if the investment is funded from abroad, the impact on the gross national product and national income may be much weaker. Therefore, to avoid creating spurious links between investment and welfare, we need to distinguish between asset ownership and asset location: The assets owned by a region need no longer be the assets located in that region, and the income generated by the assets in a region need no longer accrue to that region’s residents. To distinguish between asset location and ownership, we introduce a rudimentary representation of financial assets. Regions now accumulate not only physical capital stocks but also claims to the ownership of physical capital. These ownership claims are financial assets of some kind. Thus 13
14
Elena I. Ianchovichina and Robert A. McDougall
international income receipts and payments emerge as part of the system of accounting for financial assets. With capital internationally mobile, we need to determine regional capital stocks. This is most satisfactorily done in a dynamic model. First, tracing out the investment and capital stock time paths is the best way to assure ourselves that the end-of-simulation capital stocks are reasonable. Second, the immediate impact of the earlier period investments required to achieve the end-of-simulation stocks in regional economies is itself of some interest. Accordingly, we make the model dynamic and incorporate the stock flow or intrinsic dynamics of investment and capital accumulation. Likewise, we incorporate the intrinsic dynamics of saving and wealth accumulation. The key features of this extension are endogenous regional capital stocks, international financial assets and liabilities, international investment and income flows, and intrinsic dynamics of physical and financial asset stocks. While introducing these new features we seek to preserve the strengths of the standard GTAP model, including the abilities to work with empirical rather than highly stylized databases and to solve the model in a reasonable time on reasonable computing platforms, while preserving a detailed regional and sectoral disaggregation, a money metric of utility, and an associated decomposition. The GDyn model is suitable for medium- and long-run policy analysis, in which the comparative statics of the end-of-simulation solution is supplemented with time paths leading to the solutions. It has enough dynamics and a sufficient treatment of financial assets to support this analysis, but not enough to support short-run macroeconomic dynamics or financial or monetary economics. A salient technical feature of GDyn is the treatment of time. Many dynamic models treat time as an index, so that each variable in the model has a time index. In GDyn, time itself is a variable, subject to exogenous change along with the usual policy, technology, and demographic variables. Section 2 elucidates the mechanics and motivation of this treatment, and Section 3 applies it to the capital accumulation equation. This lays the groundwork for the discussion in Section 4 of wealth accumulation, financial asset determination, and foreign income flows. Section 5 describes the investment theory, incorporating lagged adjustment of capital stocks and adaptive expectations for the rate of return. Section 6 discusses the properties of the complete model and the existence of and convergence toward a long-run equilibrium. The chapter concludes in Section 7 with a summary of the strengths and limitations of the model.
Theoretical Structure of Dynamic GTAP
15
We encourage the reader to consult the glossary for a list of variables and coefficients and the TABLO code for the model equations.1 Doing so will help the reader follow the notation and relate the chapter to the solution program source code. We give each equation appearing in the model in two or three forms: the levels equation (if appropriate) in mathematical notation, the differential (change) equation in mathematical notation, and the differential equation as coded in the model. The coded equations are close but not literal transcriptions from the source code.
2. Time As noted in Section 1, a key technical feature of GDyn is the treatment of time not as a discrete index but as a continuous variable. However, because the continuous-time treatment may be less familiar to many readers, we first provide an overview of the more familiar discrete-time approach and then contrast the two. Within the very large class of dynamic economic models, we confine our discussion to recursively solvable AGE models. In discussing solution methods, we assume the use of the GEMPACK suite of economic-modeling software. We use a simplified wealth accumulation equation to illustrate and contrast the discrete- and continuous-time approaches. In a closed economy with a single capital good, which constitutes the sole economic asset and hence the sole vehicle for saving, real wealth may be defined as the size K of the capital stock, and the evolution of the capital stock through time is given by an integral equation,2 T K = K0 + I (τ)dτ, (2.1) T0
where K 0 denotes the capital stock at some base time T0 , and I, net investment.
2.1 The Discrete-Time Approach Within a recursively solvable discrete-time framework, there is typically a concept of a time period. A given database refers to a given time period; 1 2
The model code is available on the Web site at https://www.gtap.agecon.purdue.edu/ models/Dynamic/applications.asp. Note that equation (2.1) combines features that might be separately represented by the capital and wealth accumulation equations in a more complex setting.
16
Elena I. Ianchovichina and Robert A. McDougall
a simulation takes the database to the next time period, with simulation results representing changes between the initial period and the next. Within such a framework, the database might include a representation of the economy in the current period, together with some extra data pertaining to the next period. The representation of the economy might contain values as of the start of the period, or as of the midpoint of the period, or average values over the period. The extra data might be only the period length or might include, for example, values of stocks at the start of the next period. Suppose that the database contains a representation of the economy at the start of the period, together with the period length. We have from equation (2.1), by the mean value theorem and assuming a continuous time path for investment I, K = K 0 + (T − T0 )I (Tm ),
(2.2)
for some Tm between T0 and T , where we now interpret the base time T0 as the start of the period represented by the initial database. For small T − T0 , we have I(Tm ) ≈ I(T0 ), so K ≈ K 0 + I 0L ,
(2.3)
where L denotes the interval length T − T0 . By differentiating equation (2.3), we obtain the percentage change in the capital stock k within the simulation: k ≈ 100
I 0L . K0
(2.4)
We may calculate the right-hand side as a formula outside the model and apply it as a shock to k, or to avoid performing a separate calculation before the simulation, we may include a capital accumulation equation within the solution program, writing k ≈ 100
I 0L h, K0
(2.5)
where h is an artificial variable, sometimes called a homotopy variable, which is always exogenous and always receives a shock of 1 in a dynamic simulation. Note that the coefficients I0 and K0 refer to the start-of-simulation database and are not updated within the simulation. Note that equation (2.1) is true only approximately, not exactly. This is not because of a linearization error arising as a result of the passage from the levels to the change equation: Indeed, there is no such error, because the levels equation (2.3) is itself linear. The change equation instead inherits an
Theoretical Structure of Dynamic GTAP
17
error from the levels equation because the levels equation is itself inexact. The error cannot be reduced by refinements in the solution procedure, such as using smaller step sizes, because it is inherent in the levels equation. The only way to reduce the error is by revising the simulation strategy and using more simulations with shorter time intervals. Once the time interval is set, we have an irreducible inaccuracy in the accumulation equation. Readers familiar with the discrete-time approach may object that their own favorite discrete-time model does not suffer from this particular inaccuracy. In general, it appears that it is possible to change the form of the inaccuracy, but not to eliminate it. Suppose, for example, that the database represents the average state of the economy through the period, together with start-of-period and end-of-period stocks. Then we can derive exact equations for the start-of-period and end-of-period stocks for the next period, given initial-period and next-period investment. To calculate the next-period average capital stock value, however, we need to know how investment is distributed in time through the next period, but we cannot possibly know this. Therefore the determination of the through-period average capital stock is necessarily approximate. For sufficiently small time steps, this inaccuracy does not matter much, but for larger time steps, we must replace equation (2.5) by some other, more complex equation that offers a better approximation over longer periods. For example, in our closed economy we may equate investment with saving. Then, we have I = S/, where S denotes nominal net saving, and is the price of investment goods. If S AP is the average propensity to save, we then have S = S AP Y and I = S AP Y/, where Y denotes nominal income. Writing Y as the product of real income YR and some price index PY , investment is given by I =
S AP P Y YR .
Substituting equation (2.6) into equation (2.1), we have T S AP (τ)P Y (τ)YR (τ) dτ. K = K0 + (τ) T0
(2.6)
(2.7)
Now it is possible to solve equation (2.7) in terms of initial and final values of the variables under the integral, only with the aid of various supplementary assumptions. For example, one might assume that real income YR maintains some constant growth rate between one period and the next; the average propensity to save, S AP , maintains some constant time rate of change; and the prices PY and jump immediately to their final values. With prices
18
Elena I. Ianchovichina and Robert A. McDougall
being liable to overshooting, we might prefer this assumption to a steadygrowth assumption. The resulting equation will obviously be quite different from and far more complex than equation (2.5). Less obviously, it will, like that equation, include period-length-dependent parameters. Thus by making assumptions about time paths of variables between adjacent periods, we might derive a longer-run wealth accumulation equation. The details of the assumptions are not important. The point is that, to implement the discrete-time approach for longer time intervals, we would need to make strong assumptions about the time paths of various economic variables between time periods (e.g., that the variables involved are typically endogenous to the system and that the assumptions must be applied not at run time but in developing the accumulation equation). The method we have outlined is just one of many ways to implement a discrete-time treatment of capital accumulation, but it serves to illustrate some common features of this approach. r The database represents the economy in some period of time, possibly
but not necessarily at a single time point within the period.
r The capital accumulation equation includes coefficients derived not
r
r r
r
from the current but from the start-of-simulation database. It may also include some current coefficients, although in our illustrative example it does not. The capital accumulation equation includes parameters that depend on the size of the time step for the simulation. In our illustration, the time step size itself is L. Given the size of the time step, there is some inaccuracy built into each experiment that cannot be removed by refining the solution procedure. Major changes in the step size are liable to require revision not only of the parameters but also of the form of the capital accumulation equation. For longer time intervals, the accumulation equations are liable to embody strong assumptions about time paths of endogenous variables.
In conclusion, the discrete-time treatment of capital accumulation is viable, but it is apt to suffer from some problems, including inaccuracy, special assumptions about investment paths, and inflexibility in the size of the time step. Fortunately, there is an alternative. Capital accumulation lends itself naturally to a continuous-time approach, described next.
Theoretical Structure of Dynamic GTAP
19
2.2 The Continuous-Time Approach Returning to equation (2.1), we now reinterpret the database as representing the economy at some point in time. Both stock data and flow data refer to the same time point. In addition we treat T not as a discrete index but as a variable within the model. Totally differentiating equation (2.1), we then obtain the equation k = 100
I t, K
(2.8)
where k represents percentage change in the capital stock, and t represents change in time. This is very similar in form to the discrete-time equation (2.5). There are, however, two differences: The time variable t replaces the homotopy variable h, and the equation uses the current rather than initial values of investment I and the capital stock K. These differences have major consequences. First, the new equation, being the linearized form of equation (2.1), involves a linearization error, but not an irreducible error. Thus the error in the calculation of the capital stock may be made as small as desired by refining the solution procedure; for example, by increasing the number of subintervals. Second, because there is no irreducible error, the equation is equally valid for any time interval. Third, because the length of the time interval is given by a variable (t) rather than by a parameter (L), the time-interval length is determined at run time rather than in the database. In contrast to the discrete-time approach, our approach does the following: r uses the database to represent the economy at a point in time r in a multistep solution, uses no coefficients derived from the start-ofr r r r
simulation database, but only current values involves no parameters that depend on the length of the time interval involves no irreducible inaccuracies in dynamic relations uses the same accumulation equation for any time interval relies on no prior assumptions about the time paths of endogenous variables
The notion of time as a variable can be explained in terms of the sources of change in an economy. An economy may change not only in response to changes in external circumstances such as technology, policy, or endowments but also through the intrinsic dynamics of its stock-flow relationships. In the presence of nonzero net investment or saving, the passage of time
20
Elena I. Ianchovichina and Robert A. McDougall
leads to change in the stock of capital goods or of wealth. Furthermore, with adaptive expectations or lagged adjustment, the passage of time leads to the revision of expectations or the adjustment of target variables toward equilibrium. Such changes, arising not from changes in external circumstances but autonomously through the passage of time, can be captured in time terms using the time variable t in the equation system. The shock to t defines the change in time through the simulation. Shocks to other exogenous variables represent accompanying changes in external circumstances.
3. Capital Accumulation We now begin applying the time treatment described in Section 2 to the GDyn equation system. We start with the capital accumulation equation, deriving the capital stock variable used both in the financial assets theory presented in Section 4 and the investment theory discussed in Section 5. We begin with the integral equation for the capital stock, QK = QK 0 +
TIME
QCGDSNET × dτ,
(2.9)
TIME 0
where QK(r) represents the capital stock in region r, QK_0(r) is the capital stock at some base time TIME_0, TIME is current time, and QCGDSNET(r) is net investment. Totally differentiating equation (2.9), we obtain QK(r) ×
qk(r) = QCGDSNET(r) × time, 100
(2.10)
where qk(r) represents percentage change in the capital stock in region r, and time is change in time. Multiplying both sides by 100 times the price of capital goods, we obtain VK(r) × qk(r) = 100 × NETINV(r) × time,
(2.11)
where VK(r) denotes the money value of the capital stock in region r, and NETINV(r) represents the money value of net investment. In a static simulation, with time equal to zero, we see from equation (2.11) that the percentage change in the capital stock qk is also zero. Sometimes, however, we wish to impose some nonzero change in capital stocks. To that end we introduce into the accumulation equation a region-generic shift factor SQKWORLD and a region-specific factor SQK(r). Incorporating
Theoretical Structure of Dynamic GTAP
21
those factors, we obtain the final version of the levels equation, QK(r) = SQKWORLD × SQK(r) TIME QCGDSNET(r)dτ . × QK 0(r) +
(2.12)
TIME 0
The differential equation is then given by the following equation, VK(r) × qk(r) = VK(r)[sqkworld + sqk(r)] + 100 × NETINV(r) × time,
(2.13)
represented as follows in the model code: Equation KBEGINNING #capital accumulation # (all,r,REG) VK(r)∗ qk(r) = VK(r)∗ [sqkworld + sqk(r)] + 100∗ NETINV(r)∗ time.
4. Financial Assets and Associated Income Flows As discussed in the introduction, to model international capital mobility we need to distinguish between asset location and ownership. For this purpose, we introduce financial assets. In GDyn, regional households do not own physical capital; only firms do. Households own not physical capital but financial assets, which represent indirect claims on physical capital. In this section, we show how the model determines agents’ financial assets and liabilities, as well as the associated income receipts and payments. We begin with a discussion of the treatment’s general features in subsection 4.1 and follow with a note on notation in subsection 4.2. Stock-flow accumulation relations determine two key financial asset variables presented in subsection 4.3. With those as constraints, we use an atheoretic mechanism to determine the composition of firms’ liabilities and regional households’ assets in subsection 4.4. We complete the module with equations for the assets and liabilities of the global financial intermediary in subsection 4.5 and for income flows associated with financial assets in subsection 4.6.
4.1 General Features In addition to the prime motivation to take account of international capital mobility, several other requirements have shaped the treatment of financial assets in GDyn. For reasons discussed in subsection 5.1, we do not enforce rate-of-return equilibration over the short-run. This means that we need to represent gross ownership positions. It is not enough, for example, to know
22
Elena I. Ianchovichina and Robert A. McDougall
a region’s net foreign assets. We must know both its gross foreign assets and its gross foreign liabilities, because their rates of return may differ. To limit the burden of data construction for the extended model, and because data on foreign assets and liabilities are limited and inconsistent, we prefer a treatment of foreign assets that is parsimonious in its data requirements. We also want the treatment to accommodate the salient empirical regularity of local specialization – that countries do not hold globally balanced asset portfolios, but specialize strongly in holding local assets. With the new treatment, we do not aim to give a full or accurate representation of financial variables. The financial assets in GDyn are there not to provide a good representation of financial assets in the real world, but to let us represent international capital mobility without creating leaks in the foreign accounts. Our treatment of financial assets accordingly is minimalist and highly stylized. Influenced by these considerations, we determine some broad features of the financial assets module. First, we elect not to adopt a full financetheoretic treatment of financial assets, but to take an ad hoc or heuristic approach. The attraction of a finance-theoretic approach is that it would let us account in a principled way for investors’ holding assets with different rates of return, rather than only the highest yielding asset. It would recognize that investors are concerned not only with return but also with risk. It would relate their decisions on risk-return tradeoffs and their consumption and saving behavior to the same set of underlying preferences, preserving thereby the rigor of the welfare analysis. Yet introducing a finance-theoretic treatment would add greatly to the complexity of the model and create perhaps as many difficulties as it would solve. There are a number of paradoxes in international financial behavior, empirical regularities that are difficult to account for theoretically. Most relevant here, it is difficult to account for observed disparities between countries’ rates of return, which far exceed those predicted with simple financetheoretic models, plausible behavioral parameter settings, and observed risk levels. This does not rule out the finance-theoretic approach, but it does make the cost-benefit balance less attractive. On balance then, we elect not to implement such a treatment in this version of GDyn, while acknowledging its attractiveness as an area for future research. After this basic decision, there are several other design decisions to make. First, we must decide which physical assets should back financial assets – in other words, to which assets should financial assets represent indirect claims. To allow for international capital mobility, we must include physical capital in this set. We may also include primary factors, also referred to as “endowment commodities” in GTAP jargon, other than labor. In the GTAP
Theoretical Structure of Dynamic GTAP
23
version 4 Data Base (McDougall, Elbehri, and Truong, 1998), there are two of these endowment commodities: (1) agricultural land and (2) other natural resources such as mineral deposits, fisheries, and forests. Although it would be more logical to let all these commodities back financial assets, it is easier to let only physical capital back financial assets. In this version of the model, we take the easier approach. Accordingly, in GDyn, firms own physical capital, but rent land and natural resources. Conversely, regional households own land and natural resources, which they lease to firms, and financial assets, which may be construed as indirect claims on physical capital. The next question is which classes of financial assets to represent in the model. In the real world there are three broad classes of financial assets – money, debt, and equity – which are divided in turn into many subclasses. Recognizing more asset classes would potentially improve the realism of the model. However, for reasons discussed earlier, realism in the representation of financial assets is not a priority for this model. In light of this, and consistent with our stance that the role of the financial asset module is to support international capital mobility rather than to depict the financial sector realistically, we include in the model just one asset class, equity. Accordingly, in GDyn, firms have no liabilities and only one asset, physical capital. By the fundamental balance sheet identity (assets = liabilities + proprietorship), shareholder equity in the firm is equal in value to the physical capital that the firm owns. Next we ask which agents can hold equity in firms. The simplest design would be to let all regional households hold equity in firms in all regions. This, however, would require bilateral data on foreign assets and liabilities. Unfortunately, the available data, especially the foreign direct investment data, are insufficient and internally inconsistent. To minimize the data requirements, we adopt instead the fiction of a global trust that serves as a financial intermediary for all foreign investment. In GDyn, regional households do not hold equity directly in foreign firms, but only in local firms and the global trust. In turn the global trust holds equity in firms in all regions. The trust has no liabilities and no assets other than its equity in regional firms. Therefore, by the balance sheet identity, total equity in the trust is equal in value to total equity held by the trust. A minor defect of this treatment is that it leads the model to misreport foreign asset holdings. We identify each region’s equity in the global trust with its foreign assets, when in fact some portion of it represents indirect ownership of local assets. This misreporting is trivial for small regions, but more considerable for large regions such as the United States. In terms of analysis this defect has minimal effect on the results of a simulation because regional investment is driven by rates of return and not affected
24
Elena I. Ianchovichina and Robert A. McDougall Region r
WQHTRUST(r)
WQHHLD(r) Global Trust WQHFIRM(r)
冱r WQHTRUST(r) = 冱r WQTFIRM(r)
WQ_FIRM(r)
WQTFIRM(r)
Figure 2.1. Wealth linkages.
by the treatment of ownership. However, the treatment will result in rental income from capital ownership being biased toward the rental obtained in the home country; the extent of this bias will depend on the share of the home region in the global trust. Figure 2.1 summarizes the financial asset framework. Firms in each region r have a value WQ _FIRM(r), of which the local regional household owns WQHFIRM(r) and the global trust WQTFIRM(r). In turn the global trust is owned by the regional households, each region r owning equity WQHTRUST(r). The total financial wealth of the regional household comprises equity WQHFIRM(r) in local firms and equity WQHTRUST(r) in the global trust. We discuss these relations further in subsections 4.3, 4.4, and 4.5. The concept of income from and investment in physical and financial assets remains to be discussed. We count as income the earnings of the asset, but not capital gains or losses arising from asset price changes. For physical capital, we also exclude physical depreciation from the definition of income, just as in the standard GTAP model. For equity in firms or in the global trust, we count as investment the money value of net change in the quantity of the entity’s assets, but exclude capital gains.
Theoretical Structure of Dynamic GTAP
25
This treatment has two merits. First, it imposes consistency between income and investment in financial assets: Both exclude capital gains, so saving, calculated as total investment in financial assets, is consistent with income. Second, it supports a simple decomposition of change in proprietorship. Consider an entity that has no liabilities, but owns several assets. Let WAi denote the value of assets of type i, and W = i WAi is the total asset value. Then percentage change w in total asset value is given by the equation W = i WAi (p Ai + q Ai ), where p Ai denotes percentage change in the price of asset i, and q Ai denotes percentage change in the quantity. We can use this equation to decompose this change in total asset value into two components: the money value of net change in the quantity of the entity’s assets, (1/100) i WAi q Ai , and the money value of change in the prices of the quantity’s assets, (1/100) i WAi p Ai . Now by the balance sheet identity, total proprietorship in the firm is equal to total asset value W, so w = p Q + q Q , where pQ and qQ denote percentage change in the price and volume, respectively, of the firm’s stock. We can decompose this into an investment component, (1/100)Wq Q , and a our conventional definition capital gain component, (1/100)WpQ . Then, by of investment, Wq Q = i WAi q Ai , so Wp Q = i WAi p Ai (i.e., the price of equity in the firm is proportional to an index of prices of the firm’s assets). Thus, the price and quantity components of change in total proprietorship equate to the corresponding components of change in total assets. Another way to look at this is to imagine that firms and the trust fully distribute their net earnings as dividends to shareholders, and they fund their net asset purchases entirely through new stock issues. Under this supposition, the value of dividends coincides with the GDyn definition of income, and the value of stock issues coincides with the GDyn definition of financial investment.
4.2 Notation We use a systematic notational convention to present this accounting framework. Percentage change variables are written in lowercase, whereas uppercase variables are data coefficients, parameters, levels variables, or ordinary change variables. In general, the first character of a variable or a coefficient shows its type: W (wealth) for asset values, and Y for income flow. The second character identifies the asset type: In the current version of the model, this is always Q for eQuity. The third character indicates the sector that owns the asset or receives the income it generates, whereas the fourth character identifies the sector that owes the asset or pays the associated income. For
26
Elena I. Ianchovichina and Robert A. McDougall
example, F designates investment in regional firms, T denotes investment in the global trust, and H stands for investment by the regional household. Thus, a name beginning with WQHF refers to the wealth in equity owned by the regional household and invested in domestic firms, whereas a name beginning with YQHF refers to the income from equity paid to the regional household by the domestic firms. An underscore is used in cases where the distinction pertaining to a particular character is not in point. The underscore is left out if it is located at the end of the name.
4.3 Asset Accumulation The financial assets module revolves around two key variables: the ownership value of firms in region r and the equity holdings of the household in region r. Both these equations are given, directly or indirectly, by accumulation relations. In GDyn, firms buy intermediate inputs, hire labor, and rent land, but own fixed capital. They have no debt. In accounting terms, they have no liabilities and no assets, except fixed capital. Conversely, only firms own fixed capital. Therefore the ownership value WQ_FIRM(r) of firms in region r is equal to the value of their fixed capital, which is the value of all local fixed capital, which in turn is equal to the product of the corresponding price and quantity: WQ FIRM(r) = VK (r) = PCGDS(r) × Q K (r),
(2.14)
where PCGDS(r) denotes the price of capital goods in region r. Differentiating, we obtain wq f(r) = pcgds(r) + qk(r),
(2.15)
where wq_f(r) denotes percentage change in WQ_FIRM(r), and pcgds(r), the percentage change in PCGDS(r). In the model, we write Equation REGEQYLCL #change in VK(r) [qk]# (all,r,REG) wq_f(r) = pcgds(r) + qk(r) + swq_f(r); where swq_f(r) is a region-specific shift factor.3 Thus the total equity value of each region’s firms is given indirectly by the capital accumulation equation (2.13). 3
The variable swq_f (r) is a shift factor used for modeling purposes. It is exogenous and equal to zero in the model.
Theoretical Structure of Dynamic GTAP
27
For future use we note that, by the conventions discussed in subsection 4.1, the price PQ_FIRM(r) of equity in firms in region r is proportional to the price of capital goods in region r, pq f(r) = pcgds(r),
(2.16)
where pq_f denotes the percentage change in PQ_FIRM. As with capital stocks and investment, we use the variable time to capture the intrinsic dynamics of regional wealth and savings. For the regional household’s ownership of domestic assets, we have the accumulation equation: TIME QQHFIRM(r)dτ, (2.17) WQHFIRM(r) = PQ FIRM(r) TIME 0
where PQ_FIRM(r) is the price of stocks in local firms in region r, and QQHFIRM(r) is the number of stocks purchased by the regional household. Similarly, for the regional household’s equity in the global trust, we have TIME QQHTRUST(r)dτ, (2.18) WQHTRUST(r) = PQTRUST TIME 0
where PQTRUST is the price of equity in the global trust, and QQHTRUST(r) is the volume of equity purchases by the regional household. Then the total wealth of the regional household is given by equation (2.19): TIME QQHFIRM(r)dτ WQHHLD(r) = PCGDS(r) TIME 0
+ PQTRUST
TIME
QQHTRUST(r)dτ. (2.19) TIME 0
Differentiating, and substituting for pq_f from equation (2.16), we obtain WQHHLD(r) × wqh(r) = WQHFIRM(r) × pcgds(r) + WQHTRUST(r) × pqtrust + 100 × (VQHFIRM(r) + VQHTRUST(r)) × time,
(2.20)
where pqtrust denotes the percentage change in PQTRUST, and VQHFIRM(r) denotes the value of new investment by the regional household in domestic firms in region r, VQHFIRM(r) = PCGDS(r) × QQHFIRM(r).
(2.21)
28
Elena I. Ianchovichina and Robert A. McDougall
VQHTRUST(r), the value of new investment by the regional household in the global trust, is as follows: VQHTRUST(r) = PQTRUST(r) × QQHTRUST(r).
(2.22)
Total investment by the regional household in domestic and foreign equity is equal to saving by the regional household; that is, VQHFIRM(r) + VQHTRUST(r) = SAVE(r), where SAVE(r) denotes savings in region r. So equation (2.20) simplifies to WQHHLD(r) × wqh(r) = WQHFIRM(r) × pcgds(r) + WQHTRUST(r) × pqtrust + 100 × SAVE(r) × time.
(2.23)
In the model code, we write as follows: Equation REGWLTH#change in wealth of the household [wqh(r)]# (all,r,REG) WQHHLD(r)∗ wqh(r) = WQHFIRM(r)∗ pcgds(r) + WQHTRUST(r) ∗ pqtrust + 100.0∗ SAVE(r)∗ time + WQHHLD(r)∗ swqh(r); where swqh(r) is a region-specific shift factor in the wealth of the region.
4.4 Assets and Liabilities of Firms and Households In subsection 4.3, we determined the value WQ_FIRM of equity in firms in each region. As shown in Figure 2.1, this equity has two components: equity belonging to the local regional household, WQHFIRM(r), and that belonging to the global trust, WQTFIRM(r): WQ FIRM(r) = WQHFIRM(r) + WQTFIRM(r).
(2.24)
Differentiating, we obtain WQ FIRM(r) × wq f(r) = WQHFIRM(r) × wghf(r) +WQTFIRM(r) × wqtf(r),
(2.25)
where wqhf(r) and wqtf(r) denote percentage changes in WQHFIRM(r) and WQTFIRM(r), respectively.
Theoretical Structure of Dynamic GTAP
29
Equation (2.25) appears in the model as Equation EQYHOLDFNDLCL #total value of firms in region r# (all,r,REG) WQ_FIRM(r)∗ wq_f(r) = WQHFIRM(r)∗ wqhf(r) + WQTFIRM(r) ∗ wqtf(r). Also in subsection 4.3, we determined the wealth in equity of the regional households, WQHHLD. As shown in Figure 2.1, this also has two components – equity in domestic regional firms, WQHFIRM, and in the global trust, WQHTRUST: WQHHLD(r) = WQHFIRM(r) + WQHTRUST(r).
(2.26)
Differentiating, we obtain WQHHLD(r) · wqh(r) = WQHFIRM(r) · wqhf(r) +WQHTRUST(r) × wqht(r),
(2.27)
where wqhf(r), and wqht(r) denote percentage changes in WQHFIRM(r) and WQHTRUST(r), respectively. This equation appears in the model as Equation EQYHOLDWLTH #total wealth of the household# (all,r,REG) WQHHLD(r)∗ wqh(r) = WQHFIRM(r)∗ wqhf(r) + WQHTRUST(r) ∗ wqht(r). Thus far, for each region r we have two accounting identities, equations (2.24) and (2.26), and three variables to determine: WQHFIRM, WQTFIRM, and WQHTRUST. Equivalently, for each region the identities suffice to determine the net value of foreign assets, WQHTRUST(r) − WQTFIRM(r) = WQHHLD(r) − WQ FIRM(r) (2.28) but not gross foreign assets and liabilities: WQHTRUST(r) and WQTFIRM(r). Obviously many different gross foreign asset positions are consistent with the net position. In this model, we do not make use of portfolio allocation theory, so we have no theory explaining the gross ownership position. Over the long-run, rates of return on capital are equalized across regions. With no portfolio allocation theory, investors care only about returns, so with returns equalized the allocation of assets is arbitrary. Over the short-run, we allow interregional differences in rates of return (subsection 5.1). We need investors to hold several assets (because net foreign ownership positions must be nonzero), but we have no theory explaining why investors would hold any assets other
30
Elena I. Ianchovichina and Robert A. McDougall
than the highest yielding one. Accordingly, we can determine portfolio allocation over the short- or long-run only by applying a heuristic rule. In selecting a portfolio rule, we have some constraints to guide us. First and most obviously, the three variables WQHFIRM(r), WQHTRUST(r), and WQTFIRM(r) must satisfy the two identities (2.24) and (2.26). Furthermore, we want to obtain positive values for those three variables. This is possible, provided that WQHHLD(r) and WQ_FIRM(r) are positive. Although it is possible in the real world to short-sell stocks, we do not observe large, long-lasting negative equity holdings. If we nevertheless allowed negative holdings in the model, they would be liable to generate strange welfare results. If, for example, we allowed the global trust to hold negative equity in Taiwan, then the income of the trust and, consequently, the foreign asset income of each region would vary not directly, but inversely with Taiwanese capital rentals. Given the real-world absence of stable negative equity holdings, this inverse relationship would be unrealistic. Finally, we want the allocation rule to preserve as nearly as possible the initial allocation of each region’s wealth between domestic and foreign assets. One of the objectives of the asset treatment is to allow the model to respect the empirical regularity that regions tend to specialize in their own domestic assets. If the initial database respects this, we want updated databases to respect it too. One possible approach is to assume that each region allocated its wealth between domestic and foreign assets in fixed proportions. This assumption is simple and in some ways appealing, but it has one defect: It makes it too easy for foreign liabilities to become negative. For example, a negative shock to productivity in Taiwan might cause the value of capital located in Taiwan to fall more rapidly than the value of equity owned by the Taiwanese. With the fixed shares approach, the value of domestic equity owned by the Taiwanese might easily come to exceed the value of the Taiwanese capital stock, so that the value of foreign ownership of Taiwanese industry would become negative. As discussed earlier, we wish to avoid such outcomes. If, conversely, we assumed that the composition of the source of funds were fixed in each region, so that foreign and domestic equity in local capital varied in fixed proportion, we would be assured that foreign ownership of local capital would not turn negative. However, growth in the local capital stock might easily lead to negative local ownership of foreign assets. To avoid negative values in both gross foreign assets and gross foreign liabilities, we need a more sophisticated approach. We find this in entropy theory. In particular, cross-entropy minimization gives us a way of dividing a strictly positive total into strictly positive components, subject to various constraints, while staying as close as possible to the initial shares. A full
Theoretical Structure of Dynamic GTAP
31
exposition of the relevant concepts would take us too far afield here. For example, Kapur and Kesavan (1992) present a modern treatment emphasizing aspects of interest to economists. Cross entropy is an indicator of the degree of divergence between two partitions Si , i = 1, . . . , n of a total value. Writing Si (0) for the initial shares and Si (1) for the final shares, the cross entropy is
S i (1) log
i
S i (1) . S i (0)
(2.29)
This takes a minimum when, for all i, Si (1) = Si (0); that is, when the final shares are equal to the initial shares (Kapur and Kesavan 1992). The advantages of the cross-entropy approach become apparent when we impose constraints on the final shares. For a wide variety of constraints, the constrained optimization problem leads to a simple and transparent set of first-order conditions. In addition, with strictly positive initial shares, we are guaranteed, constraints permitting, strictly positive final shares. We are concerned with two sets of shares: the shares of domestic and foreign equity in domestic wealth and the shares of domestic and foreign funds in ownership of local capital. With each of these we associate a crossentropy measure. For shares in domestic wealth in region r, the cross entropy is CEHHLD(r) = WQHFIRMSH(r) · log
WQHFIRMSH(r) WQHFIRMSH 0(r)
+WQHTRUSTSH(r) · log
WQHTRUSTSH(r) , WQHTRUSTSH 0(r)
(2.30)
where WQHFIRMSH(r) denotes the current share of local firms; WQHTRUSTSH(r), the current share of the global trust in the equity portfolio of the household in region r; and WQHFIRMSH_0(r) and WQHTRUSTSH_0(r) denote the initial levels of those shares. By definition, we have WQHFIRM(r) =
WQHFIRM(r) , WQHHLD(r)
WQHFIRMSH 0(r) =
WQHFIRM 0(r) , WQHHLD 0(r)
WQHTRUSTSH(r) =
WQHTRUST(r) , WQHHLD(r)
WQHTRUSTSH 0(r) =
WQHTRUST 0(r) . WQHHLD 0(r)
(2.31)
32
Elena I. Ianchovichina and Robert A. McDougall
Substituting these into equation (2.30), we obtain WQHFIRM(r) WQHFIRM 0(r) WQHTRUST(r) + WQHTRUST(r) × log WQHTRUST 0(r) WQHHLD(r) −WQHHLD(r) × log . WQHHLD 0(r) (2.32)
WQHHLD(r) × CEHHLD(r) = WQHFIRM(r) × log
Because WQHHLD(r) and WQHHLD_0(r) are given, maximizing CEHHLD(r) is equivalent to maximizing FHHLD(r) = CEHHLD(r) + WQHHLD(r) × log
WQHHLD(r) . WQHHLD 0(r) (2.33)
Then, WQHHLD(r) × FHHLD(r) = WQHFIRM(r) WQHFIRM(r) + WQHTRUST(r) × log WQHFIRM 0(r) WQHTRUST(r) × log . (2.34) WQHTRUST 0(r) Similarly, maximizing the cross entropy associated with the local capital ownership shares is equivalent to maximizing FFIRM(r), where WQ FIRM(r) × FFIRM(r) = WQHFIRM(r) WQHFIRM(r) + WQTFIRM(r) × log WQHFIRM 0(r) × log
WQTFIRM(r) . WQTFIRM 0(r)
(2.35)
We seek to minimize a weighted sum of the two cross entropies: WSCE(r) = RIGWQH(r) × WQHHLD(r) × CEHHLD(r) + RIGWQ F(r) × WQ FIRM(r) × CEFIRM(r).
(2.36)
The two cross entropies are weighted by the corresponding total values, WQHHLD(r) and WQ_FIRM(r), and explicitly by the rigidity parameters RIGWQH(r) and RIGWQ_f(r). If RIGWQH(r) is assigned a high value and RIGWQ_f(r) a low one, then the solution will, if possible, keep the allocation of household wealth nearly fixed and put most of the onus of adjustment
Theoretical Structure of Dynamic GTAP
33
on the source shares for equity in local firms. If RIGWQ_F(r) is assigned a high value and RIGWQH(r) a low one, the equity source shares will tend to remain near their initial values, and the household wealth allocation shares do most of the adjusting. From the foregoing, minimizing WSCE is equivalent to minimizing the somewhat simpler equation: F = RIGWQH(r) × WQHHLD(r) × FHHLD(r) + RIGWQ F(r) × WQ FIRM(r) × FFIRM(r) ⎛ ⎞ WQHFIRM(r) ⎜WQHFIRM(r) × log WQHFIRM 0(r) ⎟ ⎜ ⎟ ⎜ = RIGWQH(r) × ⎜ WQHTRUST(r) ⎟ ⎟ ⎝+WQHTRUST(r) × log WQHTRUST 0(r) ⎠ ⎞ ⎛ WQHFIRM(r) ⎜WQHFIRM(r) × log WQHFIRM 0(r) ⎟ ⎟ ⎜ ⎟ +RIGWQ F(r) × ⎜ ⎜+ WQTFIRM(r) × log WQTFIRM(r) ⎟ . (2.37) ⎝ WQTFIRM 0(r) ⎠ To determine the three wealth variables, we minimize this objective function subject to the constraints (2.26) and (2.24). The Lagrangean contains corresponding multipliers: XWQHHLD(r) for the household wealth constraint (2.26) and XWQ_FIRM(r) for the firm value constraint (2.24). The first-order conditions include the two constraints and three equations corresponding to the three net wealth variables. Thus, differentiating the Lagrangean with respect to foreign equity in domestic capital, WQTFIRM(r), we obtain the first-order condition WQTFIRM(r) + 1 . (2.38) XWQ FIRM(r) = RIGWQ F(r) × log WQTFIRM 0(r) Differentiating again, we obtain xwq f(r) = RIGWQ F(r) × wqtf(r),
(2.39)
where xwq_f(r) denotes change in the Lagrange multiplier XWQ_FIRM(r). In TABLO code, we have Equation EQYHOLDFNDHHD #equity holdings of trust in the firms [wqtf(r)]# (all,r,REG) xwq_f(r) = RIGWQ_F(r)∗ wqtf(r).
34
Elena I. Ianchovichina and Robert A. McDougall
Likewise, for domestic ownership of foreign equity, we have the levels form of the first-order condition, WQHTRUST(r) +1 (2.40) XWQHHLD(r) = RIGWQH(r) × log WQHTRUST 0(r) and the differential form of the first-order condition, xwqh(r) = RIGWQH(r) × wqht(r) + swqht(r),
(2.41)
where xwqh(r) denotes change in the Lagrange multiplier XWQHHLD(r), and swqht(r) is a region-specific shift variable. In TABLO code we have Equation EQYHOLDHHDFND #equity holdings of households in trust[xwqh(r)]#(all,r,REG) xwqh(r) = RIGWQH(r)∗ wqht(r) + swqht(r). Finally, for domestic ownership of domestic equity, we have the levels form of the first-order condition, XWQHHLD(r) + XWQ FIRM(r) = (RIGWQH(r) + RIGWQ F(r)) WQHFRIM(r) +1 , × log WQHFIRM 0(r) (2.42) the differential form of the first-order condition, xwqh(r) + xwq f(r) = (RIGWQH(r) + RIGWQ F(r)) × wqhf(r), (2.43) and the TABLO code specification, Equation EQYHOLDHHDLCL #shift variable wealth of firms [xwq_f(r)]# (all,r,REG) [RIGWQH(r) + RIGWQ_F(r)]∗ wqhf(r) = xwqh(r) + xwq_f(r) + swqhf(r). where swqhf(r) is a region-specific shift variable. Note that, substituting for wqtf from equation (2.39), and for wqht from equation (2.41) into equation (2.43), we obtain (RIGWQH(r) + RIGWQ F(r)) × wqhf(r) = RIGWQH(r) × wqht(r) + RIGWQ F(r) × wqtf(r). (2.44)
Theoretical Structure of Dynamic GTAP
35
This equation shows that the adjustment in WQHFIRM(r) is an average of the adjustments in WQTFIRM(r) and WQHTRUST(r). Note also that if, for example, we assign a high value to RIGWQH(r) and a low value to RIGWQ_F(r), then xwqh(r) will assume a relatively large value and xwq_f(r) a relatively small value, so that xwqh(r) ≈ RIGWQH(r) · wqhf(r), and wqhf(r) ≈ wqht(r) = RIGWQH(r) − 1 xwqh(r); that is, the household wealth allocation shares are nearly fixed, as previously asserted and discussed further in Chapter 3.
4.5 Assets and Liabilities of the Global Trust There are three accounting identities associated with the global trust. First, the value of assets owned by the global trust, WQTRUST, is equal to the sum across regions of foreign ownership of firms: WQTFIRM(r). (2.45) WQTRUST = r
In percentage change form, we have WQTFIRM(r) × wqtf(r), WQTRUST × wqt =
(2.46)
r
where wqt is the percentage change in WQTRUST, and in the TABLO code, Equation TOTGFNDASSETS #change in the value of assets owned by global trust# WQTRUST∗ wqt = sum{s, REG, WQTFIRM(s)∗ wqtf(s)}. The second identity is that the value of the trust, WQ_TRUST, is equal to the sum of the regions’ equity in the trust; that is, to the sum across regions of ownership of foreign assets: WQHTRUST(r). (2.47) WQ TRUST = r
In percentage change form, equation (2.47) transforms into WQ TRUST × wq t = WQHTRUST(r) × wqht(r),
(2.48)
r
where wq_t is the percentage change in WQ_TRUST and in the TABLO code is given as
36
Elena I. Ianchovichina and Robert A. McDougall
Equation TOTGFNDPROP #value of trust as total ownership of trust# wq_t = sum{s, REG, [WQHTRUST(r)/WQ_TRUST]∗ wqht(s)}. Finally, the total value of the trust is equal to the total value of its assets: WQ TRUST = WQTRUST.
(2.49)
This equation as written would be redundant in the model, because it is implicit in other relations. The accumulation equations, together with the equivalence of global investment and global saving, ensure that the total value of physical capital is always equal to the total value of financial asset ownership by regions: so WQ FIRM(r) = WQHHLD(r). (2.50) r
r
Then WQ TRUST =
WQHTRUST(r)
r
=
WQHHLD(r) − WQHFIRM(r)
r
=
WQHHLD(r) −
r
=
WQHFIRM(r)
r
WQ FIRM(r) −
r
=
WQHFIRM(r)
r
WQ FIRM(r) − WQHFIRM(r)
r
=
WQTFIRM(r)
r
= WQTRUST.
(2.51)
To verify that simulation results satisfy the identity, we include the following equation: WQTRUST = WTRUSTSLACK × WQ TRUST,
(2.52)
where WTRUSTSLACK denotes an endogenous slack variable. In percentage change form, wqt = wq t + wtrustslack,
(2.53)
where wtrustslack denotes percentage change in WTRUSTSLACK. In the TABLO code, we have
Theoretical Structure of Dynamic GTAP
37
Equation GLOB_BLNC_SHEET # ownership by the trust equals ownership for the trust # wqt = wq_t + wtrustslack. Provided that the model database respects the asset accounting identities, and assuming no errors in the equations, the variable wtrustslack is endogenously equal to zero in any simulation. Thus the result for the slack variable provides a check on the validity of the model. Figure 2.1 illustrates these accounting relations. Corresponding to equation (2.46) for asset values, we have a price equation. As discussed in subsection 4.1, we can divide growth in assets and in proprietorship into matching investment and capital gain components. For the global trust, equating the capital gain components of assets and proprietorship yields the equation WQTFIRM(r) pcgds(r). (2.54) pqtrust = WQTRUST r In the code, this becomes Equation PKWRLD #change in the price of equity in the global fund# WQTRUST∗ pqtrust = sum{r, REG, WQTFIRM(r)∗ pcgds(r)}.
4.6 Income from Financial Assets Having determined stocks of financial assets in the previous subsections, we now determine the associated income flows. We do so in three stages. First, we determine payments from firms to households and to the global trust. Second, we calculate the total income of the global trust and determine payments from the trust to regional households. Third, we calculate the equity income of regional households as the sum of receipts from local firms and from the global trust. For an overview of the equity income flows, we refer to Figure 2.2. Firms in region r distribute to shareholders equity income payments YQ_FIRM(r), of which YQHFIRM(r) goes to the local regional household and YQTFIRM(r) to the global trust. Summing these receipts YQTFIRM(r) across regions, we obtain the total income YQTRUST of the global trust. The trust distributes this total income among the regional households, with region r receiving an amount YQHTRUST(r). Thus the total equity income of region r, YQHHLD(r), is the sum of receipts YQHFIRM(r) from local
38
Elena I. Ianchovichina and Robert A. McDougall INCOME(“r1”)
INCOME(“r2”)
YQHHLD(“r1”)
YQHHLD(“r2”)
YQHTRUST(“r1”)
YQHTRUST(“r2”)
YQTRUST
YQHFIRM(“r1”)
YQTFIRM(“r1”)
YQHFIRM(“r2”)
YQTFIRM(“r2”)
YQ_FIRM(“r2”)
YQ_FIRM(“r1”)
Figure 2.2. Income linkages.
firms and receipts YQHTRUST(r) from the global trust. This is summed with nonequity factor income and indirect taxes, which yields total regional income INCOME(r). We begin the detailed discussion with payments by firms. Firms buy intermediate inputs, hire labor, and rent land, but own fixed capital. By the zero pure profits condition, their profits are equal to the cost of capital services, excluding any factor usage or income taxes, less depreciation. These profits accrue to shareholders. Thus total income payments by firms in region r to shareholders, YQ_FIRM(r), are equal to net after-tax capital earnings: YQ FIRM(r) = VOA(“capital”, r) − VDEP(r),
(2.55)
where VOA(“capital”, r) is the value of capital earnings, and VDEP(r) is the value of capital depreciation. Differentiating, we obtain YQ FIRM(r) × yq f(r) = VOA(“capital”, r) × (rental(r) + qk(r)) −VDEP(r) × (pcgds(r) + qk(r)),
(2.56)
where yq_f(r) denotes the percentage change in income payments by firms in region r, and rental(r) denotes the percentage change in the rental price
Theoretical Structure of Dynamic GTAP
39
of capital. In the code, this equation is written as Equation REGINCEQY #income from capital in firms in region r# (all,r,REG) YQ_FIRM(r)∗ yq_f(r) = sum{h, ENDWC_COMM, VOA(h,r)∗ [ps(h,r) + qo(h,r)]} − VDEP(r)∗ [pcgds(r) + qk(r)]. To relate the equation in the TABLO code to its mathematical form, note that ENDWC_COMM is a set with just one element: “capital”, with ps(“capital”, r) = rental(r) and qo(“capital”, r) = qk(r). Firms distribute payments among shareholders in proportion to their shareholdings. The local regional household owns WQHFIRM, and the global trust owns WQTFIRM with a total equity value WQ_FIRM (see subsection 4.4). So for payments YQHFIRM(r) to the local regional household, we have WQHFIRM(r) YQHFIRM(r) = × YQ FIRM(r). (2.57) WQ FIRM(r) Differentiating, we obtain yqhf(r) = yq f(r) + wqhf(r) − wq f(r),
(2.58)
where yqhf(r) denotes the percentage change in YQHFIRM(r). In the TABLO code this equation is given as Equation INCHHDLCLEQY #income of the household from local firms]# (all,r,REG) yqhf(r) = yq_f(r) + wqhf(r) − wq_f(r). Similarly, payments to the global trust, YQTFIRM(r), are given by YQTFIRM(r) =
WQTFIRM(r) × YQ FIRM(r). WQ FIRM(r)
(2.59)
Differentiating, we obtain yqtf(r) = yq f(r) + wqtf(r) − wq f (r),
(2.60)
where yqtf(r) is the percentage change in YQTFIRM(r). In the TABLO code, we write Equation INCFNDLCLEQY #income of trust from equity in region r# (all,r,REG) yqtf(r) = yq_f(r) + wqtf(r) − wq_f(r).
40
Elena I. Ianchovichina and Robert A. McDougall
In the second stage, we compute total income receipts and the various income payments of the global trust. The total income of the trust, YQTRUST, is equal to the sum of equity receipts from firms in each region. In levels, we express this as YQTFIRM(r), (2.61) YQTRUST = r
in percentage changes as yqt =
YQTFIRM(r) r
YQTRUST
× yqtf(r),
(2.62)
where yqt denotes the percentage change in YQTRUST, and in the TABLO code as Equation INCFNDEQY #change in the income of the trust# yqt = sum{r, REG, [YQTFIRM(r)/YQTRUST]∗ yqtf(r)}. The trust distributes its income among its shareholders, so that each region r receives income YQHTRUST(r) in proportion to its ownership share. This is expressed in the levels equation, YQHTRUST(r) =
WQHTRUST(r) × YQTRUST, WQ TRUST
(2.63)
the differential equation, yqht(r) = yqt + wqht(r) − wq t,
(2.64)
where yqht(r) denotes the percentage change in YQHTRUST, and in the TABLO code as Equation REGGLBANK #income of household r from its shares in the trust# (all,r,REG) yqht(r) = yqt + wqht(r) − wq_t. In the third and final stage we compute the financial asset income of regional households. Total equity income YQHHLD(r) of regional household r equals the sum of equity income received from domestic firms and from the global trust: YQHHLD(r) = YQHFIRM(r) + YQHTRUST(r).
(2.65)
Theoretical Structure of Dynamic GTAP
41
In percentage changes, this equation transforms into yqh(r) =
YQHFIRM(r) YQHTRUST(r) × yqhf(r) + × yqht(r), YQHHLD(r) YQHHLD(r)
(2.66)
where yqh(r) denotes percentage change in YQHHLD. In the TABLO code, it is given as Equation TOTINCEQY #total income from equity of households in r#(all,r,REG) yqh(r) = [YQHFIRM(r)/YQHHLD(r)]∗ yqhf(r) + YQHTRUST(r)/ YQHHLD(r)]∗ yqht(r).
5. Investment Theory In this section we describe a lagged adjustment, adaptive expectations theory of investment. Investors act so as to eliminate disparities in expected rates of return not instantaneously, but progressively through time. Moreover, their expectations of rates of return may be in error, and these errors are also corrected progressively through time. Finally, in estimating future rates of return, they allow for some normal rate of growth in the capital stock; this normal rate too is an estimated rate that investors adjust through time.
5.1 The Required Rate of Growth in the Rate of Return In a simple perfect adjustment model of investment, profit-maximizing investors would keep rates of return uniform across regions, because any differences in rates of return would be immediately eliminated by a reallocation of capital from regions with lower rates of return to regions with higher rates. This equalization would apply to net rates of return, so that we might write, for each region r, RORNET(r) = RORCOMM, where RORNET(r) denotes the net rate of return on capital in region r, and RORCOMM the common world rate of return. If we allow for region-specific risk premia RRISK(r), then we postulate equalization not of the actual net rates of return RORNET(r) but of the risk-adjusted rates RORNET(r) − RRISK(r), so that, for all regions r, RORNET(r) = RORCOMM + RRISK(r). Furthermore, as we find later, it is convenient to express the investment theory in terms of gross rather than net rates of return; anticipating this, we write RDEP(r) for the depreciation
42
Elena I. Ianchovichina and Robert A. McDougall
rate in region r and obtain, for the gross rate of return the equilibrium condition, RORGROSS(r) − RORCOMM − RRISK(r) − RDEP(r) = 0.
(2.67)
In principle, the gross rate of return RORGROSS(r) includes both an earnings component and a capital gains component: RORGROSS(r) =
RENTAL(r) + RG PCGDS(r), PCGDS(r)
(2.68)
where RENTAL(r) denotes the rental price of capital in region r, and RG_PCGDS(r) denotes the rate of growth in the purchase price of capital. In practice, with a period-by-period solution method, we do not know the rate of growth in the purchase price of capital,4 so we neglect it and define the gross rate of return as the earnings rate only: RORGROSS(r) =
RENTAL(r) . PCGDS(r)
(2.69)
Differentiating equation (2.69), we obtain the percentage change equation: rorga(r) = rental(r) − pcgds(r),
(2.70)
where rorga(r) denotes percentage change in RORGROSS. In the TABLO code, this equation is represented as follows: Equation RATERETURNP #identity for rate of return# (all,r,REG) rorga(r) = rental(r) − pcgds(r). We now consider the investment response to sudden (i.e., instantaneous) price changes. For example, sudden price changes may occur as the result of sudden tax rate changes. Sudden changes in output or input prices affect the capital rental price, RENTAL(r), and thereby the rate of return. In a perfect adjustment model with capital gains, these changes in the rate of return must be offset by some sudden change in PCGDS or RG_PCGDS, or by some sudden offsetting influence on RENTAL, so as to maintain international equality in rates of return as defined in equation (2.68). 4
In fact, we can estimate the backward-looking growth rate, limH->0- (PCGDS(r,T + H)PCGDS(r,T ))/H, where PCGDS(r,T ) denotes the value of PCGDS(r) at time T. This, however, is liable to differ from the forward-looking growth rate, limH->0 + (PCGDS(r,T + H)PCGDS(r,T ))/H, which is the one needed in the rate-of-return formula.
Theoretical Structure of Dynamic GTAP
43
Suppose initially that the supply of capital goods is perfectly elastic. Then a first-round improvement in profitability (i.e., a first-round positive effect on RENTAL) leads to an increase in the capital stock – increasing output supply and possibly increasing demand for noncapital inputs – and thereby negating the first-round effect on RENTAL. If the initial shock is sudden, then so also must be the increase in the capital stock, implying an infinite rate of investment over an infinitesimal time period. Of course, in the real world capital stocks do not adjust in this manner. Instantaneous adjustment of capital stocks is precluded by gestation lags, adjustment costs, imperfect elasticity of the supply of capital, and other factors. In addition, in a CGE model, even if other realistic features are lacking, the supply of capital is typically imperfectly elastic. If we rule out infinite rates of investment, how can rate-of-return equalization be maintained in the face of sudden shocks affecting profitability? The answer is through sudden changes in the price of capital goods. A sudden improvement in earnings leads to a sudden increase in demand for capital goods, and that in turn leads to a sudden increase in the price of capital goods. This price increase helps stabilize the rate of return in two ways. First, it reduces the earnings rate RENTAL(r)/PCGDS(r). Second, it leads to a decrease in the rate of capital gain, RG_PCGDS. As demand for capital goods eases through time after the initial spike, or the supply of capital goods gradually rises, the price of capital goods tends to fall through time after its initial increase. In our model, we cannot capture the capital gains effect of an increase in demand for capital goods, but we can capture the earnings rate effect. Thus the way appears open in principle to use a perfect adjustment mechanism for investment. However, because we do not capture all the effects of the increase in demand for capital, it is likely that the model will require unrealistically large increases in the price of capital goods and in the level of investment. Indeed, there are several reasons why the model would tend to exaggerate investment volatility, some of which have already been mentioned and some not: r The model does not capture the capital gain effect of capital goods
price changes.
r As we typically use it in dynamic simulations, the model assumes
perfect capital mobility within regions. Accordingly, it overstates the elasticity of the supply of capital goods. r The model does not incorporate other real-world effects such as gestation lags or adjustment costs.
44
Elena I. Ianchovichina and Robert A. McDougall
For all these reasons, the perfect adjustment approach is unrealistic in the context of this model. We pursue accordingly a lagged adjustment approach. Recalling equation (2.67), we rewrite it as RORGROSS(r) − RORGTARG(r) = 0,
(2.71)
where RORGTARG(r) denotes the target rate of return in region r. To move to a lagged adjustment approach, we replace this in turn by RRG RORG(r) = LAMBRORG(r)∗ log
RORGTARG(r) , RORGROSS(r)
(2.72)
where RRG_RORG(r) denotes the required rate of growth in the rate of return, and LAMBRORG(r) denotes a coefficient of adjustment. Differentiating, we obtain rrg rorg(r) = LAMBRORG(r)∗ [rorgt(r) − rorga(r)],
(2.73)
where rrg_rorg(r) denotes (absolute) change in the required rate of growth in the rate of return in region r, and rorgt(r) denotes the percentage change in the target rate of return. Note that this is not the final form of the equation. We present it in subsection 5.3, following further theoretical development. Referring back to equation (2.67), we note that the target rate of return includes both region-specific components RRISK(r) and RDEP(r) and a region-generic component RORCOMM. In the present context there is another possible region-generic component, a worldwide drift in rates of return such as to accommodate the global level of investment. We do not represent all these components explicitly in the model, but instead write simply RORGTARG(r) = SDRORTWORLD + SDRORTARG(r),
(2.74)
where SDRORTWORLD denotes a region-generic component in the target rate of return, and SDRORTARG(r) a component specific to region r. Differentiating, we obtain DRORT(r) = SDRORTW + SDRORT(r),
(2.75)
where DRORT(r) denotes the absolute change in the target rate of return; SDRORTW, a region-generic shift; and SDRORT(r), a region-specific shift. We use here the absolute rather than the percentage change form for the target rate, to ensure that any worldwide shift SDRORTW leads to equal percentage-point changes in rates of return in different regions; equivalently, to ensure that any cross-region differentials are maintained in percentagepoint rather than percentage terms (so, for example, we might maintain a
Theoretical Structure of Dynamic GTAP
45
risk premium of two percentage points, but not a risk premium equivalent to 20% of the rate of return). We have then DRORT(r) = SDRORTW + SDRORT(r),
(2.76)
or in TABLO code, Equation NET_ROR #equilibrium condition for rate of return# (all,r,REG) DRORT(r) = SDRORTW + SDRORT(r). We relate the absolute-change variable DRORT to the percentage-change variable rorgt with the equation RORGTARG(r) × rorgt(r) = DRORT(r),
(2.77)
and in the code, Equation GROSS_ROR #identity for target gross rate of return# (all,r,REG) RORGTARG(r)∗ rorgt(r) = DRORT(r).
5.2 The Expected Rate of Growth in the Rate of Return Having determined in subsection 5.1 the required rate of growth in the rate of return, we now relate it to the level of investment through an equation linking the expected rate of growth in the rate of return to investment, with a condition that the expected rate should be equal to the required rate. This brings us to one of the central elements of the investment theory in GDyn, the expected rate-of-return schedule. Investors understand that, the higher the level of the capital stock at any given time, the lower the rate of return at that time. Accordingly, the rate of return expected to prevail at any future time depends on the level of the capital stock at that time. Consequently, the expected rate of growth in the rate of return depends on the rate of growth in the capital stock or, equivalently, on the level of investment. We describe investors’ understanding of the investment environment through a rate-of-return schedule, relating the expected rate of return to the size of the capital stock: QK(r) −RORGFLEX(r) RORGEXP(r) = , (2.78) RORGREF(r) QKF(r) where RORGEXP(r) denotes the expected gross rate of return, and RORGFLEX(r) denotes a positive parameter, representing the absolute magnitude of the elasticity of the expected rate of return with respect to the size
46
Elena I. Ianchovichina and Robert A. McDougall
of the capital stock. RORGREF(r) is a reference rate of return in region r, and QKF(r), a reference capital stock. Investors expect that if the actual capital stock QK is equal to the reference stock QKF, then the rate of return will be equal to the reference rate RORGREF. If the capital stock exceeds the reference stock, the expected rate of return is less than the reference rate. If the capital stock is less than the reference stock, the expected rate is greater than the reference rate. In dealing with expectations, as in equation (2.78), there are two relevant times: the time at which the expectations are held and the time to which they refer. We call these, respectively, the expectation time and the realization time. So for example, in describing an investor in 2000 holding an expectation about the rate of return in 2005, the expectation time is 2000, and the realization time is 2005. In the theory underlying the investment module, expectation time is always just the current time, TIME, for the model. For example, if the model represents the state of the world economy in the year 2000, then the expectation time is 2000. However, realization time, TREAL, may be either the current or some future time. In the model itself, as opposed to the underlying theory, expectation time and realization time are always equal to the current time. So in the model equations TREAL would be redundant, and we use only the current time, TIME. To complete our description of investor expectations in equation (2.78), we need to specify how the reference rate of return and the reference capital stock depend on realization time. We postulate that the reference rate of return is independent of realization time, whereas the reference capital stock grows at some normal rate KHAT(r): QKF(r) = QKO(r)e KHAT(r) TREAL ,
(2.79)
where QKO(r) denotes the reference capital stock at some base time TREAL = 0. Under this treatment, the normal rate of growth KHAT(r) is the rate at which the capital stock can grow without (as investors expect) affecting the rate of return. If the capital stock grows at a rate greater than KHAT(r), investors expect rates of return to decline through time. If the capital stock grows at less than KHAT(r), investors expect rates of return to increase. The specification of expectations in equations (2.78) and (2.79), although simple, is intended to approximate the actual investment schedule. In particular, it allows a range between zero and infinity for the gross rate of return, RORGROSS(r). This allows, realistically, the net rate of return to be negative sometimes. Whether the specification is locally model consistent depends on the setting of the normal growth rate KHAT(r) and the elasticity
Theoretical Structure of Dynamic GTAP
47
RORGFLEX(r). As discussed in subsection 5.4, we allow model-consistent adjustment of KHAT(r), but RORGFLEX(r) is fixed. We can set it initially at a locally model-consistent value, but through a simulation or series of projections, it typically becomes more or less inconsistent. This is undesirable, but also unavoidable without a considerable increase in the complexity of the theory. To find the expected rate of growth in the rate of return, we differentiate equation (2.78) with respect to realization time. Substituting for QKF(r) from equation (2.79), we obtain ERG RORG(r) = −RORGFLEX(r) × (RG Q K (r) − KHAT(r)),
(2.80)
where ERG_RORG(r) denotes the expected rate of growth in the rate of return in region r, and RG_QK(r), the rate of growth in the capital stock. The rate of growth in the capital stock is RG Q K (r) =
QCGDS(r) − RDEP(r)Q K (r) dQK(r)/dTIME = Q K (r) Q K (r) QCGDS(r) − RDEP(r), (2.81) = Q K (r)
where QCGDS(r) denotes the level of investment in region r. Substituting the expression for RG_QK(r) into equation (2.80), we obtain ERG RORG(r) = −RORGFLEX(r) QCGDS(r) − RDEP(r) − KHAT(r) . × Q K (r)
(2.82)
Totally differentiating equation (2.82), we obtain erg rorg(r) = −RORGFLEX(r) × [IKRATIO(r) ×(qcgds(r) − qk(r)) − DKHAT(r)],
(2.83)
where erg_rorg(r) denotes the absolute change in the expected rate of growth in the rate of return in region r; IKRATIO(r), the ratio QCGDS(r)/QK(r) of gross investment to the capital stock; qcgds(r), the percentage change in gross investment; and DKHAT(r), the absolute change in the normal rate of growth in the capital stock. Then, in the TABLO code we have Equation EGROWTH_ROR #rule for expected rate of growth in rate of return# (all,r,REG) erg_rorg(r) = − RORGFLEX(r)∗ IKRATIO(r)∗ [qcgds(r) − qk(r)] + RORGFLEX(r)∗ DKHAT(R).
48
Elena I. Ianchovichina and Robert A. McDougall R
RORGROSS RORGTARG RORGREF
(A)
QK
QKT QKF
QK
Figure 2.3. Actual investment schedule.
As equation (2.83) shows, the expected rate of growth in the rate of return varies inversely with the level of investment. The level of investment is given implicitly by the condition that the expected rate of growth is equal to the required rate: ERG RORG(r) = RRG RORG(r).
(2.84)
We depict some aspects of the investment theory in Figure 2.3. Each point in the figure represents a (capital stock, rate-of-return) pair (QK, R). The curve (A) represents the expected rate-of-return schedule for realization time equal to expectation time. It is downward sloping, with the slope related to the elasticity RORGFLEX, a vertical asymptote at QK = 0, and a horizontal asymptote at R = 0. It passes through the point (QK, RORGROSS) representing the current capital stock and rate of return, and also through the reference point (QKF, RORGREF). For a realization time greater than the expectation time, the curve would be similar to (A), but dilated about the vertical axis. Assuming a positive normal growth rate, the curve would dilate rightward as realization time advanced. As drawn in Figure 2.3, the actual rate of return RORGROSS exceeds the target rate RORGTARG. From equation (2.72), this implies that the required rate of growth in the rate of return is negative. On the diagram in Figure 2.3, this implies some required downward vertical speed. By inverting equation (2.82) and equating the required rate of growth in the rate of return to the expected rate, we find the required level of investment: QCGDS(r) = Q K (r) × [RDEP(r) + KHAT(r) − (RORGFLEX(r))−1 × RRG RORG(r)].
(2.85)
Here QK(r)×RDEP(r) is the investment level required to maintain the capital stock QK(r) at its current level, QK(r)×KHAT(r) is the further
Theoretical Structure of Dynamic GTAP
49
investment required to keep pace with the rightward dilation of the rate of return curve (A), and QK(r) × (RORGFLEX(r)) − 1 × RRG_RORG(r) is the further investment required to maintain the required vertical speed down the curve.
5.3 Adaptive Expectations In practice, the investment theory as presented to this point in equations (2.73), (2.83), and (2.84) has a significant disadvantage. Using information in the benchmark data, we can calculate the actual rate of return, RORGROSS(r), in the initial year. The rate of return and the equations of the model allow us to determine the level of investment QCGDS(r). However, the benchmark data also specify the level of investment. In general this level will be inconsistent with the level calculated with the theory. Consider, for example, the region with the highest rate of return in the database. In this region the actual rate of return exceeds the target rate, so the required rate of growth in the rate of return RRG_RORG(r) is negative. This, in turn, implies that the normal rate of growth in the capital stock KHAT(r) and investment QCGDS(r) should be high. However, it may be that the level of investment recorded in the database is not particularly high. In this case, theory and data are inconsistent. We can resolve this inconsistency by modifying either the data or the theory. One approach to modifying the data involves equalizing rates of return across countries in the database. This conflicts with one of our objectives for the dynamic model, specifically that it should work with databases that conform closely to observed statistics, rather than requiring a heavily recalibrated or stylized database. Another approach would be to account for investment-level anomalies through risk premia. We can do this readily in the database, without touching the flows data, by adjusting the target rates. This option is sometimes appealing. We do not wish, however, to force it on users. Rather we recognize that reality is under no obligation to respect our or any other investment theory and that, for a multitude of reasons, observed investment levels will surely differ from any theoretical prediction. We therefore extend the theory so that it does not prescribe investment levels, but accommodates the observed investment levels over the short-run, while still maintaining the old theory’s long-run properties. We achieve the desired relaxation by letting investors react to expected rather than actual rates of return. With this approach we can account for any observed level of investment by setting the expected rate of return so as to warrant that investment level. At the same time, by incorporating an adjustment mechanism that draws the expected rate of return gradually
50
Elena I. Ianchovichina and Robert A. McDougall
toward the actual rate, we retain the long-run properties of the simpler theory, including long-run equalization of rates of return. Furthermore, this way of accounting for observed investment levels has some theoretical appeal. Investment is undertaken with the expectation of deriving returns over some period of time. Thus, investors are concerned with the rate of return not only at the moment of purchasing an asset but also throughout its life. Investors’ expectations are also “sticky” or “sluggish.” When the observed rate of return changes, investors are unsure whether this change is transient or permanent. They adjust their expectations of future rates of return only with a lag. At first investors make a small adjustment; then if the change in the actual rate persists, they make further changes in expectations, until eventually the expected rate conforms to the observed rate. Earlier, in subsection 5.1, we represented investors’ reactions to current returns through equation (2.73). To let investors react to the expected rate of return rather than the actual rate, we replace the actual rate-of-return variable rorga in equation (2.73) with the expected rate variable rorge. At the same time, we enforce the condition that the expected rate of growth in the rate of return will be equal to the required rate by replacing the required rate rrg_rorg with the expected rate erg_rorg. This changes equation (2.73) into the following equation: erg rorg(r) = LAMBORG(r) × [rorgt(r) − rorge(r)].
(2.86)
In the code, we implement this equation as Equation INVESTMENT # rule for investment # (all,r,REG) erg_rorg(r) = LAMBRORG(r)∗ [rorgt(r) − rorge(r)]. We now need to specify an error-correction mechanism that brings the expected rate, rorge, closer through time to the actual rate, rorga. We recall equation (2.78) for the expected rate-of-return schedule and note a few points: r Even before we introduced the theoretical extension in this subsection,
we already had a concept of an expected rate of return.
r Previously, however, the expected rate-of-return schedule was such
that, at the current capital stock and the current time, the expected rate of return was equal to the actual rate of return. Now we allow the expected rate of return to differ from the actual rate.
Theoretical Structure of Dynamic GTAP
51
r As specified by the expected rate-of-return schedule, the expected rate
of return is conditional on the capital stock and also on realization time. This rules out a simple adjustment rule for the expected rate of return, such as rorga(r) × time. rorge(r) = 100 × LAMBRORGE(r) × log rorge(r)
(2.87)
This adjustment rule would represent investors as perversely ignoring the effects of investment and economic growth on the rate of return. Rather than an adjustment rule for the rate of return itself, we need an adjustment rule for the rate-of-return schedule, shifting so that through time the expected current rate of return draws closer to the actual current rate. From equations (2.78) and (2.79), we note that the position of the rateof-return schedule is given by the reference rate of return, RORGREF(r), and the base time value of the reference capital stock, QKO(r). To specify an error-correction mechanism for the rate-of-return schedule, we define the warranted reference rate of return, RORGFWARR(r), as value for the reference rate that equates the expected rate RORGEXP(r) to the actual rate RORGROSS(r) in equation (2.78). Then the warranted reference rate of return is given implicitly by the equation: Q K (r) −RORGFLEX(r) RORGROSS(r) = . (2.88) RORGFWARR(r) Q K F (r) From equations (2.78) and (2.88), we have RORGROSS(r) RORGFWARR(r) = . RORGREF(r) RORGEXP(r)
(2.89)
We postulate an error-correction process through which the reference rate of return draws closer through time to the warranted rate: RORGFWARR(r) × time, rorgf(r) = 100 × LAMBRORGE(r) × log RORGREF(r) (2.90) where rorgf(r) denotes the percentage change in the reference rate of return, and LAMBRORGE(r) is an adjustment coefficient. Substituting from equation (2.89), we obtain RORGEXP(r) time rorgf(r) = −100 × LAMBRORGE(r) × log RORGROSS(r) = −100 × LAMBRORGE(r) × ERRRORG(r) × time,
(2.91)
52
Elena I. Ianchovichina and Robert A. McDougall
where ERRRORG(r) is a measure of error in the expected rate of return, ERRRORG(r) = log(RORGEXP(r)/RORGROSS(r)). Having specified this error-correction mechanism for the expected rateof-return schedule, we can now derive the error-correcting equation for the expected rate of return itself. By substituting equation (2.79) into equation (2.78), we obtain −RORGFLEX(r) Q K (r) RORGEXP(r) = . (2.92) RORGREF(r) QKO(r)e KHAT(r)TIME At this point, we add one final feature. For various reasons, users may sometimes wish to intervene in the expectation-setting process. They may wish to add some additional shock to the expected rate of return or to deactivate the expectations rule, so as, for example, to set the investment level directly. To allow this, we add a shift factor SRORGEXP(r) to the expected rate equation. This variable is normally exogenous and zero, but it may be given a nonzero value to add exogenous shocks to the expectationsetting process, or it may be endogenized to disable the expectations rule so that, for example, the investment level may be set directly. This gives us the final form for the levels equation: RORGEXP(r) = SRORGEXP(r) RORGREF(r) −RORGFLEX(r) Q K (r) × . QKO(r)e KHAT(r)TIME
(2.93)
Differentiating equation (2.93), we obtain rorge(r) = rorgf(r) − RORGFLEX(r) × (qk(r) − 100 × KHAT(r) × time) + srorge(r),
(2.94)
where srorge(r) denotes the percentage change in the expected rate shift factor. Substituting for rorgf from equation (2.91), we obtain rorge(r) = −RORGFLEX(r) × (qk(r) − 100 × KHAT(r) × time) −100 × LAMBRORGE(r) × ERRRORG(r) × time + srorge(r). (2.95) This equation shows three sources of change in the expected rate of return: (1) divergence between the actual rate of growth in the capital stock, qk(r)/[100 · time], and the normal growth rate KHAT(r); (2) a correction for the observed error in the expected rate; and (3) an exogenous shift
Theoretical Structure of Dynamic GTAP
53
factor. We implement this in the model as follows: Equation EXPECTED_ROR #rule for expected gross rate of return# (all,r,REG) rorge(r) = −RORGFLEX(r)∗ [qk(r) − 100.0∗ KHAT(r)∗ time] −100.0∗ LAMBRORGE(r)∗ ERRRORG(r)∗ time + srorge(r).
5.4 The Normal Rate of Growth in the Capital Stock As noted in subsection 5.2, whether the expected growth rate ERG_RORG is model-consistent depends in part on the normal growth rate KHAT. In some early versions of the model, we treated KHAT as a fixed parameter, calibrating it before each base-case projection to ensure that it was consistent with the long-run behavior of the model. This had two disadvantages. It forced us to calibrate the parameter anew for each base-case projection, which was somewhat onerous. It also held KHAT constant within each projection, which was not always appropriate. For example, holding that variable constant would not be appropriate to a projection involving slow technological progress through the 1980s, but faster progress through the 1990s. We could avoid this problem by setting several different KHAT values for different periods within the projection, but that would involve yet more calibration simulations. To avoid these problems, we now treat the normal growth rate KHAT as an updatable coefficient within the model and provide an adjustment mechanism to bring it toward a model-consistent value through the course of a simulation. We postulate an adjustment mechanism DKHAT(r) = 100 × LAMBKHAT(r) × (KHAPP(r) − KHAT(r)) × time, (2.96) where LAMBKHAT(r) denotes a coefficient of adjustment, and KHAPP(r) the apparent current normal growth rate in region r. By the apparent current normal growth rate, we mean the normal growth rate implied by current changes in the capital stock and the rate of return and by the assumed elasticity RORGFLEX. If the rate of return is currently constant, then it appears that the capital stock is growing at the normal rate, so the apparent normal rate is equal to the actual rate. If the rate of return is rising, then the apparent normal rate is greater than the actual rate. If the rate of return is falling, the apparent normal rate is lower than the actual rate.
54
Elena I. Ianchovichina and Robert A. McDougall
To calculate the apparent normal rate, we return to the expected investment schedule equation (2.78), assume that it agrees with the actual schedule, and solve for the apparent value KHAPP of the normal growth rate KHAT. We thus obtain KHAPP(r) = RORGFLEX(r)−1 ARG RORG(r) +
QCGDS(r) − RDEP(r). Q K (r) (2.97)
This shows that the apparent normal growth rate KHAPP(r) is equal to the actual growth rate QCGDS(r)/QK(r) − RDEP(r), plus an adjustment RORGFLEX(r) − 1 ARG_RORG(r) for current growth in the rate of return. Substituting into equation (2.96), we obtain DKHAT(r) = 100 × LAMBKHAT(r) × (RORGFLEX(r)−1 × ARG RORG(r) + QCGDS(r)/Q K (r) − RDEP(r) − KHAT(r)) × time. (2.98) Now adapting equation (2.10), we have QCGDS(r) − RDEP(r) × time. qk(r) = 100 × Q K (r)
(2.99)
Also, by definition of ARG_RORG(r) we have rorga(r) = 100 × ARG RORG(r) × time.
(2.100)
Substituting into equation (2.98), we obtain DKHAT(r) = LAMBKHAT(r) × (qk(r) + RORGFLEX(r)−1 rorga(r) + 100 × KHAT(r) × time).
(2.101)
Translating into TABLO code, we have in the model Equation KHATGROWTH # behavioral equation for estimated normal rate growth rate # (all,r,REG) DKHAT(r) = LAMBKHAT(r)∗ [qk(r) + (1/RORGFLEX(r))∗ rorga(r) −100.0∗ KHAT(r)∗ time] + SDKHAT(r). The exogenous shift variable SDKHAT(r) is included for modeling purposes. Figure 2.4 shows two rate-of-return curves: the expected rate-of-return curve (E), passing through the current capital stock and the expected current rate-of-return (QK, RORGEXP), and the warranted curve (A), passing
Theoretical Structure of Dynamic GTAP
55
R
RORGROSS
RORGTARG RORGREF RORGEXP RORGEREF
(A) (E) QK QKEF QKT QKF
QKW
QK
Figure 2.4. Actual and expected investment schedules.
through the current capital stock and actual current rate of return (QK, RORGROSS). As before, the expected investment curve dilates rightward through time at a rate given by the normal rate of growth in the capital stock KHAT, or leftward, if KHAT is negative. Yet now it also dilates vertically, so as to draw closer to the warranted curve (A). The shape of the curve is such that any vertical dilation is equivalent to a horizontal dilation, and vice versa. Specifically, a vertical dilation by a factor V is equivalent to a horizontal dilation by a factor V. Therefore we may say simply that the curve dilates inward or outward, at a rate depending on the normal growth rate KHAT, but adjusted so as to draw closer to the warranted curve (A). As the expected rate curve dilates outward, so too does the warranted rate curve, at a rate described by the apparent normal growth rate KHAPP. If the error in expectations is zero (RORGEXP(r) = RORGROSS(r)) and the expected normal growth rate KHAT agrees with the apparent rate KHAPP, then the expected rate and warranted rate curves (E) and (A) coincide and also dilate outward together at the same rate so as to remain coincident. If the error in expectations is zero (RORGEXP(r) = RORGROSS(r)), but the apparent normal growth rate KHAPP exceeds the expected normal growth rate KHAT, then the expected rate curve (E) and the warranted rate curve (A) are initially coincident, but the warranted rate curve dilates outward faster than the expected rate curve. Through the normal rateadjustment process, the normal rate accelerates toward the apparent rate, pushing the velocity of the expected rate curve closer to that of the warranted rate curve, whereas the rate-of-return adjustment process pushes the position of the expected rate curve closer to that of the warranted rate curve.
56
Elena I. Ianchovichina and Robert A. McDougall Table 2.1. Investment module DRORT (r) = SDRORTW + SDRORT (r) RORGTARG(r) · rorgt(r) = DRORT (r) rorge(r) = − RORGFLEX (r)[qk(r) − 100 · KHAT (r) · time] − 100 · LAMBRORGE(r) · ERRRORG(r) · time + srorge(r) erg rorg(r) = LAMBRORG(r)[rorgt(r) − rorge(r)] DKHAT (r) = LAMBKHAT (r) [RORGFLEX (r) − 1 rorga(r) + qk(r) + 100 · KHAT (r) · time] erg_rorg(r) = − RORGFLEX (r){IKRATIO(r)[qcgds(r) − qk(r)] − DKHAT (r)}
(2.76) (2.77) (2.95) (2.86) (2.101) (2.83)
If the expected normal growth rate KHAT agrees with the apparent growth rate KHAPP, but the expected rate of return RORGEXP exceeds the actual rate RORGROSS, then the expected rate curve (E) lies outside the warranted rate curve (A). Then the expected rate curve dilates outward at less than the normal rate, allowing the warranted rate curve to catch up with it.
5.5 Summary Equations (2.86), (2.83), (2.101), and (2.95), shown in Table 2.1, comprise the investment theory of adaptive expectations and jointly determine the forward-sloping regional supply of investment funds. With this set of equations, there is perfect capital mobility only over the long-run as regional rates of return gradually adjust toward a common target rate. Equation (2.95) both determines the expected rate of return rorge and (in Fig. 2.4) governs the position of the expected rate-of-return curve (E). It lets the expected rate curve (E) dilate outward at a rate governed partly by equation (2.101) and partly by a catch-up component drawing toward the warranted rate curve (A). Equation (2.101) coordinates the movements of curves (A) and (E) so that, abstracting from the catch-up effect, their velocities draw together. Equation (2.86) specifies the required rate of growth in the expected rate of return – the required vertical velocity of the point (QK, RORGEXP) in Figure 2.4. Equation (2.83) translates this into a required level of investment, or horizontal velocity within the figure, given the vertical velocity and the requirement that the point lie on the expected rate curve (E). Thus equations (2.86), (2.83), (2.101), and (2.95) determine regional investment and, via the accumulation equation (2.13), regional capital stocks in GDyn. To illustrate the disequilibrium nature of the adjustment mechanism in this model, let us assume initial equality between the actual, expected, and
Theoretical Structure of Dynamic GTAP
57
target rates of return. This equality implies that the actual and expected rate-of-return schedules overlap and move together in response to changes in the normal rate of growth in the capital stock KHAT(r). If there is a positive shock to productivity, the actual rate of return increases, and the warranted rate-of-return schedule (A) moves to the right of the expected rate-of-return schedule (E). The model detects the acceleration in economic growth via equation (2.101) in the initial period of the shock, which leads to an increase in regional investment via equation (2.83). In the next period, a further increase in the expected normal growth rate KHAT(r) leads to a further rise in regional investment. Graphically, this rise is represented by an outward dilation of the expected rate-of-return curve. In addition, via the second term of equation (2.95), investors realize that the expected rate of return is lower than the actual rate of return. This leads both to a further outward dilation of the expected rate-of-return curve toward the warranted rate-of-return curve and, through equation (2.86), a decrease in the required rate of growth in the rate of return, RRG_RORG. Equation (2.83) translates this into a rightward movement along the expected rate curve. These four equations, together with the equations (2.76) and (2.77) for rorgt, form an equation subsystem. With a normal closure, the subsystem takes as given variables qk, SDRORT, srorge, and time, and one degree of freedom of qcgds, because qcgds is constrained by the requirement that the money value of world investment must equal world saving. It determines the variables qcgds and SDRORTW. The variables DKHAT, DRORT, erg_rorg, rorge, and rorgt are internal to the subsystem. In its relations with the rest of the equation system, this subsystem has some notable features. It may be surprising that the capital stock qk(r) helps determine the investment level. Referring back to the derivation of equation (2.83), we see that achieving a given expected rate of growth in the rate of return entails achieving a certain rate of growth in the capital stock. The level of investment required to achieve that rate of growth depends on the size of the capital stock. The capital stock thus serves as a scaling factor for investment. Equally notable is the absence of certain links from the equation system. We expect the actual rate of return to affect the expected rate, yet in the expected rate-of-return equation (2.95), the variable rorga(r) does not affect rorge(r). Equally, we expect investment to affect the capital stock, yet in the capital accumulation equation (2.13), the variable qcgds(r) does not affect qk(r). The explanation is that these links do exist in the theoretical structure, but through data coefficients rather than variables. The level of
58
Elena I. Ianchovichina and Robert A. McDougall
the actual rate of return affects the coefficient ERRRORG(r), which appears in the expected rate-of-return equation and affects the variable rorge(r). Similarly, the level of gross investment affects the coefficient NETINV(r), which appears in the capital accumulation equation and affects the variable qk(r).
5.6 Alternative Investment Determination In some simulations, the user may wish to disable the investment theory described in the preceding subsections and instead impose specific investment targets. For example, he or she may wish to use investment forecasts from macroeconomic models or to model sudden (perhaps dramatic) fluctuations in investment, such as those observed during financial crises. Imposing investment targets on all regions is harder than it may seem at first. Through the identity that world saving is equal to world investment, it would implicitly impose a target on world saving. To accommodate that, the user would need to change the treatment of saving in the closure. In this section we consider a more limited objective – imposing targets on regional shares in world investment – while allowing the usual saving theory to determine its level.5 To enable this, we use an equation QCGDS(r) = SQCGDSREG(r) × SQCGDSWORLD,
(2.102)
representing investment in region r as the product of a region-specific factor SQCGDSREG(r) and a region-generic factor SQCGDSWORLD. Differentiating, we obtain qcgds(r) = sqcgdsreg(r) + sqcgdsworld,
(2.103)
where sqcgdsreg(r) and sqcgdsworld denote percentage changes in SQCGDSREG(r) and SQCGDSWORLD, respectively. In the TABLO code, this equation is given as Equation GDI # region specific determination of investment # (all,r,REG) qcgds(r) = sqcgdsreg(r) + sqcgdsworld. When we wish to target the regional allocation of investment, we exogenize sqcgdsreg and endogenize either srorge or SDRORT. At the same time, 5
GDyn inherits from GTAP the fixed propensity to save treatment, which is adequate in the medium to long-run (5 to 15 years), but is inadequate in terms of depicting saving behavior in the very long-run. Saving behavior in GDyn is discussed in detail in Chapter 14.
Theoretical Structure of Dynamic GTAP
59
we exogenize SDRORTW and endogenize sqcgdsworld, letting sqcgdsworld adjust so that global investment remains equal to global saving. If we wish to target the investment allocation in all periods, it does not matter whether we endogenize srorge or SDRORT. If, however, we wish to target it only in earlier periods, but let the investment theory drive it in later periods, then the choice of variable does matter. If in the earlier periods we endogenize srorge, the model achieves the investment targets by adjusting expected rates of return. In the later periods, with srorge exogenous, the expected rates converge toward the actual rates according to the usual GDyn theory. So under this treatment, the imposed investment allocation is transient. If, however, in the earlier periods we endogenize SDRORT, the model achieves the investment targets by adjusting target rates of return. In the later periods, with SDRORT exogenous, the differentials in the target rates remain in place unless and until we shock them back toward equality. So under this treatment, the imposed investment allocation is persistent.
6. Properties and Problems Having completed the presentation of the GDyn theoretical structure, we now discuss some properties of the system and issues arising in using it: r r r r r
existence and stability of long-run equilibrium cumulative and comparative dynamic results path dependence one-way relations capital account volatility and the propensity to save
6.1 Long-Run Equilibrium In the GDyn investment theory (Section 5), expected, target, and actual rates of return may all differ over the short-run. In the long-run equilibrium, these three rates are all equal and constant over time, as is also the normal growth rate for the capital stock: RORGEXP(r) = RORGTARG(r) = RORGROSS)(r), ∀r (2.104) ˙ ˙ ˙ RORGEXP(r) = RORGTARG(r) = RORGROSS(r) = 0, ∀r KHAT(r) = 0, DKHAT(r) = 0, ∀r
(2.105) (2.106)
60
Elena I. Ianchovichina and Robert A. McDougall
These conditions imply in turn a constant investment-capital ratio. They are the same conditions characterizing the equilibrium solution of a multiregion q-investment model with convex adjustment costs. Ianchovichina (1998) demonstrates the existence and stability of the long-run equilibrium. Here we provide a numerical illustration. We use a three-region aggregation of the GTAP 3 Data Base (McDougall 1997) featuring the United States, the European Union (EU), and all other regions aggregated into a Rest of World region (ROW). The initial data (1992) reveal regional differences in rates of return, RORGROSS(r) (Fig. 2.5); normal rates of growth in capital KHAT(r) (Fig. 2.7); investment-capital ratios (Fig. 2.8), as well as sizable errors in expectations ERRRORG(r) (Fig. 2.6). In short, the benchmark data depict a world in disequilibrium. We test the long-run properties of the model over a 100-year period. The simulation represents the changes in the three economies occurring solely due to the passage of time. It depicts the movement from the initial disequilibrium state toward a long-run equilibrium. For simplicity, we assume zero regional risk premia. Figure 2.6 suggests that in the GTAP 3 Data Base, investors underestimated returns to capital in the United States and the rest of the world and overestimated returns to capital in the European Union. As investors realize their errors in predicting these returns, they adjust their expectations in an upward direction in the case of the United States and the ROW region, and in a downward direction in the case of the EU (via equation [2.100]). As a result, investment in the United States and ROW increases, whereas investment in the European Union declines (via equations [2.86] and [2.83]). It takes approximately 12 years for the model to eliminate errors in expectations (Fig. 2.6) and interregional differences in rates of return (equilibrium condition [2.104]; Fig. 2.5). However, because KHAT(r) is neither zero nor a constant in 2004, this is only a temporary equilibrium. Positive and nonconstant KHAT(r) (Fig. 2.7) implies that the expected investment schedule (2.78) will overshoot the warranted one (2.88) and over time will start moving back. We observe this type of oscillating behavior in Figures 2.5–2.8 around 2004. Only after further reduction in KHAT via equation (2.101), leading to a reduction in the investment-capital ratio via equation (2.83), will the model permanently eliminate errors in expectations and differences in interregional rates of return. Figure 2.5 shows the convergence of the regional rates of return RORGROSS(r) toward the target rate RORGTARG(r), and Figure 2.6 shows the elimination of errors in expectations ERRRORG(r) over time. Figure 2.7 displays the normal rate of growth in the capital stock KHAT(r) in its
Theoretical Structure of Dynamic GTAP
61
14
12
10
6
4
2
USA
EU
ROW
2087
2082
2077
2072
2067
2062
2057
2052
2047
2042
2037
2032
2027
2022
2017
2012
2007
2002
1997
0 1992
Percent
8
TARGET
Figure 2.5. Actual and target rates of return. Source: Ianchovichina (1998).
movement toward 0 over the long-run, whereas Figure 2.8 demonstrates the process of adjustment toward constant investment-capital ratios. The four figures suggest that the stability conditions of the model are satisfied over time.
6.2 Cumulative and Comparative Dynamic Results GDyn is designed as a recursive dynamic model. To obtain projections through time, you run a sequence of simulations, one for each time period in the projection. To obtain comparative dynamic results, you run two sequences of simulations, one representing a base-case projection and the other representing a variant projection. From the period-by-period results, you then calculate cumulative results for each projection. Finally you find the difference between the two series of cumulative results to obtain comparative dynamic results.
62
Elena I. Ianchovichina and Robert A. McDougall
0.2
0.15
0.1
Level
0.05
2087
2082
2077
2072
2067
2062
2057
2052
2047
2037
2042
2032
2027
2022
2017
2012
2007
2002
1992
1997
0
-0.05
-0.1
-0.15 USA
EU
ROW
Figure 2.6. Errors in expectations. Source: Ianchovichina (1998).
The formulas used for calculating cumulative results and differences in results are different for the different kinds of variables distinguished in GEMPACK: change and percentage change. For a change variable dV, the cumulative change over two periods 1 and 2, dV02 = dV01 + dV12 ,
(2.107)
where the subscript 01 denotes changes between the start and end of period 1, and 12, denotes changes between the end of period 1 and the end of period 2. For a percentage change variable v, we have a more complex formula, v 02 = 100
v 12 v 01 1+ −1 . 1+ 100 100
(2.108)
This procedure works for most of the variables in the model, but not for all. In particular, it does not work for the equivalent variation, EV(r), and
Theoretical Structure of Dynamic GTAP
63
0.04
0.035
0.03
0.025
Level
0.02
0.015
0.01
0.005
-0.005 USA
EU
ROW
Figure 2.7. Normal rate of growth in capital stock, KHAT(r). Source: Ianchovichina (1998).
associated variables. The problem is that the variable is defined so that, in say the first period, the equivalent variation variable is EV01 = E (U1 , P 0 ) − E (U0 , P 0 ),
(2.109)
where E is the expenditure function, U is utility, P denotes prices, and the subscripts 0 and 1 refer to values at the beginning and end of the first period. In the second period, we have EV12 = E (U2 , P 1 ) − E (U1 , P 1 ).
(2.110)
Then the cumulative equivalent variation for the first and second periods is derived from EV02 = E (U2 , P 0 ) − E (U0 , P 0 ),
(2.111)
but we cannot calculate this from EV01 and EV12 . Thus we cannot calculate valid cumulative results for the equivalent variation nor, consequently, can we obtain valid comparative dynamic results. Similarly, we cannot calculate
2088
2084
2076
2080
2072
2068
2060
2064
2052
2056
2044
2048
2040
2032
2036
2028
2024
2016
2020
2012
2004
2008
2000
1992
1996
0
64
Elena I. Ianchovichina and Robert A. McDougall
0. 0.1 0.09 0.08 0.07
0.05 0.04 0.03 0.02 0.01
USA
2088
2080
2084
2076
2068
2072
2064
2056
2060
2052
2044
EU
2048
2036
2040
2032
2028
2020
2024
2016
2008
2012
2004
1996
2000
0 1992
Ratio
0.06
ROW
Figure 2.8. Investment–capital ratio. Source: Ianchovichina (1998).
valid comparative dynamic results for the equivalent variation decomposition (Huff and Hertel 1996). This does not mean that we cannot obtain comparative dynamic results for equivalent variation. To obtain them, however, we need some computational machinery beyond the cumulating and differencing procedures used for other variables. The calculation of welfare in GDyn is discussed in Chapter 6 of this book.
6.3 Path Dependence GDyn is inherently a path-dependent model. That is, in GDyn, the effects of changes in exogenous variables depend not only on the overall changes in exogenous variables but also on the paths followed by them. When GDyn is used in a dynamic mode by shocking the time variable, the effects of
Theoretical Structure of Dynamic GTAP
65
economic shocks depend not only on the size but also on the timing of the shocks. Path dependence is built into the theory in three places: wealth accumulation (subsection 4.3), the partial adjustment treatment of the capital stock (subsection 5.1), and the adaptive expectations treatment of the expected rate of return and the normal growth rate (subsections 5.3 and 5.4). In GDyn a region’s wealth depends on its past history; it cannot be determined from other current variables, such as income. The final level of regional wealth in any simulation depends on the original level and on the time paths of the exogenous variables within the simulation. For example, technological progress in a given region normally leads to an increase in its wealth, but the increase in wealth is greater if the technological progress occurs mostly near the beginning of the period than if it occurs mostly toward the end. Other path dependencies arise from the lagged adjustment treatment of the capital stock and from the adaptive expectations treatment of investment. Similarly, regional capital stocks cannot be inferred from other current variables for two reasons. Globally, the money value of net physical investment is equal to saving, so the money value of the global capital stock is determined by wealth accumulation (and capital gain), not by an equilibrium condition. In addition, the distribution of capital across regions is given not by an equilibrium condition, but by a partial adjustment process, as described in subsection 5.1. Investors do redistribute capital to equalize rates of return, but only gradually. Past shocks therefore affect the current international distribution of capital more if they occur in the more distant past and less if they occur in the more recent past. Finally, the level of investment depends not on the actual rate of return but on the expected rate. The expected rate of return cannot be inferred from other current variables, but adjusts toward the actual rate with a lag, as described in subsection 5.3. Here then is yet another adjustment process whose results depend not only on the size of the changes in its inputs but also on their timing. Given GDyn’s objectives, this path dependence must be construed not as a bug but a feature. Indeed, if we should extend GDyn to provide a better treatment of short-run dynamics, bringing in more macroeconomic content such as that found in such models as G-CUBED or FAIR, path dependencies will become more pervasive rather than less. In short, path dependence in GDyn is here to stay. Nevertheless, and this is why one might be tempted to construe it as a bug, path dependence imposes some practical inconveniences. It places on the
66
Elena I. Ianchovichina and Robert A. McDougall
user an onus to represent accurately the time paths of exogenous variables, in circumstances where doing so would otherwise be unnecessary. Users need to take it into account in several places in their computational strategy. First, you need to set periods, within the overall projection time interval, to capture sufficient detail about the time profile of the shocks. With the continuous-time approach used in GDyn, you can run, say, a tariffreduction scenario over a single 10-year interval and get sensible and meaningful results. If, however, you want the tariff cuts to be back-loaded and not implemented at an even pace, then you need to use several shorter intervals, so that you may specify lower rates of tariff reduction in the earlier intervals and higher rates in the later intervals. Second, even if you wish to apply shocks evenly through time, you may wish to avoid long time intervals, if you do not like the rule TABLO uses to distribute shocks between steps. TABLO-generated programs distribute shocks so that the change in the levels variable is the same for all steps (Harrison and Pearson 1998, see chapter 4, “GEMSIM and TABLOGenerated Programs”). To take an extreme example, if you shock a variable by 300%, using a two-step solution procedure, your TABLO-generated program shocks the variable by an amount equivalent to 150% of the initial level in each step; that is, by 150% in the first step (going from 1 to 2.5 times the initial level), and 60% in the second step (going from 2.5 to 4 times the initial level). For most percentage change variables, in most applications, a more appealing default assumption is that the percentage change in the variable is constant across steps. For example, it is more natural to assume that the population grows at a constant rate through time (for example, by 1% per year) than that it changes at a constant rate (for example, by 200,000 persons per year). Likewise, in the extreme example given previously, we would typically prefer by default to shock the variable by 100% in each step, rather than by 150% in the first step and 60% in the second. GEMPACK wizards may perhaps know some way to coerce TABLO to use equal percentage changes. No such way, however, is apparent from the published documentation. The shock-splitting rule does not much matter when the shocks are small, but it does matter when they are large. More specifically, it matters when the total shock in a simulation is large, even if the shock is broken up into small pieces in individual steps. One way to work around the problem is to avoid long intervals always, even if all shocks are evenly distributed through time. Finally, path dependence rules out some common closure-swapping strategies. In GEMPACK, a common expedient is to let the model determine
Theoretical Structure of Dynamic GTAP
67
the change in some instrument variable required to achieve a given change in a target variable, by making the naturally exogenous instrument variable endogenous, and the naturally endogenous target variable exogenous. For example, we may determine the rate of technological progress required to achieve a given improvement in welfare by endogenizing the technological change variable and exogenizing the welfare variable. If we then run a second simulation, using the natural closure and shocking the technological change variable according to the results from the first simulation, we get – with a path-independent model – the same results as in the first simulation. We can then investigate the effects of changes in other elements of the scenario on welfare as on other variables, using the natural closure and the calibrated technological change shock. With a path-dependent variable this approach does not work. The trouble is that the path of the technological change variable is different in the two simulations. In the second simulation, technology changes evenly throughout the simulation interval; in the first, it changes so as to keep GDP moving evenly through the interval. This difference is liable to affect the simulation results. In GDyn, for example, a front-loaded improvement in technology has more effect on end-of-interval wealth than the same total improvement distributed evenly through the interval. What we need (but, at the time of writing, do not have) for this problem is an automated algorithm for finding the constant rate of growth in an instrument variable (or the constant rates of growth in a set of instrument variables) that achieves given total growth in a target variable (or a set of target variables). Such a tool would be useful not only for single-simulation but also for multi-simulation projections. For example, in a projection made up of five 2-year simulations, each involving different tariff shocks, we would like to be able to find the constant rate of technological progress, through the complete 10-year projection interval, required to achieve a given welfare improvement over the interval. Such a multi-simulation facility would be useful even with path-independent models.
6.4 One-Way Relations A novel emergent feature of GDyn, relative to standard GTAP, is the appearance of what we here describe as one-way relations. In standard GTAP, as perhaps in most GEMPACK-implemented models, if an exogenous variable A affects an endogenous variable X, we can swap A and X in the closure and determine in a simulation the change in A required to bring about a given change in X. Of course, this may not work if X is not monotonic in A, but it works most of the time, and of course, it always works for sufficiently small
68
Elena I. Ianchovichina and Robert A. McDougall
changes in X if X is locally nonstationary in A. In GDyn, however, it may easily fail. That is, there are relations between variables A and X such that the solution program can determine the change in X arising from a given change in A, but not the change in A required to bring about a given change in X – no matter how well behaved mathematically the relation is between A and X. These one-way relations appear when one variable affects another not through the equation system but through data updates. For example, investment, of course, affects the capital stock, yet the investment variable (qcgds) does not appear in the relevant equation (2.13). Instead, the equation contains the investment coefficient NETINV. In a single-step simulation, qcgds has no effect on qk. In a multistep simulation, it affects NETINV at each update of the data files and thereby affects qk. Now consider the effect of a shock to a normally exogenous, investmentrelated variable, for example, the target rate shift variable SDRORT. In each period, this leads through the investment module to some change in the investment variable qcgds, but that change in qcgds has no effect in the current period on qk. If we try to change the closure to find the SDRORT value consistent with a given change in qk, we find it impossible to do so. If we exogenize qk(r) for some region r, and endogenize SDRORT(r), we make the model singular, because we thereby make exogenous all variables in equation (2.13). The only natural way to exogenize qk(r) is to swap it with sqk(r), and that does not achieve the larger purpose, because it does not allow qk(r) to determine qcgds(r) or SDRORT(r). Another one-way relation is that between the actual rate of return RORGROSS and the expected rate RORGEXP. According to the GDyn theory, changes in RORGEXP cause changes in RORGROSS, yet rorga does not appear in the equation for rorge (2.95). In a single-step simulation, indeed, rorga has no effect on rorge. In a multistep simulation, however, it affects ERRRORG at each data file update and thereby affects rorge. If you shock some exogenous variable so as to increase the actual rate of return – if, for example, you apply a positive shock to labor supply qo(“labor”,r) – this has no effect on the expected rate rorge(r) in a singlestep simulation, but does affect it in a multistep simulation. However, if you want to find the labor supply change needed to achieve a given change in the expected rate of return, you find that you cannot exogenize rorge(r) and endogenize qo(“labor”,r). To do so would create a singular system: The twoequation subsystem comprising the capital accumulation equation (2.13) and the expected rate-of-return equation (2.95) for region r would contain only one endogenous variable, qk(r). The only natural way to exogenize
Theoretical Structure of Dynamic GTAP
69
rorge(r) is to endogenize srorge(r), and this does not achieve the purpose of determining labor supply endogenously. Given that closure swaps do not work at all across these one-way relations, we evidently need some new computational machinery to let us target the naturally endogenous variables in them.
6.5 Capital Account Volatility and the Propensity to Save GDyn inherits from standard GTAP its specification of the regional household demand system and, in particular, the treatment of saving. As in standard GTAP, there is a fixed average propensity to save. In other words, saving is a fixed proportion of income in each region. One unwelcome implication is that the capital account and net foreign liabilities are highly volatile and can grow without bound in GDyn simulations. In the real world, for reasons that are poorly understood, saving and investment are highly correlated across countries, and international capital flows are much smaller and more stable than simple theory would suggest (Feldstein and Horioka 1980). In GDyn, we do not impose any such correlation, so relatively modest economic shocks can lead to unrealistically large international capital flows, as well as unrealistically large changes in regions’ net foreign liabilities. The second problem is that as economies with high savings rates, like China, grow, there is a glut of global savings and, as a result, of investment and capital in the world. Because of excessive investment, rates of return to capital fall without bound. This problem prevents us from running simulations with GDyn over the very long-run. Chapter 14 evaluates the behavior of foreign assets and liabilities in GDyn, comparing it to historical data on these indicators, and proposes a technique to address this shortcoming of the model.
7. Concluding Remarks This chapter presented a set of new equations added to the GTAP model to construct GDyn, a dynamic AGE model of the world. The new theory offers a disequilibrium approach to modeling endogenously international capital mobility in a dynamic applied general equilibrium setting, and it takes into account stock-flow dynamics and foreign asset income flows. The method can be especially attractive to policy modelers as it permits a recursive solution procedure, a feature that allows easy implementation of dynamics into any static AGE model without imposing limitations on the model’s size.
70
Elena I. Ianchovichina and Robert A. McDougall
Key to the proposed approach is investors’ adaptive expectations about potential returns to capital. This type of expectation emphasizes errors in investors’ assessment of potential returns to capital – such as those observed during financial crises. It can also be shown that it ensures the convergence of the model toward a stable equilibrium and offers the flexibility of tailoring the model to observed data. Despite some limitations of GDyn, such as the lack of equity-for-debt substitution, the absence of bilateral detail, and the lack of forward-looking behavior, the model offers a unique and simple treatment of international capital mobility in a dynamic AGE context. It captures endogenously the economy-wide effects of capital and wealth accumulation and the income effects of foreign property ownership. References Feldstein, M. and C. Horioka. 1980. “Domestic Saving and International Capital Flows.” Economic Journal 90(358), 314–29. Harrison, W. J. and K. R. Pearson. 1998. An Introduction to GEMPACK. GEMPACK Document No. GPD-1 (4th ed.). Melbourne, Australia: Centre of Policy Studies and Impact Project, Monash University. Hertel, T. W. (ed.). 1997. Global Trade Analysis Modeling and Applications. Cambridge: Cambridge University Press. Hertel, T. W. and M. E. Tsigas. 1997. “Structure of GTAP.” In T. W. Hertel (ed.), Global Trade Analysis Modeling and Applications (pp. 13–73). Cambridge: Cambridge University Press. Huff, K. and T. W. Hertel. 1996. Decomposing Welfare Changes in GTAP. Technical Paper No. 5, Center for Global Trade Analysis. West Lafayette, IN: Purdue University. Ianchovichina, E. I. 1998. International Capital Linkages: Theory and Application in a Dynamic Computable General Equilibrium Model. Ph.D. thesis, Department of Agricultural Economics, Purdue University. Kapur, J. N. and H. K. Kesavan. 1992. Entropy Optimization Principles with Applications. New York: Academic Press. McDougall, R. A. (ed.). 1997. Global Trade, Assistance, and Protection: The GTAP 3 Data Base. Center for Global Trade Analysis. West Lafayette, IN: Purdue University. McDougall, R. A., A. Elbehri, and T. P. Truong (eds.). 1998. Global Trade, Assistance, and Protection: The GTAP 4 Data Base. Center for Global Trade Analysis. West Lafayette, IN: Purdue University. McDougall, R., M. E. Tsigas, and R. Wigle. 1997. “Overview of the GTAP Data Base.” In T. W. Hertel (ed.), Global Trade Analysis Modeling and Applications (pp. 74–124). Cambridge: Cambridge University Press.
THREE
Behavioral and Entropy Parameters in the Dynamic GTAP Model Alla Golub and Robert A. McDougall
1. Introduction The dynamic theory in Chapter 2 describes various new parameters governing international capital mobility. This chapter examines what we can learn from country-panel data about the magnitude of these additional parameters, corresponding calibration procedures, and the manipulation of the parameters with an aggregation program. The new parameter file containing parameters used in the dynamic theory is a GEMPACK header array file; its contents are listed in Table 3.1. The first new parameter listed in Table 3.1 is INC. This parameter is the initial income level across simulations and is used to calculate welfare measures in multiperiod simulations. It is represented in the unit US$ millions. The rest of the parameters in Table 3.1 can be grouped according to their role in the model: lagged adjustment parameters, elasticity of rate of return to capital with respect to capital stock, and parameters determining the allocation of regional wealth and composition of regional capital. These three parameter types are discussed in turn in this chapter.
2. Parameters Determining Lagged Adjustments The investment theory presented in Chapter 2 is expressed in terms of gross rather than net rates of return, and it allows only zero or positive gross rates of return. However, net of depreciation, rates of return may be negative, and they may decline to the negative of the depreciation rate. The long-run equilibrium in the GDyn model is defined as the convergence of the net rates of return to capital stock across regions. If region-specific risk premia are We are thankful to Thomas Hertel, Terrie Walmsley, Ken Foster, Elena Ianchovichina, and Paul Preckel for their valuable comments and suggestions.
71
72
Alla Golub and Robert A. McDougall Table 3.1. Contents of the dynamic parameters file
Coefficient name
Dimensions
Description
INC LAMBKHAT
REGa REG
LAMBRORGE LAMBRORG RORGFLEX
REG REG REG
RIGWQH
REG
RIGWQ_F
REG
initial income coefficient of adjustment in estimated normal growth rate coefficient of adjustment in expected rate of return coefficient of adjustment in rate of return elasticity of rate of return to capital with respect to capital stock rigidity of allocation of wealth by regional household rigidity of source of funding of enterprises
a
REG denotes number of regions.
allowed, then the long-run equilibrium in the model is defined as the convergence of risk-adjusted net rates of return to capital stock across regions. In the absence of risk premia, and if the depreciation rates are the same across regions, the convergence of net rates of return guarantees the convergence of gross rates of return. In this section, we construct cross-country time-series data on net rates of return to capital to test the convergence hypothesis and determine the speed of convergence in rates of return across countries. We then use the results to set the lagged adjustment parameters in the model in accordance with the observed behavior.
2.1 Data As in the standard GTAP model, the GDyn model is a real assets model; that is, there is no financial market. The gross rate of return to capital for each country is defined as the ratio of gross operating surplus to the capital stock, and the net rate of return to capital is the ratio of net (of depreciation) operating surplus to capital stock.1 To determine parameters that will quantify the degree of capital mobility in the GDyn model, the rates of return to capital are constructed in accordance with these definitions. Net rates of return to capital are often used to compare companies’ profitability across countries. Walton (2000a, 2000b) uses the net rate of return to capital to compare profitability of the corporate nonfinancial sector in the United Kingdom with profitability of the corporate nonfinancial sector in other countries. In these studies the rates of return were constructed using 1
In both definitions, the capital stock in the model is net capital stock.
Behavioral and Entropy Parameters in the Dynamic GTAP Model
73
data available in the national accounts of 19 countries. These rates of return are rather sparse for our purposes (the time period covered is too short for econometric investigation and the panel is unbalanced) and represent returns for nonfinancial corporations only, whereas rates of return to capital in the GDyn model represent the overall profitability of the economy. Therefore these studies are considered here for illustrative purposes only. There are two important features of the data highlighted in Walton (2000a, 2000b) that we should keep in mind when choosing data for our analysis. First, annual rates of return are calculated as the ratio of the operating surplus to capital employed. Profits, the main source of the operating surplus, are defined fairly precisely and measured reasonably consistently. However, capital employed is not defined as precisely; the definitions and methods used to estimate capital stock vary from country to country. Most of the national statistics data on capital are compiled using the perpetual inventory method (PIM), which is discussed in detail later. This method involves adding gross fixed capital formation to, and deducting consumption of fixed capital from, an initial estimate of capital stock. The variations come from the estimates of useful service lives by capital type and country and are influenced by the business cycle and technological change. As a result, the differences between countries’ rates of return, constructed using national account data, can reflect both differences in profitability and differences in calculation methods. Because countries have estimated profitability consistently over time, relative changes in net rates of return should reflect real changes in their economies and hence could be used in the cross-country profitability comparisons undertaken in Walton (2000a, 2000b). However, to test for convergence of rates of return to capital, we need a capital stock series that is constructed using uniform (across countries) assumptions. In addition, not many countries have data on profitability and/or capital stock, and time coverage varies from country to country (Walton 2000b). Our first step in constructing rates of return is to define profits associated with the use of capital stock. Using the income approach to gross domestic product, GDP can be represented as a sum of value added at factor costs plus indirect taxes. The value added at factor costs consists of labor earnings, capital earnings, and land earnings. Although time-series data on value added at factor costs with good country coverage are available from many sources (for example, the World Development Indicators (WDI) database supported by the World Bank), the labor earnings data are problematic. For this reason, the analysis begins with a set of countries for which these data are most readily available – the Organization for Economic Cooperation and Development (OECD) countries.
74
Alla Golub and Robert A. McDougall
Time-series data on the gross operating surplus at current prices are obtained from the SourceOECD database, Annual National Accounts Volume II – Detailed Tables – Main Aggregates Volume 2004 release 01 (SourceOECD). This database covers detailed national account data for most OECD countries, including components of value added, from 1970 to the present. It lists four components of GDP: (1) compensation of employees; (2) taxes less subsidies on production and imports; (3) gross operating surplus and gross mixed income; and (4) a statistical discrepancy, which is small or zero for most of the countries considered. The sum of the gross operating surplus and gross mixed income is used as a proxy for capital earnings. Note that this measure overestimates capital earnings because it includes land earnings, returns to natural resources, and that portion of self-employed labor earnings that is not accounted for by imputed wages. Land earnings should not be a big problem because they are expected to be small relative to capital earnings in developed countries. However, the potential inclusion of self-employed labor may result in a larger error in the capital earnings measure, but this error is expected to be much smaller in the OECD countries than in developing countries. The gross operating surplus measure also includes depreciation of capital stocks. As noted earlier, the convergence in GDyn is modeled as convergence in net rates of return. To test for convergence of net rates of return, we will construct net operating surplus measures. The second step in constructing rates of return to capital is to define capital stocks. Several alternative sources for capital stock data for the OECD countries could be used. In all these sources the capital stock estimates are derived using the perpetual inventory method (PIM). The first data source for capital stock data for the OECD countries is the OECD itself. Until 1997, OECD published annual data in a report titled Flows and Stocks of Fixed Capital (Statistics Directorate OECD, various years).2 However, in 1997, production of these data ceased because of the move to the new system of national accounts. Some countries are now starting to produce these data again, but not in a sufficient amount for the OECD to resume publishing them. The data are available for only a few of the OECD countries and cover different time periods for different countries. The data come from national account statistics, and as noted earlier, the assumptions made to construct these series differ from country to country.
2
The PIM and the estimation procedures used by the OECD countries are described in the manual issued by OECD (OECD 1993).
Behavioral and Entropy Parameters in the Dynamic GTAP Model
75
A second data source is Larson et al. (2000), who constructed capital stock time-series data for 62 industrial and developing countries for the period 1967–92. The main objective of this database was to provide sectoral and economy-wide capital stock data for countries both within and outside of the OECD. The same method was used in the calculations to facilitate comparisons across countries. Larson et al. (2000) constructed fixed-capital series based on national account investment data, using a modified version of the PIM. A third data source for capital stock data is Nehru and Dhareshwar (1993), who constructed capital stock time-series data for 92 developing and industrial countries from 1960 to 1990. Because the OECD capital stock data are sparse and constructed using assumptions that differ from country to country, we eliminated this source from consideration and made a choice between the Larson et al. (2000) and Nehru and Dhareshwar (1993) databases. The PIM used in the construction of these databases can be generalized in the following equation: K t = s t I t + s t−1 I t−1 + · · · + s t−L I t−L ,
(3.1)
where Kt is capital stock at the end of year t, It is investment made during year t, L is the lifetime of the capital good, t − L is the vintage of the oldest surviving capital asset, and sj is the productivity of investment of age j, 0 < s j < 1 for 0 < j < L; s0 = 1 and sj = 0 for j ≥ L. The main difference between the Larson et al. (2000) and Nehru and Dhareshwar (1993) methods for constructing the databases is in the assumptions made about the path of sj . Nehru and Dhareshwar (1993) assume that sj follows a geometric decay pattern with the rate of decay fixed at 4%, which is equivalent to the assumption of an infinite lifetime L of capital assets and a 4% decline in productivity every year. The method used to define productivity in Larson et al. (2000) is more general and closer to the one used in the OECD data. To restrict productivity to be non-negative, this method assumes a finite lifetime L of capital assets and a curvature parameter β bounded from above by 1, as shown in the following equation (Larson et al. 2000): s j = (L − j )/(L − βj ),
0 ≥ j ≥ L.
(3.2)
To generate economy-wide capital stocks from investment data, Larson et al. (2000) set β = 0.7 and defined service life L as a stochastic variable with a mean of 20 and standard deviation of 8 years. Analysis of equation (3.2) shows that productivity falls with the age of assets, and when β is positive but less than unity, the depreciation accelerates with the time of asset use. Figure 3.1 illustrates these points and the differences between productivity paths assumed in Nehru and Dhareshwar (1993) and Larson et al. (2000).
76
Alla Golub and Robert A. McDougall
1 0.9
Relative productivity
0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0
2
4
6
8
10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 Time in use
Nehru and Dhareshwar (1993)
Larson et al. (2000)
Figure 3.1. Comparison of the relative productivity paths in Nehru and Dhareshwar (1993) and Larson et al. (2000). The geometric decay path in Nehru and Dhareshwar (1993) is based on a 4% decay rate. Larson et al. (2000) built the path of productivity of total economy-wide fixed capital assuming 20 years of service life and curvature parameter β = 0.7. Source: Authors’ calculations.
To construct a capital stock series, an assumption about the initial value of capital is needed. Many techniques to seed the initial values are discussed in Nehru and Dhareshwar (1993) and Larson et al. (2000). However, if the investment series are sufficiently long, and given that the productivity of old capital is low, the contribution of old capital to the current capital stock should be small. This view is supported by the analysis of sensitivity of constructed capital stock series with respect to the choice of initial values in Larson et al. (2000). Because the Larson et al. (2000) assumptions to measure economy-wide capital stocks are more realistic, and capital stock series constructed using their method are less sensitive to the choice of initial value, we chose this database to construct rates of return to capital. The choice of countries and years to be included in the analysis is dictated by the availability of data in both the OECD records on gross operating surplus and the capital stock data in the Larson et al. (2000) database. To construct a net operating surplus, depreciation should be subtracted from the gross operating surplus. Data
Behavioral and Entropy Parameters in the Dynamic GTAP Model
77
on depreciation are available from SourceOECD; however, these data are not consistent with the depreciation assumed in the calculations of net capital stocks in Larson et al. (2000). To calculate a net operating surplus that is consistent with the net capital stock data, the depreciation Dt is recovered from the Larson et al. (2000) time series using the following formula: D t = K t−1 − K t + I t .
(3.3) 3
It is then subtracted from the gross operating surplus. To calculate the values of depreciation (Dt ), current capital (Kt ), investment (It ), and capital in the previous year, (Kt −1 ) should be measured in constant prices. The capital stock data in Larson et al. (2000) are given in 1990 US$, whereas investments are given in 1990 local currency units (LCU). A real exchange rate et = 1/Et (P tUS /p t ) is constructed, where Et is the nominal exchange rate in US$ to local currency, P tUS is the dollar deflator, and pt is the domestic deflator.4 Using the real exchange rate, capital stocks are converted from 1990 US$ to 1990 LCU, and depreciation in 1990 LCU is calculated. Using the dollar deflator, capital stock and depreciation are converted into current US$. The gross operating surplus is also converted from current local currency to current dollars using SourceOECD exchange rates.5 The net operating surplus in current dollars is calculated by subtracting depreciation from the gross operating surplus, and then the net rates of return are calculated as a ratio of net operating surplus and net capital stock for 20 OECD countries from 1970 to 1992.6 The net rates-of-return series are shown in Figure 3.2. To simplify the figure, the rates of return are shown for only nine OECD countries. The time series of the other 11 countries are in the range between the rates of return in Portugal and Finland. The rates of return are very high in the beginning of the time period in Turkey, Greece, South Korea, and Portugal, but then decrease. This feature concurs with our expectations that the rates of return 3
4
5
6
For some reason, the calculated depreciation in 1980 is negative for all countries in the Larson et al. (2000) database. For each country, we use the arithmetic average of 1979 and 1981 depreciations for the 1980 depreciation. Investments series and all determinants of the real exchange rate are given in Larson et al. (2000), including the exchange rate, which is the market exchange rate from the International Monetary Fund. For European countries adopting the Euro, the SourceOECD database lists gross operating surplus in Euros. For this reason, SourceOECD exchange rates of US$ to local currency are different from ones obtained from the International Monetary Fund. These countries are Australia, Austria, Belgium-Luxemburg, Canada, Denmark, Finland, France, Greece, Ireland, Italy, Japan, Netherlands, New Zealand, Norway, Portugal, South Korea, Sweden, Turkey, United Kingdom, and the United States.
78
Alla Golub and Robert A. McDougall
60%
Net rate of return to capital
50%
Finland Greece Japan Portugal South Korea Sweden Turkey United Kingdom United States
40%
30%
20%
10%
0% 1970 1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
-10% Year
Figure 3.2. Net rates of return to capital in OECD countries. Source: Authors’ calculations.
to capital are higher in the least developed countries because capital is a scarce resource. As these countries’ economies grow, capital expands and its marginal product falls, and the rates of return to capital decline. Note that in Finland in 1991, the return to capital is negative, although small in absolute magnitude. This may be because the Larson et al. (2000) method overestimates the depreciation, or possibly it is negative simply because of a decline in capital earnings.
2.2 Convergence of Rates of Return to Capital Convergence of different productivity measures is a popular topic (Bernard and Jones 1996a, 1996b; Nin et al. 2004). This section draws on the econometric techniques used in these studies to focus on the question of convergence in rates of return. Assuming the absence of risk premia, the long-run equilibrium in the GDyn model is defined as the convergence of the ratios of capital earnings to capital stock across regions. Thus an important research question is whether these measures actually converge. An initial look at the dispersion of the rates of return to capital across countries in Figure 3.2 shows that it does appear to decline over time. Time-series evidence can also be used to examine the convergence of the rates of return by applying
Behavioral and Entropy Parameters in the Dynamic GTAP Model
79
the test for unit roots in panel data. With such a short time series, unit root testing for pairs of countries would appear to be out of the question; however, the technique of testing unit roots in panel data would be appropriate (Bernard and Jones 1996b; Levin and Lin 1992). In conducting the unit root test, the United Kingdom is chosen as a benchmark country, and deviations from the United Kingdom’s rate of return for 19 OECD countries are constructed. Consider the following general model: rr it = μi + ρ · rr it−1 + εit ,
(3.4)
where rrit is the difference between country i and the benchmark country rates of return, with error term εit ∼ iid (0, σε2 ) and drift μi ∼ iid(μ, σμ2 ). Let ρˆ and t be the OLS parameter estimate and t-statistic from the regression with equation (3.4) above, respectively. Bernard and Jones (1996b) show that, under the null hypothesis of a unit root and nonzero drift, t approaches the standard normal distribution. We are testing the null hypothesis of no convergence, which means that the deviation of the rate of return to capital from a benchmark country is a nonstationary process with nonzero drift. The alternative hypothesis is that rates of return to capital are converging in the sense that deviations of rates of return to capital from the benchmark country across countries are stationary processes. Table 3.2 reports estimates of country-specific drifts μi , together with results of the test ρˆ = 1. The results show that ρˆ is significantly less than 1, providing evidence against the null hypothesis of no convergence. Because all considered countries are developed countries, this result is expected and similar to what was found in the literature on convergence of productivity for OECD countries (Bernard and Jones 1996b). The estimate of ρˆ reported in Table 3.2 implies a convergence rate of 9% per year for the net rates of return to capital in the OECD countries. It is also important to mention that model (3.4) allows for countryspecific intercepts. For 16 of the 19 (without benchmark) countries, the intercepts are not statistically different from zero. This result has implications for the target rate and closure in the dynamic model: If depreciation rates are assumed to be equal, a common target rate can be set in the initial database for 17 (16 plus the United Kingdom) countries. For three other countries – Turkey, Ireland, and Greece – the intercept is different from zero. One possible interpretation is that, even if the deviations from the benchmark are stationary, there is a nonzero long-run value of the deviation, which may, in turn, suggest country-specific target rates for these three countries in the GDyn. Another possible explanation is that we have too few
80
Alla Golub and Robert A. McDougall Table 3.2. Time-series test for convergence of rates of return to capital
for 20 OECD countries∗
Parameter Australia (μ1 ) Austria (μ2 ) Belgium-Luxemburg (μ3 ) Canada (μ4 ) Denmark (μ5 ) Finland (μ6 ) France (μ7 ) Greece (μ8 ) Ireland (μ9 ) Italy (μ10 ) Japan (μ11 ) Netherlands (μ12 ) New Zealand (μ13 ) Norway (μ14 ) Portugal (μ15 ) South Korea (μ16 ) Sweden (μ17 ) Turkey (μ18 ) United States (μ20 ) ρ R-square Test ρ = 1 ∗
Estimate
Std. Error
t-value
Pr > |t|
0.001 −0.002 0.003 0.002 −0.001 −0.004 −0.001 0.010 0.007 0.002 −0.003 0.000 0.003 0.000 0.003 0.005 −0.002 0.017 0.003 0.907 0.984 5.410
0.003 0.003 0.003 0.003 0.003 0.003 0.003 0.004 0.003 0.003 0.003 0.003 0.003 0.003 0.003 0.004 0.003 0.006 0.003 0.017
0.320 −0.880 0.920 0.600 −0.450 −1.300 −0.240 2.280 2.480 0.850 −1.110 0.160 0.940 0.130 0.860 1.280 −0.610 2.880 1.180 52.900
0.749 0.382 0.361 0.546 0.651 0.195 0.810 0.023 0.014 0.395 0.269 0.870 0.350 0.899 0.388 0.200 0.543 0.004 0.240 <.0001
<.0001
United Kingdom is a benchmark country.
data points to observe complete convergence, or zero long-run deviation from the benchmark, for these countries.
2.3 Calibration of Lagged Adjustment Parameters The degree of capital mobility is defined as the speed with which differences in the risk-adjusted rates of return across regions are eliminated. Low speeds of convergence imply smaller changes in capital flows and slower capital reallocation. The time-series analysis of the convergence of rates of returns in the previous section can be used to calibrate parameters in the model to mimic the degree of capital mobility observed in the data. In this section we describe the calibration procedure. The disequilibrium mechanism determining the regional supply of investments and the path of capital stocks in the model consists of three
Behavioral and Entropy Parameters in the Dynamic GTAP Model
81
lagged adjustments: (1) adjustment of the expected rate of return toward the actual rate of return, (2) adjustment of the expected rate of return toward the target rate of return, and (3) adjustment of perceived-by-investors normal growth in capital stock toward the actual normal growth rate. The speed of each adjustment depends on the corresponding parameter: LAMBRORGE, LAMBRORG, and LAMBKHAT (see Table 3.1). The higher the magnitude of these parameters, the faster is the adjustment process toward equilibrium in the model. In the short- and medium-run, larger LAMBRORG leads to larger changes in the required rate of growth in the rate of return, larger LAMBRORGE leads to larger changes in the expected rate of return, and larger LAMBKHAT leads to larger changes in the perceived-by-investors normal rate of growth in capital stocks KHAT. As a result, larger magnitudes for any of the adjustment parameters are reflected in more volatile investments, capital, and GDP. In the long-run, the differences between the target and expected rate of return disappear, the expected and actual rates of return converge, and KHAT is equal to actual growth in capital stock; therefore, in the long-run, the magnitudes of the lagged adjustment parameters do not matter.7 Whereas the first two adjustment mechanisms are introduced into the model to reflect investor behavior, the third mechanism is built in for convenience and may be viewed as a substitute for calibration. Consider a specific scenario wherein we know that in the long-run, the normal rate of growth in capital stocks is zero. In such situations we can turn off the mechanism permitting adjustment of investors’ perceived normal growth in capital stock toward the actual normal growth rate and set the initial KHAT to zero. Ianchovichina (1998, p. 80) considered the stability properties of the GDyn model and found that, the lower the magnitude of the parameter LAMBKHAT, the higher the length of run over which the model is stable. Because the role of the parameter LAMBKHAT is different from that of the parameters LAMBRORGE and LAMBRORG, and because higher magnitudes of LAMBKHAT may affect the stability of the model, we set LAMBKHAT to some small number and manipulate LAMBRORGE and LAMBRORG to achieve the desired speed of convergence. In the following example, based on a 3 × 3 aggregation of the GTAP 5.4 Data Base (Dimaranan and McDougall 2002), we show how we calibrate parameters to achieve the desired degree of capital mobility in the model. 7
In the long-run equilibrium, the perceived-by-investor normal growth rate in capital stock KHAT is equal to the actual normal growth rate and to actual growth in capital stock qk(r).
82
Alla Golub and Robert A. McDougall
0.13 0.12 0.11 0.1 0.09 0.08 0.07 0.06 0.05 0.04 1995
2000
2005 1 NAM
2010 2 EUN
2015
2020
3 ROW
Figure 3.3. Convergence of net rates of return to capital when LAMBRORG = 0.4 and LAMBRORGE = 0.4. Source: Authors’ simulations with the GDyn model.
In this aggregation there are three regions – North America (NAM), European Union (EUN), and the Rest of World (ROW) – and three sectors (food, manufacturing, and services). Initially we set LAMBRORG and LAMBRORGE to 0.4 and LAMBKHAT to 0.2.8 Parameters LAMBRORGE and LAMBRORG are set equal because they determine speeds of similar adjustments in the model − convergence of the expected rate to the target rate of return, and convergence of the expected rate to the actual rate of return. In the case of no risk premia and a uniform across-regions depreciation rate, net rates of return to capital converge to the (net of depreciation) target rate. The convergence of net rates of return is shown in Figure 3.3. To determine the speed of convergence in this simulation, we estimate equation (3.4) using simulated net rates of return over 22 years. We define ROW as the benchmark country and define rrNAM t as the difference between the rates of return in NAM and ROW and rrEUN t as the difference between the rates of return in EUN and ROW. We estimate equation (3.4) in deviations of rates of return from the benchmark country. The estimated speed of convergence is 6% per year. To achieve a greater degree of capital mobility we gradually increase LAMBRORG and LAMBRORGE, keeping 8
We keep LAMBKHAT at 0.2 in the subsequent simulations for the reasons discussed earlier.
Behavioral and Entropy Parameters in the Dynamic GTAP Model
83
0.13 0.12 0.11 0.1 0.09 0.08 0.07 0.06 0.05 0.04 1995
2000
2005 NAM
2010 EUN
2015
2020
ROW
Figure 3.4. Convergence of net rates of return to capital when LAMBRORG = 0.5 and LAMBRORGE = 0.5. Source: Authors’ simulations with the GDyn model.
them equal. It is important to keep in mind that the expected rates of return in the initial database are a function of the parameter LAMBRORG. For every new value of LAMBRORG, we recalculate expected rates of return in the initial database. For this specific aggregation, with LAMBRORG and LAMBRORGE set to 0.5, we achieve the speed of convergence, found in subsection 2.3, of 9%. The corresponding convergence of the rates of return obtained from the GDyn simulation is shown in Figure 3.4. In this example we used a 3 × 3 aggregation of the GTAP 5.4 Data Base. To achieve the desired 9% convergence rate in 7 × 7 aggregations, LAMBRORG and LAMBRORGE were set to 0.4. Thus, to achieve the desired speed of convergence for every new aggregation, the test (3.4) should be repeated. It is also important to keep in mind that the convergence rate of 9% per year was obtained using OECD data only. Most likely, the speed of convergence would be lower if we included countries outside OECD, and hence a speed of convergence of 9% per year represents the upper bound of the desirable convergence of the net rates of return in the GDyn model. Without estimating equation (3.4) on data that includes countries outside OECD, we cannot say anything about the lower bound. Given that the GTAP Data Base includes all countries, a speed of convergence of 9% overestimates
84
Alla Golub and Robert A. McDougall
the degree of capital mobility, and some lower value is probably more desirable. The lower speed of convergence is also desirable because, for some aggregations of the GTAP Data Base, high values of LAMBRORG and LAMBRORGE may lead to the model being unsolvable. Finally, let us consider situations when a low value of LAMBRORGE is desirable. Consider the situation when a region is characterized by a very large expected rate of return and a very low actual rate such that the error in investors’ expectations is large. Large errors in investors’ expectations lead to a large decline in the expected rate of return rorge(r), determined by the following expression in the model (equation [2.95] in Chapter 2): rorge(r) =−RORGFLEX(r) × [qk(r)−100.0×KHAT(r)×time] −100.0×LAMBRORGE(r)×ERRRORG(r)×time + srorge(r), where ERRRORG(r) = ln(RORGEXP(r)/RORGROSS(r)) is an error in investors’ expectations. Note that the contribution of the error to the change in the expected rate of return depends on the magnitude of the parameter LAMBRORGE: The larger the parameter, the greater the contribution of the error to rorge(r). The large decline in the expected rate of return, in turn, leads to a large positive change in the required rate of growth in the rate of return, and because required and expected rates of growth are equal, this decline also leads to a large positive change in the expected rate of growth in the rate of return (equation [2.86] in Chapter 2): erg rorg(r) = LAMBRORG(r)∗ (rorgt(r) − rorge(r)). This large positive change in the expected rate of growth in the rate of return determines the change in investment qcgds(r) through equation (2.83) in Chapter 2: erg rorg(r) = −RORGFLEX(r)∗ {IKRATIO(r)∗ [qcgds(r) − qk(r)] −DKHAT(r)}. To achieve large positive changes in the rate of growth in the rate of return, investors should reduce their investment. The larger the change in the rate of growth in the rate of return to capital erg_rorg(r), the larger the reduction in investment should be undertaken. In some cases, the required decline in investment is so large that the level of investment may become negative, which is not allowed in the model simply because we do not observe such cases in the real world.
Behavioral and Entropy Parameters in the Dynamic GTAP Model
85
The following example based on a 7 × 7 aggregation of the GTAP 5.4 Data Base illustrates this case. In this aggregation Japan is a separate region and is characterized by a large error in investors’ expectations. This large error in investors’ expectations about rates of return to capital in Japan leads to a large decline in the expected rate of return, positive change in the expected rate of growth in the rate of return, and a large decline in investment. In the second year of the simulation the decline in investment is so large that one of the inputs in production of capital goods in Japan becomes negative. This is equivalent to destroying capital stock. Because we do not allow such situations – spending on investment is restricted to be non-negative – the model cannot be solved. As noted earlier, the contribution of the error to the change in the expected rate of return depends on the magnitude of the parameter LAMBRORGE. In this example, the LAMBRORGE was initially set to 0.4. By decreasing LAMBRORGE, we can reduce the influence of the error in investors’ expectations on rorge(r), erg_rorg(r), and then, finally, on qcgds(r). At LAMBRORGE set to 0.2 for Japan, the investment declines much more slowly, and the situation in which the level of input in the production of capital goods becomes negative is avoided. Although the lower value of LAMBRORGE for Japan allows avoiding a negative level of investment, it creates asymmetry in a sense that the degree of capital mobility for Japan in this simulation is lower than for other regions. In situations where such asymmetry is not desirable, LAMBRORGE could be set to the smaller value of 0.2 in our example, uniformly across regions. A better solution would be to set some exogenous minimum level of gross investment and then let investors choose the maximum of this minimum investment level and investment determined by the dynamic theory (see Appendix).
3. Elasticity of Rate of Return to Capital with Respect to Capital Stock The parameter RORGFLEX in Table 3.1 represents the negative of the elasticity of the expected rate of return with respect to the size of capital stock. The effect of different values of RORGFLEX on the expected rate-of-return schedule is shown in Figure 3.5. A large value for RORGFLEX – for example, 10 – implies that a 1% increase in the capital stock is expected to reduce the rate of return to capital by 10%; hence, the expected rate of return is sensitive to the supply of capital goods. Inversely, the supply of new capital goods is not very sensitive to changes in the expected rate of return.
86
Alla Golub and Robert A. McDougall
30%
Expected rate of return
25%
20%
15%
10%
5%
0% 0
50
100
150
200
250
Capital stock RORGFLEX=0
RORGFLEX=1
RORGFLEX=2
Figure 3.5. The effect of different values of the elasticity of rate of return on the expected rate of return schedule. Source: Authors’ calculations.
Using a production function approach to GDP, it can be shown that RORGFLEX is the inverse of the elasticity of substitution between labor and capital in a constant elasticity of substitution (CES) production function. The following example illustrates this point. If we produce GDP via a CES production function where capital and labor are the sole inputs, we can write Y = (αK −ρ + (1 − α)L −ρ )−1/ρ .
(3.5)
If we maximize profits defined as = pY − wL − rK, the first-order conditions (FOCs) are Y 1+ρ = r/p α K 1+ρ Y (1 − α) = w/p , L
(3.6)
where r is rental price of capital, w is labor wage, p is price level, and factor rental rates are r/p and w/p for capital and labor, respectively. Taking
Behavioral and Entropy Parameters in the Dynamic GTAP Model
87
logarithms of the FOCs we get
r Y ln α + (1 + ρ) ln = ln K p w Y . ln(1 − α) + (1 + ρ) ln = ln L p
(3.7)
From the FOCs, the negative of the elasticity of the return to capital with respect to the size of capital stock, RORGFLEX, is d ln(r/p ) = 1 + ρ = 1/σ, (3.8) d ln K where σ is the elasticity of factor substitution of a CES production function. Expression (3.8) for RORGFLEX is derived under the assumption that the price of capital goods is equal to the GDP price index p. This assumption allows defining the rate of return to capital as r/p. Note that the rate of return to capital in (3.8) is the actual, not the expected, rate of return to capital. Here and later we assume that investors’ understanding of the relationship between the actual rate of return and size of the capital stock and the relationship between the expected rate of return and size of the capital stock is characterized by the same magnitude of the curvature parameter RORGFLEX. This assumption is reasonable because it simply implies that investors know the economy’s production possibility and build their expectations in accordance with these possibilities. Having data on inputs and prices, we could obtain the elasticity of factor substitution, and hence the parameter RORGFLEX could be obtained from joint estimation of FOCs (3.7) and the production function (3.5). However, the production structure in the GTAP model is much more complex than the simple CES production function approach described here. Instead of econometric estimation, we discuss two approaches to setting RORGFLEX. The first approach allows us to obtain model-consistent or the perceived (by investors in the model) elasticity of the rate of return with respect to the size of capital stock. This approach is based on a post-aggregation calibration procedure for each region that is done each time a new aggregation is made. In this calibration simulation we shock the capital stock in each region by shocking the exogenous shift variable sqk(r), the region-specific shock to capital stock. Then the elasticity, given in (3.8), is calculated as the ratio of the obtained percent change in the actual rate of return rorga(r) and capital stock qk(r). This simulation is conducted in a comparative static mode; that is, the variable time is not shocked, and all parts of the model that depend on the variable time are excluded. As a result, the obtained change in the RORGFLEX = −
88
Alla Golub and Robert A. McDougall
rate of return is not conditional on the parameter LAMBRORGE, which is desirable because LAMBRORGE is not accurately known. It should be noted that calibrated RORGFLEX is conditional on the model structure, the database, and parameters. As a consequence, the true elasticity of the rate of return in the initial period, computed with the initial database, becomes inconsistent in the later periods of the simulation as we update the database. Ideally, RORGFLEX should be recalibrated every period, but doing so would tremendously complicate the use of the model. However, because the full model uses CES production functions, and the elasticities of substitution in these production functions are constant, we expect that the changes in the true elasticity will be relatively minor. If so, RORGFLEX can be calibrated just once before the simulation, but on a post-aggregation database. The setting of RORGFLEX can be further simplified, because in many simulations with the model the calibrated elasticity was close to 1 for all considered regions (see, for example, Ianchovichina 1998). Hence, RORGFLEX can be set to 1 uniformly for all regions independent of the aggregation. The second approach determines the RORGFLEX parameter not by postaggregation calibration, but by applying simple aggregation formulas (see the later discussion). This approach is consistent with the view that different economies are characterized by different elasticities of the rate of return to capital with respect to capital stock; it would require deciding about the magnitude of RORGFLEX in each country/region. Note that this decision may be purely judgmental or may be based on the calibration of RORGFLEX parameters once for all disaggregated regions in the GTAP Data Base and then aggregated for specific scenarios. Magnitudes of RORGFLEX obtained with this approach are unlikely to match exactly the model-consistent values of RORGFLEX. When RORGFLEX diverges from the true flexibility of the capital stock, the divergences affect the behavior of the model in several ways. Consider two situations: first when RORGFLEX is set to the model-consistent value as described by the first approach, and second when RORGFLEX is set to some different value. The differences between the second and first situations are errors caused by the inconsistent value of RORGFLEX.9 Comparing these two situations, first there is an error in the change in the expected rate of growth in the rate of return (equation [2.83] in Chapter 2): erg rorg(r) = −RORGFLEX(r)∗ {IKRATIO(r)∗ [qcgds(r) − qk(r)] − DKHAT(r)}. 9
In this discussion, the errors are different from “errors in investors’ expectations” in the structure of the dynamic model.
Behavioral and Entropy Parameters in the Dynamic GTAP Model
89
However, it is not troublesome. This equation determines the change in real investment qcgds(r) in a region. Because expected and required growth in the rate of return are equal, the expected rate of growth in the rate of return (equation [2.86] in Chapter 2) is erg rorg(r) = LAMBRORG(r)∗ (rorgt(r)− rorge(r)), and the change in investment qcgds(r) is given implicitly: −RORGFLEX(r)∗ {IKRATIO(r)∗ [qcgds(r) − qk(r)] − DKHAT(r)} = LAMBRORG(r)∗ (rorgt(r) − rorge(r)).
(3.9)
For different values of RORGFLEX, the same qcgds(r) can be obtained by simply changing parameter LAMBRORG in expression (3.9). That is, the inconsistency in RORGFLEX can be eliminated by adjusting the parameter LAMBRORG. Although we do not do this, we could adjust the parameter LAMBRORG because it is not accurately known. Returning to the comparison of the two situations − model-consistent and model-inconsistent elasticity of the rate of return with respect to the size of capital stock − there is an error due to the model-inconsistent value of RORGFLEX in the adjustment of DKHAT (equation [2.101] in Chapter 2): DKHAT(r) = LAMBKHAT(r)∗ [qk(r) + RORGFLEX(r)−1∗ rorga(r) − 100∗ KHAT(r)∗ time]. Provided that the rate of return does not change rapidly (rorga(r) is small), this error is small. Over the long-run, the rate of return is unlikely to change rapidly, so this error is unlikely to interfere greatly with the elimination of errors in the perceived normal rate of growth in the capital stock. Finally, there is an error in the equation for the expected rate of return, in the term representing adjustment of the expected rate of return to abnormal growth in the capital stock (equation [2.95] in Chapter 2): rorge(r) = −RORGFLEX(r)×[qk(r)−100.0×KHAT(r)×time] − 100.0×LAMBRORGE(r)×ERRRORG(r)×time+srorge(r). Again, over the long-run the error is likely to be small, because there is unlikely to be prolonged rapid abnormal growth in the capital stock. It seems then that errors in RORGFLEX may not interfere greatly with the convergence properties of the model. However, they will affect the behavior of the model when the economy is far from equilibrium: that is, when there is rapid abnormal capital accumulation or when the rate of return is
90
Alla Golub and Robert A. McDougall
0.14
actual rate of return
0.12
0.1
0.08
0.06
0.04
0.02
0 1997
2002
2007
2012 NAM
EUN
2017
2022
ROW
Figure 3.6. Divergence of net rates of return to capital when RORGFLEX = 5, five times larger than the model-consistent value. Source: Authors’ simulations with the GDyn model.
changing rapidly. The consequences of errors in RORGFLEX are not clear a priori, and the behavior of the model should be tested under realistic scenarios, with various treatments of RORGFLEX. To illustrate a possible problem, we use the 3 × 3 aggregation, discussed earlier. For the three regions in this aggregation – NAM, EUN, and ROW – the post-aggregation calibrated RORGFLEX is close to 1. However, for illustration purposes, we set RORGFLEX = 5 for all three regions.10 Figure 3.6 shows that when the elasticity of the rate of return is set much larger than its model-consistent value, rates of return to capital diverge. So the setting of this parameter requires some care. If the second approach is adopted – that is, RORGFLEX parameters are not determined by post-aggregation calibration – and if countries in a region are characterized by different elasticities of the rate of return to capital with respect to capital stock, then the issue arises of how to aggregate RORGFLEX across countries. The aggregation method is described in Chapter 4. 10
The expected rates of return in the initial database are a function of the parameter RORGFLEX. In this simulation the expected rates of return in the initial database are consistent with the new RORGFLEX = 5, as required by the investment theory described in Chapter 2.
Behavioral and Entropy Parameters in the Dynamic GTAP Model
91
4. Parameters Determining Composition of Wealth and Capital in a Region The investment theory of GDyn determines how much will be invested in any given region in each period. These regional investments include both domestic investment and foreign investment via the global trust. Chapter 2 describes how regional savings are allocated between investments in domestic and foreign assets in the model. To remind the reader, equity in a region’s firms WQ_FIRM(r) has two components: equity owned domestically (WQHFIRM(r)) and equity owned by foreigners (WQTFIRM(r)), where WQ FIRM(r) = WQHFIRM(r) + WQTFIRM(r).
(3.10)
The value of financial claims held by a regional household WQHHLD(r), or regional wealth, also has two components: ownership of foreign equity or equity of regional household in the global trust (WQHTRUST(r)) and ownership of domestic equity (WQHFIRM(r)), where WQHHLD(r) = WQHFIRM(r) + WQHTRUST(r).
(3.11)
Thus, for each region we have two accounting identities, but three unknowns. Equations (3.10) and (3.11) determine net foreign assets, but not gross foreign assets. Because rates of return differ across regions in the short- to mediumrun, it is necessary to know the gross foreign asset position to determine income flows from foreign ownership and therefore how regional wealth is affected by a given model simulation. One natural way to pin down a region’s gross foreign asset position would be to adopt a portfolio approach, based on a balancing of risks and returns associated with domestic and foreign assets. However in this model, agents are not risk averse, and there is no endogenous mechanism for generating risk. Therefore, we are forced to turn to an atheoretic rule. This rule takes into account a set of restrictions. First, WQHFIRM(r), WQHTRUST(r), and WQTFIRM(r) should be positive.11 Second, equations (3.10) and (3.11) should hold. Third, the three variables should satisfy the empirical regularity first documented by Feldstein and 11
The restriction that three variables WQHFIRM(r), WQHTRUST(r), and WQTFIRM(r) are positive is imposed in the model. However, as we see later in the case of BelgiumLuxemburg, there are real-world situations when negative values for WQHFIRM(r) are possible.
92
Alla Golub and Robert A. McDougall
Horioka (1980), namely that regions tend to specialize their portfolios strongly in their own domestic assets. The determinants of international portfolio diversification have attracted considerable attention in the literature (for a literature review, see Lewis 1999). Most studies find that international diversification is much lower than that predicted by portfolio allocation models. This is termed the “home bias effect.” Kraay et al. (2000) demonstrated that, under reasonable assumptions, the probability that international crises occurs twice a century is enough to generate a set of country portfolios that are roughly consistent with the data (i.e., a home bias in investments). If regions specialize their portfolios strongly in domestic assets in the initial database, we would like to preserve this relationship over the course of a simulation. In GDyn this is achieved with entropy theory. Cross-entropy minimization gives us a way of dividing a strictly positive total into strictly positive components subject to various constraints, while staying as close as possible to the initial shares. Specifically, this method guarantees that (1) while regional households’ equity is changing over time, the split between equity in local firms and equity in foreign firms stays as close as possible to the split in the initial database and (2) while firms’ capital in a region is changing over time, the split between capital belonging to foreigners and capital belonging to local households stays as close as possible to the split in the initial database. The cross-entropy minimization is summarized by the following equation: (RIGWQH(r) + RIGWQ F (r))∗ wqhf(r) = RIGWQH(r)∗ wqht(r) + RIGWQ F (r)∗ wqtf(r),
(3.12)
where wqhf(r) is the percentage change in equity held by the regional household in domestic firms (WQHFIRM(r)), wqht(r) is the percentage change in equity held by the regional household in the global trust (WQHTRUST(r)), and wqtf(r) is the percentage change in equity held by foreigners in a region (WQTFIRM(r)). RIGWQH(r) and RIGWQ_F(r) are rigidity parameters. The relative magnitude of the rigidity parameters is important: If RIGWQH(r) is assigned a high value and RIGWQ_F(r) a low value, then from equation (3.4) wqhf(r) ≈ wqht(r). That is, the allocation of household wealth is nearly fixed, and most of the adjustment is put on shares in local capital. The opposite happens if RIGWQH(r) is assigned a low value and RIGWQ_F(r) a high value. Setting RIGWQH(r) and RIGWQ_F(r) as equal assumes equal adjustment in household wealth shares and regional firms’ capital shares.
Behavioral and Entropy Parameters in the Dynamic GTAP Model
93
Indirectly, these parameters determine the allocation of regional savings between foreign and local assets and the split of new investment in a region between domestic and foreign investments. Note that SAVE(r) = VQHFIRM(r) + VQHTRUST(r).
(3.13)
Savings in a region SAVE(r) are distributed between investments into domestic firms VQHFIRM(r) and foreign firms VQHTRUST(r). Similarly, NETINV(r) = VQHFIRM(r) + VQTFIRM(r)
(3.14)
represents investment in a region. This may be derived from either the regional household (VQHFIRM(r)) or from foreigners (VQTFIRM(r)). It can be shown that the change in equity held by regional households in domestic firms (WQHFIRM(r)∗ wqhf(r)/100) is determined by the percent change in the price of old equity pcgds(r) and new investment by regional households in domestic firms:12 WQHFIRM(r)∗ wqhf(r)/100 = WQHFIRM(r)∗ pcgds(r)/100 + VQHFIRM(r).
(3.15)
Similarly, WQHTRUST(r)∗ wqht(r)/100 = WQHTRUST(r)∗ pqtrust/100 + VQHTRUST(r).
(3.16)
Equation (3.16) states that the change in equity held by the regional household in the global trust is determined by the change in price of the old equity in the trust pqtrust(r) and new investment by the regional household in the global trust VQHTRUST(r). Finally, WQTFIRM(r)∗ wqtf(r)/100 = WQTFIRM(r)∗ pcgds(r)/100+VQTFIRM(r). (3.17) Equation (3.17) says that the change in equity held by the global trust in a region is determined by the change in price of the old equity pcgds(r) and new investment in a region VQTFIRM(r). Let us assume that the effect of a change in the price of capital goods is small. Then, if RIGWQH(r) is assigned a high value and RIGWQ_F(r) a low value, the split of regional 12
Note, equations (3.15–3.17) and the investment variables are not in the GDyn model code explicitly. They are introduced here to show the effect of the rigidity parameters on the investments and savings allocation. Also, these equations are not technically correct because they should carry a time variable.
94
Alla Golub and Robert A. McDougall
(a) 2,500,000
1,500,000
1,000,000
500,000
9 19 8 9 20 9 0 20 0 0 20 1 02 20 0 20 3 0 20 4 0 20 5 0 20 6 0 20 7 0 20 8 0 20 9 10 20 1 20 1 12 20 1 20 3 1 20 4 1 20 5 1 20 6 1 20 7 1 20 8 19 20 2 20 0 2 20 1 2 20 2 2 20 3 24 20 2 20 5 26 20 27
0 19
Value in millions of 1997 US$
2,000,000
year VQHFIRM(EUN)
VQTFIRM(EUN)
VQHTRUST(EUN)
SAVE(EUN)
NETINV(EUN)
Figure 3.7a. Composition of investments and savings in the European Union (RIGWQH(r)/RIGWQ_F (r) = 10). Source: Authors’ simulations with the GDyn model.
savings between investing locally and investing abroad will be nearly constant and will fluctuate in the neighborhood of the split of regional household wealth WQHHLD(r) between wealth in local assets WQHFIRM(r) and wealth in assets abroad WQHTRUST(r) in the initial database; most of the adjustment will be forced onto the composition of capital and composition of investment in a region, that is on VQTFIRM(r) and VQHFIRM(r). The following hypothetical example, based on a 3 × 3 aggregation of the GTAP 5.4 Data Base, illustrates the importance of the relative magnitude of the rigidity parameters. To perturb the model, it is assumed that there is an economy-wide 5% productivity shock per year to the European economy (EUN) for the first 5 years of the 30-year simulation. Two simulations are compared: (1) rigidity parameters are equal, and (2) rigidity parameter RIGWQH(r), determining the composition of local wealth and distribution of savings, is set 10 times larger than RIGWQ_F(r). Figures 3.7a and 3.7b show that levels of investment, as well as savings, in a region are equal in both simulations. The positive shock to the EUN economy leads to a rapid increase in investments in EUN. The difference between the two simulations is in how this increase is financed. When EUN wealth and regional savings compositions are more rigid compared to EUN capital composition (RIGWQH(r)/RIGWQ_F(r) = 10), the increase in EUN investment is financed mostly from abroad (Fig. 3.7a), and foreigners
Behavioral and Entropy Parameters in the Dynamic GTAP Model
95
(b) 2,500,000
1,500,000
1,000,000
500,000
9 19 8 9 20 9 0 20 0 0 20 1 0 20 2 03 20 0 20 4 0 20 5 06 20 0 20 7 0 20 8 0 20 9 1 20 0 1 20 1 12 20 1 20 3 1 20 4 1 20 5 16 20 1 20 7 1 20 8 1 20 9 2 20 0 2 20 1 22 20 2 20 3 2 20 4 2 20 5 2 20 6 27
0
19
Value in millions of 1997 US$
2,000,000
-500,000 year VQHFIRM(EUN)
VQTFIRM(EUN)
VQHTRUST(EUN)
SAVE(EUN)
NETINV(EUN)
Figure 3.7b. Composition of investments and savings in the European Union (RIGWQH(r)/RIGWQ_F = 1). Source: Authors’ simulations with the GDyn model.
receive most of the benefits of higher returns in EUN. When RIGWQH(r)/ RIGWQ_F(r) = 1, the increase in EUN investment is financed almost equally from domestic and foreign sources. That is, the share of EUN savings invested locally increases after the positive shock to the EUN economy, and the share of EUN savings invested abroad decreases. In this way EUN investors participate more fully in the benefits from a positive shock to their domestic economy. The setting of the rigidity parameters affects the distribution of EUN wealth between local and foreign assets (Figs. 3.7c and 3.7d) and of ownership shares of EUN capital in the long-run. This means that the long-run paths of foreign income payments and gross national product (GNP) also depend on the relative magnitude of the rigidity parameters. In short, they are critical to any dynamic general equilibrium analysis with this model.
4.1 Econometric Model and Data To estimate the relative magnitude of the rigidity parameters, we rearrange equation (3.12), dividing both sides by the sum RIGWQH(r) + RIGWQ_F(r), yielding wqhf(r) = αwqht(r) + βwqtf(r) + e(r) s.t.
α + β = 1,
(3.18)
96
Alla Golub and Robert A. McDougall
(c) 0.85
0.8
share
0.75
0.7
0.65
0.6
19
98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 20 13 20 14 20 15 20 16 20 17 20 18 20 19 20 20 20 21 20 22 20 23 20 24 20 25 20 26 20 27
0.55
year
WQHFIRM/WQHHLD(EUN)
WQHFIRM/WQ_FIRM(EUN)
Figure 3.7c. European Union wealth in local assets as a share of European Union wealth and as a share of European Union capital (RIGWQH(r)/RIGWQ_F = 10).
(d) 0.85 0.8
0.7 0.65 0.6
24 20 26
20
22 20
20 20
18 20
16 20
14 20
12 20
10 20
08 20
06 20
04 20
02 20
00 20
98
0.55
19
share
0.75
year WQHFIRM/WQHHLD(EUN)
WQHFIRM/WQ_FIRM(EUN)
Figure 3.7d. European Union wealth in local assets as a share of European Union wealth and as a share of European Union capital (RIGWQH(r)/RIGWQ_F = 1).
Behavioral and Entropy Parameters in the Dynamic GTAP Model
97
where the coefficient α = RIGWQH(r)/(RIGWQH(r) + RIGWQ_F(r)) and β = RIGWQ_F(r)/(RIGWQH(r) + RIGWQ_F(r)). Note that equation (3.18) does not assert a causal relationship between wqhf(r), wqht(r), and wqtf(r). Rather we are just looking at the relative strength of correlations between the variables. Variables affecting the growth of WQHFIRM(r), and potentially the growth of the two other components WQTFIRM(r) and WQHTRUST(r), are not in the model. Therefore it should not be surprising if the explanatory power of the model (3.18) is low. To estimate equation (3.18) national wealth measures are required. Fortunately, such wealth measures were previously constructed, and analysis based on these measures yields meaningful results. Kraay et al. (2000) constructed country wealth measures to examine how countries hold their financial wealth. Using the same dataset, Kraay and Ventura (2000) studied current account responsiveness to changes in the terms of trade, transfers from abroad, and fluctuations in production. Calderon et al. (2003) also used a similar dataset to explore the roles of risk and returns in the evolution of net foreign asset positions of industrial and developing countries. Calderon et al. (2003) found that for upper and middle income countries and countries with moderate capital account restrictions, there is a longrun relationship between net foreign assets (relative to country wealth) and the relative measure of returns on domestic investment, the relative measure of investment risk, and the ratio of foreign to domestic wealth. In GDyn, net foreign assets are defined as country wealth minus country assets, WQHHLD(r) –WQ_FIRM(r). In principle, observations in Calderon et al. (2003) could be used to build a new module that would replace the atheoretic entropy method. However, to adopt this theoretical approach, we would need not only to remove the entropy module but also to change the simple expectations mechanism presented in Chapter 2, which would lead to a totally new and much larger model. Therefore, we opt to retain the current specification and use this dataset to estimate the key rigidity parameters. Returning to the wealth measures and the econometric model (3.18), three growth rates are needed: growth of WQHFIRM(r), WQHTRUST(r), and WQTFIRM(r). We use the country portfolio database constructed by Kraay et al. (2000). The database covers 68 countries, listed in Table 3.3, including all industrial countries and a substantial number of developing countries from 1966 to 1997. The database contains estimates of domestic capital stock, domestic equity owned by foreign residents, foreign equity owned by domestic residents, loans issued by domestic residents and owned by foreign residents, and loans issued by foreign residents and owned by domestic residents. Gross assets abroad WQHTRUST(r) is foreign equity
98
Alla Golub and Robert A. McDougall
Table 3.3. Kraay et al. (2000) database: Countries and period covered for each country Country
Code
Period
Obs.
East Asia and the Pacific (EAP) China CHN 1981–97
17
Indonesia
IDN
1966–94
29
Korea Malaysia
KOR MYS
1969–97 1976–94
29 19
Philippines Singapore Thailand
PHL SGP THA
1967–97 1966–7 1969–97
31 32 29
Country
Code
Period
Obs.
Industrial countries (INDC) Australia AUS 1966–97 Austria AUT 1967–97 BelgiumBLX 1967–77, Luxembourg 1986–96 Canada CAN 1966–97 Switzerland CHE 1983–97
32 31 22
Latin American and the Caribbean (LAC) Argentina ARG 1966–74, 30 1976–89, 1991–7 Bolivia BOL 1966–85, 31 1987–97 Brazil BRA 1966–7 32 Chile CHL 1967–73, 15 1977–9, 1981–5 Colombia COL 1967–94 28 Costa Rica CRI 1966–95 30 Dominican DOM 1969–84, 25 Republic 1986–94 Ecuador ECU 1966–96 31 Guatemala GTM 1966–94 29 Honduras HND 1966–97 32 Jamaica JAM 1968–95 28
32 14
Mexico Nicaragua
MEX NIC
Germany Denmark Spain
DEU DNK ESP
1968–97 1968–96 1966–97
30 29 32
PER SLV TTO
Finland
FIN
1966–97
32
Peru El Salvador Trinidad and Tobago Uruguay
France United Kingdom Greece Ireland
FRA GBR
1968–97 1966–97
30 32
Venezuela
VEN
GRC IRL
1966–96 1966–97
31 32
Italy Japan
ITA JPN
30 23
Netherlands Norway
NLD NOR
1968–97 1971–86, 1991–7 1966–97 1975–97
Middle East and North Africa (MENA) Algeria DZA 1966–91 26 Egypt EGY 1988, 2 1989 Iran IRN 1966–82 17 Israel ISR 1969–97 29
32 23
Jordan Morocco
JOR MAR
New Zealand Portugal
NZL PRT
1973–97 1971–97
25 27
Oman Saudi Arabia
OMN SAU
URY
1966–97 1966–81, 1983 1975–93 1966–97 1974–94
32 17
1967–73, 1980–97 1974–97
25
1966–89 1966–82, 1988–97 1973–89 1966–9, 1981, 1985–9
19 32 21
24
24 27 17 10
(continued)
Behavioral and Entropy Parameters in the Dynamic GTAP Model
99
Table 3.3 (continued) Country
Code
Period
Obs.
Country
Code
Period
Obs.
Sweden United States
SWE USA
1966–96 1969–97
31 29
Syria Tunisia Turkey
SYR TUN TUR
1966–87 1966–97 1966–98
21 32 32
Sub-Saharan Africa (SSA) Cote d’Ivoire CIV 1970–85, 1987
17
South Asia (SA) Bangladesh BGD
1972–81, 1983 1966–97
11
Cameroon
CMR
10
India
IND
Congo
COG
1979, 1986–93, 1995 1993–6
4
Sri Lanka
LKA
1966–75, 1980–97 1966–71, 1974–97
28
Lesotho
LSO
1980–94
15
Pakistan
PAK
Mauritius Senegal
MUS SEN
24 13
South Africa
ZAF
1974–97 1968–70, 1972–81 1968–97
32
30
30
owned by domestic residents plus loans issued by foreign residents and owned by domestic residents. Gross foreign liabilities WQTFIRM(r) is domestic capital owned by foreigners and loans issued by domestic residents and owned by foreign residents. WQHFIRM(r) is simply the difference between domestic capital and gross foreign assets WQTFIRM(r). For an overview of the data sources, methodology, and assumptions used to construct the database, we refer the reader to Kraay et al. (2000). The sources used to construct the Kraay et al. (2000) database are relatively standard: Penn World Tables for initial stocks of domestic capital, International Monetary Fund’s (IMF) Balance of Payments Statistical Yearbook and other sources for direct and portfolio equity stocks and flows and debt stocks and flows, and the World Bank’s Global Development Finance for debt stocks and flows for developing countries. Although the discussion in Kraay et al. (2000) focuses on how the financial wealth of the country is distributed across holdings of domestic capital and various foreign assets, we are interested in the relative rigidity of the allocation of domestic wealth and composition of capital. Figures 3.8 and 3.9 show the distribution of gross foreign assets as a share of wealth and gross foreign liabilities as a share of capital, respectively, pooling the available 1,717 observations with nonmissing values for all countries and years. Gross foreign assets positions are small: For 75% of the observations, gross foreign assets as
100
Alla Golub and Robert A. McDougall
50
40
30 % 20
10
0 0
0.1
0.2
0.3
0.4
0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 Gross foreign assets as a share of wealth
1.3
1.4
1.5
1.6
1.7
Figure 3.8. Distribution of gross foreign assets as a share of wealth, pooling the available 1,717 observations with nonmissing values for all countries and years in the Kraay et al. (2000) database.
share of wealth are less than 14%. Gross foreign liabilities as a share of capital are somewhat larger, but for 75% of the sample still represent less than 25% of capital. On both graphs, observations with values greater than 1 represent Belgium-Luxemburg. Toward the end of the sample period, BelgiumLuxemburg keeps most of its wealth abroad and at the same time borrows more than the size of its capital stock, which results in observed gross foreign assets and liabilities shares that are greater than 1. Table 3.4 shows variation in gross foreign assets and liabilities across regions and over time. Claims on foreign assets represent only 4.3% of the wealth in developing countries, whereas foreign claims on domestic assets consist of 11% of domestic capital. For industrial economies, claims on foreign assets and claims of foreigners on domestic assets are balanced and represent 14.8% of wealth and 14.9% of capital, respectively. For industrial countries, the shares of gross foreign assets and gross foreign liabilities in wealth are increasing over time, indicating increasing integration of capital markets. In contrast, the data for developing countries show no strong pattern. It is interesting to note that gross foreign assets and liabilities consist primarily of loans rather than equity (Kraay et al. 2000). However, the
Behavioral and Entropy Parameters in the Dynamic GTAP Model
101
45
40
35
30
25 % 20
15
10
5
0 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.95 1.05 1.15 1.25 1.35 1.45 1.55 1.65 1.75 Gross foreign liabilities as share of capital
Figure 3.9. Distribution of gross foreign assets as a share of capital, pooling the available 1,717 observations with nonmissing values for all countries and years in Kraay et al. (2000) database.
composition of gross foreign assets and liabilities is of less interest for our discussion, because in the GDyn there is only one class of financial assets – equity. This is driven by the consideration that the role of financial assets in the GDyn model is to support international capital mobility, rather than to represent the financial sector per se. Thus the data on gross foreign assets and liabilities, consisting of both equity and loans, are chosen for the analysis.
4.2 Empirical Results We start our analysis by analyzing the relationship between growth rates wqhf(r), wqht(r), and wqtf(r). Figures 3.10a–3.10c show correlations between these variables in high income, middle income, and low income countries, respectively. Countries on the horizontal axis are ordered by average per capita income over the sample period.13 Figure 3.10a shows that 13
Eight countries with less than 14 observations are omitted from the analysis here and in the econometric analysis. These countries are Bangladesh (BGD), Switzerland (CHE),
102
Alla Golub and Robert A. McDougall Table 3.4. Foreign assets as share of wealth and foreign liabilities as a share of
capital, across regions and over time 1966–73 1974–81 1982–9 1990–7 1966–97 Gross foreign assets as a share of wealth Industrial countries Developing countries East Asia and the Pacific Latin America and the Caribbean Middle East and North Africa South Asia Sub-Saharan Africa
0.076 0.025 0.044 0.018 0.043 0.007 0.096
0.088 0.045 0.046 0.032 0.114 0.011 0.061
0.152 0.043 0.035 0.038 0.130 0.008 0.054
0.211 0.046 0.050 0.049 0.065 0.012 0.056
0.148 0.043 0.045 0.038 0.098 0.010 0.061
Gross foreign liabilities as a share of capital Industrial countries 0.061 Developing countries 0.114 East Asia and the Pacific 0.130 Latin America and the Caribbean 0.102 Middle East and North Africa 0.186 South Asia 0.062 Sub-Saharan Africa 0.243
0.082 0.114 0.125 0.122 0.136 0.051 0.167
0.154 0.117 0.082 0.161 0.171 0.058 0.138
0.218 0.102 0.080 0.134 0.161 0.071 0.105
0.149 0.110 0.086 0.135 0.160 0.061 0.148
Weighted averages for each 8-year period are computed using an unbalanced panel. For foreign assets share, the weight is country wealth. For foreign liabilities share, the weight is country capital. Source: Authors’ calculations based on the Kraay et al. (2000) database.
for the majority of industrial countries correlations between growth in gross foreign assets (wqht(r)) and growth in gross foreign liabilities (wqtf(r)) are much stronger than correlations between wqhf(r) and wqht(r) or between wqhf(r) and wqtf(r). This finding suggests a high degree of integration in the global economy. The story for developing countries presented in Figures 3.10b and 3.10c is very different from the one for industrial economies. Overall, correlations are smaller, and the correlations between growth rates of gross foreign assets and liabilities (wqht(r) with wqtf(r)) are much smaller than correlations between composites of capital (wqhf(r) with wqtf(r)) and wealth (wqhf(r) with wqht(r)) compared to Figure 3.10a. What makes the pattern for developing countries so different? There are two possible explanations. First, the developing countries are less integrated in the global economy. Comparison of Figures 3.10b and 3.10c reveals that as we move across the per capita income spectrum – the countries are Chile (CHL), Cameroon (CMR), Congo (COG), Egypt (EGY), Saudi Arabia (SAU), and Senegal (SEN).
Behavioral and Entropy Parameters in the Dynamic GTAP Model
103
(a) 1
0.8
0.6
0.4
0.2
0 GRC PRT IRL ESP
ITA
AUT FIN GBR BLX NLD NZL JPN FRA DNK DEU SWE AUS NOR CAN USA
-0.2
-0.4 country wqhf with wqht
wqhf with wqtf
wqht with wqtf
Figure 3.10a. Correlations between wqhf(r), wqht(r), and wqtf(r) for industrial countries. (b) 0.9
0.7
0.5
0.3
0.1
-0.1
-0.3
-0.5 IND
PAK
IIDN HND
CHN
PHL
BOL
VIC MAR
wqhf with wqht
LKA country
SLV
wqhf with wqtf
DOM
GTM CIN TUN
DZA
JAM
ECU
wqht with wqtf
Figure 3.10b. Correlations between wqhf(r), wqht(r), and wqtf(r) for low per capita income developing countries.
PER
104
Alla Golub and Robert A. McDougall
(c)
0.9
0.7
0.5
0.3
0.1
-0.1
-0.3 THA JOR COL TUR CRI ZAF SYR BRA MYS URY KOR IRN MUS MEX ARG VEN OMN SGP ISR TTO country wqhf with wqht
wqhf with wqtf
wqht with wqtf
Figure 3.10c. Correlations between wqhf(r), wqht(r), and wqtf(r) for middle per capita income developing countries.
ordered by average over the 1966–97 period in per capita income from India (IND) on Figure 3.10b to Trinidad and Tobago (TTO) on Figure 3.10c – correlations for developing countries become more and more like ones of industrial countries. As developing countries become richer, the correlations between growth rates of external positions become higher, suggesting a higher degree of integration in the global economy. Our second explanation is based on the nature of the data. The data for developing countries are much more sparse and incomplete, and a set of assumptions is used to construct the database (see Kraay et al. 2000). These factors could also contribute to the differences that we see between Figure 3.10a and Figures 3.10b and 3.10c. The globalization effect observed in industrial countries creates a problem for us. Because correlation between the independent variables in econometric model (3.18) is very strong – Sweden (SWE) is an extreme case – there is a multicollinearity problem, and we cannot distinguish the estimated
Behavioral and Entropy Parameters in the Dynamic GTAP Model
105
coefficients from zero in the model (3.18). The integration into the world economy masks the relationship between growth in composites of wealth and capital. To overcome the multicollinearity problem, we rewrite the restriction as β = 1 − α, substitute the restriction into the model (3.18), and rearrange the equation: wqhf(r) − wqht(r) = β(wqtf(r) − wqht(r)) + e(r).
(3.19)
Equation (3.19) allows comparison of the rigidity of allocation of wealth and the composition of capital even in the case of industrial countries where correlations between wqht(r) and wqtf(r) are strong. The ordinary least squares (OLS) method, the standard linear regression procedure, assumes that errors are uncorrelated and the variance of error terms is homoskedastic. When data are time series, as in equation (3.19), these assumptions could be violated and need to be tested before applying OLS. If regression disturbances follow the autoregressive scheme or their variance is not constant through time, the least squares estimator’s variances are biased and could not be used to test the hypothesis. We use the Durbin-Watson test for first-order autocorrelation. To see whether variance is nonconstant through time, we test for the presence of the autoregressive conditional heteroskedasticity (ARCH) process developed by Engel (1982).14 ARCH models recognize successive periods of relative volatility and stability, and they treat heteroskedasticity as a variance to be modeled. Note that here we are not interested in the variance itself; rather we want to correct the standard errors to be able to accurately test a statistical hypothesis. We found the presence of first-order autocorrelation in all industrial countries and some developing countries.15 The null hypothesis of constant variance is rejected only in the cases of the United Kingdom, Israel, Ireland, and Greece. In the United Kingdom and Ireland, the test result is driven by a large squared error in 1997. In 1997, the proportionate change in equity in domestic firms wqhf held by regional households is abnormally large by absolute magnitude and is negative; it is driven by the shift in composition of wealth toward assets abroad in both countries. Dropping 1997 from the estimation results in homoskedastic variance. In the cases of Greece and 14
15
The standard test for ARCH of order q process is used, where the OLS squared residuals in model (3.10) are regressed on a constant and q lags. Then, we compare the N ∗ R2 value (where N is sample size and R2 is R-square measure of fit) with a χ2 distribution with q degrees of freedom (see Shazam user’s reference manual, for example). First-order autocorrelation was found in all industrial countries, with the exception of Belgium-Luxemburg where we did not test for autocorrelation because of a break in the time-series data.
106
Alla Golub and Robert A. McDougall
Ireland, it seems that the test for nonconstant variance reflects the autoregressive error process, because after the correction for first-order autocorrelation the hypothesis of homoskedasticity cannot be rejected, whereas in the case of the United Kingdom and Israel, heteroskedasticity persists after correction for first-order autocorrelation. For these four countries the ARCH (1) model was estimated (Engel 1982).16 The results of this estimation for coefficient β in equation (3.19) are very similar to results obtained from estimation with a simple correction for first-order autocorrelation in terms of the magnitude of obtained estimates of β and are identical in terms of test β = 1 outcomes. Because results of the more complex ARCH are similar to results of the simpler method, for Israel we report results based on OLS (where no autocorrelation is found), and results for the other three countries are based on the maximum likelihood estimation method, which allows correction for first-order autocorrelation. Estimation results for equation (3.19) are presented in Table 3.5. The estimates of coefficient β in equation (3.19) and its significance level are reported in column 2. Where necessary, we correct for first-order autocorrelation.17 The associated autocorrelation coefficient and its significance level are shown in column 3. The Durbin-Watson statistics calculated after the correction for first-order autocorrelation and model fit are shown in columns 4 and 5, respectively. In column 6 we report results from testing the hypothesis that coefficient β is equal to 1. This would be the case if the composition of capital is rigid and the allocation of wealth is flexible. If coefficient β is not different from zero, we also test if α = 1 − β is not different from 1 to see if the allocation of wealth is rigid, given that the composition of capital is flexible. Based on the results reported in Table 3.5, countries can be grouped into four categories. The first group of countries is characterized by the rigid composition of capital and the flexible allocation of wealth. This group includes most of the developing countries in our sample, except for Singapore (SGP), and 13 of 20 industrial countries.18 For these countries, coefficient β is highly significant and in most cases is not statistically different 16 17
18
Higher order ARCH coefficients in the variance equation are not significant. If regression disturbances follow an autoregressive scheme, the least squares estimator of the regression coefficients is not asymptotically efficient, and the estimator’s variances are biased and cannot be used for testing hypothesis. In these cases we introduce an autocorrelation coefficient into the model and estimate it together with coefficient β using the full unconditional maximum likelihood estimation method. Although Singapore is a highly industrialized country with high per capita income, we include this country in East Asia and the Pacific group simply because it was included in this group in the Kraay et al. (2000) database.
Behavioral and Entropy Parameters in the Dynamic GTAP Model
107
Table 3.5. Regression results for the model wqhf (r)t – wqht(r)t =
β (wqtf (r)t − wqht(r)t ) + et estimated for each country
Coeff. of Country Estimated β autocorrelation 1 2 3
DW 4
R-square 5
F-test β=1 6
INDC AUS AUT BLX CAN DEU DNK ESP FIN FRA GBR GRC IRL ITA JPN NLD NOR NZL PRT SWE USA
1.223∗∗∗ 0.425 2.731∗ 1.100∗∗∗ 0.343 1.240∗∗∗ 1.183∗∗∗ 0.703∗∗∗ 0.783 −0.342 1.078∗∗∗ 1.309∗ 1.024∗∗ 0.291 2.249∗∗∗ 0.391∗∗ 1.011∗∗∗ 0.918∗∗∗ 0.477 0.184
1.999 1.729 2.323 1.895 2.141 1.831 1.790 1.720 1.892 1.700 1.782 1.660 1.762 1.470 1.503 2.023 1.983 1.715 1.607 2.426
0.947 0.297 0.145 0.670 0.573 0.218 0.601 0.322 0.395 0.139 0.900 0.374 0.438 0.660 0.322 0.148 0.793 0.602 0.314 0.509
15.81∗∗∗ 1.69 1.29 0.3 4.00∗ 0.33 1.11 1.37 0.20 6.76∗∗ 1.42 0.22 0.00 4.12∗ 4.17∗∗ 12.80∗∗∗ 0.01 0.17 3.18∗ 42.80∗∗∗
LAC ARG BOL BRA COL CRI DOM ECU GTM HND JAM MEX NIC PER SLV TTO URY
0.943∗∗∗ 0.841∗∗∗ 1.056∗∗∗ 0.949∗∗∗ 0.685∗∗∗ 1.029∗∗∗ 0.765∗∗ 0.938∗∗∗ 0.899∗∗∗ 1.222∗∗∗ 1.065∗∗∗ 1.229∗∗∗ 0.956∗∗∗ 1.103∗∗∗ 0.916∗∗∗ 0.865∗∗∗
1.861 2.041 1.943 1.533 1.653 1.694 1.883 2.232 2.004 1.919 1.703 2.686 1.540 2.147 1.862 1.877
0.783 0.756 0.866 0.921 0.227 0.867 0.751 0.700 0.612 0.749 0.620 0.689 0.801 0.677 0.875 0.787
−0.648∗∗∗ −0.571 −0.446∗∗ −0.760 −0.309 −0.573∗∗∗ −0.454∗∗ −0.457 −0.449∗ −0.523∗∗∗ −0.733∗∗∗ −0.568 −0.567∗∗∗ −0.567∗∗∗ −0.601 −0.357∗ −0.523∗∗∗ −0.567∗∗∗ −0.768∗∗∗
−0.256
−0.372∗∗
−0.333∗
F-test 1−β = 1 7
0.93
1.09
2.64 0.44
0.69 5.26∗∗∗
2.64 2.17
3.12∗ 86.67∗∗∗ 0.47 0.89 1.74 0.11 9.90∗∗∗ 105.14∗∗∗ 0.28 0.59 2.56 0.22 1.07 0.14 0.55 1.12 1.97
Obs. 8 31 30 20 31 29 28 31 31 29 31 30 31 29 15 31 22 24 26 30 28 27 31 31 27 29 23 31 28 31 27 31 15 20 31 20 23 (continued)
108
Alla Golub and Robert A. McDougall Table 3.5 (continued)
Country 1
Estimated β 2
Coeff. of autocorrelation 3
EAP CHN IDN KOR MYS PHL SGP THA
1.141∗∗∗ 0.997∗∗∗ 0.901∗∗∗ 0.610∗∗ 0.884∗∗∗ 0.162 0.653∗∗∗
MENA DZA IRN ISR JOR MAR OMN SYR TUN TUR
1.035∗∗∗ 0.975∗∗∗ 1.037∗∗∗ 0.494∗ 0.449∗∗∗ 0.741∗∗∗ 0.926∗∗∗ 0.905∗∗∗ 0.811∗∗∗
−0.285 −0.497∗
SA IND LKA PAK
1.020∗∗∗ 0.688∗∗∗ 1.019∗∗∗
−0.465∗∗∗
SSA CIV MUS ZAF
0.788∗∗ 0.624∗∗∗ 1.294∗∗∗
−0.606∗∗ −0.367∗ −0.337∗ −0.385∗
−0.280 −0.288
−0.770∗∗∗
DW 4
R-square 5
F-test β=1 6
F-test 1−β = 1 7
Obs. 8
2.044 1.784 1.835 1.566 1.660 1.713 1.777
0.781 0.999 0.614 0.424 0.283 0.112 0.654
1.02 0.58 1.27 5.10∗∗ 2.12 6.56∗∗ 14.44∗∗∗
0.25 51.12∗∗∗
16 28 28 18 30 31 28
1.883 1.213 1.635 1.548 2.020 1.236 1.905 1.87 2.038
0.829 0.974 0.366 0.152 0.377 0.393 0.927 0.739 0.677
0.15 0.37 0.02 4.12∗ 21.78∗∗∗ 1.19 1.34 0.80 3.42∗
1.898 1.791 2.085
0.947 0.548 0.888
0.17 6.21∗∗ 0.08
31 26 28
2.298 1.642 1.516
0.378 0.683 0.586
0.61 17.28∗∗∗ 1.77
15 23 29
3.96∗ 14.51∗∗∗
25 16 28 23 25 16 20 31 31
∗∗∗ , ∗∗ ,
and ∗ denote significance levels at 0.01, 0.05, and 0.1, respectively. The estimation method is ordinary least squares. Where it is necessary, we corrected for first-order autocorrelation and used full unconditional maximum likelihood estimation method. Because no intercept term is used, the R-square is redefined. In column 7 the F-test is reported only (1) to see if α = 1 − β is different from 1 in the cases when coefficient β is not different from zero; and (2) to see if α = 1 − β is different from 1 in the cases when β is different from zero, but at the same time less than 1.
from 1, meaning that α = 1 − β is zero and the allocation of wealth is very flexible. The second group consists of countries where the coefficient β is small in absolute magnitude and statistically is not different from zero, and α = 1 − β is not different from 1. These are industrial countries and include Germany (DEU), United Kingdom (GBR), Japan (JPN), Sweden
Behavioral and Entropy Parameters in the Dynamic GTAP Model
109
(SWE), United States (USA), and Singapore (SGP). In these economies the composition of wealth is rigid, and the composition of capital is flexible. The third group is represented by countries wherein both compositions are rigid. Here, the coefficients β and α are both significant and less than 1. Depending on their relative magnitude, one composition is more rigid than another. This group consists of one industrial country (Norway [NOR]) and some of the developing countries: Bolivia (BOL), Ecuador (ECU), Malaysia (MYS), Thailand (THA), Jordan (JOR), Morocco (MAR), Sri Lanka (LKA), and Mauritius (MUS). Finally, the fourth group consists of Austria (AUS) and France (FRA). In these countries, the coefficient β is not different from zero, but at the same time it is not different from 1. Similarly, α = 1 − β is neither different from 1 nor from zero. We interpret this as representing the case where both the capital and wealth compositions are equally flexible or equally rigid. Having determined the relative rigidities of the composition of capital and allocation of wealth in the countries in the Kraay et al. (2000) database, the question of how to generalize results for all other countries needs to be addressed. To answer this question we create a balanced panel that covers the 1975–94 period (20 years) and includes 40 countries. Comparing countries by region analyzed in Table 3.5, these 40 countries represent 17 of 20 industrial countries (INDC), 11 of 16 Latin American countries (LAC), 5 of 7 East Asia and the Pacific countries (EAP), 3 of 9 Middle East and North Africa countries (MENA), 2 of 3 South Asia countries (SA) and 2 of 3 countries in Sub-Saharan Africa (SSA) region. Thus, all regions of the Kraay et al. (2000) database are represented relatively well in the panel, except MENA. To see whether the shorter time period covered in this panel, as compared to the longer periods covered in the separate country regressions in Table 3.5, could potentially influence the result of the panel estimation, we repeated the Table 3.5 estimation for each country for 1975–94. We found that estimates of β and results of the tests β = 1 are very robust with respect to the choice of time interval. In estimating the econometric equation (3.19) we use a pooling technique similar to one described in Kmenta (1986). Specifically, we assume that the model is cross-sectionally heteroskedastic and timewise autoregressive. Because the model (3.19) is likely to suffer from the omitted variables problem – variables that influence all cross-sectional units – we also assume that error terms are cross-sectionally correlated. However, the estimation of model (3.19) under these three assumptions about error terms appears to be problematic. The problem is that for this particular dataset, in which the number of cross-sectional units is twice as large as the number of time
110
Alla Golub and Robert A. McDougall Table 3.6. Regression result for each region and overall panel
Region 1
Estimated β 2
R-square 3
Assuming cross-sectional correlation 0.907 INDC 0.948∗∗∗ 0.825 EAP 0.970∗∗∗ 0.839 LAC 0.884∗∗∗ 0.545 MENA 0.772∗∗∗ 0.943 SA 1.108∗∗∗ 0.656 SSA 0.653∗∗∗ Assuming cross-sectional independence INDC 0.996∗∗∗ EAP 0.956∗∗∗ LAC 0.920∗∗∗ MENA 0.732∗∗∗ SA 1.021∗∗∗ SSA 0.676∗∗∗ 0.730 0.716 All 0.947∗∗∗ ∗∗∗ , ∗∗ ,
F-test β = 1 4
Number of Countries 5
Obs. 6
9.767∗∗∗ 0.463 19.742∗∗∗ 6.082∗∗ 0.205 21.000∗∗∗
17 5 11 3 2 2
340 100 220 60 40 40
0.693 0.796 4.296∗∗ 8.366∗∗∗ 0.269 13.925∗∗∗
17 5 11 3 2 2
340 100 220 60 40 40
6.364∗∗
40
800
and ∗ denote significance levels at 0.01, 0.05, and 0.1, respectively.
periods, the variance–covariance matrix is close to singular and cannot be inverted. There are two possible approaches to overcoming the problem. The first approach is to estimate model (3.19) under the three assumptions, but with smaller number of cross-sectional units; that is, using subsets of 40 countries. The second approach is to drop the assumption of crosssectional dependence. As we will see, the results are consistent across these two approaches and also with the results reported in Table 3.5. The estimation results of equation (3.19) as a cross-sectionally correlated and timewise autoregressive model for each of the six geographic regions are presented in the first part of Table 3.6. For the group of industrial countries (INDC), the estimated coefficient β is 0.948, which is close to 1, but different from 1 statistically. From regression results for each INDC country reported in Table 3.5, we see that the estimated β varies from country to country and, for 8 of the 20 countries, is different from 1. Thus, the result obtained using the pooling technique is consistent with the results based on a separate regression for each INDC country, suggesting that in INDC the composition of capital is much more rigid than the allocation of wealth. For the Latin America (LAC) region, the estimated β is 0.884, and the hypothesis β = 1 is strongly rejected. The latter is somewhat surprising
Behavioral and Entropy Parameters in the Dynamic GTAP Model
111
because in separate regression models for each country (Table 3.5) the hypothesis β = 1 cannot be rejected for the majority of Latin America countries. However, both pooled and individual country models support the hypothesis that for the LAC region the composition of capital is more rigid than the allocation of wealth. In the MENA region, where pooled data include only Israel (ISR), Tunisia (TUN), and Turkey (TUR), the estimated β is 0.772 and statistically different from 1. Again, this is somewhat surprising given that in separate regressions for these three countries (Table 3.5), β is statistically different from 1 only in Turkey (TUR). However, the results based on the pooling technique do not alter the conclusion achieved with country-by-country regressions: that the composition of capital is more rigid than the composition of wealth in the MENA region. In support of this conclusion, we test and reject the hypothesis that β = 0.5, which is the test for equal rigidity of the two compositions. In East Asia and the Pacific (EAP) and South Asia (SA), estimated β is not statistically different from 1, which is consistent with the results reported in Table 3.5 for countries in these regions. In the Sub-Saharan Africa (SSA) region, the estimate of β is significantly different from zero but less than 1. To see whether compositions of capital and wealth have similar degrees of rigidity, we test the hypothesis β = 0.5 and reject it only at the 10% significance level. This suggests that for SSA the rigidities of composition of capital and wealth are very similar. Note that the results obtained for MENA, SA, and SSA could be generalized to other countries (not included in the estimation or in the Kraay et al. (2000) database) in these regions with caution, because panel data used in estimation for each of these regions are not representative. Now we estimate the regression model (3.19) by pooling all 40 countries together assuming timewise autoregression, cross-sectional heteroskedasticity, but cross-sectional independence. First, we restrict the slope coefficients to be the same for all six regions and then allow heterogeneous slopes to test for differences across regions. The results are reported in the second part of Table 3.6. The results reported in the first and second parts of Table 3.6 are very similar in terms of the magnitude of the estimated β and the results of the test β =1, except for INDC countries. Taking into account cross-sectional correlation among the group of INDC countries appears to be important, particularly when testing whether β is significantly different from 1. When we ignore the omitted variables problem for industrial countries, the estimate of β is not statistically different from 1. Finally, restricting all β to be equal for all countries results in an estimate of β very close to 1, suggesting that the composition of
112
Alla Golub and Robert A. McDougall
capital is much more rigid than the allocation of wealth for all countries considered.
4.3 Rigidity Parameters This section summarizes findings about the relative rigidity of the composition of capital and the allocation of wealth in terms of recommendations for setting rigidity parameters in the GDyn model. The capital stock of a region consists of two assets: domestic capital owned by domestic residents and gross foreign liabilities. The wealth of a region is also held in a two-asset portfolio composed of domestic capital owned by domestic residents and gross foreign assets represented by shares in the global trust. Changes in the compositions of wealth and capital over time are determined by investors’ decisions on how much of their savings should be invested domestically and how much should be sent abroad. In the real world, when forming their portfolios, investors take into account not only relative returns to capital but also risk. In GDyn, investors will reallocate capital from regions with lower rates of return to regions with higher rates of return; however, the model does not account for the risk-related portion of this investment decision. Therefore, to determine the composition of capital and allocation of wealth, we adopt the atheoretic rule (3.12). Indirectly, this rule determines the allocation of regional savings between foreign and local assets and the split of new investment in a region between domestic and foreign investments. The relative magnitude of the rigidity parameters in (3.12) determines the relative rigidity of the composition of wealth and capital. The rigidity parameters are weights in a weighted sum of two cross entropies – one associated with local capital ownership shares and another with wealth allocation shares. This weighted sum is minimized to keep the composition of capital and the allocation of wealth as close as possible to the split in the initial database. Only non-negative values of the rigidity parameters are consistent with cross-entropy minimization, and only one of the rigidity parameters can be zero, not both.19 Although one of the rigidity parameters can be set to be zero in the model, such a parameterization makes the model more fragile. If, for example, RIGWQH(r) is set to zero and there is high investment in a region, the local household may be required to invest more than 100% of its savings in local firms. In this sense, a zero value is not desirable. 19
Setting both parameters to zero will eliminate relationship (3.12) from the model.
Behavioral and Entropy Parameters in the Dynamic GTAP Model
113
Table 3.7. Rigidity parameters in the GDyn model Country 1 INDC AUS AUT BLX CAN DEU DNK ESP FIN FRA GBR GRC IRL ITA JPN NLD NOR NZL PRT SWE USA ALL INDC LAC ARG BOL BRA COL CRI DOM ECU GTM HND JAM MEX NIC PER SLV TTO URY ALL LAC
Estimated β 2
F-test β = 1 3
1.223∗∗∗ 0.425 2.731∗ 1.100∗∗∗ 0.343 1.240∗∗∗ 1.183∗∗∗ 0.703∗∗∗ 0.783 −0.342 1.078∗∗∗ 1.309∗ 1.024∗∗ 0.291 2.249∗∗∗ 0.391∗∗ 1.011∗∗∗ 0.918∗∗∗ 0.477 0.184 0.948∗∗∗
15.81∗∗∗ 1.69 1.29 0.3 4.00∗ 0.33 1.11 1.37 0.20 6.76∗∗ 1.42 0.22 0.00 4.12∗ 4.17∗∗ 12.80∗∗∗ 0.01 0.17 3.18∗ 42.80∗∗∗ 9.767∗∗∗
0.943∗∗∗ 0.841∗∗∗ 1.056∗∗∗ 0.949∗∗∗ 0.685∗∗∗ 1.029∗∗∗ 0.765∗∗ 0.938∗∗∗ 0.899∗∗∗ 1.222∗∗∗ 1.065∗∗∗ 1.229∗∗∗ 0.956∗∗∗ 1.103∗∗∗ 0.916∗∗∗ 0.865∗∗∗ 0.884∗∗∗
3.12∗ 0.47 0.89 1.74 0.11 9.90∗∗∗ 0.28 0.59 2.56 0.22 1.07 0.14 0.55 1.12 1.97 19.74∗∗∗
F-test 1 − β = 1 4
0.93
1.09
2.64 0.44
0.69 5.26∗∗∗
2.64 2.17 3278.64∗∗∗
86.67∗∗∗
105.14∗∗∗
1143.07∗∗∗
RIGWQH 5
RIGWQ_F 6
0.01 1 0.01 0.01 1 0.01 0.01 0.01 1 1 0.01 0.01 0.01 1 0.01 1 0.01 0.01 1 1 0.05
1 1 1 1 0.01 1 1 1 1 0.01 1 1 1 0.01 1 1 1 1 0.01 0.01 1
0.01 0.189 0.01 0.01 0.01 0.01 0.307 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.13
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 (continued)
114
Alla Golub and Robert A. McDougall Table 3.7 (continued)
Country 1
Estimated β 2
F-test β = 1 3
EAP CHN IDN KOR MYS PHL SGP THA ALL EAP
1.141∗∗∗ 0.997∗∗∗ 0.901∗∗∗ 0.610∗∗ 0.884∗∗∗ 0.162 0.653∗∗∗ 0.970∗∗∗
1.02 0.58 1.27 5.10∗∗ 2.12 6.56∗∗ 14.44∗∗∗ 0.463
MENA DZA IRN ISR JOR MAR OMN SYR TUN TUR
1.035∗∗∗ 0.975∗∗∗ 1.037∗∗∗ 0.494∗ 0.449∗∗∗ 0.910∗∗∗ 0.926∗∗∗ 0.905∗∗∗ 0.811∗∗∗
0.15 0.37 0.02 4.12∗ 21.78∗∗∗ 0.14 1.34 0.80 3.42∗
SA IND LKA PAK
1.020∗∗∗ 0.688∗∗∗ 1.019∗∗∗
SSA CIV MUS ZAF ALL
0.788∗∗ 0.624∗∗∗ 1.294∗∗∗ 0.947∗∗∗
∗∗∗ , ∗∗ ,
F-test 1 − β = 1 4
RIGWQH 5
RIGWQ_F 6
0.25 51.12∗∗∗ 468.93∗∗∗
0.01 0.01 0.01 1 0.01 1 1 0.01
1 1 1 1 1 0.01 1 1
0.01 0.01 0.01 1 1 0.01 0.01 0.01 0.233
1 1 1 1 1 1 1 1 1
0.17 6.21∗∗ 0.08
0.01 1 0.01
1 1 1
0.61 17.28∗∗∗ 1.77 6.364∗∗
0.01 1 0.01 0.06
1 1 1 1
3.96∗ 14.51∗∗∗
1996.26∗∗∗
and ∗ denote significance levels at 0.01, 0.05, and 0.1, respectively.
Using the data on composition of wealth and capital stocks for 68 developing and industrial countries (Kraay et al. 2000), we investigated the relative rigidity of the composition of capital and wealth. Our findings are summarized in Table 3.7 in terms of recommendations for setting the rigidity parameters in GDyn. When setting the parameters, we take into account restrictions on the parameter space as discussed earlier. For the majority of countries, including all developing countries except Singapore (SGP) and 13 industrial countries, the composition of capital
Behavioral and Entropy Parameters in the Dynamic GTAP Model
115
is much more rigid than the allocation of wealth. The coefficient β = RIGWQ_F(r)/(RIGWQH(r) + RIGWQ_F(r)) is not different from 1, and the coefficient α = RIGWQH(r)/(RIGWQH(r) + RIGWQ_F(r)) is not different from zero. For these countries we could reasonably set RIGWQH(r) equal to zero and RIGWQ_F(r) equal to 1. However, because RIGWQH(r) = 0 makes the model rather fragile, we advise setting RIGWQH(r) to a small positive number, for example 0.01. In many cases, the estimated β is greater than 1, which suggests that α = 1 − β is negative. It is not a problem when β is not significantly different from 1 and the coefficient α is negative, but not different from zero. However, cases such as Australia (AUS) and the Netherlands (NLD) are problematic because β is significantly greater than 1 (see columns 2 and 3 in Table 3.7), and the coefficient α is significant and negative, which leads to a negative RIGWQH(r). Because negative rigidity parameters are not allowed in the model, RIGWQH(r) is set to 0.01 and RIGWQ_F(r) to 1 for Australia (AUS) and the Netherlands (NLD). In the second group of countries the composition of wealth is much more rigid than the composition of capital. In the case of Germany (DEU), United Kingdom (GBR), Japan (JPN), Sweden (SWE), United States (USA), and Singapore (SGP), β is not different from zero and α is not different from 1. For these countries we set RIGWQH(r) to 1 and RIGWQ_F(r) to 0.01. In the third group, consisting of Bolivia (BOL), Ecuador (ECU), Malaysia (MYS), Thailand (THA), Jordan (JOR), Morocco (MAR), Sri Lanka (LKA), Mauritius (MUS), and Norway (NOR), both compositions are rigid. For these countries we tested if α = β = 0.5, and for all nine countries except for Bolivia (BOL) and Ecuador (ECU) where β is close to 1, we could not reject the hypothesis that both capital and wealth have a very similar degree of rigidity. For these countries, we set RIGWQ_F(r) = RIGWQH(r) = 1. In Bolivia and Ecuador, the composition of capital is slightly more rigid than the composition of wealth. For these countries RIGWQ_F(r) is set to 1, and RIGWQH(r) is calculated as (1 − β)/β. Finally, in Austria (AUT) and France (FRA) both splits are very flexible. Because we cannot set rigidity parameters to zero in the model, we make them equal, in this case setting them both to 1. Using the regression results reported in Table 3.6, we set rigidity parameters for countries that are not part of the Kraay et al. (2000) database. From the very last row in Table 3.6, we conclude that in the “average” country the composition of capital is much more rigid than the allocation of wealth. However, β is significantly less than 1. The parameter RIGWQ_F(r) is set to 1, and RIGWQH(r) = (1 − β)/β = 0.06. Regional βs reported in Table 3.6
116
Alla Golub and Robert A. McDougall
may also be used. For industrial countries (INDC), Latin America countries (LAC), and East Asia and the Pacific (EAP), RIGWQ_F(r) is set to 1, and RIGWQH(r) is calculated as RIGWQH(r) = (1 − β)/β, where βs are estimates reported in the first part of Table 3.6. This calculation results in RIGWQH(r) = 0.05 for INDC, RIGWQH(r) = 0.13 in LAC, and RIGWQH(r) = 0.01 in EAP. For countries in the Middle East and North Africa (MENA), South Asia (SA), or Sub-Saharan Africa (SSA), the “average” country setting of the rigidity parameters − RIGWQ_F(r) = 1 and RIGWQH(r) = 0.06 − is more desirable because results for these three regions in Table 3.6 are based on only two or three countries. Because in the majority of countries considered the composition of capital is much more rigid than the allocation of wealth, for post-aggregation regions RIGWQ_F(r) can be set to 1 and RIGWQH(r) set to 0.06. Although there are differences in the relative rigidities of the composition of capital and allocation of wealth across countries, in the majority of countries the split between capital belonging to foreigners and capital belonging to local households is much more rigid than the split between equity in local firms and equity in foreign firms. The exceptions to this rule are offered by six industrialized countries. This apparent empirical regularity warrants further theoretical and empirical investigation. One possible explanation for this phenomenon may be based on asymmetric information. In the majority of countries domestic investors may have much better information about the investment opportunities in the domestic economy than do foreign investors. When good times come, domestic investors reallocate their portfolio to capture higher returns generated by domestic assets – and they do it faster and with greater ease than foreign investors. Similarly, when returns to domestic capital fall, domestic investors are likely to be the first to reallocate their portfolios toward foreign assets. This asymmetric information would result in the observed flexible allocation of wealth and rigid composition of capital.
4.4 Aggregation Issues The solution to the aggregation issue of the rigidity parameters is straightforward. Because in the majority of countries the split between capital belonging to foreigners and capital belonging to local households is much more rigid than the split between equity in local firms and equity in foreign firms, for post-aggregation regions RIGWQ_F(r) = 1 and RIGWQH(r) = 0.06. Region-specific parameters reported in Table 3.7 may also be used for regions or a country in a specific region. If the user would like to use different
Behavioral and Entropy Parameters in the Dynamic GTAP Model
117
values of the parameters at the country level reported in Table 3.7 to determine parameters for a region (i.e., aggregation of countries), an aggregate method is required at the regional level. In our aggregation method we use the wealth of the regional household (coefficient WQHHLD(r)) to aggregate RIGWQH(r) and the value of the domestic capital stock (coefficient WQ_FIRM(r)) to aggregate RIGWQ_F(r).
5. Conclusions and Summary This chapter develops econometric underpinnings for the behavioral and entropy parameters in the GDyn model to add realism to projections with the model. The long-run equilibrium in the model is defined as the convergence of the risk-adjusted net rates of return to capital across regions. In this chapter, the rates of return to capital are constructed using gross operating surplus obtained from the SourceOECD database and capital stock documented in Larson et al. (2000). These returns are used to test the hypothesis of convergence in the rates of return across countries and to measure the degree of international capital mobility. Based on econometric analysis, the null hypothesis of no convergence is rejected. The speed of convergence in net rates of return to capital in 20 OECD countries is 9% per year. Most likely, this speed of convergence would be lower if we included countries outside OECD, and hence a speed of convergence of 9% per year represents the upper bound of the desirable convergence of the net rates of return in the model. By changing the parameters determining the speed of (1) the lagged adjustment of the expected rate of return toward actual rate of return and (2) the lagged adjustment of the expected rate toward the target rate, the desirable degree of capital mobility can be achieved in the model. However, the same magnitude of the parameters may result in different degrees of capital mobility depending on regional aggregation. Therefore the lagged adjustment parameters should be calibrated on simulated rates of return for every new regional aggregation. The calibrations of lagged adjustment parameters for 3 × 3 and 7 × 7 aggregations of the GTAP Data Base demonstrated that the speed of convergence of 9% can be achieved when the lagged adjustment parameters are set to 0.5 and 0.4, respectively. This may suggest that in scenarios based on a more disaggregated GTAP Data Base, the desirable degree of capital mobility in the model can be achieved by setting the lagged adjustment parameters to some lower value. This chapter also discusses the approach to setting the elasticity of the rate of return with respect to the size of capital stock. The best solution is to
118
Alla Golub and Robert A. McDougall
calibrate the elasticity on the post-aggregation database. The second-best solution is to set it to 1. Setting the elasticity of the rate of return far from its model-consistent value may adversely affect convergence properties of the model. The lagged adjustment mechanisms determine regional investments, which include both domestic investment and foreign investment via the global trust. Savings of the regional household, in turn, are spent on investments in the domestic economy and investments in the global trust. Periodby-period decisions about the investments and savings composition affect the composition of capital and allocation of wealth of a region. In this chapter the parameters determining the relative rigidities of composition of capital and allocation of wealth in the GDyn model are estimated using a country portfolios database. Although there are differences in the relative rigidities of the composition of capital and allocation of wealth across countries, in the majority of countries the split between capital belonging to foreigners and capital belonging to local households is much more rigid than the split between equity in local firms and equity in foreign firms. The results of the econometric investigation are used to set the rigidity parameters in the model. References Bernard, A. B. and C. I. Jones. 1996a. “Comparing Apples to Oranges: Productivity Convergence and Measurement across Industries and Countries.” American Economic Review 86(5), 1216–38. Bernard, A. B. and C. I. Jones. 1996b. “Productivity across Industries and Countries: Time Series Theory and Evidence.” Review of Economics and Statistics 78(1), 135–46. ´ C., N. Loayza, and L. Serv´en. 2003. Do Capital Flows Respond to Risk and Calderon, Return? Policy Research Working Paper Series 3059. Washington, DC: World Bank. Dimaranan, B. V. and R. A. McDougall. 2002. Global Trade, Assistance, and Production: The GTAP 5 Data Base. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Available at http://www.gtap.agecon.purdue.edu/databases/ v5/v5 doco.asp. Engel, R. F. 1982. “Autoregressive Conditional Heteroskedasticity with Estimates of the Variance of United Kingdom Inflation.” Econometrica 50, 987–1007. Feldstein, M. and C. Horioka. 1980. “Domestic Savings and International Capital Flows.” Economic Journal 90, 314–29. Ianchovichina, E. I. 1998. International Capital Linkages: Theory and Application in a Dynamic Computable General Equilibrium Model. Ph.D. thesis, Department of Agricultural Economics, Purdue University. Kmenta, J. 1986. Elements of Econometrics, 2nd ed. New York: Macmillan. Kraay, A., N. Loayza, L. Serven, and J. Ventura. 2000, July. Country Portfolios. Working Paper Series No. 7795: 1-61. Washington, DC: National Bureau of Economic Research.
Behavioral and Entropy Parameters in the Dynamic GTAP Model
119
Kraay, A. and J. Ventura. 2000. “Current Accounts in Debtor and Creditor Countries.” Quarterly Journal of Economics 95, 1137–66. Larson, D. F., R. Butzer, Y. Mundlack, and A. Crego. 2000. “A Cross-Country Database for Sector Investment and Capital.” World Bank Economic Review 14, 371–91. Levin, A. and C.-F. Lin. 1992. Unit Root Tests in Panel Data: Asymptotic and Finite-Sample Properties. Discussion Paper 92–23. San Diego: Department of Economics, University of California, San Diego. Lewis, K. K. 1999. “Trying to Explain Home Bias in Equities and Consumption.” Journal of Economic Literature 37, 571–608. Nehru, V. and A. Dhareshwar. 1993. “A New Database on Physical Capital Stock: Sources, Methodology and Results.” Revista de Analisis Economico 8(1), 37–59. Nin, A., T. W. Hertel., K. Foster, and A. Rae. 2004. “Productivity Growth, Catching-Up and Uncertainty in China’s Meat Trade.” Agricultural Economics 31, 1–16. OECD. 1993. Methods Used by OECD Countries to Measure Stocks of Fixed Capital. Paris: OECD. SourceOECD. Annual National Accounts Volume II – Detailed Tables – Main Aggregates Volume 2004 release 01. Available at http://iris.sourceoecd.org/vl=158758/cl=20/nw= 1/rpsv/home.htm. Statistics Directorate OECD. 1983, 1987, 1991, 1994, 1996, 1997. Flows and Stocks of Fixed Capital. Paris: OECD. Walton, R. 2000a, January. “International Comparison of Profitability.” Economics Trends 554. Walton, R. 2000b, December. “International Comparison of Company Profitability.” Economic Trends 565.
FOUR
An Overview of the Dynamic GTAP Data Base The Data Base Construction and Aggregation Programs Robert A. McDougall, Terrie L. Walmsley, Alla Golub, Elena I. Ianchovichina, and Ken Itakura 1. Introduction The GDyn Data Base, used with the GDyn model, is based on the GTAP Data Base, which describes the global economy in a given year (2001 for the GTAP 6 Data Base; Dimaranan 2006). The GTAP Data Base is augmented with additional data required for GDyn, the dynamic version of the GTAP model. This chapter covers the preparation of these additional elements of the GDyn Data Base1 and the manipulation and aggregation of the GDyn Data Base with various utility programs. The focus is on the additional data requirements, with minimal discussion of the dynamic parameters (see Chapter 3). Readers interested in the underlying GTAP Data Base are referred to Dimaranan (2006). The GTAP Data Base (Dimaranan, 2006) is changed in the following three ways to obtain a database for GDyn: 1. Five new arrays, listed in Table 4.1, are added to the standard GTAP data file. The values of the regional savings array also change from those in the standard GTAP Data Base, although the dimensions of the array do not change. These changes are discussed in Section 2. 2. A new parameters file contains the seven new parameters used in the dynamic theory (Table 4.2). The methodology for estimating 1
At various times we have to refer to specific versions of the GDyn Data Base. We use a version-numbering system of the form :. The element is of the form .. We declare a new major version to mark a major difference in data content; a new minor version marks a change in format or a minor change in content. For example, version 6:1.0 of the GDyn Data Base is based on release 6 of the GTAP Data Base. Version 6:2.0 is based on the same GTAP release, but incorporates more recent GDyn-specific data. At the time of writing, the current version is 6:2.0.
120
An Overview of the Dynamic GTAP Data Base
121
Table 4.1. New arrays in the standard data file Header name
Coefficient name
Dimensions
Units
KHAT
KHAT
REG
year − 1
RRGT YQHT
RORGTARG YQHTRUST
REG REG
year − 1 $US Millions
YQTF
YQTFIRM
REG
$US Millions
YQHF
YQHFIRM
REG
$US Millions
Description normal capital stock growth rate target gross rate of return income of region from global trust income of global trust from firms in region income of region from local firms
econometrically, and in some cases calibrating these parameters, was examined in detail in Chapter 3. In this chapter the procedure for assigning values to these parameters in the standard GDyn Data Base is outlined. These changes are discussed in Section 3. 3. One parameter array from the standard GTAP parameters file – RORDELTA – is not used in the new investment theory and can be deleted from the file.
2. The GDyn Data File We construct the new GDyn data as follows. First, we procure statistics for foreign income receipts and payments; from these and standard GTAP data Table 4.2. Contents of the dynamic parameters file Coefficient name
Dimensions
Description
INC LAMBKHAT
REGa REG
LAMBRORGE
REG
LAMBRORG RORGFLEX
REG REG
RIGWQH
REG
RIGWQ F
REG
initial income coefficient of adjustment in estimated normal growth rate coefficient of adjustment in expected rate of return coefficient of adjustment in rate of return elasticity of rate of return to capital with respect to capital stock rigidity of allocation of wealth by regional household rigidity of source of funding of enterprises
a
REG denotes number of regions in the GDyn model and GTAP Data Base.
122
Robert A. McDougall et al.
we obtain the arrays: YQHT, the income from equity owned by each regional household in the global trust; YQTF, the income from equity owned by the global trust in each region’s firms; and YQHF, local household income from the ownership of equity in local firms. We then recognize that foreign income and receipts affect regional income. Because regional income must equal regional expenditure, regional expenditure must also change when foreign income is incorporated. Accordingly, we calculate new values for initial regional income, and regional expenditures are adjusted through a change in regional savings. The adjusted data on saving are stored in the standard data file under the header SAVE. Second, the normal rate of growth in the capital stock (KHAT) is estimated. The normal rate of growth in the capital stock is perceived by investors and assumed to converge to (1) the actual normal rate of growth in the capital stock and (2) the actual capital stock growth rate in the long-run equilibrium. To determine this perceived normal growth rate of capital, time-series estimates for investment are collected and used to construct time-series data for capital stock. The normal growth rate of capital (KHAT) is then set equal to the resulting geometric average growth rate of capital. Third, the target gross rate of return, RORGTARG, is set uniformly across regions and equal to the world average actual gross rate of return. In previous versions of the GDyn Data Base (6:1.0 and lower) the expected gross rate of return (RORGEXP) was also included in the GDyn Data Base. In the release 6:2.0 of the GDyn Data Base, RORGEXP is calculated within the GDyn model code. This change was made to allow users to specify different values of LAMBRORG, on which RORGEXP depends. RORGEXP is set individually in each region so as to reconcile the investment level implied by the investment theory with the level recorded in the database. Each of these stages is discussed in detail in the following subsections.
2.1 Income and Saving For the GDyn Data Base version 6:2.0, the data reference year is 2001.2 The standard country GDP dataset is based on the macro country dataset used in building the standard GTAP 6 Data Base, supplemented by estimates from the CIA World Factbook. For foreign income receipts and payments, we use the World Development Indicators (WDI) dataset from the World Bank. For the GDyn 2
Version 5:1.0 is based on data from 1997.
An Overview of the Dynamic GTAP Data Base
123
1.2
1
0.8
0.6
0.4
0.2
N Aus ew tr Ze alia al a C nd hi n Ja a pa K Ph o n ilip rea Si pin ng es a Th por ai e la nd Sr Ind i L ia a C nka an ad a U S M A Ve exi ne co Ar zeu ge la nt in B a U raz ru il gu Au ay s Fi tria nl a Fr nd G anc er e m an y U Ire K la nd N et Ita he ly rl Po and rtu s g Sp al ai T So u n ut rke h y Af ric a
0
WDI 1997
Kraay et al. (2000) 1997
WDI 2001
Figure 4.1. Comparison of share of foreign ownership based on income and wealth. Source: GDyn Data Base versions 5:1.0 and 6:2.0 and Kraay et al. (2000).
6:2.0 Data Base, we use the 2002 edition of the WDI. We extract two data series: r Foreign income receipts (BoP, current US$) r Foreign income payments (BoP, current US$) We obtain foreign receipts data for 154 countries and foreign payments data for 155. The shares of the foreign income payments and receipts in total income are also used to determine the share of foreign and domestic ownership of regional capital stocks. An alternative would be to use wealth/ownership data to imply the shares of income.3 Unfortunately, the set of countries for which wealth/ownership data are available is small, and data for the correct base year are unavailable. For example the Kraay et al. (2000) database contains 1997 data for only 38 countries, significantly less than the 154 countries for which data on income are available. For this reason we choose to use income data and to imply wealth from this data. The foreign ownership shares of regional capital obtained from the foreign income payments and receipts WDI data for 1997 and 2001 are compared to the shares from Kraay et al. (2000) in Figure 4.1. The data show that 3
This approach was used in earlier versions of the GDyn Data Base.
124
Robert A. McDougall et al.
the share of foreign ownership is higher when wealth (Kraay et al. 2000), as opposed to income from the WDI, is used to calculate the shares and that the 2001 shares of foreign ownership are higher than the 1997 shares, reflecting the fact that the share of foreign ownership has grown over this period. We draw comfort from the fact that the shares obtained from the Kraay et al. data and the ones from the WDI data follow the same cross-country pattern. New foreign income and saving data are constructed in five steps: r r r r r
Fill in missing values in the foreign income receipts data. Fill in missing values in the foreign income payments data. Balance foreign income receipts and payments. Aggregate foreign income data from countries to GTAP regions. Calculate new values for income earned by domestic capital, total regional income, and saving in each region.
The average ratio of foreign income receipts to GDP (FYRFFACT) is calculated for countries for which we have foreign income data. Foreign income receipts data are extended to cover all standard countries, assuming that, for the missing countries, the ratio of foreign income receipts to GDP is equal to the average ratio for countries for which data are available; that is, we calculate the missing value of foreign income receipts for country c as FYRFFACT∗GDP(c). We take the GDP values from the standard country GDP dataset. We use a similar calculation to extend the foreign income payments data to cover all standard countries. The foreign income data are unbalanced: The world total for foreign income receipts is not equal to the world total for foreign income payments. A new common target total is calculated as the geometric mean of the two new totals.4 The data are then balanced by rescaling both receipts and payments to match the common target total, adjusting the values for all countries. The balanced foreign income data are then aggregated across countries to the GTAP regions. This gives us the arrays YQHT and YQTF. Income from capital in each region (VOA) is then obtained from the standard GTAP standard data file. Subtracting foreign income payments (YQTF) from this figure, we then obtain estimates for local income from domestic capital in each region (YQHF). In so doing, we assume implicitly that all foreign income payments are income from capital. Taking account 4
Total of all factor income payments/receipts for all standard countries (including countries for which data were available and countries for which data were computed).
An Overview of the Dynamic GTAP Data Base
125
of the new foreign income payments and receipts data, we calculate new values for total income, INC, and for saving in each region, SAVE: SAVE(r) = INCOME(r) − PRIVEXP(r) − GOVEXP(r) ∀r ∈ REG, where INCOME(r) =
(4.1)
VOA(i, r) + YQHHLD(r)
i∈ENDWNA
+ TAXREV(r) ∀r ∈ REG.
(4.2)
2.2 Normal Rates of Growth in Capital Stocks For the normal rate of growth in the capital stock, KHAT, we use timeseries investment data (referred to as the macro times-series data5 ), as well as investment and capital stocks data for 2001 from the GTAP Data Base (referred to as the single-year macro dataset6 ). For version 6:2.0 of the GDyn Data Base, the macro time-series data are available for 117 countries for the period 1990–2005 from the WDI, whereas the 2001 investment data from the standard GTAP Data Base are available for 226 countries7 (we refer to this list of 226 countries as the GTAP standard country classification8 ). 5 6 7
8
The macro time-series dataset contains time-series investment, GDP, and other macro variables. In this chapter, however, we are only interested in the investment data. The single-year macro dataset also contains other macro data, including GDP, investment, capital stocks, consumption, and government spending. The single-year macro dataset for 2001 uses its own country classification, which is incomplete relative to the list of 226 countries on which the GTAP Data Base is based. We fill in the gaps using GDP data from the World Bank that are consistent with the GTAP data. For investment, we calculate a scaling factor equal to the ratio of the two sums over the countries covered in the single-year macro dataset. The numerator is the sum of investment values from the macro dataset; the denominator is the sum of GDP values from the standard country GDP dataset. We apply this scaling factor to GDP values from the standard country GDP data set to fill in the investment values missing from the single-year macro dataset. We use similar methods to fill in the missing countries in the macro data for capital stocks, GDP, and the rest. These procedures are undertaken as part of the standard GTAP Data Base construction procedures (Dimaranan 2006). The GTAP standard country classification includes Aruba, Afghanistan, Angola, Anguilla, Albania, Andorra, Netherland Antilles, United Arab Emirates, Argentina, Armenia, American Samoa, Antigua and Barbuda, Australia, Austria, Azerbaijan, Burundi, Belgium, Benin, Burkina Faso, Bangladesh, Bulgaria, Bahrain, Bahamas, Bosnia Herzagova, Belarus, Belize, Bermuda, Bolivia, Brazil, Barbados, Brunei, Bhutan, Botswana, Central African Republic, Canada, Switzerland, Chile, China, Cote D’Ivoire, Cameroon, Congo, Peoples Democratic Republic of Congo, Cook Islands, Colombia, Comoros, Cape Verde, Costa Rica, Cuba, Cayman Islands, Cyprus, Czech Republic, Germany, Djibouti, Dominica, Denmark, Dominican Republic, Algeria, Ecuador, Egypt, Eritrea, Spain, Estonia, Ethiopia, Finland, Fiji, Falklands, France, Faroe islands, Micronesia, Gabon, United Kingdom, Georgia,
126
Robert A. McDougall et al.
For each region we define the normal rate of growth in the capital stock as the rate at which the capital stock in that region can grow, without any increase or decrease in the rate of return. We store a region-specific parameter KHAT representing the normal rate of growth as estimated by investors in the data file. We proceed in five steps: r Fill in missing values and disaggregate country groups in the macro
time-series data.
r From the capital stock data in the single-year macro dataset and the
investment data in the macro time-series dataset, construct time-series data for capital stocks. r Aggregate the capital stock time-series data from countries to GTAP regions. r Calculate the geometric average rates of growth in capital stocks. r Set KHAT equal to those average growth rates. As mentioned earlier, the macro time-series dataset is available for 117 countries/regions. The dataset is therefore incomplete and somewhat aggregated relative to the standard GTAP country classification of 226 countries (notably, it treats the Soviet Union as a single country). We use data from the completed single-year macro dataset to fill in missing values and disaggregate aggregate values. We fill in the missing values in the investment time series by calculating, for each year in the time-series data, a country-generic investment scaling factor equal to the ratio of two values for total investment in countries Ghana, Gibraltar, Guinea, Guadeloupe, Gambia, Guinea Bissau, Equatorial Guinea, Greece, Grenada, Greenland, Guatemala, French Guiana, Guam, Guyana, Hong Kong, Honduras, Croatia, Haiti, Hungary, Indonesia, India, Ireland, Iran, Iraq, Iceland, Israel, Italy, Jamaica, Jordan, Japan, Kazakhstan, Kenya, Kyrgyzstan, Cambodia, Kiribati, St Kitts, South Korea, Kuwait, Laos, Lebanon, Liberia, Libya, Saint Lucia, Liechtenstein, Sri Lanka, Lesotho, Lithuania, Luxembourg, Latvia, Macau, Morocco, Monaco, Moldova, Madagascar, Maldives, Mexico, Marshalls, Macedonia, Mali, Malta, Myanmar, Mongolia, North Marianas, Mozambique, Mauritania, Montserrat, Martinique, Mauritius, Malawi, Malaysia, Mayotte, Namibia, New Caledonia, Niger, Norfolk Island, Nigeria, Nicaragua, Niue, Netherlands, Norway, Nepal, Nauru, New Zealand, Oman, Pakistan, Panama, Peru, Philippines, Palau, Papua New Guinea, Poland, Puerto Rico, North Korea, Portugal, Paraguay, Occupied Palestine, French Polynesia, Qatar, Reunion, Romania, Russian Federation, Rwanda, Saudi Arabia, Serbia Montenegro, Sudan, Senegal, Singapore, Saint Helena, Solomon Islands, Sierra Leone, El Salvador, San Marino, Somalia, St Pier Miq, Sao Tome, Suriname, Slovakia, Slovenia, Sweden, Swaziland, Seychelles, Syria, Turks Caicos, Chad, Togo, Thailand, Tajikistan, Tokelau, Turkmenistan, Timor Leste, Tonga, Trinidad and Tobago, Tunisia, Turkey, Tuvalu, Taiwan, Tanzania, Uganda, Ukraine, Uruguay, United States, Uzbekistan, St Vincent, Venezuela, UK Virgin Islands, USA Virgin Islands, Vietnam, Vanuatu, Wallis Futuna, Samoa, Yemen, South Africa, Zambia, and Zimbabwe.
An Overview of the Dynamic GTAP Data Base
127
covered by the time-series data: one value taken from the time-series dataset (numerator) and the other from the single-year macro dataset (denominator). We then apply these scaling factors to investment values for the required countries taken from the single-year macro dataset. To disaggregate timeseries investment estimates for country aggregates, we use shares calculated from the investment data in the single-year dataset. We fill in the missing countries in the GDP time series by calculating for each year in the timeseries data a country-generic GDP scaling factor, using the same approach as that taken for the investment scaling factor. To disaggregate time-series investment estimates for country aggregates, we use shares calculated from the GDP data in the single-year dataset to apportion the aggregated investment time-series data across the countries in the aggregated region. We then use the single-year macro dataset and the disaggregated (226 country level) macro time-series dataset to construct time-series data for capital stocks in each country. Because the two input datasets may use different units, we first put them on a common basis. To do so we rescale the investment time series using a country-specific scaling factor: the ratio of the GDP value from the single-year macro dataset to the GDP value for the same reference year from the macro time-series dataset. This rescaling ensures that the 2001 time-series investment data are equal to the investment in the 2001 single-year macro dataset, and it accounts for any differences in units. We then construct the capital stock time-series data using the perpetual inventory method (PIM), taking for capital stock base values, the capital stock data from the single-year macro dataset; for gross investment, the rescaled investment time series, and for the depreciation rate, a separately supplied country-generic value. We aggregate the capital stock time-series data from countries to GTAP regions. For each region, we set KHAT equal to the average actual rate of growth between the first and last year covered by the capital stock time series. The normal growth rates (KHAT) are within the range of 1.9% to 14.5%, except for Mozambique, which has a KHAT of 22%.9
2.3 Target Rate of Return Finally we include in the GDyn data file a region-dimension array RRGT. This is the array containing data on the target gross rate of return used by investors in allocating capital between regions (RORGTARG). 9
This is most likely due to the very low value of capital stocks in Mozambique in 2001 relative to the high value of investment in 2001.
128
Robert A. McDougall et al.
For each region, we set the target rate of return equal to the world average rate of return, which depends on the global earnings to capital (EKTOT) relative to the global value of capital (VKTOT): ⎡ RORGTARG =
VOA(i, r)
EKTOT ⎢ i∈Capital r∈REG =⎣ VKTOT VK (r)
⎤ ⎥ ⎦,
(4.3)
r∈REG
where VOA(i, r) is the value of output at agents’ prices of capital in region r, and VK(r) is the value of capital stocks in region r. Thus, in the database, although not necessarily in the model theory, target rates of return are uniform across regions.
3. Parameters The new behavioral parameters must then be included in a new dynamic parameters file. For all the arrays in the parameters file, except for initial income INC, the parameter values are based on the results of the empirical analysis undertaken in Chapter 3. For a more detailed explanation of the parameters used in the GDyn model the reader is referred to Chapter 3. Here we summarize the GDyn parameters and provide a brief account of their chosen default values in the GDyn Data Base.
3.1 INC The first new parameter, listed in Table 4.2, is INC. This parameter is the initial income level in US$ millions in the initial GDyn Data Base. It is used for calculating welfare measures in multiperiod simulations and remains unchanged over the simulation time.
3.2 Lagged Adjustment Parameters The lagged adjustment parameters include LAMBRORG, the coefficient of adjustment in the actual rate of return; LAMBRORGE, the coefficient of adjustment in the expected rate of return; and LAMBKHAT, the coefficient of adjustment in the perceived normal growth rate of capital stock. In the standard GDyn 6:2.0 Data Base, LAMBRORG and LAMBRORG are set equal to 0.4, and LAMBKHAT is set equal to 0.2 for all regions. Using the calibration procedures outlined in Chapter 3, we find that these values lead to empirically consistent convergence of the rates of return in a typical GDyn
An Overview of the Dynamic GTAP Data Base
129
aggregation (approx. 10 regions) and reduce the occurrence of negative investment. Those readers wishing to calibrate these parameters to their own aggregation or to alter the values to obtain alternative rates of convergence are referred to Chapter 3 for further details on the calibration procedure and effects associated with altering these parameters.
3.3 Flexibility of the Gross Rate of Return RORGFLEX is the elasticity of the rate of return to capital with respect to capital stock. This parameter is set equal to 1 for all regions, which was shown in Chapter 3 to be close to the true value of RORGFLEX given the settings for LAMBRORG stated earlier. The true values of RORGFLEX can be obtained via calibration that follows the techniques outlined in Chapter 3. It is highly recommended that users do not alter RORGFLEX arbitrarily, because large differences between the chosen and true values can adversely affect the convergence properties of the model (see Chapter 3).10
3.4 Rigidity Parameters There are two rigidity parameters in the GDyn model: RIGWQH, which is the rigidity of the allocation of wealth by regional household, and RIGWQ_F, the rigidity of source of funding of enterprises. We use the empirical estimates of rigidity parameters (as specified in Table 3.7) to obtain rigidity parameters for the 226 standard GTAP countries. We used the following procedure to obtain the rigidity parameters. Rigidity estimates for 57 countries and 5 regions, including Benelux, were provided in Chapter 3. After applying the rigidity parameters of Benelux to Belgium and Luxembourg, data for 59 countries are available. Of these 59 countries, 42 countries are contained in the GTAP 6 Data Base. For these 42 countries in the GTAP 6 Data Base the country-specific rigidity estimates come from Table 3.7. The rigidity parameters for the four regions11 from Table 3.7 were then assigned to the 16 GTAP regional groupings used in the production of the GTAP 6 Data Base according to Table 4.3.12 10
11 12
It should also be noted that calibrated RORGFLEX is dependent on LAMBRORG, and hence changes in LAMBRORG affect the true value of RORGFLEX. If users choose calibrated RORGFLEX, as opposed to RORGFLEX = 1, and wish to alter LAMBRORG, they are advised to recalibrate RORGFLEX to obtain new true value. All Industrialized countries (INDC), All Latin America (LAC), All East Asia Pacific (EAP), and All. Note that the rigidity estimates for the remaining 17 (59 less 42) countries not in the GTAP 6 Data Base were not used because of aggregation problems. The alternative would have been to allocate estimates to all 226 countries and then aggregate the parameters of the
130
Robert A. McDougall et al.
Table 4.3. Allocation of rigidity parameters across 16 GTAP regional groupings Based on the following countries/ regions from Table 3.7 RIGWQH RIGWQ F Oceania East Asia South East Asia South Asia North America South America Central America Caribbean Europe Eastern Europe Former Soviet Union Middle East North Africa Southern Africa Central and Eastern Africa Western Africa
All ALL EAP ALL EAP All All All LAC CRI, SLV, GTM, HND, NIC All All All All All All All All All
0.06 0.01 0.01 0.06 0.06 0.13 0.01 0.06 0.06 0.06 0.06 0.06 0.06 0.06 0.06 0.06
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
The remaining standard countries (226 less 42) were then assigned the rigidity parameters of the region (one of 16) to which they belonged.13 The rigidity parameters for the 226 countries were then aggregated up to the 87 regions in the GTAP 6 Data Base, using simple averages. Because all countries not covered in the GTAP Data Base as separate regions are assigned regional averages, the weights used to aggregate are irrelevant.
4. Creating Aggregations of the GDyn Data Base and Parameters The aggregation program developed for the GDyn model allows the user to aggregate up the database and parameters from the GTAP regional and sectoral level to an aggregation that can be used in their analysis/application. In developing the aggregation procedure, issues arise concerning the choice of aggregation formulas. It is not obvious that all parts of the aggregate
13
countries to obtain regional estimates. This method was not used because of the lack of country estimates to aggregate to those regions and the difficulties in obtaining data for weighting in such an aggregation. Because all the regional rigidity parameters were similar, this allocation method did not yield any real differences between regions; hence where countries within the region were not included in the econometrics the parameters for “ALL” were used.
An Overview of the Dynamic GTAP Data Base
131
database should be constructed by aggregating the original database. Alternative methods such as calibration may be appropriate for some parameters. Furthermore, for some parameters, it is not obvious which aggregation formulas should be used. In general, we calculate aggregate parameters as weighted averages of disaggregate parameters, but the choice of weights is not always obvious. In earlier versions, the parameters KHAT and RORGFLEX were set by an iterative procedure involving calibration simulations with the GDyn model. This process was done after aggregation. This complicated procedure, however, goes against our goal of providing a quick and easy aggregation procedure. Therefore in the latest version of the GDyn Data Base, we determine the aggregate KHAT and RORGFLEX parameters not by post-aggregation calibration but by applying simple aggregation formulas. With this new approach, KHAT and RORGFLEX are unlikely to match exactly the underlying or “true” parameters. This raises no particular problems for KHAT. The model theory does not require that it should agree with the true normal rate of growth in the capital stock, because KHAT is a self-correcting parameter that varies through time to eliminate divergences between the perceived and true rate of normal growth. In contrast, RORGFLEX is not self-correcting, and problems associated with the aggregation of RORGFLEX and the other parameters may arise, as discussed in Chapter 3.
4.1 KHAT The parameter KHAT is updatable and therefore held in the data file. In aggregating it, we use capital stock weights calculated from the array VKB in the data file. VKB(r) .KHAT(r) , ∀r ∈ REG, ∀n ∈ AREG AKHAT(n) = AVKB(n) ∀r in region n (4.4) where r AKHAT(n) is the aggregated KHAT for region n r VKB(r) is the value of capital stocks for country r (in aggregated region
n)
r AVKB(n) is the aggregated value of capital stocks for the region n
132
Robert A. McDougall et al.
r KHAT(r) is the normal growth rate of capital for country r (in aggre-
gated region n).
4.2 RORGFLEX The parameter RORGFLEX represents the elasticity of the expected rate of return with respect to the capital stock. The aggregate elasticity is φ=
∂ log R E , ∂ log K
(4.5)
where RE represents the expected (gross) rate of return and K, the capital stock. Letting the subscript i run over the corresponding disaggregate regions, we have RE =
Ei R E ,i K i E = = , K K K i i
(4.6)
where E represents earnings; therefore, d logR E =
SE ,i d logR E ,i +
i
SE ,i d logK i − d logK ,
(4.7)
i
where SE,i represents the share of region i in aggregate capital earnings. Then, assuming uniform relative change in the capital stock – that is, for all disaggregate regions i, d logKi = d logK – we have d log R E =
S E ,i d log R E ,i ,
(4.8)
i
so φ=
i
S E ,i
∂ log R E ,i ∂ log R E ,i = S E ,i = S E ,i φi . ∂ log K ∂ log K i i i
(4.9)
Therefore, for RORGFLEX, earnings weights are appropriate. More precisely, because the rate of return in question is the expected and not the actual rate of return, we should use expected, not actual, earnings weights. Note that the assumption of uniform capital stock growth is not innocuous. We should have obtained a different aggregation formula if we had assumed, for example, uniform relative change in the rate of return.
An Overview of the Dynamic GTAP Data Base
133
4.3 LAMBRORGE The parameter LAMBRORGE is used to define the error-correction effect for the expected rate of return: ∂ log R E = E R E , ∂t
(4.10)
where E is a measure of error. Recalling equation (4.1), and putting ∂ log∂t K i = ∂ log K , we have again: ∂t S E ,i d log R E ,i , (4.11) d log R E = i
Then, we have 1 ∂ log R E ,i S E ,i E i ∂t 1 = S E ,i E i R E ,i . E i
R E =
(4.12)
Then, assuming uniform errors Ei (or no correlation between errors Ei and speed of adjustment R E ,i ), we have R E = S E ,i R E ,i . (4.13) i
So again, expected earnings weights are appropriate. Given that the parameters for all regions are set equal to 0.2, the aggregation procedure is not relevant under the standard settings; however, if changes are made to specific countries/regions, this procedure would become relevant. This is also true for the aggregation of LAMBRORG and LAMBKHAT.
4.4 LAMBRORG The parameter LAMBRORG represents the derivative of the required rate of growth in the rate of return with respect to a measure of divergence between the expected and the target rate of return. At any time t, investors hold a perception or expectation not only of the current rate of return but also of the rate of return at future times. They require that this expected rate of return should grow through time at a rate that depends on the divergence between the expected and the target rate. If the expected rate exceeds the target rate, the expected rate should grow at a positive rate; if the target rate exceeds the expected rate, the expected rate
134
Robert A. McDougall et al.
should decline. The derivative of the rate of growth in the expected rate of return with respect to the measure D of divergence is LAMBRORG or R R . Now, R R =
∂ d log R E ∂D dt
so, again recalling equation (4.1) and assuming R R =
(4.14) ∂ log K i ∂t
d log R E ,i ∂ S E ,i . ∂D i dt
=
∂ log K ∂t
, we have (4.15)
Then, ignoring any effect of change in divergence on the earnings shares, we have S E ,i R R,i . (4.16) R R = i
So once again, we use expected earnings weights. As in the case of LAMBRORGE, LAMBRORG parameters for all regions are set equal to 0.2.
4.5 LAMBKHAT The parameter LAMBKHAT or Kˆ represents the derivative of the expected normal rate of growth in the capital stock with respect to a measure of error in expectations: ∂ d log Kˆ ∂E dt d log Kˆ i ∂ = S K ,i ∂E i dt = S K ,i K ,i .
Kˆ =
(4.17)
i
For LAMBKHAT we use capital stock weights. Again the parameters for all regions are set equal to 0.2.
4.6 RIGWQH The parameter RIGWQH describes the rigidity of the allocation of the wealth of the regional household between local and domestic assets. To aggregate it we use the wealth of the regional household (coefficient WQHHLD) for weights.
An Overview of the Dynamic GTAP Data Base
135
4.7 RIGWQ F The parameter RIGWQ_F describes the rigidity of the sourcing of funds of enterprises between locally and foreign sources. To aggregate it we use the value of the domestic capital stock (coefficient WQ_FIRM) for weights.14 References Dimaranan, B. V. (ed.). 2006. Global Trade, Assistance, and Production: The GTAP 6 Data Base. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Kraay, A., N. Loayza, L. Serven, and J. Ventura. 2000, July. Country Portfolios. Working Paper Series No. 7795: 1–61. Washington, DC: National Bureau of Economic Research.
14
If we derived the weights to be used with RIGWQH and RIGWQ_F with the same level of care as the earlier parameters, we would obtain some strange formulas, in which the weights sum to less than 1, except in the special case in which the composition of wealth (for RIGWQH) or sourcing of funds (for RIGWQ_F) is uniform across disaggregate regions within disaggregate regions. This is similar to the situation we meet with domestic/import or factor substitution elasticities (details available on request). In those situations in standard GTAP, we ignore the complexities and use plain weights summing to 1; we do the same here. Then the choice of weighting coefficient can be justified by this argument: For RIGWQH, as the share of some disaggregate region in the aggregate region’s wealth approaches 1, the aggregate region’s rigidity parameter should approach that disaggregate region’s parameter; similarly for RIGWQ_F and domestic capital.
FIVE
A Baseline Scenario for the Dynamic GTAP Model Terrie L. Walmsley, Betina V. Dimaranan, and Robert A. McDougall
1. Introduction The development of a baseline scenario is an important component of assessing the impact of a policy issue with a dynamic model. A baseline depicts how the world economy might be expected to change, over a given period of time, if the policy were not implemented. The baseline scenario should therefore reflect as closely as possible the changes expected to occur in the world economy, excluding the particular policy of interest. The components of the baseline will depend on the regional/sectoral aggregation chosen for the study and the policy being examined. A good baseline will include projections for macroeconomic variables – such as real GDP, population, technological change, and primary factor growth rates – for each of the regions being examined, as well as key policies that have already been agreed on or are expected to affect the regions/sectors or the policy scenario being examined. New baselines are often developed each time a new policy issue is addressed. For example, a baseline built to examine China’s accession to the WTO will not be suitable for examining the impact of free trade agreements within Southern Africa, most notably because the regional aggregation will differ between the two simulations but also because the policies we would want to include in the baseline would differ considerably. Moreover, the key macroeconomic variables may differ depending on the policy simulations; for example, tracking changes in investment might be important for examining the impact of a policy on investment in the East Asian economies, whereas tracking changes in population or labor force might be important if you were examining the impact of improved health policies in Southern Africa. Although it is quite difficult to develop a common baseline for use with all policy scenarios, some elements, particularly the basic macroeconomic projections, are common to many policy scenarios. There are significant 136
A Baseline Scenario for the Dynamic GTAP Model
137
benefits to be gained, in terms of time saved and comparability of results, from sharing a common set of forecasts for these variables. In this chapter we discuss the construction and implementation of a baseline scenario for the GDyn model. The chapter utilizes the baseline used in Chapter 8 by Walmsley, Hertel, and Ianchovichina to examine China’s accession to the WTO as an example. The chapter is divided into five sections. After the introduction, the next two sections discuss the macroeconomic and policy data used in the creation of the baseline. They provide an in-depth account of the data, including the macroeconomic and policy projections, and of the methods used to fill in missing projection values and then extrapolate the macroeconomic projections. Because the macroeconomic projections are expected to be common across many baselines, a full set of projections has been developed for all countries, which can then be aggregated into any subset of GTAP regions. In contrast, the policy projections are specific to the simulation being undertaken, and hence we concentrate on the methods used to develop these projections. The fourth section then outlines how these projections are implemented into the model, and the fifth section examines alternative ways in which technological change can be treated when creating a baseline. Finally, we conclude the chapter in the fifth section.
2. Macroeconomic Projections This section discusses the macroeconomic projections: projected values of gross domestic product, gross domestic investment, capital stocks, population, skilled labor, and unskilled labor for the period 2001 to 2020 for as many of the 226 standard GTAP countries as possible. The section is divided into two subsections: The first contains a description of the projections, including the macroeconomic projections and macroeconomic data, and the second subsection describes the procedures used to fill in missing data.
2.1 Sources of Projections Projections Projections were obtained for gross domestic product, gross domestic investment, population, total labor, and skilled labor as follows: r Gross domestic product, gross domestic investment, and population pro-
jections were available for 133 countries/regions for the period 1992 to 2010 (with projections from 1998 to 2010).1 1
Projections data are consistent with forecasts used in Global Economic Prospects, The World Bank.
138 Terrie L. Walmsley, Betina V. Dimaranan, and Robert A. McDougall r Labor force projections in the form of number of male and female
workers were available for 205 countries/regions. Projections were provided on a 5-year basis from 1990 to 2020. Data on male and female workers were added together to obtain projections for the total labor force.2 r Skilled labor projections were obtained from two sources. For the less developed countries, projections of the share of secondary and tertiary educated labor as a proportion of the population were obtained for 71 developing countries. These were 5-year projections from 1990 to 2020 and were obtained from Ahuja and Filmer (1995). For the developed economies, skilled labor projections were based on projected skilled labor shares for 12 developed/developing regions over the period 1994 to 2050. These data were obtained from the CPB (1999).
Other Data Required In addition to projections, macroeconomic data for the base or initial year (2001) were also collected for all standard countries. GDP and population data were obtained for each of the countries from the World Bank, supplemented by data from the CIA World Factbook.3 Other macroeconomic variables, including gross domestic investment and capital stocks, were either obtained directly from the World Bank or GDP shares were used to estimate their value. This base-year data were used to scale data, fill in missing values, and obtain capital stock projections. In this chapter, these data are usually referred to as the base-year data.
2.2 Missing Data In all cases, the projections obtained from the various sources listed earlier were incomplete and in some cases incompatible. Some processing was required to put the projections into a common format and to ensure that there were values for all 226 standard countries and for all years of interest (1995 to 2020). This section outlines the assumptions made and the steps taken to achieve this objective. The macroeconomic projections are discussed in turn. 2 3
Projections data are consistent with forecasts used in Global Economic Prospects. These data are consistent with those used in the production of the GTAP 6 Data Base (Dimaranan 2006).
A Baseline Scenario for the Dynamic GTAP Model
139
Gross Domestic Product, Gross Domestic Investment, and Population A number of steps were undertaken to obtain gross domestic product, gross domestic investment, and population for all 226 standard countries. These steps included extrapolating, disaggregating regions, filling in projections for missing countries, scaling, and finally calculating yearly growth rates. Each of these stages is discussed in turn. Extrapolating. Because projections were only available for the period 1998 to 2010, the first step was to determine the growth rate that should be used to extrapolate from 2010 to 2020. First, it was assumed that the per capita growth rate would be used to extrapolate,4 with the population projections used to determine per capita growth rates. The growth rate used for extrapolation was the average growth rate in the final 5 years, usually 2005 to 2010. There were some cases, the United States for example, where data were not available for the entire period. In this case the last 5 years of available projections were used to extrapolate. Disaggregating Regions. Although most of the projections data were for individual countries, in a limited number of cases projections data were provided for an aggregate region. For example, projections were obtained for Belgium-Luxembourg, rather than for these two countries individually. For these aggregate regions the projections are divided across the individual countries in those regions using the relevant base-year share. For example, the real GDP projections of Belgium-Luxembourg are divided using macroeconomic data on their respective levels of GDP in the base year, and population projections are divided according to macroeconomic data on population, thus assuming that the growth rates for all countries within the region are the same as the growth rate for the region as a whole. Filling in Missing Countries. The next step involved providing projections for those standard countries for which projection data were not available. Data were missing for only 18 very small countries. The growth rates of these 18 countries were assumed to equal the average growth rate of the countries for which data were available. It was necessary to estimate the 4
In the case of gross domestic investment, a check was made to ensure that the ratio of gross domestic investment to gross domestic product did not change dramatically after 2007. The difference between extrapolating based on growth rates and per capita growth rates was not significant because population growth rates were also extrapolated.
140 Terrie L. Walmsley, Betina V. Dimaranan, and Robert A. McDougall
growth rates for these 18 countries to ensure that the estimated growth of the Rest of World (ROW) region was not overly biased toward the growth rate of one particular country in the region when aggregated. Scaling. The projections data were based on 1992 prices. In addition, projections for the base year were often inconsistent with actual data obtained from the GTAP Data Base. To ensure consistency between the projections and the GTAP Data Base, all projections were scaled so that the base-year projection (1995) was equal to the equivalent value in the GTAP 4 or 5 Data Base. Calculating Growth Rates. Finally the projections were converted into yearly growth rates. The resulting growth rates for real gross domestic product, gross domestic investment, and population are shown in Figures 5.1–5.2, 5.3–5.4 and 5.5–5.6 respectively, for a selection of regions.5 Figure 5.1 shows that on average the growth rate for real GDP in North America is higher than in Europe. Japan shows a large decline in growth rate in 1998/99, which only gradually returns to its previous levels in 2005. This poor performance reflects the difficulties experienced by Japan in the 1990s as a result of a crisis in its own economy and in Asia in general. Similar patterns can be seen in gross domestic investment (Figure 5.3). Figure 5.2 again shows the negative effects of the Asian and Brazilian crises on Southeast Asia and Latin America. China’s impressive growth rates over the 1990s are also clear. Similar patterns can also be seen in gross domestic investment (Figure 5.4). Figure 5.5 illustrates the low and declining population growth rates currently being experienced across the developed economies. Population growth rates in the developing economies are much higher (Fig. 5.6).
Labor Projections Projections were obtained for total labor, skilled labor, and unskilled labor. As was the case for gross domestic product, gross domestic investment, and population, a number of steps were undertaken to obtain labor force projections for all standard countries. These included disaggregating regions, filling in projections for missing countries, and filling in data for missing years. The three types of labor projections are discussed in turn.
Total Labor Force Projections Disaggregating Regions. Although most of the labor projections data were for individual countries, in a limited number of cases labor projections 5
The projections were aggregated into 11 regions.
141 5 4
Growth in Real GDP
3 2 1 0 -1 -2 -3 -4
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 Time 1 NAmerica
2 WEurope
3 Japan
Figure 5.1. Gross domestic product growth rates: North America, Western Europe, and Japan.
142 Terrie L. Walmsley, Betina V. Dimaranan, and Robert A. McDougall 15
5
0
-5
-10
19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 20 13 20 14 20 15 20 16 20 17 20 18 20 19 20 20
Growth in Real GDP
10
Time 4 China
7 SEA
8 SoAsia
9 LatinAM
10 AfrMidE
Figure 5.2. Gross domestic product growth rates: China, South East Asia, South Asia, Latin America, and Africa-Middle East.
were provided for regional aggregates. For example, labor projections were obtained for a region “other Europe.” For these composite regions the labor projections were divided across the countries contained in those regions using population shares. Thus it was assumed that the growth rates for all countries within the region were the same as the growth rate for the region as a whole. Filling in Missing Countries. The next step involved providing labor force projections for those standard countries for which no labor force projections data were available. Labor projections were filled in by assuming that the growth rates in those regions where no projections were available were equal to the average growth rate of the countries for which data were obtained. Filling in the Missing Years. Labor force projections were available in 5-year intervals from 1990 to 2020. Thus it was necessary to fill in projections for the intermediate years. To do this it was first necessary to find the average yearly growth rate for each of these 5-year periods. It was assumed that growth rates for a particular country were equal for all years within the 5-year interval. These growth rates were then used to obtain projections for each year. This is the reason for the resulting 5-year steps in the labor force growth rates.
143
5
4
Growth in Gross Domestic Investment
3
2
1
0
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 -1
-2
-3
-4
Time 1 NAmerica
2 WEurope
3 Japan
Figure 5.3. Gross domestic investment growth rates: North America, Western Europe, and Japan.
144 Terrie L. Walmsley, Betina V. Dimaranan, and Robert A. McDougall
30 20 10 0
19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 20 13 20 14 20 15 20 16 20 17 20 18 20 19 20 20
Growth in Gross Domestic Investment
40
-10 -20 -30 Time 4 China
7 SEA
8 SoAsia
9 LatinAM
10 AfrMidE
Figure 5.4. Gross domestic investment growth rates: China, South East Asia, South Asia, Latin America, and Africa-Middle East.
Skilled Labor Projections As mentioned earlier, skilled labor force projections were obtained from two sources: Five-year projected shares of labor force were obtained for the developing countries, whereas yearly projected shares were obtained for a number of developed regions. To obtain a complete set of projections, a number of steps had to be applied to both sets of data. For developing countries, missing years in the skilled labor shares needed to be filled in 1.4 1.2
0.8 0.6 0.4 0.2
97
98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 20 13 20 14 20 15 20 16 20 17 20 18 20 19 20 20
-0.2
19
19
96
0
19
Growth in Population
1
Time 1 NAmerica
2 WEurope
3 Japan
Figure 5.5. Population growth rates: North America, Western Europe, and Japan.
A Baseline Scenario for the Dynamic GTAP Model
145
3
2
1.5
1
0.5
0
19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 20 13 20 14 20 15 20 16 20 17 20 18 20 19 20 20
Growth in Population
2.5
Time 4 China
7 SEA
8 SoAsia
9 LatinAM
10 AfrMidE
Figure 5.6. Population growth rates: China, South East Asia, South Asia, Latin America, and Africa-Middle East.
and projections determined. For the developed economies, regional skilled labor shares were attributed to the developed countries and projections calculated. The two sets of projections were combined and any missing countries filled. Each of these stages is discussed in turn. Developing Countries. Projections for developing countries are initially given as projected shares of the labor force in 5-year intervals from 1990 to 2020. Data are available for both tertiary and secondary education, and tertiary education was used to estimate skilled labor growth rates. To obtain the final projections, two steps were undertaken. First, it was necessary to fill in projections for the intermediate years by calculating the average yearly growth rate for each of these 5-year periods. It was then assumed that within a given 5-year period yearly growth rates of shares for a particular country are equal. These growth rates were then used to obtain the projected shares for each year. Second, the shares were combined with the total labor force projections to determine the projected number of people with tertiary and secondary education. Developed Countries. The share of skilled labor in the total labor force was obtained for 12 regions. These regions included both developing and developed countries. Twenty-five developed economies were then given the projected shares of the region in which they were located. These shares were
146 Terrie L. Walmsley, Betina V. Dimaranan, and Robert A. McDougall 1.5
0.5
96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 0 20 5 06 20 07 20 08 20 09 20 10 20 11 20 12 20 13 20 14 20 1 20 5 16 20 17 20 18 20 19 20 20
0
19
Growth in Skilled Labor
1
-0.5 -1 -1.5 -2
Time 1 NAmerica
2 WEurope
3 Japan
Figure 5.7. Skilled labor growth rates: North America, Western Europe, and Japan.
then combined with the labor force projections determined earlier to obtain projected skilled labor. Combining and Filling. The next step involved providing skilled labor force projections for those standard countries in which no skilled labor projections data were available. Skilled labor projections were filled in by taking the projected value of labor for that country and multiplying it by the total average share of skilled labor in total labor. The total average share of skilled labor in total labor uses both the projections for the developed and developing countries. The resulting projections for skilled labor are depicted in Figures 5.7 and 5.8. Figure 5.7 shows that the growth rates, and ultimately the number of skilled workers, are clearly declining in the developed economies because of the decline in labor force and populations. Figure 5.9 shows that growth rates in unskilled labor are negative in both Western Europe and Japan. Growth rates in North America are positive but less than 1%, because the decline in population is not as severe in North America as it is in Japan and Western Europe, perhaps partly because of higher migration. In the developing economies, however, the decline in both skilled and unskilled labor is less severe. High growth rates in skilled and unskilled labor are declining over time, but the growth rates are remaining positive and high (Figs. 5.8 and 5.9). The growth rates of skilled labor are generally much higher than those of unskilled labor, perhaps reflecting policies designed to increase education and skill levels.
A Baseline Scenario for the Dynamic GTAP Model
147
10 9 Growth in Skilled Labor
8 7 6 5 4 3 2 1
19
19
96 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 20 13 20 14 20 15 20 16 20 17 20 18 20 19 20 20
0 Time 4 China
7 SEA
8 SoAsia
9 LatinAM
10 AfrMidE
Figure 5.8. Skilled labor growth rates: China, South East Asia, South Asia, Latin America, and Africa-Middle East.
Unskilled Labor Projections Once total labor and skilled labor projections were determined, unskilled labor projections were calculated as the difference between the total labor and skilled labor projections. The resulting projections for unskilled labor are depicted in Figures 5.9 and 5.10.
1.2 1
0.6 0.4 0.2
97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 20 13 20 14 20 15 20 16 20 17 20 18 20 19 20 20
19
96
0 19
Growth in Unskilled Labor
0.8
-0.2 -0.4 -0.6 Time 1 NAmerica
2 WEurope
3 Japan
Figure 5.9. Unskilled labor growth rates: North America, Western Europe, and Japan.
148 Terrie L. Walmsley, Betina V. Dimaranan, and Robert A. McDougall 3.5
2.5 2 1.5 1
97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 20 13 20 14 20 15 20 16 20 17 20 18 20 19 20 20
19
0
96
0.5
19
Growth in Unskilled Labor
3
Time 4 China
7 SEA
8 SoAsia
9 LatinAM
10 AfrMidE
Figure 5.10. Unskilled labor growth rates: China, South East Asia, South Asia, Latin America, and Africa-Middle East.
3. Changes in Policy on the Baseline Policy projections are another important element of a legitimate baseline scenario. The policy projections that will be incorporated into the baseline will depend primarily on the issue being examined. For example, if you are interested in free trade agreements among Southeast Asian economies, it would be important to incorporate any agreements that have already been ratified – such as the ASEAN FTA. However if you are examining agreements between the EU and South Africa then the ASEAN FTA agreement is of limited interest.6 It is up to the user to decide whether a particular agreement should be included in the baseline. The policy projections incorporated into the baseline scenario examined in this chapter reflect the issue being discussed: in this case, China’s accession to the WTO. The following is a list of the agreements incorporated into the baseline for examining China’s accession to the WTO: r the removal of tariffs under the implementation of the Uruguay Round
(UR)
r the implementation of the Multi-Fiber Arrangement (MFA) r some pre-WTO accession tariff reductions implemented by China
before 2000 6
Although it is likely that the reader will not be investigating the same issue, many of the questions addressed here are relevant to the other policies that may be included in the baseline.
A Baseline Scenario for the Dynamic GTAP Model
149
The first subsection contains a description of the original sources for the policy projections. In the second subsection, each of the policy agreements is discussed in turn – outlining the assumptions made and any problems with the data.
3.1 Data Sources The following data on the UR agreements and on China’s accession were obtained: r Post-UR tariff estimates were obtained from Francois and Strutt
(1999). These estimates were based on post-UR information from the GTAP 3 Data Base and the GATT/WTO Integrated Database (IDB). These estimates were in the form of post-Uruguay tariff revenue. These tariff revenues had been updated to reflect the GTAP 4 Data Base by McDougall, Elbehri, and Truong (1998). r Estimates of the weighted average applied tariff rates offered by China and Taiwan for their accession to the WTO were obtained (Martin et al. 2000). These are based on the final agreement between China and the United States and were available for 43 of the 57 GTAP commodities and 64 of the 66 GTAP regions. Tariff rates were provided for 2000, 2001, and 2007. The rates for 2000 and 2001 reflect tariff reductions already undertaken by China in the lead-up to accession and therefore should be incorporated into the baseline. Similar information was provided for Taiwan’s accession agreement. r Removal of export tax equivalents was used to simulate the effects of reducing the quotas on textiles and wearing apparel under the MFA. These were obtained from the GTAP 4 Data Base.
3.2 The Policy Agreements The timing of the policy projections in the baseline scenario are of utmost importance. Table 5.1 provides an overview of the timing of various policies included in the baseline scenario.
Uruguay Round Since the Uruguay Round data were first incorporated into the GTAP 3 Data Base (Francois and Strutt 1999), a number of problems needed to be
150 Terrie L. Walmsley, Betina V. Dimaranan, and Robert A. McDougall Table 5.1. Timing of policy projections Tariffs 1995–2005 UR: tariff reductions for all regions except China and Taiwan (no shocks to agriculture)
Quotas MFA: USA and EU quotas increased on exports of textiles and wearing apparel for all regions except Taiwan and China
1995–2001 Pre-WTO tariff reductions: undertaken by China before accession.
addressed to update the data. As a result, these tariff revenues were adjusted in two ways: r Large differences between the actual tariff revenue provided in ver-
sions 3 and 4 of the GTAP Data Base for beverages and tobacco led to substantially different shocks being applied to simulate the UR depending on which version of the GTAP Data Base was being used. The UR tariff revenue for beverages and tobacco was adjusted to ensure that the final shock obtained in version 4 was the same as that obtained in version 3. r The original post-UR tariffs calculated by Francois and Strutt (1999) assumed that all commitments would be implemented. However, it had become increasingly clear that agreements made with respect to agriculture were unlikely to be undertaken. Hence no changes were made to agriculture under the UR agreement in the baseline. Each year the tariff is assumed to fall by the same amount so that the specified tariff level is achieved in the final year. The reduction in tariffs per the Uruguay Round agreement occurred between 1995 and 2005. The reductions are quite small as most of the UR agreement had already been implemented before 1995.
China’s Pre-WTO Tariff Cuts Projections for China’s (and Taiwan’s) tariff rates were provided for 2000, 2001, and 2007. Each year the tariff is reduced by the same amount so that the specified tariff level is achieved in the final year. China’s average tariff rates before WTO accession are shown in Table 5.2. The pre-accession changes in tariffs are fairly large; hence not including them in the baseline would overestimate the impact of China’s accession to the WTO.
A Baseline Scenario for the Dynamic GTAP Model
151
Table 5.2. China’s average tariff rate pre-accession
Crops Livestock Food and beverages Resources Textiles Wearing apparel Metals and chemical products Autos Electronics Other manufacture Utilities Trade and transport Other services
1995
2000
2001
4% 7% 22% 8% 58% 76% 19% 129% 22% 23% 0% 0% 4%
0% 3% 20% 6% 34% 32% 15% 32% 13% 18% 0% 0% 4%
0% 3% 20% 6% 34% 32% 15% 32% 13% 18% 0% 0% 4%
Quotas Tariffs and quotas were treated slightly differently in the baseline scenario. As mentioned earlier the reduction in tariffs was assumed to occur gradually over several years. For the quota rents, however, the growth in quotas and their eventual removal have not been gradual but have been severely backloaded over time. This is consistent with how the Agreement on Textiles and Clothing (ATC) is currently being implemented by North America and Western Europe. Figure 5.11 illustrates how the removal of export quota rents has been implemented in the baseline. 120
100
80
60
40
20
0 1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
Figure 5.11. Rate at which quota rents (%) on wearing apparel and textiles are being removed over time under the ATC.
152 Terrie L. Walmsley, Betina V. Dimaranan, and Robert A. McDougall
4. Implementing the Baseline 4.1 Aggregation At this stage we have projections for gross domestic product, gross domestic investment, population, labor force, and skilled and unskilled labor for 226 standard countries and for each year from 1995 to 2020. For GTAP Data Base users and in particular for the GDyn model the projections and growth rates are then aggregated to obtain projections for each of the 11 GTAP regions used in this example: North America, Western Europe, Japan, China, Taiwan, other Newly Industrialized Economies, South East Asia, South Asia, Latin America, Africa and the Middle East, and Rest of World.
4.2 Closure The growth rates in real GDP, gross domestic investment, population, and skilled and unskilled labor, as well as the shocks to tariffs and quotas, are then incorporated into the baseline scenario. With the exception of real GDP and gross domestic investment, these projections affect variables that are exogenous in the standard closure of the model; therefore no special changes need to be made to introduce the growth rates as shocks to these variables. Real GDP and gross domestic investment are now considered in greater detail.
Real GDP and Technological Change Real GDP is usually determined endogenously by the model. As such, to include these projected growth rates, another variable must be made endogenous. From the growth accounting literature we know that growth in GDP depends on growth in endowments of labor, land, and capital and on technological change. In this baseline we have estimates of growth in labor (skilled and unskilled) and capital (via projections of gross domestic investment), and land is assumed to be fixed (zero growth). If we also have estimates of growth in real GDP then technological change must be endogenously determined. Hence, real GDP (qgdp) is set exogenously and “shocked” by the projected growth rate, and country-wide technological change is determined endogenously as the residual between growth in the country’s real GDP and its endowment growth (we use afereg). The resulting growth rates of technological change are depicted in Figures 5.12 and 5.13.
A Baseline Scenario for the Dynamic GTAP Model
153
3.5 3
Technological Progress
2.5 2 1.5 1 0.5 0 -0.5 -1 -1.5 -2 Time 1 NAmerica
2 WEurope
3 Japan
Figure 5.12. Technological change growth rates: North America, Western Europe, and Japan.
There are a number of points to note regarding the resulting changes in technological change: r The constant technological changes in the periods 1996–2000, 2011–
15, and 2016–20 reflect the fact that those periods were aggregated in the simulation; to make the figures more easily readable, the resulting 5-year change was then broken down into a constant yearly change. r If we concentrate on the period before 2002, we see that technological change also captures some of the decline resulting from the various crises in Asia, Brazil, and Japan. This occurs because the decline in investment (or capital stocks) cannot fully explain the decline in real GDP. Moreover the growth in labor relates to the growth in the supply of skilled and unskilled labor and not to changes in demand for labor – thus there is an assumption of full employment. This assumption does not accurately reflect the real situation during this period, and hence technological change is capturing some of this decline in demand for labor. r Growth rates in technological change are higher and more persistent than one would expect, especially in the developing economies. This occurs because many developing economies are currently experiencing very high growth rates in real GDP with high technological change. These very high growth rates in real GDP are then extrapolated to 2020; hence the technological change remains high.
154
8
6
Technological Progress
4
2
0 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
2012 2013 2014 2015 2016 2017 2018 2019 2020
-2
-4
-6
-8
Time 4 China
7 SEA
8 SoAsia
9 LatinAM
10 AfrMidE
Figure 5.13. Technological change growth rates: China, South East Asia, South Asia, Latin America, and Africa-Middle East.
A Baseline Scenario for the Dynamic GTAP Model
155
r The average growth rate of technological change has increased over
time. The growth rates of real GDP and gross domestic investment are projected to stabilize over time, whereas population and labor are projected to decline. Therefore the only way to maintain these high growth rates in real GDP is for increases in technological progress to compensate for the declining labor force. Another reason for the increase in average technological progress over time is the downward bias in the initial period caused by the Japanese, Asian, and Brazilian crises. These features of the growth in technological change raise concerns about the assumptions made in creating the baseline, particularly in relation to real GDP. It may be more appropriate to allow the model to determine real GDP endogenously and to incorporate more appropriate projections of technological change into the model. Ultimately the decision rests with the user and whether he or she wishes to target the real GDP projections or to target technological change and allow the model to determine growth rates in real GDP. Moreover if the policy issue being considered focuses on a particular sector or set of sectors, the user would be advised to incorporate sectorspecific technological shocks into the baseline. Incorporation of sectorspecific technological change may also improve the resulting endogenously determined country-wide growth in technology.
Gross Domestic Investment Gross domestic investment (qcgds) is also determined endogenously in the standard closure of the model. GDI is driven by the investment mechanisms in the model – namely (1) the elimination of errors in expectations, (2) the gradual movement of actual rates of return to target, and (3) the gradual movement of the growth rate of capital to the normal growth rate of capital. To set investment exogenously, one of these mechanisms must be disrupted. This can be achieved by taking either one of these steps: r Allow errors in expectations to change independently of the gradual
elimination of these errors imposed by the model. That is, allow investment to determine the errors in expectations by endogenizing srorge. In this case if the investment mechanisms are allowed to resume (i.e., investment becomes endogenous), then these errors in expectations will gradually be eliminated.
156 Terrie L. Walmsley, Betina V. Dimaranan, and Robert A. McDougall r Allow the actual rate of return to change independently of the gradual
movement toward the target rate of return by imposing an endogenous risk premium. This is achieved by endogenizing SDRORT. In this case even once the investment mechanisms are allowed to resume (i.e., investment becomes endogenous once again), the accumulated risk premium is permanent and will not be eliminated over time. In the example of China’s accession, the first mechanism was used. As in the case of real GDP the user may prefer to allow the investment mechanisms in the model to determine gross domestic investment rather than use the simple extrapolation process outlined earlier to obtain projections.
Foreign Investment in China Finally, as outlined in Chapter 8, foreign ownership in China was also adjusted to reflect the fact that it had increased significantly more in China than was justified by the changes in the rates of return. These changes in foreign ownership, it was argued, were the result of policies designed to open up the Chinese economy to foreign investors and reduce the perceived risk of investing there. Because the simulation being undertaken was designed to examine the impact of China’s accession on foreign direct investment, it was thought that the baseline should be adjusted to more accurately reflect the higher level of foreign investment in China before its accession. The mechanisms in the model that determine how much investment is owned by domestic and foreign investors were turned off (by endogenizing swqht) for China, and the level of foreign ownership (wqht) was set exogenously.
5. Conclusions In this chapter we built a baseline scenario to analyze the impact of China’s accession to the WTO. The baseline included a number of elements – macroeconomic projections and expected policy changes – that were expected to affect the analysis of China’s accession. The choice of what to include in a baseline ultimately depends on the issue being examined and the user’s view of how the world will look over the period being examined. However, some elements, particularly macroeconomic projections, are common to many baselines. For this reason this chapter has focused primarily on these common elements. It is important to note that, although the baseline itself is generally not of any interest to the analysis, it does affect the shares in the database and
A Baseline Scenario for the Dynamic GTAP Model
157
hence the results of the analysis. Care should therefore be taken to ensure a reasonably accurate representation of the world – or at least the countries and/or sectors of interest. References Ahuja, V. and D. Filmer. 1995, July. Educational Attainment in Developing Countries; New Estimates and Projections Disaggregated by Gender. World Bank Policy Research Working Paper 1489. Washington, DC: World Bank. Anderson, K., B. Dimaranan, T. Hertel, and W. Martin. 1997a. “Asia-Pacific Food Markets in 2005: A Global, Economy-Wide Perspective.” Australian Journal of Agricultural and Resource Economics 41(1): 19–44. Anderson, K., B. Dimaranan, T. Hertel, and W. Martin. 1997b. “Economic Growth and Policy Reform in the Asia-Pacific: Trade and Welfare Implications by 2005.” Asia-Pacific Economic Review 3: 1–18. Bach, C. F., B. Dimaranan, T. W. Hertel, and W. Martin. 2000, May. “Market Growth, Structural Change and the Gains from the Uruguay Round.” Review of International Economics 8(2). CPB. 1999, December. WorldScan: The Core Version. The Hague: CPB Netherlands Bureau for Economic Policy Analysis. Dimaranan, B.V. (ed.). (2006). Global Trade, Assistance, and Production: The GTAP 6 Data Base. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Dimaranan, B. V. and R. A. McDougall. 2002. Global Trade, Assistance, and Production: The GTAP 5 Data Base. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Dimaranan, B. V. and T. L. Walmsley. 2002. “Chapter 18A: Macroeconomic Data.” In B. V. Dimaranan and R. A. McDougall (eds.), Global Trade, Assistance, and Production: The GTAP 5 Data Base. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Francois, J. and A. Strutt. 1999, June. “Post Uruguay Round Tariff Vectors for GTAP v.4.” Unpublished memorandum. Martin, W., B. Dimaranan, T. Hertel, and E. Ianchovichina. 2000. Trade Policy, Structural Change and China’s Trade Growth. Working Paper No. 64. Stanford, CA: Institute for Economic Policy Research. McDougall, R. A., A. Elbehri, and T. P. Truong. 1998. Global Trade Assistance and Protection: The GTAP 4 Data Base. West Lafayette, IN: Center for Global Trade Analysis, Purdue University.
SIX
Welfare Analysis in the Dynamic GTAP Model Terrie L. Walmsley, Robert A. McDougall, and Elena I. Ianchovichina
1. Introduction The use of welfare measures is now commonplace in many AGE applications. The calculation and decomposition of welfare have become important parts of the analytical tools used to evaluate and/or compare the impact of policies on regions. In dynamic models, in which the importance of capital ownership features significantly, welfare measures are clearly superior to the use of real GDP as a measure of assessing the benefits of a policy. This is because GDP is defined as goods and services produced in the country, rather than as ownership of produced goods and services. However, calculating and decomposing welfare in a dynamic model are much more difficult tasks than in the traditional comparative static models. The aim of this chapter is to develop a method for decomposing welfare in the GDyn model presented in Chapter 2. In fulfilling this aim, a number of practical problems associated with decomposing welfare in the GDyn model arise. These problems are discussed and a solution is proposed. It involves a comparative static simulation that allows us to decompose the welfare change resulting from a policy shock at any year during the simulation horizon. The method is illustrated using an example.
2. The Welfare Decomposition The welfare decomposition used in the GDyn model is based on the welfare decomposition outlined in Hertel and Huff (2001) for the comparative static GTAP model. Hanslow (2000) generalized it, showing that the decomposition can be applied to other models; for example, with multiple households and/or foreign income flows resulting from foreign capital ownership, remittances, or foreign aid. 158
Welfare Analysis in the Dynamic GTAP Model
159
The decomposition presented in this chapter is based on the decomposition used in the comparative static GTAP model, extended using Hanslow’s (2000) simple generalization to reflect the welfare effects of additional income flows from foreign capital ownership – a key feature of the GDyn model. The incorporation of foreign capital income flows into the GTAP model and the inclusion of this term in the welfare decomposition are important features of the GDyn model that significantly enhance its ability to examine policy issues related to investment. In the GDyn model welfare is decomposed into the following components1 : r r r r r r r r
changes in preferences (CNTdpar) allocative efficiency (CNTalleffr) terms of trade (CNTtotr) price of investment relative to saving (CNTcgdsr) nonaccumulable endowments (CNTendwnar) ownership of capital endowments (CNTfeqyr) technological changes (CNTtechr) population (CNTpopr)
The standard GTAP model does not distinguish between capital owned by the regional household and capital located in the region. This assumption severely limits the ability of the comparative static GTAP model to investigate the long-run impact of a policy or the impact of a policy on investment. Hence the only component in the GDyn model that differs from the welfare decomposition in the comparative static GTAP model is the ownership of capital endowments (CNTfeqyr). In the standard GTAP model this component is equal to the changes in welfare resulting from capital located in the region, whereas in the GDyn model there is a sharp distinction drawn between ownership and location of capital; hence the capital component of welfare (CNTfeqyr) relates to changes in welfare caused by changes in capital owned by the regional household. This is determined in equation 6.1 by adding changes in welfare from net capital located in the region ([sum(i,ENDWC_COMM, VOA(i,r)∗ [qo(i,r) − pop(r)]) − VDEP(r)∗ [qk(r) − pop(r)]) to net income2 earned on capital owned but located 1 2
The equations for this decomposition are shown in appendix to this chapter. Ianchovichina (1998) also shows a detailed derivation of the welfare decomposition for the GDyn model. Income earned on capital owned but located abroad less capital located at home but owned by foreign households. This term is always zero in the standard comparative static GTAP model.
160 Terrie L. Walmsley, Robert A. McDougall, and Elena I. Ianchovichina
abroad (YQHTRUST(r)∗ (yqht(r) − pop(r)) − YQTFIRM(r)∗ (yqtf(r) − 3 pop(r)) ): CNTfeqyr(r) = [0.01∗ EVSCALFACT(r)]∗ [sum(i, ENDWC COMM, VOA (i,r)∗ [qo(i,r) − pop(r)]) − VDEP(r)∗ [qk(r) − pop(r)] + YQHTRUST (r)∗ (yqht(r) − pop(r)) − YQTFIRM(r)∗ (yqtf(r) − pop(r))]
(6.1)
where EVSCALFACT(r) is the EV scale factor; it is related to the elasticity of utility with respect to nominal household income of household h; VOA(i,r) is the value of capital endowment i at agent’s prices located in region r; qo(i,r) is the proportionate change in the quantity of capital endowment i (used in all sectors) by region r; pop(r) is the proportionate change in the population of region r; VDEP(r) is the value of depreciation on the capital endowment; qk(i,r) is the proportionate change in the quantity of capital located in region r; YQHTRUST(r) is the value of income earned on equity owned by households of region r and located abroad; YQTFIRM(r) is the value of income earned on equity owned by foreigners (via the trust) and located in region r; yqht(r) is the proportionate change in income earned on equity owned by households of region r and located abroad; yqtf(r) is the proportionate change in income earned on equity owned by foreigners (via the trust) and located in region r. A second difference between the standard comparative static and the GDyn models is that in the GDyn model we generally undertake two simulations – a baseline and a policy simulation – whereas a static simulation involves just one policy simulation. Rather than focus on the growth rates of each of the variables of interest, the difference between the growth rates determined in the policy and the baseline simulations are calculated and cumulated over time. For example, if skilled labor increases by 2% per year 3
Summed over s, where s is the set of regions. Note that this is the fully flexible form; in the GDyn model foreign ownership is dealt with through a global trust, and hence foreign ownership is not tracked on a bilateral basis. This is primarily due to the difficulty in obtaining good quality bilateral data on foreign ownership.
Welfare Analysis in the Dynamic GTAP Model
161
in the baseline and by 3% in the policy, the difference is 0.980392% per year.4 The cumulative difference is 0.980392% in the first year, 1.970396% in the second year,5 and 2.970106% in the third year. These cumulated differences allow us to focus on the impact of the policy relative to the baseline. The welfare decomposition therefore shows the cumulative difference between welfare effects in the policy and in the baseline.
3. Problems Decomposing Welfare in the GDyn Model There are a number of problems with this decomposition of welfare in the GDyn model. The first one relates to the timing of the benefits or costs of a policy. For example, a policy may have a negative impact in the short-run, a positive impact in the medium-run after structural adjustment has occurred, and larger gains in the long-run with capital accumulation. Hence whether the policy is considered to have a positive impact or not will depend on when welfare is examined. Moreover, two countries may have achieved the same long-run impact on welfare, but one may have obtained the benefits earlier than the other country. A second problem is the fact that in a recursive dynamic model there is no discount factor, and hence welfare cannot be cumulated over time in a meaningful way. In response to these two problems we can only recommend that the user show the changes in welfare over time or apply an externally determined discount factor to cumulate welfare. A third problem with the welfare decomposition is that it is highly path dependent. This means that the result depends on the path taken, rather than being a true reflection of the actual impact of the policy. The welfare decomposition is path dependent because it does not have a levels equivalent like most of the other equations in the GDyn model; instead welfare is a change variable dependent on the levels and percentage changes resulting from the simulation. The method of cumulating differences in the GDyn model to isolate changes caused by the policy ensures that this path dependency is an issue for the welfare decomposition.6 4
5
6
The difference between base and policy growth rates is determined by (P − B)/B × 100, where P is the growth rate during the policy (3%) and B is the growth rate in the base case (2%). Cumulative growth rate for the second year is given by (1 + P1/100)(1 + P2/100). Hence the cumulative difference is given by [(1 + P1/100)(1 + P2/100) − (1 + B1/ 100)(1 + B2/100)]/(1 + B1/100)(1 + B2/100) × 100. This path dependency is related to, but not identical, to the path dependency discussed in Chapter 2. In Chapter 2, path dependency refers to the final state of the world, which depends not only on the cumulative shocks but also on the time path of the shocks.
162 Terrie L. Walmsley, Robert A. McDougall, and Elena I. Ianchovichina
An example will help illustrate the point. We use a small 3 × 3 aggregation and application of the GDyn model that involves a simple shock to region-specific technological change in the Rest of World (ROW) and the baseline scenario discussed in Chapter 5.7 Table 6.1 shows the welfare decomposition obtained from the cumulative differences in 2020 of a dynamic simulation. If we examine the welfare decomposition outlined in Hertel and Huff (2001) and Hanslow (2000), we note that each of these contributions to welfare is a change in quantity (or relative prices in the case of the terms of trade contribution). For example, the contribution to welfare of nonaccumulable8 endowments (CNTendwnar(r)) is shown in equation 6.2: CNTendwnar(r) = [0.01∗ EVSCALFACT(r)]∗ [sum(i, ENDWNA COMM, VOA(i,r)∗ [qo(i,r) − pop(r)])],
(6.2)
where VOA(i,r) is the value of nonaccumulable endowment i located in region r; qo(i,r) is the proportionate change in the quantity of nonaccumulable endowment i located in region r; pop(r) is the proportionate change in the population of region r. Equation 6.2 illustrates that the contribution of nonaccumulable endowments to welfare depends on the sum of changes in the quantity supplied of all nonaccumulable endowments. The growth in the quantity of nonaccumulable endowments is usually determined by the baseline scenario imposed. For example, in the baseline scenario presented in Chapter 5, the growth of unskilled labor in North
7
8
In Chapter 6, path dependency refers to a problem with certain variables in the model, including the welfare decomposition that cannot be defined in the levels but only in differentials. As a result, reported cumulative changes in these variables depend not only on the initial and final states but also on the solution path between them. The application CH6HO3 × 3 gdyn v32 97.zip and the welfare decomposition of that application Ch6HO3 × 3 gdyn v32 97wd.zip can be downloaded from the GTAP Web site at http://www.gtap.agecon.purdue.edu/models/Dynamic/applications.asp. Readers are advised to go through Chapter 7 first and only do the welfare decomposition after completing the exercises in Chapter 7. Nonaccumulable endowments are those endowments in the model that do not accumulate over time. In the dynamic version of the GTAP model this includes land, skilled and unskilled labor, and natural resources, but does not include capital because investment adds to capital over time.
163 Table 6.1. Welfare decomposition obtained from cumulative differences in 2020 of a dynamic simulation
North America European Union Rest of World a b c
EV welfare
CNTdpar changes in preferences effecta
CNTalleffr allocative efficiency effect
CNTtotr terms of trade effect
CNTcgdsr capital goods effect
CNTendwnar nonaccumulable endowment effectb
CNTfeqyr foreign equity effect
CNTtechr technological change effectc
CNTpopr population effect
6736.4 −27008.0 −909594.0
0 0 0
−324.95 −1864.06 −106827.00
−18887.40 −20857.30 40806.48
−5048.22 189.86 −2578.86
3693.75 1245.25 −76082.90
19271.13 −10292.10 −203950.00
5655.13 4027.75 −454230.00
2377.00 542.59 −106732.25
Changes in preferences (CNTdpar) – this effect is zero because preferences for private consumption, saving, and government at the top level of the GDyn model have not been altered. Nonaccumulable endowments (CNTnar) – this effect is nonzero because path dependency has not yet been removed. Technological changes (CNTtechr) – this effect is related to the policy shock, which involved a large decrease in technological change in the rest of the world.
164 Terrie L. Walmsley, Robert A. McDougall, and Elena I. Ianchovichina
America between 1997 and 2002 is forecasted to be 8.07%. Because most policies do not affect this growth rate, North America’s growth of unskilled labor with the policy is also likely to be 8.07% over the same period. Thus the impact of the policy on the growth rate of unskilled labor, or the difference between the baseline and the policy, is zero. It would therefore be reasonable to expect that the contribution of unskilled labor growth to the change in welfare resulting from the policy would also be zero; because the same applies to all other nonaccumulable endowments, CNTendwnar(r) should equal zero. However this is not the case. Table 6.1 shows the value of CNTendwnar(r) obtained from the regionspecific technological change in the rest of world using a 3 × 3 aggregation of the GTAP Data Base and the baseline scenario discussed in Chapter 5. The results indicate that the contribution of nonaccumulable endowments to the change in welfare can be quite significant. By 2002, nonaccumulable endowments accounted for between 1% and 2% of the total change in welfare, and by 2020, 54% of North America’s cumulative welfare change from the policy is explained by nonaccumulable endowments. This paradoxical result is caused by path dependency in the welfare decomposition. CNTendwnar(r) is calculated for each region in both the baseline and the policy scenarios; then the difference is taken to determine the impact of the policy by itself. Although the baseline and the policy simulations both have the same proportionate changes in nonaccumulated endowments, the changes in those endowments may differ.9 Then when the cumulative difference between these changes in contributions in the baseline and policy are taken to isolate the change caused by the policy, a nonzero difference is obtained. There is other evidence of this path dependency in Table 6.1. Given that the policy shock is a negative shock to technological change in the rest of the world, we would expect to see a negative contribution to welfare in ROW. So where do the contributions listed under North America and the European Union come from? They are the result of the technological change (afereg) calibrated in the baseline to obtain the change in real GDP. Again, when this technological change interacts with different values in the baseline and policy and when cumulative differences are taken, spurious difference occurs due to path dependency. The same also applies to the population effect, which changes by the same amount in both the policy and the baseline. Given that equivalent variation is, by definition, the change 9
The reason for this is that the coefficients get updated and reflect price changes that differ between the two simulations.
Welfare Analysis in the Dynamic GTAP Model
165
Policy A Base
B
Figure 6.1. The welfare decomposition.
in welfare at constant initial prices, these changes due to differences in value shares result in an incorrect measure of welfare. These variations in the welfare decomposition, due to path dependency, will occur to some extent in all components of the welfare decomposition and therefore will bias the aggregate welfare results. In the next section we examine a possible solution to this problem.
4. A Comparative Static Simulation for Calculating Welfare To remove any path dependency from the welfare decomposition, after the baseline and policy simulation are completed, a comparative static simulation is undertaken. The other benefit of using a comparative static simulation is that we are examining the difference in welfare between the baseline and the policy at a single point in time, and hence a discount rate is not required to cumulate welfare over time. We have designed the comparative static simulation in such a way as to find the difference between the baseline and the policy results in a given year (e.g., 2020). In terms of Figure 6.1 this is the difference between the base and policy lines or the distance AB. The benefit of the comparative static simulation is that it starts with a common initial database – the updated baseline database in 2020 – and applies only those shocks that are related to the policy itself to obtain the updated policy database (i.e., move from A to B in Fig. 6.1), so that all path dependency in the welfare decomposition is removed. The cost associated with this method is that it does not allow us to distinguish between two countries that have achieved the same long-run welfare, but one country obtained these benefits earlier than the other. As mentioned earlier, the size
166 Terrie L. Walmsley, Robert A. McDougall, and Elena I. Ianchovichina
and magnitude of the welfare impact of a policy will depend on the point in time at which welfare is measured. We choose to decompose welfare in the long-run when all structural adjustment has taken place and investment has augmented capital stocks available for production. If the user wishes to have detailed information on the path of welfare over time, then he or she could undertake a comparative static simulation for each period; this can be very time consuming, however. The types of shocks that need to be incorporated into the comparative static simulation are not as obvious as one might first suspect. We start by outlining those shocks that would be required to re-create a database identical to the updated policy data created in the dynamic simulation, beginning with the 2020 baseline data. These shocks include the following: 1. Policy variables: All policy shocks imposed and cumulated over all periods before the year being examined must be included in the comparative static simulation. 2. Closure swaps: If any closure swaps were made between the baseline and the policy in any period during the simulation, then these closure swaps need to be taken into account. For example, if tax revenue replacement was assumed in the policy but not in the baseline, then the tax that was used to replace any lost tax revenue needs to be set exogenously and shocked by the cumulative amount required to ensure tax revenue replacement for the policy change. 3. Accumulation variables: Any variable that accumulated over time, such as capital stocks or wealth, must also grow by the amount it accumulated over the period. This includes the variables qk and wqh. 4. Time-dependent variables: Any variable that depends on or changes over time as part of the dynamic mechanisms, such as the expected rate of return (rorge) and the change in the normal growth rate (DKHAT) must also be included. 5. Numeraire shock The numeraire must also be shocked. The numeraire is a composite price and therefore also suffers from path dependency. Equation 6.3 illustrates the equation for the numeraire: psavewld = sum(r,REG, [NETINV(r)/GLOBINV]∗ pcgds(r));
(6.3)
where psavewld is the numeraire, NETINV(r)/GLOBINV is the share of net investment in region r relative to global net investment, and pcgds(r) is the price of capital goods by region.
Welfare Analysis in the Dynamic GTAP Model
167
Table 6.2. Welfare results from a comparative static simulation and a
dynamic simulation
North America European Union Rest of World
Comparative static
Dynamic
Difference (%)
2456.6 −22606.0 −1011024.0
6736.4 −27008.0 −909594.0
174 19 −10
Because the shares (NETINV(r)/GLOBINV) are updated and differences between the baseline and policy values of the share can arise, the variable pcgds(r) may have to be altered slightly to compensate and to ensure that the numeraire (psavewld) continues to equal 0. In truth all prices will have to adjust slightly as a result of this path dependency in the numeraire. To make this adjustment, we swap the numeraire with a price that is not a composite price – for example, the price of capital in one region10 – and shock this price by the cumulative difference found in the dynamic model simulation. To check that one has done this comparative static simulation correctly, the user should compare the final database from the dynamic simulation (the updated database after the policy) with the updated data from the comparative static simulation. Differences will usually occur at the 5th and 6th decimal place due to inaccuracies in computing.11 Larger differences may result at the 3rd and 4th decimal place if there is significant path dependency due to a more complicated experiment. This may occur when the time intervals are longer than 1 year. The user will need to take care when judging whether these differences are too great.
5. A Comparison of the Welfare Decomposition under the Comparative Static and Dynamic Simulations The resulting welfare calculations obtained using this procedure are provided in Table 6.2. This table also shows the change in welfare if path 10
11
The noncomposite price is chosen because it is not path dependent; that is, it is not a valueweighted aggregation of other prices. In the case of a composite price, any differences in the paths followed by the value shares force the composite price to differ between the baseline and the policy (even if the individual prices, which are aggregated, change by the same amount in both the baseline and policy). Choosing this composite price as the numeraire and thereby forcing it to be zero in both the policy and the baseline would therefore lead to small differences in all the prices in the baseline and policy in order to negate the differences in the paths of the value shares. In the case of a noncomposite price it is not an aggregate of other prices and therefore this does not happen. GEMPACK uses single precision.
168 Terrie L. Walmsley, Robert A. McDougall, and Elena I. Ianchovichina
dependency in the welfare decomposition had not been removed. The results show that welfare can be significantly different if path dependency in the welfare decomposition is not taken into account. For example, in the case of North America welfare is less than half of the value obtained when path dependency was ignored. For this reason it is important to remove the path dependency from the calculations. Table 6.3 shows the decomposition of welfare into the eight contributions outlined in Section 2. The largest noticeable differences in welfare contributions are nonaccumulable endowments and population, which are nonzero in the dynamic simulation (Table 6.1) and zero in the comparative static welfare decomposition (Table 6.3). However, all of the welfare contributions differ between the dynamic and comparative static welfare simulations, and in many cases these differences are quite large. For this reason it is very important to take account of the path dependency in the welfare decomposition.
6. Conclusions This chapter presented a welfare decomposition technique for use with the GDyn model. We considered three issues that biased the welfare results. The first one relates to the fact that the impact of the policy on welfare will depend on when welfare is examined. The second issue relates to the problem of cumulating welfare results over time without a feasible discount factor to aggregate over time. The third problem is that the welfare decomposition is highly path dependent. To remove path dependency in the welfare decomposition and determine the cumulated welfare change, a comparative static simulation was developed that removed the path dependency and determined the change in welfare resulting from the policy at a particular point in time. The comparative static simulation is designed to find the difference between the baseline and the policy results in a given year (e.g., 2020). If one wishes to obtain the welfare path, the techniques must be applied at each year of the simulation period.
Appendix: Code for Decomposing Welfare in the GDyn Model Welfare is EV(r) = CNTdpar(r) + CNTalleffr(r) + CNTtotr(r) + CNTcgdsr(r) +CNTendwnar(r) + CNTfeqyr(r) + CNTtechr(r) + CNTpopr(r)
169 Table 6.3. Welfare decomposition
North America European Union Rest of World a b c d
EV welfare
CNTdpar changes in preferences effecta
CNTalleffr allocative efficiency effect
CNTtotr terms of trade effect
2456.56 −22606.34 −1011023.57
0.00 0.00 0.00
−261.27 −1393.22 −132707.19
−20718.81 −24384.74 44703.52
CNTcgdsr CNTendwnar capital nonaccumulable goods endowment effect effectb −2722.51 −1113.96 6749.31
0.00 0.00 0.00
CNTfeqyr foreign equity effect 26159.14 4285.58 −198727.22
CNTtechr technological CNTpopr change population effectc effectd 0.00 0.00 −731072.00
0.00 0.00 0.00
Changes in preferences (CNTdpar) – this effect is zero because preferences for private consumption, saving, and government at the top level of the GDyn model have not been altered. Nonaccumulable endowments (CNTnar) – this effect is now zero because the policy shock does not affect nonaccumulable endowments and the path dependency has been removed. Technological changes (CNTtechr) – this effect is related to the policy shock, which involved a large decrease in technological change. Population (CNTpopr) – this effect is zero because the policy shock does not affect population growth.
170 Terrie L. Walmsley, Robert A. McDougall, and Elena I. Ianchovichina
where: Contribution due to changes in preferences is: CNTdpar(r) = −[0.01 ∗ UTILELASEV(r) ∗ INCOMEEV(r)] ∗ [DPARPRIV(r) ∗ loge(UTILPRIVEV(r)/UTILPRIV(r)) ∗ dppriv(r) + DPARGOV(r) ∗ loge(UTILGOVEV(r)/UTILGOV(r)) ∗ dpgov(r) + DPARSAVE(r) ∗ loge(UTILSAVEEV(r)/UTILSAVE(r)) ∗ dpsave(r)]; Contribution due to changes in allocative efficiency is: CNTalleffr(r) = [0.01 ∗ EVSCALFACT(r)] ∗ [sum(i,NSAV COMM, PTAX(i,r) ∗ [qo(i,r) − pop(r)]) + sum(i,ENDW COMM, sum(j,PROD COMM,ETAX(i,j,r) ∗ [qfe(i,j,r) − pop(r)])) + sum(i,TRAD COMM, sum(j,PROD COMM,IFTAX(i,j,r) ∗ [qfm(i,j,r) − pop(r)])) + sum(i,TRAD COMM, sum(j,PROD COMM,DFTAX(i,j,r) ∗ [qfd(i,j,r) − pop(r)])) + sum(i,TRAD COMM, IPTAX(i,r) ∗ [qpm(i,r) − pop(r)]) + sum(i,TRAD COMM, DPTAX(i,r) ∗ [qpd(i,r) − pop(r)]) + sum(i,TRAD COMM, IGTAX(i,r) ∗ [qgm(i,r) − pop(r)]) + sum(i,TRAD COMM, DGTAX(i,r) ∗ [qgd(i,r) − pop(r)]) + sum(i,TRAD COMM, sum(s,REG, XTAXD(i,r,s) ∗ [qxs(i,r,s) − pop(r)])) + sum(i,TRAD COMM, sum(s,REG, MTAX(i,s,r) ∗ [qxs(i,s,r) − pop(r)]))]; Contribution due to changes in terms of trade is: CNTtotr(r) = [0.01 ∗ EVSCALFACT(r)] ∗ [sum(i,TRAD COMM, sum(s,REG, VXWD(i,r,s) ∗ pfob(i,r,s))) + sum(m,MARG COMM, VST(m,r) ∗ pm(m,r)) − sum(i,TRAD COMM, sum(s,REG, VXWD(i,s,r) ∗ pfob(i,s,r))) − sum(m,MARG COMM, VTMD(m,r) ∗ pt(m))];
Welfare Analysis in the Dynamic GTAP Model
171
Contribution due to changes in price of investment goods relative to the saving goods (type of terms of trade effect) is given as: CNTcgdsr(r) = [0.01 ∗ EVSCALFACT(r)] ∗ [NETINV(r) ∗ pcgds(r) − SAVE(r) ∗ psave(r)]; Contribution due to changes in nonaccumulable endowments is: CNTendwnar(r) = [0.01 ∗ EVSCALFACT(r)] ∗ [sum(i,ENDWNA COMM, VOA(i,r) ∗ [qo(i,r) − pop(r)])]; Contribution due to changes in capital ownership is: CNTfeqyr(r) = [0.01 ∗ EVSCALFACT(r)] ∗ [sum(i,ENDWC COMM, VOA(i,r) ∗ [qo(i,r) − pop(r)]) − VDEP(r) ∗ [qk(r) − pop(r)] + YQHTRUST(r) ∗ (yqht(r) − pop(r)) − YQTFIRM(r) ∗ (yqtf(r) − pop(r))]; Contribution due to changes in technology is: CNTtechr(r) = [0.01 ∗ EVSCALFACT(r)] ∗ [sum(i,PROD COMM, VOA(i,r) ∗ ao(i,r)) + sum(j,PROD COMM, VVA(j,r) ∗ ava(j,r)) + sum(j,PROD COMM, sum(i,ENDW COMM, VFA(i,j,r) ∗ afe(i, j, r))) + sum(j,PROD COMM, sum(i,TRAD COMM, VFA(i,j,r) ∗ af(i,j,r))) + sum(m,MARG COMM, sum(i,TRAD COMM, sum(s,REG, VTMFSD(m,i,s,r) ∗ atmfsd(m,i,s,r)) + sum(i,TRAD COMM, sum(s,REG, VIMS(i,s,r) ∗ ams(i,s,r)))]; Contribution due to changes in population is: CNTpopr(r) = 0.01 ∗ INCOMEEV(r) ∗ pop(r).
References Hanslow, K. 2000. A General Welfare Decomposition for CGE Models. GTAP Technical Paper No. 19. West Lafayette, IN: Center for Global Trade Analysis, Purdue University.
172 Terrie L. Walmsley, Robert A. McDougall, and Elena I. Ianchovichina Hertel, T. W. and K. Huff. 2001. Decomposing Welfare Changes in GTAP, GTAP Technical Paper No. 5 (revised version). West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Ianchovichina, E. I. 1998. International Capital Linkages: Theory and Application in a Dynamic Computable General Equilibrium Model. Ph.D. thesis, Department of Agricultural Economics, Purdue University.
SEVEN
Implementing the Dynamic GTAP Model in the RunDynam Software Ken Itakura, Elena I. Ianchovichina, Csilla Lakatos, and Terrie L. Walmsley
1. Introduction The purpose of this chapter is to introduce the reader to the publicly available software – RunDynam – that is used to carry out the applications presented in Part III of this book. The RunDynam program is based on the GEMPACK suite of software (Harrison and Pearson 1998), which is specifically designed to solve nonlinear general equilibrium models. Other general equilibrium models solved using the GEMPACK software suite include the standard GTAP model and the Monash model of Australia. The RunDynam program has been specially tailored to the needs of the GDyn model and other dynamic models. It offers the user a great deal of flexibility in constructing simulations. It is available from the Centre of Policy Studies at Monash University, Australia.1 Using the RunDynam software, you can examine the data, construct and modify simulations, solve simulations, and examine results. Users who wish to alter the underlying theory of the model will need to acquire additional software from the Centre of Policy Studies at Monash University, Australia. In addition, those who wish to make their own aggregations will have to purchase the GTAP Data Base from the Center for Global Trade Analysis, Purdue University.2 Altering the standard model and the data aggregation process are not discussed in this chapter. The RunDynam program requires a PC running Microsoft Windows XP or later, with at least 512MB of RAM and at least 1GB of free disk space.
1 2
See http://www.gempack.com.au. See http://www.gtap.agecon.purdue.edu/.
The authors would like to thank Anna Strutt and Horacio Santander for their help with the welfare decomposition section of this chapter.
173
174
Ken Itakura et al.
The chapter is organized as follows. Section 2 demonstrates how to install the program and load up the applications. Section 3 shows how to view the data. Section 4 provides an introduction to running simulations. Section 5 examines how to view simulation results. Section 6 outlines the notation used to denote the chapters presenting the applications in Part III of this book.
2. Installing the RunDynam Software and Loading up Applications 2.1 Installing RunDynam Follow these steps to install RunDynam: r Double-click on the install EXE in the RunDynam CD-ROM. r The RunDynam Installation program starts and a Welcome pop-up
box appears on your screen. Click Next.
r You will be asked which directory you wish to install RunDynam.
Default destination folder is c:\RunDynam. Click Next.
r If you are ready to install, then click Next. r Once installation is completed, click Finish.
The RunDynam icon will then appear on your desktop ready for use.
2.2 Downloading the RunDynam Application Archive Once you have installed the RunDynam software you can obtain the applications for this chapter and the other chapters in this book, from the GTAP Web site (https://www.gtap.agecon.purdue.edu/models/Dynamic/ applications.asp). Each application is a RunDynam zipped file. Download the zipped files from the Web site, and place them into the following RunDynam sub-directory c:\RunDynam\archive.
2.3 Opening RunDynam The first step is to open up RunDynam. This is achieved by double-clicking on the RunDynam icon on your desktop. If a dialogue box appears to ask if you would like to load a model and simulation from the zip archive, click No.
Implementing the Dynamic GTAP Model in the RunDynam Software 175
The following should appear at the top of your screen:
Once inside RunDynam, you will notice two sets of toolbars at the top of the page. The first toolbar (File, Zip, Tasks, Rational Expectations, View, Options, Run Preferences, and Help) is similar to those you may have seen in other Windows programs. This toolbar is referred to as the main menu. This main menu allows you to carry out various functions such as opening, saving, and viewing data and seeking help. The second toolbar (Introduction, Model/Data, Sim Overview, Closure/Shock, Results, and Other files) looks like a notebook or card index. Each of the pages in the notebook or card index contains an essential element of the simulation design. Each card is referred to as a page (i.e., the Model/Data page). Each label is referred to as a tab. You click on a tab to go to the required page.
2.4 Restoring the Ingredients of a Simulation At this stage, we will use a simulation that was created previously for RunDynam. It uses the GDyn model to simulate the effects of a productivity shock on the Rest of World (ROW). The ingredients of this simulation have been saved to a zipped file and can be restored into RunDynam by following these instructions: r On the main menu first choose Zip | Restore Ingredients from ZIP
Archive. A drop-down menu will appear. Move down in the dropdown menu until you find Restore Ingredients from ZIP Archive and then click. r A “ZIP File to Restore From” box will appear. Choose the following file from the archive directory where you downloaded them: Ch7HO3 × 3_gdyn_v3_97.zip, and click on the OPEN button. r An information box appears and shows the history of the ZIP archive. Click OK.
176
Ken Itakura et al.
r Another information box appears and asks you to specify the directory
where you want to restore the files in the ZIP archive. Click OK.
r Choose the following directory: C:\RunDynam\HO3 × 3. Click OK.
If the directory you specified does not exist, the software will create it for you. Click Yes.
r A confirmation box appears. Click OK. r After the files are unzipped, you will be asked, “Do you want to load
these Simulation Details into RunDynam now?” Click Yes.
r The software reminds you that the restored application is now the
current application. Click OK. Note that under the directory you specified three new folders: C:\RunDynam\HO3 × 3\data, C:\RunDynam\HO3 × 3\model and C:\RunDynam\HO3 × 3\tabetc have automatically been created to store the files. r Click the tab labeled Model/Data to turn to the Model/Data page. r Listed at the top in blue should be the model. Check that the model is
gdyn.exe.
Implementing the Dynamic GTAP Model in the RunDynam Software 177
r Also make sure you have all the data ingredients in the white box in
the middle. The data ingredients should be as follows:
Notice again that the model (gdyn.exe) is stored in the directory, model, and the various data input files are stored in the directory, data.
3. Viewing the Data In this section, we use RunDynam to look at the GDyn model and the associated data files. Two examples are undertaken to examine the sets data and the core data. In each of these examples, we first look at how the data are used by the underlying model or TABLO file, and then we examine the data files themselves. The underlying equations of the GDyn model are defined in the TABLO file (gdyn.tab). The TABLO file is the human-readable version of the executable file gdyn.exe. As we saw earlier, four header array files (extension ∗ .HAR) are required by the GDyn model: One contains the sets, one contains the base data, and the other two files contain the parameters for the standard GTAP model and the dynamic extension, respectively. The header array file containing the base data includes both the standard GTAP data and some additional data required for the dynamic extension. Example 3.1. Viewing the Set Data r The TABLO code can be viewed by choosing View | Main TABLO file
from the main menu. TABmate opens a copy of the gdyn.tab file, and the copy may be labeled tab1.tab.
178
Ken Itakura et al.
The TABLO file contains the essential elements of the model, including its variables, coefficients, and equations. For the software to understand the model equations set out in gdyn.tab, a number of rules must be adhered to, including the following: r All variables, coefficients, sets, parameters, and files referred to in the
TABLO file must first be defined in the TABLO file.
r All coefficients must either be read from a file or derived from other
coefficients that have been previously defined and read.
r To assist the user of the model, comments have been added to the
TABLO file; these comments are always placed between two exclamation marks. You will need to refer to the TABLO file constantly when seeking any information about equations, variables, and coefficients or when interpreting your results. r In TABmate (or other editor), select Search | Find from the main menu
to find the word File. You should find the following statement (F3 will repeat the search): File GTAPSETS # file with set specification #; This statement defines a file with the logical name GTAPSETS. The words between the # #’s are the label given to this logical file name. The name indicates to the user what is contained in the file (i.e., the set information).
Implementing the Dynamic GTAP Model in the RunDynam Software 179 r If you now use Search | Find to find Set, you should find a list of set
statements, such as Set REG # regions in the model # maximum size 10 read elements from file GTAPSETS header “H1”; The second line of each set declaration defines the set and gives it a name. The third line states that the set should contain no more than 10 elements and is read from header H1 of the file with the logical filename GTAPSETS. r If you now use Search | Find to find REG you will see that many
of the variables, coefficients, and equations are defined over the set REG. You will also see variables defined in terms of the other sets (e.g., TRAD COMM). r You can now close the TABmate by clicking File | Exit. r Back in RunDynam, you should still be on the page labeled Model/Data. If you look at the white box containing the data files, you will see the name of the file corresponding to the logical name GTAPSETS. File GTAPSETS = C:\RunDynam\HO3 × 3\data\gdset.har This tells us that gdset.har is the file, with the logical name GTAPSETS, containing the set information. r To open this file, highlight the GTAPSETS file in the white box, and
click the right-hand button of your mouse to get a menu. Then select View this file.
Once in ViewHAR you will see a menu bar and a table containing the sets. Each row in the table corresponds to a set.
180
Ken Itakura et al.
In the table of sets the first column relates to the header, the second to the type of data, and the third to the size of the data; the last column is a brief explanation of what is in the set. The 1C in the second column means that the information contained in the header are characters. The 3 length 12 in the third column means that there are 3 elements that may contain up to 12 characters. As we know from the TABLO code (tab1.tab, copy of gdyn.tab), the set of regions (REG) should be located under header H1. The comment in the last column is the same comment seen between the # #’s in the TABLO code. r You can view the regions by double-clicking anywhere on the row
labeled header H1.
r Click on Contents on the main menu bar or click anywhere on the
data to return to the list of sets.
r Now look at the traded commodities (TRAD COMM) in this dataset. r You can now close the header array file by clicking File | Exit.
Example 3.2. Viewing the Core Data Before examining the data, let us look again at the TABLO file (tab1.tab, a copy of gdyn.tab). r Choose View | Main TABLO file from the main menu. The core data
can be found in the file with the logical file name GTAPDATA.
r Using Search | Find in TABmate, search for GTAPDATA. You should
find the following statement. File GTAPDATA # file containing all base data #;
Implementing the Dynamic GTAP Model in the RunDynam Software 181
This statement defines a header array file with logical file name GTAPDATA. Now let us look at an equation containing some data. r In TABmate, select Search | Find and look for TOTINCEQY. This will
take you to the following equation labeled TOTINCEQY: Equation TOTINCEQY # This equation determines the change in total income from equity# (all,r,REG) yqh(r) = [YQHFIRM(r)/YQHHLD(r)]∗ yqhf(r) + [YQHTRUST(r)/YQHHLD(r)]∗ yqht(r); The first line is the equation name. The second line is a comment providing some information on what the equation does. The third line states that this equation holds for each region r in the set REG (i.e., USA, EU, and ROW as seen in Example 3.1). The rest of the lines define the equation. In standard GTAP notation, the convention is to use upper case for levels and lower case for percentage changes or deviations from the base case. Here, we are interested in the initial database or the levels coefficients (shown in upper case). We can find out what these coefficients are and how they are calculated by searching the TABLO file. Because all variables and coefficients must be defined before use, we should be able to find a definition by searching upward. r Select Search | Find from the main menu of the TABmate and look for
YQHHLD. Remember to change the find box to search upward. This is done by clicking on Back in the box labeled Direction (alternatively you can use CTL-Home to move you back to the beginning of the TABLO file, and search from top or gloss). Keep searching (F3 to search again) until you find the following definition and formula for YQHHLD(r): Coefficient (all, r, REG) YQHHLD(r)# regional household equity income #; Formula (all, r, REG) YQHHLD(r) = YQHFIRM(r) + YQHTRUST(r); This tells us that YQHHLD(r) is defined as a coefficient and is the income earned on equity by the regional household. It is equal to the sum of
182
Ken Itakura et al.
two other coefficients, YQHFIRM(r) and YQHTRUST(r). This coefficient is often referred to as a derived coefficient because it is not read directly from the database, but is derived from other coefficients. r You can now use Search | Find to find YQHFIRM(r) and
YQHTRUST(r). You should find the following statements in the TABLO file: Coefficient (ge 0) (all,r,REG) YQHFIRM(r) # income of region r from local firms #; Update (all,r,REG) YQHFIRM(r) = yqhf(r); Read YQHFIRM from file GTAPDATA header “YQHF ”; Coefficient (ge 0)(all,r,REG) YQHTRUST(r) # regional income from global trust #; Update (all,r,REG) YQHTRUST(r) = yqht(r); Read YQHTRUST from file GTAPDATA header “YQ HT ”; These statements tell us that YQHFIRM(r) and YQHTRUST(r) are also coefficients defined as income earned by region r from local firms and from the global trust, respectively. Unlike YQHHLD(r) these coefficients are read directly from headers “YQHF” and “YQHT,” respectively, in the file with the logical file name GTAPDATA. r If you scroll up or down from here, you will notice a number of other
coefficients that are also read from the file GTAPDATA.
r You can now close the TABLO file by clicking File | Exit.
Back in RunDynam you should still be on the page labeled Model/Data. In the white box containing the data files, you should see the name of the file corresponding to the logical file named GTAPDATA. r To view this file, highlight the GTAPDATA file and click the right-hand
button of your mouse to get a menu. Then select View this file.
Implementing the Dynamic GTAP Model in the RunDynam Software 183
You should see the following table containing the base data:
r Look for header YQHF containing the values of equity income earned
from local firms. View the data by double-clicking on the header “YQHF”. Data should be available for each of the three regions. Repeat the same for header “YQHT”. r The two parameter files can be viewed in the same way.
4. Running a Simulation The purpose of this section is for you to gain some hands-on experience doing simulations with RunDynam. An increase in total factor productivity in the Rest of World illustrates the use of RunDynam. For this purpose we examine the remaining pages of RunDynam. The next two pages, Sim Overview and Closure/Shock, relate to the elements required to undertake a simulation with the GDyn model. The third page, Results, provides an easy way to view the results of the simulation. The final page, Other Files, lists the data files used in the simulation and informs you as to whether they are updated during a typical simulation. First, we provide a recipe outlining the basic ingredients required to conduct a simulation using RunDynam. The simulation undertaken here is a productivity shock. Two of the main elements of the simulation – the Model (gdyn.tab) and the Data – have already been discussed in the previous section. Therefore, the focus here is on the two pages labeled Sim Overview and Closure/Shock. We examine the Sim Overview page, the Closure/Shock page, a base-case shock file, a policy shock file, and a closure file. Finally, we run the simulation.
4.1 Simulation Overview Page r Click on the Sim Overview tab on the second level of the toolbar to
move to the Sim Overview page.
184
Ken Itakura et al.
This page contains a number of important aspects of the simulation, including the label for the starting year, the number of periods being examined, the length of these periods, and the solution method. The page should look something like the figure provided here.
Starting from the top and moving down, this is what you will see: r The box labeled Start from data for year (4 digits) is the first year of
the simulation. You may assume this is the current year, or it may be the year corresponding to when the data were collected. In this case we start at year 1997. Check that this says 1997. r In the next box labeled Number of Periods for base case (1 or 2 digits), you are required to determine how many periods you want to examine. In this case we are looking at five periods. Check that this figure is correct. r Below this box is a drop-down menu for the length of each period. In the GDyn model, a period may differ from 1 year. In this case, the length of each period, except the last one, is 5 years. r Following this is a list labeled Simulation and Sim names (3 Chars). In the GDyn model three simulations are undertaken: the base case, the base rerun, and the policy. At this stage you are required to give simulation names to all three of these experiments. In this case, use the names BAS, BRR, and POL for the Base Case [B], Base Rerun [R], and Policy [P] simulations, respectively. Later, when you do more
Implementing the Dynamic GTAP Model in the RunDynam Software 185
simulations, you may want to give these ones different labels to avoid writing over previous simulation results. A description of the three simulations follows: 1. The base-case simulation represents how we might expect the economy to look without the policy shock (i.e., without the productivity shock). Depending on what we know about the future state of the world economy, it could include our beliefs about population and labor growth rates or the state of tariff reductions over the simulation period. How to develop a baseline was discussed in Chapter 5. 2. The base-rerun simulation takes the policy closures and the base-case shocks. It is a calibration simulation. If any variables are endogenous in the base-case closure and exogenous in the policy closure, the program automatically takes the values of these variables from the base-case simulation and includes them as exogenous shocks to variables in the base-rerun simulation. The usefulness of the base-rerun simulation is that it allows you to reverse any calibration done in the baseline. For instance in Chapter 5 we exogenized real GDP (qgdp) to ascertain changes in region-wide technological change (afereg) to obtain that change in the real GDP. In the base-rerun simulation the values for the region-wide technology obtained in the base case are now applied in the base rerun to afereg to endogenously determine real GDP. If the base rerun has worked correctly, the resulting changes in real GDP in the base rerun should equal those applied in the base case, subject to small differences due to path dependency. The fact that real GDP is now endogenous means that, in the policy simulation, real GDP can now respond endogenously to the policy shock. 3. The policy simulation examines the effects of the policy shock, which is applied in addition to the other base-case shocks. The policy shock then interacts with the other changes expected to occur in the world economy. Both simulations are undertaken so that the difference between the two scenarios can be calculated, and hence the effects of the policy shock isolated. r The next step requires us to specify the current working directory: C:\RunDynam\HO3 × 3\. Check that this is in fact the working directory. If not, you can change it by clicking on the change button.
186
Ken Itakura et al. r Next you are asked to specify a mapping file: GDyn.map. The map-
ping file contains a list of variables you would like to place into a spreadsheet file for further analysis. You can view the mapping file by clicking on the button labeled Edit. r Finally, you are asked to specify a solution method. In this case we use Gragg: 2–4–6 steps extrapolation. You must also include automatic accuracy. It should appear in brackets next to the solution method. If not, automatic accuracy can be chosen by clicking on the Change button. In the bottom left-hand corner click on the box labeled automatic accuracy and check the options (keep the default options). Then click OK.
4.2 Closure/Shock Page Now we can proceed to the closure and shock file. r Click on the tab labeled Closure/Shock to move to the next page. This
page contains a table similar to the one provided here:
The table lists the shock and closure files for each period of the simulation. You will notice that the labels in the first column of the table relate to the information provided on the previous page: Sim Overview. The labels 2002, 2007, 2012, 2017, and 2020 are the last years for each of the periods. For example, the first period starts at 1997 and continues for 5 years to the beginning of 2002.
Implementing the Dynamic GTAP Model in the RunDynam Software 187
Base-case and policy shocks and closures must be specified for every period. Closures and shocks are specified in files (with extension ∗ .CLS for closure, extension ∗ .BSH for base shocks, and extension ∗ .PSH for policy shocks). If these are set up correctly, all the files (with extensions ∗ .CLS, ∗ .BSH, and ∗ .PSH) are indicated in black font. Any files in red do not exist in the working directory specified. Any files in green do exist, but they refer to another (constituent) file that does not exist. r To view a base-case shock file, place the cursor on the file you wish to
open, click the right-hand button on the mouse, and select Edit.
The shock file contains a list of variables and the corresponding shocks to these variables that are imposed in the base-case scenario. As mentioned previously the base-case scenario indicates what we might expect to happen in the world economy during the simulation period. In this case there are forecasts for the rate of real GDP growth (qgdp), population growth (pop), skilled and unskilled labor growth (qfactsup), shocks to import tariffs (tms), and export subsidies (txs).3 Finally, time is shocked by 5 to indicate that a 5-year period is being simulated; hence this first simulation is from the beginning of 1997 to the beginning of 2002. All the shocks refer to changes over a 5-year period. 3
If there are any variable names you are uncertain of, you can always go back to the TABLO file to find out what they are.
188
Ken Itakura et al.
In some cases the shocks are located in separate files (e.g., ENDW002.shk), which are referred to hereafter as the base-case shock file. These files can be viewed by selecting File | Open from the main menu in TABmate. If one of these files were missing, Y97 02.BSH would appear in green in the table on the Closure/Shock page. The policy shock files are given in the final column of the table on the Closure/Shock page. r Again the policy shock file can be viewed by clicking on the file
(AFEREG.PSH), clicking the right-hand button on the mouse, and selecting Edit. There are two differences between the policy shock file and the base-case shock file: 1. The policy shock file includes only the policy shocks – in this case, a single shock to productivity (afereg) – even though in the policy simulation all shocks from both the base-case shock file and the policy shock file are imposed. 2. The term ashock is used to shock this variable as compared to the basecase shock file in which the term “shock” is used. The term ashock stands for additional shock. Thus afereg(“row”) is shocked in the base
Implementing the Dynamic GTAP Model in the RunDynam Software 189
case and by an additional −5% in the policy simulation in the period 2002 to 2007. r Finally, the closure file (POL.CLS) can be viewed by clicking on the file
with the right click and selecting Edit. The closure file is used to set out which of the variables are exogenous or fixed and which are determined endogenously within the model. To close the model, the number of endogenous variables must equal the number of equations; otherwise the model will not solve. In the closure file, there is a list of exogenous variables, followed by the statement “Rest Endogenous.” The file POL.CLS specifies the standard Gdyn closure. A comparison of the policy and base-case closures will reveal that we are calibrating region-wide technological change to target forecasted real GDP.
4.3 Running the Simulation r Choose Tasks | Run Base, Base Rerun, and Policy to run the base-case,
base-rerun, and policy simulations consecutively.
r An information box will then appear telling you, “Beginning simula-
tion(s).” Click OK.
r If successful a message will appear:
r Click OK. Notice you have jumped onto the Results page.
4.4 Altering the Simulation As you become more confident with the program, you will wish to make your own simulations. We suggest making small changes to current applications – changing closures, shocks, and time periods – and then eventually making your own aggregations. To alter the closures and/or shocks, you simply edit the files we have been examining. To alter the length of periods or extend the time horizon, simply edit the Sim Overview page. Any changes to the length or extension of time will automatically adjust the Closure/Shock
190
Ken Itakura et al.
page; however, you will then have to alter/make new shock and closure files. Note that if you alter the length of the time periods, you will also need to alter the shock to time in the shock files – the program does not do this automatically, and it is easy to forget this step. When you start making your own simulations you should also do the following: 1. Rename the three-letter extensions given to the base, base rerun, and policy on the Sim Overview page. This will stop you from overwriting previous results, because changing these extensions will store the results under alternative names. 2. Also edit the simulation description by clicking on edit sim description on the Sim Overview page. This allows you to write a short description of the simulation, which will help you remember what this simulation was when you come back to it at a later date. 3. Save the ingredients of your simulations. r One method is to save the details of the simulation by selecting File | Save simulation details as. Then provide a name for the simulation. These details can then be reloaded at any time by selecting File | Load simulation details and selecting the relevant file. The disadvantage with this method is that it only saves the simulations details,4 not the ingredients5 themselves; thus if you change an ingredient you will be changing the simulation. r An alternative method is to zip up the ingredients of the simulation. This is done by selecting File | Save ingredients as Zip archive. We suggest you do this here. Save the ingredients as Example1f.zip. The disadvantage/advantage of this method is that it does not save the results. With all the ingredients and simulation details saved, the results are just a click away. If you wish to keep the results you need to zip them up yourself.
5. Viewing the Results Results are obtained for each period of the base-case, base-rerun, and the policy simulations and can be viewed in a variety of ways. The Results page is divided into two parts. The first section allows you to look at the results for all periods at once, whereas the second shows results for individual periods. 4 5
The simulation details refer to the setup in RunDynam and includes information recorded on the Sim Overview and Closure/Shock pages. The ingredients are the closure and shock files.
Implementing the Dynamic GTAP Model in the RunDynam Software 191
5.1 Viewing the Results for All Periods The first section looks something like this:
There are three parts to this first section: 1. Results for the base-case, the base-rerun, and the policy simulations and the difference between the base rerun and the policy can be viewed by selecting one of these options in the first column. The difference between the two simulations shows the effects of the policy shock and is comparable to the output of a comparative static model. 2. The results for some period(s) or all periods can be displayed in the AnalyseGE program, a spreadsheet, or within the ViewSOL program. Clicking the AnalyseGE box starts the AnalyseGE program. Similarly, you can see the simulation results by clicking the Spreadsheet box or the Graph/ViewSOL box. 3. The third part relates to style. Results can be displayed year-on-year (or in the GDyn model period-on-period) or cumulative.6 Year-on-year results show the percentage change in the variable occurring during that period, whereas cumulative results show the total percentage change in that variable occurring between the initial period and the period specified. Example 5.1: Viewing the Results Using ViewSOL The ViewSOL program is similar to the program ViewHAR used previously to examine the data. ViewSOL is much more adept at viewing the results than the spreadsheet. To view the results using ViewSOL you must first select the simulation you wish to view. r Click on the Differences – Policy v Rerun option. A small black dot
should appear in the white circle next to the selected option.
r Next select the style: year-on-year or cumulative results. Select year-
on-year by clicking on Year-on-year in the Style box. The choice of style at this stage is not important because both are available from within ViewSOL. However, the choice of simulation is important. 6
Note that when using AnalyseGE you can only view cumulative results.
192
Ken Itakura et al.
If you select Base Case then only the Base Case results will be available; however, if you select either Policy or Difference, all simulations will be available to view once in ViewSOL. r Now click on the box labeled Graph/ViewSol.
You are now in ViewSOL. The screen will look something like the following:
r Once in ViewSOL you will see a menu bar and a table of results. The
r r
r
r r
7 8
table is divided into four columns. The first one shows the variable name, the second displays the size of the variable or its dimensions, the third shows the number of variables in the list, and the final column provides a short description of the variable. You will notice that the cumulative difference results appear regardless of the options you selected in RunDynam. Your selections need to be made once again inside ViewSOL. To view the policy simulation you need to select Time series . . . | Show.. Pert | Perturbed Solution.7 To view the year-on-year results for the policy simulation you need to select Time series . . . | Show.. YonY | Year-on-year. Very little appears to change, except you will notice that the phrase “SEQ4 Diff Cum d” in the top right-hand corner has changed to read “SEQ4 Pert YonY p.” Check that this has changed.8 You can now view one of the variables by double-clicking on the variable name. Move down and find qgdp, the variable for percentage change in real GDP. You can now click on Contents in the main menu to move back to the table of variables. You can also check that the value of the shock to afereg was as expected. Find and click on afereg. Remember that this shock will depend on the shock imposed in the base-case scenario and in the policy scenario.
Note that if you opened the base-case simulation in RunDynam this option would not be available. If you cannot see this, click on the small box (next to the cross) in the very top right-hand corner to enlarge the screen.
Implementing the Dynamic GTAP Model in the RunDynam Software 193 r The cumulative results can also be viewed by selecting Time series . . .
| Show.. YonY | Cumulative. Again check that the phrase in the top right-hand corner reads “SEQ4 Pert Cum P.” r You can now exit from ViewSOL. Example 5.2: Viewing the Results Using AnalyseGE Cumulative results can also be examined in AnalyseGE, which is a program that is well suited for analysis (see Pearson, Hertel, and Horridge 2002). The program brings together the model code (tab file), the underlying data, and the solution file. Open the cumulative differences in AnalyseGE by selecting differences, all and cumulative, and then click on the AnalyseGE button.
AnalyseGE will load the model tab file, the cumulative differences between the baseline and the policy for the final period (2020), and the initial database, which in this case is the updated data for the period preceding the final period (i.e., 2017). Using the gloss and evaluate/decompose features in AnalyseGE, you can analyze the results for each equation. For example, follow these steps: r Search for rorge, the expected rate of return. r Once you have found rorge, select the Gloss button to see all occur-
rences of that variable.
194
Ken Itakura et al.
r The line numbers are given in red on the left-hand side, and the
highlighted lines indicate instances where the variable is on the lefthand side of the equation. Click on line 1754 to move to that line and the equation. r You can now view the value of rorge (cumulative difference) by placing your cursor on the variable (left click) and then using the right click to call up the menu.
r Clicking on Evaluate (selection or coeff/variable at cursor) will give
you the value of rorge.
r Using the decompose options will allow you to decompose the equation
or selection. Right click on rorge and select decompose all or one side of this equation. r You will be asked now to decompose; select RHS, intelligent and first toggle position.
Implementing the Dynamic GTAP Model in the RunDynam Software 195 r Once selected, ViewHAR will open up and you will obtain the follow-
ing:
r This is a decomposition of the equation for rorge.
rorge(r) = RORGFLEX(r)∗ [qk(r) − 100.0∗ KHAT(r)∗ time] − 100.0∗ LAMBRORGE(r)∗ ERRRORG(r)∗ time + srorge(r); r Intelligent decomposition breaks the equation up according to the
brackets and assigns a label (TempCoeff) to each section of the equation. The coefficients from each section of the equation are used for labeling purposes. For example, 1 e1 RORGFLEX refers to the first part of the equation – RORGFLEX(r) ∗ [qk(r) – 100.0 ∗ KHAT(r) ∗ time]; 2 e1 LAMBRORGE refers to the second part – RORGFLEX(r) ∗ 100.0 ∗ LAMBRORGE(r) ∗ ERRRORG(r) ∗ time. r So this tells us that in NAM qk rises relative to KHAT. Further decomposition will show that in this case qk is 1.66, because the cumulative difference in time and hence 100.0 ∗ KHAT(r) ∗ time is zero. r Although intelligent decomposition is usually best, this is not always the case. Try decomposing complete by variable to see the other features of AnalyseGE. r Close AnalyseGE and return to the RunDynam results page.
5.2 Viewing the Results for Some Years Example 5.3. Viewing Some of the Results Using AnalyseGE In some cases you may not want the cumulative differences for all of the years.
196
Ken Itakura et al.
r On the results page of RunDynam if you select some and then click on
AnalyseGE, RunDynam will ask for an RSL file. To obtain an RSL file you will need to cancel and go back to RunDynam. r Next select Tasks | Run SS Jobs for selected Years.
r RunDynam will then ask you to select the starting and end years and
the simulation.
Implementing the Dynamic GTAP Model in the RunDynam Software 197
r Select 2002 as the initial year and 2012 as the end year; then click OK.
Check the prerequisites and click OK.
r RunDynam will then run the SS job. Now when you select AnalyseGE
there is an RSL file for 2002 to 2012 that you can load up.
5.3 Viewing the Results for Individual Periods This section allows you to view the base-case and policy simulation results for each period, as well as the updated data and log files for the end of each period. It also allows you to examine the GDYNView, GDYNVol, and Welfare results for individual periods. The second part of the Results page looks like this. Example 5.4: Viewing the Updated Data To look at the 2012 updated database for the base-case simulation you must do the following: r First select which of the simulations you would like to view – base case
or policy. In this case, select Base.
198
Ken Itakura et al.
r Second, select the output type you wish to view. You can choose
from the Solution file containing the results for the variables; the Log file containing a log of the simulation, including any errors that may have occurred; Accuracy Summary of the simulation; or the GTAPDATA containing the updated database. In this case, select the updated database by clicking on GTAPDATA. r Third, select the label corresponding to the period you are interested in. In this case it is 2002. r Finally click on the view button to view these results. The ViewHAR program for viewing header array files is automatically opened. This should look familiar because you have used the same tool to look at the core database. r Exit from ViewHAR by selecting File | Exit. Release 9 of GEMPACK also makes it possible to undertake postsimulation processing. Hence, in addition to the Solution file and GTAPDATA, users now have access to a more detailed updated data file through WELVIEW, GDYNView, TAXRATES, and GDYNVol.9 When taking a closer look, you will be able to find that in fact GDYNView, GDYNVol, and TAXRATES are all placed at the end of the TABLO file inserted between the following two statements: PostSim(Begin); PostSim(End); Postsimulation processing thus makes it possible to carry out calculations with values of variables and coefficients that depend on previous simulation results within a single TABLO file. GDYNView gathers certain parts of the global data, with important information about certain macro variables and others regarding trade, transport, and protection. 9
GDYNVol output is placed in the updated GTAPDATA file; no special GDYNVol file is produced.
Implementing the Dynamic GTAP Model in the RunDynam Software 199
Follow the same steps to examine the GDYNView file and the GDYNVol file for selected periods. These files are similar to those in RunGTAP (Pearson and Nin Pratt 1999), with some minor modifications to include useful data for the GDyn model, such as rates of return and foreign income. The formulas for GDYNView, GDYNVol, and WELVIEW are included in the standard GDyn tab file as post-sim processing. Example 5.5: Viewing the Welfare Decomposition The welfare decomposition (WELVIEW) results are viewed in the same way as the GTAPView and GTAPVol results. Note, however, that the simulation will give zero results if time is shocked. Valid results will only appear in WELVIEW when time = 0 (i.e., when you undertake the comparative static simulation to create a valid welfare decomposition). Users wanting to undertake the welfare decomposition for the dynamic simulation should rerun this simulation (CH7HO3 × 3 gdyn v35 97.zip), but set automatic accuracy at 90%10 and then refer to Chapter 6 for details on the special closure and shocks used in the baseline. Here is a summary of the steps required to undertake a welfare decomposition simulation: r Rerun the dynamic simulation (CH7HO3 × 3 gdyn v35 97.zip), but
set automatic accuracy at 90%. Open the results in ViewSOL and leave open for later use.11 r Model/Data page: Change base data to the updated data file from the final year of the base rerun (e.g. dat-basb-brrr-2020.har). r Sim Overview page: Change labels so as not to overwrite your files. r Year and time period (you need just one year: start 2019 and length 1 year) r Alter file names (WDB, WDR, WDP). Although you will not need all three it is very easy to mistakenly overwrite one of the simulations by clicking the wrong button. r Closure/Shock page: Change the closure file by swapping the following (remember to save the new closure under a different name and load into the baseline): 10
11
Note that higher automatic accuracy helps ensure accuracy of the comparative static welfare decomposition simulation. The application CH6HO3 × 3 gdyn v35 97.zip and the welfare decomposition of that application CH6HO3 × 3_gdyn_v35_97wd.zip can be downloaded from the GTAP Web site at https://www.gtap.agecon.purdue.edu/models/ Dynamic/applications.asp. Closing the file will not delete it. However you will need to reopen it later, and switching between results can be time consuming. An alternative way of loading up other sequences of results is to use time series | load sequence. See help for details.
200
Ken Itakura et al. r Time variables
swap srorge = rorge; swap SDKHAT = DKHAT; swap sqk = qk; swap swqh = wqh; r Numeraire, e.g.,
swap psavewld = ps(“food”,“ROW”); r Closure/Shock page: Change the shock file (again remember to save
and load into RunDynam)
r Shock the following variables by the amount from the solution file for
r r r r
the cumulative difference between the policy and the base rerun. r rorge r DKHAT r qk r wqh ps for one commodity and region [e.g., ps(“food”,“ROW”)] Shock all the policy variables (by the cumulative differences from the dynamic policy simulation) Run base Open WELVIEW for the base year to view the welfare decomposition. (Note that if all the welfare results are zero, then there is a shock to time and you have made a mistake.)
This will give the (more-or-less) correct EV and decomposition. To check, you can compare the resulting database, from the comparative static welfare simulation, with the final database from the dynamic simulation. There will be differences, but they should be relatively small – less than 0.01% of the values for a simple simulation like this one. If the differences are larger, then try increasing the accuracy of the dynamic and comparative simulations.
6. Book Applications Several applications are provided with this book and with the RunDynam software. The file name containing the application is labeled Ch# model v# ##.zip, where Ch# refers to the chapter number in this book (e.g., Ch7 refers to the application used in Chapter 7), model refers to the model variant used (e.g., GDyn, Savings, or Farm variant, etc.), v# refers to
Implementing the Dynamic GTAP Model in the RunDynam Software 201
the tab file version used if relevant (e.g., v3 is the latest version of GDyn.tab at the time of this writing), and ## refers to the year of the initial GTAP 5 Data Base used (e.g., 97 is 1997, which corresponds to the GTAP 5 Data Base, as it has a reference year of 1997). These files are available on the GTAP Web site at https://www.gtap.agecon.purdue.edu/models/Dynamic/applications .asp. Chapter 6 Chapter 7 Chapter 8 Chapter 9 Chapter 10 Chapter 14 Appendix A
Ch6HO3 × 3_gdyn_v35_97.zip12 C6HO3 × 3_gdyn_v35_97wd.zip Ch7HO3 × 3_gdyn_v35_97.zip Ch8_gdyn_r19_95.zip Ch9_gdyn_r19_95.zip Ch10_dfarm_95.zip Ch14_gdyns_34_97.zip Ch7HO7 × 7_gdyn_v31c_97.zip
References Harrison, J. and K. Pearson. 1998. Getting Started with GEMPACK: Hands-on Examples. GEMPACK Document No. 8. Melbourne: Centre of Policy Studies and Impact Project, Monash University. Pearson, K. and A. Nin Pratt. 1999. Hands-on Computing with RunGTAP and WinGEM to Introduce GTAP and Gempack. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Pearson, K., T. Hertel, and M. Horridge. 2002. AnalyseGE: Software Assisting Modellers in the Analysis of their Results. Melbourne: Centre of Policy Studies and Impact Project, Monash University.
12
This is the same application as Ch7HO3 × 3_gdyn_v35_97.zip, but with automatic accuracy of 90%.
PART III
APPLICATIONS OF DYNAMIC GTAP
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
EIGHT
Assessing the Impact of China’s WTO Accession on Investment Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina
1. Introduction During the 1980s and 1990s, China carried out significant reforms to restructure and open up its economy to foreign trade. These reforms helped spur a period of rapid growth, with annual growth in real per capita GDP averaging 6.04% over the period 1978–95 (Maddison 1998). However, reforms to the legislative framework governing foreign investment in China have proceeded more slowly. Entry into certain sectors, as well as the share of foreign ownership, is significantly limited, and joint ventures typically remain the best option for foreign investors. In an effort to boost foreign investment, the Chinese government began offering a number of special incentives in the early 1990s, including duty drawbacks on imported intermediate inputs and capital goods used for the production of exports, exemptions and reductions in the rate of income taxes paid on profits,1 and preferential tax rates for foreign enterprises that reinvest their profits (China Council 2000). These incentives paid off, with foreign investment increasing sharply in 1992–3 (Fig. 8.1). Indeed, by 1994 China accounted for approximately 20% of all foreign direct investment (FDI) in developing countries (Garbaccio 1995). However, by 1996 the rate of growth in FDI had begun to slow (Fig. 8.1). A survey of the Chinese economy by The Economist (2000) explored some 1
Incentives are numerous and depend on the sector and/or the region within China where the investment is undertaken, as well as the purpose of the production activity (e.g., export sales). The duration of incentives also varies and may be for a period of up to 5 years (China Council 2000).
Original version published in Pacific Economic Review, 11(3), 315–39, 2006. The application related to this Chapter is Ch8 gdyn r19 95.zip and is available for download on the Web site at https://www.gtap.agecon.purdue.edu/models/Dynamic/applications.asp.
205
206
Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina
60,000
50,000
US$ Millions
40,000
30,000
20,000
10,000
0 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002
Figure 8.1. Foreign investment in China. Sources: IMF Balance of Payments Statistics, 1999 and China’s Statistical Yearbook, 2002 (US$ millions).
of the reasons behind the FDI slowdown in China. In many cases investors’ high hopes for this market were slow to materialize, with the absence of a rules-based economy making it difficult for outsiders to operate effectively in China. Informal relationships and corruption still hindered many business transactions by foreigners, and inefficient state enterprises still dominated many key sectors of the economy. China’s push for accession to the WTO was partly an attempt to remedy some of these fundamental problems, thereby promoting foreign investment. Indeed, the recent upturn in FDI provides some evidence that investors have responded positively to China’s accession. It is this linkage between WTO accession and foreign investment that provides the focal point for this chapter. China’s bid to rejoin the General Agreement on Tariffs and Trade (GATT) began in 1986 and culminated in the Doha meetings of the WTO, when membership was finally approved by that body. The final accession agreement reflects the importance to the U.S. and European negotiators of increasing access for foreign investors – particularly in the services sectors. In the words of Aaditya Mattoo of the World Bank, “China’s GATS commitments represent the most radical services reform program negotiated in the WTO. China has promised to eliminate over the next few years most restrictions on foreign entry and ownership, as well as most forms of discrimination against foreign firms” (Mattoo 2001). Clearly the stage is set for increases in foreign investment.
Assessing the Impact of China’s WTO Accession on Investment
207
Domestic investment is also expected to rise as a result of WTO accession. In the 1990s FDI inflows surged because domestic firms were uncompetitive, domestic savings were inefficiently allocated, and laws put domestic companies at a disadvantage relative to foreign companies (Huang 2002).2 Tax advantages for export processing have benefited mainly foreign-owned companies, whereas domestic content requirements have restricted the use of imported intermediate inputs by companies selling locally and have been a particular problem for the domestic automobile industry. In preparation for WTO accession, China reformed its export-processing system and extended tax benefits to all companies producing exports, not only the foreign-owned ones. Commitments as part of WTO accession have led to the removal of the local content requirement and to reforms that will ensure fair treatment for both domestic and foreign companies (no discrimination) and more efficient use of domestic savings. Therefore, it is expected that China’s WTO accession will have a positive impact on domestic investment. China’s accession to the WTO has generated a large amount of interest, and extensive research has already been undertaken on this topic.3 Most of this work has focused on the trade implications of accession – largely abstracting from the likely impact on investment. Analysis of the impact of China’s WTO accession on investment has been much more limited. McKibbin and Tang (2000) offered an early assessment of this issue. They used the G-cubed model to examine the effect of three possible developments in China: (1) the liberalization of trade, (2) the liberalization of financial markets, and (3) the possibility that financial liberalization might result in a financial crisis within China. They found that financial liberalization could result in significant gains, particularly for China itself. They did caution, however, that many of the problems affecting the Asian economies at the time of the Asian crisis, such as a fragile banking 2 3
The result was an increase in FDI even in industries in which Chinese firms have been strong in the past, such as textiles, herbs, and others. Anderson et al. (2000) used the GTAP model to determine some early estimates of the effects of China’s accession. Ianchovichina and Martin (2001, 2003) updated these estimates by taking into account duty drawbacks in China. Wang (1997a, 1997b, 2001, 2002) has also undertaken a number of studies examining the effects of China’s WTO accession using the GTAP Data Base. In general, the results show that world trade increases substantially as a result of China’s accession. The main winner from China’s accession is China itself. North America and many of the other developed nations also gain as a result of increased exports, particularly of agricultural products (Wang 1997a). The removal of quotas for China’s apparel and textile exports under the ATC agreement appears to be a significant contributor to the benefits accruing to North America. These results are consistent with the findings by other authors (Hertel et al. 1996; Harrison, Rutherford and Tarr 1996; Walmsley and Hertel 2001; Yang 1996).
208
Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina
system, were also features of the Chinese economy, and therefore they argued that financial liberalization would be much riskier than liberalization of trade. However, there are a number of limitations to this analysis. First, they used a very old database (1991), which reflects a very different Chinese economy than that poised to join the WTO. A second limitation is that their analysis was not based on China’s actual offer to the WTO. Instead their accession scenario was highly stylized, representing the full removal of protection in China from their 1991 base. Finally, McKibbin and Tang (2000) did not reflect the system of preferences for export enterprises in China.4 Walmsley and Hertel (2001) offer a more recent assessment of China’s WTO accession. Although their paper focuses on the uncertainty in the timing of the abolition of textile and apparel restrictions on China’s exports, they also report findings related to changes in foreign investment resulting from China’s WTO accession. Their analysis shares several important features with the present chapter. However, they, too, fail to account for the duty drawback regime currently in place in China. In addition, Walmsley and Hertel (2001) ignore the problem of revenue replacement. Another significant limitation of their paper has to do with the construction of the baseline scenario in which the authors show foreign investment declining after 1997. This does not square with what has been observed since 1999. Moreover, Walmsley and Hertel (2001) focus only on the issue of trade liberalization, whereas this chapter extends the analysis to encompass liberalization of trade and investment in services. We address many of the shortcomings of the Walmsley and Hertel (2001) paper and focus specifically on the liberalization and productivity gains in the services sector. The revised version of the GDyn model used in this chapter places international capital mobility at the forefront, thereby providing a useful vehicle for exploring the impact of China’s WTO accession on investment and economic growth in China. We also deal directly with the problem of duty drawbacks by modeling the export-processing sectors separately. Furthermore, we assume that the Chinese government will
4
Ianchovichina (2003) found that the failure to account for duty drawbacks could lead to a substantial overstatement of the impact of WTO accession on China’s light manufacturing industry. For example, she showed that failure to account for China’s duty exemptions in analyzing WTO accession resulted in overstating the increase in China’s trade flows by 40% and in exports of selected sectors by 90 percent. This limitation is even more severe when one considers the fact that the duty exemptions have been extended to investment goods imported by joint ventures and foreign-owned companies.
Assessing the Impact of China’s WTO Accession on Investment
209
replace lost tariff revenue with increased consumption taxes. This assumption seems quite reasonable, because it reflects a continuation of the trend over the past decade in which consumption taxes rose from 18% to 79% of total revenue. Moreover, the accession agreement used in the chapter is based on the August 1999 offer agreed to by China and the United States and obtained from Martin et al. (2000).5 In addition to reducing tariffs on goods, we examine the liberalization of rules governing direct trade in services and the impact of potential productivity gains in the services and automobiles sectors. Finally, special attention is given in the baseline to tracking the growth rate of foreign ownership over the period 1995–2002 and reflecting the fact that the removal of restrictions on foreign ownership and trade reform in China in anticipation of WTO accession has already had a significant effect on the share of foreign investment. Our basic macroeconomic results confirm the findings of earlier studies (Ianchovichina and Martin 2001, 2003; Wang, 2001), with China expected to gain most from accession relative to other countries. The new insight provided by our chapter relates to the implications for investment and ownership of productive assets. We find that accession significantly boosts investment, doubling the extent of foreign ownership of Chinese assets relative to the no-accession baseline. Central to this increase in foreign ownership is the expected boost in productivity of the services sectors owing to the liberalization of rules governing foreign investment in these sectors. The next section of this chapter discusses the details underlying the design of the accession scenario. Section 3 examines the results of this scenario, comparing them to the baseline. Finally Section 4 summarizes the findings and offers concluding remarks.
2. Modeling China’s Accession to the WTO The effects of China’s accession are examined over the period 1995 to 2020. We begin with the GTAP 4 Data Base (McDougall, Elbehri, and Truong 5
In most studies on China’s WTO accession, including Ianchovichina and Martin (2001) and Wang (2001), the bilateral agreement between the United States and China served as the basis for determining the extent of liberalization due to WTO accession. The agreement cannot be represented as a move to free trade or a simple proportional cut in protection as many studies have tried to do (for a survey, see Gilbert and Wahl 2001). More recently, Ianchovichina and Martin (2003) used the final multilateral commitments, which became publicly available after the Doha meeting. Their results confirm that the U.S.-China bilateral agreement indeed serves as a good basis for assessing the extent of liberalization in China as part of WTO accession.
210
Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina
1998)6 augmented with additional data for the GDyn model. This time frame is divided into a number of unequal intervals to provide the highest resolution over the accession period. These intervals are as follows: 1995– 2000,7 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2010–15, and 2015–20. Two simulations are undertaken: one representing the baseline (without China’s accession) and one representing the policy scenario (with China’s accession). The baseline scenario provides a picture of what we expect the world economy to look like without China’s accession to the WTO8 and is similar to the baseline described in Chapter 5. The policy simulation examines the effect of China’s accession on both domestic and foreign investment. We assume that accession commences in 2002 and that it takes 5 years to implement the agreement; hence accession is expected to be completed by the beginning of 2007, but the effects on foreign ownership and income payments are expected to linger for another decade, as we will see in our results that extend to 2020. For comparison purposes and because the reader may find some elements of the simulation more convincing than others, we examine both the total effect of China’s accession and the individual contribution of five components, representing five different elements of China’s accession agreement, to the changes in investment and welfare. These components are as follows: 1. 2. 3. 4.
the gradual reduction in risk premium expected by foreign investors the removal of tariffs and quotas as outlined in the agreement the liberalization of cross-border trade in services the expected improvements in productivity in the automobile sector resulting from accession 5. the liberalization of rules governing the foreign commercial presence in the services sector and associated sector-specific productivity impacts In the remainder of this section, we discuss the policy scenario in detail, including the shocks used to incorporate each of these aspects and the assumptions made.9 6 7 8 9
See the Appendix for a list of regions and commodities used in this application. The period starts from the beginning of 1995 and ends at the beginning of 2000. The baseline scenario uses the model adjusted to take account of duty drawbacks in China. Note that the liberalization of services (points 3 and 5) is discussed simultaneously under the heading “Services Liberalization,” because of their interrelated nature. However, in analyzing the results we chose to separate the effects of services trade liberalization and productivity improvements in the services sector for the reasons outlined earlier – namely that readers may be less convinced of China’s ability to attain the productivity gains assumed.
Assessing the Impact of China’s WTO Accession on Investment
211
2.1 The Gradual Reduction in Risk Premia Expected by Foreign Investors Data on the value of foreign investment in China obtained from IMF Balance of Payment Statistics (1999) and China’s Statistical Yearbook (2002: Fig. 8.1) suggest that foreign ownership in China has increased significantly more over the pre-accession period than was justified by the changes in expected rates of return resulting from the pre-accession tariff cuts and the assumption of minimizing any changes in the portfolio shares of foreign and domestic investment in GDyn.10 The increase in the share of foreign ownership in China may have been the result of a number of factors, including tax incentives for foreign direct investment, removal of restrictions on foreign direct investment in some industries (mostly notably, light manufacturing and electronics), and a decline in risk premia, which have occurred as China gradually opened up its economy and accession to the WTO became increasingly assured. Despite creating incentives for FDI in the 1990s, as of 2002 China still had a long way to go in terms of removing restrictions to foreign direct investment: Joint ventures remained the best way for foreigners to invest in China, and certain sectors such as services were closed to foreign investment. We therefore assume that foreign ownership will continue to increase relative to the baseline11 as risk premia continue to fall, although the effect will diminish gradually. Hence by 2007, any affinity by foreign investors to increase foreign ownership in China, caused by China’s opening up to foreign investment, will most likely have been eliminated.12 10
11
12
According to foreign investment data, the growth rate of domestic ownership in 2000 was 10% less than the model would have predicted, given the actual and expected rates of return in the baseline scenario. In 2001 it was 9% less than the model predicted. Some might argue that the increased share of foreign ownership should be viewed as part of the baseline scenario, reflecting the fact that it is the result of China’s broader attempts to open up its economy and introduce a rules-based system. Alternatively, it might be argued that the two are inseparable, and thus at least part of the decline in risk premia is related to accession. We find that a good way to think of this question is to consider the true “counterfactual”: What would happen if China’s accession had failed? Would this trend continue, or is it more likely that foreign investors would reconsider their positions in China? We believe that the latter would have been more likely. For this reason we have chosen to include a continued rise in foreign ownership as part of China’s accession. Of course, we separate this effect for purposes of analysis and permit the reader to apply his or her own judgment in this matter. This type of approach, which uses historical data to infer future changes, was formalized by Dixon and Rimmer (2002). Baldwin, Francois and Portes (1997) adopt a similar approach in modeling the impacts of EU enlargement on investment in Central and Eastern Europe.
212
Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina
Table 8.1. Tariff barriers, services liberalization, and productivity gains for Chinaa
Sectors
Cross-border services liberalizatione
1995 Tariffs pre- Tariffs posttariffsb accessionc accessiond
Crops 3.92 Livestock 7.33 Food and beverages 21.51 Extraction 8.50 Textiles 57.52 Wearing apparel 76.12 Metals and chemical 19.26 products Autos 129.23 Electronics 22.16 Other manufactures 23.48 Household utilities 0.12 Trade and transport 0.00 Other services 4.20
−0.39 2.30 20.90 5.45 30.39 31.84 14.82
−0.39 2.30 20.25 3.46 9.25 15.80 10.87
31.34 12.75 17.37 0.16 0.00 4.20
20.37 4.93 11.06 0.15 0.00 4.20
Productivityf
20.0
5.90 (1.15% p.a.) 0.92 (0.18% p.a.) 10.48 (2.01% p.a.)
4.6 4.6 4.6
a
Table 8.1 does not include details of shocks to incorporate the change in the risk premium or the removal of MFA quotas. b Tariff rates in GTAP 4 Data Base (1995). c Tariff rates after reductions in baseline. This is due to the fact that China had already undertaken considerable trade liberalization before its accession. d Tariff rates after accession, as per WTO accession agreement (as of August 1999). e Cumulative shocks. Yearly shocks in brackets. f Cumulative shocks. Shocks are applied as per Figure 8.3.
2.2 Tariffs and Quotas Constructing an accurate accession scenario is a complex task – even when it comes to tariff cuts. The tariff reductions agreed to by China as part of the accession offer were obtained from Martin et al. (2000) and are based on China’s offer as of August 1999. However, in many cases, these reductions had already been implemented as part of China’s preparation for WTO accession. The approach that we take is to compare the offer to China’s 1997 applied tariffs. Where the binding is lower, the offer is taken as a change in policy. Table 8.1 shows China’s tariff rates before and after accession. In the case of Taiwan (China), the cuts are based on their announced target of 4% average tariffs for manufactures.13 13
No data were available on Taiwan’s (China) offer for agriculture at the time of writing, and hence no shocks were applied. More recently data on Taiwan’s accession agreement have become available. Results using Taiwan’s actual accession agreement can be found in
Assessing the Impact of China’s WTO Accession on Investment
213
The reduction in tariffs is assumed to occur in equal installments over the entire period and to refer only to imported intermediate goods used in the production of commodities for domestic consumption, as well as to other imports for final private and government consumption. China’s tariffs on imported intermediate goods used for the production of exports and investment goods remain unchanged and are zero, reflecting the presence of duty exemptions. To achieve this type of tariff reduction we extended the GDyn model to take into account duty exemptions on imported inputs and capital goods used in the production of China’s exports.14 Finally, to avoid biasing our accession results by permitting the government to run an increased deficit as a result of accession, we offset these tariff cuts with increases in the current consumption tax, thereby maintaining unchanging total tax revenue, relative to net national income, over the period 2002–7. We also assume the quotas on China’s textiles and clothing exports to North America and the European Union will be removed by the beginning of 2007. Based on experience with implementation of the Agreement on Textiles and Clothing under the Uruguay Round Agreement of the WTO (Spinanger 1999), we assume that liberalization of the quotas on imports from China and Taiwan (China) will be back-loaded, with the bulk of the impact not felt until the final two years. Specifically, we follow the method used in Walmsley and Hertel (2001), except that accession does not commence until 2002.
2.3 Productivity in the Automobiles Sector The automobile sector is currently one of the most heavily protected industries in China, with an average pre-accession tariff rate for vehicles of about 70%. In addition, barriers to internal trade in automobiles have arisen in response to the efforts of individual provinces to foster their own auto industry. The government also sets prices and limits competition to favor local producers (Francois and Spinanger 2002). As a consequence, the opening of this sector to world trade is expected to lead to substantial restructuring and rationalization of the motor vehicle industry. Francois and Spinanger (2002) point out that the 13 plants currently producing sedans in China average only 39,000 units per year – roughly one-tenth of the minimum
14
Ianchovichina and Walmsley (2003). The results for China, the focus of this chapter, do not change significantly as a result of including Taiwan’s actual accession agreement. We incorporate duty drawbacks into the model following the method outlined in Ianchovichina (2003).
214
Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina
4.5 Autos Productivity
4
Services Productivity Alternative Services Productivity
% Change in Productivity
3.5 3 2.5 2 1.5 1 0.5 0 2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
Figure 8.2. Automobile and Service Sector Productivity shocks (% point differences from baseline growth rates). “Alternative services productivity” refers to the case where service sector productivity peaks at 2.7% (this corresponds to the annual productivity gain estimated by Mai et al., 2003).
efficient scale, by world standards – and that WTO accession could bring as much as a 20% increase in total factor productivity in the auto sector. In this chapter, we adopt the Francois and Spinanger productivity gain estimate, and just as they do we assume that the increase in productivity of the auto sector is expected to occur in the assembly of automobiles, rather than in the production of parts. Because our analysis is dynamic, we must also distribute these gains over time. For this purpose, we assume that the gains in productivity occur slowly, peaking in the last year of accession (2006) and then falling back to the baseline rates by 2010, as illustrated in Figure 8.2.
2.4 Services Liberalization As noted in the introduction, China’s commitments in the services sector “represent the most radical services reform program negotiated in the WTO” (Mattoo 2001: 1) and, in this sense, are the most significant part of China’s WTO accession package. Therefore, omitting the services commitments from the evaluation of the WTO accession15 seriously underestimates the gains from trade reform in China. 15
The vast majority of studies have abstracted from commitments in services when evaluating the economy-wide impacts of China’s WTO accession.
Assessing the Impact of China’s WTO Accession on Investment
215
The WTO groups services commitments into four “modes”: (1) the delivery of services across national borders (direct trade in services), (2) the consumption of services abroad (e.g., tourism), (3) commercial presence (i.e., foreign direct investment in the services sector in China), and (4) the movement of natural persons (foreigners providing services in China in person). For purposes of our study, we focus on modes 1 and 3, because they appear to be quantitatively the most important and have the strongest implications for foreign investment – particularly commitments with respect to mode 3, the commercial presence in the services sector. Cross-border supply of services: According to Mattoo (2001: 7) China has committed to allow the cross-border supply of professional services, to remove quantitative limitations on accountancy firms, to offer taxation services outside of “economically developed areas,” and to improve urban planning and legal services. Cross-border supply of education services is also fully open. Commitments on cross-border supply in the area of logistics, distribution, and transport are mixed. Although nothing is offered at the wholesale distribution level, the cross-border supply of retail services through mail-order is allowed – presumably opening the door to retail ecommerce. In the area of transport services, there are provisions relating to the cross-border supply of airline services, as well as maritime services to the 130 ports that are open to foreign shipping (Wenping and Findlay 2001). Francois and Spinanger (2002) provide estimates of the tariff equivalents of these nontariff barriers before and after China’s accession for wholesale and retail trade, transportation, communication, construction, finance, insurance and real estate services, other commercial services, and other services (such as public health). We use these tariff equivalents to estimate the reduction in price resulting from the liberalization of services. The tariff equivalents are aggregated using weights obtained from the GTAP Data Base to determine the pre- and post-accession tariff equivalents for the three services sectors used in this analysis (see Table 8.1 for shocks for China). As with tariffs, we assume that the liberalization of direct trade in services occurs gradually over the accession period (2002–7). To introduce the impact of these cuts in tariff equivalents on direct trade in services, we use the approach developed in Chapter 9. Foreign Commercial Presence in Services: From the point of view of foreign investment, the most important aspect of the services agreement is that pertaining to commercial presence, because it has a direct impact on the constraints facing firms interested in investing in China’s services sector. Of special interest are the anticipated changes in telecommunications,
216
Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina
transport, logistics,16 and financial services.17 In the telecommunications sector, foreign companies can provide value-added and mobile telephone services only in (and between) the three richest metropolitan areas (Shanghai, Guanzhou, and Beijing) and only through joint ventures with majority Chinese ownership. With WTO accession, by 2004, foreign companies may also provide fixed-line services, and by the end of the accession period geographic restrictions will be eliminated and equity restrictions relaxed (Mattoo 2001). These reforms promise to make a relatively competitive mobile phone market even more efficient, and fixed-line services are expected to undergo major increases in productivity (Pangestu and Mrongowius 2001). Because most of the reforms to rules governing the foreign establishment of a commercial presence in services are phased in gradually, we implement the asymmetric, triangular pattern of productivity shocks (Fig. 8.2). The maximum deviation from baseline occurs in 2007, when productivity growth is assumed to be 1% per year relative to the baseline rate of growth.18 This WTO-induced effect subsequently diminishes and disappears by 2010, leaving total factor productivity in the services sector 4.58% higher than under the baseline in 2010. This productivity enhancement appears to be rather modest in light of the major reforms being undertaken by China and of the findings of Mai, Horridge, and Perkins (2003), which use historical simulations to estimate potential productivity growth in services in China over the accession period. They predict a productivity growth of 2.7% per year in services over a 16
17
18
China’s commitments in the logistics area are particularly striking. By 2006, the distribution services sector would be largely open to foreign investment, with few ownership restrictions and no geographic restrictions (Mattoo 2001). This is a sector that has hitherto been largely in the domain of the state planners, with each sector organizing its own warehousing and storage activities. At the local level, government and enterprise functions were not separated, so the administration of the logistics sector was severely fragmented (Wenping and Findlay 2001). The infusion of foreign supply chain expertise could have a profound impact on productivity in this sector. Road and rail transport services are also scheduled for full liberalization by 2007. In short, there is enormous scope for productivity gains in the key transport and logistics activities that lubricate a modern economy. Before accession, foreign banks could only operate in a few regions, and they could only accept deposits from nonresidents and only in foreign currencies. Under accession, this sector will be gradually and fully liberalized. First the geographic restrictions will be eliminated, then the local currency restriction will be dropped in selected regions, and by 2006 the banking services will be fully liberalized. Similar liberalization will be undertaken in the insurance services sector. These changes promise to greatly increase competition and productivity in financial services. Note there is no sector-specific productivity differential in the baseline. Region-specific productivity is calibrated to achieve forecasted GDP growth.
Assessing the Impact of China’s WTO Accession on Investment
217
10-year period following accession. We therefore also explore the effect of a higher rate of productivity growth, in which productivity growth is assumed to peak at 2.7%/year relative to the baseline rate of growth.19 As we see later, even these smaller cumulative productivity shocks are sufficient to play a dominant role in the overall accession story.
3. Impact of China’s WTO Accession In this section, the results of China’s accession are discussed. We begin by providing an overview of the macroeconomic effects of accession on capital accumulation and investment before turning to the specific question at hand – namely the impact on domestic and foreign ownership. The section concludes with a summary of the long-run welfare consequences of China’s accession.
3.1 Macroeconomic Impact on Capital Accumulation A comprehensive picture of the global impact of China’s accession is given in Table 8.2, which shows the cumulative percentage changes relative to the baseline at the beginning of 2020, thus highlighting the long-run effects of China’s entry into the WTO. Overall the results concur with the findings of other research: China and Taiwan (China) are the primary beneficiaries of China’s accession in terms of both production and welfare. In the long-run, with the additional accumulation of capital, these benefits are even more pronounced. Here, we trace the major mechanisms determining the changes in investment, capital, and real GDP. As a result of China’s accession the rental price of capital is 4.3% higher than the baseline by 2007. The increase in capital’s rental price is the result of increased demand for capital by all industries, but particularly in the wearing apparel and auto sectors, where exports increase as a result of the removal of tariffs and quotas (and productivity gains in the case of autos), and in the services sectors, where production has increased as a result of the services liberalization and productivity gains. The effect of liberalization in the cross-border trade in services on the rental price of capital is small (0.1% increase in real GDP)20 when compared to the potential impact of 19
20
This leads to a cumulative change in services productivity of 12.794%, which is substantially higher than the total factor productivity change of 4.58%. Mai et al. (2003) estimated productivity growth of 2.7% per year. However, Mai et al.’s estimates may be overestimates because they abstract from reform at the border in their study. This is because cross-border trade in services is a relatively small share of total trade.
218 Table 8.2. Impact of China’s accession (2020)a
China Taiwan (China) North America Western Europe Japan Other NICs South East Asia South Asia Latin America Africa and Middle East ROW a
Real GDP
Capital stocks
Change in trade balance % of GDP
Actual rate of return
Regional wealth in domestic assets
Regional wealth in foreign assets
Foreign wealth located in region
Wealth of H’hold
Welfare (p.c. in utility)
22.52 4.06 −0.23 −0.27 −0.46 −0.10 −1.08 −0.86 −0.31 −0.43 0.05
25.82 9.94 −0.65 −0.87 −0.92 −0.07 −1.92 −1.09 −0.65 −0.85 −0.28
11.22 2.09 −0.31 −0.39 −0.43 −0.12 0.05 −0.55 −0.23 −0.27 −0.34
−2.37 −0.86 0.36 0.57 0.46 0.29 0.30 0.42 0.25 0.44 0.41
21.05 9.72 0.15 0.35 0.08 0.83 −1.22 −0.76 0.78 0.50 0.81
−19.66 −5.05 3.45 1.47 2.83 2.26 5.42 4.93 2.98 2.53 1.71
126.54 26.79 −3.04 −0.75 −2.59 −0.58 −7.44 −6.15 −1.37 −1.49 −0.09
9.00 3.16 0.41 0.73 0.83 1.01 0.56 −0.75 0.79 0.55 0.82
15.99 3.06 0.08 0.00 0.02 0.28 −0.21 −0.66 0.10 0.05 0.48
Unless otherwise stated, results represent cumulative percent changes from baseline from all five components of China’s accession. Source: Authors’ simulations with GDyn.
Assessing the Impact of China’s WTO Accession on Investment
219
6 Risk Premium 5
Tariffs
Cum % Differences Relative to Baseline
Cross-Border Services Liberalization 4
3
Autos Productivity Services Productivity
2
1
0
-1
-2
-3 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020
Figure 8.3. Cumulative percentage differences from baseline in China’s actual rate of return. Each curve shows the cumulative percent difference between the baseline and the policy. The policy is assumed to be the particular aspect referred to plus all the previous shocks listed. Source: Authors’ simulations with GDyn.
productivity growth in services, which accounts for 12.2% of the 22.5% increase in real GDP. In addition, the price of capital goods in China gradually declines over the accession period.21 This decline is primarily fueled by declines in the prices of domestic intermediates purchased by the capital goods sector. These prices fall as a result of the removal of tariffs on their imported intermediates and the productivity gains in the automobile and service sectors. Again, the liberalization of cross-border trade in services plays a minor role in the decline in the price of capital goods. The combined effect of the rise in the rental price of capital and the decline in the price of capital goods causes the rate of return on investment in China to increase.22 Figure 8.3 illustrates the effect of the various features of China’s accession on the rate of return. The removal of tariffs and the 21
22
This decline would have been much larger if we had not incorporated the preexisting duty drawbacks on imported intermediate imports purchased for the production of capital goods. The rate of return is the ratio of these two prices. It ignores capital gains (losses) because GDyn is a recursive dynamic model.
220
Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina
25 Risk Premium
Cum % Differences Relative to Baseline
Tariffs 20
Cross-Border Services Liberalization Autos Productivity Services Productivity
15
10
5
0 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020
Figure 8.4. Cumulative percentage differences from baseline in China’s real GDP. Each curve shows the cumulative percent difference between the baseline and the policy. The policy is assumed to be the particular aspect referred to plus all previous shocks. Source: Authors’ simulations with GDyn.
productivity shocks appear to have the largest impact on the rates of return in China, whereas the effect of liberalization of cross-border trade in services is minimal because of its limited effect on the price of capital goods and the rental price. Investment and hence capital stocks increase in response to the higher rates of return in China as a result of WTO accession. These larger capital stocks serve to gradually bring the rate of return down and closer to the global rate of return after 2007 when accession is complete. Eventually, the rate of return in China falls below that in the baseline scenario. This is the result of the capital accumulation mechanisms in the model, which allocate investment in such a way as to eliminate differences in the rates of return across regions. As a result of WTO accession, 2020 capital stocks increase by 25.8% and real GDP by 22.5%, relative to the baseline (Table 8.2). Figure 8.4 divides the effect on real GDP according to the five components of China’s accession. The components associated with improvements in the productivity in the services and automobile sectors have the greatest impact on real GDP.
Assessing the Impact of China’s WTO Accession on Investment
221
4.5
4
3.5
3
%
2.5
2
1.5
1
Baseline
Risk premium
Tariffs
Cross-Border Services Liberalization
Autos Productivity
Services Productivity
0.5
0
Figure 8.5. Share of foreign ownership of Chinese assets. Each curve shows the share of foreign ownership of Chinese assets over time resulting from the particular aspect referred to plus all previous shocks. Source: Authors’ simulations with GDyn.
As mentioned previously, it was assumed that any tax revenue lost as a result of lower tariffs would be replaced with consumption taxes. The power of these taxes rises by 1.32% in 2007. As the productivity gains in autos and services take hold, this revenue replacement requirement disappears.
3.2 Domestic and Foreign Ownership In this chapter we are particularly interested in the impact of WTO accession on domestic and foreign ownership of those capital stocks. Figure 8.5 shows that the share of foreign ownership in China grew strongly between 1995 and 2002. After 2002, however, growth in foreign ownership in the baseline occurs at a more modest rate (Fig. 8.6). As a consequence, the share of foreign investment in the baseline falls between 2002 and 2020. This decline in the share occurs because China’s income and hence savings are growing faster than in other regions; hence Chinese households can fund more of China’s investment. With rates of return increasing as a result of China’s accession, foreign investment is attracted to China (Fig. 8.6), and the share of foreign ownership in Chinese assets also rises (Fig. 8.5) as China’s regional investment
222
Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina
400 Baseline 350
Risk Premium Tariffs
Cumulative % Changes
300
250
Cross-Border Services Liberalization Autos Productivity Services Productivity
200
150
100
50
0 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020
Figure 8.6. Gross foreign ownership of Chinese assets. Each curve shows the cumulative percent change of foreign ownership of Chinese assets, in the baseline and under each of the policy simulations undertaken.Source: Authors’ simulations with GDyn.
rises faster than saving. Table 8.2 reports that foreign wealth located in Chinese assets increases by approximately 126% relative to the baseline by 2020. Most of this increase is due to the boost that China’s accession to the WTO is expected to give to productivity (associated with the rise in the foreign commercial presence in the service sectors), followed by the reduction in risk premia23 and tariff liberalization (Fig. 8.6). Foreign and domestic ownership are little affected by the liberalization of cross-border trade in services. Each of the elements of WTO accession raises the share of foreign ownership relative to the baseline (Fig. 8.5). However, it is only when restrictions on the foreign commercial presence in services are eliminated that the share rises in absolute terms. Eventually, income and hence saving in China rise sufficiently to cause the domestic share of this investment to rise and the foreign share to fall. 23
In Figure 8.6 the line labeled “Risk Premium” shows the effect on foreign ownership if we assume that the current rising trend in foreign ownership continues. In this case the marked slowdown in growth of foreign ownership in the baseline is reduced. The decline in the share of foreign ownership still occurs, although the effect is delayed.
Assessing the Impact of China’s WTO Accession on Investment
223
30 Baseline 25
Risk premium Tariffs Cross-Border Services Liberalization
20
Autos Productivity
%
Services Productivity 15
10
5
0
Figure 8.7. Share of Chinese wealth located in foreign assets. Each curve shows the share of Chinese assets located abroad over time resulting from the particular aspect referred to plus all previous shocks. Source: Authors’ simulations with GDyn.
China’s accession also increases the total wealth of Chinese households. Figure 8.7 shows the share of this wealth allocated to foreign assets. In the baseline, Chinese households increase their holdings of foreign assets. However, with China’s accession, higher relative rates of return at home make domestic investments more attractive, and the share of Chinese wealth located in foreign assets falls as the share in domestic firms rises. The increase in capital flows into China is mirrored in the country’s trade balance relative to GDP – as reported in Figure 8.8. The increase in capital flows24 causes the trade balance to initially decline. By 2020 when foreign capital flows level off, the increase in income payments to foreign owners of capital causes the trade balance to improve. In the other regions, by 2020, the trade balance declines as imports rise faster than exports in the developed economies or, in the case of the developing economies, as exports decline (Table 8.2). 24
Financial capital flows, which also affect the capital account, and hence the trade balance, are not included in this model. All foreign capital flows relate to foreign investment in physical capital.
224 0.03
Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina
Risk Premium Tariffs
Cum % Differences Relative to Baseline
0.02
Cross-Border Services Liberalization Autos Productivity
0.01
Services Productivity
0
-0.01
-0.02
-0.03
-0.04
-0.05 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020
Figure 8.8. Cumulative difference from baseline of the change in China’s trade balance relative to GDP. Each curve shows the change in the trade balance relative to GDP between the baseline and the policy. The policy is assumed to be the particular aspect referred to plus all previous shocks. Source: Authors’ simulations with GDyn.
It is important to reiterate that domestic and foreign shares of China’s investment are held constant, subject to constraints discussed in Chapter 2, and hence are only affected by the relative ability of domestic and foreign households to fund investment (i.e., their savings). The model does not take into account other factors that may influence the decisions of foreign and domestic investors, such as differences in their expectations, delays in reforms to China’s inefficient state enterprises or financial markets, and many of the other problems outlined in The Economist (2000) that may continue to cause foreign investors to be cautious, even after accession. Moreover capital flows relate to investment in physical goods, not the movement of financial capital. Consequently the model does not take account of the possibility that as China opens up its financial markets there will be an outflow of financial capital as Chinese investors seek to diversify their own portfolios. This would increase the extent to which foreign investors could fund new investments in China. All countries, other than China and Taiwan (China), increase their longrun foreign ownership, which is invested primarily in China and to a
Assessing the Impact of China’s WTO Accession on Investment
225
lesser extent in Taiwan (China) (Table 8.2). By 2020 China reduces its foreign investments abroad relative to the baseline, choosing instead to invest domestically (Table 8.2). In contrast to China, real GDP, investment, and wealth in the South East and South Asian economies are adversely affected by China’s accession to the WTO. This is primarily due to increased competition in the wearing apparel and textiles markets as China’s quotas are removed. Total household wealth in South East Asia, however, rises due to the large increase in foreign investment. In contrast, the Newly Industrialized Economies (NIEs) supply the growing Chinese sectors with intermediate inputs; hence real GDP, capital, and foreign ownership in the NIEs fall only marginally. The other developed and developing countries also increase their ownership of foreign assets. Note that if productivity in the services sectors were to increase to a maximum of 2.7% by 2007 (a cumulative increase in productivity of 12% over the baseline), as compared to 1% in the simulation outlined earlier, foreign investment in China would more than triple as a result of China’s accession. Real GDP in China would increase by 35.7% as compared to the current increase of 22.5%.
3.3 Welfare For the reasons set out in Chapter 6, we have opted for a relatively simple, straightforward approach to welfare analysis in which a single comparative static simulation is performed in the year 2020 to determine the difference in static welfare – at that point in time – with and without China’s accession. The final column of Table 8.2 reports the percentage change in the representative regional household’s static utility in 2020 owing to China’s WTO accession. This change is largest for China (16%), followed by Taiwan (China) with a 3.1% increase in 2020 welfare. Other changes are less than 1%. With the exception of South and South East Asia, all regions gain from China’s WTO accession. The first column in Table 8.3 reports the equivalent variation (EV) associated with the changes in regional household utility, and the remainder of that table uses the Huff/Hertel decomposition technique to explain the sources of welfare changes in each region. In the case of China, we see that US$57 billion of the gain in 2020 is due to improved allocative efficiency: Lower tariffs improve the allocation of resources between domestic production and imports. The second column of Table 8.3 shows that China’s terms of trade deteriorate as it exports more to pay for its increased imports. In addition, the abolition of the Multi-Fiber Agreement (MFA) quotas on
226
Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina Table 8.3. Decomposition of welfare (2020, $US million) Total Allocative Terms of welfare efficiency trade
China Taiwan (China) North America Western Europe Japan Other NICs South East Asia South Asia Latin America Africa and Middle East ROW Total
Capital
Foreign Tech ownership changea
Otherb
103,433 7,763 5,802 −122 1,074 1,527 −1,032 −2,570 1,169 528
57,625 1,275 −454 −3,970 −5,117 −204 −88 −2,110 281 −730
−9,687 2,457 8,787 2,971 836 1,318 574 −554 1,780 1,888
40,408 5,912 −12,794 −9,829 −9,464 −135 −3,729 −908 −3,554 −2,387
−13,255 −1,886 10,404 9,432 16,743 567 2,349 758 1,860 1,045
30,559 609 0 0 0 0 0 0 0 0
−2,218 −604 −141 1,275 −1,925 −18 −139 243 802 712
6,078 123,649
2,807 49,315
4,218 14,588
−1,072 2,449
−868 27,150
0 31,168
992 −1,022
a
Due to the productivity shocks in the automobile and services sectors. “Other” includes changes in welfare caused by changes in the price of saving relative to the price of capital goods and by changes in marginal utility. Source: Authors’ simulations with GDyn.
b
textiles and apparel means that the associated exporter rents get dissipated in lower export prices. Because WTO accession stimulates investment in China, the capital stock is higher, which contributes US$40.4 billion to welfare (Table 8.3). However, a substantial portion of this new investment is foreign owned, and so the associated income payments go overseas. This flow translates into an offsetting US$13 billion deduction from China’s welfare in the foreign ownership column of Table 8.3. The other major contributor to welfare in 2020 is improved productivity. This accounts for a US$31 billion increment to static 2020 welfare. The OECD countries tend to experience a decline in allocative efficiency (Table 8.3). This is a result of the change in production of taxed or subsidized sectors, including automobiles (Japan), trade and transport (USA), and/or agriculture (EU25 ) sectors, resulting from the production changes in China. The impact in other regions is mixed. Most striking is the US$2 billion decline in 2020 static efficiency in South Asia, where the volume of trade declines under China’s accession. Apart from China and South Asia, the 25
The fall in allocative efficiency in the EU, resulting from productivity gains in China’s automobile sector, is driven by an increase in agricultural production, which is heavily subsidized. Although the production of automobiles falls as expected, the rise in imports of automobiles, which are subject to high tariff rates, causes welfare to rise.
Assessing the Impact of China’s WTO Accession on Investment
227
Table 8.4. Welfare decomposition by element of WTO accession (2020 $US millions)
Total China Taiwan (China) North America Western Europe Japan Other NICs South East Asia South Asia Latin America Africa and Middle East ROW Total
Risk Tariffs Services trade Auto Services premia and quotas liberalization productivity productivity
103,433 7,763 5,802 −122 1,074 1,527 −1,032 −2,570 1,169 528
−132 7 18 137 135 3 7 0 −3 −4
13,848 6,418 5,395 2,464 1,863 797 −2,320 −1,504 −188 −217
522 521 246 199 66 89 2 0 51 67
34,746 141 −2,002 −2,518 −2,376 317 2,001 −816 625 888
54,448 675 2,144 −405 1,386 321 −722 −250 685 −207
6,078 123,649
−4 164
210 26,767
53 1,817
3,922 34,928
1,897 59,972
Source: Authors’ simulations with GDyn.
terms of trade improve for all other regions.26 The capital and foreign ownership columns are largely offsetting in the non-China regions, because the lower domestic capital stock is offset by increased earnings on foreign investments. Thus far, we have discussed the economic mechanisms underlying the welfare impacts. Yet which elements of the accession package contribute most to these welfare gains? In Table 8.4, the welfare (EV) results are decomposed according to the impact on welfare of each element of the accession package. Overall, the changes in China’s welfare are primarily (85%) due to productivity improvements in the automobile and services sectors. The reduction in tariffs by China and the expansion of U.S. and EU quotas on textiles and wearing apparel account for 13% of the total welfare gains or US$13.8 billion.27 Hence 52% of the gains to China are from productivity gains in services, an effect that is missed by most of the studies on China accession in the literature.28 26
27 28
The increase is particularly strong for North America and Europe, where the demand for exports rises and the elimination of textile and apparel quotas on imports results in lower prices for those products. This finding is consistent with other studies that take into account the duty drawbacks currently in place in China (Ianchovichina and Martin 2001). China’s accession may also result in productivity gains in other manufacturing sectors. In textiles and footwear, Claro (2001) states that the adoption rates for foreign technology are between 30 and 62% for collective enterprises. Mai et al. (2003) also estimate significant productivity gains in the manufacturing sector. These are not captured here.
228
Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina
In contrast, 52% of the gains to the developed countries are the result of the elimination of tariffs and quotas, as compared to only 13% for China. However, the improvements in productivity in China still account for a substantial proportion (almost 40%) of the welfare gains of the other regions. Hence for both China and the other regions these productivity gains are an important aspect of China’s accession agreement.
4. Conclusion Foreign investment was a focal point in China’s negotiations for accession to the WTO. China has aggressively pursued foreign investment over the 1990s – albeit with significant limitations on foreign entry in certain sectors, particularly services. Accordingly, increasing access for foreign investors in the services sectors has been of paramount importance to the U.S. and European negotiators in their dealings with China. In addition, WTO accession will lead to a reduction in tariffs, elimination of quotas on exports of textiles and apparel, and a reduction in the barriers to cross-border supply of services. This chapter focuses on these elements of China’s WTO accession, comparing two alternative time paths of investment and ownership in the Chinese economy over the coming two decades: a no-accession baseline (counterfactual) scenario, and projections under WTO accession. We used a modified version of the dynamic global applied general equilibrium model, GDyn, which explicitly models capital accumulation and foreign ownership, as well as taking account of China’s existing duty exemption system. Our results show that China’s accession to the WTO would boost the Chinese economy and raise rates of return. The resulting increase in capital stocks would be financed by increased domestic investment and foreign investment from industrialized and newly industrialized economies in East Asia, North America, and Europe. Central to the increase in China’s capital stock are the anticipated productivity gains in the automobile and services sectors. In the automobile sector, this improvement is expected to be fueled by a rationalization of production, as the number of production facilities falls, but the length of run for any given facility increases (Francois and Spinanger 2002). In the case of services, the productivity gains are expected to come from the opening of the Chinese market to foreign investment. In particular, telecommunications, banking, insurance, logistics, and transportation are expected to experience substantial productivity gains as foreign investors induce changes in the organization and operation of these activities. Rather than model the elimination of these barriers to commercial presence directly, we adopt a
Assessing the Impact of China’s WTO Accession on Investment
229
“dual” approach, whereby we postulate an associated gain in productivity as part of our accession scenario, thereby observing how much foreign investment such gains are likely to sustain. Although this part of our analysis is clearly speculative, these productivity shocks are well within the plausible range identified by Mai et al. (2003). Because the services component of the accession package is a critical factor in determining the overall impact on GDP and welfare – both in China and in its trade and investment partner economies – further research on the likely productivity gains after accession needs to be undertaken. The overall size of the potential gains to China is quite substantial. We estimate that GDP could be nearly 22.5% higher in 2020 as a result of WTO accession. The static welfare gains are lower (16% in 2020) because a substantial share of the additional investment comes from overseas. Nevertheless, these impacts are quite large and are far larger than those predicted by earlier studies, which have ignored the impact of accession on productivity in the services sector of China, as well as abstracting from capital accumulation and foreign investment. Future research should be directed toward reducing the uncertainty associated with the impact of accession on productivity in services.
Appendix: Aggregation of GTAP Data Base Table 8.A1. Sector aggregation 50 sectors
GTAP code
This study
1 2 3 4 5 6 7 8 9 10 11 12 13 14
pdr wht gro vf osd cb pfb ocr ctl oap rmk wol for fsh
crops crops crops crops crops crops crops crops lstk lstk lstk lstk extrprds extrprds
Paddy rice Wheat Cereal grains nec Vegetables, fruit, nuts Oil seeds Sugar cane, sugar beet Plant-based fibers Crops nec Bovine cattle, sheep and goats Animal products nec Raw milk Wool silk-worm cocoons Forestry Fishing
(continued )
230
Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina Table 8.A1 (continued) 50 sectors
GTAP code
This study
15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
col oil gas omn cmt omt vol mil pcr sgr ofd bt tex wap lea lum ppp pc crp nmm is nfm fmp mvh otn ele ome omf ely gdt wtr cns tt osp osg dwe
extrprds extrprds extrprds extrprds lstk lstk foodbev lstk foodbev foodbev foodbev foodbev textiles wearapp othmnfcs extrprds metlchem extrprds metlchem metlchem metlchem metlchem metlchem autos othmnfcs electronics othmnfcs othmnfcs houseutils houseutils houseutils houseutils tradetrans othsvces othsvces houseutils
Coal Oil Gas Minerals nec Bovine cattle, sheep, and goat Meat products nec Vegetable oils and fats Dairy products Processed rice Sugar Food products nec Beverages and tobacco products Textiles Wearing apparel Leather products Wood products Paper products, publishing Petroleum, coal products Chemical, rubber, plastic products Mineral products nec Ferrous metals Metals nec Metal products Motor vehicles and parts Transport equipment nec Electronic equipment Machinery and equipment nec Manufactures nec Electricity Gas manufacture, distribution Water Construction Trade, transport Financial, business, recreatio Public admin and defense, educ Dwellings
Assessing the Impact of China’s WTO Accession on Investment Table 8.A2. Regional aggregation 45 Regions
GTAP code
This study
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
aus nzl jpn kor idn mys phl sgp tha vnm chn hkg twn ind lka ras can usa mex cam ven col rap arg bra chl ury rsm gbr deu dnk swe fin reu eft cea fsu tur rme mar rnf saf rsa rss row
ROW ROW Japan OthNICs SEA SEA SEA OthNICs SEA SEA China OthNICs Taiwan SoAsia SoAsia SoAsia NAmerica NAmerica NAmerica LatinAM LatinAM LatinAM LatinAM LatinAM LatinAM LatinAM LatinAM LatinAM WEurope WEurope WEurope WEurope WEurope WEurope WEurope ROW ROW AfrMidE AfrMidE AfrMidE AfrMidE AfrMidE AfrMidE AfrMidE ROW
Australia New Zealand Japan Korea Indonesia Malaysia Philippines Singapore Thailand Viet Nam China Hong Kong Taiwan India Sri Lanka Rest of South Asia Canada United States of America Mexico Central America and Caribbean Venezuela Colombia Rest of the Andean Pact Argentina Brazil Chile Uruguay Rest of South America United Kingdom Germany Denmark Sweden Finland Rest of European Union EFTA Central European Associates Former Soviet Union Turkey Rest of Middle East Morocco Rest of North Africa South African Customs Union Rest of southern Africa Rest of sub-Saharan Africa Rest of World
231
232
Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina References
Anderson, K., J. Francois, T. W. Hertel, B., Hoekman, and W. Martin. 2000. Potential Gains from Trade Reform in the New Millennium. Paper presented at the Third Annual Conference on Global Economic Analysis, June 27–30, Monash University, Australia. Ahuja, V. and D. Filmer. 1995, July. Educational Attainment in Developing Countries; New Estimates and projections Disaggregated by Gender. World Bank Policy Research Working Paper 1489. Washington, DC: World Bank. Baldwin, R. E., J. F. Francois, and R. Portes. 1997. “The Costs and Benefits of Eastern Enlargement: The Impact on the EU and Central Europe.” Economic Policy (April), 127–76. China Council. 2000. China Business Guide. Beijing: China Council for the Promotion of International Trade / China Chamber of International Commerce. China’s Statistical Yearbook. 2002. Beijing: China Statistics Press. Claro, S. 2001. Tariff and FDI Liberalization: What to Expect from China’s Entry into the WTO? Paper presented at the Eighth Annual Conference on Empirical Investigations in International Trade, November 9–11, Purdue University, West Lafayette, IN. CPB. 1999, December. WorldScan: The Core Version. The Hague: CPB Netherlands Bureau for Economic Policy Analysis. The Economist. 2000. “A Survey of China: Now Comes the Hard Part,” April 8–14. Dixon P. B. and M. T. Rimmer, 2002. Dynamic General Equilibrium Modelling for Forecasting and Policy: A Practical Guide and Documentation of MONASH. Amsterdam: North-Holland Publishing Company. Francois, J. and D. Spinanger. 2002. Regulated Efficiency, WTO Accession and the Motor Vehicle Sector in China. Paper prepared for the conference WTO Accession, Policy Reform and Poverty Reduction, June 28–29, Beijing. Francois, J. and A. Strutt. 1999, June. “Post Uruguay Round Tariff Vectors for GTAP v.4.” Unpublished memorandum. Garbaccio, R. F. 1995. US-China Trade Relations: 1972–95. Paper prepared for the 4th International Symposium on Societies and Economy in East Asia, Oct. 2–3, Osaka, Japan. Gilbert, J. and T. Wahl. 2001. “Applied General Equilibrium Assessments of Trade Liberalization in China.” World Economy 25(5): 697–731. Harrison, G. W., T. F. Rutherford, and D. G. Tarr. 1996. “Quantifying the Uruguay Round.” In W. Martin and A. Winters (eds.), The Uruguay Round and the Developing Economies (pp. 216–52). Cambridge: Cambridge University Press, Hertel, T. W. (ed.). 1997. Global Trade Analysis: Modeling and Applications. Cambridge: Cambridge University Press. Hertel, T. W. and K. Huff. 2001. Decomposing Welfare Changes in GTAP. GTAP Technical Paper No. 5 (rev. version). West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Hertel, T. W., W. Martin, K. Yanagishima, and B. Dimaranan. 1996. “Liberalizing Manufactures Trade in a Changing World Economy.” In W. Martin and A. Winters (eds.), The Uruguay Round and the Developing Economies (pp. 183–215). Cambridge: Cambridge University Press. Huang, Y. 2002. Selling China: Foreign Direct Investment during the Reform Era. Cambridge: Cambridge University Press.
Assessing the Impact of China’s WTO Accession on Investment
233
Ianchovichina, E. I. 2003. GTAP-DD: A Model for Analyzing Trade Reforms in the Presence of Duty Drawbacks/ GTAP Technical Paper No. 21. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Ianchovichina, E. I. and W. Martin. 2001. “Trade Liberalization in China’s Accession to WTO.” Journal of Economic Integration 16(4), 421–44. Ianchovichina, E. I. and W. Martin. 2003. Economic Impacts of China’s Accession to the WTO. Policy Research Working Paper 3053. Washington, DC: World Bank. Ianchovichina, E. I. and T. L. Walmsley. 2003. Impact of China’s WTO Accession on East Asia. Policy Research Working Paper 3109. Washington, DC: World Bank. Lejour, A. 2000. China and the WTO: The Impact on China and the World Economy. Paper presented at the Third Annual Conference on Global Trade Analysis, June 27–30, Melbourne, Australia. Maddison, A. 1998. Chinese Economic Performance in the Long Run. Paris: Development Centre Studies, OECD. Mai, Y., M. Horridge, and F. Perkins. 2003. Estimating the Effects of China’s Accession to the World Trade Organisation. CoPS General Paper No. G-137. Paper presented at the Sixth Annual Conference on Global Trade Analysis, June 12–14, Scheveningen, The Hague,. Martin, W., B. Dimaranan, T. Hertel, and E. Ianchovichina. 2000. Trade Policy, Structural Change and China’s Trade Growth. Working Paper No. 64. Stanford: Institute for Economic Policy Research. Mattoo, A. 2001. China’s Accession to the WTO: The Services Dimension. Paper presented at the Workshop on WTO Accession, Policy Reform, and Poverty Reduction, October, Beijing. McDougall, R. A., A. Elbehri, and T. P. Troung (eds.). 1998. Global Trade, Assistance and Protection: The GTAP 4 Data Base. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. McKibbon, W. J., and K. K. Tang. 2000. “Trade and Financial Reform in China: Impacts on the World Economy.” World Economy 23(8), 979–1003. Pangestu, M. and D. Mrongowius. 2001. Facing the Challenge of WTO Accession: Development and Reform of China’s Telecommunications Services Sector. Paper presented at the Workshop on WTO Accession, Policy Reform, and Poverty Reduction, October, Beijing. Spinanger, D. 1999. “Textiles beyond the MFA Phase-Out.” World Economy, 22(4), 455–76. Walmsley, T. L. and T. W. Hertel. 2001. “China’s Accession to the WTO: Timing is Everything.” World Economy 24(8), 1019–49. Wang, Z. 1997a. “China and Taiwan, China Access to the World Trade Organization: Implications for U.S. Agriculture and Trade.” Agricultural Economics, 17, 239–64. Wang, Z. 1997b. Impact of China’s WTO Accession on the Labor Intensive Exports and Implications for U.S. Agricultural Trade – A Recursive Dynamic CGE Analysis. Paper presented at the 1997 AAEA Meeting, July 28–31, Toronto. Wang, Z. 2001. The Impact of China’s WTO Accession on Patterns of World Trade. Paper prepared for the International Agricultural Trade Research Consortium, May 18–20. Wang, Z. 2002. WTO Accession, “Greater China” Free Trade Area and Economic Relations across the Taiwan Strait. Paper presented at the Fifth Conference on Global Economic Analysis, June 5–7, Taipei.
234
Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina
Wenping, L. and C. Findlay. 2001. Logistics in China: Accession to the WTO and Its Implications. Paper presented at the Workshop on WTO Accession, Policy Reform, and Poverty Reduction, October, Beijing. Yang, Y. 1996. “China’s WTO Membership: What’s at Stake?” World Economy 19(6), 661–82.
NINE
Dynamic Effects of the “New-Age” Free Trade Agreement between Japan and Singapore Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura
1. Introduction In the 1990s, there was a flood of regional trade agreements. By 2001, more than 130 such agreements were in place (WTO 2000). The European Union, the North America Free Trade Agreement (NAFTA), and MERCOSUR have been particularly effective at promoting intraregional trade. This success has led other countries to explore options for such regional agreements, and in December 1999, Japan and Singapore established a Joint Study Group to examine the feasibility and desirability of establishing an FTA. After a favorable report from the Study Group, negotiations commenced in early 2001 (Joint Study Group 2000a), and the Japan-Singapore FTA came into effect in November 2002. The main elements of the Japan-Singapore FTA involve bilateral liberalization and facilitation of trade through reduction of tariff and nontariff barriers, as well as the mutual recognition of national standards, streamlining of customs procedures, facilitation of increased services trade, and the establishment of an exemplary framework for foreign investment. This “new-age” FTA also envisioned increased collaboration on intellectual property, education and training, media and broadcasting, and tourism. This trade agreement is particularly significant, because it is viewed by many as providing a possible template for future FTAs in the region, for example the FTA between Japan and Korea (KIEP 2000). Japan already trades quite intensively with Singapore and Korea. Based on the Brown-Kojima-Drysdale export intensity index, Japan exported about Original version published in Journal of Economic Integration, December, 16(4), 446–8, 2001. The application related to this Chapter is Ch9 gdyn r19 95.zip and is available for download on the Web site at https://www.gtap.agecon.purdue.edu/models/Dynamic/ applications.asp.
235
236
Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura
7 6
4 3 2 1 1997
1995
1993
1991
1989
1987
1985
1983
1981
1979
1977
1975
1973
1971
1969
1967
0 1965
Index
5
Year Korea
Singapore
ROW
Figure 9.1. Japanese export bias: Total merchandise trade. Source: Authors’ computations.
twice as much to these countries as one would expect, solely on the basis of their world import shares in 1998. However, there is some indication that the relative attraction of trading with Korea and Singapore has been diminishing. Figure 9.1 reports Japan’s total merchandise trade export bias to Korea, Singapore, and the Rest of World (ROW). Drysdale and Garnaut (1982) uses this measure to capture the determinants of export shares after taking into account the size, openness, and product composition of the importing market. Anderson and Norheim (1993) argue that this bias component offers a proxy for the kind of transactions costs that FTAs are intended to lower. The higher this bias, the more attractive the export destination relative to alternative markets. As can be seen from Figure 9.1, the relative attractiveness of exporting from Japan to Korea and Singapore has been cut in half over the past 30 years. Figure 9.2 reports the export bias from Korea and Singapore to Japan, which also have been falling. Thus, although trading costs between Japan and these partners may have fallen significantly over the past 30 years, there is some evidence that trading costs with other partners have fallen more rapidly. In light of this observation, it is perhaps not surprising that these three countries have initiated discussions aimed at lowering trading costs among their respective economies. The goal of this chapter is to provide a quantitative assessment of the dynamic effects of one of these new-age FTAs – namely, that between Japan and Singapore. The chapter is organized as follows. In the next section, we outline the key elements of the Japan-Singapore FTA and discuss our approach to
Dynamic Effects of the “New-Age” Free Trade Agreement
237
7 6
4 3 2 1 1997
1995
1993
1991
1989
1987
1985
1983
1981
1979
1977
1975
1973
1971
1969
1967
0 1965
Index
5
year Korea
Singapore
Figure 9.2. Korea and Singapore export biases to Japan. Source: Authors’ computations.
quantifying them. Given the relatively low level of industrial tariffs on most trade between these two countries, we devote considerable effort to quantifying the nontariff elements of this agreement. In Section 3, we outline the dynamic modeling approach taken in this chapter. It is aimed explicitly at capturing the impact of these new-age FTAs not only on trade but also on international investment flows. To analyze the potential impact of these FTAs, it is important to have a view of the evolution of regional trade and growth in the absence of the agreements. This baseline is established in Section 4 of the chapter. Section 5 reports the results and analysis based on a comparison of the baseline with the counterfactual FTA simulations and is followed by the conclusions.
2. Quantifying the New-Age Agreement 2.1 Trade and Tariffs As with other such regional trade agreements, the FTA between Japan and Singapore includes the bilateral elimination of tariffs (Joint Study Group 2000a, 2000b). Table 9.1 reports estimated average bilateral tariffs levied by Singapore and Japan on one another’s exports. The tariff estimates for Singapore are based on applied rates for 1999, as reported in the WTO Singapore Trade Policy Review (2000). Applied tariffs in Singapore are now zero for all goods, except for alcoholic beverages (shown as other food products in Table 9.1). This reflects some liberalization from the 1995 applied rates in the GTAP 4 Data Base, and these tariff cuts in Singapore are
238
Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura Table 9.1. Bilateral tariffs and composition of imports Imports from Japan
Commodity Rice Other grains Other crops Meat Other food Fish Textiles and wearing apparel Leather Extraction Chemicals and minerals Other manufactures Automobiles Machinery and equipment Utilities Construction Trade and transport Business and financial services Total (millions $US)
Imports from Singapore
Trade share
Tariff rate
Trade share
Tariff rate
0.0 0.0 0.1 0.0 0.6 0.1 0.6 0.0 7.4 8.2 5.9 4.5 59.5 0.1 0.0 1.3 11.7 $33,731
0.0 0.0 0.0 0.0 1.7 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
n.t. n.t. 0.5 0.1 1.8 0.1 0.2 0.1 1.0 6.6 2.7 0.02 32.8 0.1 0.0 48.9 5.1 $18,066
n.t. n.t. 1.8 47.7 21.3 1.3 6.6 6.1 0.2 1.2 0.1 0.0 0.1 0.0 0.0 0.0 0.0
Note: n.t. = no trade.
implemented as part of the baseline described in Section 4. Based on the nonagricultural tariffs in Table 9.1, implementation of the FTA will have no direct impact on Singapore’s imports of merchandise commodities from Japan – hence the calls for a new-age FTA. The tariff estimates for Japan in Table 9.1 reflect the lower of 1995 applied rates, as obtained from the GTAP 4 Data Base, and WTO bindings under the Uruguay Round (UR). In cases where the bindings are below 1995 tariffs, we reduce them to the level of the post-UR bindings as part of the baseline simulation. Note from the trade share entries in Table 9.1 that Japan does not import any grains from Singapore. Bilateral imports of meats and other food products are modest, but face a very high average tariff. It is clear from this table why food and agriculture represent a very sensitive part of this agreement. Given the very high tariffs facing these products imported from other destinations, the incentive for transshipment through Singapore is likely to be substantial under an FTA. This raises the prospect of significant enforcement costs associated with the rules of origin for this FTA. For this reason, it is quite likely that agriculture will be left out of the final
Dynamic Effects of the “New-Age” Free Trade Agreement
239
FTA agreement – or perhaps it will be implemented in a delayed fashion. Of course, textiles, apparel products, and leather goods also face non-negligible tariffs so that substantial expansion of imports from Singapore is expected under the FTA. In our study we assume that implementation of the FTA is undertaken in 2005. At the time this work was undertaken, it was unclear how fast Japan and Singapore would move on this agreement. By placing it after completion of the Uruguay Round we also simplified our simulation design. For this reason, we focus on the projected 2005 trade shares in Table 9.1 in evaluating the potential impact of this FTA (see Section 4 for details underlying the baseline projections). Based on the figures in Table 9.1, it is clear that this trade relationship is highly concentrated, with the bulk of Japanese exports to Singapore involving machinery and equipment. This is followed in importance by business and financial services, chemical and mineral products, extractive products, and other manufactures. Singapore’s exports to Japan are concentrated in the services sector, followed by machinery and equipment and petroleum/chemical and mineral products. Clearly this trading relationship involves a great deal of intra-industry trade that should receive a substantial boost from any reduction in nontariff trade costs.
2.2 Customs Automatization This brings us to a second aspect of the Japan-Singapore FTA that lends itself to quantification, namely the reduction of customs costs for bilateral trade between these two partners. In building the case for efforts to streamline customs procedures, the Joint Study Group (2000b) cites UNCTAD research indicating that customs paperwork and procedures costs add up to 7% of the global value of trade. This is likely a considerable overstatement of these costs in the case of Japan-Singapore trade. Nevertheless, in an era of increasing regional integration and vertical specialization in production, small trade costs can have a significant impact on intra-industry trade. Furthermore, any costs above 1–2% will represent a more substantial barrier to trade than industrial tariffs. The Joint Study Group (2000b) has focused on a proposal to reduce customs clearance costs by implementing an electronic trade document exchange system (ETDS) that will increase the speed of customs clearance, reduce the cost of dispatching information and documents, and ensure security of associated documents. Singapore currently has such a system in place, so the emphasis is on extending this technology to Japan’s customs procedures.
240
Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura
At the heart of the ETDS proposal resides commercially operated electronic document exchange servers that will facilitate the exchange of customs documents among importers, exporters, and the Japanese customs authorities. Introduction of an electronically harmonized custom system will reduce the time and cost spent on customs paperwork, processing, and shipments. Customs automatization will also improve efficiency in shipments of products by eliminating the time spent waiting for customs clearance at ports. Our estimates of the savings in time and direct costs due to customs automatization are based on research conducted in conjunction with the Ministry of Economy, Trade, and Industry (METI) and the Mitsubishi Research Institute (MRI) in Japan. Through a careful study of the direct costs associated with current customs procedures, as well as the costs of connecting to the ETDS system, the MRI estimates that introduction of customs automatization in Japan would lower the effective merchandise prices for all trading partners by 0.201% for exports and by 0.203% for imports. Additional reductions in direct costs arise when a trading partner of Japan also implements the electronic custom system to synchronize the customs clearance. For the case of the Japan-Singapore FTA, the effect of linking the two systems is expected to generate additional reductions in effective prices amounting to 0.065% in Japanese imports from Singapore and 0.013% in Singaporean imports from Japan. It should be noted that these cost savings refer solely to reduced paperwork, storage, and transit expenses. However, in addition to the direct cost savings of ETDS, there are indirect savings associated with the elimination of customs-related delays in merchandise flows between these two countries. Hummels (2000) emphasizes that such time savings can have a profound effect on international trade by reducing both “spoilage” and inventory holding costs. He argues that spoilage can occur for many types of goods. The most obvious might be agricultural and horticultural products that physically deteriorate with the passage of time. However, products with information content (newspapers), as well as highly seasonable (fashion) goods, may also experience spoilage. Hummels points out that inventory costs include not only the capital costs of the goods while they are in transit but also the need to hold larger inventories to accommodate variation in arrival time. The latter has become increasingly important because of increased use of “just in time” production techniques. To estimate the value of time savings in international trade, Hummels uses a detailed dataset that he has assembled that includes information on modal choice (air vs. sea), modal prices (shipping rates), and modal
Dynamic Effects of the “New-Age” Free Trade Agreement
241
Table 9.2. Time savings per day and price reductions due to customs automatization
(percent ad valorem) Price reduction (%)
Rice Other grains Other crops Meat Other food Fish Textiles and wearing apparel Leather Extraction Chemicals and minerals Other manufactures Automobiles Machinery and equipment
No linking effect
With linking effect
Opportunity cost of a day in trade (% ad valorem)
Exports (a)
Imports (b)
Exports (c)
Imports (d)
0 0 0 0 0 0.14 0.14
0.201 0.201 0.201 0.201 0.201 0.431 0.435
0.203 0.203 0.203 0.203 0.203 0.406 0.409
0.214 0.214 0.214 0.214 0.214 0.619 0.627
0.268 0.268 0.268 0.268 0.268 0.741 0.749
0.13 0.30 0.94 0.29 0.16 0.51
0.422 0.707 1.795 0.686 0.465 1.072
0.398 0.649 1.609 0.631 0.436 0.971
0.604 1.107 3.027 1.069 0.679 1.750
0.723 1.309 3.549 1.266 0.811 2.061
Source: Authors’ computation based on Hummels (2000) and data from MRI.
shipping times at the 10-digit, HS level for U.S. imports. It includes approximately one million observations per year over the entire 1974–98 period. Hummels (2000) estimates a discrete-choice model, wherein the probability of choosing air over sea transport depends on relative freight rates and the associated time savings. He finds that the average value of firms’ willingness to pay for one day saved in trade is estimated to be 0.5% ad valorem (i.e., one-half percent of the value of the good itself). However, this value of time savings varies widely by product category, with the lowest values for bulk commodities and the highest values for intermediate goods. The second column of Table 9.2 (value of one day saved) reports the percentage ad valorem value of a day saved in trade at the level of commodity aggregation used in the present study. The smallest value is 0.13%/day for leather, whereas the value of a one-day reduction in transit reaches nearly 1% (0.94%) per day for petrochemical and mineral products. This value is also quite high for machinery and equipment (0.51%/day). Hummels’ estimates for agricultural products are not significantly different from zero and are therefore omitted.
242
Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura Table 9.3. Reductions in bilateral prices due to customs
automatization Imports
Exports
Japan Singapore ROW
Japan
Singapore
ROW
n.a. (d) (b)
(c) n.a. 0
(a) 0 0
Note: See Table 9.2 for values of (a), (b), (c), and (d), by sector.
The MRI estimates that the amount of time that would be saved by customs automatization would be 1.7 days for exports and 1.5 days for imports. These translate into effective price reductions of 0.85% and 0.75%, respectively, based on the average valuation of 0.5% ad valorem per day. Additional time savings are likely if the countries at both ends of the transaction adopt the same ETDS system, as is the case with Japan and Singapore. The additional saving on lead-time is estimated to be an additional 1.3 days for exports and 2 days for imports, or 0.65% and 1% reductions in the average effective prices of exports and imports, respectively. By applying these time savings associated with ETDS to Hummels’ estimates of the value of time savings, by commodity, we obtain the price reductions associated with customs automatization shown in columns (a) to (d) in Table 9.2. These estimated price reductions vary depending on whether the linking effect is present. Columns (a) and (b) apply to trade between Japan and ROW, whereas columns (c) and (d) apply to trade between Japan and Singapore. The full set of bilateral shocks associated with customs automatization is summarized in Table 9.3. Note from this table that Singapore’s trade with ROW is unaffected by the implementation of ETDS in Japan.
2.3 E-Commerce Another important element of the proposed FTA is the section aimed at improving security and harmonizing standards governing Business to Business and Business to Consumer e-commerce between Japan and Singapore. The goal of this part of the agreement is to make e-commerce between the two countries as safe and acceptable to customers as domestic e-commerce is presently. Accordingly, we have taken estimates of the extent of B-to-B e-commerce penetration in the domestic Japanese market (column two of Table 9.4), along with the MRI estimated reduction in
Dynamic Effects of the “New-Age” Free Trade Agreement
243
Table 9.4. Reduction in price due to e-commerce
Rice Other grains Other crops Meat Other food Fish Textiles and wearing apparel Leather Extraction Chemicals and minerals Other manufactures Automobiles Machinery and equipment Utilities Construction Trade and transport Business and financial services
E-commerce penetration (%)
Potential reduction in average price (%)
0.95 0.95 0.95 0.95 0.95 0.95 2.80 2.80 0.82 0.20 0.80 14.20 6.59 0.10 0.04 0.20 0.20
0.09 0.09 0.09 0.09 0.09 0.09 0.27 0.27 0.08 0.02 0.08 1.39 0.65 0.01 0.00 0.02 0.02
Note: Price reduction is computed as follows: e-commerce penetration multiplied by reduction in margin as % of final price. Source: Authors’ computation based on estimates from MRI.
wholesale-retail margins (projected to be reduced from 19.6% to 4.9% of prices in the presence of e-commerce). We then computed the potential reduction in average effective price across all transactions that might be attained on products traded between Singapore and Japan (column three of Table 9.4). This reduction varies by sector, depending on the degree of ecommerce penetration. It is highest for automobiles and parts, which show a 1.39% reduction in average price.
2.4 Services Trade The previously discussed aspects of the FTA – tariff reductions, customs automatization, and e-commerce – largely affect the cost of merchandise trade between Japan and Singapore. However, the FTA also proposes liberalization of services trade. Here, quantification is quite difficult, because data on services trade and potential barriers are rather scarce. For purposes of this study, we follow the recent work of Joseph Francois (1999), who has estimated two gravity models of trade – one for business services and one for
244
Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura Table 9.5. Estimated tariff equivalents for services (percent ad valorem) Regions
Business services
Construction
9.0 9.4 7.2 20.6 19.5 7.1 8.1 6.9 5.2 13.5 36.8 5.6 4.4 19.1 15.4 0.4 19.2
9.9 18.5 24.5 29.9 41.1 28.5 28.8 9.6 17.8 61.7 57.5 26.3 9.6 52.1 41.9 11.1 29.1
North America Western Europe Australia and New Zealand Japan China Taiwan Korea Indonesia Other South East Asia India Brazil Other Latin America Other Middle East and North Africa CEECs and Russia South Africa Other Sub-Saharan Africa Rest of World Source: Francois (1999) and authors’ estimation.
construction services – using bilateral services export data from the United States (Bureau of Economic Analysis 1999). Francois’ gravity models permit him to predict what trade would be in the absence of barriers to trade – using Hong Kong and Singapore as “free-trade” benchmarks. By positing an import demand function he is then able to obtain tariff equivalents for the unobserved trade barriers for services trade in business and finance and construction. Estimation results are reported in Table 9.5. As can be seen, Japan’s estimated tariff equivalent of 20.6% is relatively high for business services. The tariff equivalent for construction imports into Japan is even higher (29.9%), although other countries have much more restrictive trade barriers in this sector. We seek to quantify the services trade liberalization portion of the JapanSingapore FTA by eliminating – on a bilateral basis – these services trade barriers. Because all of the barriers in Table 9.5 are measured relative to Singapore and Hong Kong, this liberalization once again does not affect Singapore. In contrast, it lowers the effective price of business and financial services exported from Singapore to Japan by 20.6%, and for constructions services the price drop is 29.9%. Of course it should be noted that most of the biggest barriers to trade in services arise in the trade and transport
Dynamic Effects of the “New-Age” Free Trade Agreement
245
sector (Hoekman 1995), and we have ignored it altogether because of a lack of protection estimates. We also ignore prospective liberalization of investment and the movement of persons providing services – which are major vehicles for delivering services to foreign markets. In short, this quantification is quite limited and should be seen as providing a lower bound on potential impacts of the services component of the FTA between Japan and Singapore.
3. Analytical Framework It has now become standard practice to use AGE models to analyze the likely impact of free trade agreements (Francois and Shiells 1994). Because of the economy-wide nature of FTAs, it hardly makes sense to examine any given sector in partial equilibrium isolation; the interplay between sectors becomes a key aspect of such regional trade agreements. Their explicit incorporation of bilateral trade flows also makes AGE models well suited to analyzing the consequences of preferential trade arrangements. Finally, their neoclassical theoretical foundations facilitate AGE models’ analysis of the tradeoff between greater openness, on the one hand, and potential trade diversion, on the other. Accordingly, we use the AGE approach in this study. To capture the dynamic effects of the new-age FTA between Japan and Singapore, as well as the potential impacts on international investment flows and wealth, we use the GDyn model outlined in Chapter 2. Taking account of foreign capital ownership is especially important in East Asia, where international investment has boomed in recent decades with the outsourcing of production from Japan and other high-wage economies. Explicit modeling of the ownership of regional investment in Japan and other Asian economies enables the determination of the accumulation of Japanese wealth by foreigners. In addition it can also track Japan’s ownership of domestic and foreign assets. Income accruing from the ownership of these foreign and domestic assets can then be appropriately incorporated into total regional income and hence the computation of welfare for Japan, Singapore, and the rest of the world. In this chapter we use a 17-region, 17-commodity aggregation of the GTAP 4 Data Base. The regions and commodities are listed in the Appendix of this chapter.
3.1 Treatment of Unobserved Trade Costs As we saw earlier, a key feature of the proposed new-age FTA between Japan and Singapore is a series of measures intended to lower nontariff trade costs
246
Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura
between the FTA members. Yet many of these trade costs (e.g., the costs of customs clearance) are not found explicitly in the database. How can we introduce these nontariff shocks and analyze their likely impact on trade flows? The approach we have taken is to introduce the notion of an “effective price” of commodity i, imported from country r at domestic prices in destination market s: PMS∗irs . This is related to the observed price, PMSirs , as follows: PMS∗ = PMS/AMS. The technical coefficient AMS is unobserved and is equal to 1 in initial equilibrium. Changes in its value capture the impact of nontariff measures on the price of imports from a particular exporter. Thus an increase in AMSirs ensures a fall in the effective domestic price of good i exported from r to s. To ensure a balanced dataset, a compensating quantity adjustment is required, so we define the “effective quantity” of exports associated with this price: QXS∗ = QXS · AMS. Therefore, the product of observed price and quantity equals the product of effective price and quantity, and trade balance is maintained. When this theory is incorporated into the GDyn model, and the import price and demand equations are totally differentiated and placed in percentage change form (denoted by lower case variables), we obtain revised equations (9.1) and (9.2): Import Demand Equation is i · [pmsirs − amsirs − pimis ]; qxsirs = −amsirs + qimis − σm
Composite Import Price Equation is θiks · [pmsiks − amsiks ], pimis =
(9.1)
(9.2)
k
where i : elasticity of substitution among imports of i; σm qxs irs : percentage change in bilateral exports of i from r to s; qimis : percentage change in total imports of i into s; pmsirs : percentage change in price of imports of i from r in s; pimis : percentage change in average import price of i in s; ams irs : percentage change in effective price of i from r in s due to change in unobserved trade costs.
From equations (9.1) and (9.2), we can see that a shock to the new variable can be seen to have three distinct effects. First, from the import demand equation, we see that a 1% shock to amsirs will lower the effective price of imports of good i from exporter r imported into country s, thereby inducing substitution toward this exporter and away from other exporters,
Dynamic Effects of the “New-Age” Free Trade Agreement
247
i as governed by the elasticity of substitution: σm . However, there is a second effect in the same equation, which works in the opposite direction. Because the effective quantity of the good has also increased, less is required to meet the needs of the importer. Finally, from the composite import price equation, we can see that a 1% shock to amsirs will lower the average import price, thereby encouraging an expansion of imports at the expense of domestic purchases. Although the total impact on imports is uncertain in theory, given the values of the trade elasticities in GTAP, we expect a reduction in trade costs to increase both observed expenditures on imports and the share of imports from the FTA partner to which this reduction in trade costs is applied.
4. Baseline To establish the impact of the prospective Japan-Singapore FTA, we must begin by developing a baseline to show what the world economy would look like without the FTA imposed. This gives us two time paths for each variable of interest: first, a path that shows how the variable would change over time without the free trade agreement; and second, a path that shows how the variable would change with the free trade agreement. The difference between the two paths shows the effect of the free trade agreement. Typically these differences are cumulated and then plotted against time to illustrate the impact of the FTA on a given variable. The baseline scenario used in this chapter is based on the baseline developed in Chapter 5. It contains information on macroeconomic variables as well as expected policy changes over the 1995–2020 period. The macroeconomic variables in the baseline include observations or projections for real gross domestic product, gross investment, capital stocks, population, skilled and unskilled labor, and total labor. By way of illustration, Figure 9.3 shows the growth rates in real GDP between 1995 and 2007 for Japan, Singapore, Korea, and China. Higher GDP growth tends to translate into higher growth in trade – both for imports and exports, ceteris paribus. In the baseline, postcrisis growth rates are positive but quite low for Japan, relative to Korea and Singapore. China’s growth remains very strong out to 2007 under this baseline. The specification of policies in the baseline is very important for our FTA analysis. For example, as tariffs come down worldwide, under the implementation of the Uruguay Round agreement, the potential for trade diversion is reduced. This is because the remaining preference margin is smaller in the wake of lower MFN tariffs. The policies included in the
248
Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura Table 9.6. Baseline policies Imports
Exports
1995–2000
Uruguay Round tariff reductions for all regions except China and Taiwan (no agriculture) Singapore reduces all tariffs to zero (except on beverages and tobacco) Pre-WTO accession tariff reductions undertaken by China before 2000
USA and EU quotas increased on textiles and wearing apparel to all countries except China and Taiwan
2000–5
Uruguay Round tariff reductions for all regions China and Taiwan WTO accession
USA and EU quotas increased on textiles and wearing apparel to all countries including China and Taiwan
2005–20
None
None
Time 1 Japan
2 Korea
3 Singapore
7 China
Figure 9.3. Growth in real GDP, %: Baseline. Source: Authors’ computations.
2007
2006
2005
2004
2003
2002
2001
2000
1999
1998
1997
12 10 8 6 4 2 0 -2 -4 -6 -8 1996
Growth in Real GDP
baseline are those that are expected to occur within the region. They are summarized in Table 9.6. The aim here is to develop a realistic policy scenario for the free trade simulations undertaken here. Uruguay Round tariff commitments are assumed to be honored by all countries. However, because of the presence of “dirty tariffication” in agriculture (Ingco 1996), it is assumed that there will be no further effective liberalization in agriculture from measured levels of protection in 1995. China and Taiwan are assumed to join the WTO, with their accession offers
Dynamic Effects of the “New-Age” Free Trade Agreement
249
phased in over the 2000–5 period. This accession also gives them quota-free access to the North American and European textile and apparel markets by 2005. However, the liberalization of these quotas is assumed to be heavily back-loaded, with most of the liberalization occurring after 2002 (Francois and Spinanger 2001).
5. Results It is the aim of this chapter to examine the relative importance of the various components of the Japan-Singapore FTA, as well as their combined effect on international trade, investment flows, and growth in these two economies. Toward this end, four FTA simulations were undertaken. Each one adds another dimension of the FTA, thereby permitting an assessment of each part of this prospective agreement. The first simulation simply involves the removal of tariffs between these two trading partners. The next three simulations successively add further new-age features of the JapanSingapore FTA: the liberalization of direct trade in business services and construction, the implementation of improved security and common standards for e-commerce between Japan and Singapore, and, finally, modern, Web-based customs clearance procedures designed to automate this aspect of international trade in Japan.
5.1 Impacts on Singapore Figure 9.4 shows that all four of these components of the Japan-Singapore FTA lead to higher rates of return on investment in Singapore. The tariff cuts (largely in Japan) boost the demand for Singaporean products, thereby raising returns to capital in that economy. The time profile of this effect is shown by the shaded area at the bottom of the bars in Figure 9.4 (tariff only). The increase in the rate of return encourages additional investment – both domestically and by foreigners – and the additional investment eventually brings the rate of return back down to that attained in the baseline simulation. Indeed, the FTA rate of return actually falls slightly below its baseline level before rebounding after 2015. Eventually all rates of return are equalized due to perfect capital mobility. Yet this is only attained in the very long run. The reduction in barriers to Singapore’s direct exports of services to Japan has a similar effect to that of tariff cuts on the rate of return. This may be seen by considering the incremental effect shown by the second set of bars in Figure 9.4 (tariffs plus services). Judging from the gap between
250
Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura
2 1.5 1 0.5 0 -0.5 -1 2006
2007
2008
Tariff only
Tar+Ser
2009 Tar+Ser+Ecom
2010
2015
2020
Tar+Ser+Ecom+Cust
Figure 9.4. Effect of Japan-Singapore Free Trade Agreement on the rate of return in Singapore (cumulative percent differences from baseline). Source: Authors’ computations.
the rate-of-return effect under the tariffs-only simulation and that under the tariffs and services simulation, services liberalization is somewhat more important for the rate of return. Unlike the one-sided trade liberalization measures, the e-commerce and customs automatization shocks affect both the demand for Singaporean products in Japan and the cost of Japanese imports in Singapore. By lowering the cost of investment goods in Singapore, there is an added boost to the rate of return. Not only has the rental rate on capital risen – due to increased demand for Singaporean products in Japan – but the cost of investing in Singapore has also fallen. This is particularly true of customs automatization, which lowers the effective price of Japanese machinery and equipment in Singapore. As a consequence, these new-age features of the FTA contribute the majority of the change in the rate of return in Singapore. Because of the higher rates of return over the 2006–10 period, the increased investment in Singapore dominates the increase in national savings as a result of higher incomes. Therefore Singapore’s trade balance deteriorates, relative to the baseline simulation. This is shown in Figure 9.5. The deterioration reaches its peak in 2008, after which it begins to improve. This improvement reflects the fact that rates of return fall back to their baseline levels and the increase in foreign wealth invested in Singapore gives rise to larger foreign income payments – thereby requiring higher levels of exports relative to the baseline.
Dynamic Effects of the “New-Age” Free Trade Agreement
251
1,000 500 0 -500 -1,000 -1,500 -2,000 -2,500 2006
2007 Tariff
2008 Tar+Ser
2009 Tar+Ser+Ecom
2010
2015
2020
Tar+Ser+Ecom+Cust
Figure 9.5. Effect of Japan-Singapore Free Trade Agreement on Singapore’s trade balance (cumulative differences from baseline in US $ millions). Source: Authors’ computations.
Figure 9.6 summarizes the long-run impact of the Japan-Singapore FTA by reporting the cumulative difference between the FTA and baseline simulations in 2020 for a variety of macroeconomic variables of interest, including real GDP, capital stock, exports, imports, and foreign ownership. The higher rates of return due to the FTA give rise to a large increase in foreign ownership (3.7% by 2020), as well as higher capital stocks and GDP. 3 2.5 2 1.5 1 0.5 0 Real GDP
Capital Stocks
Tariff Only
Tar+Ser
Imports
Tar+Ser+Ecom
Exports
Foreign wealth located in firms
Tar+Ser+Ecom+Cust
Figure 9.6. Effect of Japan-Singapore Free Trade Agreement on Singapore’s real GDP, capital, exports and imports, and foreign ownership in Japan (cumulative percent differences from baseline).
252
Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura
1.2 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 Tariff only
Tar+Ser
Foreign wealth of households
Tar+Ser+Ecom
Tar+Ser+Ecom+Cust
Domestic wealth of households
Household wealth
Figure 9.7. Effect of Japan-Singapore Free Trade Agreement on the wealth of Singaporean households. Unlike the previous figure this shows the total (not the additional) cumulative percent difference between the base case and the policy resulting from the simulation. Thus by adding customs automatization (Tar + Ser + Ecom + Cust) benefits are positive overall and higher when compared with the e-commerce simulation (Tar + Ser + Ecom). This is shown by the larger positive numbers under Tar + Ser + Ecom + Cust than under Tar + Ser + Ecom.
Not surprisingly, both exports and imports are also higher in 2020, although their rates of increase are much smaller than for foreign investment. This result indicates the importance of this new-age FTA for foreign investment, as well as for trade. Given the importance of this FTA for investment and capital stocks, it is interesting to focus specifically on the resulting changes in the wealth of Singaporean households. These changes are highlighted in Figure 9.7. The line connecting the triangular dots in this figure illustrates the cumulative percent difference between the base case and FTA simulation for the total wealth of Singaporean households. This shows that total wealth rises due to the FTA, and all four elements of the FTA contribute to this increase. However, Singaporean wealth is divided among domestic equity and foreign equity. Not surprisingly, Singaporean households choose to invest more in their own country and less in foreign countries as the domestic rate of return rises. Table 9.8 reports the impacts of the FTA on Singapore’s trade with Japan as well as with ROW (all other countries combined). The figures in the table represent cumulative changes in 2020 trade volumes in millions of $US. Percentage changes are reported in parentheses. The biggest volume changes in Singapore’s exports to Japan are in machinery and equipment
Dynamic Effects of the “New-Age” Free Trade Agreement
253
($1,514 million), followed by business and finance services and other food products. There are large percentage changes in meat products (418%), leather, and textiles and apparel, but the initial trade flows are small for these products and so the volume change is also small. However, these large percentage changes indicate a very substantial preference margin for Singaporean goods exported to Japan. This preference margin, in turn, signals the potential for trade being routed from third countries through Singapore to Japan; hence the interest in strong rules of origin on the part of food and light manufactures producers in Japan. In contrast, the percentage changes in Singapore’s imports from Japan are much more uniform, with the largest changes stimulated by the combined benefits of e-commerce and customs automatization for auto imports from Japan (12.6% increase in bilateral imports) and imports of machinery and equipment. Because of its predominant role in Japanese exports to Singapore, the increased volume in the machinery and equipment sector ($2,905 million) is by far the largest change in Singapore’s bilateral imports from Japan. Finally, Singapore’s imports from ROW rise across the board – indicating that this FTA package is not leading to the diversion of trade. Table 9.9 reports the changes in 2020 volume of output ($US millions) in Singapore across the four cumulative simulations. Although individual components of the FTA lead to output declines in a few cases, when combined with customs automatization, almost all sectors increase their output levels in 2020. (Other crops experience a very small decline in output.) Typically in the case of such FTAs one sees winners and losers – so it is surprising that there is no significant contraction of output across the sectors shown in Table 9.10. This is due to the growth effects of the FTA. The increase in capital stock available in Singapore, coupled with the relative balanced import effects, permits the simultaneous expansion of nearly all sectors of the economy.
5.2 Results for Japan The impacts of the FTA on Japan have a distinctly different character than those for Singapore. Japan’s exports to Singapore represent only 3.2% of total trade. Therefore, the strictly bilateral measures, including tariff cuts, reduced services trade barriers, and e-commerce regulations, have a relatively minor impact on aggregate output, trade, investment, and GDP. Rather, the impacts of the FTA on Japan are driven largely by the customs automatization process, which affects the cost of trading with all partners.
254
Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura
Table 9.7. Effect of Japan-Singapore free trade agreement on capital, real GDP, exports,
imports, and equity ownership in 2020 (cumulative percent differences from baseline)
Singapore Japan Korea Malaysia Thailand Rest of S.E. Asia China Hong Kong Taiwan South Asia Australia-New Zealand Canada USA Mexico Chile Western Europe Rest of World
Capital stocks
Real GDP
Real Real Overseas Equity held exports imports Wealth holdings by foreigners
1.81 0.33 0.10 0.27 0.32 0.17 −0.01 0.23 0.20 −0.02 0.02
1.67 0.20 0.11 0.31 0.33 0.13 0.04 0.27 0.18 0.01 0.03
1.00 1.93 0.08 0.15 0.30 0.09 0.10 0.08 0.10 −0.02 0.01
0.64 1.82 −0.05 0.01 0.11 0.00 0.00 0.01 −0.08 −0.04 −0.02
0.31 0.34 −0.03 0.00 −0.04 −0.01 −0.04 −0.26 −0.05 −0.06 −0.04
−0.77 −0.27 −0.08 −0.10 −1.53 −0.18 0.20 −0.66 −0.16 0.07 −0.06
2.66 1.45 0.06 0.37 1.72 0.33 −0.40 0.15 0.21 −0.19 −0.02
−0.04 −0.03 −0.02 0.03 −0.07 −0.05
−0.01 0.01 0.00 0.03 −0.01 −0.01
−0.07 −0.03 −0.08 0.03 −0.09 −0.04
−0.07 −0.01 −0.10 −0.01 −0.04 −0.02
−0.06 −0.07 −0.08 −0.04 −0.06 −0.05
0.07 0.08 0.10 −0.06 0.01 0.06
−0.24 −0.25 −0.28 −0.01 −0.19 −0.17
Source: Authors’ simulation.
This may be seen in Figure 9.8, which reports the impact of individual components of the FTA on the rate of return on investment in Japan. The cumulative effects of the first three (bilateral) elements of the agreement are negligible when compared to the impact of customs automatization. This reform boosts rates of return in Japan by increasing efficiency in the economy, which gives rise to a capital inflow (recall that this was also the case in Singapore). As a result, Japan’s trade balance deteriorates relative to the baseline (Figure 9.9). However, in the long-run, increased foreign income payments dictate an increase in exports, so that we observe the same U-shaped pattern for the trade balance change, relative to the baseline, that we saw for Singapore. The long-run impacts of the Japan-Singapore FTA on other macroeconomic variables in the Japanese economy are reported in the second row of Table 9.7. Whereas the investment and GDP results dominated the aggregated trade volume effects in Singapore, in Japan, where customs automatization lowers the cost of trade with all partners and the trade/GDP ratio is much smaller, the trade volume changes dominate. Specifically, Japan’s
255 Table 9.8. Effect of Japan-Singapore free trade agreement on imports in 2020 (import volume change in millions of US$,
percent change in parentheses) Japan’s imports from Singapore Rice n.t. Other grains Other crops Meat Other food Fish Textiles and wearing apparel Leather Extraction Chemicals and minerals
n.t. 10.83 (7.88) 95.67 (417.74) 688.24 (124.82) 2.17 (10.58) 38.15 (58.89) 54.41 (79.58) 21.67 (6.25) 294.75 (14.40)
Singapore’s imports from
ROW’s imports from
ROW
Japan
ROW
Japan
Singapore
ROW
0.06 (1.12) −5.32 ( −0.09) 75.25 (0.45) 46.55 (0.27) 42.74 (0.11) 27.27 (1.38) 1001.94 (1.71) 295.80 (1.57) 1671.95 (1.24) 1167.93 (1.83)
0.00 (1.72) 0.00 (0.78) 0.39 (0.44) 0.28 (1.90) 29.23 (9.08) 1.01 (2.46) 8.86 (3.35) 0.79 (5.81) 164.83 (3.85) 400.53 (7.91)
10.51 (2.04) 2.29 (1.11) 40.87 (0.74) 40.81 (1.89) 151.64 (1.81) 1.72 (0.86) 81.92 (0.89) 31.23 (1.32) 336.71 (0.61) 341.02 (0.68)
−0.39 ( −0.77) −0.11 ( −0.75) −7.51 ( −0.78) −2.60 ( −0.41) −22.50 ( −0.64) 6.77 (0.61) −4.41 ( −0.04) 4.71 (0.92) 723.81 (1.12) 3417.29 (3.63)
0.00 (0.12) 0.00 (0.08) −37.26 ( −1.05) −26.95 ( −1.94) 12.57 (0.13) −2.12 ( −0.68) −38.69 ( −0.74) 0.17 (0.02) 13.46 (0.08) −159.51 ( −0.27)
−2.05 ( −0.02) 17.81 (0.02) 120.01 (0.05) 166.32 (0.14) 758.77 (0.20) 12.95 (0.07) 216.13 (0.04) 100.95 (0.06) −1595.51 ( −0.10) −3219.44 ( −0.22) (continued )
256 Table 9.8 (continued) Japan’s imports from
Other manufactures Automobiles Machinery and equipment Utilities Construction Trade and transport Business and financial services Source: Authors’ simulation.
Singapore’s imports from
ROW’s imports from
Singapore
ROW
Japan
ROW
Japan
Singapore
ROW
48.13 (4.87) 1.87 (23.99) 1513.65 (13.45) 0.17 (0.91) 0.18 (110.04) 160.70 (0.87) 1300.06 (67.57)
950.83 (1.59) 745.02 (2.83) 3839.40 (2.72) 16.28 (1.19) 0.54 (0.65) 1395.34 (0.85) −386.34 ( −0.73)
95.53 (2.98) 284.58 (12.60) 2905.27 (9.12) −0.25 ( −1.10) 0.00 ( −0.84) −5.02 ( −0.82) 5.72 (0.09)
219.82 (0.82) 66.51 (1.08) 1610.66 (0.72) 30.83 (0.78) 0.38 (0.77) 55.55 (0.85) 558.88 (1.64)
235.35 (0.78) 342.25 (0.42) 9509.30 (2.49) −14.77 ( −1.86) −0.06 ( −1.53) −656.26 ( −1.06) −377.84 ( −1.46)
−3.04 ( −0.01) 92.80 (1.97) 747.49 (0.31) −12.40 ( −0.27) 0.32 (0.29) 22.97 (0.05) −14.49 ( −0.13)
−213.62 ( −0.03) −760.29 ( −0.10) −10050.1 ( −0.33) 35.94 (0.01) 26.41 (0.04) 647.75 (0.07) 1608.75 (0.27)
257 Table 9.9. Changes in 2020 volume of output ($US millions) in Singapore across the four cumulative simulations Japan
Rice Other grains Other crops Meat Other food Fish Textiles and wearing apparel Leather Extraction
Singapore
Sim1
Sim2
Sim3
Sim4
Sim1
Sim2
Sim3
Sim4
−8.02 ( −0.01) −0.32 ( −0.02) −4.34 ( −0.01) −33.75 ( −0.03) −192.29 ( −0.04) −6.79 ( −0.02) 6.02 (0.00) −0.50 ( −0.00) 49.51 (0.01)
−5.99 ( −0.01) −0.10 ( −0.01) −3.08 ( −0.01) −26.41 ( −0.02) −169.37 ( −0.03) −5.42 ( −0.02) 23.84 (0.01) 2.57 (0.01) 127.46 (0.02)
−1.67 ( −0.00) −17.11 ( −0.94) −32.32 ( −0.08) −189.46 ( −0.15) −221.83 ( −0.04) −19.90 ( −0.06) −680.72 ( −0.33) −150.86 ( −0.69) 365.43 (0.07)
−2.15 ( −0.00) −18.56 ( −1.02) −35.27 ( −0.09) −204.81 ( −0.16) −234.60 ( −0.04) −21.60 ( −0.06) −745.69 ( −0.36) −161.52 ( −0.74) 458.46 (0.09)
0.06 (0.50) 0.04 (0.59) −9.82 ( −0.50) 46.42 (2.20) 590.29 (4.80) 0.05 (0.02) 19.88 (0.36) 36.54 (4.11) 46.20 (0.17)
0.05 (0.46) 0.04 (0.54) −10.88 ( −0.55) 44.08 (2.09) 575.92 (4.69) 0.03 (0.01) 2.98 (0.05) 34.85 (3.92) 16.24 (0.06)
0.06 (0.53) 0.04 (0.53) −10.78 ( −0.55) 45.52 (2.16) 590.61 (4.81) 0.08 (0.03) 2.57 (0.05) 35.35 (3.98) −2.34 ( −0.01)
0.11 (0.95) 0.07 (1.05) −12.35 ( −0.63) 45.72 (2.16) 635.42 (5.17) 0.17 (0.06) 3.49 (0.06) 45.51 (5.12) 173.12 (0.65)
(continued )
258 Table 9.9. (continued) Japan
Chemicals and minerals Other manufactures Automobiles Machinery and equipment Utilities Construction Trade and transport Business and financial services
Singapore
Sim1
Sim2
Sim3
Sim4
Sim1
Sim2
Sim3
Sim4
−9.20 ( −0.00) 10.46 (0.00) 38.24 (0.02) 101.85 (0.01) −7.13 ( −0.00) −14.65 ( −0.00) 6.27 (0.00) −5.46 ( −0.00)
67.90 (0.01) 38.08 (0.01) 82.10 (0.04) 295.65 (0.03) 6.90 (0.00) 17.45 (0.00) 85.43 (0.01) −186.80 ( −0.02)
2088.55 (0.25) 107.54 (0.02) 450.15 (0.22) 5466.27 (0.57) 330.48 (0.08) 2499.17 (0.27) 688.18 (0.06) 1014.56 (0.09)
2242.25 (0.27) 163.65 (0.03) 312.65 (0.16) 6054.28 (0.63) 355.18 (0.09) 2777.50 (0.30) 716.83 (0.06) 1124.29 (0.10)
111.74 (0.17) 7.11 (0.03) −3.90 ( −0.05) −78.11 ( −0.03) 15.70 (0.07) 55.10 (0.14) 8.27 (0.01) 29.55 (0.05)
93.42 (0.14) −18.05 ( −0.07) −10.90 ( −0.13) −381.93 ( −0.16) 37.32 (0.16) 188.19 (0.46) −23.60 ( −0.02) 1155.20 (2.11)
−234.33 ( −0.36) −32.40 ( −0.12) 12.42 (0.15) −867.75 ( −0.37) 46.73 (0.20) 237.39 (0.58) 150.42 (0.15) 1228.67 (2.24)
203.16 (0.31) 108.48 (0.40) 86.38 (1.07) 2029.25 (0.86) 131.05 (0.56) 455.69 (1.12) 491.67 (0.50) 1484.53 (2.71)
Note: Sim1: Tariff only; Sim2: Tariffs and services; Sim3: Tariffs, services, and e-commerce; Sim4: Tariffs, services, e-commerce, and customs automatization. Source: Authors’ simulation.
Dynamic Effects of the “New-Age” Free Trade Agreement
259
Table 9.10. Welfare effects of FTA: Equivalent variation in 2020 in millions of US$ Full FTA Customs (% Change Tariffs Services E-commerce automatization Full FTA in Welfare) Japan −85.0 236.7 Korea −4.9 −6.0 Singapore 55.0 115.3 Malaysia 2.6 −4.0 Thailand −1.8 0.3 IndPhlViet −1.9 −8.1 China −4.1 −5.9 Hong Kong −0.2 −2.6 Taiwan −2.3 −2.1 SoAsia −2.7 −3.2 AusNZL −1.7 −4.1 Canada −2.4 −5.9 USA −20.2 −33.8 Mexico −0.6 −1.0 Chile −1.4 −2.7 WEurope −21.2 −39.2 Remainding countries −12.5 −27.5 World total −105.3 206.1
170.6 −10.6 55.5 1.1 −15.7 1.9 −12.3 3.5 −3.3 0.6 1.4 0.6 −11.0 −0.9 0.3 −21.2 6.8 167.3
6597.5 258.6 171.0 135.9 263.6 232.8 288.4 79.3 207.1 45.0 70.2 20.4 588.2 7.9 15.4 212.9 36.8 9231.2
6919.7 237.1 396.8 135.6 246.3 224.7 266.1 80.1 199.5 39.7 65.8 12.7 523.1 5.5 11.7 131.3 3.6 9499.2
0.146 0.058 0.668 0.162 0.168 0.088 0.040 0.092 0.078 0.010 0.019 0.002 0.008 0.002 0.021 0.001 0.000
Source: Authors’ simulation.
0.18 0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 2006
2007 Tariff only
2008
2009
Tar+Ser
Tar+Ser+Ecom
2010
2015
2020
Tar+Ser+Ecom+Cust
Figure 9.8. Effect of Japan-Singapore Free Trade Agreement on the rate of return in Japan (cumulative percent differences).
260
Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura
6,000 4,000 2,000 0
-2,000 -4,000 -6,000 -8,000 2006
2007
2008
Tariff only
Tar+Ser
2009 Tar+Ser+Ecom
2010
2015
2020
Tar+Ser+Ecom+Cust
Figure 9.9. Effect of Japan-Singapore Free Trade Agreement on Japan’s trade balance (cumulative differences, US $ millions).
exports are projected to be nearly 2% higher, relative to baseline, in 2020. Capital stocks and wealth are about one-third of a percent higher in the wake of the FTA, whereas Japan’s GDP gets a modest 0.2% cumulative boost by the year 2020. The modest macroeconomic effects are mirrored at the sector level. Whereas bilateral imports from Singapore receive a considerable boost – particularly for primary products and light manufactures (recall Table 9.8) – the subsequent changes in Japanese output are quite small and reflect a shift toward Japan’s comparative advantage in durable goods production (Table 9.9). The largest output volume declines in Japan are for textiles and wearing apparel ( −$746 million), other food, and meat products. However, the only decline in excess of 1%, relative to the baseline, is for other grains ( −1.02%; Table 9.9). As was the case with Singapore, trade with ROW increases, with the only declines in Japanese imports from ROW coming in other grains and in business and finance services (Table 9.8).
5.3 Results for Rest of World The remaining rows of Table 9.7 report the impact of the Japan-Singapore FTA on the macroeconomic performance of countries outside the FTA. The automatization of customs procedures increases trade throughout the Asia-Pacific region and the rest of the world, thus boosting real GDP in all regions except for Canada, Western Europe, and the residual region in the final row of this table. All of the Asian economies gain in terms of real GDP – with the largest impact felt in Thailand and Malaysia, two economies that
Dynamic Effects of the “New-Age” Free Trade Agreement
261
trade a great deal with Singapore and Japan. These increases in real GDP also fuel increased foreign investment, with the stock of foreign-owned equity in Thailand rising by 1.7% as a result of the FTA. The increase in foreign ownership in Singapore, Japan, and Thailand is financed by a modest increase in outward FDI by the United States, Canada, Mexico, China, and South Asia, as reported in the final column of Table 9.7. Many of the other Asian economies reduce their foreign ownership to increase investment in their domestic economies.
5.4 Welfare Effects A natural question to ask in relation to any free trade agreement is the following: Does it leave the world as a whole better off? Given the multiregion, multiperiod nature of this study, we face the challenging problem of aggregating benefits over countries and over time periods. For the reasons set out in Chapter 6, we focus on welfare at a particular point in time – in this case, we choose 2020 – at which point the investment story has played itself out. We then compute the static equivalent variation, for the representative household in each region, associated with the cumulative changes that have occurred between the baseline and the FTA simulations. These dollar values represent the annual increase (or decrease) in real income stemming from the presence of the FTA. The simple sum of these EV measures is our annual measure of the change in world welfare. Equivalent variations for each region and for the world as a whole are reported in Table 9.10. Here, each of the first four columns corresponds to one of the four components of the FTA. (The final column reports the per capita percentage changes in welfare due to the FTA. This is discussed later.) The traditional, bilateral tariff elimination associated with most FTAs generates global welfare losses, because the only significant bilateral tariffs remaining are Japan’s tariffs on primary products and light manufactures. When taken on their own, elimination of these tariffs creates costly trade diversion, with increased imports coming from Singapore at the expense of lower cost suppliers elsewhere. Singapore’s welfare rises as a result of the terms of trade gain that they experience. However, Japan’s welfare falls, as does that of all other countries. As one moves from this traditional tariff-based FTA to the new-age elements that focus less on commercial policy and more on improving efficiency, the prevalence of regional benefits increases. In the case of services trade liberalization, both Japan and Singapore gain and the overall benefits outweigh the costs, generating a global welfare gain of $206 million. In the case of e-commerce, the gains are spread across more than
262
Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura
half of the trading partners. Finally, in the case of the customs automatization component of the FTA, all regions gain. In fact, the last component dominates all of the others, and consequently, all regions benefit from the FTA. The very favorable outcome from customs automatization has several explanations. First, it is the only FTA measure that is nondiscriminatory. Customs automatization benefits all trading partners. Of course, the “linking benefit” derived from two countries synchronizing their systems gives an additional margin of preference to Japan–Singapore trade. However, eventually other countries that implement this system will also obtain this linking effect. Second, unlike tariff cuts, which lead to lost revenue, customs automatization saves time and hence lowers the effective price of the product. There is no lost revenue – apart from the cost of implementation – so the liberalizing country is unlikely to experience a loss in welfare. This raises the question: If customs automatization is such a windfall, why has it not already been implemented? One answer is that, like many administrative reforms, the barriers to reform are not merely economic. A second, more interesting, answer is that the direct benefits of customs automatization are quite small, and the costs are non-negligible. The Mitsubishi Research Institute (MRI) estimates that the cost of running the new system will be $36.7 million/year. It is only when the indirect benefits – specifically the opportunity costs of time in trade – are taken into account that customs automatization becomes an important feature of the FTA. To date, this particular barrier to trade has received scant attention. Hence the importance of Hummels’ (2000) work in quantifying the ad valorem value of time savings in trade. The final column of Table 9.10 reports the percentage changes in per capita welfare in each region of the world as a result of the FTA. Unlike the EV measure, this controls for economic size when making comparisons across regions. Not surprisingly, Singapore is the largest per capita winner from the FTA. Singapore is a very open economy, trade with Japan is quite important, and Singapore receives a substantial preference margin on tariffs, services trade, and e-commerce, as well as a linking benefit associated with Japan’s customs automatization measures. More surprising is the fact that Thailand and Malaysia gain relatively more from the FTA than does Japan. This occurs, despite the fact that they are not directly included in the FTA. The reason for these large gains is their relatively high trade dependence on Japan and Singapore, both of which are importing more from all destinations as a result of the agreement. Japan’s imports rise as a result of customs automatization, whereas Singapore’s imports rise in response to the increased demand for
Dynamic Effects of the “New-Age” Free Trade Agreement
263
its own products in the Japanese markets, as well as higher incomes. The subsequent increase in demand for products from Thailand and Malaysia gives rise to substantial terms of trade gains for both economies.
6. Conclusions This study has sought to quantify the dynamic benefits of Japan’s new-age FTA currently under negotiation with Singapore. We find that the impact of the FTA on investment, capital accumulation, and economic growth is significant, particularly in Singapore. Furthermore, global benefits from the proposed FTA are substantial – on the order of $9.5 billion/year by 2020. All regions of the world gain from this agreement, although 70% of the gains are captured by Japan, which is the region undertaking most of the reforms. It is interesting to note that if the FTA were implemented as a traditional trade agreement with tariff cuts being the centerpiece – perhaps adding some liberalization of rules governing direct trade in services – none of this would be true. The global welfare gains would be uncertain, trade diversion would be significant, and most nonparticipating regions would lose from the agreement. It is only when the new-age features – e-commerce and customs automatization – are added that benefits to the other regions begin to appear and the global gains become pronounced. In closing, it is important to note the limitations of this study. First, since this work began, the agreement has been finalized and was put into effect in November 2002. In particular, agriculture was left out of the agreement because of the problem of enforcing rules of origin on these heavily protected products. Because agriculture generates relatively few of the gains in our study, its omission is unlikely to have a substantial impact on our results. Similarly, we have found that the results are quite robust with respect to the baseline. The main developments between 2002 and 2005 in our baseline are China’s accession to the WTO and elimination of the textiles and apparel export quotas. Neither has a direct bearing on the Japan–Singapore trade relationship, and results not reported here show that China’s accession makes little difference for the direct impacts of this FTA on the member economies. Despite our best attempts to quantify new-age features of the JapanSingapore FTA, we have omitted a number of important elements of this agreement. Specifically, we have not incorporated the effects of liberalization of direct trade in transport and telecommunications services where barriers are potentially quite large. We have also failed to quantify the impact of liberalizing rules governing investment and the movement of natural persons.
264
Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura
These are central modes of delivery for the rapidly growing services sector, and their omission surely leads to an understatement of the impacts of the FTA on efficiency, investment, and growth. Finally, although there are many potential benefits of the FTA between Japan and Singapore, some costs have been neglected. Several elements of the FTA will involve implementation costs. In addition, customs automatization will involve recurring costs of about $37 million/year (MRI estimate). However, these are small when compared with the potential gains. Perhaps of greater concern are the costs associated with verifying that the products granted preferential treatment under the FTA in question are indeed produced in the partner country. This issue is of particular concern in the case of food products, textiles, apparel, and leather products under the JapanSingapore FTA. Japan’s tariffs in these sectors are still high, and given the high volume of re-exports from Singapore, the potential incentive for other countries to export foodstuffs and light manufactures through Singapore to Japan would be substantial. Very tight rules of origin that would prevent such transshipment could also prove costly to the businesses involved, thereby frustrating trade.
Appendix: Aggregation of GTAP Data Base Table 9.A1. Sector aggregation 50 sectors
GTAP code
This study
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Pdr Wht Gro vf Osd cb Pfb Ocr Ctl Oap Rmk Wol For Fsh Col Oil Gas
rice othgrains othgrains Othcrops Othcrops Othcrops Othcrops Othcrops meat meat othfood Othcrops extract fish extract extract extract
Paddy rice Wheat Cereal grains nec Vegetables, fruit, nuts Oil seeds Sugar cane, sugar beet Plant-based fibers Crops nec Bovine cattle, sheep and goats, horses Animal products nec Raw milk Wool, silk-worm cocoons Forestry Fishing Coal Oil Gas
Dynamic Effects of the “New-Age” Free Trade Agreement
265
Table 9.A1 (continued) 50 sectors
GTAP code
18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
Omn Cmt Omt Vol Mil Pcr Sgr Ofd bt Tex Wap Lea Lum Ppp pc Crp Nmm is Nfm Fmp Mvh Otn Ele Ome Omf Ely Gdt Wtr Cns tt Osp Osg Dwe
Minerals nec Bovine cattle, sheep and goat, horse meat products Meat products nec Vegetable oils and fats Dairy products Processed rice Sugar Food products nec Beverages and tobacco products Textiles Wearing apparel Leather products Wood products Paper products, publishing Petroleum, coal products Chemical, rubber, plastic products Mineral products nec Ferrous metals Metals nec Metal products Motor vehicles and parts Transport equipment nec Electronic equipment Machinery and equipment nec Manufactures nec Electricity Gas manufacture, distribution Water Construction Trade, transport Financial, business, recreational services Public admin and defense, education, health Dwellings
This study extract meat meat othfood othfood rice othfood othfood othfood Texwap Texwap leather Omnfcs Omnfcs pchemineral pchemineral pchemineral extract extract extract Autos Machequip Machequip Machequip Omnfcs utilities utilities utilities construction Tradetrans Busfinance utilities utilities
266
Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura Table 9.A2. Regional aggregation 45 regions
GTAP code
This study
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
aus nzl jpn kor idn mys phl sgp tha vnm chn hkg twn ind lka ras can usa mex cam ven col rap arg bra chl ury rsm gbr deu dnk swe fin reu eft cea fsu tur rme mar rnf saf rsa rss row
AusNZL AusNZL Japan Korea Indphlviet Malaysia Indphlviet Singapore Thailand Indphlviet China HongKong Taiwan SoAsia SoAsia SoAsia Canada USA Mexico ROW ROW ROW ROW ROW ROW Chile ROW ROW WEurope WEurope WEurope WEurope WEurope WEurope WEurope ROW ROW ROW ROW ROW ROW ROW ROW ROW ROW
Australia New Zealand Japan Korea Indonesia Malaysia Philippines Singapore Thailand Viet Nam China Hong Kong Taiwan India Sri Lanka Rest of South Asia Canada United States of America Mexico Central America and the Caribbean Venezuela Colombia Rest of the Andean Pact Argentina Brazil Chile Uruguay Rest of South America United Kingdom Germany Denmark Sweden Finland Rest of European Union EFTA Central European Associates Former Soviet Union Turkey Rest of Middle East Morocco Rest of North Africa South African Customs Union Rest of southern Africa Rest of sub-Saharan Africa Rest of World
Dynamic Effects of the “New-Age” Free Trade Agreement
267
References Ahuja, V. and D. Filmer. 1995, July. Educational Attainment in Developing Countries: New Estimates and Projections Disaggregated by Gender. World Bank Policy Research Working Paper 1489. Washington, DC: World Bank. Anderson, K. and H. Norheim. 1993. “Is World Trade Becoming More Regionalized?” Review of International Economics 1(2), 91–109. Brown, A. J. 1949. Applied Economics: Aspects of the World Economy in War and Peace. London: George Allen-Unwin. Bureau of Economic Analysis. 1999. “International Accounts Data U.S. International Services: Cross-Border Trade & Sales through Affiliates, 1986–99.” Available at http://www.bea.doc.gov/bea/di/1000serv/intlserv.htm. Central Intelligence Agency. 1997. The World Factbook 1997–1998. Washington, DC: Brassey’s. CPB. 1999, December. WorldScan: The Core Version. The Hague: CPB Netherlands Bureau for Economic Policy Analysis. Dee, P. and K. Hanslow. 2000. Multilateral Liberalisation of Services Trade. Staff Research Paper. Melbourne, Australia: Productivity Commission. Drysdale, P. 1967. Japanese-Australian Trade. Ph.D. dissertation, Australian National University, Canberra. Drysdale, P. and R. Garnaut. 1982. “Trade Intensities and the Analysis of Bilateral Trade Flows in a Many-Country World.” Hitsubashi Journal of Economics 22(2), 62–84. Fan, M. and Y. Zheng. 2000. “China’s Trade Liberalisation for WTO Accession and Its Effects on China – A Computable General Equilibrium Analysis.” Unpublished memo. Feldstein, M. and C. Horioka. 1980. “Domestic Saving and International Capital Flows.” Economic Journal 90, 314–29. Francois, J. 1998 September. Scale Economies and Imperfect Competition in the GTAP Model. GTAP Technical Paper No. 14. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Francois, J. 1999. “A Gravity Approach to Measuring Services Protection.” Unpublished manuscript, Erasmus University, Rotterdam. Francois, J. and C. Shiells. 1994. Modeling Trade Policy: Applied General Equilibrium Assessments of North American Free Trade. Cambridge: Cambridge University Press. Francois, J. and D. Spinanger. 2001. With Rags to Riches but Then What? Paper presented at the Fourth Annual Conference on Global Economic Analysis, June 27–9, West Lafayette, IN. Francois, J. and A. Strutt. 1999, June. “Post Uruguay Round Tariff Vectors for GTAP v.4.” Unpublished memo. Harrison, J., M. Horridge, and K. Pearson. 1999. Decomposing Simulation Results with Respect to Exogenous Shocks. Paper presented at the Second Annual Conference on Global Economic Analysis, June 20–2, Denmark. Harrison, W. J. and K. R. Pearson. 1996. “Computing Solutions for Large General Equilibrium Models Using GEMPACK.” Computational Economics 9, 83–127. Hertel, T. W. 1992. Introducing Imperfect Competition into the SALTER Model. Department of Agricultural Economics Staff Paper No. 93–3. West Lafayette, IN: Purdue University.
268
Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura
Hertel, T. W. 1997. Global Trade Analysis: Modeling and Applications. Cambridge: Cambridge University Press. Hoekman, B. 1995. “Assessing the General Agreement on Trade in Services.” In W. Martin and L. A. Winters (eds.), The Uruguay Round and the Developing Economies. World Bank Discussion Paper 307 (pp. 327–64). Washington, DC: World Bank. Hummels, D. 2000. “Time as a Trade Barrier.” Unpublished manuscript, Purdue University. IDE-JETRO. 2000. Toward Closer Japan-Korea Economic Relations in the 21st Century. Available at http://www.ide.go.jp/English/Lecture/pressmenu/pressE000606.html. Ingco, M. 1996. “Tariffication in the Uruguay Round: How Much Liberalization?” World Economy, 19(4), 425–47. Joint Study Group. 2000a. Japan-Singapore Economic Agreement for a New-Age Partnership. Japan Study Group Report. Available at http://www.mofa.go.jp/region/asiapaci/singapore/econo b.html. Joint Study Group. (2000). Report on the Free Trade Agreement between Japan and Singapore. Ministry of Foreign Affairs, Japan and Singapore. Kawasaki, K. 1999. Foundations and Applications of Applied General Equilibrium Analysis: A Simulation Analysis on Economic Structural Reform. Tokyo: Nippon-Hyoronsha Co., Ltd. Kojima, K. 1964. “The Pattern of International Trade among Advanced Countries.” Hitsubashi Journal of Economics 5(1). Korea Institute for International Economic Policy (KIEP). 2000. Economic Effects of and Policy Directions for Korea-Japan FTA. Washington, DC: KIEP. Martin, W., B. Dimaranan, and T. Hertel. 1999. “Trade Policy, Structural Change and China’s Trade Growth.” Unpublished memo. McDougall, R. A., A. Elbehri, and T. P. Truong. 1998. Global Trade Assistance and Protection: The GTAP 4 Data Base. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Nakajima, T. and Kwon, O. 2001. An Analysis of the Economic Effects of Japan-Korea FTA. Nigata, Japan: Economic Research Institute for Northeast Asia (ERINA). Tsutsumi, M. 2000. Regional Economic Integration and China’s Participation to WTO. JCER Discussion Paper No. 60. Tokyo: Japan Center for Economic Research. World Trade Organization. 2000. Trade Policy Review: Singapore. Geneva: WTO.
TEN
Resource Use and Technological Progress in Agriculture Elena I. Ianchovichina, Roy Darwin, and Robin Shoemaker
1. Introduction The world’s rate of population growth is slowing, but total population is still increasing at about 80 million per year and is expected to reach 10 billion by the middle of the 21st century (World Bank 1999). Most of this growth will take place in developing countries, particularly in Asia and Africa. These projected increases in population, along with growth in per capita incomes and associated changes in demand for agricultural commodities, are expected to increase pressures on natural resources both through the expansion of land under cultivation and through more intense use of resources already employed in agricultural production. In support of these expectations, a recent study by Evenson et al. (1999) estimates that without the development of high-yielding varieties of crops, prices for developing country consumers would likely be much higher than they are today. For example, technological advances in the cultivation of rice have reduced costly food imports by 8% and have eliminated the need to convert millions of hectares of forestland to agricultural uses, as would have been required had yields remained at 1960 levels. This evidence and the rise of food and resource prices in 2008 highlight the central role of agricultural research in ensuring sustainable rural development and food security in high-growth developing countries and in reducing the strain on forest ecosystems. Research on technical advances in agriculture is abundant. Norton and Davis (1981) and Alston (1993) provide exhaustive surveys of the literature Original version published in Ecology Economics 38(2), 2001. The application related to this chapter is Ch10 dfarm 95.zip and is available for download on the Web site at https: //www.gtap.agecon.purdue.edu/models/Dynamic/applications.asp.
269
270
Elena I. Ianchovichina, Roy Darwin, and Robin Shoemaker
on gains from agricultural research. Very few of the reviewed studies, however, examine the impacts of agricultural research in an international context and the effects of technological spillovers across regions. A more recent paper by Frisvold (1997) has filled in this gap in the literature by analyzing the open economy aspects of agricultural research in a multiregion general equilibrium context with the GTAP model (Hertel 1997). The paper nicely illustrates how international spillovers are important aspects of agricultural research. It also extends the analysis of spillovers in a single-commodity, single stage of production setting (Edwards and Freebairn 1984) by considering the full multiregion general equilibrium implications of such spillovers. The methodology adopted by Frisvold overcomes two of the limiting assumptions of the single-commodity partial equilibrium models used by most returns-to-research studies, namely that prices and production of all other commodities are fixed, and that research results from one region do not affect the productivity in others. The methodology does not, however, account for the dynamic effects of economic and population growth or calculate land-use changes between the agricultural sector and the rest of the economy. The methodology used by Frisvold was extended by Darwin et al. (1995, 1996) in the Future Agricultural Resources Model (FARM) to include the competition for land among agriculture, forestry, and all other sectors in the economy. In this chapter we extend the GDyn model presented in Chapter 2 to include the farm and agricultural detail in FARM. The resulting model is the dynamic FARM model (D-FARM). We use it to examine how agricultural total factor productivity (TFP) interacts with forestland use and timber harvest rates. Our results support the findings by Evenson et al. (1999) that a slowdown in agricultural TFP will raise world prices and lower world production of agricultural commodities while expanding farmland usage. This expansion in demand for farmland leads to the permanent conversion of forestland into farmland and increases the environmental threat from deforestation. The loss in productivity in agriculture affects welfare in all regions negatively, with the bulk of the problems faced by regions in which agriculture accounts for a higher share of the gross domestic product. The chapter is structured as follows. Section 2 describes the modeling framework and solution procedure and discusses the data and parameters. Section 3 discusses the design of and the results from the baseline simulation. Section 4 focuses on the effects of a slowdown in agricultural TFP with spillover effects. We summarize the major findings in Section 5.
Resource Use and Technological Progress in Agriculture
271
2. Methodology This chapter uses an extended version of the GDyn model (D-FARM) to estimate how TFP interacts with forestland use and timber harvest rates during the period from 2000 to 2007. This section briefly describes the model’s structure, outlines the solution procedure, and discusses the data and parameters.
2.1 The Model D-FARM has 12 regions and 18 commodities, 11 of which are agriculturerelated products (Table 10.1). It shares all features of the standard GDyn model except its treatment of land. Therefore, we dedicate the rest of this section to the description of those features of D-FARM that do not belong to GDyn. Land in D-FARM is divided into six land classes based on the length of the growing season − the longest continuous period of time in a year that soil temperature and moisture conditions support plant growth. Land classes 1 and 2 have growing seasons of 100 days or less. Land class 1 occurs where cold temperatures limit growing seasons, mainly polar and alpine areas. Canada and the former Soviet Union contain 79.3% of the world’s endowment of land class 1. Growing seasons in land class 2, which represent mainly semi-desert and desert areas, are limited by low precipitation levels. Land class 3 has growing seasons of 101–65 days and is 13% of all land. About half of the land class 3 endowment is located in Canada and the former Soviet Union. Land class 4 represents only 10.2% of land and has growing seasons ranging from 166 to 250 days. About 29% of land class 4 is located in Africa, and another 27.6% in the United States and Europe. Land class 5, which has growing seasons of 251–300 days, is only 7.7% of the world land area. Most of it (78.8%) is located in Africa, Latin America, and Asia. Land class 6, located mainly in the tropical areas of Africa, Asia, and Latin America, accounts for 20% of all land and has year-round growing seasons. Each land class in D-FARM supplies services to 26 commodity-producing sectors according to constant elasticity of transformation (CET) functions (Table 10.1). Eight of these sectors are manufacturing and service sectors. The remaining 18 are crops, livestock, and forestry sectors specific to the land class. For example, land class 1 supplies services to crop sector 1, livestock sector 1, forestry sector 1, and to the eight manufacturing and services sectors. Thus there are six land-class specific crop, livestock, and forestry sectors. Each manufacturing and services sector uses all land classes,
272
Elena I. Ianchovichina, Roy Darwin, and Robin Shoemaker
Table 10.1. Regional, sectoral, factor, and commodity aggregation for D-FARM A. Regional Aggregation 1. ANZ-Australia and New Zealand 2. CAN-Canada 3. USA-United States and Canada 4. JPN-Japan 5. OEA- Other East Asia: Korea, China Hong Kong, Taiwan 6. SEA-South East Asia: Indonesia, Malaysia, Philippines, Thailand, Singapore 7. EU-European Union 8. FSU-Former Soviet Union 9. OEU-Other Europe 10. LAM-Latin America 11. AFR-Africa 12. OAS-Other Asia B. Sectoral Aggregation 1. CRP-Crops (six sectors) 2. LIV-Livestock (six sectors) 3. FOR-Forestry (six sectors) 4. COG-Coal, oil and gas 5. MIN-Other minerals 6. FMM-Fish, meat, and milk 7. OPF-Other processed food 8. TCF-Textiles, clothing, and footwear 9. NMM-Other nonmetallic manufactures 10. OMN- Other manufactures 11. SRV-Services 12. CGDS-Capital goods formation C. Factor Aggregation 1–6 Six land classes 7. Water 8. Skilled Labor 9. Unskilled Labor 10. Capital 11. Natural Resource Factor
D. Commodity Aggregation 1. PDR-Paddy rice 2. WHT-Wheat 3. GRO-Other grains 4. V F-Vegetables, fruits, nuts 5. OSD-Oilseeds 6. C B-Sugar cane and sugar beet 7. PFB-Plant-based fibers 8. OCR-Other crops 9. LIV-Livestock: Wool, Other livestock Products 10. FOR-Forestry 11. COG-Coal, oil, and gas 12. MIN-Other minerals 13. FMM-Fish, meat, and milk: Fishing, Meat products, Milk products 14. OPF-Other processed foods: Processed rice, Other food products, Beverages and tobacco 15. TCF-Textiles, clothing, and footwear: Textiles, Wearing Apparel 16. NMM-Other nonmetalic manufactures: Lumber and wood products, Pulp, paper and printed products Petroleum and coal products Chemicals, rubber, and plastic Nonmetalic mineral products 17. OMN-Other manufactures: Primary iron and steel, Fabricated nonferrous metals, Transport industries, Other machinery and equipment Other Manufacturing 18. SRV-Services: Electricity, gas, water, Construction, Trade and Transport, Other services
whereas each crop, livestock, and forestry sector uses only one land type. The CET functions, which restrict land’s mobility among sectors, allow land to shift among economic sectors without losing sight of its inherent productivity differences.
Resource Use and Technological Progress in Agriculture
273
The crop sectors are multi-output sectors producing their own mix of eight crop commodities (Table 10.1). The mix is determined by CET functions with Allen partial elasticities less than zero. Regional production of the eight crop varieties is the sum of production of the six crop sectors. Each livestock and forestry sector produces only one aggregate commodity – livestock and forestry, respectively. Regional livestock and forestry outputs are also obtained by summing production across the six livestock and forestry sectors associated with the different land classes. In all other sectors, the final composite input equals output, and production is not land-class specific.
2.2 The Data The economic data by region, sector, and commodity are from version 4E (with upgraded energy content) of the GTAP Data Base (McDougall et al. 1998). Economic values of inputs and outputs were distributed to the land classes based on their respective shares in 1990 as derived by FARM’s geographic information system (Darwin, 1999; Darwin et al. 1995, 1996). This GTAP Data Base was enriched with financial data required by the investment theory in GDyn (Chapter 4). To capture the longer term effects of agricultural productivity, we develop a baseline scenario that traces the growth of the world economy from 1995 to 2007. This baseline uses estimates of annual growth rates for regional GDP, gross domestic investment (GDI), population, skilled labor, and unskilled labor (Tables 10.5 – 10.7). They are based on the projections collected and discussed in Chapter 5. The policy projections in the baseline encompass the implementation of the Uruguay Round (UR), tariff reductions implemented in 2000 by the Chinese before China’s World Trade Organization (WTO) accession, the implementation of its accession to the WTO, the implementation of the Agreement on Textiles and Clothing, and finally shocks to tariff rates required to simulate a slow decrease in tariffs after the completion of the Uruguay Round (see Chapter 5). Most parameters in D-FARM are inherited from GTAP (Hertel 1997) and are based on a review of the literature. These include the Allen partial elasticities for primary factors, imported intermediates (Table 10.4), and the price and income elasticities for private consumption (Tables 10.2 and 10.3). In Darwin et al. (1996) the CET functions for land services have Allen partial elasticities of −1.0. In this study we set these elasticities 50% lower to reflect that land movements are more restricted within the shorter time horizon
274 Table 10.2. Own-price elasticities of demand in the Dynamic Future Agricultural Resources Model at initial equilibrium, by region
PDR WHT GRO VF OSD CB PFB OCR LIV FOR COG MIN FMM OPF TCF NMM OMN SRV
ANZ
CAN
USA
JPN
OEA
SEA
EU
FSU
OEU
OAS
LAM
AFR
−.076 −.082 −.080 −.110 −.069 −.237 −.159 −.161 −.165 −.782 −.659 −.782 −.104 −.348 −.590 −.728 −.669 −.155
−.071 −.071 −.071 −.066 −.062 −0.61 −.061 −.062 −.095 −.785 −.710 −.787 −.094 −.246 −.590 −.725 −.680 −.175
−.116 −.116 −.116 −.070 −.067 −.067 −.067 −.069 −.066 −.885 −.776 −.885 −.079 −.317 −.668 −.834 −.782 −.174
−.070 −.070 −.070 −.268 −.264 −.264 −.264 −.264 −.598 −1.00 −.898 −1.00 −.516 −.401 −.756 −.939 −.893 −.229
−.072 −.069 −.067 −.222 −.226 −.283 −.214 −.202 −.217 −.407 −.280 −.784 −.254 −.227 −.371 −.460 −.467 −.278
−.059 −.074 −.071 −.172 −.174 −.173 −.174 −.151 −.185 −.520 −.435 −.498 −.163 −.165 −.361 −.444 −.448 −.204
−.075 −.113 −.105 −.190 −.151 −.167 −.147 −.189 −.173 −.880 −.728 −.870 −.174 −.354 −.634 −.676 −.755 −.326
−.058 −.058 −.058 −.167 −.167 −.167 −.167 −.167 −.085 −.398 −.310 −.398 −.095 −.185 −.266 −.369 −.311 −.102
−.010 −.078 −.067 −.179 −.182 −.188 −.206 −.192 −.113 −.616 −.364 −.532 −.124 −.317 −.497 −.691 −.661 −.284
−.046 −.048 −.057 −.163 −.135 −.135 −.135 −.148 −.129 −.316 −.320 −.313 −.168 −.179 −.213 −.353 −.328 −.188
−.067 −.082 −.053 −.190 −.202 −.172 −.186 −.188 −.147 −.501 −.339 −.400 −.146 −.215 −.343 −.443 −.407 −.206
−.116 −.062 −.066 −.141 −.123 −.115 −.136 −.137 −.130 −.295 −.351 −.277 −.180 −.164 −.209 −.360 −.297 −.173
Source: GTAP 4 Data Base (McDougall, Elbehri, and Truong 1998).
275 Table 10.3. Income elasticities for private consumption in the Dynamic Future Agricultural Resources Model, by region
PDR WHT GRO VF OSD CB PFB OCR LIV FOR COG MIN FMM OPF TCF NMM OMN SRV
ANZ
CAN
USA
JPN
OEA
SEA
EU
FSU
OEU
OAS
LAM
AFR
0.131 0.137 0.135 0.194 0.134 0.380 0.270 0.273 0.262 1.121 0.992 1.121 0.155 0.624 0.927 1.120 1.068 1.079
0.130 0.130 0.130 0.128 0.128 0.128 0.128 0.128 0.153 1.114 0.998 1.114 0.138 0.450 0.925 1.114 1.076 1.075
0.168 0.168 0.168 0.141 0.141 0.141 0.141 0.141 0.118 1.118 1.004 1.118 0.117 0.520 0.941 1.118 1.128 1.046
0.162 0.162 0.162 0.355 0.355 0.355 0.355 0.355 0.689 1.089 0.992 1.089 0.611 0.498 0.881 1.089 1.065 1.112
0.435 0.435 0.413 0.734 0.641 0.527 0.420 0.815 1.054 1.321 1.102 1.183 0.668 0.667 0.934 1.285 1.180 1.155
0.174 0.370 0.365 0.635 0.634 0.661 0.662 0.479 0.650 1.167 1.022 1.183 0.570 0.597 0.919 1.263 1.066 1.223
0.147 0.185 0.177 0.291 0.236 0.276 0.226 0.291 0.253 1.129 0.994 1.121 0.236 0.551 0.922 1.123 1.055 1.078
0.187 0.187 0.187 0.543 0.543 0.543 0.543 0.543 0.281 1.151 0.920 1.151 0.297 0.627 0.875 1.151 1.063 1.117
0.206 0.205 0.200 0.499 0.445 0.284 0.347 0.415 0.357 1.342 1.147 1.385 0.290 0.637 1.018 1.333 1.185 1.103
0.313 0.301 0.311 0.593 0.732 0.723 0.734 0.661 0.637 1.524 1.013 1.343 0.557 0.691 0.909 1.289 1.148 1.221
0.195 0.267 0.183 0.527 0.504 0.580 0.529 0.537 0.447 1.221 1.030 1.232 0.403 0.621 0.943 1.216 1.145 1.182
0.686 0.353 0.384 0.642 0.641 0.633 0.650 0.637 0.624 1.514 0.915 1.535 0.618 0.689 0.939 1.343 1.215 1.256
Source: GTAP 4 Data Base (McDougall, Elbehri, and Truong 1998).
276
Elena I. Ianchovichina, Roy Darwin, and Robin Shoemaker Table 10.4. Allen partial elasticities for primary factors (σ)
and between domestic and imported commodities (ω) Sectors Crops 1 ... Crops 6 Livestock 1 ... Livestock 6 Forestry 1 ... Forestry 6 – COG MIN FMM OPF TCF NMM OMN SRV
σ
Commodities
ω
0.24 0.24 0.24 0.24 0.24 0.24 0.20 0.20 0.20 – 0.20 0.20 0.29 1.12 1.26 1.26 1.26 1.40
PDR WHT GRO VF OSD CB PFB OCR LIV FOR COG MIN FMM OPF TCF NMM OMN SRV
2.20 2.20 2.20 2.20 2.20 2.20 2.20 2.20 2.78 2.80 2.80 2.80 2.29 2.45 3.32 2.05 3.33 1.94
Note: Crop 1 . . . Crop 6 is defined as Crop 1, Crop 2, Crop 3, Crop 4, Crop 5, and Crop 6; Livestock 1 . . . Livestock 6 is defined as Livestock 1, Livestock 2, Livestock 3, Livestock 4, Livestock 5, and Livestock 6; and Forestry 1 . . . Forestry 6 is defined as Forestry 1, Forestry 2, Forestry 3, Forestry 4, Forestry 5, and Forestry 6. Source: GTAP 4 Data Base (McDougall, Elbehri, and Truong 1998).
of one year. Because there are few estimates of Allen partial elasticities of substitution for crop supplies, we set the values of these elasticities to −1. This reduces the CET functions to Cobb-Douglas ones and means that the revenue shares for crop services and livestock services, for example, received by landowners and the revenue shares received for wheat, other grains, and nongrains by crop producers within a region are constant, but not equal, across all levels of revenue (Darwin et al. 1995). We set the parameters determining the speed of adjustment in the investment theory of the model as in Chapter 3. Darwin et al. (1995) conducted a sensitivity analysis to test the importance of parameter specification to model results. They concluded that measures of total or sectoral world production are not very sensitive to changes in parameters. The signs of changes remained the same in all cases when these elasticities were changed by 50% in either direction.
277 Table 10.5. Gross domestic product: annual growth rates (percent)
ANZ CAN USA JPN OEA SEA EU FSU OEU OAS LAM AFR
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
3.8 2.3 2.4 1.4 8.9 8.3 2.3 −5.2 3.4 5.0 1.3 2.9
3.2 1.5 2.4 3.5 7.9 7.2 1.7 −4.3 2.5 5.3 3.5 4.7
2.8 3.8 3.8 0.9 7.2 3.4 2.5 1.2 2.4 4.9 5.1 2.8
3.1 2.9 3.9 −2.9 2.6 −7.2 2.8 −3.4 2.6 2.5 1.9 2.9
2.4 2.1 3.1 −1.1 4.3 −0.9 2.0 −5.5 2.3 1.9 −1.0 2.6
2.8 2.5 2.1 0.3 5.5 3.0 2.7 0.6 2.7 3.2 2.4 3.8
3.2 2.6 2.4 0.9 6.1 4.4 2.6 2.4 3.0 3.9 3.9 4.0
3.4 2.8 2.5 1.5 6.4 5.1 2.7 4.0 3.0 4.2 4.3 4.1
3.4 2.6 2.6 2.3 6.5 5.5 2.7 4.8 3.1 4.4 4.3 4.1
3.4 2.7 2.5 2.1 6.6 5.8 2.6 5.2 3.1 4.4 4.4 4.1
3.4 2.9 2.6 2.2 6.6 5.7 2.6 5.2 3.2 4.4 4.5 4.2
3.4 2.7 2.5 2.1 6.7 5.7 2.5 5.2 3.2 4.4 4.5 4.2
Source: Chapter 5.
278 Table 10.6. Target vs. achieved annual gross domestic investment growth rates (percent)
ANZ CAN USA JPN OEA SEA EU FSU OEU OAS LAM AFR a
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
1.7a −1.3b 1.3 −1.7 5.4 2.3 2.1 −0.9 13.1 9.8 13.6 10.2 4.2 1.2 −5.0 −7.8 9.3 6.1 28.3 24.5 3.9 0.9 7.5 4.4
5.0 4.2 −2.4 −3.0 5.7 5.0 6.7 6.0 7.0 6.2 10.3 9.6 −0.7 −1.4 −15.0 −15.6 7.2 6.4 4.8 4.1 4.3 3.5 6.6 5.8
1.2 2.2 17.0 18.0 7.1 8.1 −4.8 −4.0 2.8 3.8 −3.2 −2.3 2.5 3.4 −5.0 −4.1 4.2 5.1 4.7 5.7 10.4 11.4 5.1 6.1
3.1 4.6 5.6 7.2 6.6 8.2 −8.4 −7.0 0.4 2.0 −21.2 −20.0 4.6 6.2 −12.5 −11.2 6.6 8.2 0.3 1.8 3.9 5.5 3.1 4.7
2.6 3.7 4.4 5.6 4.2 5.4 −4.2 −3.1 4.5 5.7 −7.5 −6.4 3.5 4.6 −9.5 −8.5 3.2 4.4 1.5 2.7 −5.1 −4.0 4.5 5.7
2.7 2.8 3.5 3.7 2.0 2.2 0.3 0.4 5.8 6.0 0.8 0.9 3.3 3.5 2.0 2.1 4.5 4.7 3.5 3.6 4.2 4.3 5.7 5.9
4.2 4.0 3.4 3.2 2.8 2.6 1.0 0.8 6.6 6.4 4.0 3.8 2.8 2.6 4.0 3.8 4.6 4.4 4.3 4.1 5.9 5.7 6.0 5.8
4.7 4.4 3.2 2.9 1.6 1.3 2.9 2.5 7.2 6.8 5.8 5.5 3.4 3.0 5.0 4.6 4.7 4.4 5.0 4.6 5.3 4.9 5.9 5.5
4.6 4.5 2.8 2.7 1.6 1.6 2.1 2.1 7.2 7.1 6.2 6.1 3.2 3.1 6.0 5.9 4.8 4.7 5.0 4.9 5.3 5.2 5.9 5.9
4.5 4.4 3.0 2.8 1.9 1.7 1.8 1.7 7.4 7.2 6.6 6.4 3.2 3.0 7.0 6.8 4.9 4.7 5.0 4.8 5.4 5.2 6.1 5.9
4.5 4.4 3.3 3.1 1.7 1.6 2.1 1.9 7.3 7.2 6.6 6.4 3.2 3.1 7.0 6.8 4.9 4.7 5.0 4.9 5.4 5.3 6.1 5.9
4.5 4.4 3.0 2.8 1.6 1.4 2.1 1.9 7.6 7.4 6.6 6.4 3.1 3.0 7.0 6.8 4.9 4.8 5.0 4.9 5.4 5.2 6.1 5.9
Target annual gross domestic growth rates. Annual gross domestic growth rates achieved with D-FARM. Source: Chapter 5. b
Resource Use and Technological Progress in Agriculture
279
3. Simulation Design and Baseline Results As mentioned in Section 2 our baseline simulation strategy is to rely on outside projections for the macroeconomics and to use D-FARM to determine sectoral and trade results by projecting the world economy between 1995 and 2007. Tailoring the model to outside macro forecasts involves a two-step procedure for calibrating a technical change scenario to achieve the GDP and GDI growth rates in Tables 10.5 and 10.6, respectively. The first step covers the period between 1995 and 2000. It involves calibrating (a) a regional factor-saving technological change parameter, F (or afereg in GDyn), which is used as an instrument to achieve a GDP target in each region, and (b) a risk premium parameter, which is used to achieve the GDI targets and represents changes in investors’ expectations about returns to capital in each region during the period of the Asian crisis and the subsequent financial turmoil in Latin America and the former Soviet Union. A shock to afereg specifies a technological change specific to each region, uniform across industries within each region, applying to all factors other than physical capital, and uniform across the factors to which it applies. The final database from this first step of the calibration procedure serves as a starting point for the projections until 2007. The second step of the calibration procedure covers the period between 2000 and 2007. It involves again calibrating the regional factor-saving technological change parameter afereg, and a parameter B (or bfereg in GDyn), thereby defining regional technological bias toward physical capital, which is used to achieve a regional GDI target. Average factor productivity growth rates for afereg in this period are shown in Table 10.7. Unlike afereg, parameter bfereg does not represent an improvement or deterioration in technology, but merely a change in bias of technology toward the use of physical capital. The realized annual GDP growth rates match the target annual GDP growth rates (Table 10.5). Table 10.6 shows target, and below them, in italics, achieved annual GDI growth rates. A comparison of the numbers in this table indicates that the target GDI growth rates are in most cases very close to the achieved GDI growth rates. Total factor productivity is exogenous both in the baseline and policy simulations.1 The calibration to the GDP and GDI targets in Table 10.4 produced growth rates in technology that we used both in the base and policy runs. Rates of growth in agricultural TFP were assumed to be slightly faster (0.7% per year) than for nonagriculture in all regions based on evidence from Bernard and Jones (1993). 1
Total factor productivity parameters were treated as endogenous variables only for the purpose of calibrating to the GDP targets in Table 10.7.
280
Elena I. Ianchovichina, Roy Darwin, and Robin Shoemaker
Table 10.7. Macroeconomic scenario: Annual growth rates for selected variables (percent)
Region ANZ CAN USA JPN OEA SEA EU FSU OEU OAS LAM AFR a b
Populationa
Unskilled labora
Skilled labora
Average productivity growth: aferegb
0.8 0.8 0.9 0.2 0.7 1.4 0.1 0.3 0.2 1.6 1.4 2.4
1.1 0.9 1.0 −0.2 0.5 1.4 −0.2 0.7 0.1 2.4 1.3 2.6
0.9 0.9 1.0 −0.6 3.3 6.4 0.0 1.0 0.3 5.1 5.5 3.3
2.6 1.8 1.5 0.8 5.4 5.7 2.0 4.5 2.1 1.7 4.0 1.7
Source: Chapter 5. Source: Calibration performed with D-FARM.
Because the agricultural technical change scenario is an important determinant of the baseline results, we compare agricultural TFP growth rates with growth rates in public agricultural research expenditures over the past 20 years for selected regions in our model.2 The growth rates in Table 10.8 provide only a rough estimate of historical growth rates of agricultural research expenditures, because private expenditures might be sizable, especially in developed regions. The numbers in Tables 10.7 and 10.8 suggest that the average agricultural productivity growth rates assumed in the baseline are in line with the historical average growth in agricultural public expenditures. The only exception is Latin America, where our study assumes an agricultural TFP growth rate that is more than a percentage point higher than the historical growth rate for public agricultural research expenditures. The results presented next are not predictions. Our main purpose is to show the ecological-economic information generated by the proposed framework given our assumptions. By examining these results, we obtain a better understanding of the interrelationships among economic and ecological variables. The baseline results represent the effects of world population growth and economic growth based on our trade policy scenario and international capital movements. The baseline results suggest that, given the current consensus 2
Data on public agricultural research expenditure for the other regions in our study were not available.
Resource Use and Technological Progress in Agriculture
281
Table 10.8. Growth rates in public agricultural research expenditures and
agricultural productivity Public agricultural research expenditures
Region Developed regions Other East Asia – China South East Asia Latin America Africa
1971
1991
Average growth: 1971–91 (percent)
4320 457 861 507 1158a
6956 1494 3502 944 2068a
2.41 6.10 7.27 3.16 2.94
Average Ag. productivity growth in baselinec (percent)
Average Ag. productivity growth in alternative case (percent)
2.60b 6.10 6.40 4.70 2.40
1.90b 5.40 5.70 4.00 1.70
a
This number is the sum of public agricultural expenditure in Sub-Saharan and North Africa (incl. West Asia). b Numbers in this column equal those in the last column of Table 10.7 plus 0.7 percentage points. c This is an estimate based on the results in the last column of Table 10.7 for Australia and New Zealand, Canada, United States, Japan, and the European Union. Source: Table 3.12 in Alston et al. (1999). Numbers in columns 2 and 3 of this table are in millions of 1985 international dollars.
about the growth of the world economy represented by the macroeconomic assumptions in our study and our assumptions about agricultural technical change, food security problems are not likely to be exacerbated over the next decade. Prices of all farm and food commodities, for example, decline (Table 10.9).3 Figure 10.1 shows the time paths of yearly percentage changes in the world prices of rice, wheat, other grains, and vegetables. Average production growth rates for all major crops vary between 2.7 and 3.0% per annum and are higher than world population growth rates (Table 10.7). Thus the world is not expected to experience shortages of food supplies over the medium term (e.g., through 2007). The only exception may be Africa, where slight pressure on crop prices occurs over the period (not shown). However, the slight rise in the price of world forestry products indicates that pressure on the world’s forest resources will continue under our baseline projections. Other evidence supports this as well. Although forestland is projected to increase across all land classes except in the northernmost forests in Canada and the former Soviet Union (Table 10.10), timber harvest 3
The results, reported in Table 10.9, depict changes in real, not nominal, prices. All prices in the model are relative to the price of a global savings commodity, chosen as the numeraire in our model.
282
Elena I. Ianchovichina, Roy Darwin, and Robin Shoemaker Table 10.9. World prices and production by commodity
Baseline average annual percentage change in:
Average annual percent change from baseline due to a slowdown in agricultural TFP in:
Commodity
Price
Output
Price
Output
1. PDR-Paddy rice 2. WHT-Wheat 3. GRO-Other grains 4. V F-Vegetables, fruits, nuts 5. OSD-Oilseeds 6. C B-Sugar cane and sugar beet 7. PFB-Plant-based fibers 8. OCR-Other crops 9. LIV-Livestock 10. FOR-Forestry 11. COG-Coal, oil, and gas 12. MIN-Other minerals 13. FMM-Fish, meat, and milk 14. OPF-Other processed foods 15. TCF-Textiles, clothing, footwear 16. NMM-Other nonmetalic mnfc. 17. OMN-Other manufactures 18. SRV-Services
−1.2 −0.5 −0.9 −0.9 −0.4 −0.1 −0.2 −1.0 −0.7 0.3 6.1 1.0 −0.3 −0.3 −0.2 0.9 0.4 −0.1
4.2 3.0 2.7 3.1 3.2 3.4 4.1 2.9 3.2 3.6 3.3 3.6 2.3 2.5 3.2 3.0 3.3 3.0
0.91 0.83 0.89 0.82 0.91 0.89 0.88 0.88 0.67 0.08 −0.35 −0.07 0.28 0.22 0.05 −0.06 0.00 0.00
−0.23 −0.13 −0.11 −0.19 −0.16 −0.20 −0.18 −0.15 −0.14 −0.08 −0.03 −0.03 −0.08 −0.09 −0.07 −0.03 −0.01 −0.02
Source: Authors’ simulations with D-FARM.
rates increase as well (Table 10.11).4 Hence, economic depletion of forests intensifies in our baseline scenario. These results are consistent with those obtained in a FARM-based study of the impacts of population growth on forests in moist tropical areas of Asia, Africa, and Latin America (Darwin et al. 1996). In our current baseline scenario, however, forest depletion cannot be attributed to the conversion of forestland to agricultural land because cropland and pasture generally decline (Table 10.10).
4. Effects of a Slowdown in Agricultural TFP To test the importance of technological progress in agriculture we project a less optimistic agricultural TFP growth scenario. Specifically, we assume that, after the year 2000, in all regions agricultural TFP growth slows down 4
Estimated percentage changes in timber harvest rates are estimated by subtracting percent changes in forestland from percent changes in forest output.
Resource Use and Technological Progress in Agriculture
283
0 01
20
02
03
20
20
20
04
20
05
20
06
20
07
-0.2 2
-0.4
% yearly changes
-0.6
-0.8
-1
-1.2
-1.4
-1.6
-1.8 Rice
Wheat
Other grains
Fruits & vegetables
Figure 10.1. Yearly growth rates in world prices. Source: Authors’ simulations with D-FARM.
so that there is no difference between TFP growth rates in agriculture and in the rest of the economy. We implement this by eliminating 0.7% percentage points per year from the growth in the agricultural TFP parameter in all regions. The comparison between the baseline and this policy simulation reveals possible effects of a slowdown in agricultural research. Lower TFP in the agricultural sector relative to the baseline implies higher agricultural and processed food commodity prices. World farm commodity prices rise by about 0.9% per annum on average relative to baseline in the absence of faster TFP growth in agriculture after 2000 (Table 10.9). The increase in world commodity prices is driven by the decline in world production of farm and food commodities relative to the baseline (Table 10.9). To see how these results compare to recent findings in the literature we look at the results produced by Evenson et al. (1999), who focus on agricultural research and productivity in India’s rice sector. These authors found that without technological progress, which led to the development of high-yielding varieties of rice over the past 40 years, rice prices in India would likely be as much as 40% higher than they are today. Our study also
284
Elena I. Ianchovichina, Roy Darwin, and Robin Shoemaker
Table 10.10. Quantity of land services demanded by type of activity in the baseline:
Average yearly percentage changes Forested land Region ANZ CAN USA JPN OEA SEA EU FSU OEU OAS LAM AFR
Cropland
Pastures
Type 1
Type 2
Type 3
Type 4
Type 5
Type 6
−0.42 −0.13 0.02 −0.32 −0.24 −0.15 −0.31 −0.10 −0.46 −0.08 −0.22 −0.04
0.24 −0.15 −0.25 −0.70 0.30 −0.04 −0.63 0.20 0.22 0.04 0.13 0.09
0.18 −0.02 0.15 N/A 0.30 N/A 0.08 −0.10 0.01 0.56 0.76 0.53
0.11 0.15 0.10 N/A 0.50 N/A 0.40 0.04 0.31 0.54 0.90 0.47
0.40 0.43 0.40 0.78 1.32 1.38 0.73 0.08 1.14 1.29 2.04 0.83
0.58 1.02 0.81 0.47 1.60 1.18 0.87 0.07 1.18 1.30 2.24 0.82
0.54 N/A 0.76 0.81 1.49 1.44 0.80 0.20 1.03 2.22 1.12 1.07
0.11 N/A 0.76 0.49 1.20 1.31 0.61 N/A 0.80 1.35 2.26 1.26
Source: Authors’ simulations with D-FARM.
suggests that, if the growth in agricultural productivity falls to the rate of technological growth in the rest of the economy, in 35 to 40 years the price of rice in Other Asia, a region in our study dominated by the economy of India, may be 40 to 47% (or 0.97% per annum on average) higher than the baseline. In addition, in Table 10.12 we report changes in commodity prices Table 10.11. Annual changes in timber harvest rates as a result of economic and
population growth in baseline (percent) Timber harvest rate on land types
ANZ CAN USA JPN OEA SEA EU FSU OEU OAS LAM AFR
1
2
3
4
5
6
2.59 1.70 2.19 N/A 5.83 N/A 2.52 2.31 3.20 2.89 2.33 2.94
2.76 1.64 2.27 N/A 5.95 N/A 2.36 2.23 2.93 3.05 2.41 3.08
2.72 1.64 2.06 0.62 5.37 3.48 2.19 2.19 2.65 2.86 1.39 2.69
2.74 1.53 2.22 0.89 5.19 3.39 2.22 2.21 2.55 2.62 1.50 2.57
2.63 N/A 2.15 1.01 5.24 3.56 2.38 2.17 2.68 3.08 2.40 2.64
2.71 N/A 2.13 0.88 5.34 3.51 2.30 N/A 2.61 2.77 1.42 2.70
Source: Authors’ simulations with D-FARM.
Resource Use and Technological Progress in Agriculture
285
Table 10.12. Projected cumulative changes in commodity prices due to slowdown in
agricultural TFP growth over a 40-year period relative to baseline (percent)
ANZ CAN USA JPN OEA SEA EU FSU OEU OAS LAM AFR
PDR
WHT
GRO
VF
OSD
CB
PFB
OCR
LIV
20.4 44.1 48.8 46.4 39.5 53.6 37.3 39.5 13.9 47.2 42.5 47.2
36.5 39.5 39.5 38.8 35.8 45.6 31.4 40.3 31.4 44.1 41.0 45.6
38.8 45.6 45.6 39.5 40.3 52.0 33.6 43.3 30.7 42.5 41.0 48.0
39.5 40.3 43.3 38.0 40.3 56.1 27.2 36.5 25.1 40.3 39.5 43.3
41.0 43.3 44.9 42.5 41.8 50.4 35.1 41.0 35.1 44.9 41.0 48.8
41.0 44.1 46.4 42.5 42.5 48.8 32.2 43.3 32.9 44.1 40.3 46.4
39.5 41.0 39.5 43.3 38.0 59.4 35.1 40.3 34.3 43.3 42.5 44.9
38.8 44.1 44.9 41.8 44.9 50.4 35.1 38.8 34.3 44.9 41.0 45.6
32.9 28.6 25.1 14.5 32.9 35.8 19.7 17.7 21.1 37.3 51.2 34.3
Source: Authors’ simulations with D-FARM.
expected in 40 years for all other regions. These results suggest that in the case of a slowdown in agricultural productivity South East Asia might face the largest increases in the prices of most crops relative to the baseline. In almost all cases the prices of farm products in this region are expected to be 50% higher at the end of 40 years compared to the baseline. In addition to experiencing higher commodity prices, all regions may need to convert more forestland into farmland as agricultural productivity slows down over the period 2000–7 and the farm sectors use more irrigation water and land compared to the baseline (Tables 10.13 and 10.14). For instance, the quantity of farmland demanded in all regions is expected to increase relative to the baseline (Table 10.14), driving up the regional prices or rents of farmland (Table 10.14). Given that about 75% of farmland is cropland, the TFP slowdown means higher prices of cropland compared to pastureland relative to the baseline due to the stronger demand for cropland services as shown in Table 10.14. The increased demand for farmland and its accompanying increase in land rents lead to higher farmland income in all regions relative to the baseline (Table 10.15). Farm income also increases in the alternative case relative to the baseline (Table 10.15). However, the increases in farm income are expected to be less than half of the increases in farmland income. This suggests that in all regions any benefits from the slowdown will accrue mostly to landowners, not farm workers. Indeed, in poorer regions such as Africa and Other Asia, income and wages of unskilled workers in these economies decline relative to the baseline (Table 10.15). The loss of productivity in
286
Elena I. Ianchovichina, Roy Darwin, and Robin Shoemaker
Table 10.13. Cumulative percentage changes relative to baseline in the price of water
and quantity of forestland demanded by land type, 2007 Region ANZ CAN USA JPN OEA SEA EU FSU OEU OAS LAM AFR
Price of water
Type 1
Type 2
Type 3
Type 4
Type 5
Type 6
3.4 0.3 5.0 2.9 13.0 4.3 3.9 7.3 4.3 13.7 7.2 11.7
0.1 0.1 0.1 N/A −0.2 N/A −0.5 −0.5 0.0 −0.1 −0.8 −0.4
0.0 −0.8 −0.9 N/A −0.8 N/A −1.1 −1.3 −6.7 −0.7 −1.2 −0.9
−4.0 −2.3 −2.5 −1.6 −2.5 −2.0 −2.4 −3.1 −2.8 −3.0 −2.9 −2.7
−4.7 −3.9 −3.8 −1.3 −2.7 −1.7 −2.8 −3.9 −3.0 −2.6 −2.8 −2.4
−3.9 N/A −2.8 −2.2 −2.5 −1.8 −3.3 −4.1 −2.6 −4.0 −2.8 −2.7
−2.7 N/A −2.7 −1.3 −2.1 −1.7 −2.7 N/A −2.2 −2.4 −3.0 −2.8
Source: Authors’ simulations with D-FARM.
agriculture leads to welfare losses in all regions (Table 10.15). Most negatively affected are regions with high population growth rates, in which agriculture represents a high proportion of total GDP, such as Africa and Other Asia; in these regions in 1995 crops and livestock accounted for 11.9 and 10.8% of GDP, respectively. Table 10.14. Cumulative percentage changes in land prices and usage relative to
baseline, 2007 Farmland
Cropland
Grazeland
Region
Price
Quantity
Price
Quantity
Price
Quantity
ANZ CAN USA JPN OEA SEA EU FSU OEU OAS LAM AFR
17.4 22.3 24.1 15.1 13.5 10.2 24.7 24.2 20.6 15.6 17.8 14.4
0.2 0.7 0.7 1.6 0.2 0.2 1.4 0.3 0.6 0.2 0.2 0.2
23.8 22.6 24.8 15.1 15.7 10.0 25.0 25.0 21.6 17.1 18.3 15.6
0.6 0.8 0.7 1.6 0.3 0.2 1.5 0.2 0.9 0.3 0.2 0.2
13.7 20.6 21.2 15.7 8.5 12.5 23.6 20.8 19.1 11.1 17.0 10.5
0.0 0.5 0.9 1.8 0.2 1.2 1.1 0.7 0.0 0.0 0.2 0.1
Source: Authors’ simulations with D-FARM.
Resource Use and Technological Progress in Agriculture
287
Table 10.15. Cumulative percentage changes in income, wages, and welfare relative to
baseline, 2007 Farmland Farm Change in Aggregate welfare effect ($US millions) income income Income Wagesa welfare (percent) Region (2) (3) (4) (5) (6) (7) ANZ CAN USA JPN OEA SEA EU FSU OEU OAS LAM AFR
17.7 23.2 25.0 17.0 13.8 10.5 26.4 24.6 21.3 15.9 18.0 14.5
8.3 10.2 13.0 8.0 6.3 5.6 9.7 11.1 7.6 7.0 7.3 6.5
0.1 −0.1 0.3 0.0 0.1 0.1 0.1 −0.3 −0.1 −0.9 −0.4 −0.5
0.3 0.1 0.3 0.0 0.3 −0.3 0.3 0.1 0.1 −0.5 −0.4 −0.7
−0.1 −0.2 0.0 −0.1 −0.5 −0.7 −0.1 −0.3 −0.3 −1.2 −0.6 −1.3
−351 −1279 2882 −5764 −9510 −3995 −4236 −1386 −3279 −18579 −9991 −7118
a Wages of unskilled labor. Source: Authors’ simulations with D-FARM.
Table 10.16. Cumulative percentage changes in timber harvest rates as a result of a
slowdown in agricultural TFP relative to baseline, 2007 Timber harvest rate on land types
ANZ CAN USA JPN OEA SEA EU FSU OEU OAS LAM AFR
1
2
3
4
5
6
1.2 0.7 1.4 N/A 0.4 N/A 1.5 1.5 0.5 0.8 0.7 0.6
1.3 1.7 1.9 N/A 0.5 N/A 1.9 1.4 6.5 1.1 0.6 0.6
1.8 1.7 3.1 1.8 1.6 1.3 2.6 2.6 2.0 2.0 2.3 2.0
1.5 2 1.9 1.4 1.8 1.4 2.5 2.2 2.3 2.1 1.9 2.2
2.3 N/A 2.1 1.2 1.7 1.0 2.2 2.3 2.0 1.3 2.1 1.8
2.7 N/A 2.2 1.4 1.5 1.1 2.5 N/A 2.2 1.7 2.1 1.5
Source: Authors’ simulations with D-FARM.
288
Elena I. Ianchovichina, Roy Darwin, and Robin Shoemaker
Our results also show that the environmental impacts of a slowdown in agricultural TFP are negative. The quantity of forestland demanded declines across most land classes and regions (Table 10.13). This supports the hypothesis that growth in agricultural TFP helps prevent the permanent conversion of forestland to other uses. In addition, timber harvest rates increase across all land classes with a slowdown in agricultural TFP (Table 10.16). This indicates that growth in agricultural TFP helps reduce environmental or ecological damages that forestry production generates on land remaining in forests. Thus, slower global growth of agricultural TFP would likely have adverse environmental effects because it is associated both with reductions in forestland and increases in environmental or ecological damages on remaining forestland.
5. Conclusions This chapter is unique in that it analyzes the global effects of economic and population growth and the impact of a slowdown in agricultural TFP growth on agricultural and forest resources using the D-FARM model. The modeling framework captures growth effects, productivity differences in land resources across regions, the split between skilled and unskilled labor, and the presence of sector-specific factor inputs. Given the current assumptions about growth trends in the world economy, we find that the world is not likely to face food shortages and that agricultural activities are likely to present little threat to forest areas over the next decade. In support of Evenson et al.’s findings, our results suggest that prices of agricultural commodities may rise and that forestland may be converted to farmland if growth in agricultural productivity falls to the rate of technological growth in the rest of the economy compared to the baseline. We expect the largest increases in agricultural crop prices to occur in South East Asia. Such losses in agricultural productivity would lead to welfare losses in all regions, with the bulk of the problems faced by regions in which agriculture still accounts for a higher share of the gross domestic product. Slower agricultural TFP growth could have negative environmental effects because it is associated both with reductions in forestland and increases in environmental or ecological damages on remaining forestlands. References Ahuja, V. and D. Filmer. 1995. Educational Attainment in Developing Countries: New Estimates and Projections Disaggregated by Gender. World Bank Policy Research Working Paper No. 1489. Washington, DC: World Bank.
Resource Use and Technological Progress in Agriculture
289
Alston, J., 1993. “Research Benefits in a Multimarket Setting: A Review.” Review of Marketing and Agricultural Economics 39, 29–52. Alston, J., P. Pardey, and V. Smith (eds.). 1999. Paying for Agricultural Productivity. Baltimore: John Hopkins University Press. Bernard, A. and C. Jones. 1993. Productivity across Industries and Countries: Time Series Theory and Evidence. Working Paper No. 93–17. Cambridge, MA: MIT Economics Department. CPB. 1999. WorldScan: The Core Version. The Hague: CPB Netherlands Bureau for Economic Policy Analysis. Darwin, R. F. 1999. “A FARMer’s View of the Richardian Approach to Measuring Agricultural Effects of Climatic Change.” Climate Change 41(3–4), 371–411. Darwin, R., M. Tsigas, J. Lewandrowski, and A. Raneses. 1995. World Agriculture and Climate Change: Economic Adaptations. Agricultural Economic Report No. 703. Washington, DC: U.S. Department of Agriculture, Economic Research Service. Available from http://www.ers.usda.gov/publications/aer703/aer703.pdf. Darwin, R., M. Tsigas, J. Lewandrowski, and A. Raneses. 1996. “Land Use and Cover in Ecological Economics.” Ecological Economics 17, 157–81. Edwards, G. and J. Freebairn. 1984. “The Gains from Research into Tradeable Commodities.” American Journal of Agricultural Economics 66, 41–9. Evenson, R. E., C. E. Pray, and M. W. Rosegrant. 1999. Agricultural Research and Productivity Growth in India. Research Report No. 109. Washington, DC: International Food Policy Research Institute. Fan, M. and Y. Zheng. 2000. “China’s Trade Liberalization for WTO Accession and Its Effects on China – A Computable General Equilibrium Analysis.” Unpublished manuscript. Francois, J. and A. Strutt. 1999, June. “Post Uruguay Round Tariff Vectors for GTAP v.4.” Unpublished memo. Frisvold, G. B. 1997. “Multimarket Effects of Agricultural Research with Technological Spillovers.” In T. W. Hertel (ed.), Global Trade Analysis Modeling and Applications (pp. 321–46). Cambridge: Cambridge University Press. Harrison, J. W. and K. R. Pearson. 1995. Solutions for Large General Equilibrium Models Using GEMPACK. Paper No. IP-64. Melbourne: Centre of Policy Studies and the IMPACT Project, Monash University. Hertel, T. W. (ed.). 1997. Global Trade Analysis Modeling and Applications. Cambridge: Cambridge University Press. Martin, W., B. Dimaranan, T. Hertel, and E. Ianchovichina. 2000. Trade Policy, Structural Change and China’s Trade Growth. Working Paper No. 64. Stanford: Center for Economic Research on Economic Development and Policy Reform, Stanford University. McDougall, R. A., A. Elbehri, and T. P. Truong. 1998. Global Trade Assistance and Protection: The GTAP 4 Data Base. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Norton, G. and J. Davis. 1981. “Evaluating Returns to Agricultural Research: A Review.” American Journal of Agricultural Economics 63, 685–99. World Bank. 1999. World Development Indicators. Geneva: World Bank.
ELEVEN
Global Economic Integration and Land-Use Change Alla Golub and Thomas W. Hertel
1. Introduction and Motivation Land-use change – in particular deforestation – has contributed substantially to current greenhouse gas concentrations in the atmosphere. In addition, agriculture and forestry have a potentially very significant role to play in stabilizing greenhouse gas concentrations. According to results from the EMF-21 (Energy Modeling Forum) study models, land-based mitigation strategies are a significant part of the mitigation portfolio required for a climate stabilization policy, accounting for anywhere from 18 to 72% of total abatement by 2050 and 15 to 44% of total abatement by 2100 (Rose et al. 2007). The efficiency of any strategy, however, depends critically on the baseline land-use changes. In turn, those changes depend on many factors, the most important of which are population and per capita income growth, as well as the degree of integration in the global economy. In this chapter, we modify and extend the GDyn model to investigate the role of population and per capita income growth and of global economic integration in determining long-run patterns of land-use change, where that change is decomposed by agro-ecological zone (AEZ). We are able to isolate the impact of each of these three factors on the development of land use in the long-run through a series of carefully designed simulations. Although the impact of population and income growth on the derived demand for land is relatively predictable, once supply-side constraints are brought to bear, the picture becomes more complex. International trade serves to moderate the changes in land rents across regions of the world, transmitting demand growth from the fast-growing, land-constrained countries (e.g., China) to the slower growing, land-abundant countries such as Australia and Original version published in Journal of Economic Integration, September, 23(3), 463–88, 2008.
290
Global Economic Integration and Land-Use Change
291
New Zealand. Trade is also shown to have a significant impact on the composition of land-using activities, inducing significant shifts between crops, livestock, and forestry uses. In our analysis we consider the period from 1997 to 2025. Although 28 years may not be a long enough time period in which to observe dramatic changes in climate, it is long enough for climate change policy to have a significant impact on the economy – and for changes in land use to have a significant impact on emissions. Therefore this analysis is highly relevant for those seeking insight into the drivers of land-use change over this intermediate time horizon. Examples of previous studies that have investigated the tradeoffs between different land-use decisions are Adams et al. (1996), using the Forest and Agriculture Sectors Model (FASOM), Darwin et al. (1995), using the Future Agricultural Resources Model (FARM), Chapter 10 using a dynamic extension of FARM, and Ahammad and Mi (2005) using the modified global trade and environment model (GTEM). The work presented in this chapter is quite similar to that of Chapter 10 and of Ahammad and Mi (2005) in that we incorporate land use based on AEZs into a recursive dynamic, general equilibrium model. Chapter 10 explores the implications of technological change in agriculture for the pattern of land use in a baseline simulation. Ahammad and Mi (2005) explore the implications of climate policy for land use. However, neither of these studies isolates the impacts of fundamental drivers of supply and demand underlying their baseline projections of land use. Therein lies a key contribution of this chapter. In addition, we introduce an econometrically estimated consumer demand system, which is aimed at capturing the changing patterns of consumer purchases as incomes rise. Taken together, these extensions of traditional global general equilibrium analysis allow us to go much further in understanding the impact of global economic growth and integration on land-use change.
2. Methodology 2.1 Modeling Approach For this chapter, we build on the earlier paper by Golub, Hertel, and Sohngen (2007). The authors undertook several modifications to the GDyn model (outlined in Chapter 2) to capture the most important determinants of supply and demand for land and to facilitate long-run projections of the sort desired for climate change policy analysis. Here we briefly describe
292
Alla Golub and Thomas W. Hertel
those that are most important for the issue of global economic integration and land-use change. The most important driver of the demand for land is consumer demand because it influences the scale and location of each production activity. In Golub, Hertel, and Sohngen (2007), consumer demand is modeled with an implicit directly additive demand system (AIDADS) developed by Rimmer and Powell (1996). The AIDADS demand system is rank 3, meaning that it is very flexible in its ability to represent the nonhomothetic demand for consumer goods, which is especially important for fast-growing developing countries that account for an increasing share of global economic growth. Furthermore, it has been shown to outperform competing demand systems in the prediction of observed demands – particularly the demand for food – across a wide range of income levels (Cranfield et al. 2003). In the AIDADS system, the predicted budget share is the sum of subsistence and discretionary budget shares: I n − p n γ p kn γk αk + βk exp(Un ) + , (11.1) s kn = In 1 + exp(Un ) In where s kn is the average budget share spent on good k in country n, pkn is the price of good k in country n, In is total per capita expenditures in country n, Un is per capita utility in country n, γ k is an estimated parameter reflecting the subsistence level of good k, and p n γ is minimally sustainable per capita income in country n. Parameters αk and βk represent bounds on marginal budget shares at very low (i.e., close to subsistence) and very high income levels, respectively. The AIDADS parameters estimated by Reimer and Hertel (2004) using the GTAP 5 Data Base, representing the world economy in 1997, are reported in Table 11.1. The estimation results are consistent with one’s intuition regarding how the composition of consumption is likely to differ across income levels. The estimated subsistence levels γ k for “Meat, dairy, fish,” “Utilities, other housing services,” and “Transport, communication” are zeros, implying that these three categories are not necessary for survival. In contrast, subsistence level for “Grains, other crops” is 0.298, highlighting the importance of grains and crops at the lowest income levels. The estimated lower (αk ) and upper (βk ) bounds on marginal budget shares reported in Table 11.1 also make sense. At the low income levels, 8.4 and 12.2 cents of each additional dollar are spent on “Grains, other crops” and “Meat, dairy, fish,” respectively. The upper bound on marginal budget share, βk , for “Grains, other crops” is zero, implying that expenditures on this consumption category are not increasing at high income levels. As per
Global Economic Integration and Land-Use Change
293
Table 11.1. Parameter estimates for AIDADS demand system Consumed good Grains, other crops (Crops) Meat, dairy, fish (MeatDairy) Processed food, beverages, tobacco (OthFoodBev) Textiles, apparel, footwear (TextAppar) Utilities, other housing services (HousUtils) Wholesale/retail trade (WRTrade) Manufactures, electronics (Mnfcs) Transport, communication (TransComm) Financial and business services (FinService) Housing, education, health, public services (HousOthServ)
γk
αk
βk
0.298 0.000 0.142 0.030 0.000 0.078 0.002 0.000 0.014 0.086
0.084 0.122 0.138 0.068 0.035 0.132 0.169 0.115 0.030 0.108
0.000 0.026 0.032 0.030 0.047 0.238 0.099 0.097 0.118 0.313
Source: Reimer and Hertel (2004), based cross-section estimation using the GTAP 5 Data Base.
capita income grows, marginal budget shares decline for food categories, as well as for “Textiles, apparel, footwear,” “Manufactures, electronics,” and “Transport, communication,” whereas they grow for “Utilities, other housing services,” “Wholesale/retail trade,” “Financial and business services,” and “Housing, education, health, public services” – categories that are generally viewed as luxuries at low income levels. To implement AIDADS in our model, we regionalize the parameters in Table 11.1 to allow this demand system to reproduce the observed, base period budget shares, while retaining the overall pattern of the estimated Engel effects.1 These base budget shares for the crops, meat and dairy products, processed food, beverages, and tobacco, and all other goods (aggregated for purposes of this table only) are reported in Table 11.2 – at producer prices (i.e., wholesale/retail margins are in the “other” category). Our choice of regional aggregation scheme is driven by our focus on the derived demand for land due to income growth. The 78 regions of the GTAP 5.4 Data Base (Dimaranan and McDougall 2002) are aggregated to 11 regions according to the mapping reported in the Appendix of this chapter. This aggregation, although parsimonious, represents a broad spectrum of income levels and development across regions. In the poorest countries it is clear that food expenditures still command a large share of per capita income (45% in South Asia), and much of this comes in the form of purchases of land-intensive goods (crops and meats/ dairy). In contrast, food’s share in the total budget is quite small in the 1
This involves rescaling the αk and βk parameters for each good, while preserving their ratio. Details are available in Golub (2006).
294 Table 11.2. Budget shares for consumed goods grouped in four categories Consumed good
Scenario/year
ANZ
China HYAsia ASEAN SAsia
Crops
1997 benchmark 0.0095 0.2029 0.0263 2025, PE projections 0.0059 0.0924 0.0137 2025 baseline projections 0.0060 0.1025 0.0141
0.0879 0.2473 0.0058 0.0492 0.0136 0.0529 0.0876 0.1008 0.0626 0.1907 0.0038 0.0422 0.0081 0.0325 0.0706 0.0802 0.0720 0.2132 0.0038 0.0433 0.0080 0.0315 0.0715 0.0841
Meat & dairy
1997 benchmark 0.0318 0.1570 0.0220 2025, PE projections 0.0253 0.1764 0.0166 2025 baseline projections 0.0255 0.1756 0.0168
0.0744 0.1166 0.0195 0.0821 0.0508 0.1318 0.0838 0.0852 0.0696 0.1370 0.0163 0.0766 0.0400 0.1105 0.0747 0.0874 0.0694 0.1346 0.0164 0.0770 0.0403 0.1115 0.0751 0.0874
Processed food, beverages, 1997 benchmark 0.0887 0.1228 0.0850 and tobacco (OthFoodBev) 2025, PE projections 0.0710 0.0805 0.0649 2025 baseline projections 0.0715 0.0799 0.0655
0.1237 0.0832 0.0455 0.1269 0.0629 0.1266 0.1235 0.1394 0.1071 0.0429 0.0382 0.1175 0.0498 0.1015 0.1082 0.1309 0.1074 0.0427 0.0384 0.1182 0.0501 0.1015 0.1081 0.1297
All other goods
0.7141 0.5529 0.9291 0.7418 0.8727 0.6887 0.7051 0.6746 0.7607 0.6293 0.9417 0.7637 0.9020 0.7555 0.7465 0.7014 0.7512 0.6095 0.9415 0.7615 0.9016 0.7556 0.7453 0.6988
1997 benchmark 0.8700 0.5173 0.8667 2025, PE projections 0.8978 0.6507 0.9048 2025 baseline projections 0.8971 0.6420 0.9036
Source: Base data, GTAP 5.4 Data Base and authors’ simulations.
NAM
LAM
WEU
EIT
MENA
SSA
Global Economic Integration and Land-Use Change
295
richest countries (less than 10% of income, when expenditures are measured at producer prices), and most of this represents expenditure on highly processed goods. Thus the geographic pattern of global income growth is very important in determining the global demand for land in agriculture. Rapid income growth in South Asia and China will have a much more significant impact on the derived demand for land than will equivalent growth in North America or Europe. To reflect the fact that land is a heterogeneous endowment, Golub, Hertel, and Sohngen (2007) bring in climatic and agronomic information by introducing AEZs (Lee et al. 2005). This database enhances the standard GTAP Data Base by disaggregating land endowments into AEZs.2 The AEZs represent six different lengths of growing period (6 × 60-day intervals). The concept “length of growing period” refers to the number of days within the year of temperatures above 5◦ C when moisture conditions are considered adequate for crop production. This approach evaluates the suitability of each AEZ for the production of crops, livestock, and forestry based on currently observed practices, so that the competition for land within a given AEZ across uses is constrained to include activities that have been historically observed to take place in that AEZ.3 Indeed, if two uses (e.g., citrus groves and wheat) do not presently appear in the same AEZ, then they will not compete in the land market. Тhe different AEZs then enter as inputs into national production function for each land-using sector (e.g., wheat). With a sufficiently high elasticity of substitution in use,4 the returns to land across AEZs, but within a given use, will move closely together.5 Of course, there are many other factors beyond agronomic ones that limit land mobility within an AEZ. These include costs of conversion, managerial inertia, and unmeasured benefits from crop rotation. Therefore, we further limit the potential for movement of land from one use to another within an AEZ via the constant elasticity of transformation (CET) frontier. In the model, the allocation of land is determined through a nested CET, multistage optimization structure (Ahammad and Mi 2005). Owners of the particular type of land (AEZ) first decide on the allocation of land between agriculture 2
3 4 5
Lee et al. (2005) adopt the FAO/IIASA convention of agro-ecological zoning. Agroecological zoning refers to segmentation of a parcel of land into smaller units according to agro-ecological characteristics (e.g., moisture and temperature regimes, soil type, landform, etc). In other words, each zone has a similar combination of constraints and potentials for land use. In this work AEZs are static. However, they may shift due to climate change. The elasticity of substitution among AEZs in production is set to 20. Information on the relative importance of each AEZ in each of 11 regions considered in this study and the importance of each sector within AEZ can be found in Golub et al. (2007).
296
Alla Golub and Thomas W. Hertel
and forestry to maximize the total returns from land. Then, based on the return to land in crop production, relative to the return on land used in ruminant livestock production, the landowner decides on the allocation of land between these two broad types of agricultural activities. Golub, Hertel, and Sohngen (2007) set the elasticities of transformation among uses based on econometric evidence on the responsiveness of land use to changes in land rental differentials.
2.2 Baseline Assumptions We adopt the baseline assumptions developed by Hertel et al. (2006). The starting point of our simulation is the world economy in 1997, as depicted in the GTAP 5.4 Data Base (Dimaranan and McDougall 2002). In our simulations from 1997 to 2025, labor force, population, and productivity growth are all exogenous to the model. Projections of labor force (skilled and unskilled labor) growth rates for 1998–2025 are taken from Chapter 5. The historical real GDP and population growth rates for the 1998–2004 period are constructed using the World Development Indicators database. The real GDP path for 2005 −25 is driven by our assumptions about productivity growth in various sectors of the economy. Productivity growth rates in nonland-using sectors are based on our assumptions about economy-wide labor productivity growth in each region, adjusted for productivity differences across sectors using estimates reported in Kets and Lejour (2003). For a detailed description of the productivity growth in non-land-using sectors the reader is referred to Hertel et al. (2006). Productivity growth rates in agriculture, which play a key role in determining the location of future agricultural production, are based on Ludena et al. (2007). Those authors project rapid catching up in nonruminant productivity, whereas divergence (more rapid growth in productivity in the rich countries) is more common for ruminant livestock. In the case of crop productivity growth, the industrialized economies show continued strong growth, as is the case for China and South Asia. East and Southeast Asia are projected to continue a recent trend of negligible productivity growth in agriculture. In the absence of better information, productivity growth rates in forestry are assumed to be equal to the average of productivity growth rates in crops and ruminants, weighted by the share of their output in total output of crops and ruminants. This is a “neutral” assumption that does not have an effect on the allocation of land between agriculture and forestry. Productivity growth in forestry processing is predicted based on results from the global timber model (Sohngen and Mendelsohn 2006).
Global Economic Integration and Land-Use Change
297
2.3 Simulation Design The channels through which global economic growth and integration affect global patterns of land use are varied and complex. To isolate these channels on an individual basis, we conduct a sequence of simulations that may, in turn be contrasted with the baseline simulation to gain insight into these mechanisms. The first two simulations focus on the demand side of global economic growth. In so doing, they abstract from supply-side considerations. This is accomplished by fixing primary factor prices in each region, allowing supplies to adjust freely to meet any level of demand required.6 These demand-side forces are separated into two component parts: population growth at constant per capita income and per capita income growth at constant population. The first simulation, that involving the population shock, permits us to explore the impact of differential population growth across the globe on the demand for land-using commodities, and hence the demand for land. In the second demand-side simulation, we shock per capita utility, reflecting the growth in real income anticipated by the baseline scenario. With real income rising, the budget share equation (11.1) dictates how the pattern of household expenditure will change, thereby shifting the demand for commodities, and hence the derived demand for land, in each region. In the relatively wealthy regions, where the income elasticities of demand for food products are quite low, we expect stronger increases in the derived demand for land in the forest products sector. In contrast, in the poorest countries, where the income elasticity of demand for food is still high, strong income growth may generate substantial increases in food demand, and hence the demand for land used in agriculture. The third simulation represents our baseline scenario. As such, it shares the same population and per capita income growth with scenarios one and two, but it now brings to bear the supply-side constraints in the economy. Specifically, we fix land endowment and restrict the supply of accumulable capital according to the theory of the model, which, coupled with exogenous projections of the skilled and unskilled labor forces in each region, permits us to endogenize endowment prices.7 By bringing the supply side into the 6 7
In these two simulations the endowment prices are fixed and GE links are broken. The walraslack variable is swapped with the numeraire psavewld. Golub, Hertel, and Sohngen (2007) included the possibility of accessing new lands and found that the potential for accessing new forest lands plays a small role in dampening the growth in global land rents. For this reason and for simplicity of the exposition, in this chapter we have not incorporated the potential to convert inaccessible forests to commercial forests or agricultural lands. However, it is important to note that, although accessing new forest lands plays a small role in dampening the growth in global land rents,
298
Alla Golub and Thomas W. Hertel
picture, we reflect the fundamental scarcity of global land endowments, thereby requiring land rents to adjust to achieve a global general equilibrium. This has a strong impact on the pattern of land use by region and AEZ. Up to this point the simulations all focus on the impact of economic growth on the derived demand for land. However, we also believe that global integration, in the form of increased trade, can be an important determinant of the pattern of global land use. In the baseline, imports and exports adjust to equilibrate the global demand and supply of land. Thus, as we will see, rapidly growing, land-scarce regions, such as China, tend to increase imports of land-intensive crops – thereby effectively “importing land use.” In contrast, land-abundant, slower growing economies, such as Australia and New Zealand, will tend to “export” land use. Yet how important is this interregional arbitrage in the services of land? Would land use look very different in a global economy in which agricultural and forestry products were largely nontraded? We explore this issue in our final simulation. Because it is not possible to solve our global general equilibrium model in the complete absence of trade, we adopt a more modest restriction on trade flows. In particular, we eliminate the possibility of firms and households substituting imported for domestic goods in response to domestic scarcity. Thus, if 95% of the wheat consumed by Chinese households is currently domestic in origin, that ratio must be preserved over the course of the baseline simulation under the model specification adopted in our final simulation. Specifically, this final stylized simulation (we refer to it as the restricted trade simulation) corresponds to a baseline simulation with zero trade elasticities. The population and productivity growth assumptions are identical across the two simulations. By comparing baseline and alternative simulations, we are able to assess the role of globalization in land-use change by the year 2025.
3. Results8 Changes in demand for land projected based on the first two stylized scenarios are reported in Table 11.3. This discussion begins with the top panel of Table 11.3.
8
the deforestation in tropics has contributed to 15–25% of annual global greenhouse gas emissions (Gibbs et al. 2007). Unlike other applications in this volume, here we are not looking at several policies and comparing them against the baseline. Rather the simulations are designed to isolate impacts of the factors affecting the development of land use in the long-run.
Global Economic Integration and Land-Use Change
299
Table 11.3. Cumulative growth in the demand for land from 1997 to 2025, percent Variable
ANZ
China HYAsia ASEAN
SAsia
NAM LAM
Impact of Population Growth on the Demand for Land Population 22.99 19.65 2.8 40 46.33 22.74 40.8 growth Total change in demand for land, by AEZ AEZ1 21.9 15.1 – – AEZ2 22.2 17.2 – – AEZ3 22.5 17.5 3.8 – AEZ4 22.2 17.6 3.5 33.9 AEZ5 19.6 17.7 3.4 33.7 AEZ6 17.7 16.8 3.6 31.1
43.4 43.4 43.1 42.8 41.8 42.1
21.6 21.5 20.2 21.1 20.8 19.9
35.2 34.7 34.7 34.8 34.6 34.5
WEU
EIT
MENA
SSA
−0.74
−3.27
58.63
78.56
1.7 3.5 3.1 2.7 2.6 2.5
1.9 −0.7 –0.6 −0.5 −0.7 0.5
49.0 51.3 46.6 49.4 50.4 –
64.8 65.7 66.7 63.9 65.9 62.5
Impact of Income Growth on the Demand for Land Utility 82.72 784.22 126.51 125.74 237.93 64.94 43.16
93.23 177.3
74.88
51.99
Total change in demand for land, by AEZ AEZ1 60.7 288.4 – – AEZ2 60.6 290.5 – – AEZ3 60.5 290.5 46.0 – AEZ4 61.5 290.1 39.4 67.9 AEZ5 68.0 289.1 35.8 68.6 AEZ6 72.1 282.4 40.5 71.1
49.4 66.8 56.2 49.0 48.4 46.2
53.7 48.2 56.2 50.4 48.3 –
50.1 46.0 35.6 39.4 36.5 45.8
141.0 139.8 130.5 134.1 139.1 134.6
44.3 44.0 49.1 45.1 46.3 50.3
35.8 36.6 36.6 35.6 36.5 36.0
111.1 103.3 102.0 102.6 103.3 106.0
Note: “–” indicates that a specific AEZ is not present in a region. Source: Authors’ simulations.
3.1 Population Growth The first row reports the cumulative population changes by region from 1997 to 2025. These population growth rates are the highest in Sub-Saharan Africa and the Middle East and North Africa, followed by South Asia. With perfectly elastic supplies of endowments, the changes in population are translated into changes in demand for land. Within a region, the changes in demand for land across AEZs are very similar and follow closely the population growth rates. They deviate from the population growth rates because of the presence of intermediate inputs and international trade – both of which break the direct link from consumer demand for food and forestry products to the derived demand for land. The largest increases in land demand are observed in the fast-growing population regions: Africa and the Middle East, as well as South Asia.
3.2 Growth in Real Income Now turn to the second panel of Table 11.3, which shows changes in demand for land when we hold population constant and simply perturb per capita
300
Alla Golub and Thomas W. Hertel
utility by the cumulative growth rate observed over the baseline simulation. As with the first simulation, we render the endowments perfectly elastic such that the prices of these endowments – and hence the prices of the produced goods and services – do not change. The fastest growth in per capita utility is achieved in rapidly growing China, followed by South Asia. Recall from equation (11.1) that increased utility causes households to move their discretionary spending away from the lower bound (αk ), toward the upper (βk ). This results in a decline in the share of spending on food and a rise in the share of spending on nonfood products (including forest products). The 2025 budget shares resulting from this partial equilibrium simulation are reported in the second row of Table 11.2 and may be compared with the base period shares immediately above them. For example, the crops budget share in China is more than cut in half, whereas the livestock/dairy budget share rises slightly. Returning to the second panel of Table 11.3, we see that, with perfectly elastic supply, land employed in production would expand the most in the low-income, fast-growing regions; there is nearly a 300% cumulative growth in land requirements in the case of China. Note, however, that this growth is now far less rapid than the overall growth in demand (change in utility in the second panel of Table 11.3), simply because food’s overall share in consumers’ budgets is falling (Table 11.2). Now let us look more closely at the drivers of land-use change in our baseline simulation. In contrast to population growth and income growth scenarios, in our baseline the land endowment is fixed, and total changes in land used in commercial production in each AEZ and region are zeros. However, there are important changes in land use within the land already employed in production. Table 11.4 reports changes in consumer demand for crops, livestock, and forestry. As population and incomes rise, consumer demands for crops, ruminants, and forestry products also rise in all regions, with the strongest increases in China, followed by South Asia (Table 11.4). Among the three sectors of crops, ruminants, and forestry, the strongest growth in demand is predicted for forestry products, which reflects rising demands for furniture, construction, and paper products. The production patterns in the baseline scenario are also reported in Table 11.4. The differential growth rates in consumption and production serve to highlight the importance of intermediate inputs and international trade for these land-using sectors. For example, demand for crops in China grows rapidly, whereas production expands much more slowly in the baseline scenario, with the difference being accommodated through
Global Economic Integration and Land-Use Change
301
Table 11.4. Cumulative growth rates in consumption and production of crops, ruminants,
and forestry, from 1997 to 2025 Sector
Scenario ANZ China HYAsia ASEAN SAsia NAM LAM WEU EIT MENA SSA
Crops
baseline restricted trade Ruminants baseline restricted trade Forestry baseline restricted trade
Consumption 125 28 104 47
29 54
181 130
9 −13
61 17
60 71
883 707
77 77
105 89
199 153
120 112
1243 1214
136 124
207 197
471 464
baseline 393 restricted 148 trade Ruminants baseline 82 restricted 89 trade Forestry baseline −14 restricted 20 trade
89 184
−27 14
633 436 72 156
Crops
56 69
28 34
78 81
107 89
99 104
91 97
109 115
55 55
169 163
144 105
120 115
106 105
90 86
90 91
188 191
169 168
176 175
3 42
Production 111 248 139 125
146 106
131 65
95 94
34 91
139 112
38 64
45 68
201 168
104 108
142 148
41 38
142 134
50 97
128 128
21 5
74 27
344 339
−7 9
69 52
34 1
96 20
45 64
141 87
Source: Authors’ simulations.
international trade. At the same time, the opposite is observed in the Americas. These changes in regional trade balances, by commodity, for the baseline scenario are reported in the top panel of Table 11.5. Here, we see that China is expected to increase its annual net imports of crops by about $275 billion by the end of the baseline projections period. ASEAN, South Asia, High Income Asia, and Middle East and North Africa also increase their net imports of crops. When combined, this results in a very substantial increase in net crop export requirements from the rest of the world. The largest share of this increase in crops is supplied by North America, followed by Latin America and Europe. North America and Western Europe are two regions with high rates of technological progress in crops, low population growth rates, relatively low per capita income growth rates, and low income elasticities of demand for food. All these factors support the expansion of crops exports from these two regions. The direction of change in the sectoral trade balances for ruminants and forestry is similar to that for crops, but of much smaller magnitude because
302
Alla Golub and Thomas W. Hertel
Table 11.5. Change in trade balance, by sector ($US billions), from 1997 to 2025 ANZ China HYAsia ASEAN SAsia NAM LAM WEU Crops 48 −275 −53 Ruminants 3 0 −1 NonRuminants −1 −4 −7 PrFood 13 −143 −148 Forestry 0 −32 3 NatResources 22 −352 −12 Mnfcing −37 901 −733 Trans/Comm 6 132 −49 OthSvces 26 −350 −334 Total 80 −124 −1334 Crops Ruminants NonRuminants PrFood Forestry NatResources Mnfcing Trans/Comm OthSvces Total
8 −13 1 0 −1 −2 10 −21 0 −1 12 −107 3 69 14 59 13 −108 60 −125
−8 0 1 −36 2 −85 −559 −28 −122 −835
−66 −5 −1 −13 4 −80 −179 −20 407 46 −13 −1 −1 −30 −2 −70 −31 3 237 92
EIT MENA SSA Total
Baseline −112 263 77 78 −6 −42 30 −57 −1 9 2 −3 2 −8 0 −1 −5 −24 103 −33 −24 −10 −12 −18 −13 167 151 −7 −9 −80 20 −64 −6 −3 4 11 8 −1 9 −4 −178 −31 4 0 213 322 26 −66 222 418 −185 −574 −78 −159 −43 −449 108 178 −1 115 191 24 −9 676 71 246 259 −57 −314 −45 73 −17 87 1222 414 −470 −17 0 96 0 Restricted trade −9 44 2 0 3 1 0 −3 9 −7 59 94 −2 −1 1 −109 −69 −2 144 698 25 89 166 14 46 99 183 151 997 327
6 −10 −15 −3 0 −1 −3 −1 0 −70 −22 −18 3 0 0 −19 129 265 −486 −120 −197 90 94 2 −175 −121 −71 −656 −51 −35
−4 −12 0 0 0 −2 17 −25 −1 0 21 −35 38 −414 −5 498 8 −10 75 0
Source: Authors’ simulations.
of their lesser importance in global trade. High Income Asia, ASEAN, South Asia, Middle East and North Africa, as well as Europe, expand their net imports of ruminants. This expansion is satisfied by exports from Americas and Australia/New Zealand. Rapidly growing demand for forestry in China and South Asia results in increasing imports of forestry by these regions. This imbalance in forestry trade is met by increasing exports from Western Europe, Sub Saharan Africa, and the natural resource-rich Economies in Transition. Now let us turn to our alternative simulations in which we restrict the degree of global economic integration. Results from the restricted trade scenario are reported in the lower panel of Table 11.5. In this simulation imported and domestic use of each commodity grow in fixed proportion. That is, China and South Asia cannot increase their absorption of imported crops without increasing their absorption of domestically produced crops – in effect, the extent of global economic integration of goods and services is frozen at base period levels. Before considering these entries
Global Economic Integration and Land-Use Change
303
Table 11.6. Impact of global economic integration on annual growth rates in land rents
from 1997 to 2025, percent Land-using sector Scenario ANZ China HYAsia ASEAN SAsia NAM LAM WEU EIT MENA SSA All sectors
Baseline 7.08 restricted 3.09 trade
5.41 9.13
−0.74 3.73
2.64 6.24
6.96 7.90
6.02 2.79
5.11 3.47
2.79 2.81 0.86 2.12
2.53 5.13
5.35 3.83
Source: Authors’ simulations.
in detail, return to Table 11.4, which reports consumption and production increases under the restricted trade scenario. Now crop consumption in China grows much less rapidly (130% vs. 181% cumulative growth), whereas production must grow much more rapidly (184% vs. 89% cumulative growth) than under the baseline scenario to satisfy domestic demand in the absence of increased economic integration. The mirror image of this result is shown in the column for Australia/New Zealand, where consumption is now higher and production growth far lower than under the baseline scenario. Freezing the extent of global economic integration has significant consequences for the patterns of consumption and production of land-using commodities. The changes in trade balance by commodity reported at the bottom of Table 11.5 underscore these effects. In the restricted trade scenario, the increases in net imports of crops by China, South Asia, ASEAN, and High Income Asia are much smaller. Indeed, their combined trade balance for crops is now $463 billion higher compared to the baseline scenario. This, in turn, reduces the net export requirements on the part of the Americas, Western Europe, and Australia and New Zealand. Similarly, ruminant, forestry, and food imports into Asia are much lower. When importers cannot substitute away from more expensive domestic production toward imports, domestic production in Asia must expand, and consumption must be reduced (through higher prices). What is the effect of the global economic integration on the land market? The most direct measure to consider is the change in aggregate land rents. These figures are reported in Table 11.6 as average annual growth rates in average land rents across the agriculture and forestry sectors. These growth rates vary by region, and they only reflect conditions in agriculture and forestry. In the baseline scenario, pressure on land rents is very high in the agriculture- and forestry-exporting countries, as they seek to fill the gap between consumption and production of land-intensive goods in the rapidly growing, densely populated markets – particularly in Asia. In
304
Alla Golub and Thomas W. Hertel
Australia and New Zealand the average across-sectors annual growth rate in land rents is 7% over the projections period. The growth rates in land rents in the Americas are somewhat slower, but still very high, and these growth rates rival those in the rapidly growing economies of China and South Asia themselves. In the slower growing, more mature economies of High Income Asia, where the income responsiveness of consumer demand for food is weak and agricultural productivity growth rates are very low, land rents are declining, signaling a strong incentive to convert these lands to other uses (although we do not model this possibility). The second row of results in Table 11.6 explores the impact of freezing the degree of economic integration in the commodity markets at base period levels. Now there is a much greater divergence in land rents between the net exporting and net importing regions. Aggregate land rental rates in Australia/New Zealand rise by 3%/year, and growth is even slower in North America. In contrast, China’s land rents grow at the brisk pace of more than 9%/year, as local production is called on to satisfy most of the growth in domestic demand. Land rents also rise more strongly in High Income Asia, because imports can no longer be used to offset potential declines in inefficient domestic production. Clearly, trade is a very important force influencing land markets.
3.3 A Detailed Look at Land Use Having ascertained that global economic integration through trade has a very large impact on the aggregate demand for land, it is interesting to take a closer look at the pattern of land use induced by this form of globalization. We do so by examining land use by AEZ, as reported in Table 11.7. Entries in this table are the percentage change in land use in each category, weighted by the economic importance of that category (i.e., land rents) in a given AEZ. Thus, the changes across uses (agriculture vs. forestry) within a given AEZ are directly comparable. We focus here on the tradeoff between land in agriculture vs. land in forestry, because that is the most important distinction from the point of view of net greenhouse gas emissions.9 Consider the first column in the top panel of Table 11.7. This reports the rental-share-weighted percentage change in land use in agriculture and 9
The movement of land from forestry into agriculture accounts for a large share of the accumulated CO2 concentrations in the atmosphere. Meanwhile, increasing land cover in forestry offers potential carbon sequestration benefits (Sohngen et al. 2009).
Global Economic Integration and Land-Use Change
305
Table 11.7. Revenue share-weighted cumulative growth rates in demand for land by AEZ
from 1997 to 2025 Agro-ecological zones
ANZ
China HYAsia ASEAN SAsia
NAM
LAM WEU
EIT
MENA SSA
Baseline Agriculture AEZ1 AEZ2 AEZ3 AEZ4 AEZ5 AEZ6 Forestry AEZ1 AEZ2 AEZ3 AEZ4 AEZ5 AEZ6
0.00 0.00 0.20 2.66 20.39 27.14
0.00 0.03 −0.08 −0.13 −0.38 −0.87
−8.90 −6.48 −3.58 −6.69
0.00 0.00 0.00 −0.03 −0.20 0.08 −2.59 0.13 −16.94 0.38 −21.34 0.87
0.00 0.00 9.77 6.93 3.71 7.18
−0.17 −0.10 −0.32 −0.34 −0.62 −0.52 −1.48 −2.06 −1.33 0.00 0.00 0.00 0.34 0.53 2.10
1.37 1.60 9.66 3.60 5.66 11.59
1.51 3.41 3.33 1.57 3.32 2.35
0.00 5.17 5.39 2.82 2.52 1.24
−4.56 −4.05 −4.16 −4.76 −3.63 −7.13
−1.68 −1.06 −2.45 −1.93 −1.58
−1.64 −1.21 −0.32 −0.60 −0.42 −0.99
0.17 −1.36 0.10 −1.58 0.32 −8.81 0.62 −3.48 1.50 −5.36 1.34 −10.38
−1.49 −3.30 −3.22 −1.55 −3.21 −2.30
0.00 −4.92 −5.11 −2.74 −2.45 −1.23
4.78 4.22 4.34 5.00 3.77 7.68
1.71 1.07 2.51 1.96 1.61 0.00
1.67 1.23 0.32 0.60 0.42 1.00
Restricted Trade Scenario Agriculture AEZ1 AEZ2 AEZ3 AEZ4 AEZ5 AEZ6 Forestry AEZ1 AEZ2 AEZ3 AEZ4 AEZ5 AEZ6
0.00 0.00 0.13 1.66 10.28 12.11
0.00 0.05 0.08 0.10 0.15 0.35
1.38 0.88 0.44 0.92
0.00 0.00 −0.13 −1.63 −9.32 −10.81
0.00 −0.05 −0.08 −0.10 −0.15 −0.35
0.00 0.00 −1.36 −0.88 −0.44 −0.92
0.82 1.32 6.14
0.18 0.11 0.27 0.55 1.31 1.12
1.02 1.12 6.19 2.42 3.74 7.17
1.37 3.05 2.98 1.40 2.96 2.09
0.00 7.95 6.18 3.08 2.74 1.35
5.06 2.31 2.12 2.54 2.07 4.98
0.98 0.50 1.17 0.81 0.63
2.15 1.81 0.72 1.64 0.99 2.21
0.00 0.00 0.00 −0.81 −1.30 −5.79
−0.18 −0.11 −0.27 −0.55 −1.29 −1.11
−1.00 −1.10 −5.83 −2.36 −3.60 −6.69
−1.36 −2.96 −2.90 −1.38 −2.88 −2.04
0.00 −7.37 −5.82 −2.99 −2.67 −1.34
−4.82 −2.26 −2.07 −2.48 −2.03 −4.75
−0.97 −0.50 −1.15 −0.80 −0.63 0.00
−2.11 −1.78 −0.71 −1.61 −0.98 −2.16
Source: Authors’ simulations.
forestry in Australia and New Zealand, by AEZ, in the baseline simulation. Clearly there is a strong movement of land from commercial forestry into agriculture. In High Income Asia, low productivity growth in agriculture (Ludena et al. 2007) results in a shift in land into forestry production. In the Americas and Western Europe there is a fairly substantial shift of commercial activity from forestry to agriculture. The opposite is the case in the Economies in Transition, Middle East and North Africa, and SubSaharan Africa regions.
306
Alla Golub and Thomas W. Hertel
-10.54 (minimum) 0.00 0.53 (median) 2.76 12.42 (maximum)
Figure 11.1. Impact of global trade integration on land-use change: Revenue-shareweighted changes in land used in forestry, percent (difference between baseline and restricted trade scenarios). Source: Authors’ simulations.
The lower panel of Table 11.7 shows the impact on land use under the restricted trade scenario, whereby the extent of global economic integration in the goods and services markets is frozen at base period levels. The first thing to note is that these numbers are quite different – varying both in magnitude and in sign. This is hardly a surprise, given the significant difference in aggregate land rents reported in Table 11.6. Several distinct patterns are evidenced in a comparison of the two sets of results in Table 11.7. First, by limiting trade opportunities, the high-productivity agriculture, Asia-exporting regions of Australia/New Zealand and North America experience a much smaller expansion in agricultural area. This is accompanied by smaller reductions in commercial forestry activity. In contrast, the fast-growing Asian economies are forced to devote more land to agriculture. This leads to numerous sign reversals, with country/AEZs now expanding agricultural activity rather than contracting. The same is true for the Middle East and Africa (MENA and SSA). In Western Europe, the no-trade scenario marginally accentuates the movement of land from forestry to agriculture, as Europe is forced to satisfy more of its own demands for food. Yet there is a strong reversal in the direction of land movement between forestry and agriculture in Eastern Europe and the Commonwealth of Independent States region (EIT). Figures 11.1 and 11.2 offer a visual summary of the differences between the top and bottom panels of Table 11.7 for forestry and agriculture, respectively. In particular, these figures map the deviation from the baseline in
Global Economic Integration and Land-Use Change
307
-12.11 (minimum) -2.78 -0.53 (median) 0.00 15.02 (maximum)
Figure 11.2. Impact of global trade integration on land-use change: Revenue-shareweighted changes in land used in agriculture, percent (difference between baseline and restricted trade scenarios). Source: Authors’ simulations.
cumulative land-use change, by region/AEZ, due to trade. For example, in the case of land use for commercial forestry production in AEZ6 of the Australia/New Zealand region, the baseline rental-share-weighted percentage change is −21.34% (Table 11.7, top panel). Deducting the restricted trade outcome (−10.81%) from the baseline value, we obtain −10.54%, which is the minimum entry in Figure 11.1, appropriately shading the southeastern corner of Australia and much of New Zealand. From this figure it is clear that global economic integration through international trade leads to a substantial shift in forestry activity from Australia/New Zealand and the Americas into Northern Europe, Central Asia, and even parts of Africa. In contrast, Figure 11.2 shows that increased international economic integration through trade leads to strong increases in agricultural activity in Australia/New Zealand and the Americas, with the largest reductions coming in Asia. In short, global economic integration has a very important impact on the pattern of land use around the world.
4. Summary This chapter aims to shed light on the role of global economic integration in land-use change. We do so by using a dynamic general equilibrium model that has been modified to incorporate the most important economic features driving global land demand and supply: an econometrically
308
Alla Golub and Thomas W. Hertel
estimated, international demand system for commodities and a new database and modeling framework that characterizes land use by agroecological zone. We use this model to simulate a baseline period from 1997 to 2025 during which land rents worldwide rise sharply and the global allocation of land between agriculture and forestry changes rather significantly in some regions. Through a series of restricted simulations of the model, we are able to isolate the impact on land markets of the following elements of growth and globalization: population growth, real income growth, and international trade. Of the two demand-side factors, real income growth is shown to be the more important. International trade plays a very substantial role in mediating between the land-abundant, slower growing economies of the Americas and Australia/New Zealand and the land-scarce, rapidly growing economies of Asia. If the degree of integration in the global economy is frozen at base period levels, a significant divergence in agriculture/forestry land rents arises across regions. In summary, when combined, the forces of globalization are expected to play a large role in determining the pattern of land-use change. We close this chapter with a discussion of some of the most salient limitations of our approach. First, we found that income growth in poor countries will generate substantial increases in food demand and the demand for agricultural land. It should be noted, however, that we do not take into account potential changes in income distribution. We have implicitly assumed that incomes of all households in each region rise in equal proportions. Second, it is assumed in the model that only crops, ruminant livestock, and forestry compete for land, while all other sectors do not use land. Thus, the baseline omits residential, commercial, and recreational demands for land. Another important component of the contemporaneous land-use baseline that is absent from current work is the demand for biofuels. Finally, we do not believe that present implementation of the land-use component of this model is sufficiently detailed. Although the 11 region/6 AEZ/crop-livestockforestry breakdown does yield some interesting heterogeneity across AEZs, in future work we would like to use a larger number of AEZs as well as more crops to capture the heterogeneity of land use across AEZs. For example, reducing the length of the growing period in each AEZ from 60 to 10 days, and distinguishing boreal, temperate, and tropical climates would yield a total of 108 AEZs. We believe this number would be manageable in future analyses and would considerably enrich the physical detail of the ecological constraints on production in the model at a relatively low cost.
Global Economic Integration and Land-Use Change
309
Appendix Aggregation of GTAP regions Region Australia and New Zealand (ANZ) China (CHN) High Income Asia (HYAsia) Association of Southeast Asian Nations (ASEAN) South Asia (SAsia) North America (NAM) Latin America (LAM)
Western European Union Europe (WEU) except Turkey
Economies in Transition (EIT)
Middle East and North Africa (MENA) Sub Saharan Africa
GTAP regions Australia, New Zealand China Hong Kong, Japan, Korea, Taiwan Indonesia, Malaysia, Philippines, Singapore, Thailand, Vietnam Bangladesh, India, Sri Lanka, and the rest of South Asia Canada, United States Mexico, Central America and Caribbean, Colombia, Peru, Argentina, Brazil, Chile, Uruguay, Venezuela, and the rest of Andean Pact Austria, Belgium, Denmark, Finland, France, Germany, United Kingdom, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain, Sweden, Switzerland, and the rest of EFTA Albania, Bulgaria, Croatia, Czech Republic, Hungary, Malta, Poland, Romania, Slovakia, Slovenia, Estonia, Latvia, Lithuania, Cyprus, Russian Federation, and the rest of the former Soviet Union Turkey, the rest of Middle East, Morocco, and the rest of North Africa Botswana, the rest of SACU, Malawi, Mozambique, Zambia, Zimbabwe, the rest of Southern Africa, Tanzania, Uganda, the rest of sub-Saharan Africa, and the rest of the world
References Adams, D. M., R. J. Alig, J. M. Callaway, B. A. McCarl, and S. M. Winnett. 1996. The Forest and Agricultural Sector Optimization Model (FASOM): Model Structure and Policy Applications. Research Paper PNW-RP-495. Portland, OR: U.S. Department of Agriculture, Forest Service, Pacific Northwest Research Station. Ahammad, H. and R. Mi. 2003. Land Use Change Modeling in GTEM: Accounting for Forest Sinks. Australian Bureau of Agricultural and Resource Economics. Paper presented at EMF 22: Climate Change Control Scenarios, May 25–7, Stanford University.
310
Alla Golub and Thomas W. Hertel
Cranfield, J. A. L., J. S. Eales, T. W. Hertel, and P. V. Preckel, 2003. “Model Selection when Estimating and Predicting Consumer Demands Using International, Cross Section Data.” Empirical Economics 28(2), 353–64. Darwin, R., M. Tsigas, J. Lewandrowski, and A. Raneses. 1995. World Agriculture and Climate Change: Economic Adaptations. Agricultural Economic Report No. 703. Washington, DC: U.S. Department of Agriculture, Economic Research Service. Available from http://www.ers.usda.gov/publications/aer703/aer703.pdf. Dimaranan, B. V. and R. A. McDougall, 2002. Global Trade, Assistance, and Production: The GTAP 5 Data Base. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Available at http://www.gtap.agecon.purdue.edu/databases/v5/v5 doco .asp. Gibbs H. K., S. Brown, J. O. Niles, and J. A. Foley. 2007. “Monitoring and Measuring Tropical Forest Carbon Stocks: Making REDD a Reality.” Environmental Research Letters 2(4), 1–13. Global Timber Market and Forestry Data Project. 2004. Available at http://www.agecon .ag.ohio-state.edu/people/sohngen.1/forests/GTM/index.htm. Golub, A. 2006. Projecting the Global Economy to 2025: A Dynamic General Equilibrium Approach. Ph.D. dissertation, Purdue University. Golub, A., T. W. Hertel, and B. Sohngen. 2007. Projecting Land Use Change in the Dynamic GTAP Framework. Paper presented at the Tenth Annual Conference on Global Economic Analysis, June 7–9, West Lafayette, IN. Hertel, T. W., C. E. Ludena, and A. Golub. 2006, November. Economic Growth, Technological Change and Patterns of Food and Agricultural Trade in Asia. ERD Working Paper No. 86. Manila: Asian Development Bank. Kets, W. and A. M. Lejour, 2003. Sectoral TFP Growth in the OECD. CPB Memorandum 58. The Hague: CPB. Lee, H.-L., T. W. Hertel, B. Sohngen, and N. Ramankutty. 2005. Towards an Integrated Land Use Data Base for Assessing the Potential for Greenhouse Gas Mitigation. GTAP Technical Paper No. 25. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Available at https://www.gtap.agecon.purdue.edu/resources/ res display.asp?RecordID=1900. Ludena, C. E., T. W. Hertel, P. V. Preckel, K. Foster, and A. Nin. 2007. “Productivity Growth and Convergence in Crop, Ruminant and Non-Ruminant Production: Measurement and Forecasts.” Agricultural Economics 37, 1–17. Reimer, J. J. and T. W. Hertel. 2004. “Estimation of International Demand Behavior for Use with Input-Output Based Data.” Economic Systems Research 16(4), 347–66. Rimmer, M. T. and A. A. Powell, 1996. “An Implicitly Additive Demand System.” Applied Economics 28, 1613–22. Rose, S., H. Ahammad, B. Eickhout, B. Fisher, A. Kurosawa, S. Rao, K. Riahi, and D. van Vuuren. 2007. Land in Climate Stabilization Modeling: Initial Observations. Stanford: Energy Modeling Forum. Sohngen, B., A. Golub, and T. Hertel. 2009. “The Role of Forestry in Carbon Sequestration in General Equilibrium Models.” In T. Hertel, S. Rose, and R. Tol (eds.), Economic Analysis of Land Use in Global Climate Change Policy (pp. 279–303). London: Routledge. Sohngen, B. and R. Mendelsohn. 2003. “An Optimal Control Model of Forest Carbon Sequestration.” American Journal of Agricultural Economics 85(2), 448–57.
Global Economic Integration and Land-Use Change
311
Sohngen, B. and R. Mendelsohn. 2006. “A Sensitivity Analysis of Carbon Sequestration.” In M. Schlesinger (ed.), Human-Induced Climate Change: An Interdisciplinary Assessment (pp. 227–37). Cambridge: Cambridge University Press. Sohngen, B., C. Tennity, M. Hnytka, and K. Meeusen. 2009. “Global Forestry Data for the Economic Modeling of Land Use.” In T. Hertel, S. Rose, and R. Tol (eds.), Economic Analysis of Land Use in Global Climate Change Policy (pp. 49–71). London: Routledge.
TWELVE
The Contribution of Productivity Linkages to the General Equilibrium Analysis of Free Trade Agreements Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer 1. Introduction Despite the economic arguments in favor of conducting trade negotiations multilaterally under the auspices of the WTO, there have been more than 100 regional trade agreements signed since the WTO was created in 1995 (WTO 2002). The European Union, for example, signed regional trade agreements with 20 nations between 1991 and 2001 – most of them developing countries (WTO 2002). Not to be left out, the United States continues to explore the possibility of extending NAFTA further to the south, with a free trade agreement (FTA) of the Americas as the ultimate goal. A free trade agreement between the United States and Chile was concluded in 2003, and discussions are under way with other nations. The boom in regional agreements has also spread to East Asia. For example, officials from ASEAN and China have endorsed the concept of a free trade agreement between those two regions. Japan, long a staunch advocate of multilateralism, concluded an FTA with Singapore in 2001 (see Chapter 9) and has also been involved in active negotiations with other countries, including Mexico (signed in 2004), ASEAN (signed in 2008), Korea, and China. Alongside this surge in interest in FTAs has been a corresponding increase in the number of quantitative analyses of such agreements, most of which employ AGE models. There are several reasons why these models are used. In evaluating alternative models of NAFTA, Francois and Shiells (1994) conclude that AGE models are preferable to partial equilibrium approaches because the latter fail to capture the economy-wide nature of FTAs, in which some sectors expand while others contract due to competition for a common pool of labor and capital. The alternative of macro-econometric models for FTA analysis is less appealing because they generally lack sufficient sectoral detail. Additionally, because FTAs involve multiple countries by definition, it is natural to use a multiregion AGE model in these studies. 312
The Contribution of Productivity Linkages
313
AGE-based FTA studies are not without criticism, however. For example, Kehoe (2002) has recently evaluated the AGE-based studies in the Francois and Shiells volume. He finds that they greatly underestimated the increases in trade resulting from NAFTA. This raises the question whether AGE models miss important mechanisms that promote trade growth. One recurring theme is that AGE models underpredict the changes associated with FTAs because they ignore effects related to productivity linkages, procompetitive effects, and investment dynamics. Many of these linkages are currently being studied in the empirical international trade literature, and we focus on three here. First, it is well known that more openness to imports may result in a reduction in a firm’s pricecost markups and a movement of firms down their average total cost curve. This is often referred to as the “procompetitive effect” of trade liberalization (Hertel 1994; Markusen 1981).1 Levinsohn (1993) offers evidence on such market disciplinary effects for Turkey, and a recent study by Ianchovichina et al. (2000) provides econometric evidence regarding the procompetitive effects of trade liberalization in Australia. Second, exporting may also be associated with improvements in productivity (Bernard and Jensen 2001). Newly opened foreign markets may enable domestic firms to expand such that they move down their average total cost curve. Overall productivity may also increase because only the most efficient firms survive in the new environment. Furthermore, there can be “learning by exporting,” which relates to productivity improvements resulting from the knowledge and experience gained in export markets (Aw, Chung, and Roberts 2000). Although there is less theoretical literature to draw on than in the case of procompetitive effects, there is a great deal of new empirical work in this area. For example, a recent study by Bernard and Jensen (2001) provides statistical evidence of export-productivity links for U.S. firms. Third, foreign direct investment (FDI) is another important channel through which trade liberalization can lead to increases in firm productivity (Blalock 2001).2 For example, a multinational may share new ideas and processes with a local firm, thereby trying to improve the efficiency 1
2
Another import-productivity channel relates to foreign intermediate inputs, which may be cheaper, of a different variety, higher quality, or more technologically advanced, thereby improving local firm productivity (Sj¨oholm 1999). So-called new-age FTAs tend to include measures designed to facilitate FDI among member countries. Furthermore, by reducing the cost of investment goods and boosting rental rates on capital, FTAs often increase FDI independent of any facilitation measures (Hertel, Walmsley, and Itakura 2001).
314
Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer
and quality of local upstream input suppliers. In addition to “vertical” technology spillovers of this type, there exists the possibility of “horizontal” spillovers, in which local firms copy the processes or hire away the staff of a competing firm acquired by a multinational (Blalock 2001). A recent study by Chuang and Lin (1999) provides econometric evidence regarding FDI-productivity relationships of this sort in Taiwan (China). In addition, Hallward-Driemeier, Iarossi, and Sokoloff (2002) find strong correlations between FDI, exports, and firm productivity in five ASEAN nations, using detailed data from 2,700 manufacturing enterprises. Because conventional AGE analyses of trade liberalization miss these potentially important effects, the purpose of our study is to explore their contribution in the context of one of the most important East Asian FTAs, the Japan-ASEAN FTA, which was signed in 2008. We carry out this analysis using the GDyn model, outlined in Chapter 2, and the most recent econometric work in the burgeoning field of empirical international trade. However, given the uncertainty surrounding many of these studies as well as the difficulty in applying their findings to a specific FTA, this chapter must be viewed as an exploratory effort aimed at understanding the potential impacts of these additional mechanisms. By identifying which of these effects is likely to be most important in determining the overall impact of a Japan-ASEAN FTA, we hope to provide a set of priorities for future policy-oriented econometric work aimed at refining the estimates used in this chapter. The remainder of this chapter is organized as follows. In the next section, we outline the theoretical aspects of our methodology, including our implementation of the earlier described productivity effects into the GDyn model. We then describe the data used in this study as well as some key structural characteristics of the focus countries in the East Asia region. Following that is a brief description of the baseline projection and simulation design. We then proceed to an analysis of the importance of each new mechanism incorporated into this study. The final section summarizes and concludes this analysis.
2. Empirical Framework 2.1 Procompetitive Effects Ever since the path-breaking work of Rick Harris (1984), it has become increasingly common to incorporate imperfect competition and scale economies into AGE models. Harris’s work on the Canada-United States
The Contribution of Productivity Linkages
315
FTA emphasized the potential gains in scale economies from disciplining domestic markups, which forced the exit of many Canadian manufacturers and pushed the remaining firms down their average total cost curve. The potential importance of introducing these features in the context of the Japan-ASEAN FTA follows from the high degree of concentration in some of the ASEAN manufacturing sectors, as well as the inefficient scale of production in the import-competing manufacturing sectors of most developing countries (Devarajan and Rodrik 1991). A survey of the issues that arise in modeling imperfect competition in AGE models is offered by Francois and Roland-Holst (1997).3 Once one chooses to move in this direction, one must confront a bewildering array of choices: (a) entry or no-entry; (b) Bertrand or Cournot oligopoly, or perhaps monopolistic competition; (c) product differentiation by firm (DixitStiglitz) or by nation (i.e., Armington); and (d) market segmentation or integration. To date, empirical work offers very little basis for discriminating between these alternative specifications. Compounding this problem is the fact that, depending on the assumptions invoked, the findings can be reversed (e.g., Markusen and Venables 1988). Once one has determined the appropriate market structure, there still remains the nontrivial task of model calibration to observed markups and unexploited scale economies. Finally, once scale economies enter the picture, computational problems loom large, with multiple equilibria becoming much more likely. Of course, just because it is hard does not mean that such work should be avoided. It does mean, however, that work in this area requires careful theoretical consideration, and it must be tailored to the issue at hand. In light of our overall objective, we chose to focus on those industries where there is the greatest likelihood of procompetitive effects stemming from the reduction of bilateral tariffs under a Japan-ASEAN FTA. We first observe that a Japan-ASEAN FTA is likely to have little procompetitive effect on Japanese manufacturing, because Japanese tariffs are already very low, except for light manufactures where scale economies are unlikely to be significant, particularly given the size of the domestic market in Japan. In contrast, tariffs for manufactures in ASEAN are considerably higher (Table 12.3), domestic markets are smaller, and the presence of Japanese imports is quite important (Table 12.2). Thus we focus our attention in this chapter on the potential for procompetitive effects in the ASEAN manufacturing sectors. 3
Francois (1998) has also made a wide variety of these approaches readily available within the GTAP framework.
316
Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer Table 12.1. Self-sufficiency ratios, 1997
Sector Textiles/apparel Leather Paper/wood Chemical products Metal products Automotive Machinery Electrical equipment
Thailand Indonesia Malaysia Philippines Vietnam Singapore 122 180 100 93 52 75 71 134
154 573 170 87 68 39 39 100
124 60 144 91 58 63 55 158
113 143 90 70 59 27 36 116
77 367 99 38 36 5 36 53
51 39 87 97 56 50 60 130
Notes: Values represent the percent that domestic production has of total use. Boldface entries indicate self-sufficiency below 95%.
Within the ASEAN manufacturing sectors, we believe that those sectors that are already heavily involved in exporting (e.g., textiles and apparel in Thailand, Indonesia, Malaysia, and the Philippines; see Table 12.1) are likely to have few remaining unexploited scale economies. However, in those industries where most domestic consumption is supplied by imports (e.g., automobiles in Indonesia; see Table 12.1) and that are also protected by substantial tariffs, we expect significant potential for procompetitive effects. Therefore, we adopt a model of import-competing domestic industries and apply this to all ASEAN manufacturing sectors with less than 95% selfsufficiency ratios (see the boldface entries in Table 12.1). The theoretical framework that we use here is that of oligopolistic competition in the presence of firm-level product differentiation (Hertel 1994). Foreign firms are assumed to leave their markups unchanged, so the import price falls by the full amount of the tariff cut. Domestic firms incur fixed costs to produce a new variety, after which production is subject to constant returns to scale, so that average total costs decline with output. Furthermore, we assume that displaced domestic varieties will be replaced by similar imported varieties such that the varietal impact on consumer welfare from the FTA is negligible (Case III, Hertel 1994).4 With this model in mind, and simplifying the analysis to a two-sector, small, open economy model of one of the ASEAN economies, it can be shown (Hertel 1994, Proposition 5) that welfare is an increasing function of the elasticity of the domestic markup with respect to the foreign price 4
Because of the “home bias” observed in the trade data, if we assumed otherwise, welfare would fall with the tariff cuts because of the high value placed on domestic varieties (e.g., Markusen and Venables, 1988). We do not believe that to be realistic in the case of Indonesian versus Japanese autos, for example.
317 Table 12.2. Relative importance of Japan in ASEAN merchandise trade, 1997 (%) ASEAN
Singapore
Indonesia
Malaysia
Philippines
Thailand
Vietnam
Percent of all ASEAN imports coming from Japan Food & agriculture Nat. resources Textiles/apparel Leather Paper/wood Chemical products Metal products Automotive Machinery Electrical equipment
2.1 0.3 9.5 2.0 8.6 16.4 25.8 60.9 28.5 21.0
2.9 0.1 6.0 1.4 7.1 14.1 21.5 33.6 26.1 18.3
Food & agriculture Nat. resources Textiles/apparel Leather Paper/wood Chemical products Metal products Automotive Machinery Electrical equipment
18.2 37.5 9.9 6.0 25.7 9.0 15.6 11.9 11.7 9.2
7.9 19.2 2.7 9.3 7.8 5.2 6.0 0.7 5.7 7.6
0.7 1.1 12.3 1.0 9.2 15.7 26.1 59.8 32.0 25.1
1.0 0.8 8.3 0.8 11.1 15.9 27.9 61.1 25.6 18.2
1.8 0.1 8.2 4.1 6.6 18.8 18.9 64.6 27.5 31.4
4.1 0.3 12.7 4.5 9.3 24.4 33.0 74.1 34.2 24.7
2.4 3.9 10.7 1.3 5.1 8.2 11.0 59.8 19.9 18.4
26.0 8.0 9.7 3.9 30.1 15.1 19.0 16.1 15.6 13.9
22.9 37.2 34.8 8.2 27.8 11.2 20.6 40.0 16.9 8.8
Percent of all ASEAN exports going to Japan 20.5 39.8 8.6 6.4 29.1 11.1 35.8 25.6 19.2 12.6
6.5 34.0 7.8 1.7 24.9 8.5 11.3 4.6 10.4 7.5
22.8 68.8 5.8 4.5 19.2 17.2 16.9 31.3 23.0 13.5
318 Table 12.3. Average sectoral tariff rates, 1997 (%)
Food & agriculture Natural resources Light mnfcs High-tech mnfcs Merchandise total Services
Japan
ASEAN
Indonesia
Malaysia
Philippines
Thailand
Vietnam
Singapore
52.7 −1.0 7.8 0.8 7.0 22.4
15.8 0.7 11.5 5.4 5.2 6.0
9.0 2.8 11.3 8.7 6.9 6.9
21.0 1.1 12.5 5.3 5.5 5.2
17.5 1.0 12.9 5.6 5.3 6.0
31.3 0.8 19.9 13.1 11.5 5.9
32.3 4.6 29.0 14.1 16.5 6.2
4.3 0.0 0.0 0.0 0.2 0.0
Note: Average tariffs on service sectors are estimated tariff equivalents.
The Contribution of Productivity Linkages
319
(βMF ). This elasticity is itself a function of the nature of the oligopoly (e.g., βMF is larger for Cournot than Bertrand oligopoly), as well as the substitutability among varieties and market shares (Hertel 1994; see Table 12.2). The mechanism by which this welfare gain arises is precisely that described earlier: Lower foreign prices discipline domestic markups and cause output per firm to increase, with some domestic firms exiting the industry to restore market equilibrium. To be more precise, we have the following relationship between output per firm (q) and the power of the domestic markup (M) in the presence of ˆ domestic entry/exit: qˆ = −−1 F M, where ˆ denotes percentage change and F is the share of fixed in total costs. With the total number of varieties on offer fixed, the equation for markups as a function of relative foreign ˆ = βMF (pˆ F − pˆ H ), where βMF is (p F ) and home (p H ) prices is simply M the markup elasticity with respect to the foreign price, as discussed above. Combining these yields equation 12.1: qˆ = −−1 F βMF (pˆ F − pˆ H ),
(12.1)
which shows that output per firm is directly related to the markup elasticity, scaled by the inverse of the share of fixed costs in production. The presence of fixed costs, in conjunction with constant returns to scale in variable costs, means that the elasticity of output with respect to composite inputs (z) is greater than 1. In particular, we have qˆ = (1 − ˆ where −1 −1 F )z, F may be shown to equal the cost disadvantage ratio (CDR; Francois and Roland-Holst, 1997). A reduction in tariffs lowers the price of competing foreign goods and so lowers markups, thereby boosting output per firm. The resulting efficiency gain, expressed as a percentage of sectoral ˆ Multiplying both sides of (12.1) by F , we can output, is given by F q. write the sectoral efficiency gain as a function of the markup elasticity and the percentage change in relative prices:5 efficiency = −βMF (pˆ F − pˆ H ).
(12.2)
This reduced-form expression is incorporated into the GDyn model to capture the essence of the procompetitive effects that might be available under a Japan-ASEAN FTA. To keep the focus on the counterfactual FTA impacts, we do not implement this specification in the baseline itself. Rather it only applies to deviations of foreign (p F ) and home (p H ) prices from the baseline price path. 5
For this procompetitive effect (as well as the export and investment effects) the “efficiency” variable is represented as ao(j,r) in the GDyn model, which is the percentage rate of Hicks-neutral technical change in sector j of region r.
320
Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer
The key factor in our empirical implementation of this model will be the size of βMF . This could be obtained by assuming something about the oligopoly structure and the number of firms in each imperfectly competing industry in ASEAN. Alternatively we could refer to recent empirical estimates of βMF . In the spirit of this econometrically based study, we take the latter approach. One study that estimates βMF is that of Ianchovichina et al. (2000), who use a partial equilibrium model corresponding to the theoretical model discussed earlier. These authors focus on the behavior of the Australian auto industry during a period of deep tariff cuts – similar to those faced by ASEAN members under the proposed FTA. They obtain a fairly precise estimate of βMF = 0.44 for this industry using quarterly data. This means that nearly half of a 1% decline in tariffs will be absorbed by reduced markups, which is quite significant. Indeed, in some simple, partial equilibrium simulations of tariff cuts in the Australian auto sector, Ianchovichina et al. (2000) conclude that a perfectly competitive model of this industry would likely overstate the output decline by 80%. That is, by ignoring this procompetitive effect, one might erroneously predict nearly twice as great a decline in output as would actually take place following a tariff cut. If this is indicative of the kind of impact that tariff cuts on Japanese imports would have on markups in ASEAN, then we need to give it further consideration. Of course, the auto industry tends to be a special case in the ASEAN economies, with very high rates of protection.6 Additionally, because the markup elasticity depends on a number of factors, many of which are unobservable, there is no reason to believe that other sectors will have the same value of βMF . Indeed, this is what Ianchovichina (1994) finds for several other sectors in the Australian economy, where her estimates for this markup elasticity are as follows: chemical fertilizers: 0.38, plastic materials: 0.22, and steel pipes: 0.19. The high markup elasticity for autos is not surprising, because this sector is highly protected and highly concentrated, selling a highly differentiated product. The fact that chemical fertilizers exhibit such a high markup elasticity is more an indication of market power in an industry with very high entry costs, as opposed to its being a highly differentiated product. Evidence on markups across industries worldwide (Francois 1998) reinforces the point that the chemicals sector tends to be quite imperfectly competitive. The same is true for steel products. Yet it is
6
The average rate of automobile protection in ASEAN is 32.2%, as opposed to 5.8% for all other manufactures.
The Contribution of Productivity Linkages
321
hardly surprising that clay bricks show no markup elasticity. These bricks are undifferentiated products with low barriers to entry. Our approach to estimating βMF for the import-competing sectors in ASEAN relies on outside information about the relative size of markups across industries in non-OECD economies (Francois 1998). Our theoretical model shows that βMF is increasing in the size of the industry markup, so we use this as a guide for setting the relative values of this elasticity across sectors within the ASEAN region. We next restrict the range of values for βMF in our model to the estimates obtained from Ianchovichina (1994), namely 0.0 to 0.44. We then distribute the values of βMF over this range, according to the relative size of the sectoral markups. The resulting estimates are as follows: 0.44 for chemical products, 0.40 for paper and wood products, 0.29 for metal products, 0.19 for the automotive sector, 0.19 for textiles/apparel/leather products, 0.13 for the machinery sector, and 0.13 for electrical equipment. Thus, the highest markups, and hence the largest procompetitive effects, are in the chemical and wood products industries, followed by metal products. The other manufacturing sectors show lower markups and hence receive a negligible procompetitive effect.
2.2 The Exporting-Productivity Effect In the case of the export-productivity linkage, there is little theoretical literature to draw on, and one confronts a fundamental problem of causality. Are exporters more productive because they export, or are they exporting because they are more productive? Bernard and Jensen (2001) attempt to control for this problem by looking at individual firms over time. They use a panel dataset covering 50,000–60,000 individual manufacturing plants in the United States for the years between 1983–92. They find that plants that always export during this time period are 8–9% more productive than plants that never export, a result similar to those found in other studies (p. 9). Although they find that exporting does not necessarily increase plant productivity growth rates, exporting is associated with the shifting of resources from less efficient to more efficient plants. Of particular interest to our purposes here, they find that firms that start exporting tend to have productivity levels above those that never export during the period, although significantly below those who export throughout the period. As soon as firms begin exporting, however, their productivity grows until they nearly reach the level of firms that were exporting throughout the period (Bernard and Jensen 2001; Fig. 12.1). At the same time, firms that were exporters at the beginning of the period and then stopped exporting
322
Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer
at some point started out with high productivity but converged downward to the level of firms that never exported at all. This indicates that there may be some degree of reversibility, such that a firm’s relative productivity can diminish if it ceases to export. Based on these findings, if we assume that the number of firms in an industry remains fixed, we have a situation in which the overall technological prowess of the industry will rise as exports rise, but will fall in concert with decreases in overall industry exports. We incorporate this export-productivity linkage into our global AGE model as follows. Let δ > 1 be the ratio of the technology index used for export-oriented firms relative to that of firms specializing in the domestic market only. Let s X be the share of output that is exported and s D be the share of output that is used domestically (note that s D + s X = 1). Furthermore, let qˆ X be the percentage change in output that is exported and qˆ D be the percentage change in output that is used domestically. Then, as shown in Appendix 12.1, we can obtain an equation for the rate of change in overall productivity in the industry as a function of the productivity differential between exporters and domestic firms and of the differential growth in these two output markets: efficiency =
(δ − 1)s X s D (qˆ X − qˆ D ) . s D + δs X
(12.3)
Based on Bernard and Jensen’s (2001) calculations and our definition of δ, we calibrate the exporting sector to be 8% more productive, giving rise to δ = 1.08. Therefore, when the rate of change in exported output exceeds that of output for domestic consumption (qˆ X > qˆ D ), the average level of technology in the industry rises (i.e., the efficiency variable is positive). Note that this formulation can also induce efficiency losses when exported output declines relative to output for domestic consumption, because technological gains are reversible (as was found by Bernard and Jensen). As with the procompetitive effect, this efficiency effect is incorporated into the GDyn model as a reduced-form representation of the more complex underlying process by which exporting affects firm-level productivity. Furthermore, this export-productivity linkage only applies to the export and output deviations from the baseline, not to the baseline itself.
2.3 The FDI-Productivity Effect Increased levels of foreign direct investment (FDI) have the potential to transfer technology and managerial skills to a host country, thereby enhancing productivity (Blalock 2001). Some authors, such as Rodrik (1999), point out that there is little hard evidence for the more extravagant claims linking
The Contribution of Productivity Linkages
323
FDI and productivity. However, an increasing number of studies confirm that there are indeed significant, positive technological spillovers, even if we cannot always identify the precise mechanism through which these spillovers occur. For example, in a study of FDI, research and development (R&D), and spillover efficiency in Taiwan (China), Chuang and Lin (1999) use firm-level data to confirm the existence of beneficial spillovers from FDI. Specifically they find that a 1.0% increase in an industry’s FDI ratio produces a 1.40% to 1.88% increase in domestic firm productivity. As indicated in Chapter 2, in the GDyn model regional capital is owned by domestic and foreign households via a global trust. This relationship is V = VH + VF , where V is the equity value of firms in a given country, and VH and VF are the domestic- and foreign-held components of V, respectively. Thus we can write the foreign equity share as θF ≡ VF /V. We use this as a proxy for the share of FDI in total capital stock. Because we want to relate productivity changes to changes in θF , we totally differentiate this to get ˆ Using Chuang and Lin’s (1999) lower bound estimate, we can θˆF = Vˆ F − V. write the percentage change in productivity associated with a capital inflow from abroad as ˆ efficiency = 0.014(Vˆ F − V). As with Chuang and Lin’s study, we implement this reduced-form relationship only for manufacturing sectors and incorporate it into the GDyn model as an additional equation determining the change in efficiency endogenously as a function of changes in the share of foreign ownership, owing to the FTA. We do not incorporate this productivity effect into the dynamic baseline. It only plays a role in the FTA counterfactual.
3. Data and Procedures 3.1 Data and Aggregation In this analysis we employ the GTAP 5 Data Base, which has a base year of 1997 and distinguishes 57 sectors and 66 regions (Dimaranan and McDougall 2002). Among its notable features are disaggregation of service sectors and the explicit treatment of international transport margins. We aggregate the GTAP data up to 23 sectors and 19 regions (see Appendices 12.2 and 12.3, respectively).7 Our regional aggregation emphasizes the individual countries involved in the Japan-ASEAN FTA. The GTAP data 7
In certain tables in this chapter we use higher sector aggregates (e.g., food & agriculture) to save space. Definitions are in Appendix 12.2. The analysis is otherwise done in terms of the 23 sectors.
324
Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer
distinguish six ASEAN nations (Singapore, Indonesia, Malaysia, Philippines, Thailand, and Vietnam), and when we refer to “ASEAN” later, we refer to these six only. Although the GTAP data do not disaggregate Brunei, Cambodia, Laos, and Myanmar, these four nations comprise only 3.25% of ASEAN GDP (ASEAN 2002).
3.2 Trade Flows and Tariffs In this section we use the aggregated GTAP data to provide an overview of the current trade and tariff relationships between Japan and ASEAN nations. The data indicate that, although ASEAN depends on Japan for about 19% of overall imports, Japan gets only about 11% of its imports from ASEAN.8 So despite their proximity, and dissimilarities in terms of endowments and technology (which may be a source of comparative advantage), these economies are not highly integrated, especially when compared to other regions such as Europe or North America. A deeper view of the current Japan-ASEAN trade relationship can be gained from Table 12.2, which breaks down the relative importance of merchandise trade with Japan for individual ASEAN members. The top half refers to the percent of ASEAN merchandise imports that originate in Japan. Clearly, Japan is not an important supplier of agriculture, resources, or light manufactures for ASEAN, because it supplies less than 10% of total imports in nearly all cases (Table 12.2). However, Japan is quite important as a source of high-technology manufactures. In the automotive sector Japan plays a particularly dominant role, with an import share of 60.9% for ASEAN overall. The bottom half of Table 12.2 depicts the relative importance of Japan as an export destination for the different ASEAN countries. Although this varies to a great degree across ASEAN countries, Japan generally plays a fairly large role as an export destination for ASEAN food and natural resource sectors. To the extent that comparative advantage is a driver of trade, it would appear that Japan and ASEAN are natural trading partners. Japan can play a role as a high-tech supplier, whereas the ASEAN countries as a group are presently well suited to meet Japan’s need for resources, agriculture, and light manufactures. Average tariff rates for all sectors are reported in Table 12.3. Japan is quite notable for its protection of food and agriculture (52.7% average tariff), which is driven to a large extent by protection of its rice market (rice has 8
Japan’s imports of goods and services from ASEAN nations totaled $52.3 billion in 1997, whereas Japan’s imports from the Rest of World (ROW) were $395.1 billion. ASEAN imported $81.2 billion from Japan and $276.2 billion from the ROW and had $79.4 billion worth of trade within itself.
The Contribution of Productivity Linkages
325
Table 12.4. Average trade-weighted bilateral tariffs, 1997 (%) Japan Singapore Indonesia Malaysia Philippines Thailand Vietnam Japan Singapore Indonesia Malaysia Philippines Thailand Vietnam
– 1.3 5.4 1.9 5.5 13.4 11.4
0.0 – 0.2 0.2 0.2 0.2 0.6
9.6 4.5 – 7.9 3.6 8.3 3.5
8.4 5.1 11.0 – 2.3 7.4 24.8
6.2 4.5 7.8 5.5 – 3.9 19.0
16.8 11.2 15.4 11.4 8.3 – 10.9
17.5 15.3 9.4 18.6 4.4 23.6 –
a tariff equivalent of 409%); it protects its service sector to a relatively lesser extent (22.4% tariff equivalents). In contrast, Japan is fairly open with regard to light manufacturing (7.8% tariff), and its average tariff on high-tech manufactures is only 0.8%. ASEAN is more open in food and agriculture and more protective with regard to manufacturing compared to Japan. This is particularly the case for the automotive sector, where tariff equivalents range from 38% to 48% for Vietnam, Indonesia, Malaysia, and Thailand. So there appears to be a large degree of complementarity between the two regions in terms of the benefits that can accrue from reducing tariffs. An alternative view of the level of protection is provided in Table 12.4, which reports a matrix of trade-weighted, bilateral tariffs across all commodities traded between country pairs. Looking first at Japan’s column, its 1997 tariffs on goods and services from the ASEAN nations ranged from 1.3% for Singapore to 13.4% for Thailand. The top row reports ASEAN tariffs on Japanese exports. Whereas Thailand’s and Vietnam’s tariffs on Japanese exports were relatively high, other ASEAN nations appear to be fairly open, at least as far as the trade-weighted average tariff goes. Note that in our FTA simulation all Japan-ASEAN tariffs are eliminated.9 Clearly there will be a fair amount of Japan-Thailand and Japan-Vietnam trade response on the basis of the relatively large tariffs in place on both sides.
3.3 Baseline Simulation Our policy simulation results are obtained by comparing the counterfactual FTA policy scenario to our baseline. To produce meaningful results, the baseline should reflect as closely as possible the changes in the world 9
Table 12.4 also displays intra-ASEAN tariffs for 1997. As shown in Appendix 12.4, these are reduced in our baseline scenario in the manner prescribed by ASEAN’s Common Effective Preferential Tariff (CEPT) reduction program.
326
Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer
economy expected to occur over the period under study: 1997 to 2020. The baseline used in this chapter is built on the baseline developed in Chapter 5. It contains information on macroeconomic variables as well as expected policy changes. The macroeconomic variables in the baseline include projections for real GDP, gross investment, capital stocks, population, skilled and unskilled labor, and total labor. These projected macroeconomic variables were obtained for 211 countries over the period from 1997 to 2020. These projections for population, investment, skilled labor, and unskilled labor were aggregated, and growth rates were calculated to obtain the macro shocks describing the baseline. Changes in capital stocks were not imposed exogenously, but rather were determined endogenously as the accumulation of projected investment. Any changes in real GDP not explained by the changes in endowments are attributed to technological change. In addition, policy projections are also introduced into the baseline (these are summarized in Appendix 12.4). The policies included in the baseline are those that are already agreed on and are legally binding (e.g., Uruguay Round commitments and China’s WTO accession). Uruguay Round tariff commitments are assumed to be honored by all countries. Taiwan (China) and China’s accession to the WTO is phased in over two periods: a period of pre-WTO tariff reduction for 1997–2001 and the period from 2002–20. This accession also gives China and Taiwan (China) quota-free access to the North American and European textile and apparel markets by 2007. However, the liberalization of these quotas is assumed to be heavily back-loaded, with most of the liberalization occurring after 2002. The CEPT preferential tariff reduction program among ASEAN members and the Japan-Singapore FTA have also been incorporated into the baseline.
3.4 Simulation Design Once the baseline has been established, we are able to explore the impact of counterfactual policy simulations. Our simulations of the Japan-ASEAN free trade agreement involve four different model specifications, aimed at identifying the most important potential sources of productivity gain. Simulation (a) involves the complete elimination of tariffs among all countries involved in the Japan-ASEAN FTA (as well as removal of service trade barriers), but does not allow for the three new linkages described earlier. As such, it represents the standard types of effects that a conventional, dynamic AGE model would capture, including allocative efficiency, investment reallocation, and accumulation of capital stocks, as well as terms of trade effects. The remaining three simulations extend the first simulation (a) by adding
The Contribution of Productivity Linkages
327
the three additional modeling effects one at a time. Simulation (b) adds export-productivity effects, simulation (c) adds procompetitive effects, and simulation (d) adds FDI-productivity effects. These additions are cumulative in nature, and therefore simulation (d) includes all three additional effects.
4. Results We begin this section by introducing some shorthand notation regarding the productivity linkages we incorporate into our analysis. In the tables discussed in this section, columns labeled “STD” are meant to represent the difference between the baseline simulation and simulation (a). As such, “STD” refers to the effects normally captured by standard dynamic AGE models, including allocative, investment, and terms of trade effects. Next, “EXP” is the difference between simulations (a) and (b), and it captures productivity effects related to the potential expansion of export-oriented firms under an FTA. “IMP” is the difference between simulations (b) and (c), and it captures procompetitive effects related to the exposure of local, imperfectly competitive firms to foreign competition. Finally, “FDI” is the difference between simulations (c) and (d), and it refers to productivity effects related to foreign investment in local firms.
4.1 Welfare and GDP Impacts of the FTA Table 12.5 reports regional welfare changes in the year 2020 resulting from the Japan-ASEAN free trade agreement. “Welfare” is defined as the percentage change in utility of a representative regional household in 2020 owing to the FTA. Consider first the change in welfare with all effects in place (i.e., the results of simulation [d]). These are reported in the “Total” column of Table 12.5. All of the member nations experience an increase in welfare relative to the baseline. In relative terms, Thailand has the most to gain from a Japan-ASEAN FTA, with a welfare level that is 3.32 percentage points above the baseline scenario. For ASEAN nations as a whole, the welfare gain is 1.04 percentage points over the baseline, with Japan having a lower figure of 0.23 percentage points. The nations that face relatively low barriers in Japan before the FTA, including Indonesia, the Philippines, and particularly Singapore (Table 12.4), tend to experience smaller improvements in welfare (0.26, 0.24, and 0.46 percentage points, respectively). We can also examine how these results would differ had we not incorporated the additional productivity-linkage effects. Refer to the following
328
Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer Table 12.5. Overall welfare effects, 2020
Region
Total
STD (% of total)
EXP
IMP
FDI
ASEAN Indonesia Malaysia Philippines Thailand Vietnam Singapore Japan
1.04 0.26 0.47 0.24 3.32 0.62 0.46 0.23
0.26 (25) 0.11 (42) −0.06 (−12) 0.13 (54) 0.82 (25) 0.22 (36) 0.15 (33) 0.20 (86)
0.05 0.04 0.06 0.02 0.10 0.01 0.01 0.01
0.35 0.07 0.32 0.07 1.06 0.25 0.14 0.01
0.38 0.05 0.14 0.03 1.34 0.14 0.16 0.02
Note: These values represent percentage point differences from the baseline scenario in 2020. Figures in parentheses are percentage contribution of standard AGE model effects to overall welfare change. Abbreviations are as follows: standard AGE modeling effects (STD), export-productivity effect (EXP), procompetitive effect (IMP), FDI-productivity effect (FDI).
columns in Table 12.5: “STD” (standard AGE modeling effects resulting from the tariff cuts), “EXP” (the export-productivity effect), “IMP” (the procompetitive effect), and “FDI” (the FDI-productivity effect). In terms of utility, the contributions of these extra effects are generally significant. In fact, we see that only 25% of the welfare change in ASEAN is related to standard AGE modeling effects (see the values within parentheses in the STD column). This figure is higher for Japan (86%), because of the absence of procompetitive and FDI effects for that country. Thailand shows the largest overall relative gains. In that country, the most important channel for welfare change is the FDI-productivity effect (1.34 out of 3.32 percentage points), in which higher levels of foreign ownership following the FTA led to improvements in domestic firm productivity. We now move on to other macroeconomic results presented in Table 12.6. As in Table 12.5, these changes are given as percentage point differences from the baseline, allowing us to gauge differences in relative terms. We first focus on the total change in GDP (the other variables in Table 12.6 are discussed in later sections). Although Japan’s 2020 GDP is only 0.14 percentage points higher than in the baseline scenario, ASEAN’s overall change is significantly higher (3.66 percentage points), with Thailand having the largest change by far (12.41 percentage points).10 In Thailand, most of the change is due to conventional AGE modeling effects (STD), followed by the FDI-productivity effect and the procompetitive effect. This is because 10
The small change in Singapore’s GDP (0.18 percentage points) reflects the fact that it has already formed an FTA with Japan and thus does not benefit to the extent that other ASEAN nations do.
The Contribution of Productivity Linkages
329
Table 12.6. Effect of Japan-ASEAN FTA on selected macro variables, 2020 Effect
GDP
Imports
Exports
Capital
GDP
Japan Total STD EXP IMP FDI
0.14 0.15 0.01 −0.01 0.00
2.83 2.73 0.02 0.02 0.06
0.53 0.33 0.06 0.08 0.06
1.27 1.10 0.07 0.04 0.05
1.54 1.54 0.02 −0.03 0.02
0.27 0.29 0.01 −0.02 −0.01
3.66 2.06 0.08 0.77 0.75
1.56 0.73 0.12 0.47 0.24
2.46 1.91 0.07 0.20 0.29
3.28 2.92 0.12 0.13 0.11
1.23 0.93 0.10 0.11 0.08
0.99 0.83 0.03 0.10 0.04
3.36 2.59 0.09 0.34 0.35
0.18 −0.10 −0.02 0.18 0.12
0.99 0.38 0.01 0.30 0.30
0.58 0.19 0.00 0.22 0.17
4.03 2.79 0.07 0.43 0.73
6.24 4.46 0.07 0.98 0.73
5.71 3.61 0.08 1.08 0.93
2.35 2.21 0.04 0.06 0.05
2.84 2.62 0.04 0.14 0.04
2.23 1.93 0.03 0.19 0.07
Thailand 2.26 1.26 0.14 0.55 0.30
12.41 6.96 0.15 2.56 2.75
Singapore Total STD EXP IMP FDI
Capital
Philippines
Malaysia Total STD EXP IMP FDI
Exports
ASEAN
Indonesia Total STD EXP IMP FDI
Imports
15.44 10.15 0.21 1.77 3.31
23.96 16.44 0.09 4.11 3.33
16.22 10.19 0.12 3.03 2.88
Vietnam 0.30 −0.23 −0.01 0.32 0.21
2.72 2.04 0.06 0.50 0.12
4.93 4.98 −0.04 −0.45 0.44
11.11 8.27 0.43 3.22 −0.81
4.62 3.10 0.01 1.28 0.23
Note: These values represent percentage point differences from the baseline scenario in 2020.
Japan has notably high overall tariffs with respect to Thailand, and Thailand also displays relatively high tariffs with respect to Japan (Table 12.4). Figure 12.1 offers a temporal perspective regarding the changes in Thailand’s GDP. Here, deviations from the baseline attributed to each effect are provided separately. Begin by looking at the year 2006, the first year of the prospective FTA. There we see that without the procompetitive effect (IMP), we would have underestimated Thailand’s GDP change from the baseline by 0.8 percentage points. Although initially the procompetitive effect is the most important driver of Thailand’s GDP difference, by 2007 conventional AGE effects (STD) in the form of added investment take over as the most important contributor. Also observe that the FDI-productivity link (FDI) is
330
Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer
8.0
Percentage points difference from baseline
7.0
6.0 5.0
4.0
3.0 2.0
1.0 0.0 2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
-1.0 STD
EXP
IMP
FDI
Figure 12.1. Effect of Japan-ASEAN FTA on GDP, Thailand, relative to baseline.
unimportant in the first several years after implementation of the FTA, but continuously grows in importance along with the increased foreign investment until it is the second largest contributor to the growth in GDP by 2020. The sum contribution of all three additional productivity effects to GDP is 5.45 percentage points over Thailand’s baseline level in 2020, compared to 6.96 percentage points from the standard effects alone. On this basis it would appear that the productivity effects that are normally ignored in AGE analysis may indeed be important in the analysis of the Japan-ASEAN FTA, although one must bear in mind that these effects were somewhat more pronounced in Thailand than in the other nations (recall Table 12.6).
4.2 Effects on Trade and Foreign Capital Ownership Looking back at Table 12.6 we see that a Japan-ASEAN FTA results in higher overall imports and exports for all ASEAN nations, as well as Japan. Thailand has the largest increases: 15.44 and 23.96 percentage points over the baseline for imports and exports, respectively. In ASEAN most of the changes in trade volumes are due to conventional AGE modeling effects (STD). Thus relatively little is missed by ignoring effects on productivity arising from increased exports, imports, and foreign ownership of firms. Table 12.7 presents a sectoral decomposition of the changes in JapanASEAN trade resulting from the FTA. The upper half reports the differences over the baseline regarding exports from ASEAN to Japan.11 Not 11
Of course, changes involving ASEAN exports to Japan coincide exactly with changes involving Japanese imports from ASEAN.
The Contribution of Productivity Linkages
331
Table 12.7. Change in trade volume between ASEAN and Japan due to
Japan-ASEAN FTA, 2020 Total Food and ag. Nat. resources Textiles/apparel Leather Paper/wood Chemical products Metal products Automotive Machinery Electrical equipment Services
Food and ag. Nat. resources Textiles/apparel Leather Paper/wood Chemical products Metal products
STD
EXP
IMP
FDI
5,064 (77.5) −613 (−4.0) 1,458 (73.3) 709 (159.0) 1,149 (16.2) 359 (14.3) 635 (18.8) 261 (45.3) 2,789 (24.7) 4,537 (15.7) 1,206 (11.6)
Exports from ASEAN to Japan 5,250 −13 −73 (80.3) (−0.2) (−1.1) −512 −9 −51 (−3.3) (−0.1) (−0.3) 1,362 13 50 (68.5) (0.7) (2.5) 687 3 19 (153.9) (0.6) (4.2) 1,087 19 19 (15.4) (0.3) (0.3) 221 3 114 (8.8) (0.1) (4.5) 476 16 109 (14.1) (0.5) (3.2) 157 1 90 (27.3) (0.2) (15.6) 2,234 51 238 (19.8) (0.5) (2.1) 3,327 25 539 (11.5) (0.1) (1.9) 1,151 −5 114 (11.1) (−0.0) (1.1)
−99 (−1.5) −41 (−0.3) 34 (1.7) 1 (0.3) 25 (0.3) 21 (0.8) 33 (1.0) 12 (2.1) 266 (2.4) 645 (2.2) −54 (−0.5)
532 (52.9) 170 (21.3) 1,264 (187.5) 32 (118.0) 550 (62.7) 12,018 (33.0) 7,524 (47.8)
Exports from Japan to ASEAN 491 1 18 (48.9) (0.1) (1.8) 132 1 19 (16.5) (0.1) (2.4) 1,258 4 12 (186.5) (0.6) (1.7) 32 0 0 (116.3) (0.2) (1.2) 541 −1 5 (61.7) (−0.1) (0.6) 11,926 32 −161 (32.7) (0.1) (−0.4) 6,972 24 184 (44.3) (0.2) (1.2)
22 (2.2) 17 (2.2) −10 (−1.4) 0 (0.3) 4 (0.5) 221 (0.6) 343 (2.2) (continued)
332
Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer Table 12.7 (continued)
Automotive Machinery Electrical equipment Services
Total
STD
EXP
IMP
FDI
8,850 (70.1) 11,474 (36.5) 3,446 (18.1) 312 (8.2)
8,993 (71.3) 10,690 (34.0) 3,245 (17.1) 258 (6.8)
15 (0.1) 56 (0.2) 2 (0.0) 2 (0.1)
−348 (−2.8) 243 (0.8) 106 (0.6) 12 (0.3)
191 (1.5) 485 (1.5) 93 (0.5) 40 (1.0)
Notes: Values represent differences in 1997 US$ millions from baseline scenario in 2020. Values in parentheses represent the corresponding percentage point differences.
surprisingly, the FTA leads to a great deal more trade in nearly every category. In relative terms, the biggest boost comes from increased exports of leather products (159 percentage points over the baseline, or $709 million). In absolute terms, the biggest trade increase is in food and agricultural products, which increase by $5,064 million (77.5 percentage points) over the baseline, followed by electrical equipment ($4,537 million, 15.7 percentage points) and machinery ($2,789 million, 24.7 percentage points). In all of these cases the increases are related mainly to standard AGE modeling effects (STD). Exports from Japan to ASEAN are given a big boost in general (lower half of Table 12.7). In absolute terms, the largest change derives from an increase in chemical exports by $12,018 million (33.0 percentage points). In relative terms, exports of textile and apparel products have the largest increase, at 187.5 percentage points ($1,264 million). In both cases, the majority of the change results from standard effects relating to the tariff cuts. This is related to the fact that Thailand and Vietnam had particularly large initial tariffs in these particular sectors (Table 12.3). (For a disaggregation of Table 12.7, see also Appendixes 12.2 and 12.3.) Recall from Table 12.6 that implementation of the Japan-ASEAN FTA results in higher capital stocks for all the countries involved in the FTA. For these increased capital stocks, it is of interest to focus on the change in foreign ownership because it is hypothesized to drive the efficiency gain. Table 12.8 reports the change in share of foreign capital ownership by 2020, compared to the baseline. A free trade agreement between Japan and ASEAN attracts investment from abroad to all the countries involved in the FTA, resulting in a higher share of foreign capital ownership. For Thailand the share of capital
The Contribution of Productivity Linkages
333
Table 12.8. Change in share of foreign capital ownership, 2020 (percent
of total capital ownership)
Indonesia Malaysia Philippines Thailand Vietnam Singapore Japan
Total
STD
EXP
IMP
FDI
0.003 0.05 0.82 5.29 1.40 0.03 0.07
0.002 0.03 0.71 3.15 0.98 −0.08 0.08
0.001 0.00 0.01 0.05 0.00 0.00 0.00
0.00 0.01 0.08 0.96 0.36 0.06 −0.01
0.00 0.01 0.03 1.14 0.07 0.04 0.00
owned by foreign investors increases by 5.29 percentage points, relative to the baseline. The figure for Vietnam is 1.4 percentage points, whereas the remaining changes in foreign ownership share are all less than 1%. The conventional AGE modeling effects (STD) account for the majority of the increase, but the procompetitive effects (IMP) and FDI-productivity effects (FDI) also contribute a considerable amount, particularly in Thailand and Vietnam.
4.3 Effects on Efficiency Table 12.9 reports the sectoral efficiency gains in the manufacturing sectors for ASEAN countries by 2020. These values reflect the combined impact of the export-productivity, procompetitive, and the FDI-productivity effects (they are decomposed in Appendix 12.3). Here we see that the automotive industries of Thailand and Vietnam have the largest gains, at 4.50 and 3.44 percentage points over the baseline, respectively. Recall in Table 12.1 that the self-sufficiency ratios for these sectors were well below 95%, so the procompetitive effects were active and indeed account for the largest source of efficiency gain in these cases (Appendix 12.3). Although most ASEAN manufacturing industries attain higher efficiency levels due to the Japan-ASEAN FTA, some countries have sectors for which the agreement has no impact. Interestingly, there are even slight reversals in efficiency in a few cases. This happens, for example, in Vietnam’s electrical equipment sector (Table 12.9). In this case, it is the import-productivity (IMP) linkage coupled with the FDI-productivity linkage that gives rise to the technology reversal. Following the FTA, both a drop in foreign investment in the Vietnamese electrical equipment sector and a reorientation of the existing firms toward the domestic market contribute to a slight loss in overall sectoral productivity.
334
Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer Table 12.9. ASEAN sectoral efficiency gains, 2020 (%)
Sector
Thailand Indonesia Malaysia Philippines Vietnam Singapore
Textile/apparel Leather Paper/wood prod. Chemical products Metal products Automotive Machinery Electrical equipment
0.86 0.93 0.83 2.40 1.33 4.50 0.73 0.94
0.08 0.08 0.15 0.32 0.43 0.73 0.22 0.21
0.15 0.08 0.19 0.78 0.27 2.02 0.11 0.10
0.05 0.08 0.28 0.20 0.16 0.78 0.12 0.02
0.06 −0.02 0.04 0.50 0.00 3.44 0.14 −0.06
−0.01 0.02 0.13 −0.05 0.04 −0.02 0.04 0.00
Note: These changes are due solely to the EXP, IMP, and FDI effects.
4.4 Effects on Sectoral Output Finally, we move on to consider changes in sectoral output relating to the Japan-ASEAN FTA. Table 12.10 provides this information for ASEAN in both absolute and relative terms, for the final year of the simulation, 2020. In general, there are output increases in every sector in ASEAN. Electrical equipment shows the largest increase in output over the baseline: $44,451 million, or 10.2 percentage points under the total column. Moving across the columns of Table 12.10, we see that more than half of this increase is related to conventional AGE modeling effects (almost $30 billion, 6.85 percentage points), with the procompetitive effect contributing a difference of $6.4 billion from the baseline (1.46 percentage points), and the FDIproductivity effect contributing $7.9 billion (1.8 percentage points). Chemical products and the automotive sector offer two interesting cases. Here, we observe negative impacts under the standard AGE closure (STD): − $568 million or − 0.32 percentage points, and − $1,805 million or − 2.95 percentage points, respectively). In both of the sectors, the procompetitive effects (IMP) together with the FDI-productivity effects (FDI) are positive and large enough to reverse the overall output changes. Although these additional effects are important particularly for chemical products and the automotive sector, it is nevertheless the conventional AGE effects related to Japan’s tariffs that are the most important reason for the increase in output for most of the sectors. Outside of the manufacturing sectors, output by ASEAN’s food and agricultural sector grows by $3.6 billion (0.96 percentage points) over the baseline, and the corresponding value for the service sector is $32.7 billion (2.04 percentage points) over the baseline. ASEAN’s natural resources sectors have much smaller changes. As with the manufacturing sectors, it is
The Contribution of Productivity Linkages
335
Table 12.10. Sectoral output changes in ASEAN, 2020
Food and ag. Nat. resources Textiles/apparel Leather Paper/wood Chemical products Metal products Automotive Machinery Electrical equipment Services
Total
STD
EXP
IMP
FDI
3,595 (0.96) 311 (0.22) 6,436 (9.14) 1,515 (12.50) 4,051 (3.92) 4,226 (2.38) 4,673 (5.90) 2,228 (3.64) 24,300 (10.45) 44,451 (10.20) 32,702 (2.04)
3,539 (0.95) 228 (0.16) 4,505 (6.40) 1,199 (9.89) 2,635 (2.55) −568 (−0.32) 1,839 (2.32) −1,805 (−2.95) 15,632 (6.72) 29,865 (6.85) 19,083 (1.19)
2 (0.00) 14 (0.01) 84 (0.12) 20 (0.17) 145 (0.14) 185 (0.10) 195 (0.25) 163 (0.27) 668 (0.29) 347 (0.08) 714 (0.04)
143 (0.04) 46 (0.03) 824 (1.17) 185 (1.53) 570 (0.55) 3,127 (1.76) 1,622 (2.05) 2,445 (4.00) 3,495 (1.50) 6,380 (1.46) 6,540 (0.41)
−89 (−0.02) 23 (0.02) 1,024 (1.45) 111 (0.91) 701 (0.68) 1,483 (0.83) 1,016 (1.28) 1,425 (2.33) 4,505 (1.94) 7,860 (1.80) 6,364 (0.40)
Notes: Values represent differences in 1997 US$ millions from baseline scenario in 2020. Values in parentheses represent the corresponding percentage point differences.
generally the conventional AGE effects (STD) that drive most of the changes in output related to implementation of the FTA. (For a disaggregation of Table 12.10, see also Appendix 12.4.)
5. Conclusions AGE models are extensively used in the evaluation of FTAs, but they have often underpredicted the increases in trade and economic growth that followed FTA implementation (Kehoe 2002). Meanwhile, there have been a surge of new empirical trade studies demonstrating strong correlations between firm productivity, on the one hand, and exporting, importing, and investment, on the other. Because increasing these flows is a key objective of most FTAs, this raises the question: Might these additional productivity linkages have a significant impact on AGE-based analyses of FTAs? To test this hypothesis, we generalize the GDyn model to allow for the
336
Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer
productivity-enhancing effects of import competition, increased exports, and FDI-productivity linkages. We then incorporate the best econometric evidence currently available and proceed to examine the impact of the Japan-ASEAN FTA, signed in 2008. In general, we find that this FTA will result in increases in trade for most sectors of the countries involved and that the welfare of all participating countries will improve. By far the largest proportional gains accrue to Thailand, which currently has rather high bilateral tariffs on its trade with Japan. Importantly, we find that the effects normally captured by standard AGE models still play a key role in driving the results. Our conventional, dynamic AGE model captures more than half of the ensuing GDP and trade changes. Overall, we find that the procompetitive and FDI-productivity linkages were the most important, with the export-productivity linkage playing a minor role. These added effects generally serve to reinforce the direction predicted by the standard AGE model. However, addition of the procompetitive effects does lead to aggregate output increasing instead of falling in the case of the two most imperfectly competitive sectors in the ASEAN region: chemicals and automobiles. Therefore, further refinements of the associated econometric estimates would be very worthwhile. We can think of several ways that our results could be changed by future research. For example, the elasticity of productivity response to FDI employed was only 1.4%, and the estimate concerning the higher productivity of exporting firms was only 8%. It seems likely that these figures may be higher for the specific countries examined in this study, particularly those within ASEAN. Future econometric research concerning these parameters for the specific countries examined would facilitate the analysis of FTAs using the framework developed in this chapter. Additionally, sensitivity analysis concerning these parameters (perhaps based on econometric standard errors) would also aid in the progression of this literature.
APPENDIX 12.1: DERIVATION OF THE EXPORTS-PRODUCTIVITY EXPRESSION We begin with the identity: A O Q O ≡ A D Q D + A X Q X , where AO is an index of an industry’s technology, AD represents technology used by local market firms, and AX is technology used by export firms. QO, QD , and QX indicate the total output, the output for the domestic market, and output for export. Normalize such that A D ≡ 1, and let δ ≡ A X /A D . We then rewrite the above identity as A O Q O = Q D + δQ X . We totally differentiate to get
The Contribution of Productivity Linkages
337
A O dQ O + Q O dA O = dQ D + δdQ X . Divide through by A O Q O , let s D ≡ Q D /Q O and s X ≡ Q X /Q O , and multiply both sides by 100 to get qˆ D s D qˆ X s X ˆq O + aˆ O = +δ , (12.A1) AO AO where the lowercase symbols with hats refer to percentage changes (e.g., qˆ O = (dQ O /Q O ) × 100%). Based on the earlier identity we can also derive A O = s D + δs X . Using this we can restate (12.A1): aˆ O =
s D qˆ D + δs X qˆ X − qˆ O . s D + δs X
(12.A2)
Using the identity Q O ≡ Q D + Q X we can totally differentiate and show that qˆ O = s D qˆ D + s X qˆ X , which can be plugged into (12.A2). With algebraic manipulation and the fact that s D + s X = 1, we obtain the equation for the rate of change in overall productivity as a function of the productivity differential between exporters and domestic firms and the differential growth in these two markets for output: efficiency = aˆ O =
(δ − 1)s X s D (qˆ X − qˆ D ) . s D + δs X
(12.A3)
APPENDIX 12.2: AGGREGATION OF GTAP 5 DATA BASE SECTORS No.
Sectors in this study
57 GTAP sectors
1 2 3
Rice Grains Othcrops
4
Meat
5
Othfood
Paddy rice, Processed rice Wheat, Cereal grains nec Veg., fruit, nuts; Oil seeds; Sugar; Fibers; Crops nec; Wool, silk-worm cocoons Cattle, sheep, goats; Animal products nec; Meat products nec Raw milk; Veg. oils and fats; Dairy; Sugar; Food products nec; Bev & tobacco Forestry Fishing Coal, Oil, Gas, Minerals nec Textiles, wearing apparel Leather products Wood products; Paper products, publishing Petroleum, coal products; Chemical, rubber, plastic prods; Mineral products nec
6 7 8 9 10 11 12
Forestry Fish Extract Texwap Leather Paperwood Chemicals
(continued)
338
Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer
APPENDIX 12.2 (continued) No.
Sectors in this study
13 14 15
Metals Autos Machinery
16 17
Electrequip Othservice
18 19 20 21 22 23
Construction Trade Transport Comm Insfinance Pubservice
57 GTAP sectors Ferrous metals; Metals nec; Metal products Motor vehicles and parts Transport equipment nec, Machinery and equipment nec, Manufactures nec Electronic equip. Electricity; Gas; Water; Business services nec; Recr. and oth. services; Dwellings Construction Wholesale/retail trade Transport nec, Sea transport, Air transport Communication Financial services nec, Insurance PubAdmin/Defence/Health/Educat
Notes: “Food & agriculture” is 1–5, “Natural resources” is 6–8, “Light manufactures” is 9–11, and “High tech” is 12–16.
APPENDIX 12.3: AGGREGATION OF GTAP 5 DATA BASE REGIONS No.
Regions
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Japan Korea Malaysia Philippines Indonesia Vietnam Thailand Singapore Taiwan HongKong China USA Canada Mexico AusNzl CSAmerica
66 GTAP regions Japan Korea Malaysia Philippines Indonesia Vietnam Thailand Singapore Taiwan (China) Hong Kong China USA Canada Mexico Australia, New Zealand Central Am., Carib, Colombia, Peru, Venezuela, Argentina, Brazil, Chile, Uruguay, Rest of South America
The Contribution of Productivity Linkages
No.
Regions
66 GTAP regions
17
WEuro
18
SAsia
19
ROW
Austria, Belgium, Denmark, Finland, France, Germany, United Kingdom, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain, Sweden, Switzerland, Rest of EFTA Bangladesh, India, Sri Lanka, Rest of South Asia Hungary, Poland, Rest of Cent. Eur., Former S.U., Turkey, Rest of Mid-East, Moroc., Rest of N. Africa, Bots., Rest of SACU, Malawi, Moz., Tanz., Zam., Zimb., Other S. Africa, Uganda, Rest of Sub-Saha. Afr., Rest of World
339
APPENDIX 12.4: BASELINE POLICY SHOCKS Period 1997–2000
1997–2005
2002–2007
Import tariff adjustments 1. UR tariff reductions for all regions except China and Taiwan (China) (no shocks to agriculture) 2. Pre-WTO tariff reductions undertaken by China before 2002 ASEAN’s Common Effective Preferential Tariff (CEPT) reduction program (1997–2003) and Japan-Singapore Free Trade Agreements (2002) UR tariff reductions for all regions; Taiwan (China) and China’s WTO agreement included (no shocks to agriculture, except for Taiwan (China) and China
Export tax adjustments
USA and EU quotas increased on exports of textiles and wearing apparel for all regions except Taiwan (China) and China
USA and EU quotas increased on exports of textiles and wearing apparel for all regions including Taiwan (China) and China
Notes: Japan-Singapore FTA and CEPT are added to the baseline originally developed in Chapter 5. Their study is otherwise the source that should be consulted concerning the baseline.
340
Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer
References ASEAN. 2002. Unemployment. Available from www.aseansec.org. Aw, B. A., S. Chung, and M. J. Roberts. 2000. “Productivity and Turnover in the Export Market: Micro-Level Evidence from the Republic of Korea and Taiwan (China).” World Bank Economic Review 14(1), 65–90. Bernard, A. B. and J. B. Jensen. 2001. “Exporting and Productivity: The Importance of Reallocation.” Unpublished manuscript. Blalock, G. 2001. “Technology from Foreign Direct Investment: Strategic Transfer through Supply Chains.” Unpublished manuscript, University of CaliforniaBerkeley. Brooke, J. 2002. “Rumor’s of Japan’s Recovery Are, It Seems, Exaggerated.”New YorkTimes, Dec. 16. Chuang, Y. C. and C. M. Lin. 1999. “Foreign Direct Investment, R&D and Spillover Efficiency: Evidence from Taiwan’s Manufacturing Firms.” Journal of Development Studies 35(April), 117–37. Devarajan, S. and D. Rodrik. 1991. “Procompetitive Effects of Trade Reform: Results from a CGE Model of Cameroon.” European Economic Review 35, 1157–84. Dimaranan, B. V. and R. A. McDougall. 2002. Global Trade, Assistance, and Production: The GTAP 5 Data Base. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Dixit, A. K. and J. E. Stiglitz. 1977. “Monopolistic Competition and Optimum Product Diversity.”American Economic Review 67, 297–308. The Economist. 2002. “Unemployment.” Available from www.economist.com. Francois, J. F. 1998. Scale Economies and Imperfect Competition in the GTAP Model. GTAP Technical Paper No. 14. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Francois, J. F. and D. W. Roland-Holst. 1997. “Scale Economies and Imperfect Competition.” In J. Francois and K. Reinert (eds.), Applied Methods for Trade Policy Analysis: A Handbook (pp. 331–63). New York: Cambridge University Press. Francois, J. F. and C. R. Shiells. 1994. “AGE Models of North American Free Trade.” In J. F. Francois and C. R. Shiells (eds.), Modeling Trade Policy: Applied General Equilibrium Assessments of North American Free Trade (pp. 3–44). New York: Cambridge University Press. Hallward-Driemeier, M., G. Iarossi, and K. Sokoloff. 2002. Exports and Manufacturing Productivity in East Asia: A Comparative Analysis of Firm-Level Data. NBER Working Paper No. 8894. Washington, DC: NBER. Harris, R. G. 1984. “Applied General Equilibrium Analysis of Small Open Economics with Scale Economics and Imperfect Competition.” American Economic Review 74, 1016–32. Hertel, T. W. 1994. “The ‘Procompetitive’ Effects of Trade Policy Reform in a Small, Open Economy.” Journal of International Economics 36, 391–411. Hertel, T. W. (ed.). 1997. Global Trade Analysis: Modeling and Applications. Cambridge: Cambridge University Press. Hertel, T.W., T. L. Walmsley, and K. Itakura 2001. “Dynamic Effects of the New Age Free Trade Agreement between Japan and Singapore.” Journal of Economic Integration 16(4), 446–84.
The Contribution of Productivity Linkages
341
Ianchovichina, E. I. 1994. The Procompetitive Effects of Foreign Competition on Optimal Markups in the Presence of Imperfect Competition. M.S. thesis, Purdue University. Ianchovichina, E., J. Binkley, and T. W. Hertel. 2000. “Procompetitive Effects of Foreign Competition on Domestic Markups.” Review of International Economics 8(1), 134–48. Kehoe, T. J. 2002. An Evaluation of the Performance of Applied General Equilibrium Models of the Impact of NAFTA. Research Department Staff Report. Minneapolis: Federal Reserve Bank of Minneapolis. Levinsohn, J. 1993. “Testing the Imports-as-Market-Discipline Hypothesis.” Journal of International Economics 35(1/2), 1–22. Markusen, J. R. 1981. “Trade and the Gains for Trade with Imperfect Competition. Journal of International Economics 11, 531–51. Markusen, J. R. and A. J. Venables. 1988. “Trade Policy with Increasing Returns and Imperfect Competition: Contradictory Results from Competing Assumptions.” Journal of International Economics 24, 299–316. Rodrik, D. 1999. The New Global Economy and Developing Countries: Making Openness Work. Overseas Development Council Policy Essay No. 24. Washington, DC: Johns Hopkins Press. Sj¨oholm, F. 1999. “Exports, Imports, and Productivity: Results from Indonesian Establishment Data.”World Development 27(4), 705–15. World Trade Organization (WTO). 2002. “Regionalism.” Available from http://www. wto.org.
THIRTEEN
Global Demographic Change, Labor Force Growth, and Economic Performance Rod Tyers and Qun Shi
1. Introduction Recent changes in global demographic behavior, including to fertility, mortality, migration, and the sex ratio at birth, have been considerable and many were not widely anticipated in recent decades. In most countries, consistent with the central phase of the global demographic transition, infant mortality fell through the course of the last century and adult life expectancy increased, causing a surge in population growth. The declines in birth rates as part of the final phase of this transition have been particularly sharp, first in developed countries and recently in many developing countries.1 Before this century is half over, populations in Japan and some European countries are likely to be smaller than they were in 1990, with these declines in total populations being preceded by declines in the number and proportion of people of working age.2 The economic implications of these demographic trends and uncertainties are the subject of an already substantial global literature.3 Recent 1 2 3
IMF (2004: Chapter 3), Lee (2003). Bryant and McKibbin (1998), United Nations (2003). At minimum, this literature spans demography (Booth et al. 2002; McDonald and Kippen 2001), population economics (Lee 2003; Mason 2003), public economics (OECD 1996, 1998), economic history (Bloom and Williamson 1997; Williamson 1998), and growth economics (Barro and Becker 1989).
Funding for the research described in this chapter is from Australian Rural Industries Research and Development Corporation Research Contract No. ANU-51A and from Australian Research Council Discovery Grant No. DP0557889. Thanks are due to Heather Booth, Siew Ean Khoo, and Ming Ming Chan for helpful discussions about the demography; to Jeff Davis, Brett Graham, Ron Duncan, Robert McDougall, and Hom Pant for their constructive comments on the economic analysis; and to Terrie Walmsley for technical assistance with the GTAP Data Base, as well as useful discussions on the subject of baseline simulations. Useful comments were also received at the Conference on Global Economic Analysis in June 2005 from Dominique van der Mensbrugghe and Nico van Leeuwen. Jahnvi Vedi and Jyoti Pant provided research assistance in the later stages of the study.
342
Global Demographic Change
343
macroeconomic studies of demographic change have been global in scope, emphasizing the effects of aging on average saving rates and financial flows (Bryant and McKibbin 1998, 2001; Bryant et al. 2003; Faruqee and Muhleisen 2002).4 This work has clearly demonstrated the substantial implications of demographic change in some regions for economic performance in others. It has, however, fallen short of the complete demographic modeling needed to capture the three principal avenues through which demographic change influences economic performance: labor force growth,5 average saving rates, and age-specific consumption variation. This chapter examines the economic implications of population change using a complete demographic model on 14 regions, which is constructed as integral with a dynamic model of the global economy. The latter model is a development of GDyn in which regional households are disaggregated by age group and gender. Our explicit incorporation of the demographic submodel allows age-gender distributions, migration flows, and, therefore, labor force participation rates, migration rates, average saving rates, and the age-gender effects on consumption to be endogenized. It is constructed around a baseline projection through 2030 in which populations and labor forces are projected to decline in Europe and Japan and to begin declining before the end of this period in China and elsewhere in East Asia. Notably, as age distributions change, the trends in labor forces are shown to diverge substantially from those in total populations. The behavior of the model is then illustrated by considering the effects of one alternative demographic scenario. In this case, we imagine that continued improvements in public health and medical science cause life expectancy beyond 60 to grow faster than anticipated in all regions of the world. This scenario causes significant departures from the baseline and has important implications for overall economic performance. In models of the Solow-Swan type, where endogenous growth takes the form of physical capital accumulation, the decelerating population growth that is prominent in our baseline projection reduces the GDP growth rate but raises that of real per capita income. Multiple trading regions 4
5
Much of this research was organized under a project coordinated by the Brookings Institution in the United States and involving staff from the International Monetary Fund. Financing is from the Economic and Social Research Institute of the Japanese Cabinet Office. Capturing the demographic influences on the labor force change requires consideration of age- and gender-specific participation rates and rates of part-time employment, as well as age- and gender-specific migration rates.
344
Rod Tyers and Qun Shi
notwithstanding, this tendency is prominent in our model, too. There are two complicating effects of population deceleration, however. First, age distributions change so that the average age rises, which alters the pattern of international financial flows and the distribution of global investment. In addition, because consumption preferences vary with age, the pattern of consumption also changes. Second, on the supply side, changes in the size and composition of the population correspondingly change the size and composition of the labor force, although as we show, these changes are most often far from proportionate. Increased longevity complicates this picture by both accelerating the aging process and raising population and labor force growth. Although increased longevity raises aged dependency everywhere, it tends to attract new investment to those regions whose elderly have high labor force participation rates. In Section 2, we introduce the demographic submodel and discuss its population and labor force projections. We describe the extension of the GDyn model to incorporate populations disaggregated by age and gender and then the full demographic submodel. Section 3 discusses the construction of the baseline scenario. The “accelerated aging” scenario is described in Section 4, and the implications for the performance of the global economy are quantified through a comparison of its simulation results with the baseline. Section 5 offers brief concluding remarks.
2. Modeling Global Demographic Change The approach adopted follows Tyers et al. (2005), in that it applies a complete demographic submodel that is integrated within a dynamic numerical model of the global economy.6 The economic model is a development of GDyn, the standard version of which has single households in each region and therefore no demographic structure.7 The version used has regional households that are disaggregated by age group, gender, and skill level.
2.1 Demography The demographic submodel tracks populations in four age groups and two genders, yielding a total of eight population groups in each of 14 regions.8 6 7 8
See also Shi and Tyers (2004) and Tyers et al. (2005). Earlier applications of the GDyn model to the issues raised in this paper include those by Shi and Tyers (2004) and Duncan, Shi and Tyers (2005). The demographic submodel has been used in a stand-alone mode for the analysis of trends in dependency ratios. For a more complete documentation of the submodel, see Chan and Tyers (2006).
Global Demographic Change
345
The four age groups are the dependent young, adults of fertile and working age, older working adults, and the mostly retired over 60s. The resulting age-gender structure is displayed in Figure 13.1. The population is further divided between households that provide production labor and those providing professional labor.9 Each age-gender-skill group is a homogeneous subpopulation with group-specific birth and death rates and rates of both immigration and emigration.10 If the group spans T years, the survival rate to the next age group is the fraction 1/T of its population, after groupspecific deaths have been removed and its population has been adjusted for net migration. The final age group (60+) has duration equal to measured life expectancy at 60, which varies across genders and regions. The key demographic parameters, then, are birth rates, sex ratios at birth, age- and gender-specific death, immigration and emigration rates, and life expectancies at 60.11 Another key parameter is the rate at which each region’s education and social development structure transforms production worker families into professional worker families. Each year a particular proportion of the population in each production worker age-gender group is transferred to professional status. These proportions depend on the regions’ levels of development, the associated capacities of their education systems, and the relative sizes of the production and professional labor groups. In any year, for each age group a, gender group g, skill group s, region of origin r, and region of destination d, the volume of migration flow is t t R t Ma,g ,s,r,d = δd M a,g ,s,r,d N a,g ,s,d ,
∀a, g , r, d,
(13.1)
where δtd is a destination-specific factor reflecting immigration policy in R region d, set to unity in all but counterfactual simulations, and Ma,g ,s,r,d is the migration rate between r and d expressed as a proportion of the group population in region d, Na,g ,s,d . Given the migration matrix, Ma,g ,s,r,d , the population in each age, gender, and skill group and region can be constructed. We begin with the population 9
10
11
The subdivision between production and professional labor accords with the ILO’s occupation-based classification and is consistent with the labor division adopted in the GTAP Data Base. See Liu et al. (1998). Mothers in families providing production labor are assumed to produce children who will grow up to also provide production labor, whereas the children of mothers in professional families are correspondingly assumed to become professional workers. Immigration and emigration are also age and gender specific. The model represents a full matrix of global migration flows for each age and gender group. Each of these flows is currently set at a constant proportion of the population of its destination group. See Chan and Tyers (2006) and Vedi (2005) for further details.
346
Rod Tyers and Qun Shi Female population
Male Population D
D
Mo
Mo
Aged >60
Mi
Aged >60
Mi
S S D Working 40-60
Mo
D Working 40-60
Mi
S Mo
D
Mi
Working fertile 15-40
B
S D Mi
Deaths Survival Births Immigration Emigration Sex ratio at birth Figure 13.1. The demographic submodel.
Mo
D
Young 0-15
Glossary:
Mo
Mi
Working fertile 15-40
S
Mo
SRB
D S B Mi Mo SRB
Mi
S D
Young 0-15
Mo
Mi
Global Demographic Change
347
of males aged 0–14 from professional families in region d (a = 014, g = m, s = sk, r = d). S dt B t N t−1 1 + S dt sk,d 1539,m,sk,d t−1 t − D t014,m,sk,d N014,m,sk,d + M014,m,sk,r,d r t−1 t − M014,m,sk,d,r + ρd N014,m,unsk,d
t−1 t N014,m,sk,d = N014,m,sk,d +
r
1 t−1 t−1 − N014,m,sk,d − D t014,m,sk,d N014,m,sk,d , 15
∀d (13.2)
where S dt is the sex ratio at birth (the ratio of male to female births) in region d, B tsk,d is the birth rate, D t014,m,sk,d the death rate, and ρd is the rate at which region d’s educational institutions and general development transform production into professional worker families. The final term is survival to the corresponding 15–39 age group. In the corresponding equation for young males from production worker families the penultimate term is negative. For females in professional families in this age group the corresponding equation is 1 B t N t−1 1 + S dt sk,d 1539,f ,sk,d t−1 t − D t014,f ,sk,d N014,f M014,f ,sk,r,d ,sk,d + r t−1 t − M014,f ,sk,d,r + ρd N014,f ,unsk,d r 1 t−1 t−1 N014,f ,sk,d − D t014,f ,sk,d N014,f − ,sk,d , 15
t−1 t N014,f ,sk,d = N 014,f ,sk,d +
∀d. (13.3)
For adults of gender g from professional families in the age group 15–39, the equation includes a survival term from the younger age group: 1 t−1 −D t N t−1 N 15 014,g ,sk,d 014,g ,sk,d 014,g ,sk,d t−1 t M1539,g − D t1539,g ,sk,d N1539,g ,sk,r,d ,sk,d + r t−1 t − M1539,g ,sk,d,r + ρd N1539,g ,unsk,d r 1 t−1 t−1 t ∀g , d N1539,g − ,sk,d − D 1539,g ,sk,d N 1539,g ,sk,d , 25
t−1 t N1539,g ,sk,d = N 1539,g ,sk,d +
(13.4)
348
Rod Tyers and Qun Shi
where the second term is the surviving inflow from the 0–14 age group and the final term is the surviving outflow to the 40–59 age group. Again, the skill transformation term is negative in the case of the corresponding equation for production worker families. The population of professional adults of gender g in age group 40–59 follows: 1 t−1 t−1 N1539,g ,sk,d − D t1539,g ,sk,d N1539,g ,sk,d 25 t−1 t M4059,g − D t4059,g ,sk,d N4059,g ,sk,r,d ,sk,d + r t−1 t − M4059,g ,sk,d,r + ρd N 4059,g ,unsk,d r 1 t−1 t−1 ∀g , d. (13.5) − N4059,g ,sk,d − D t4059,g ,sk,d N4059,g ,sk,d , 20
t−1 t N4059,g ,sk,d = N 4059,g ,sk,d +
and for adults in the 60+ age group, the corresponding relationship is 1 t−1 t−1 t−1 t N60+,g N4059,g ,sk,d − D t4059,g ,sk,d N4059,g ,sk,d = N 60+,g ,sk,d + ,sk,d 20 t−1 t t M60+,g M60+,g + ,sk,r,d − ,sk,d,r + ρd N 60+,g ,unsk,d r
−
r
1 L t60+,g ,sk,d
t−1 N60+,g ,sk,d ,
∀g , d,
(13.6)
where the final term indicates that deaths from this group each year depend on its life expectancy at 60, L t60+,g ,sk,d . Again, the equation for aged production worker family members is the same except that the skill transformation term is negative.
Sources and Structure R Key parameters in the model are the migration rates Ma,g ,s,r,d , birth rates t t t B s,r , sex ratios at birth S r , death rates D a,g ,s,r , life expectancies at 60 L t60+,g ,s,r and the skill transformation rates ρd . The migration rates are based on recent migration records and are held constant through time.12 The skill transformation rates are based on changes during the decade prior to the base year, 1997, in the composition of aggregate regional labor forces as between production and professional workers. These are also held constant through time.13 12 13
The migration rates and the corresponding birth rates are listed in detail in Chan and Tyers (2006: Tables 2–5) and Vedi (2005). Note that, as regions become more advanced and populations in the production worker families become comparatively small, the skill transformation rate has a diminishing effect on the professional population.
Global Demographic Change
349
Asymptotic Trends in Other Parameters The birth rates, life expectancy at 60, and the age-specific mortality rates all trend through time asymptotically. For each age group a, gender group g, and region r, a target rate is identified.14 The parameters then approach these target rates with initial growth rates determined by historical observation. In year t the birth rate of region r is (13.7) B tr = B 0r + B 0Tg t − B 0r 1 − e β t , where the rate of approach β is calibrated from the historical growth rate: 0 B Tg t − B 0r 1 − e β B 1r − B 0r 0 B x¨ r = = , (13.8) P r0 B 0r so that
B 0r Pˆ r0 β = ln 1 − 0 . B Tg t − B 0r
(13.9)
The birth rates and death rates, thus calculated, are summarized in Tables 13.1 and 13.2. The corresponding life expectancies at 60 are listed in Table 13.3.
Labor Force Projections To evaluate the number of full-time equivalent (FTE) workers we first construct labor force participation rates, Pa,g,r by gender and age group for each region from ILO statistics on the “economically active population.” We then investigate the proportion of workers who are part time and the hours they work relative to each regional standard for full-time work. The result is the number of FTEs per worker, Fa,g,r . The labor force in region r is then L¯ tr =
f unsk 60+
L ta,g ,s,r
t t where L ta,g ,s,r = μta,r P a,g ,r F a,g ,r N a,g ,s,r .
a=1539 g =m s=sk
(13.10) Here is a shift parameter reflecting the influence of policy on partict ipation rates. The time superscript on P a,g ,r refers to the extrapolation of 15 observed trends in these parameters. μta,r
14 15
In this discussion the skill index, s, is omitted because birth and death rates and life expectancies at 60 do not vary by skill category in the version of the model used. Although part-time hours may well also be trending through time, we hold F constant in the current version of the model.
350
Rod Tyers and Qun Shi Table 13.1. Birth rates and the sex ratio at birth Fertility Total Total Birth: births/ completed Birth: births/ completed Sex ratio 1000 women fertility: 1000 women fertility: at birth aged 15–39a children per aged 15–39a children per male/female (1997) woman (1997) (2030) woman (2030)
Australia North America Western Europe Central Europe, FSU Japan China Indonesia Other East Asia India Other South Asia South America Mid East Nth Africa Sub-Saharan Africa Rest of world
1.05 1.05 1.06 1.06 1.06 1.10 1.05 1.06 1.05 1.05 1.05 1.05 1.03 1.05
72 88 61 51 57 76 104 99 139 160 105 137 180 128
1.8 2.2 1.5 1.3 1.4 1.9 2.6 2.5 3.5 4.0 2.6 3.4 4.5 3.2
70 85 53 51 57 58 91 90 115 144 101 130 178 102
1.7 2.1 1.3 1.3 1.4 1.4 2.3 2.3 2.9 3.6 2.5 3.3 4.5 2.6
a
Birth rates are based on UN estimates and projections as represented by the U.S. Bureau of the Census. The latter representation has annual changes in rates, whereas the UN model has them stepped every 5 years. Aggregation is by population-weighted average. Initial birth rates are obtained from the UN model by dividing the number of births per year by the number of females aged 15–39. These rates change through time according to annualized projections by the U.S. Bureau of the Census. Source: Aggregated from United Nations (2003), U.S. Department of Commerce, U.S. Bureau of the Census “International Data Base.”
Asymptotic Trends in Labor Force Participation For each age group a, gender group g, and region r, a target country is identified whose participation rate is approached asymptotically. The rate of this approach is determined by the initial rate of change. Thus, the participation rate takes the form 0 t 0 0 βt (13.11) P a,g ,r = P a,g ,r + P Tg t − P a,g ,r (1 − e ), where the rate of approach, β, is calibrated from the initial participation growth rate: 0 0 t 0 P (1 − e β ) − P − P P a,g ,r Tg t a,g ,r a,g ,r 0 = , (13.12) Pˆ a,g ,r = 0 0 P a,g P a,g ,r ,r
Global Demographic Change
351
Table 13.2. Age- and gender-specific death rates a (deaths per year per thousand in each age
and gender group) 0–14 Males
15–39
Females
Males
40–59
Females
Males
Females
1997 2030 1997 2030 1997 2030 1997 2030 1997 2030 1997 2030 Australia 1.6 0.6 1.3 0.5 1.3 North America 2.2 0.6 1.8 0.6 1.7 Western Europe 1.8 0.6 1.4 0.6 1.2 Central Europe, FSU 2.0 0.8 1.8 0.6 2.1 Japan 1.2 0.7 1.0 0.7 0.7 China 1.1 0.5 0.9 0.5 0.8 Indonesia 1.4 1.1 1.1 0.9 2.3 Other East Asia 1.4 0.7 1.1 0.6 2.3 India 8.2 3.8 9.4 4.5 1.3 Other South Asia 8.2 3.6 9.4 4.2 1.3 South America 1.8 1.4 1.4 1.0 1.3 Mid East Nth Africa 6.7 1.5 6.5 1.9 1.3 Sub-Saharan Africa 10.1 14.8 7.7 11.4 1.3 Rest of World 1.7 0.7 1.4 0.6 1.3
1.1 0.8 1.1 1.6 0.6 0.6 2.0 2.0 1.1 1.4 1.1 0.9 1.7 0.8
0.5 0.7 1.1 0.9 0.4 0.3 2.0 1.0 2.4 3.1 0.7 1.5 1.3 0.7
0.4 3.2 2.8 2 1.5 0.5 4.5 3.6 2.7 2.2 1.0 4.4 3.2 2.7 2.0 0.7 8.0 6.1 3.3 2.5 0.3 3.5 2.6 2.0 1.4 0.2 3.9 2.8 2.0 1.8 1.6 7.9 6.3 3.9 2.7 1.0 7.6 3.9 3.4 2.3 2.1 12.3 7.6 8.5 5.7 3.1 10.8 9.9 10.3 9.5 0.4 4.3 3.0 2.3 2.0 1.5 8.3 8.0 5.0 3.6 1.3 30.0 29.6 30.0 28.4 0.5 5.4 3.6 2.8 1.9
a
Aggregation is by population-weighted average. Projections of these parameters to 2020 assume convergence on target rates observed in comparatively “advanced” countries, as explained in the text. Only the endpoint values are shown here, but the model uses values that change with time along the path to convergence. Source: Aggregated from United Nations (2000) and WHO (2003).
so that
β = ln 1 −
0 ˆ0 P a,g ,r P a,g ,r 0 0 P Tg t − P a,g ,r
.
(13.13)
Target rates are chosen from countries considered “advanced” in terms of trends in participation rates. Where female participation rates are rising, therefore, Norway provides a commonly chosen target because its female labor force participation rates are higher than for other countries.16
Accounting for Part-Time Work For each age group a, gender g, and region r, full-time equivalency depends on the fraction of participants working full time, fa,g,r , and, for those working part time, the ratio of average part-time hours to full-time hours for that 16
The resulting participation rates are listed by Chan and Tyers (2006: Table 10).
352
Rod Tyers and Qun Shi Table 13.3. Baseline life expectancy at 60 (years)a Male
Australia North America Western Europe Central Europe, FSU Japan China Indonesia Other East Asia India Other South Asia South America Mid East Nth Africa Sub-Saharan Africa Rest of World
Female
Initial
2030
Initial
2030
17 19 19 15 22 16 14 17 15 15 17 16 10 18
21 21 21 15 27 17 14 18 16 16 19 17 10 22
21 23 23 20 26 18 15 19 18 15 20 18 13 23
26 25 25 21 33 21 15 19 19 17 21 19 13 28
a
Aggregation is by population-weighted average. Projections of these parameters to 2020 assume convergence on target rates observed in comparatively “advanced” countries, as explained in the text. Only the endpoint values are shown here, but the model uses values that change with time along the path to convergence. Source: Aggregated from United Nations (2000b) as presented in Chan and Tyers (2006).
gender group and region, rg,r . For each group, the ratio of FTE workers to total labor force participants is then F a,g ,r = f a,g ,r + (1 − f a,g ,r )r g ,r .
(13.14)
Preliminary estimates of fa,g,r and rg,r are approximated from OECD (1999: Table 1.A.4) and OECD (2002: Statistical Annex, Table F).17
The Aged Dependency Ratio We define and calculate four dependency ratios: (1) a youth dependency ratio is the number of children per FTE worker, (2) an aged dependency ratio is the number of persons over 60 per FTE worker, (3) a nonworking aged dependency ratio is the number of nonworking persons over 60 per 17
No data have yet been sought on part-time work in non-OECD member countries. In these cases the diversity of OECD estimates is used to draw parallels between countries and regions and thus to make educated guesses. The results are listed by Chan and Tyers (2006: Tables 11 and 12).
Global Demographic Change
353
FTE worker, and (4) a more general dependency ratio is defined that takes as its numerator the total nonworking population, including children.18 The one of most widespread policy interest is the nonworking aged dependency ratio: f unsk t N60+,g ,sk,r − L t60+,g ,sk,r
= R ANW r,t
g =m s=sk
t
Lr
.
(13.15)
2.2 The Global Economic Model The chapter extends GDyn, presented in Chapter 2. In the version used, the world is subdivided into 14 regions, one of which is China. Industries are aggregated into only three sectors: food (including processed foods), industry (mining and manufacturing), and services. To capture the full effects of demographic change, including those of aging, the standard model has been modified to include multiple age, gender, and skill groups in line with the structure of the demographic submodel. In the adapted model, these 16 groups differ in their consumption preferences, saving rates, and labor supply behavior. Unlike the standard GTAP models, in which regional incomes are split among private consumption, government consumption, and total saving via an upper level Cobb-Douglas utility function that implies fixed regional saving rates, this adaptation first divides regional incomes between government consumption and total private disposable income. The implicit assumption is that governments balance their budgets while private groups save or borrow. Private disposable income is then split among the eight age-gender groups in a manner informed by empirical studies of age- and genderspecific consumption behavior. For each age-gender group we then use a Keynesian consumption equation to split disposable income between saving and consumption expenditure. Group private saving rates then become endogenous, depending on real disposable income and the real interest rate, thereby relaxing the fixed average saving rate assumption in the standard model. Once group consumption expenditures are known, the standard GTAP CDE19 consumption preferences are applied to each, with preference parameters varying to reflect age-gender differences in tastes. Finally, consumption volumes are totaled across groups to obtain the final demand for each product, and consumption expenditures are subtracted from group 18 19
All these dependency ratios are defined in detail by Chan and Tyers (2006). This refers to the “constant difference of elasticities of substitution” demand system. See Hertel (1997) and, in particular, Huff et al. (1997).
354
Rod Tyers and Qun Shi
disposable incomes to obtain group saving levels, which are then totaled across groups to obtain regional saving.
Income Splitting The first step is to split government from private disposable income. For this we retain the original Cobb-Douglas system, this time in a two-way split, and the governments’ income shares from the original database.20 Total regional disposable income is then split among the eight age-gender groups. For this we draw from other studies the distribution of disposable income between age-gender groups for “typical” advanced and developing countries. To ensure that changes in the age-gender distribution of each region’s population alter the corresponding age-gender distribution of income, we define a set of weights, Wa,g ,r , that represent the ratio of the per capita disposable income of group (a, g), to that of the (15–39, m) group, chosen as an arbitrary standard.21 The share of the disposable income of region r enjoyed by people of gender g and age group a is thus D Ya,g ,r
YrD
=
60+
Wa,g ,r Na,g ,r . f Wa,g ,r Na,g ,r
(13.16)
a=0−15 g =m
The adopted values of Wa,g ,r are listed in Table 13.4. Their selection is guided by the empirical studies of the age distribution of income and consumption noted in the table.
Splitting Savings and Consumption Expenditure from Group Disposable Income Our reduced form approach to the intertemporal optimization problem faced by each individual centers on an exponential consumption equation that links group per capita consumption expenditure to per capita disposable income and the real interest rate, r: εc D Ya,g C a,g ,r ,r =A r rεr , (13.17) c a,g ,r = C C Na,g ,r P a,g N P a,g ,r a,g ,r ,r 20 21
This implies the assumption that all governments balance their budgets and that all saving in the original database is private. To date we have not realized the opportunity to have the age-gender distribution of income depend on the income’s factor origin. Despite intuition suggesting a link – the aged in advanced countries receive retirement income stemming from capital ownership – consistent empirical work on this distribution is unavailable.
Global Demographic Change
355
Table 13.4. Income weights, Wa,g ,r , by age-gender group 0–14
Australia North America Western Europe Central Europe, FSU Japan China Indonesia Other East Asia India Other South Asia South America Mid East Nth Africa Sub-Saharan Africa Rest of World
15–39
40–59
60+
Male
Female
Male
Female
Male
Female
Male
Female
0.60 0.40 0.50 0.50 0.60 0.60 0.50 0.60 0.50 0.50 0.40 0.50 0.50 0.60
0.60 0.40 0.50 0.50 0.60 0.60 0.50 0.60 0.50 0.50 0.40 0.50 0.50 0.60
1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
0.86 1.10 1.00 1.00 1.60 1.60 1.40 1.60 1.40 1.40 1.05 1.40 1.40 0.86
0.86 1.10 1.00 1.00 1.60 1.60 1.40 1.60 1.40 1.40 1.05 1.40 1.40 0.86
0.67 0.60 0.70 0.70 0.94 0.94 0.90 0.94 0.90 0.90 1.10 0.90 0.90 0.67
0.67 0.60 0.70 0.70 0.94 0.94 0.90 0.94 0.90 0.90 1.10 0.90 0.90 0.67
Source: Compiled from studies of consumption behavior on particular countries, including U.S. and UK: Attanasio and Banks (1998), Attanasio et al. (1999); Japan: Kitamura et al. (2001: Table 1); Mexico (standard for Latin America and an indicator for some other developing regions): Attanasio and Szekely (1998: Figure 1); New Zealand (standard for Australia and Western Europe): Gibson and Scobie (2001: Figure 1). C where P a,g ,r is a group consumption price index, group consumption expenditure is C a,g ,r and parameters εc and εr are income and interest elasticities. This equation is calibrated for each group and region based on the set of initial (1997) age-specific saving rates from per capita real disposable income listed in Table 13.5. These estimates are drawn from the same empirical studies of the age distribution of income and consumption as the income weights of Table 13.4. They are recalibrated for consistency with the overall private saving rate in each region indicated in the GTAP Data Base.22 22
The elasticities of consumption expenditure to disposable income suggested by the empirical literature seem to be poor choices as reduced forms for saving behavior in the long term because they imply high marginal saving rates. We calibrate these elasticities according to the following scenario: (a) North American per capita disposable income grows at 3%/yr for 100 years; (b) growth in all other regions is sufficient to attain North America’s per capita disposable income levels within that period; (c) when the other regions catch up, all regions attain identical group-specific saving rates; and (d) the income, consumption, and saving transitions are smooth and exponential. Our reduced-form consumption equation (13.17) can be simplified for a single age-gender group to c = A y εc r εr , where c is per capita consumption expenditure, y is per capita disposable income, and r is the real interest rate. To focus on the key elasticity, εc , imagine that the real interest rate is constant through time, so that the interest term can be embedded in the constant. Then, if per capita disposable
356
Rod Tyers and Qun Shi
Table 13.5. Initial saving rates from personal disposable income by age-gender group
(percent) 0–14
Australia North America Western Europe Central Europe, FSU Japan China Indonesia Other East Asia India Other South Asia South America Mid East Nth Africa Sub-Saharan Africa Rest of World
15–39
40–59
60+
Male
Female
Male
Female
Male
Female
Male
Female
0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0
7 14 10 4 24 35 23 36 19 7 7 8 2 5
7 14 10 4 24 35 23 36 19 7 7 8 2 5
31 19 39 18 28 40 34 40 28 10 17 19 6 23
31 19 39 18 28 40 34 40 28 10 17 19 6 23
−5 −30 −20 −6 22 31 23 32 19 7 6 7 2 −5
−5 −30 −20 −6 22 31 23 32 19 7 6 7 2 −5
Source: Compiled from studies of consumption behavior on particular countries, including Mexico: Attanasio and Szekely (1998); Japan: Kitamura et al. (2001); New Zealand: Gibson and Scobie (2001); US: Attanasio et al. (1999).
Consumption Preferences The construction of the CDE demand system requires the calibration of two sets of parameters by the method detailed by Huff et al. (1997). Its advantage over the CES, or constant elasticity of substitution, system is that it is nonhomothetic and therefore allows income elasticities of demand to vary between commodities. Elasticities of demand then depend on CDE “expansion” and “substitution” parameters, which are calibrated for each region’s aggregate household in the GTAP Data Base. We retain the calibrated values of these parameters for the eight age-gender groups. To complete the demand system we then need expenditure shares for each of the eight different age-gender groups in each region. For these shares we turn, once again, to the consumption analysis literature. Studies of consumption preferences by age group are available for a few of the identified countries, and we use those as a guide in the income grows at rate g y , the rate of growth of consumption expenditure is εc g y . If per capita consumption expenditure is initially c 0 and per capita disposable income is initially y 0 , we can calculate the group’s average saving rate in period t and invert the resulting expression to find the elasticity that is consistent with the target saving rate after T years: T εc = 1 + g y1T ln( 1−s ). For further details, see Tyers et al. (2005). 1−s 0
Global Demographic Change
357
construction of the complete matrix of expenditure shares listed in Table 13.6. The study by Weber et al. (2002) is the most detailed, and it shows only very modest variation in expenditure shares by age group when commodities are highly aggregated. Although there is considerable variation when comparisons are at a high level of detail, such as between fresh food and restaurant meals or between health and other services, the broad shares are remarkably similar.23 For presentational economy, we focus in this chapter on the three-product version.24 Age-gender group expenditure shares are drawn initially from the literature (indicated in Table 13.6) and then rendered consistent with group expenditures on the one hand and GTAP Data Base values for aggregate expenditure shares on the other by using RAS techniques to concord the shares with row and column sums in the matrix of expenditures.
Elasticities of Substitution It is well known that general equilibrium simulation results are particularly sensitive to the assumed degree of differentiation between home and foreign goods and services. In models such as this one, in which products are highly aggregated, some of this differentiation reflects regional differences in sectoral product composition. Both the complementarity of product compositions and true regional product differentiation are therefore represented in the model via the choice of the elasticities of substitution between home and foreign products. Controversy has raged over the merits of various estimates, and the view is commonly expressed that the “standard” GTAP estimates, which range between 1.0 and 4.0, are too small. We agree, because when these elasticities are used, our baseline simulation yields substantial divergence between the paths of home and trading prices in different regions. In the absence of any new trade distortions, global markets appear far more segmented in 2030 than they are at present. Newer estimates by Harrigan (1995), Trefler and Lai (1999), and Obstfeld and Rogoff (2000) all support much higher values. We therefore use 7.0 for food products, 4.0 for manufactures, and 2.2 for the less tradable services, and we retain the traditional “rule of two” for substitutability of imports by region of origin. 23
24
It is of concern that some expenditure shares for detailed products and services appear to be changing very rapidly through time. Weber et al. (2002) show that the health share is rising rapidly for the aged and that this is associated with very rapid growth in the share of expenditure on drugs by all groups but particularly the aged. The GTAP commodity classification is production oriented, based on the International Standard Industrial Classification (ISIC), and so it differs from the classification used in expenditure surveys. We use the GTAP commodities throughout, weakening the sensitivity of our analysis to differences in preferences.
358
Rod Tyers and Qun Shi Table 13.6. Private expenditure shares by age-gender group 0–14
15–39
40–59
60+
Male
Female
Male
Female
Male
Female
Male
Female
Australia Food Manufactures Services
0.18 0.07 0.74
0.18 0.07 0.74
0.10 0.19 0.71
0.10 0.19 0.71
0.10 0.19 0.71
0.10 0.19 0.71
0.18 0.05 0.77
0.18 0.05 0.77
North America Food Manufactures Services
0.12 0.12 0.76
0.12 0.12 0.76
0.05 0.16 0.79
0.05 0.16 0.79
0.06 0.16 0.78
0.06 0.16 0.78
0.11 0.09 0.80
0.11 0.09 0.80
Western Europe Food Manufactures Services
0.18 0.12 0.70
0.18 0.12 0.70
0.09 0.30 0.61
0.09 0.30 0.61
0.09 0.30 0.61
0.09 0.30 0.61
0.18 0.09 0.73
0.18 0.09 0.73
Central Europe, FSU Food 0.44 Manufactures 0.10 Services 0.47
0.44 0.10 0.47
0.26 0.27 0.47
0.26 0.27 0.47
0.26 0.27 0.47
0.26 0.27 0.47
0.43 0.07 0.50
0.43 0.07 0.50
Japan Food Manufactures Services
0.18 0.07 0.75
0.18 0.07 0.75
0.10 0.18 0.72
0.10 0.18 0.72
0.10 0.18 0.72
0.10 0.18 0.72
0.17 0.05 0.78
0.17 0.05 0.78
China Food Manufactures Services
0.47 0.13 0.40
0.47 0.13 0.40
0.26 0.35 0.39
0.26 0.35 0.39
0.26 0.35 0.39
0.26 0.35 0.39
0.47 0.10 0.43
0.47 0.10 0.43
Indonesia Food Manufactures Services
0.46 0.07 0.48
0.46 0.07 0.48
0.30 0.26 0.44
0.30 0.26 0.44
0.30 0.26 0.44
0.30 0.26 0.44
0.45 0.05 0.50
0.45 0.05 0.50
Other East Asia Food Manufactures Services
0.30 0.10 0.60
0.30 0.10 0.60
0.17 0.35 0.47
0.17 0.35 0.47
0.17 0.35 0.47
0.17 0.35 0.47
0.29 0.08 0.63
0.29 0.08 0.63
India Food Manufactures Services
0.57 0.08 0.35
0.57 0.08 0.35
0.37 0.31 0.32
0.37 0.31 0.32
0.37 0.31 0.32
0.37 0.31 0.32
0.56 0.06 0.38
0.56 0.06 0.38
Other South Asia Food Manufactures Services
0.54 0.07 0.39
0.54 0.07 0.39
0.37 0.27 0.36
0.37 0.27 0.36
0.37 0.27 0.36
0.37 0.27 0.36
0.54 0.05 0.41
0.54 0.05 0.41
(continued)
Global Demographic Change
359
Table 13.6 (continued) 0–14
15–39
40–59
60+
Male
Female
Male
Female
Male
Female
Male
Female
0.36 0.10 0.53
0.36 0.10 0.53
0.21 0.36 0.43
0.21 0.36 0.43
0.21 0.36 0.43
0.21 0.36 0.43
0.36 0.08 0.57
0.36 0.08 0.57
Mid East Nth Africa Food 0.39 Manufactures 0.07 Services 0.54
0.39 0.07 0.54
0.25 0.27 0.48
0.25 0.27 0.48
0.25 0.27 0.48
0.25 0.27 0.48
0.38 0.05 0.57
0.38 0.05 0.57
Sub-Saharan Africa Food 0.46 Manufactures 0.07 Services 0.47
0.46 0.07 0.47
0.30 0.28 0.42
0.30 0.28 0.42
0.30 0.28 0.42
0.30 0.28 0.42
0.45 0.05 0.50
0.45 0.05 0.50
Rest of World Food Manufactures Services
0.36 0.10 0.54
0.20 0.27 0.53
0.20 0.27 0.53
0.20 0.27 0.53
0.20 0.27 0.53
0.35 0.07 0.58
0.35 0.07 0.58
South America Food Manufactures Services
0.36 0.10 0.54
Source: Constructed with guidance from the results presented by: Abdel-Ghany and Sharpe (1997), Blisard (2001a and b), Blisard et al. (2003), Case and Deaton (2002), Paulin (2000), Regmi et al. (2001) and Weber et al. (2002). The shares are then modified using a RAS process to conform with aggregate expenditures by product in the GTAP Data Base.
3. Constructing the Baseline Scenario The baseline scenario represents a “best judgment” projection of the global economy through 2030. Although policy analysis can be sensitive to the content of this scenario, our focus is on the extent of departures from it that would be caused by changes in demographic scenarios. Nonetheless, it is instructive to describe the baseline for two reasons. First, all scenarios examined have in common a set of assumptions about future trends in productivity, and second, some exposition of the baseline makes the construction of departures from it clearer.
3.1 Exogenous Factor Productivity Growth Exogenous sources of growth enter the model as factor productivity growth shocks, applied separately for each of the model’s five factors of production (land, physical capital, natural resources, production labor, and professional labor). Simulated growth rates are very sensitive to productivity growth rates because the larger they are for a particular region, the larger is that region’s marginal product of capital. The region therefore enjoys higher
360
Rod Tyers and Qun Shi
levels of investment and hence a double boost to its per capita real income growth rate. The importance of productivity notwithstanding, the empirical literature is inconsistent as to whether productivity growth has been faster in agriculture or in manufacturing and whether the gains in any sector have enhanced all primary factors or merely production labor. The single set of factor productivity growth rates assumed in all scenarios are drawn from a new survey of the relevant literature (Tyers et al. 2005). Agricultural productivity grows more rapidly than that in the other sectors in China, Australia, Indonesia, Other East Asia, India, and Other South Asia. This faster growth is caused by continued increases in labor productivity in agriculture and the associated shedding of labor to the other sectors. In the other industrialized regions, the process of labor relocation has slowed down, and labor productivity growth is slower in agriculture. In the other developing regions, the relocation of workers from agriculture has tended not to be so rapid.
3.2 Investment Interest Premia In addition to exogenous productivity growth, a key aspect of the baseline projection is its allocation of investment across regions. The model takes no explicit account of investment risk and the segmented capital markets that are prevalent in developing countries, and so it tends to allocate investment to regions that have high marginal products of physical capital. These regions tend to include labor-abundant developing countries whose labor forces are still expanding rapidly, yet we know that capital market underdevelopment, capital controls, and risk considerations limit the flow of foreign investment into these regions at present and that these considerations are likely to remain important in the future. To account for investment risk and the segmented capital markets prevalent in developing countries we have constructed a “pre-baseline” simulation in which we maintain the relative growth rates of investment across regions. In this simulation, global investment rises and falls, but its allocation between regions is thus controlled. To do this an interest premium variable (GDyn variable SDRORT) is made endogenous. This creates wedges between the international and regional interest rates, the scale of which is indicated in Figure 13.2. It shows high premia for the populous developing regions of Sub-Saharan Africa, the Middle East and North Africa, and India. Premia tend to fall in regions in which labor forces are falling or growing more slowly, such as Japan and the East Asian regions. Most spectacular is the secular fall in the Chinese premium, which occurs because investment growth is maintained in China despite the eventual decline in
Global Demographic Change
361
8
6 North America
4
Sub-S Africa Mid-East N Africa
2
Western Europe
0
Central Europe FSU India
-2
Japan China
-4
Other East Asia
-6
-8 1995
2000
2005
2010
2015
2020
2025
2030
2035
Figure 13.2. Baseline interest premia. These are simulated cumulative percentage point wedges between regional and global rates of return (DROR-DRORW), the global rate being that offered by the global trust. They are derived from a pre-base simulation in which investment is constrained to grow at the same rate in all regions.
its labor force.25 The final baseline simulation then frees up investment. It and the subsequent simulations not only maintain endogenous investment, but to do this they have the time paths of the regional interest premia set as exogenous to the levels indicated in the figure.
3.3 The Baseline Population Projection Notwithstanding their dependence on a comparatively simple four-agegroup demographic model, regional population levels and age structures 25
A possible consequence of this is that our baseline investment in China, and therefore China’s projected economic growth rate, is optimistic. A more detailed study of this issue is offered by Tyers and Golley (2006).
362
Rod Tyers and Qun Shi Table 13.7. Baseline projections of labor force size and structure Labor forcea
Australia North America Western Europe Central Europe Japan China Indonesia Other East Asia India Other South Asia South America Mid East Nth Africa Sub-Saharan Africa Rest of World
% Female
% 40+
Initial
2030
Initial
2030
Initial
2030
8 182 184 181 61 570 87 127 356 134 123 103 150 79
10 250 165 148 55 592 130 178 594 265 193 176 349 131
37 40 40 47 37 37 38 41 27 28 38 24 28 36
40 42 44 46 37 36 38 40 28 28 39 23 29 34
42 42 47 44 58 34 40 37 36 32 33 30 29 38
48 47 55 53 65 47 54 51 47 44 48 42 36 48
a Measured in FTE workers (in millions). Source: Projection using the demographic model described in the text, as presented in detail by Chan and Tyers (2006).
follow closely corresponding United Nations projections.26 Baseline projections of labor force levels and age structures are summarized in Table 13.7, which shows substantial aging of labor forces in all regions. Indeed, the extent of the widespread aging is especially clear from the trends in nonworking aged dependency ratios listed in Table 13.8. Western Europe, Australia, and Japan are the regions with the “oldest” population by 2030 in terms of the aged dependency ratio. China’s dependency ratio is the highest among the developing economies. In particular, it rises fairly rapidly during this period, suggesting a significant demographic transition for the economy. Finally, baseline projections of total populations and labor forces for a selection of regions are displayed in Figure 13.3. These illustrate divergences in the growth paths of populations and labor forces that are important for economic performance. Particularly noteworthy are the declines in the labor forces of Japan, Western Europe, and China, all of which precede their associated population declines as their populations age. 26
See United Nations (2003) and the detailed comparison provided in Chan and Tyers (2006).
Global Demographic Change
363
Table 13.8. Baseline nonworking aged dependency ratios Nonworking aged/working
Australia North America Western Europe Central Europe Japan China Indonesia Other East Asia India Other South Asia South America Mid East Nth Africa Sub-Saharan Africa Rest of World
Initial
2030
0.35 0.24 0.42 0.29 0.32 0.19 0.09 0.09 0.12 0.09 0.16 0.15 0.13 0.15
0.54 0.36 0.61 0.42 0.48 0.44 0.16 0.23 0.23 0.18 0.29 0.33 0.15 0.27
Source: Base period statistics constructed from population statistics from United Nations (2003) and simulation results from the demographic model described in the text.
3.4 The Baseline Economic Projection Overall baseline economic performance is suggested by Table 13.9, which lists the projected increments to regional real per capita incomes by 2030. In part because of its comparatively young population and hence its continuing rapid labor force growth, India attracts substantial new investment and is projected to take over from China as the world’s most rapidly expanding region. This investment, combined with exogenous factor productivity growth, ensures that India is also projected to deliver the largest improvement in real per capita income through 2030. China’s growth is slower in aggregate because of its declining labor force, but its declining interest premium maintains a high level of investment growth sufficient to deliver the second largest proportional change in real per capita income. Indonesia and Other East Asia are also strong performers, while the older industrial economies continue to grow more slowly. The African regions enjoy good GDP growth performance, but their high population growth rates limit their performance in per capita terms. In the GDyn model the current account is not forced to converge on any particular steady state. As a result, it is possible for the interregional distribution of asset ownership to change so as to cause current account
364
1.05
Rod Tyers and Qun Shi Western Europe
1
1
0.95
0.95
0.9
0.9
0.85 1995
Japan
1.05
population
population
labor force
labor force
2005
2015
2025
2035
China
1.1
0.85 1995
2005
2015
2025
2035
India
1.75 1.65
1.05
1.55 1 1.45 0.95
1.35 1.25
0.9
1.15
population 0.85
0.8 1995
population
labor force
2005
2015
1.05
2025
2035
0.95 1995
labor force
2005
2015
2025
2035
Figure 13.3. Baseline population and labor force projections.
deficits to widen relative to GDP. Some imbalances do widen, although the deficits and surpluses expand more slowly than the trend of the past two decades. Most notable are expanding deficits in Japan and China, in which the aging populations reduce saving through time and deficits are required to finance continued investment. North America trends toward widening trade surplus, although it belies comparatively large changes in the net factor income component of its current account, which actually goes
Global Demographic Change
365
Table 13.9. Baseline real per capita income projection to
2030, percent change over 1997 Ranked by performance India China Indonesia Other East Asia Central Europe FSU Japan Australia Western Europe Other South Asia South America Rest of World North America Sub-Saharan Africa Mid-East Nth Africa
369 361 356 341 218 207 185 171 156 142 141 139 125 88
Source: Baseline simulation using the augmented GDyn, as described in the text.
into deficit; therefore North America’s overall current account deficit does not diverge significantly. Net factor income flows are less significant for the other regions. Finally, the product composition of global output in the baseline projection is suggested by Figure 13.4. Engel’s Law drives increased resources into the industrial and services sectors relative to the food and agricultural sector. However, this change emerges despite a relative increase in food prices, which is also shown in the figure. This result is not without controversy. Ever since the work of Lewis (1952), a measured trend toward declining prices of food commodities relative to traded manufactures has been prominent in the commodity trade literature. Grilli and Yang (1988) confirmed and updated the trend identified by Lewis, indicating a decline in relative commodity prices of 0.5% per year. This work was flawed, however, by its lack of consideration of improvements in the quality of manufactures over time. Lipsey (1994), after adjusting for quality changes, found that primary commodity prices actually increased by 0.5% per year in the last half of the 20th century. Our baseline simulation reflects this result. However, it is worth recalling that the purpose of the baseline is to provide a noncontroversial reference against which to compare the alternative scenario with different demographic behavior. Our
366
Rod Tyers and Qun Shi Global output
4
World trading prices 1.15
Food Industrial goods
3.5
Services
1.10
3
2.5
1.05
2
1.00 1.5
1
Food
0.95
Industrial goods
0.5
Services 0 1995 2000 2005 2010 2015 2020 2025 2030 2035
0.90 1997
2002
2007
2012
2017
2022
2027
Figure 13.4. Baseline global output and world trading product prices. World trading prices are relative to a common global numeraire.
emphasis is therefore on the departures of the alternative scenario from the baseline.
4. The Impact of Accelerated Population Aging The analysis presented here is centered on the baseline projection of populations, labor forces, and their structures described in the previous section. This baseline is compared with an alternative scenario that is identical in all respects except that aging is accelerated in all regions via increases in the life expectancy at 60. In particular, life expectancies at 60 grow faster than in the baseline case, by 2% per year. This alternative “accelerated aging” scenario is prompted by the conjecture of Booth (2004) that the standard demographic projections tend to ignore many new and potential developments in health science, which makes them pessimistic about future longevity.27 The effect of accelerated aging is to increase the aged population in every region, while leaving younger populations the same as in the baseline. This has two key implications. First, it raises projected overall populations so 27
An alternative approach to the measurement of the effects of aging is to compare the baseline, which embodies projected changes in age distributions, with a hypothetical scenario in which aggregate populations grow at the same rates but age distributions do not change. This approach is adopted in Tyers and Shi (2006).
Global Demographic Change
367
Table 13.10. Demographic effects in 2030 of accelerated aging,a percent departures
from the baseline scenario Population
Australia North America Western Europe Central Europe Japan China Indonesia Other East Asia India Other South Asia South America Mid East Nth Africa Sub-Saharan Africa Rest of World
Labor force
Nonworking aged dep ratio
Total
60+
Total
60+
Aged NW
5.7 5.1 7.5 7.8 7.4 5.9 4.5 4.4 3.8 3.1 4.1 3.3 2.4 3.6
23.7 23.6 25.8 31.4 21.4 27.4 31.0 25.8 26.6 26.6 24.4 25.3 34.5 20.1
1.3 2.2 2.3 3.6 5.5 1.4 4.7 3.5 3.2 3.0 3.4 2.3 2.8 3.3
24.6 24.0 26.3 32.5 22.1 29.0 31.2 26.1 27.5 26.7 24.6 25.8 36.2 21.0
22.0 20.9 22.9 26.6 14.8 25.5 25.0 21.4 22.2 22.7 20.1 22.3 30.0 15.8
a Growth in target life expectancies at 60 by 2% per year. Source: Simulation results from the model described in the text.
that it raises aggregate demand, and second, to the extent that the aged continue to work it raises labor forces and so also bolsters the supply side. Because the departure from the baseline only affects the aged population in each region, it also causes nonworking aged dependency ratios to rise in all regions.28 These demographic and labor supply effects are summarized in Table 13.10. The absolute increase in each region’s population and labor force depends on the demographic structure of its population in the base year and the aged labor force participation rate at the time. As shown in Figure 13.5, Japan enjoys the largest labor force increase. This is because Japan began with the largest aged share of its population and also the largest aged labor force participation rate.29 Because of its comparatively large aged population, Western Europe also enjoys substantially higher labor force growth. Among 28
29
Our modeling to date ignores the fiscal implications of this change because governments are modeled as maintaining patterns of expenditure that do not depend on the age distributions of populations. Moreover, governments balance their budgets in all simulations. To capture fiscal implications it will be necessary to separate out such sectors as health and retirement services on the one hand and educational services on the other. Almost half of Japan’s men aged 60+ are in the labor force. This is substantially higher than for other industrialized regions, although retirement is not a luxury enjoyed in most
368
Rod Tyers and Qun Shi Labor Force
Average Saving Rate
6
2
Japan Indonesia 5
0
Central Europe FSU India Western Europe
4
-2
North America China
3
-4
2
-6
1
-8
Japan Australia Central Europe FSU
-10
0
Western Europe North America
-12
-1 1997
2002
2007
2012
2017
2022
2027
1997
2002
2007
2012
2017
2022
2027
Figure 13.5. Accelerated aging–labor force and average saving rate, departure from baseline, %.
the regions at the other extreme is China, which has a proportionally smaller aged population and, compared with all the other regions, smaller aged labor force participation rates.30 The economic implications of accelerated aging stem not only from expanded consumption demand on the one hand and larger labor forces on the other. In North America, Western Europe, Central Europe, the former Soviet Union, and Australia, accelerated aging also raises the share of income in the hands of the over 60s, who tend to have negative saving rates. In these regions, therefore, average saving rates decline, as also shown in Figure 13.5. Indeed, the savings effect of this demographic change is so significant that real savings in most regions fall, despite an increase in real GDP and hence real income of the private households. Capital returns are raised by the population-driven demand increases and the expanded labor
30
of the developing world. A full listing of participation rates is provided by Chan and Tyers (2006). In China, formal retirement was a tradition in the state-owned enterprises that dominated the economy until recently. Recorded aged labor force participation rates actually declined during the 1990s. With the rise in private sector employment, however, it is expected that aged labor force participation rates will rise, rather than fall, as is assumed in the analysis presented.
Global Demographic Change
369 Real GDP
Investment 4
2
0
3
-2
2
-4
1
North America South America Sub-Saharan Africa Mid-East Nth Africa Western Europe India Japan China Australia
Japan -6
0
Indonesia North America India
-8
-1
Western Europe Central Europe FSU China
-10
-2
1997
2002
2007
2012
2017
2022
2027
1997
2002
2007
2012
2017
2022
2027
Figure 13.6. Accelerated aging–investment and real GDP, departure from baseline, %.
supplies, yet real financing costs also rise because proportionally less is saved worldwide. The net effect is slower growth in global investment, with the level achieved in 2030 emerging lower by more than 3% compared to the baseline. The global investment slowdown notwithstanding, GDP tends to increase everywhere because of faster growing, albeit aged, labor forces (Table 13.11). The gains are not evenly distributed, however. Regions with the most labor force expansion, and particularly Japan, enjoy higher capital returns relative to other regions. This raises their shares of global investment compared to that in the baseline projection, particularly later in the simulation, and hence it accelerates their GDP growth (Figure 13.6). This investment is reassigned mainly from China, where investment falls by more than 8% by 2030, and from Australia, where investment falls close to 3%. In both regions, labor forces are boosted the least because their aged populations have lower labor force participation rates. The redistribution of investment expands the mismatch between the location of physical capita and its ownership, causing a corresponding divergence between GDP and GNP growth. The regions that expand more rapidly as a consequence of the accelerated aging tend also to be those with slower saving growth, and so these regions experience outflows of factor income on their current accounts. When this is combined with the tendency, due to diminishing returns, for labor expansions to reduce per capita income,
370 Table 13.11. Accelerated aging – product and factor prices, percent departure from the baseline in 2030 Labor Average Real Real GDP Real per Real production Land Resource Food Manufactures Services force saving rate investment GDP pricea capita income wageb rentb rentb output output output Australia North America Western Europe Central Eur, FSU Japan China Indonesia Other East Asia India Other South Asia South America Mid East Nth Africa Sub-Saharan Africa Rest of World a
1.3 2.2 2.3 3.6 5.5 1.4 4.7 3.5 3.2 3.0 3.4 2.3 2.8 3.3
−5.1 −11.7 −11.5 −6.7 −0.5 0.3 −0.2 0.1 0.0 −2.3 −1.1 −0.1 0.6 −3.3
−2.7 −0.5 −1.6 −2.0 1.4 −8.2 0.7 −1.8 −0.9 −0.7 0.9 −2.5 1.0 0.0
−0.1 1.3 0.8 0.9 3.1 −1.9 1.4 0.8 0.6 1.0 1.7 0.3 1.9 1.1
0.3 −0.1 0.2 −0.1 0.5 −0.5 0.0 0.0 0.2 −0.1 −0.2 0.0 −0.2 −0.2
The GDP price is measured relative to the common numeraire in GDyn. Note relative values only, indicative of real exchange rate changes. Source: Simulations of the model described in the text.
b
−5.2 −4.6 −6.8 −6.6 −2.7 −6.2 −2.4 −2.1 −2.4 −1.9 −2.4 −2.1 −0.5 −2.6
−0.9 −0.3 −1.0 −1.7 −1.4 −2.1 −2.1 −1.7 −2.1 −1.6 −1.1 −1.3 −0.5 −1.4
2.7 5.4 7.9 6.6 7.8 2.9 4.3 3.6 2.8 3.0 5.3 5.4 6.2 4.3
−1.2 0.6 −0.6 0.3 0.9 −1.1 0.5 −0.1 −0.1 0.6 1.1 −0.5 1.0 0.7
1.5 2.8 4.9 3.8 4.6 2.1 2.1 1.8 1.3 1.2 2.7 3.2 3.4 2.3
−1.3 0.7 −0.3 0.4 1.1 −1.1 0.6 −0.1 0.2 0.8 1.3 −0.7 0.9 0.7
0.0 1.3 0.9 0.8 3.4 −3.1 1.8 1.1 0.6 1.0 1.8 0.5 2.1 1.2
Global Demographic Change Real per Capita Income
371 Global Output
1
3
0
2.5
Food Industrial products Services
-1
2 -2
1.5 -3
1 -4 Sub-Saharan Africa India -5
0.5
Japan North America
-6
0
China Western Europe
-7 1997
2002
2007
2012
2017
2022
2027
-0.5 1995 2000 2005 2010 2015 2020 2025 2030 2035
Figure 13.7. Accelerated aging real per capita GNP and global product output, departures from baseline, %.
declines in real per capita GNP are observed in all regions (Fig. 13.7 and Table 13.11). Turning finally to the effects on the global product mix, as Table 13.6 demonstrates, when the array of products and services is aggregated into just three groups, the differences between the consumption preferences of the old and the young are not substantial. Overall, the effect of accelerated aging is to raise the production of food and agricultural products relative to the others and that of services relative to industrial products (Fig. 13.7). The rise in food production is due primarily to the declines in per capita incomes and Engel’s Law, whereas the relative rise in services production does appear to be a response to aging preferences. Supply-side effects, via declines in real wages (Table 13.11), advantage comparatively labor-intensive manufacturing, but these seem insufficient for the change in manufacturing output to outpace that in services.
5. Conclusion Global population and labor force changes that are already built into regional age distributions will cause the populations in several key regions, including Western Europe, Japan, and China, to decline in the near future and their labor forces to decline sooner and more dramatically. Associated with these changes will be the aging of the populations in all regions. This will not
372
Rod Tyers and Qun Shi
only affect their economic performance but also the regional distribution of global investment and saving, and hence capital account flows, as well as regional production costs, comparative advantage, and therefore the pattern of global trade. To address these issues, standard GDyn is modified to accommodate eight age-gender groups within each regional household. A full demographic submodel is then incorporated and a baseline projection constructed through 2030. To illustrate the use of this model in the analysis of demographic change, an alternative scenario is constructed in which advances in health science cause life expectancies at 60 to rise more quickly in all regions. This accelerates the aging of all populations, and because the aged also contribute to the labor force, there is a relative expansion in the global labor supply. This expansion is largest in regions with either large aged shares of their populations at the outset, or high aged labor force participation rates, or both. It is therefore particularly strong for Japan. The labor supply expansions tend to cause some redistribution of global investment in favor of aging regions with high labor force participation rates. The volume of that investment is made smaller, however, by reduced global saving. Nonetheless, most regions enjoy net expansions in GDP. Because the capital attracted to the expanding regions tends to be owned elsewhere, the boost to regional GNP levels is more modest. Moreover, diminishing returns ensure that the effects on real per capita regional incomes are consistently negative. Hardest hit are Western and Central Europe and China, where aged participation rates are low, so that their populations age and expand, yet this does not increase their labor supplies and therefore their capital growth. Reduced per capita incomes ensure that global food output is boosted relative to that of both industrial products and services. In addition, because the consumption preferences of the old become more prominent, service output is accelerated relative to that of industrial products. In the form presented here, the model offers little role for governments in the aging process and its consequences. Governments balance their budgets, and age-related service expenditures are not identified explicitly. A number of potential improvements to the model therefore arise. First, increased product variety would make it possible to better reflect differences in consumption preferences across age groups. Second, health and retirement services could be identified in the model, with contributions to their supply from both governments and regional private sectors. These improvements would facilitate the analysis of the fiscal implications of aging and of changes in health and retirement sector productivity.
Global Demographic Change
373
References Abdel-Ghany, M. and D. L. Sharpe. 1997. “Consumption Patterns among the Young-Old and Old-Old.” Journal of Consumer Affairs 31(1), 90–112. Attanasio, O. P. and J. Banks. 1998. “Trends In Household Saving Don’t Justify Tax Incentives to Boost Saving.” Economic Policy: A European Forum 10(27), 547– 83. Attanasio, O. P., J. Banks, C. Meghir, and G. Weber. 1999. “Humps and Bumps in Lifetime Consumption. Journal of Business and Economic Statistics 17(1), 22–35. Attanasio, O. P. and M. Szekely. 1998, December. Household Savings and Income Distribution in Mexico. Office of the Chief Economist, Documento de Trabajo No. 390. Washington, DC: Inter-American Development Bank. Barro, R. J., and G. S. Becker. 1989. “Fertility Choice in a Model of Economic Growth.” Econometrica 57(2), 481–501. Blisard, N. 2001a. Food Spending in American Households, 1997–98. Economic Research Service Statistical Bulletin 972. Washington, DC: U.S. Department of Agriculture. Blisard, N. 2001b. Income and Food Expenditures Decomposed by Cohort, Age and Time Effects. Electronic Report from the Economic Research Service. Washington, DC: U.S. Department of Agriculture. Available at www.ers.usda.gov. Blisard, N., J. N. Variyam, and J. Cromartie. 2003. Food Expenditures by U.S. Households: Looking Ahead to 2020. Electronic Report from the Economic Research Service. Washington, DC: U.S. Department of Agriculture. Available at www.ers.usda.gov. Bloom, D. E. and J. G. Williamson. 1997. “Demographic Transitions, Human Resource Development and Economic Miracles in Emerging Asia.” In J. Sachs and D. Bloom (eds.), Emerging Asia (pp. 419–455). Manila: Asian Development Bank. Booth, H. T. 2004. “On the Importance of Being Uncertain: Forecasting Population Futures for Australia.” People and Place 12(2), 1–12. Booth, H. T., J. Maindonald, and L. Smith. 2002. “Applying Lee-Carter under Conditions of Variable Mortality Decline.” Population Studies 56(3), 325–36. Booth, H. T. and L. Tickle. 2003. “The Future Aged: New Projections of Australia’s Elderly Population.” Australasian Journal of Ageing 22(4), 196–202. Bryant, R. C. 2001. Incorporating Demographic Change in Multi-Country Demographic Models: Some Preliminary Results. Available at http://www.sensiblepolicy.com/wmhp/ home1.htm. Bryant, R. C., H. Faruqee, D. Veculescu, and E. Arbatli. 2003. Fertility Declines and Youth Dependency: Implications for the Global Economy. Washington, DC: Brookings Institution. Bryant, R. C. and W. J. McKibbin. 1998. Issues in Modelling the Global Dimensions of Demographic Change. Available at http://www.sensiblepolicy.com/wmhp/home1.htm. Case, A. and A. Deaton. 2002, May. Consumption, Health, Gender and Poverty. Research Program in Development Studies. Princeton: Princeton University. Chan, M. M. and R. Tyers. 2006, December. Global Demographic Change and Labour Force Growth: Projections to 2020. Centre for Economic Policy Research Discussion Paper, Research School of Social Sciences (RSSS). Canberra: Australian National University. Dixon, P. B. and M. Rimmer. 2002. Dynamic General Equilibrium Modelling for Forecasting and Economic Policy. No. 256 in the Contributions for Economic Analysis series. Amsterdam: Elsevier North Holland.
374
Rod Tyers and Qun Shi
Duncan, R., Q. Shi, and R. Tyers. 2005. Global Demographic Change and Demand for Food in Australia. Rural Industries Research and Development Corporation Report No 05/014. Canberra: RIRDC. Faruqee, H. and M. Muhleisen. 2002. “Population Ageing in Japan: Demographic Shock and Fiscal Sustainability.” Japan and the World Economy 15: 185–210. Gibson, J. and G. Scobie. 2001. “A Cohort Analysis of Household Income, Consumption and Savings.” New Zealand Economic Papers 35(2), 196–216. Grilli, E. and M. C. Yang. 1988. “Primary Commodity Prices, Manufactured Goods Prices and the Terms of Trade of Developing Countries: What the Long Run Shows.” World Bank Economic Review, 2(1). Harrigan, J. 1995. “The Volume of Trade in Differentiated Products: Theory and Evidence.” Review of Economics and Statistics 77(2): 283–93. Hertel, T. W. (ed.). 1997. Global Trade Analysis Using the GTAP Model. Cambridge: Cambridge University Press. Huff, K. M., K. Hanslow, T. W. Hertel, and M. E. Tsigas. 1997. “GTAP Behavioral Parameters.” In T. W. Hertel (ed.), Global Trade Analysis Using the GTAP Model (pp. 124–48). Cambridge: Cambridge University Press. IMF. 2004, September. World Economic Outlook. Washington, DC: International Monetary Fund. Kitamura, U., N. Takayama, and F. Arita. 2001, September. Household Savings and Wealth Distribution in Japan. Discussion Paper No. 38, Project on Intergenerational Equity, Institute of Economic Research. Tokyo: Hitotsubashi University. Lee, R. D. 2003. “The Demographic Transition: Three Centuries of Fundamental Change.” Journal of Economic Perspectives 17(4), 167–90. Lewis, W. A. 1952. “World Production, Prices and Trade, 1870–1960.” Manchester School of Economic and Social Sciences 20(2), 105–38. Lipsey, R. E. 1994. Quality Change and Other Influences on Measures of Export Prices of Manufactured Goods. World Bank Policy Research Working Paper 1348. Washington, DC: World Bank. Liu, J., N. Van Leeuwen, T. T. Vo, R. Tyers, and T. W. Hertel. 1998, September. Disaggregating Labor Payments by Skill Level in GTAP. Technical Paper No. 11. West Lafayette, IN: Center for Global Trade Analysis, Purdue University, West Lafayette. Available at http://www.agecon.purdue.edu/gtap/techpapr/tp-11.htm. Mason, A. (ed.). 2003. Population Change and Economic Development in East Asia: Challenges Met and Opportunities Seized. Stanford, CA: Stanford University Press. McDonald, P. and R. Kippen. 2001. “Labour Supply Prospects in 16 Developed Countries.” Population and Development Review 27(1): 1–32. Obstfeld, M. and K. Rogoff. 2000. “The Six Major Puzzles in International Macroeconomics: Is There a Common Cause?” In B. Bernanke and K. Rogoff (eds.), NBER Macroeconomics Annual 2000 (pp. 339–412). Cambridge: NBER and the MIT Press. OECD. 1996. Ageing in OECD Countries. Social Policy Studies No. 20. Paris: Organization for Economic Cooperation and Development. OECD. 1998. Maintaining Prosperity in an Ageing Society. Paris: Organization for Economic Cooperation and Development. OECD. 1999. Employment Outlook. Paris: Organization for Economic Cooperation and Development.
Global Demographic Change
375
OECD. 2002. Employment Outlook. Paris: Organization for Economic Cooperation and Development. Paulin, G. D. 2000. “Expenditure Patterns of Older Americans, 1984–97.” Monthly Labour Review 123(5), 3–28. Productivity Commission. 1999, November. Microeconomic Reforms and Australian Productivity: Exploring the Links. Commission Research Paper. Melbourne: Government of Australia. Productivity Commission. 2001, April. Resource Movements and Labour Productivity, An Australian Illustration 1994–95 to 1997–98. Staff Research Paper. Melbourne: Government of Australia. Regmi, A., M. S. Deepak, J. L. Seale Jr., and J. Bernstein. 2001. Cross Country Analysis of Food Consumption Patterns. Economic Research Service WRS-01–1. Washington, DC: U.S. Department of Agriculture. Shi, Q. and R. Tyers. 2004. Global Population Forecast Errors, Economic Performance and Australian Food Export Demand. Working Papers in Economics and Econometrics No.439. Canberra: Australian National University. Available at http://ecocomm.anu. edu.au/ecopapers. Trefler, D. and H. Lai. 1999. Gains from Trade: Standard Errors with the CUS Monopolistic Competition Model. Toronto: University of Toronto. Tyers, R. and J. Golley. 2006. China’s Growth to 2030: The Roles of Demographic Change and Investment Risk. Paper presented at the conference, WTO, China and the Asian Economies IV: Economic Integration and Development, University of International Business and Economics, June 24–5, Beijing. Tyers, R. and Q. Shi. 2007. “Global Demographic Change, Policy Responses and Their Economic Implications.” World Economy 30(4), 537–66. Tyers, R., Q. Shi, and M. M. Chan. 2005, September. Global Demographic Change and Economic Performance: Implications for the Food Sector. Report to the Rural Industries Research and Development Corporation. Canberra: RIRDC. United Nations. 2000, March. Replacement Migration: Is It a Solution to Declining and Ageing Populations? UN Population Division. New York: UN Secretariat. United Nations. 2003, February. World Population Prospects: The 2002 Revision. UN Population Division. New York: UN Secretariat. Available at www.un.org/esa/population/ publications/wpp2002. Vedi, J. 2005. The Global Economic Effects of Expanded Skilled Migration. Honours dissertation, School of Economics, College of Business and Economics, Australian National University. Weber, G., R. Miniaci, and C. Monfardini. 2002. “Changing Consumption Patterns.” In H. Siebert (ed.), Economic Policy for Ageing Societies (pp. 53–76). Berlin: Springer Verlag. WHO. 2003. Mortality Database: Table One: Number of Registered Deaths. Geneva: World Health Organization. Available at http://www3.who.int/whosis/menu.cfm?path= whosis,inds,mort&language=english. Williamson, J. G. 1998. “Growth, Distribution and Demography: Some Lessons from History.” Explorations in Economic History 35(3), 241–71.
PART IV
EVALUATION OF THE DYNAMIC GTAP FRAMEWORK
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
FOURTEEN
Household Saving Behavior in the Dynamic GTAP Model Evaluation and Revision Alla Golub and Robert A. McDougall 1. Introduction The GDyn model presented in Chapter 2 inherits from the standard GTAP model its specification of the regional household demand system and, in particular, the treatment of saving. As in the standard GTAP model, regional households in GDyn spend their income according to a Cobb-Douglas per capita utility function specified over three sources of utility: private consumption, government consumption, and real saving. Because the model is not forward-looking but is recursively dynamic, the utility function is static – it represents utility from present but not future consumption. The practice of including saving in the static utility function derives from Howe (1975) and allows regional households to value saving in atemporal settings.1 Because of the Cobb-Douglas functional form, the average propensity to save is fixed, and saving is a fixed proportion of income in each region.2 There are several unwelcome implications of this assumption. Because propensities to save are fixed and incomes are rising over time, countries in which saving substantially exceeds investment, like Japan, accumulate unrealistically large stocks of foreign assets. If like China they also exhibit high rates of growth in income, at the end of long-run GDyn simulations that span many decades, such countries may end up owning a large part of the wealth of the whole world. Although such outcomes cannot altogether 1 2
See Hertel (1997) for a discussion of this issue. In fact, according to McDougall (2002) the propensity to save is not quite fixed, but it is close enough to fixed.
We thank Elena Ianchovichina and Terrie Walmsley for many useful comments. The application related to this chapter is Ch14_gdyns_34_97.zip and is available for download on the Web site at https://www.gtap.agecon.purdue.edu/models/Dynamic/applications.asp
379
380
Alla Golub and Robert A. McDougall
be ruled out a priori, they are very strong predictions resting on a very weak empirical basis. In the real world, saving and investment are highly correlated across countries (Feldstein and Horioka 1980) and net international capital flows, and as a result, net foreign positions are much smaller. Another problem is that as economies with high savings rates, like China, grow there is a glut of global savings and, as a result, of investment and capital in the world. Because of excessive investment, rates of return to capital fall without bound. This prevents us from running simulations with the GDyn model over very long time horizons. This chapter evaluates the performance of GDyn presented in Chapter 2 by comparing the model’s projected outcomes for gross and net foreign assets and liabilities with empirical data and proposes a change to the household saving behavior to address the problems associated with the saving behavior in GDyn. We start from a review of theories explaining household saving behavior and their implementation in CGE models. In forwardlooking models, saving rates are determined by the tradeoff between utility from present consumption and utility from future consumption. However, this approach cannot be implemented in a recursive model like GDyn. In this work we adopt an approach that supports a balanced growth scenario in which regional income, wealth, and saving have the same growth rate. The approach has no particular theoretical foundation, but is practical and motivated by the stylized fact that gross foreign assets and liabilities do not diverge through time in reality nearly as much as in standard GDyn. We modify the theoretical structure of GDyn so that the saving rate in each region is endogenous and is a function of the ratio of wealth to income.
2. Theories of Household Saving Behavior There are several theories explaining people’s desire to save. The most influential theory of household saving behavior is the life-cycle hypothesis developed in Modigliani and Brumberg (1954) and further tested in Ando and Modigliani (1963). This model starts from a consumer who maximizes his or her utility, which is a function of current and future consumption, subject to a resource constraint: “As a result of this maximization the current consumption of the individual can be expressed as a function of his resources and the rate of return on capital with parameters depending on age” (Ando and Modigliani 1963, p. 56). This theory suggests that people try to smooth their consumption over their lifetimes, saving little when they are young but much more in middle age and dissaving in retirement.
Household Saving Behavior in the Dynamic GTAP Model
381
Another theory of household saving behavior suggests that people save for precautionary reasons. If people are uncertain about their future income, and access to credit is limited, people tend to save more. At the same time, stock market and real estate capital gains reduce saving rates because people feel richer. Finally, people may adjust their saving in response to fiscal policies, hence the proposition of Ricardian equivalence. Despite the fact that data support these different theories, more or less, in different regions, it would be difficult to implement them in a CGE model. In the forward-looking G-Cubed CGE model (McKibbin and Wilcoxen 1995), household behavior is modeled by infinitely lived representative agent maximizing intertemporal utility subject to intertemporal budget constraint. However, only a portion of consumption – and, as it follows, saving – is determined by these intertemporally optimizing consumers. The remainder is determined by after-tax current income and the fixed marginal propensity to save. Because at least part of the supply of savings in the G-Cubed model is determined by the tradeoff between utility from present consumption and utility from future consumption, current saving is implausibly sensitive to remote future events. In the forward-looking CGE model described in Benjamin (1994), households choose current consumption and saving based on income level, expected future price level, and the interest rate. The household choice between present and future consumption is expressed with a Cobb-Douglas function, and thus, the budget share spent on saving is fixed. In the GREEN (Burneaux et al. 1992), recursively dynamic model, like in GDyn, the saving enters atemporal extended linear expenditure system utility function (Howe 1975), and the marginal propensity to save is constant and independent of the rate of reproduction of capital. The linkages between aging profiles and saving rates in a CGE model context are explored in Tyers (2005). To investigate the economic implications of changes in population sizes and age profiles, determined by changes in fertility, mortality, and migration, this work builds a demographic model into the GDyn model presented in Chapter 2. Tyers (2005) relaxes the assumption of fixed propensity to save in GDyn by introducing endogenous age-gender specific propensities to save, which depend on real disposable income and the real interest rate. These group-specific saving rates then determine regional saving rates in each period.
3. Comparison of Historic and Simulated Assets and Liabilities How large are gross and net foreign assets and liabilities in GDyn simulations? Our basis for comparison is a country portfolio database constructed
382
Alla Golub and Robert A. McDougall
by Kraay et al. (2000) that was used in Chapter 3 of this book.3 We construct three indicators to evaluate foreign assets and liabilities in GDyn simulations: gross foreign assets as a share of household wealth, gross foreign liabilities as a share of domestic capital, and the ratio of net foreign assets to GDP. The choice of the first two indicators is driven by our interest in regional wealth allocation and the composition of capital in GDyn. The third indicator is of more general interest and is often used in discussion in the financial press and among policymakers about what size of net foreign liabilities is sustainable (Kouparitsas 2004). We compare the development through time of gross foreign positions in the Kraay et al. (2000) database and GDyn simulations. For this illustration, the GTAP 5.4 Data Base was aggregated to 22 regions and 3 sectors.4 To make the figures clear, we show results for just 11 of the 22 regions, representing 11 individual countries. Figure 14.1 shows gross foreign assets as a share of wealth in 1970 and 1997 from the Kraay et al. (2000) database together with gross foreign assets as a share of wealth in 1997 and 2024 simulated with GDyn. Years 1970 and 1997 from the Kraay et al. (2000) sample are chosen to capture changes over 27 years in gross foreign positions and because for some countries data before 1970 are not available. For GDyn simulations, the year 1997 is chosen because it is the initial year in our simulations and the benchmark year in the GTAP 5.4 Data Base, and it allows comparison of foreign positions in the GTAP and Kraay et al. (2000) databases. Year 2024 is chosen to show changes in gross foreign assets in the period of 27 years in the simulation. Similarly, Figure 14.2 shows gross foreign liabilities as a share of capital in 1970 and 1997 of the Kraay et al. (2000) database, together with gross foreign liabilities as a share of capital in 1997 and 2024 of GDyn simulations. First, there are differences in the 1997 gross foreign positions between Kraay et al. (2000) and the initial GDyn Data Base. This is because they adopt different approaches to construct wealth and gross foreign positions. Kraay et al. use historical information on investment accumulation, whereas in GDyn, regional capital is the starting point in the construction of wealth 3
4
Table 3.4 shows variation in historical gross foreign assets and liabilities across regions and over time. Figures 3.8 and 3.9 show the distribution of gross foreign assets as a share of wealth and gross foreign liabilities as a share of capital, respectively. On average, gross foreign assets and liabilities are small (see discussion of the size of gross foreign assets and liabilities in Chapter 3). Only the time variable is shocked. The sectors are agriculture, manufacturing, and services. The 22 countries/regions are Australia, New Zealand, China, Japan, Korea, South Asia, Canada, United States, South America, Austria, Denmark, Finland, United Kingdom, France, Ireland, Italy, Netherlands, Portugal, Sweden, Turkey, Rest of Europe, and Rest of World.
Household Saving Behavior in the Dynamic GTAP Model
383
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0 Korea
Japan
Turkey
China Kraay 1970
US
France Netherlands Italy
Kraay 1997
GDyn 1997
Sweden
UK
Ireland
GDyn 2024
Figure 14.1. Gross foreign assets as share of wealth in Kraay et al. (2000) database and in GDyn simulation. For Japan, the first observation in the Kraay et al. database is for 1971. For Sweden, the last observation is for 1996. For Japan and Sweden, we show 1971 and 1996 gross foreign assets, respectively. Source: Authors’ simulation and Kraay et al. (2000) database.
and its components. The ownership shares in regional capital of domestic and foreign residents are then determined by foreign income receipts and payments. Although the differences between the initial starting point of simulation and the corresponding gross foreign positions recorded in Kraay et al. (2000) are important, they are not the focus of this discussion. Figures 14.1 and 14.2 reveal important differences in the development of gross foreign positions over time between the historical data and the GDyn simulation data. First, note that, although on average gross foreign positions are small, in some countries, like the United Kingdom and Ireland, gross foreign positions can be large. Growth in gross foreign positions in the United Kingdom and Ireland was very significant between 1970 and 1997 (Figs. 14.1 and 14.2). The large size of gross foreign positions in GDyn simulations need not be of concern as long as their size is comparable with historical data.
384
Alla Golub and Robert A. McDougall
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0 Korea
Japan
Turkey
China
Kraay 1970
US
France Netherlands
Kraay 1997
GDyn 1997
Italy
Sweden
UK
Ireland
GDyn 2024
Figure 14.2. Gross foreign liabilities as share of capital in Kraay et al. (2000) and in GDyn simulation. For Japan, the first observation in the Kraay et al. database is 1971. For Sweden, the last observation is in 1996. For Japan and Sweden, we show 1971 and 1996 gross foreign liabilities, respectively. Source: Authors’ simulation and Kraay et al. (2000) database.
Second, as can be seen from Figures 14.1 and 14.2, if historical gross foreign assets and liabilities grow over time, they grow together, which results in small net foreign positions. In contrast, in GDyn simulations, growth in gross foreign assets corresponds to decline in gross foreign liabilities, and vice versa. The faster the growth in gross foreign assets, the more rapid the decline in gross foreign liabilities, as in Korea and Japan. The faster the growth in gross foreign liabilities, the faster the decline in gross foreign assets, as in Turkey and the United States. This relationship results in very large net foreign positions in GDyn simulations. To further illustrate this statement, we compare historical and simulated ratios of net foreign positions to GDP. The net foreign asset to GDP ratios calculated from the Kraay et al. (2000) database are shown in Figure 14.3. Kraay et al. (2000) provide a detailed discussion and presentation of net foreign asset positions. To save space, we show ratios for just 12 of 68 countries covered in Kraay et al. (2000). These 12 countries are chosen to represent all 6 regions of the Kraay et al. database (see Table 3.3) and to present extremes as well as common tendencies in
Household Saving Behavior in the Dynamic GTAP Model
385
1.5
1
0.5
0
-0.5
-1
19
66 19 67 19 68 19 69 19 70 19 71 19 72 19 73 19 74 19 75 19 76 19 77 19 78 19 79 19 80 19 81 19 82 19 83 19 84 19 85 19 86 19 87 19 88 19 89 19 90 19 91 19 92 19 93 19 94 19 95 19 96 19 97
-1.5
USA India
China New Zealand
Japan Switzerland
Korea Jamaica
Turkey Brazil
France Singapore
Figure 14.3. Net foreign position to GDP ratio for selected 12 countries from 1966 to 1997. Source: Kraay et al. (2000).
the database. Switzerland’s net foreign assets and New Zealand’s net foreign liabilities are two extreme cases among the industrialized countries. Net foreign position to GDP ratios of all countries in this group are in the range from −0.7 to 0.5. For the East Asia and Pacific countries the net foreign assets to GDP ratio ranges from −0.34 (Malaysia) to 0.86 (Singapore), with most observations in the range from −0.34 to 0.02. Among Latin American and Caribbean countries, Jamaica is an outlier, with relatively large net foreign liabilities (see Fig. 14.3). Net foreign positions relative to GDP of other Latin American and Caribbean countries range from −0.53 to 0.07. In the Middle East and North Africa (MENA) region, countries’ net foreign positions relative to GDP are in the range from −0.6 to 0.25, with Saudi Arabia’s net foreign positions as high as 1.34. In South Asia, net foreign assets to GDP ratios are in the range from −0.18 (Pakistan) to 0.02 (Bangladesh). Finally, net foreign position to GDP ratios of SubSaharan Africa (SSA) countries are in the range from −0.92 (Congo) to 0.15 (Lesotho). To summarize, historical net foreign assets are small. Simulated net foreign assets as a share of GDP are shown in Figure 14.4. For this simulation we use the same 22 × 3 aggregation described earlier. To make the figure clear, only 10 regions are shown. Net foreign assets as
386
Alla Golub and Robert A. McDougall
15
10
5
0
-5
19
97 20 01 20 05 20 09 20 13 20 17 20 21 20 25 20 29 20 33 20 37 20 41 20 45 20 49 20 53 20 57 20 61 20 65 20 69 20 73 20 77 20 81 20 85
-10
China
Japan
Korea
SA
US
SAM
Turkey
France
Sweden
Italy
Figure 14.4. Net foreign position to GDP ratios in a GDyn simulation. For this simulation the GTAP 5 Data Base was aggregated to 22 regions. To make the figure clear only 10 regions are shown. The ratios of other 12 regions are in the range between ratios of Japan and Turkey. Source: Authors’ simulation.
a share of GDP of the other 12 regions in this aggregation are between the ratios of Japan and Turkey. As can be seen from a comparison of Figures 14.3 and 14.4, net foreign positions in GTAP simulations are unrealistically large. In less than 10 years of the simulation, the ratios become outside the ( −1, 1) range and continue to grow in absolute value. Fixed propensities to save in the model coupled with rising incomes lead to unrealistically large international capital flows. They create too much saving and, as a result, too much investment and capital in the world; they allow chronic rapid saving or dissaving and, as a result, too great foreign assets or liabilities in individual regions. To overcome this problem we modify household saving behavior by endogenizing saving rates. The development of wealth of a region, which is a sum of its capital stock and net foreign assets, is determined by its saving behavior. For illustrative purposes, consider a country where capital and income grow at the same rate. Rapid growth in net foreign liabilities relative to regional income will lead to a decrease in wealth relative to income, whereas fast-growing net foreign assets relative to income increase regional wealth-to-income ratios. Our treatment is based on a simple historical observation that country
Household Saving Behavior in the Dynamic GTAP Model
387
20 18 16 14
ratio
12 10 8 6 4 2 0
1997 2002 2007 2012 2017 2022 2027 2032 2037 2042 2047 2052 2057 2062 2067 2072 2077 2082 2087 year China
Japan
Korea
US
France
Italy
Netherland
Turkey
Figure 14.5. Wealth-to-income ratios in a long-run GDyn simulation. Source: Authors’ simulation.
wealth-to-income ratios are in a specific range and do not change rapidly over time. As an illustration, we compare historical wealth-to-income ratios calculated from the Kraay et al. (2000) database with ratios in GDyn simulations. Rapidly growing wealth-to-income ratios in GDyn simulations are shown in Figure 14.5. By comparison, the historical wealth-to-income ratios in Figure 14.6 are very stable. To show this, we regress historical wealthto-income ratios on a time trend. The regression results are reported in Table 14.1. The time trend is not statistically significant in the majority of countries, with the exception of countries in the East Asia and Pacific region and a few Latin American and Caribbean countries. Figure 14.6 shows historical wealth-to-income ratio over time for selected countries and illustrates the regression result.
4. New Household Saving Behavior in GDyn We assume that regional households try to maintain target wealth-toincome ratios WYRT(r). The actual wealth-to-income ratio is given by WYRA(r) =
WQHHLD(r) , INCOME(r)
(14.1)
388
Alla Golub and Robert A. McDougall
6
5
4
3
2
1
0
1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 China
Korea
US
Turkey
Finland
Singapore
Brazil
Germany
Figure 14.6. Historical wealth-to-income ratios of selected countries constructed using the Kraay et al. (2000) database.
where WQHHLD(r) and INCOME(r) denote household wealth and income. Differentiating (14.1), we obtain the percentage change equation: wyra(r) = wqh(r) − y(r),
(14.2)
where wyra(r) denotes the percentage change in the actual wealth-to-income ratio, wqh(r) the percentage change in the wealth of the regional household, and y(r) the percentage change in income. If the wealth-to-income ratio is too low or too high compared to the target wealth-to-income ratio in a region, then the regional household gradually adjusts its actual wealth-toincome ratio toward the target. Similar to the lagged adjustment mechanism in the rates of return presented in Chapter 2, we define a lagged adjustment mechanism for the wealth-to-income ratio: RRG WYR(r) = LAMBWYR(r)∗ log
WYRT(r) , WYRA(r)
(14.3)
where RRG_RORG(r) denotes the required rate of growth in the wealthto-income ratio, and LAMBWYR(r) is a coefficient of adjustment.
Household Saving Behavior in the Dynamic GTAP Model
389
Table 14.1. The wealth-to-income ratio regressed on a time trend Country 1 INDC Australia Austria Canada Switzerland Germany Denmark Spain Finland France United Kingdom Greece Ireland Italy Japan Netherlands Norway New Zealand Portugal Sweden USA LAC Bolivia Brazil Colombia Costa Rica Dominican Republic Ecuador Guatemala Honduras Jamaica Mexico Nicaragua Peru Slovenia Trinidad and Tobago Uruguay
Intercept 2
Time trend 3
First lag 4
Second lag 5
R-square 6
Obs. 7
0.588∗∗ 0.902∗∗∗ 0.558∗∗∗ 2.231∗∗ 0.501∗∗ 0.979∗∗ 0.661∗∗ 1.026∗∗ 0.900∗∗ 0.796∗∗ 0.564∗∗ 0.688∗∗ 1.248∗∗∗ 1.655∗∗ 0.845∗∗ 0.903 0.921∗ 1.472∗∗∗ 0.701∗∗ 0.431∗∗
−0.007 0.013 0.000 0.146∗∗∗ −0.001 −0.010 0.008 −0.000 0.006 −0.004 −0.002 −0.002 −0.003 0.000 −0.000 −0.005 −0.002 −0.010 −0.005 −0.005∗∗∗
1.326∗∗∗ 1.096∗∗∗ 1.377∗∗∗
−0.495∗∗∗ −0.462∗∗ −0.669∗∗∗
1.051∗∗∗ 1.173∗∗∗ 1.163∗∗∗ 1.270∗∗∗ 1.015∗∗∗ 1.161∗∗∗ 1.199∗∗∗ 1.155∗∗∗ 0.966∗∗∗ 0.534∗∗ 1.165∗∗∗ 0.754∗∗∗ 0.593∗∗∗ 0.878∗∗∗ 1.351∗∗∗ 0.855∗∗∗
−0.200 −0.425∗∗ −0.442∗∗ −0.528∗∗∗ −0.338∗ −0.470∗∗ −0.428∗∗ −0.451∗∗ −0.329∗
0.895 0.788 0.819 0.666 0.909 0.779 0.792 0.781 0.713 0.724 0.771 0.718 0.610 0.356 0.755 0.762 0.361 0.618 0.843 0.904
30 29 30 14 28 27 30 30 28 30 29 30 28 15 30 22 24 26 29 28
0.909∗∗ 0.524∗∗∗ 0.268 0.514∗∗∗ 0.300∗∗∗ 0.575∗∗ 0.307∗∗ 0.312∗ 0.672∗∗ 0.667∗∗∗ 0.694∗∗ 0.692∗ 0.197∗∗ 0.576∗∗ 0.125
0.035∗∗ 0.024∗∗∗ −0.002 0.010∗∗ 0.067∗∗∗ −0.005 −0.002 −0.002 0.003 0.007∗ 0.002 0.063∗∗∗ 0.002 −0.004 0.021∗∗
0.761 0.822 0.763 0.758 0.941 0.653 0.490 0.680 0.479 0.646 0.030 0.393 0.680 0.364 0.708
19 30 26 29 16 30 28 31 27 30 15 19 31 20 16
0.428∗∗ 0.878∗∗∗ 1.055∗∗∗ 0.425∗∗ 0.754∗∗∗ 0.687∗∗∗ 0.749∗∗∗ 0.621∗∗∗ 0.894∗∗∗ 0.180 0.729∗∗∗ 0.568∗∗∗ 0.487∗∗
−0.412∗∗ −0.432∗∗ −0.590∗∗∗
−0.414∗∗ −0.243
−0.449∗∗
(continued)
390
Alla Golub and Robert A. McDougall Table 14.1 (continued)
Country 1 EAP China Indonesia Korea Malaysia Philippines Singapore Thailand MENA Algeria Iran Israel Jordan Morocco Oman Syria Tunisia Turkey SA India Sri Lanka Pakistan SSA Cote d’Ivoire MUS South Africa
Intercept 2
Time trend 3
First lag 4
0.018∗∗ 0.013∗∗ 0.018∗∗ 0.014∗ 0.010∗∗ 0.009∗ 0.003
0.330 1.019∗∗∗ 1.295∗∗∗ 0.410∗ 1.022∗∗∗ 1.378∗∗∗ 1.232∗∗∗
0.518∗∗ −0.065 0.710∗∗ 0.237∗ 0.354∗∗ 0.370∗ −0.927∗∗ 0.315∗ 0.400∗
0.015 0.067∗ −0.003 −0.009∗∗∗ 0.001 0.021 0.019∗∗ −0.007∗∗ 0.003
0.572∗∗∗ 0.960∗∗ 0.681∗∗∗ 0.870∗∗∗ 0.842∗∗∗ 0.538∗∗ 1.305∗∗∗ 0.835∗∗∗ 0.761∗∗∗
0.300∗∗ −0.034 0.151
−0.004∗∗ 0.019∗ −0.000
1.196∗∗∗ 0.533∗∗ 1.176∗∗∗
−0.396∗∗
0.006 −0.002 0.001
1.165∗∗∗ 0.710
−0.336
0.441 0.305∗∗∗ 0.291∗∗∗ 0.703∗∗ 0.311∗∗ 0.332∗∗ 0.175
0.877∗∗∗ 0.218 0.543∗∗
Second lag 5
−0.464∗∗ −0.634∗∗∗ −0.416∗ −0.554∗∗∗ −0.384∗ −0.414 −0.365 0.384
−0.379∗∗
R-square 6
Obs. 7
0.533 0.908 0.965 0.582 0.883 0.961 0.870
16 27 27 18 29 30 27
0.762 0.917 0.537 0.761 0.447 0.814 0.816 0.865 0.662
25 15 28 23 23 16 19 31 31
0.885 0.946 0.834
30 17 22
0.013 0.817 0.541
16 22 29
∗∗∗ , ∗∗ ,
and ∗ denote significance levels at 0.01, 0.05, and 0.1, respectively. Number of lags is chosen to whiten the errors. Source: Authors’ calculations using Kraay et al. (2000) data.
Differentiating (14.3) we obtain rrg wyr(r) = LAMBWYR(r)∗ (wyrt(r) − wyra(r)),
(14.4)
where rrg_wyr(r) denotes the (absolute) change in the required percentage point rate of growth in the wealth-to-income ratio WYRA(r), and wyrt(r) is the percentage change in the target wealth-to-income ratio. The latter is typically exogenous and zero. If the actual wealth-to-income ratio is higher or lower than the target ratio in a region, then the regional household will decrease or increase,
Household Saving Behavior in the Dynamic GTAP Model
391
respectively, its saving rate to move toward the target ratio. The actual rate of growth in the wealth-to-income ratio is defined as ARG WYR(r) =
dWQHHLD(r)/dTIME dWYRA(r)/dTIME = WYRA(r) WQHHLD(r) −
dINCOME(r)/dTIME . INCOME(r)
(14.5)
Expression (14.5) says that the actual rate of growth in the wealth-to-income ratio is the difference between the rate of growth in household wealth and the rate of growth in income. From Chapter 2, change in household wealth is given in the model by equation (14.6): WQHHLD(r)∗ wqh(r) = WQHFIRM(r)∗ pcgds(r) +WQHTRUST(r)∗ pqtrust(r) + 100∗ SAVE(r)∗ time,
(14.6)
where WQHFIRM(r) denotes household equity in domestic firms; pcgds(r), the percentage change in the price of equity in domestic firms; WQHTRUST(r), the household equity in the global trust; pqtrust(r), the percentage change in the price of equity in the global trust; and SAVE(r), the time rate of (net of depreciation) household saving. Expression (14.6) says that change in household wealth is determined by changes in the prices of old assets plus the rate of new saving both in domestic assets and abroad. We assume that in forming its expectation of rate of growth in its wealth, the regional household takes into account only changes in wealth that are due to new saving, ignoring current changes in asset prices (or, equivalently, assuming that asset prices are constant). Under this assumption, expected change in household wealth is dWQHHLD(r) = SAVE(r)∗ dTIME.
(14.7)
From (14.7), the expected rate of growth in household wealth is dWQHHLD(r)/dTIME SAVE(r) = . WQHHLD(r) WQHHLD(r)
(14.8)
In estimating the rate of growth in the wealth-to-income ratio, the regional household takes the perceived long-run economic growth of its own region into account. The normal rate of growth in income, which we denote as YHAT(r), is a long-run concept and is different from the actual percent change in income in every period of simulation. Using (14.8) and
392
Alla Golub and Robert A. McDougall
YHAT(r) we now can define the expected rate of growth in the wealth-toincome ratio: ERG WYR(r) =
SAVE(r) − YHAT(r). WQHHLD(r)
(14.9)
When the long-run steady state is known, we can set YHAT(r) to its actual long-run value at the beginning of the simulation. For example, if a simulation is characterized by the static steady state with no growth in endowments and no technological progress in the long-run, then we can set YHAT(r) equal to zero in the initial period and keep it equal to zero during the simulation. In scenarios where the long-run economic growth is not known precisely, the household is allowed to adjust YHAT(r) so that as the economy converges toward a steady growth equilibrium, YHAT(r) converges toward the actual steady equilibrium growth rate. We postulate a gradual adjustment mechanism in YHAT(r): DYHAT(r) = LAMBYHAT(r)∗ (y(r) − 100∗ YHAT(r)∗ time),
(14.10)
where DYHAT(r) denotes absolute percentage point change in YHAT(r), and LAMBYHAT(r) is a coefficient of adjustment. Totally differentiating (14.9), we obtain erg wyr(r) =
SAVE(r) (psave(r) + qsave(r) − wqh(r)) − DYHAT(r), WQHHLD(r) (14.11)
where erg_wyr(r) denotes the absolute percentage point change in the expected rate of growth in wealth-to-income ratio; psave(r), the percentage change in the price of saving; and qsave(r), the percentage change in real saving. To achieve the required rate of growth in wealth-to-income ratio, RRG_WYR(r), the regional household should make an appropriate savings decision. Each period, the regional household decides how much of its income should be saved to achieve the required rate of growth in the wealth-to-income ratio. When making the decision, the regional household adjusts its propensity to save through the variable dpsave(r), which is exogenous in the standard GTAP treatment of regional household demand, but endogenous with the new structure. The expected rate of growth in the wealth-to-income ratio is equal to the required wealth-to-income ratio, both in levels: ERG WYR(r) = RRG WYR(r),
(14.12)
Household Saving Behavior in the Dynamic GTAP Model
393
and in differentials: erg wyr(r) = rrg wyr(r).
(14.13)
Substituting (14.3) and (14.9) into (14.12), we see that the rate of saving is given by the condition that the expected rate of growth is equal to the required rate of growth in the wealth-to-income ratio: WYRT(r) SAVE(r) − YHAT(r) = LAMBWYR(r)∗ log . WQHHLD(r) WYRA(r)
(14.14)
The four equations in the following box comprise the new savings module in the model. Box 14.1. New savings module wyra(r) = wqh(r) − y(r)
(14.2)
erg wyr(r) = LAMBWYR(r)∗ (wyrt(r) − wyra(r))
(14.4)
DYHAT(r) = LAMBYHAT(r)∗ (y(r) − 100∗ YHAT(r)∗ time) (14.10) WQHHLD(r) (erg wyr(r) + DYHAT(r)) = psave(r) + qsave(r) − wqh(r) SAVE(r) (14.11)
One new exogenous variable, wyrt(r) denotes the percentage change in the target wealth-to-income ratio. New endogenous variables include the percentage change in the actual wealth-to-income ratio, wyra(r), determined by equation (14.2); the percentage change in the propensity to save, dpsave(r), determined by equation (14.11); and the change in the expected percentage point rate of growth in the wealth-to-income ratio, erg_wyr(r), determined by equation (14.4). We also add a variable representing the absolute change in the normal percentage point rate of income growth, DYHAT(r), determined by equation (14.10).
394
Alla Golub and Robert A. McDougall
5. Data and Parameters To implement the new household saving treatment in the model, we need to set the initial value of the coefficient YHAT(r), the parameters LAMBWYR(r) and LAMBYHAT(r), and the target wealth-to-income ratio WYRT(r). We set the initial perceived long-run economic growth rate YHAT(r) equal to a geometric average of historical annual growth rates. Time-series data for gross national income (GNI) at constant local currency units are obtained from the World Development Indicators database, which covers the period from 1960 to 2004. However, we use just the last 15 years (1990–2004), because for many countries observations for earlier years are missing. Another reason to restrict the series to later years is to avoid incorporating past episodes of rapid economic growth into our current assessments. For example, annual growth rates in the Japanese economy between 1960 and 1973 were in the range of 6 to 13%, but became much lower in recent years. Having the GNI time series from 1990 to 2004, we aggregate over countries according to an aggregation scheme and then calculate the geometric average GNI growth rate, which is a proxy for YHAT(r). Note that precision in estimating YHAT(r) is not very important because it is updatable. However, as we see later, the initial YHAT(r) setting plays a role in setting the regional target wealth-to-income ratios. The parameter LAMBWYR(r) determines the initial distance between the target and actual wealth-to-income ratios and how fast the actual wealthto-income ratio moves toward the target ratio. From (14.14), the initial distance between the wealth-to-income target and actual ratios is given by the formula WYRT(r) 1 SAVE(r) = exp − YHAT(r) . WYRA(r) LAMBWYR(r) WQHHLD(r) (14.15) Other things being equal, smaller LAMBWYR(r) leads to a larger gap between target and actual wealth-to-income ratios in the initial database. If growth in regional wealth is faster than perceived normal economic growth (YHAT(r)), then the expression in parentheses is positive and WYRT(r) is greater than WYRA(r). Similarly, when growth in regional wealth is slower than perceived normal economic growth, then the expression in parentheses is negative and WYRT(r) is less than WYRA(r).
Household Saving Behavior in the Dynamic GTAP Model
395
To achieve desired wealth-to-income ratios, regions adjust their propensities to save. Consider a simple scenario with no growth in endowments and no technological progress over the long-run. For such a scenario, we can set YHAT(r) equal to zero in the initial period and keep it equal to zero during the simulation, removing the adjustment mechanism equation (14.10). The region accumulates more wealth and increases its wealth-toincome ratio, moving it closer to the target. As the distance between the target and actual wealth-to-income ratios diminishes, less growth in wealth is needed, and the saving rate declines. When WYRA(r) = WYRT(r), the regional household saves only to cover its capital depreciation, and the net saving rate SAVE(r) is zero. How much adjustment should we allow in wealth-to-income ratios? An answer to this question will help us set parameter LAMBWYR(r). As discussed in Section 3 and as shown by the insignificant time trends in wealth-to-income ratios in Table 14.1, these ratios are very stable for most countries. This suggests that the target wealth-to-income ratios should be set close to the actual wealth-to-income ratios calculated in the initial database. The size of historical wealth-to-income ratios is also helpful in setting target wealth-to-income ratios. The ratio is larger for industrialized countries than for other countries in the sample. The ratios for industrialized countries in the period 1966–97 are in the range from 1.5 to 5.5, with the exception of Switzerland’s, which is in the range from 4.35 to 6.72.5 For other countries the ratios are in the range from 0.65 to 4 in the East Asia and Pacific region, from 0.5 to 3.7 in Latin America and the Caribbean region, from 0.5 to 3.7 in the Middle East and North Africa, from 0.5 to 1.6 in South America, and from 0.5 to 2.7 in Sub-Saharan Africa. From these observations, it appears that a reasonable upper bound on the target wealth-to-income ratio is 5.5. We now present a specific example illustrating how we set LAMBWYR(r). We use the 22 × 3 aggregation of the GTAP 5 Data Base mentioned in Section 3. Following equation (14.15), the target wealth-to-income ratio is given as ∗ 1 SAVE(r) − YHAT(r) . WYRT(r) = WYRA(r)∗exp LAMBWYR(r) WQHHLD(r) (14.16) 5
This range is for 1983–96 because the data for Switzerland in the Kraay et al. (2000) database are available only for those years.
396
Alla Golub and Robert A. McDougall
Table 14.2. Calculation of target wealth-to-income ratios for 22 × 3 aggregation of
the GTAP 5.4 Data Base when LAMBWYR(r) = 0.05
Region Australia New Zealand China Japan Korea South Asia Canada United States South America Austria Denmark Finland France United Kingdom Ireland Italy Netherlands Portugal Sweden Europe Turkey ROW
WYRAa
YHATb
SAVE/ WQHHLDc
SAVE/ WQHHLD − YHAT
WYRT
2.842 2.476 2.657 4.013 3.296 2.894 2.528 2.626 3.013 4.064 2.356 3.325 3.531 3.050 2.111 3.534 3.593 2.769 2.286 3.676 3.075 3.770
0.037 0.030 0.073 0.013 0.053 0.050 0.028 0.032 0.029 0.021 0.025 0.019 0.020 0.027 0.060 0.015 0.024 0.027 0.017 0.024 0.030 0.034
0.031 0.024 0.100 0.051 0.081 0.065 0.039 0.028 0.022 0.022 0.046 0.027 0.027 0.026 0.047 0.021 0.051 0.012 0.040 0.027 0.031 0.013
−0.007 −0.006 0.027 0.039 0.028 0.014 0.010 −0.003 −0.007 0.001 0.021 0.009 0.007 −0.002 −0.013 0.006 0.027 −0.014 0.023 0.003 0.001 −0.021
2.489 2.187 4.548 8.688 5.737 3.866 3.107 2.470 2.614 4.140 3.588 3.942 4.099 2.946 1.632 3.973 6.179 2.077 3.613 3.918 3.132 2.475
a
Wealth-to-income ratio, based on GTAP 5.4 Data Base. Geometric average growth calculated using WDI database. c Wealth growth rate, based on GTAP 5.4 Data Base. Source: Authors’ calculations using sources indicated in the table. b
The inputs into the calculation, together with the resulting target wealth-toincome ratios when LAMBWYR(r) = 0.05, are given in Table 14.2. For most of the regions in this aggregation, the calculated target wealth-to-income ratios are within the plausible range. The exceptions are Japan, Korea, and the Netherlands. In those countries unusually high target wealth-to-income ratios result from a combination of high initial ratios (WYRA(r)) and rapid initial growth in the ratios, calculated as SAVE(r)/WQHHLD(r) − YHAT(r). The latter is driven by slow growth in income in Japan, rapid growth in wealth in Korea, and both slow growth in income and rapid growth in wealth in the Netherlands. It is possible that for these countries the 1997
Household Saving Behavior in the Dynamic GTAP Model
397
savings-to-wealth ratio recorded in GTAP 5.4 Data Base is abnormally high. Yet what is really important is that even these relatively high target wealthto-income ratios impose an upper bound on actual wealth-to-income ratios in GDyn simulations, and these bounds are much lower than the ratios that would be observed with fixed saving rates, as illustrated in Figure 14.5. How sensitive are the target ratios with respect to LAMBWYR(r)? As can be seen from equation (14.16), as LAMBWYR(r) becomes very small, WYRT(r) approaches infinity. Large LAMBWYR(r) leads to WYRT(r) = WYRA(r). This is supported by Figures 14.7 and 14.8. With LAMBWYR(r) below 0.05 (Fig. 14.7), WYRT(r) becomes very large. In contrast, with LAMBWYR(r) larger than 0.1 (Fig. 14.8), target wealth-to-income ratios are not sensitive to LAMBWYR(r), but remain very close to initial actual ratios WYRA(r). Taking into account the historical fact that wealth-to-income ratios are stable for most countries, setting LAMBWYR(r) to 0.1 or larger seems plausible. However, a high setting of the parameter would lead to rapid changes in savings rates in early years of the simulations; for example, rapid reductions in saving in Japan, Korea, China, and the Netherlands and rapid increases in the residual Rest of World region. However, although it is plausible that such changes will occur eventually, there is little reason to expect them to take place rapidly and in the near future. We therefore prefer to avoid such obtrusive phenomena in our scenario results. It is also possible that, for some scenarios, setting the adjustment parameter too high may make the model unsolvable. Based on these considerations, we conclude that the acceptable range for LAMBWYR(r) is around 0.05 to 0.1. Finally, we need to set LAMBYHAT(r). This plays the same role in the formation of expected growth in income in our new saving treatment as LAMBKHAT(r) plays for expected growth in the capital stock in GDyn investment theory, presented in Chapter 2. Because LAMBYHAT(r) has a similar role to LAMBKHAT(r), we set LAMBYHAT(r) equal to LAMBKHAT(r), that is, to 0.2. As noted earlier, the introduction of the adjustment mechanism for the expected growth in income (equation [14.10]) is for convenience only. In some cases, it may prove more convenient to delete the adjustment mechanism and provide a fixed expected growth rate. In particular, high settings of the adjustment parameter lead to faster adjustment toward equilibrium and larger changes in the model and may make the model unsolvable. Specifically, large changes in saving may lead to negative utility from savings in the top-level Cobb-Douglas regional household demand system. In such cases, the solution is to calibrate YHAT(r) for the specific simulation and set it consistent with long-run trends in the scenario.
398
Alla Golub and Robert A. McDougall
250
target wealth-to-income ratio
200
150
100
50
0 0.01
0.02
0.03
0.04
0.05
LAMBWYR Japan
Korea
United States
France
United Kingdom
Netherlands
Europe
Figure 14.7. Target wealth-to-income ratios in GDyn simulations when LAMBWYR(r) is relatively small. Source: Authors’ calculation. 10 9
target wealth-to-income ratio
8 7 6 5 4 3 2 1 0 0.05 0.07 0.09 0.11 0.13 0.15 0.17 0.19 0.21 0.23 0.25 0.27 0.29 0.31 0.33 0.35 0.37 0.39
LAMBWYR Japan
Korea
United States
France
United Kingdom
Netherlands
Europe
Figure 14.8. Target wealth-to-income ratios in GDyn simulations when LAMBWYR(r) is relatively large. Source: Authors’ calculation.
Household Saving Behavior in the Dynamic GTAP Model
399
6. Two Simulations and Final Notes We conclude this chapter with two illustrative simulations based on the 22 × 3 aggregation of the GTAP Data Base, followed by a discussion of directions for future research to improve the realism of the GDyn model. In the first illustrative simulation there is no growth in endowments and productivity, and only the time variable is shocked. Here, we assume that the long-run steady state is static; that is, there is no growth in productivity or endowments. Under these assumptions net saving tends toward zero over the long-run, as does YHAT(r), resulting in relatively high target wealth-toincome ratios. In terms of the saving rate adjustment, it is an extreme case because it requires a large adjustment in such countries as China, Japan, and Korea in this 90-year simulation. Figure 14.9 shows convergence of wealth-to-income ratios toward their targets in the long-run simulation. Figure 14.10 contrasts these wealth-to-income ratios with the unbounded ratios when the saving rates are constant. Figure 14.11 shows endogenously declining saving rates. Figure 14.12, contrasted with Figure 14.4, illustrates how ratios of net foreign assets to GDP are now bounded. These ratios are still high, however, compared to the historically observed ratios in Figure 14.3. In the second illustrative simulation, we use population, labor, and GDP growth given in the baseline provided in Chapter 5. In the longrun we assume that GDP growth rates gradually converge to levels given in Table 14.3 and that labor and population growth rates remain after 2020 as in 2020. Wealth-to-income ratios, saving rates, and the share of net foreign asset in GDP in this simulation are shown in Figures 14.13–14.15. Figure 14.16 shows the global rate of return to capital in the long-run GDyn simulations at constant and endogenous saving rates. New household saving behavior allows reduction in the rate of growth in world capital and stabilization of the global rate of return to capital. The proposed modification enforces long-run stabilization in regional wealth-to-income ratios, which is critical to obtaining a sensible baseline in light of currently divergent savings rates across countries. As a direction for future research, it would be interesting to compare investmentsaving correlations in the simulations from the model with and without new household saving behavior, and with correlations reported in the studies following Feldstein and Horioka (1980). Finally, it would be interesting to validate projections of the saving rates generated by the new mechanism against those tied to regional aging profiles and reported in Tyers (2005).
400
Alla Golub and Robert A. McDougall Table 14.3. Long-run growth rates assumed in the second
illustrative simulation Region
YHAT
Region
YHAT
Australia New Zealand China Japan Korea South Asia Canada US South America Austria Denmark
0.031 0.030 0.035 0.023 0.035 0.039 0.023 0.026 0.035 0.028 0.025
Finland France United Kingdom Ireland Italy Netherlands Portugal Sweden Europe Turkey ROW
0.032 0.024 0.026 0.035 0.025 0.027 0.027 0.028 0.026 0.035 0.036
Source: Authors’ calculations based on time-series data (1990–2004) for gross national income (GNI) at constant local currency units obtained from the World Development Indicators database.
8
7
6
5
4
3
2
1
19
97 20 01 20 05 20 09 20 13 20 17 20 21 20 25 20 29 20 33 20 37 20 41 20 45 20 49 20 53 20 57 20 61 20 65 20 69 20 73 20 77 20 81 20 85
0
China
Japan
Korea
SA
US
SAM
Turkey
France
Sweden
Italy
Figure 14.9. Wealth-to-income ratio in the GDyn simulations with new household saving behavior, LAMBWYR(r) set to 0.1, and long-run growth assumed to be zero. Source: Authors’ simulation.
Household Saving Behavior in the Dynamic GTAP Model
401
20 18 16 14 12 10 8 6 4 2 0 China
Japan
Korea
US
France
constant saving rates
Italy
Netherlands
Turkey
endogenous saving rates
Figure 14.10. Comparison of wealth-to-income ratios at the end of 90-year GDyn simulations. Source: Authors’ simulations. 0.3
0.25
0.2
0.15
0.1
0.05
09 20 13 20 17 20 21 20 25 20 29 20 33 20 37 20 41 20 45 20 49 20 53 20 57 20 61 20 65 20 69 20 73 20 77 20 81 20 85
01
05
20
20
20
19 97
0
China
Japan
Korea
SA
US
SAM
Turkey
France
Sweden
Italy
Figure 14.11. Saving rates in GDyn simulations with new household saving behavior, with long-run growth assumed to be zero. Source: Authors’ simulation.
402
Alla Golub and Robert A. McDougall
15
10
5
0
-5
19 97 20 01 20 05 20 09 20 13 20 17 20 21 20 25 20 29 20 33 20 37 20 41 20 45 20 49 20 53 20 57 20 61 20 65 20 69 20 73 20 77 20 81 20 85
-10
China
Japan
Korea
SA
US
SAM
Turkey
France
Sweden
Italy
Figure 14.12. Net foreign assets as a share of GDP in GDyn simulations with new household saving behavior, with long-run growth assumed to be zero. Source: Authors’ simulation. 6
5
4
3
2
1
19 97 20 01 20 05 20 09 20 13 20 17 20 21 20 25 20 29 20 33 20 37 20 41 20 45 20 49 20 53 20 57 20 61 20 65 20 69 20 73 20 77 20 81 20 85
0
China
Japan
Korea
SA
US
SAM
Turkey
France
Sweden
Italy
Figure 14.13. Wealth-to-income ratio in GDyn simulations with new household saving behavior, LAMBWYR(r) set to 0.1, and long-run growth rates as given in Table 14.3. Source: Authors’ simulation.
Household Saving Behavior in the Dynamic GTAP Model
403
0.3
0.25
0.2
0.15
0.1
0.05
01 20 05 20 09 20 13 20 17 20 21 20 25 20 29 20 33 20 37 20 41 20 45 20 49 20 53 20 57 20 61 20 65 20 69 20 73 20 77 20 81 20 85
20
19
97
0
China
Japan
Korea
SA
US
SAM
Turkey
France
Sweden
Italy
Figure 14.14. Saving rates in GDyn simulations with new household saving behavior, with long-run growth rates as given in Table 14.3. Source: Authors’ simulation. 15
10
5
0
-5
China
Korea
Sweden
81
85
20
20
73
77
20
20
65
69
20 France
20
61
57
Turkey
20
53
20
49 20
SAM
20
45
41
US
20
37
20
33 SA
20
20
25
29 20
21
Japan
20
17
20
20
09
13
20
20
05 20
01 20
19
97
-10
Italy
Figure 14.15. Net foreign assets as a share of GDP in GDyn simulations with new household saving behavior, with long-run growth rates as given in Table 14.3. Source: Authors’ simulation.
404
Alla Golub and Robert A. McDougall
0.11
0.1
0.09
0.08
0.07
0.06
0.05
Net rate of return at constant saving rate
41 20 45 20 49 20 53 20 57 20 61 20 65 20 69 20 73 20 77 20 81 20 85
20
33
29
25
37
20
20
20
21 20
20
13
09
05
01
17 20
20
20
20
20
19
97
0.04
Net rate of return at new household saving behavior
Figure 14.16. Global net rate of return to capital at constant and endogenous saving rates in the long-run GDyn simulations. Source: Authors’ simulation.
References Ando, A. and F. Modigliani. 1963. “The ‘Life Cycle’ Hypothesis of Saving: Aggregate Implication and Tests.”American Economic Review 53(1), 55–84. Benjamin, N. 1994. “Investments, Expectations, and Dutch Disease: A Comparative Study (Bolivia, Cameroon, Indonesia).” In J. Mercenier and T. N. Srinivasan (eds.), Applied General Equilibrium and Economic Development (pp. 235–51). Ann Arbor: University of Michigan Press. Burniaux, J., G. Nicoletti, and J. Olivera-Marins, 1992. GREEN: A Global Model for Quantifying the Costs of Policies to Curb CO2 Emissions. OECD Economic Studies 19. Paris: OECD. Feldstein, M., and C. Horioka. 1980. “Domestic Savings and International Capital Flows.” Economic Journal 90, 314–29. Hertel, T. W. and M. E. Tsigas. 1997. “Structure of GTAP.” In T. W. Hertel (ed.), Global Trade Analysis: Modeling and Applications (pp. 13–73). Cambridge: Cambridge University Press. Howe, H. 1975.“Development of the Extended Linear Expenditure System from Simple Saving Assumptions.”European Economic Review 6, 305–10. Ianchovichina, E. and R. McDougall. 2001. Structure of Dynamic GTAP. GTAP Technical Paper 17. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Available at https://www.gtap.agecon.purdue.edu/resources/download/160 .pdf.
Household Saving Behavior in the Dynamic GTAP Model
405
Kouparitsas, M. 2004. How Worrisome is the U.S. Net Foreign Debt Position? Chicago Federal Letter, No. 2002. Available at http://www.chicagofed.org/publications/fedletter/ cflmay2004 202.pdf. Kraay, A., N. Loayza, L. Serven, and J. Ventura. 2000, July. Country Portfolios. National Bureau of Economic Research Working Paper Series No. 7795:1–61. Washington, DC: NBER. McDougall, R. 2002. A New Regional Household Demand System for GTAP. GTAP Technical Paper No. 20. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Available at https://www.gtap.agecon.purdue.edu/resources/download/1593.pdf. McKibbin, W. J. and P. J. Wilcoxen. 1995. “The Theoretical and Empirical Structure of G-Cubed.” Mimeo, Brookings Institution. Modigliani, F. and R. Brumberg 1954. “Utility Analysis and the Consumption Function: An Interpretation of Cross-Section Data.” In K. K. Kurihara, ed., Post-Keynesian Economics (pp. 388–436). New Brunswick, NJ: Rutgers University Press. Tyers, R. 2005. Aging and Slower Population Growth: Effects on Global Economic Performance. Paper presented to the Experts’ Meeting on Long Term Scenarios for Asia’s Growth and Trade, November 10–11, Manila. Available at http://www.adb .org/Documents/Events/2005/Experts-Meeting/agenda.asp.
FIFTEEN
Implications for Global Economic Analysis Elena I. Ianchovichina and Terrie L. Walmsley
With the launch of the GTAP project two decades ago it became possible for economists who are not specialists in AGE modeling to start using the general equilibrium approach for economic policy analysis, while at the same time avoiding the excessive costs of collecting data and programming. Not only the large number of citations of the first GTAP book documenting the standard GTAP model (Hertel 1997) but also the large number of individuals around the world using different versions of the framework and the database for policy research and analysis are testimony to the success of this project. The GTAP network of researchers and analysts has been a key asset to the project: Members of the team not only benefited from the availability of data and advances in the modeling framework but also contributed models and data to the project. Knowledge generated by network members has been disseminated at annual conferences, and new ideas have been implemented quickly as the initial data and model lowered the costs of each new extension. In this respect GDyn is no different from other extensions of the GTAP model, although it did benefit substantially from the fact that there was a global database and standard model to start from. However, just like the GTAP model, GDyn will make it possible for others to use dynamic CGE modeling techniques to analyze specific issues of interest without incurring the costs of building a global dynamic CGE model and corresponding databases. Moreover, like GTAP, the GDyn model and database provide a standard dynamic modeling platform on which others can build. The book presents a number of applications developed by such users of GDyn. We hope that, just as did the GTAP book (Hertel 1997), this book will support the development of new applications and in the process result in improvements to the data, parameters, and economic behavior of GDyn. 406
Implications for Global Economic Analysis
407
Synopsis This book documents GDyn, including its model structure, database, parameters, construction of baseline, and software. The GDyn model represents an important extension of the GTAP model that takes into account capital accumulation and ownership over time. First, the GDyn model introduces the theory of adaptive expectations and the gradual movement toward a steady-state equilibrium into the theory determining regional investment. Second, this investment adds to capital stocks available for production over time, thereby providing the dynamic effects of a policy shock from the accumulation of capital. Third, one of the key drawbacks of the GTAP model was that all capital was assumed to be owned by the region in which it was located, and therefore all income earned on capital located in the region accrued to that region’s households. This assumption meant that the GTAP model failed to take account of the implications of foreign ownership on welfare and the trade balance. As the importance of foreign investment grows and we seek to investigate the long-run impact of policies, this assumption becomes more contentious. By incorporating capital accumulation and foreign ownership, the GDyn model addresses this drawback of the GTAP model. Finally, the GDyn model allows the user to show the transition path by which a policy affects the global economy. In undertaking these extensions to the underlying model, the GTAP Data Base has also been enhanced with additional data on foreign and domestic income flows, as well as a number of new parameters required by the theory; for instance, measuring the rate of convergence in rates of return. Underlying the GTAP project are a number of core values inherited from the Impact Project/Center of Policy Studies in Australia – public availability, documentation, replication, and ease of use. We have continued to hold onto these values throughout the development of the GDyn model and database, because we believe that the benefits to trained users far exceed the costs. With the inclusion of dynamics into the GTAP model, however, the complexity of these models has the potential to increase dramatically. In developing the GDyn model, the key developers have been acutely aware of the potential problems of building such a complex dynamic model and have to the extent possible tried to retain many of the benefits of the standard GTAP model. The standard GDyn model therefore keeps many of the features of the standard GTAP model, imposing complexity only where required. Despite these efforts the model does add additional complexity, not simply because of the new dynamic equations added to the model but also because of the need to understand
408
Elena I. Ianchovichina and Terrie L. Walmsley
how to develop more complex baseline and policy scenarios, any additional software, and the more intricate methods required to measure welfare in a dynamic setting. All these new complications are examined in this book. The book also presents a collection of major and diverse GDyn applications, along with the tools needed to replicate them. These are needed because we believe that replication is an essential ingredient of empirical research. We also hope that the book will serve to motivate others to improve the model and data and use them to expand the range of issues one can address with GDyn. The book will also be used as a principal textbook for GDyn courses. This book project is a culmination of a truly collective effort that has been sustained over a number of years. The team has benefited from the availability of the GDyn model and data (Ianchovichina and McDougall 2001), as well as a standard platform for implementing applications that will facilitate the documentation and replication of future GDyn applications. The experience with the first GTAP book (Hertel 1997) shows that standardization helps the pedagogy and replication processes, while imposing no limits on innovations related to modeling and applications. The time saved building a new model and data for dynamic CGE models is significant, allowing for the development of more complex model versions tailored to different tasks, for careful attention to the nature of the simulation design being undertaken, and to additions to the database. Experienced modelers may find the database itself of great interest. It can provide the starting point for applications using different models.
Evaluation Many of the types of issues that GDyn is especially well suited for are global and long term in nature. These are issues that cannot be analyzed to one’s satisfaction without paying attention to the transition path along the way and without taking account of capital accumulation and ownership; providing a model to take those factors into account was the primary motivation behind the development of the GDyn model. In Chapter 8, Hertel, Walmsley, and Ianchovichina use GDyn to compare alternative time paths of investment and ownership in the Chinese economy and quantify the benefits from China’s WTO accession when increases in investment flows into China – a result of accession reforms including in the services sector – are taken into account. They find much larger gains from WTO accession than those predicted by earlier studies, which ignored the impact of accession
Implications for Global Economic Analysis
409
on productivity in services and abstracted from capital accumulation and foreign investment. Ex-post analyses of the impact of China’s WTO accession suggest that China’s actual gains from the reforms were large and close to the gains suggested in Chapter 8. In Chapter 9, Hertel, Walmsley, and Itakura use GDyn to capture the dynamic effects of the “new-age” FTA between Japan and Singapore, as well as the potential impacts on international investment flows and wealth. Taking account of foreign capital ownership is especially important for this analysis because international investment has boomed in recent decades in East Asia, as Japan and other high-wage economies started outsourcing production to low-wage economies to cut production costs. Explicit modeling of the ownership of regional investment in Japan and other Asian economies allows the authors to track the accumulation of Japanese wealth by foreigners and Japan’s ownership of domestic and foreign assets. Income accruing from the ownership of these foreign and domestic assets can then be appropriately incorporated into total regional income and, hence, the computation of welfare for Japan, Singapore, and the Rest of World. The authors find that the impact of the FTA on investment, capital accumulation, and economic growth is significant – particularly in Singapore. It is critical to have a global dynamic framework that captures the accumulation of capital and other stock variables of importance in the chapters on environmental issues. In Chapter 10 Ianchovichina, Darwin, and Shoemaker extend GDyn to study the global effects of economic and population growth and the impact of a slowdown in agricultural TFP growth on farm and forest resources. In this extension land is divided into six land classes based on the length of the growing season, and land is allocated to sectors without losing sight of its inherent productivity differences. They find that slower agricultural productivity growth could have negative environmental effects because it is associated both with reductions in forestland and increases in environmental or ecological damages on remaining forestlands. Chapter 11 aims to shed light on the role of global economic integration for land use change. Golub and Hertel add to GDyn the most important economic features driving global land demand and supply, including an econometrically estimated, international demand system for commodities and an additional database and equations characterizing land use by agroecological zones. This application is a case study showing the interplay between economic and environmental issues in a global context. It is also an example of how GDyn can be extended to address environmental issues
410
Elena I. Ianchovichina and Terrie L. Walmsley
that have grown in importance in the past few years and will become even more prominent in the future. In Chapter 12 Itakura, Hertel, and Reimer investigate whether increases in productivity associated with increases in trade and investment as part of FTAs have a significant impact on the CGE results of FTAs. To test their hypothesis they incorporate econometric evidence into GDyn, which allows the exploration of FDI issues in a CGE framework. They found that the conventional, dynamic AGE model captures more than half of the ensuing GDP and trade changes. The rest could be attributed to the procompetitive and FDI-productivity linkages, whereas the export-productivity linkage played a minor role. Their application is important because it provides a good example of a simulation exercise that sets priorities for future policyoriented econometric work aimed at refining the estimates used in this chapter. In Chapter 13, Tyers and Shi focus on an important long-run issue – the global population and labor force changes that will affect not only the performance of regional economies but also the distribution of global investment and savings and, hence, capital account flows and regional production costs, and therefore the pattern of global trade. They add a demography submodule to GDyn and relax the assumption of a fixed private saving rate. Their extension offers a possibility to expand the treatment of the government sector in GDyn and to undertake a number of potential improvements to facilitate the analysis of the fiscal implications of aging. Finally, Golub and McDougall undertake a major validation exercise in Chapter 14. Their backcasting work is important because the quality of the insights generated by GDyn depends on the validity of the modeling framework, including model, data, and parameters. Golub and McDougall compare the gross and net foreign assets and liabilities in GDyn simulation to those in a country portfolio database constructed by Kraay et al. (2000) for the period between 1966 and 1997. Their results show that the fixed propensities to save in the model coupled with rising incomes lead to unrealistically large international capital flows in the very long run, which is defined as a period longer than 20 years. They create too much saving and, as a result, too much investment and capital in the world; they allow chronic rapid saving or dissaving, and therefore too great foreign assets or liabilities in individual regions. When Golub and McDougall modify household saving behavior by endogenizing saving rates, the model is able to enforce long-run stabilization in a regional wealth-to-income ratio, which is observed in the data and is critical to obtaining a sensible baseline in light of the currently divergent
Implications for Global Economic Analysis
411
saving rates across countries. Their work elevates high on the list of priorities for future work with GDyn comparisons of investment-saving correlations in GDyn simulations with fixed and endogenous savings, with those in the studies following Feldstein and Horioka (1980), and those in Chapter 13 of Tyers and Shi.
Future Directions and Conclusions GDyn is an important tool for economic policy analysis because the model and data enable economists to investigate the impact of a number of global issues that will continue to grow in importance and dominate economic policy debate in years to come: climate change; globalization, including the movement of goods, services, capital, and labor across regions; and the rapid changes in the profile of the global workforce and population. The Center for Global Trade Analysis, the home of the GTAP project, will continue to rely on the GTAP network to assist in advancing the GTAP Data Base and models to ensure that tools are publicly available to the network to investigate the important policy issues of the day. These tools, including the GDyn model and Data Base, will continue to be made publicly available, along with essential training on these tools for policy analysts and researchers around the world. It is hoped that GDyn will support the development of new applications and in the process result in improvements to the data, parameters, and economic behavior of GDyn, as has been the case with the GTAP model. References Feldstein, M. and C. Horioka. 1980. “Domestic Savings and International Capital Flows.” Economic Journal 90, 314–29. Hertel, T. W. 1997. Global Trade Analysis: Modeling and Applications. Cambridge: Cambridge University Press. Ianchovichina, E. and R. McDougall. 2001. Structure of Dynamic GTAP. GTAP Technical Paper No. 17. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Kraay, A., N. Loayza, L. Serven, and J. Ventura. 2000, July. Country Portfolios. National Bureau of Economic Research World Paper Series No. 7795: 1–61. Washington, DC: NBER.
APPENDIX
Negative Investment: Incorporating a Complementarity into the Dynamic GTAP Model Terrie L. Walmsley and Robert A. McDougall It has been found that in some cases the investment mechanisms in the GDyn model can result in gross investment falling below zero. This will occur when the investment mechanisms cause the actual rate of return to rise considerably, lowering investment to unacceptable levels; for example, if there is a large error in expectations (the expected rate of return far exceeds the actual rate) or the expected and actual rates of return are far below the target rate of return. In this appendix a complementarity is introduced to ensure that in these cases investment does not fall below a minimum level. We use the method developed by Harrison, Horridge, Pearson and Wittwer (2002) for introducing complementarities. The method is designed specifically for models implemented using the GEMPACK software. First the equations from the GDyn model are used to demonstrate the negative investment problem in the GDyn model. Following this we introduce the complementarity and show, using the example, how the complementarity works.
1. Negative Investment Here we outline an example of the case where investment falls to an unacceptable level. The example is based on a small 7 commodities1 by 7 region2 aggregation of the GDyn Data Base.3 1 2 3
Primary and Agricultural commodities; Processed Food; Natural Resources; Textiles; Manufacturing; Transport, Machinery and Equipment; and Services. US and Canada; EU; Japan; Newly Industrializing economies; South East Asia; China and Hong Kong; and Rest of World. The application related to this Appendix is Ch7HO7 × 7 gdyn v31c 97.zip and is available for download on the Web site at: https://www.gtap.agecon.purdue.edu/models/Dynamic/ applications.asp.
413
414
Terrie L. Walmsley and Robert A. McDougall
Actual Expected
Target
ROW
CHN_HKG
SEA
NIC
JPN
EUN
USC
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
Figure 1. Rates of return.
Figure 1 shows the target and actual, expected rates of return in this 7 × 7 aggregation. Expected rates of return hover around 13 and 14%, with the target rate of return equal to 13.5%. There is not a great deal of variation between the expected and target rates of return (Figure 1). In contrast, actual rates of return vary greatly across regions, with a very low 9% in Japan and 19% in South East Asia. Let us consider these two extreme cases. In South East Asia the actual rate of return is very high; it is much higher than the expected rate of return. The actual rates of return are based on returns to capital relative to the price of capital goods, whereas expected rates of return are implied or calibrated from the data. In countries where rates of return are high, theory tells us that investment should also be high as investors seek high returns. If the data agree, then the expected and actual rates of return will be equal. However, if the data show small levels of investment in South East Asia, then either the data or the theory is deemed to be incorrect. In the GDyn model, the theory is adjusted to include errors in expectations. Expected rates of return therefore reflect the rates of return implied by the
0.2
Appendix: Negative Investment
415
2,500,000
2,000,000
1,500,000
1,000,000
500,000
0 1997
2002
2007
2012
2017
2020
Figure 2. Investment in Japan over time ($US).
data, and the difference between the actual and expected rates of return are errors in expectations. Hence although rates of return are high, investment is low in South East Asia, and hence the expected rates of return are much lower than the actual rates. In the case of Japan the reverse is true: The low rates of return suggest low levels of investment, yet investment is very high, suggesting that expected rates of return are much higher. As a result of Equation 1 in Box 1, expected rates of return in South East Asia will rise, and those in Japan will fall (relative to the target). Assuming no change in the target, the rate of growth in the expected rate of return (Equation 2) is therefore negative for South East Asia and positive for Japan as errors in expectations are eliminated and the expected rate of return is also drawn toward the target. If the expected rate of return is to rise in South East Asia, then so too must investment, as investors choose to invest where returns are highest (Equation 3). The rise in investment also causes actual rates of return to fall as more investment leads to more capital stocks and lower returns. In Japan the reverse is true. Lower expected rates of return lead to lower investment (Equation 3), and hence actual rates of return rise. The fall in expected rates of return and the rise in actual rates cause the errors in expectations to be eliminated faster.
416
Terrie L. Walmsley and Robert A. McDougall
Box 1: Investment theory of GDyn 1. Rˆ E = −φ(Kˆ − dt) − μ log(R E /R A )dt 2. E = [Rˆ T − Rˆ E ] I 3. E = φ [Iˆ − Kˆ ] + φd K Rˆ A 4. ω = Kˆ + − dt φ 5. Kˆ = Idt where: R E is the expected rate of return R A is the actual rate of return R T is the target rate of return K is the quantity of capital stocks I is Investment is the normal growth rate of capital – that rate of growth consistent with no change in rates of return dt is change in years log(R E /R A ) is a measure of the errors in expectations φ is the elasticity of the rate of return with respect to capital stocks μ is the rate at which errors in expectations are eliminated E is expected rate of growth in the expected rate of return is the rate at which differences in the expected and actual rate of return are eliminated ω is the change in the normal growth rate of capital ˆ the hat represents proportionate change. Note: All variables except R T are indexed by region. The problem in the case of Japan is that the level of investment required to eliminate errors in expectations and ensure convergence to the target rate of return (at the rates given in the database) may require the level of gross investment to fall below zero. This creates problems because it is assumed that capital (to which investment is added, Equation 5) cannot be discarded, but will depreciate gradually over time (at a rate of 4% in the GDyn Data Base). In our small 7 × 7 aggregation, this occurs in Japan. Rates of return are so low that the gradual elimination of errors results in investment falling too quickly.
Appendix: Negative Investment
417
2. Complementarities To overcome this we introduce a complementarity into the model that stops gross investment falling below some minimum investment-to-capital ratio (MINIKRAT: currently set at 0.0054 ). A full list of the equations incorporated are provided in Box 2. Coefficient (parameter) MINIKRAT #Minimum ratio of Investment to Capital stock #; Read MINIKRAT from file GTAPPARMK header “MIKR”; To implement this we define the level of real gross investment QREGINV and the level of real gross investment (QREGINV_D) determined by the dynamic model Equations (1–5) and the minimum level of real investment (QMININV), which is related to the minimum investment-to-capital ratio (MINIKRAT). Variable (levels)(all,r,REG) QREGINV(r) # real GROSS investment in r (qty of “cgds” output) #; Variable (levels)(all,r,REG) QREGINV D(r) # real GROSS investment in r (qty determined by dynamics) #; Variable (levels)(all,r,REG) QMININV(r) # Minimum GROSS investment allowed in r #; Formula (initial)(all,r,REG) QMININV(r) = MINIKRAT∗ VK(r); The complementarity is then introduced in the following way: Complementarity (variable = QREGINV, lower bound = QMININV) ! Ensures QREGINV = QREGINV D when QREGINV D > QMININV else QREGINV = QMININV ! C QREGINV (all,r,REG) QREGINV(r) − QREGINV D(r); 4
Note that in older versions of the GDyn Data Base this may be set equal to 1. You must change it to 0.005 as it will not work with a value of 1.
418
Terrie L. Walmsley and Robert A. McDougall
This equation simply states that when QREGINV D < or = QMININV, QREGINV = QMININV QREGINV D > QMININV, QREGINV = QREGINV D. In most cases the dynamic equations result in a level of investment greater than the minimum, and hence the value determined by the dynamic equations is used. However, on rare occasions where the level of investment is too low, investment will be set equal to the minimum. Box 2: Equations Variable (levels)(all,r,REG) QREGINV(r) # real GROSS investment in r (qty of “cgds output”) #; Formula (initial)(all,r,REG) QREGINV(r) = sum(k,CGDS COMM, VOA(k,r)); Equation E QREGINV(all,r,REG) p QREGINV(r) = qcgds(r); Variable (levels)(all,r,REG) QREGINV D(r) # real GROSS investment in r (qty det er min ed by dynamics) # ; Formula (initial)(all,r,REG) QREGINV D(r) = sum(k,CGDS COMM, VOA(k,r)); Equation E QREGINV D(all,r,REG) p QREGINV D(r) = qcgds d(r); Coefficient (parameter) MINIKRAT # Minimum ratio of Investment to Capital stock #; Read MINIKRAT from file GTAPPARMK header “MIKR”; Variable (levels)(all,r,REG) QMININV(r) # Minimum GROSS investment allowed in r #; Formula (initial)(all,r,REG) QMININV(r) = MINIKRAT∗ VK (r); (continued)
Appendix: Negative Investment
419
Box 2 (continued) Equation E MININV(all,r,REG) p QMININV(r) = qk(r); Complementarity (variable = QREGINV, lower bound = QMININV) ! Ensures QREGINV = QREGINV D when QREGINV D > QMININV else QREGINV = QMININV ! C QREGINV (all,r,REG) QREGINV(r) – QREGINV D(r);
In the 7 × 7 application outlined earlier, Japan’s real investment is set equal to the minimum level. The dynamic equations push Japan’s investment down by 99.94%, or from 1222350 in the initial database to 66334.04, which is lower than the minimum of 75812.88. As a result, real investment is set equal to the minimum and falls by only 94.5%.5 Figure 2 shows the resulting changes in investment over time. Although investment falls in the initial period, the target rate of return also falls over time, causing investment in Japan to rise in later years. References Elbehri, A. and K. Pearson. 2005. Implementing Bilateral Tariff Rate Quotas in GTAP Using GEMPACK. GTAP Technical Paper No. 18. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Harrison, J., M. Horridge, K. Pearson, and G. Wittwer. 2002. “A Practical Method for Explicitly Modeling Quotas and Other Complementarities.” Computational Economics 23, 325–41. Hertel T. W. (ed.). 1997, Global Trade Analysis: Modeling and Applications. Cambridge: Cambridge University Press.
5
Note that real investment is affected by this complementarity within the simulation, however at the beginning of each period the value of investment and capital are used to determine the real quantities.
Glossary of GDyn Notation Terrie L. Walmsley
1. Overview Specification of the GDyn Data Base and model requires a vast amount of notation. This notation has been carefully chosen to be brief, yet descriptive. Since many of the variables used in the GDyn model are also in the standard GTAP model, this glossary restricts itself to the additional variables in the GDyn model used throughout this book. The variables correspond to the GEMPACK representation of the model, GDynv34.tab, as specified in February 2009. It lists the sets and subsets, base data, derivatives of the base data, and variables used in the model. We hope that the glossary will be useful for both new and experienced users of the GTAP framework. Some important conventions follow: a) Sets and parameters are denoted in uppercase. b) The levels form of the variables in GDyn is denoted in uppercase. Percentage changes in variables are denoted in lowercase (linearized form of the variables). For instance, QK(r) is the quantity of capital located in region r in the levels form, and qk(r) = [d QK(r)/QK(r)]∗ 100% is the linearized form of this variable. c) The GDyn Data Base comprises only value flows (in their levels form). Database values are accordingly written in uppercase. These are declared as coefficients in the GEMPACK code and are updated using percentage changes in the component prices and quantities, after each step in the solution. The database stores the minimal amount of information. No redundancies are permitted. d) Derivatives of the database variables are also in levels form. There are two types of derivatives: value flows and shares. The derivative variables naturally get updated following each update of the database. 421
422
Terrie L. Walmsley Table 1. Aide memoire for the naming conventions used in the GDyn model
Variable
Longnamea
Definition
First letter: income or wealth Y
Income
W
Wealth
Second letter: type of ownership Q Equity Third letter: owner and receiver of income H HHLD Household
T
TRUST
_ (underscore)
Trust
All owners
Last letter: investment or payer of income T TRUST Trustc F
FIRM
None
a b c
Firm
All investments
Exampleb YQTRUST (or yqt) − Income on equity owned by the trust WQHHLD(r) (or wqh(r)) − Wealth in equity owned by households in region r YQ_TRUST (or yq_t) − Income on equity paid by the trust YQHHLD(r) (or yqh(r)) − Income earned on equity owned by the regional household r YQTFIRM(r) (or yqtf(r)) − Income on equity paid to the trust by firms in region r WQ_TRUST (or wq_t) − Total wealth of trust − Sum of all households’ ownership in the trust YQHTRUST(r) (or yqht(r)) − Income on equity paid to household r by the trust YQHFIRM(r) (or yqhf(r)) − Income on equity paid to household r by domestic firms WQTRUST (or wqt) − Total wealth of trust − Sum of all investments made by the trust
In order to distinguish variables from coefficients, the last letter of a coefficient is always extended to its longname. Coefficients are denoted in upper case and percentage change variables are denoted in lower case. Note that the trust is an owner of wealth invested in regional firms and it is also owned by regional households.
e) The GDyn model consists of a system of linearized equations with all variables appearing in percentage change form. GEMPACK solves for percentage changes in (endogenous) prices and quantities, thereupon relinearizing the model and solving it once again. Before commencing with the glossary, the user may wish to review Table 1, which is provided to assist the user in understanding the naming conventions used in introducing the new income and wealth data.
Glossary of GDyn Notation
423
2. Glossary Sets and subsets Sets CGDS_COMM DEMD_COMM ENDW_COMM ENDWC_COMM ENDWM_COMM ENDWS_COMM NSAV_COMM PROD_COMM TRAD_COMM
Capital Goods Commodities (“cgds”) Demanded Commodities Endowment Commodities Capital Endowment Commodity (“capital”) Mobile Endowment Commodities Sluggish Endowment Commodities Non-Savings Commodities Produced Commodities Traded Commodities
Subsets CGDS_COMM CGDS_COMM DEMD_COMM ENDW_COMM ENDWC_COMM ENDWC_COMM ENDWM_COMM ENDWS_COMM PROD_COMM TRAD_COMM TRAD_COMM
in NSAV_COMM in PROD_COMM in NSAV_COMM in DEMD_COMM in ENDW_COMM in NSAV_COMM in ENDW_COMM in ENDW_COMM in NSAV_COMM in DEMD_COMM in PROD_COMM
Example: The 3 × 3 (3 regions × 3 sectors) economy CGDS_COMM DEMD_COMM ENDW_COMM ENDWC_COMM ENDWM_COMM ENDWS_COMM NSAV_COMM PROD_COMM REG TRAD_COMM
= {capital goods} = {land, labor, capital, food, manufacturing, services} = {land, labor, capital} = {capital} = {labor} = {land, capital} = {land, labor, capital, food, manufacturing, services, capital goods} = {food, manufacturing, services, capital goods} = {usa, eu, row} = {food, manufacturing, services}
424
Terrie L. Walmsley
BASE DATA KHAT(r)
RORGEXP(r) RORGTARG YQHFIRM(r) YQHTRUST(r) YQTFIRM(r)
Price-neutral rate of growth in the capital stock. The rate of growth in the capital stock is, agents believe, consistent with a constant rate of return on capital. Expected gross rate of return. Target gross rate of return. Income on equity paid to households r by domestic firms. Income on equity paid to regional household r by the trust. Income on equity paid to the trust by the regional firm r.
∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG
PARAMETERS INC(r)
LAMBKHAT(r) LAMBRORG(r)
LAMBRORGE(r)
RIGWQ_F(r)
RIGWQH(r)
RORGFLEX(r)
Initial equilibrium regional expenditure data INC is set equal to INCOME and does not change during a simulation. Coefficient of adjustment for the normal rate of growth. Coefficient of adjustment for the rate of return. Controls the rate at which agents aim to adjust rates of return in response to differences between expected and target rates. If the target rate exceeds the expected rate by 1%, agents aim to adjust the rate by LAMBRORG% per period. Coefficient of adjustment for the expected rate of return. Controls the rate at which agents adjust expectations of rates of return in response to differences between expected and actual rates. If the actual rate exceeds the expected rate by 1%, agents adjust the expected rate by LAMBRORGE% per period. Rigidity on total wealth in equity invested by both the trust and the regional household in the regional firm r. Rigidity on total wealth in equity owned by the regional household r and invested in both the trust and domestic firms. Flexibility of the gross rate of return. Agents expect each 1% expansion in the capital stock to reduce the gross rate of return by RORGFLEX %.
∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG
∀ r ∈ REG
∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG
Glossary of GDyn Notation
425
DERIVATIVES OF BASE DATA RORGROSS(r) WQ_FHHLDSHR(r) WQ_FIRM(r) WQ_FTRUSTSHR(r) WQ_THHLDSHR(r) WQ_TRUST WQHFIRM(r) WQHHLD(r)
WQHTRUST(r) WQT_FIRMSHR(r) WQTFIRM(r) WQTRUST YQ_FIRM(r) YQ_FHHLDSHR(r)
YQ_FTRUSTSHR(r)
YQ_TRUST YQ_THHLDSHR(r)
YQHHLD(r) YQT_FIRMSHR(r)
YQTRUST
Gross rate of return. Share of total wealth on equity of regional firm r which is owned by domestic households. Total wealth in equity invested by the domestic household and by the trust in regional firms. Share of wealth on equity in regional firm r which is owned by the trust. Share of wealth in the equity of the trust which is owned by the regional household r. Total wealth in equity invested by all regional households in the trust. Wealth in equity owned by the regional household r and invested in domestic firms. Total wealth in equity owned by regional household r and invested in both the trust and domestic firms. Wealth in equity owned by household r and invested in the trust. Share of wealth in equity owned by the trust which is invested in regional firms r. Wealth in equity owned by the trust and invested in regional firms r. Total Wealth in equity owned by the trust and invested in all regional firms. Total Income on equity paid to both the trust and the regional household by regional firms r. Share of total income on equity paid by regional firm r which is paid to domestic households (domestic share of firm income). Share of total income on equity paid by regional firms which is paid to the trust (trust’s share of firm income). Total income paid by the trust (i.e., to all regional households). Share of total income paid by the trust which is paid to the regional household r (regional household’s share of trust income). Total income on equity paid to the household r by both the domestic firms and by the trust. Share of total income in equity paid to the trust which was paid by the regional firm r (i.e., share of trust income paid by firm r). Total income in equity paid to the trust by all regional firms.
∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG
∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG
∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG
∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG
426
Terrie L. Walmsley
VARIABLES DKHAT(r) DROR(r) DRORT DRORW(r) erg_rorg(r) ERRRORG(r) pqtrust qk(r) rorga(r) rorge(r) rorgt rorwqht(r) rorwqtf (r) SDROR(r) SDRORT(r) SDRORTW SDRORW sqk(r) sqkworld srorge(r) time wq_f (r)
wq_t wqh(r)
wqhf (r)
Change in the normal rate of growth in capital. Change in the rate of return − net or gross. Change in the target rate of return − net or gross. Change in the world-wide average rate of return − net or gross. Change in the expected rate of growth in the gross rate of return. Measure of error in the rate of return. Percentage change in the price of equity invested in the trust. Percentage change in the capital stock. Percentage change in the actual gross rate of return. Percentage change in the expected gross rate of return. Percentage change in the target gross rate of return. ROR on wealth in equity owned by regional household r and invested in the trust. ROR on wealth in equity owned by the trust and invested in regional firms. Change in the region-specific shift in the rate of return. Change in the region-specific shift in the target rate of return. Change in the world-wide shift in the target rate of return. Change in the world-wide shift in the rate of return. Percentage change in arbitrary region-specific shock to capital stock. Percentage change in arbitrary region-generic shock to capital stock. Percentage change in the exogenous shift in the expected gross rate of return. Absolute change in time, measured in years. Percentage change in total wealth in equity invested by both the regional household and the trust invested in regional firm r. Percentage change in total wealth in equity invested by all regional households in the trust. Percentage change in total wealth in equity owned by regional household r and invested in both domestic firms and in the trust. Percentage change in wealth in equity owned by the regional household r and invested in domestic firms.
∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG
∀ r ∈ REG
∀ r ∈ REG ∀ r ∈ REG
∀ r ∈ REG ∀ r ∈ REG
Glossary of GDyn Notation wqht(r) wqt(r) wqtf (r) wtrustslack xwq_f (r)
xwqh(r)
yq_f (r)
yqh(r)
yqhf (r) yqht(r) yqt yqtf (r)
Percentage change in wealth in equity owned by the regional household r and invested in the trust. Percentage change in wealth in equity owned by the trust. Percentage change in wealth in equity owned by the trust and located in the regional firm r. Percentage change in the slack between wealth of trust. Percentage change in the shift variable for the total wealth in equity invested by both the domestic household and the trust in the regional firm r. Percentage change in the shift variable for the total wealth in equity owned by the regional household r from investments in both domestic firms and the trust. Percentage change in total income on equity paid to the domestic household and the trust by regional firms r. Percentage change in total income on equity paid to regional households r by both domestic firms and the trust. Percentage change in income on equity paid to regional households r by domestic firms. Percentage change in income on equity paid to regional household r by the trust. Percentage change in total income on equity paid to the trust by all regional firms. Percentage change in income on equity paid to the trust by regional firms r.
427 ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG
∀ r ∈ REG ∀ r ∈ REG
∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG
∀ r ∈ REG
Index
accession: China’s WTO, 7, 207–9, 225, 233, 326, 408–9; scenario, 208–9, 212, 229 accumulation: asset, 5; capital, 7, 13–14, 16, 18, 20, 26, 57–8, 68, 89, 161, 217, 220, 228, 263, 343, 407–9; equation, 14–20, 26–7, 36, 56–8, 68; wealth, 14–15, 18, 65, 70 actual investment schedule, 46 actual rate of return, 48–50, 57–8, 65, 68, 81–2, 87, 117, 128, 132, 156, 219, 413–4, 416 Adams, 291, 309 adaptive expectations: theory of investment, 13, 41; treatment, 65 additional shock (ashock), 52, 188 Africa, 5, 98–9, 102, 109, 111, 116, 123, 125–6, 130, 136, 142, 144–5, 147–8, 152, 226–7, 231, 244, 266, 269, 271, 281–2, 285–6, 299, 301–2, 305–7, 309, 339, 350–2, 355, 359–63, 365, 367, 369, 371, 385, 390, 395 aggregation procedure, 130–1, 133 aging: population, 364; process, 344, 372 Agreement: Free Trade (FTA), 5, 7, 136, 148, 245, 254, 261, 312, 326, 332; Multi-Fibre (MFA), 149, 150, 212, 233; North American Free Trade (NAFTA), 3; on Textiles and Clothing (ATC), 151, 207; Uruguay Round, 150, 247 agricultural productivity, 273, 280–1, 284–5, 288, 304, 409 agricultural research expenditure, 280–1 Agro-Ecological Zone (AEZ), 291, 295, 298–300, 304–5, 307–8 Ahammad, 291, 309 Allen partial elasticities (APE), 273, 276 allocation rule, 30
Alston, 269, 281, 289 alternative closure, 5 An Implicit Directly Additive Demand System (AIDADS), 292–3 AnalyseGE program, 191 Anderson, 157, 207, 232, 236, 267 Ando, 125, 380, 404 apparent current normal growth rate (KHAPP), 53–6 Armington, 315 assets: domestic, 27, 30, 92, 100, 116, 134, 245, 391, 409; financial, 5, 6, 13, 14, 20–6, 37, 101; foreign, 4, 8–9, 22–3, 29–30, 35, 69, 91, 95, 97, 99–102, 112, 116, 223, 225, 245, 379, 380–6, 399, 402, 403, 409–10; local, 22–3, 93–4, 96, 112 automatic accuracy, 186, 199, 201 autoregressive conditional heteroskedasticity (ARCH), 105–6 backcasting, 410 balance of trade/trade balance, 223–4, 246, 250–1, 254, 260, 301–3, 407 balance sheet identity, 23, 25 base rerun, 185, 190–1, 199, 200 base time, 15–16, 46, 51 basecase shock, 188 baseline: assumptions, 296; projection, 239, 281, 291, 301, 314, 343, 360, 362, 365, 366, 369, 372; scenario, 4, 136–7, 148–9, 151–2, 156, 162, 164, 208, 210, 211, 220, 247, 273, 282, 297, 300–1, 303, 325, 327–9, 332, 335, 344, 359, 367; simulation, 160, 238, 249–51, 270, 276, 291, 297–8, 300, 305, 327, 342, 357, 361, 365
429
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
430
Index
behavioral parameters, 128 Benjamin, 381, 404 Bernard, 79, 118, 279, 289, 313, 321–2, 340 Blalock, 340 Booth, 342, 366, 373 Brown-Kojima-Drysdale export intensity, 235 Brumberg, 380, 405 Bryant, 342–3, 373 Calderon, 97 calibration procedure, 71, 80, 87, 128–9, 279 capital: account volatility, 59; accumulation, 7, 13–4, 16, 18, 20, 26, 57–8, 68, 89, 161, 217, 220, 228, 263, 343, 407–9; fixed, 26, 38, 73, 76; flows, 69, 80, 223–4, 380, 386, 410; gain effect, 43; rental price, 42; services, 38; value of, 30, 38, 76, 127–8, 131, 160 capital mobility: degree of, 72, 80–2, 84–5, 117; international, 4, 7, 13, 21–3, 69–71, 101, 117, 208; perfect, 43, 56, 249 capital stock: 4, 13–14, 20, 27, 43, 56, 65, 74–7, 80–1, 114, 123, 125–8, 131, 137–8, 153, 166, 220–1, 228, 247, 251–2, 326, 332, 407, 415–16; rate of growth in, 81; series, 73, 76; value of, 127–8, 131 Center for Global Trade Analysis (CGTA), 6 Central Europe, 231–2, 266, 350–2, 355–6, 358, 361–3, 365, 367–9, 372 Centre of Policy Studies (CoPS), 233 China’s WTO Accession, 7, 207–9, 225, 233, 326, 408–9 Chuang, 314, 323, 340 Chung, 313, 340 closure: alternative, 5; file, 183, 186, 189–90, 199; GDyn, 189; policy, 185; standard, 152, 155; swap, 69, 166 Cobb-Douglas per capita utility function, 379 Coefficient of adjustment: 424; in estimated normal growth rate (LAMBKHAT), 53–4, 56, 72, 81–2, 89, 121, 128, 133–4, 397, 424; in expected rate of return (LAMBRORGE), 51–2, 56, 72, 81–5, 88, 121, 133–4, 195, 424; in normal rate of growth (LAMBYHAT), 392–4, 397; in rate of return (LAMBRORG), 44, 50–2, 56, 72, 81–5, 88–9, 121–2, 128–9, 133, 134, 195, 424; in wealth-to-income ratio (LAMBWYR), 388, 390, 393–8, 400, 402
commercial presence, 210, 215–6, 222, 228 commodity aggregation, 241, 272 Common Effective Preferential Tariff (CEPT) reduction program, 326 comparative advantage, 260, 324, 372 comparative static: GTAP model, 158–9; simulation, 158, 165–8, 199, 225 composite regions, 142 composition of wealth, 105, 109, 111–12, 114–15, 135 Constant Difference Elasticity (CDE) demand system, 356 constant elasticity: of substitution (CES), 86–8, 356; of transformation (CET), 271–3, 276, 295 consumer demand, 291–2, 299–300, 304 consumption analysis literature, 356 consumption expenditure, 353–6 consumption of services abroad, 215 consumption preferences, 344, 353, 356, 371–2 continuous time, 16 convergence: of rates of return, 73, 80; speed of, 4, 72, 81–4, 117 cost disadvantage ratio (CDR), 319 counterfactual policy simulations, 326 Cournot, 315, 319 Cranfield, 310 Cross border supply of services, 215, 228 cumulative differences, 162, 164, 193, 195, 200 cumulative results, 61–3, 191, 193 current time, 20, 46, 50 Darwin, 7, 269–70, 273, 276, 282, 289, 291, 310, 409 data: construction, 4, 6, 22; foreign income, 124; GTAP, 3, 120–1, 125, 177, 323–4; investment, 23, 75, 125–7, 211; population, 138; trade, 316; updated, 30, 167, 193, 197–9 Data Base: GDyn, 120–3, 125, 128, 130–1, 382, 413, 416–17, 421; GTAP 70, 118, 120, 122, 129–30, 135, 138, 149, 157, 201, 209, 212, 233, 237–8, 245, 268, 276, 289, 292–3, 310, 323, 340, 386, 395; updated policy, 165–6 Database: GATT/WTO Integrated (IDB), 149; updated, 30, 167, 197–8 Davis, 269, 289, 342 debt, 23, 26, 99
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
Index demand: consumer, 291–2, 299–300, 304; derived, 290, 293, 295, 297–9; equations, 246; final, 353; import, 244, 246 demand-side simulation, 297 demographic change, 4, 6–8, 343, 353, 368, 372 dependency ratio, 344, 352–3, 362–3, 367 depreciation, 24, 38, 71–2, 74–9, 82, 127, 160, 391, 395 derived coefficient, 182 Devarajan, 340 Dimaranan, 118, 120, 135–6, 157, 232–3, 268, 289, 310, 340 direct trade in services, 209, 215, 263 discount factor, 161, 168 disequilibrium approach, 69 domestic assets, 27, 30, 92, 100, 116, 134, 245, 391, 409 domestic capital stock, 97, 117, 135, 227 domestic equity, 30, 34, 91, 97, 252 domestic firms, 26–7, 92–3, 105, 207, 313, 319, 322, 337, 391, 422, 424–7 domestic wealth, 31, 97, 99 Drysdale, 236, 267 Durbin-Watson test, 105 duty drawback regime, 7, 208 dynamic baseline, 323 dynamic effects, 236, 245, 270, 407, 409 dynamic equations, 407, 418–9 dynamic FARM model (D-Farm), 270 dynamic simulation, 16, 43, 162, 166–8, 199–200 East Asia, 8, 98, 102, 106, 109, 111, 116, 129–30, 136, 142, 144–5, 147–8, 152, 225–8, 232–3, 244–5, 272, 281, 285, 288, 312, 314, 340, 343, 350–1, 355, 358, 360–3, 365, 367, 374, 385, 387, 395, 409, 413–15 Eastern Europe, 130, 211, 306 Eastern Europe and the Commonwealth of Independent States region (EIT), 299, 301–3, 305, 306 ecological-economic information, 280 econometric: analysis, 101, 117; investigation, 73, 118; standard errors, 336 Economies in Transition, 302, 305, 309 Edwards, 289 effective price, 240, 242–4, 246, 250, 262 efficiency losses, 322
431
elasticity: of the expected rate of return with respect to capital stocks (RORGFLEX), 45, 47–8, 52–4, 72, 85–90, 121, 129, 131–2, 195, 424; markup, 319, 320–1; of substitution, 86, 246–7, 295, 356; of transformation, 271, 295 Elbehri, 23, 70, 149, 157, 209, 233, 268, 276, 289, 419 Electronic Trade Document Exchange System (ETDS), 240, 242 empirical trade studies, 335 employment, 153, 343, 368 endowment: non-accumulable, 171; ownership of capital, 159 Engel, 105, 118, 293, 365, 371 entropy: parameters, 6, 117; theory, 30, 92 environmental: or ecological damages, 8, 288, 409; effects, 288, 409; issues, 409–10 equilibrium: long-run, 14, 59–60, 71–2, 78, 81, 117, 122; partial, 5, 245, 270, 300, 312, 320; steady growth, 392; steady-state, 407 equity: domestic, 34, 91, 97, 252; in firms, 23–4, 28; foreign, 28, 31, 33–4, 91, 97, 99, 252, 323; income, 37, 40, 181, 183; income earned on, 160; income from, 26, 41, 122, 181; negative, 30; portfolio, 31; price of, 25, 27, 37, 391, 426; shareholder, 23; total, 23, 26, 37, 39; wealth in, 26, 29, 424–7 equity-for-debt substitution, 70 equivalent variation (EV), 62–4, 160, 163–4, 168–71, 200, 225, 261–2, 377 errors in expectation, 60, 155–6, 414–6 Evenson, 269–70, 283, 288–9 exogenous productivity, 360 expectation time, 46, 48 expected investment schedule, 54–5, 60 expected rate of growth in the rate of return, 45, 47–8, 50, 84–5, 88–9 expected rate of return (rorge), 45–6, 49–53, 56–8, 65, 68–9, 72, 81, 84–7, 89, 117, 121, 132–4, 166, 193–5, 200, 413, 415–6, 426 expenditure: agricultural research, 280–1; consumption, 353–6; matrix of, 357; private, 280; public, 280; regional, 122, 424 export: subsidies, 187; tax equivalents, 149 extended linear expenditure system, 381, 404 Faruqee, 343, 373–4
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
432
Index
FDI-productivity effect, 327–8, 333–4 Feldstein, 70, 91, 118, 267, 399, 404, 411 final demand, 353 financial assets, 5–6, 13–14, 20–6, 37, 101 financial sector, 23, 101 Findlay, 215–6, 234 firms: domestic, 26–7, 92–3, 105, 207, 313, 319, 322, 337, 391, 422, 424–7; foreign, 23, 92–3, 116, 118, 206; liabilities of, 5; local, 23–4, 27, 31, 33, 37, 39, 92, 112, 116, 118, 121–2, 182–3, 314, 327; regional, 23, 26, 29, 92, 422, 425–7 fixed capital, 26, 38, 73, 76 foreign assets, 4, 8–9, 22–3, 29–30, 35, 69, 91, 95, 97, 99–102, 112, 116, 223, 225, 245, 379, 380–6, 399, 402–3, 409–10 Foreign Direct Investment (FDI), 23, 156, 205–7, 211, 215, 232, 252, 261, 313–14, 322–3, 327–36, 410 foreign equity, 28, 31, 33–4, 91, 97, 99, 252, 323 foreign income: data, 124; payments, 95, 123–5, 250, 254; receipts, 121–2, 124, 383 foreign investment, 7, 23, 91, 93, 112, 118, 156, 205–6, 208–11, 215–16, 221, 223, 225, 227–9, 235, 252, 261, 327, 330, 333, 360, 407, 409 foreign liabilities, 8, 9, 22, 30, 69, 99–102, 112, 382, 384–6 foreign ownership, 7, 29, 30, 35, 91, 123–4, 156, 160, 205, 209–11, 217, 221–2, 225–8, 251, 261, 323, 328, 330, 332–3, 407 foreign wealth, 250 Forest and Agricultural Sectors Model (FASOM), 291, 309 forward-looking behavior, 70 Francois, 149–50, 157, 211, 213–15, 232, 243–4, 267, 289, 312–13, 315, 319, 340 Free Trade Agreement (FTA), 5, 7–8, 136, 148, 235–40, 242–7, 249–54, 259–64, 268, 312–20, 323, 325–36, 339, 409–10 Freebairn, 270, 289 Frisvold, 270, 289 Future Agricultural Resources Model (FARM), 270, 273, 282, 284, 286–7, 289, 291 future time, 45–6, 133 Garbaccio, 232
GATT/WTO Integrated Database, 149 G-Cubed Model, 381 GDyn: closure, 189; Data Base, 120–3, 125, 128, 130–1, 382, 413, 416–17, 421 GDYNView, 197–9 GDYNVol, 197–9 gender, 8, 343–5, 348–9, 351–4 General Agreement on Tariffs and Trade (GATT), 149 General Agreement on Trade in Services (GATS), 206 General Equilibrium Modelling Package (GEMPack), 5, 10, 15, 62, 66–7, 70–1, 167, 173, 198, 201, 267, 289, 413, 419, 421–2 global climate change, 6 global investment, 36, 59, 60, 344, 360, 369, 372, 410 global rate of return, 220, 399 global savings, 69, 281, 380 global trust, 23–4, 26–31, 35, 37–40, 91–3, 112, 118, 121–2, 160, 182, 323, 361, 391 globalization, 8, 308 Golub, 8–9, 71, 120, 290–3, 295–7, 310, 379, 409–10 gravity model, 243–4 GREEN, 381, 404 greenhouse gas emissions, 304 Grilli, 365, 374 Gross Domestic Investment (GDI), 58, 144, 155, 279 Gross Domestic Product (GDP), 7, 67, 73–4, 81, 86–7, 122, 124–5, 127, 136, 138–42, 152–3, 155–6, 158, 164, 185, 187, 189, 192, 205, 216–20, 223–5, 229, 247–8, 251, 253–4, 260–1, 273, 279, 286, 296, 324, 326–30, 336, 343, 363–4, 368–70, 372, 382, 384–6, 399, 402–3, 410 gross investment, 47, 58, 85, 127, 247, 326, 413, 416–17 Gross National Product (GNP), 369, 371–2 gross rate of return, 42, 45–6, 53, 72, 121–2, 127, 424, 426 growth: in capital stocks, 81, 126; labor force, 142, 343–4, 363, 367; per capita income, 290, 297, 301; population, 8, 139–40, 187, 269–70, 280–2, 284, 286, 288, 296, 297, 299–301, 308, 342–3, 363, 399, 409; productivity, 216–17, 219, 279–81, 296, 298, 304–5, 321, 359–60, 363, 409; in the
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
Index growth (cont.): rate of return, 44–5, 47–8, 50, 54, 57, 81, 84–5, 88–9, 133; rates, 146, 155, 281; in real GDP, 153; real income, 8, 308; Total Factor Productivity (TFP), 7–8, 279–80, 282–3, 285, 288, 310, 409 GTAP: 4 Data Base, 70, 149, 157, 209, 212, 233, 237–8, 245, 268, 276, 289; 5 Data Base, 118, 157, 201, 292–3, 310, 323, 340, 386, 395; 6 Data Base, 120, 122, 129–30, 135, 138, 157; model, 3, 5, 13–14, 24, 69, 72, 87, 120, 158–9, 162, 173, 177, 207, 270, 353, 379, 406–7, 411, 421; parameters, 121 Hanslow, 158–9, 162, 171, 267, 374 Harrigan, 357, 374 Harris, 314 Harrison, 10, 70, 201, 207, 232, 267, 289, 413, 419 header array file (HAR), 71, 177, 180–1, 198 Hertel, 3, 7–8, 10, 13, 64, 70–1, 119, 137, 157–8, 162, 172, 193, 201, 205, 207–8, 213, 225, 232–3, 235, 267–8, 270, 273, 289–93, 295–7, 310–13, 316, 340–1, 353, 374, 379, 404, 406, 408–11, 419 Home bias effect, 92 Horioka, 69–70, 92, 118, 267, 380, 399, 404, 411 Horridge, 193, 201, 216, 233, 267, 413, 419 household: local, 112, 122; regional, 8, 21, 23–4, 26–9, 37, 39–40, 69, 72, 91–4, 105, 117–18, 121–2, 129, 134, 159, 181, 225, 327, 343–4, 372, 379, 387–8, 390–2, 395, 422, 424–7; wealth, 32–3, 35, 92, 94, 225, 382, 388, 391 Howe, 381, 404 Huang, 232 Huff, 70, 158, 162, 172, 225, 232, 353, 356, 374 Hummels, 240–2, 262, 268 Ianchovichina, 3, 7, 9, 13, 60–4, 70–1, 81, 88, 118, 120, 157–9, 172–3, 205, 207–9, 213, 227, 233, 269–70, 289, 313, 320–1, 341, 379, 404, 406, 408–9, 411 Impact Project/Center of Policy Studies, 407 imperfect competition, 314–15
433
import demand, 244, 246 income: earned on equity, 160; from equity, 26, 41, 122, 181; flows, 5–6, 14, 21, 37, 69, 91, 158–9, 365, 407; tax, 38, 205 income growth: per capita, 290, 297, 301; real, 8, 308 interest: premium, 360, 363; rate, 353–5, 360, 381 intermediate goods, 213, 241 international capital: flow, 69, 380, 386, 410; mobility, 4, 7, 13, 21–3, 69–71, 101, 117, 208; movement, 280 international demand system for commodities, 308, 409 international investment flows, 237, 245, 409 International Monetary Fund (IMF), 206, 211, 342, 374 international spillovers, 270 intra-industry trade, 239 investment: allocation of, 58, 360; data, 23, 75, 125–7, 211; flows, 7, 237, 245, 249, 408–9; foreign, 7, 23, 91, 93, 112, 118, 156, 205–6, 208–11, 215–16, 221, 223, 225, 227–9, 235, 252, 261, 327, 330, 333, 360, 407, 409; Foreign Direct (FDI), 23, 156, 205–7, 211, 215, 232, 252, 261, 313–14, 322–3, 327–36, 410; global, 36, 59, 60, 344, 369, 372, 410; gross, 47, 58, 85, 127, 247, 326, 413, 416–17; Gross Domestic (GDI), 58, 144, 155, 279; level of, 43–5, 47–9, 56–7, 65, 84–5, 363, 416, 418; model of, 41, 409; net, 19, 20, 166; value, 127, 419 investment schedule: actual, 46; expected, 54–5, 60 Itakura, 7–8, 120, 173, 235, 312–14, 340, 409–10 Japan-Singapore Free Trade Agreement, 235–6, 239–40, 247, 249, 251, 254, 260, 326, 339 Jensen, 313, 321–2, 340 Jones, 78–9, 118, 279, 289 Kapur, 31, 70 Kehoe, 313, 341 Kesavan, 31, 70 Kets, 296, 310 Kmenta, 109, 119 Kouparitsas, 405
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
434
Index
Kraay, 9–10, 92, 97–102, 104, 106, 109, 111, 115, 118–19, 123–4, 135, 382–5, 387–8, 390, 395, 405, 410–11 labor: force growth, 142, 343–4, 363, 367; productivity, 296, 360; projections, 138, 140, 142, 146–7; skilled, 137–8, 140, 144–7, 160, 273, 326; unskilled, 4, 137, 140, 146–7, 152–3, 162, 164, 187, 247, 273, 288, 296–7, 326 lagged adjustment: approach, 44; of capital stocks, 14; parameters, 71–2, 81, 117, 128 Lagrangean, 33 Lai, 357, 375 land-intensive: crops, 298; goods, 293, 303 Larson, 75–8, 117, 119 Latin America, 98, 102, 109–11, 116, 129, 140, 142, 144–5, 147–8, 152, 226–7, 244, 271, 279–82, 301, 309, 355, 385, 387, 395 Lejour, 233, 296, 310 Levin, 79, 119, 313, 341 Levinsohn, 313, 341 Lewis, 92, 119, 365, 374 liberalization of services, 210, 215, 243 life expectancy, 342–3, 345, 348–9, 352, 366 life-cycle hypothesis, 380 Lin, 70, 79, 118–19, 172, 312–15, 319, 321, 323, 325, 327, 329, 331, 333, 335, 337, 339, 340, 341, 375, 404 Lipsey, 365, 374 local assets, 22–3, 93–4, 96, 112 local capital, 30–2, 92, 112 local firms, 23–4, 27, 31, 33, 37, 39, 92, 112, 116, 118, 121–2, 182, 183, 314, 327 local household, 112, 122 logical file name, 178, 180–2 long-run equilibrium, 14, 59–60, 71–2, 78, 81, 117, 122 long-run stabilization, 399, 410 low-wage economies, 409 Ludena, 296, 310 macroeconomic projections, 136–8, 156, 157 Maddison, 233 Mai, 214, 217, 227, 229 marginal propensity to save, 381 markup elasticity, 319, 320–1 Markusen, 313, 315–16, 341
Martin, 126, 157, 207, 209, 212, 227, 232–3, 268, 289 Mattoo, 206 maximum likelihood estimation method, 106, 108 McDougall, 8–9, 13, 23, 60, 70–1, 81, 118, 120, 136, 149, 157–8, 209, 233, 268, 273–6, 289, 293, 296, 310, 323, 340, 342, 379, 404–5, 408, 410, 411, 413 McKibbin, 207–8, 342–3, 373, 405 Mendelsohn, 296, 310, 311 Mi, 291, 309 Middle East and North Africa, 98, 102, 109, 116, 244, 299, 301–2, 305, 309, 360, 385, 395 model: comparative static GTAP, 158–9; discrete choice, 241; dynamic FARM (D-Farm), 270; Future Agricultural Resources Model (FARM), 270, 273, 282, 284, 286–7, 289, 291; G-Cubed, 381; of investment, 41, 409; partial equilibrium, 270, 320; path-dependent, 64; recursive dynamic, 61, 161, 219; regression, 111 movement of natural persons, 215 Mrongowius, 216, 233 Muhleisen, 343, 374 Multi-Fibre Agreement (MFA), 149, 150, 212, 233 negative equity, 30 Nehru, 75, 76, 119 net investment, 19, 20, 166 net rate of return, 41, 46, 72, 404 Newly Industrialized Economies (NIEs), 152, 225 non-accumulable endowments, 171 nontariff barriers, 215, 235 nontariff trade costs, 239, 245 Norheim, 236, 267 normal growth rate: apparent current (KHAPP), 53–56; of the capital stock (KHAT), 46–9, 52–7, 59–60, 63, 81, 84, 121, 125–7, 131–2, 195, 424 normal rate of growth in income (YHAT), 391 North America, 3, 82, 130, 140, 144, 146–7, 151–3, 164, 167–8, 207, 213, 226–8, 235, 244, 249, 267, 295, 301, 304, 306, 309, 324, 326, 340, 350–2, 355–6, 358, 361–9, 371
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
Index North American Free Trade Agreement (NAFTA), 3, 312–13, 341 Norton, 269, 289 Numeraire shock, 166 oligopolistic competition, 316 omitted variables problem, 109, 111 Ordinary Least Squares (OLS), 79, 105–6 Organization for Economic Cooperation and Development (OECD), 74–80, 83, 117, 119, 226, 233, 310, 352, 374–5, 404 ownership: of capital endowment, 159; foreign, 7, 29, 30, 35, 91, 123–4, 156, 160, 205, 209–11, 217, 221–2, 225–8, 251, 261, 323, 328, 330, 332–3, 407 Pangestu, 233 parameters: behavioral, 128; entropy, 6, 117; GTAP, 121; lagged adjustment, 71–2, 81, 117, 128; rigidity, 32, 92–5, 97, 112–16, 118, 129–30 partial equilibrium, 5, 245, 270, 300, 312, 320; model, 270, 320; simulation, 300, 320 path dependent/dependency, 59, 64–7, 161–9, 185 Pearson, 5, 10, 66, 70, 173, 193, 201, 267, 289, 413, 419 perfect adjustment model of investment, 41 perfect capital mobility, 43, 56, 249 Perkins, 216, 233 Perpetual Inventory Method (PIM), 74–5 physical capital, 13, 21–4, 36, 223, 279, 343, 359–60 policy: closure, 185; projections, 4, 5, 137, 148–50, 273, 326; shock, 4, 13, 158, 164, 166, 169, 183, 185, 187, 188, 191, 407; simulation, 136, 160, 164–5, 185, 188–92, 197, 200, 210, 222, 279, 283, 325–6; variables, 200 pooling technique, 109–11 population: data, 138; growth, 8, 139–40, 187, 269–70, 280–2, 284, 286, 288, 296, 297, 299, 300–1, 308, 342–3, 363, 399, 409 Powell, 292, 310 price: of capital goods, 20, 26–7, 43, 87, 93, 166, 219, 220, 226, 414; effective, 240, 242–4, 246, 250, 262; of equity, 25, 27, 37, 391, 426; food, 365; food and resource, 269; resource, 269; of saving, 226, 392
435
private expenditure, 280 procompetitive effect, 313, 315, 316, 319–22, 327–9, 333–4, 336 product differentiation, 315–16, 357 productivity: agricultural productivity, 273, 280–1, 284–5, 288, 304, 409; gains, 208–10, 212, 216–17, 219, 221, 226–9; growth, 216–17, 219, 279–81, 296, 298, 304–5, 321, 359–60, 363, 409; linkages, 313, 327, 335; in services, 229, 408; shock, 216–17, 220, 226, 229 projections: baseline, 138, 146–7; labor, 138, 140, 142, 146–7; macroeconomic, 136–8, 156, 157; policy, 4, 5, 137, 148–50, 273, 326; skilled labor, 138, 146–7; unskilled labor, 147 propensity to save, 9, 17, 58–9, 69, 379, 381, 392–3, protection, 198, 208–9, 245, 248, 320, 324–5 public expenditure, 280 rate of growth in capital stocks, 81 rate of growth in the rate of return: expected, 45, 47–8, 50, 84–5, 88–9; required, 44, 48, 133 rate of return: actual (rorga), 48–50, 57–8, 65, 68, 81–2, 87, 117, 128, 132, 156, 219, 413–14, 416; expected (rorge), 45–6, 49–53, 56–8, 65, 68–9, 72, 81, 84–7, 89, 117, 121, 132–4, 166, 193–5, 200, 413, 415–16, 426; global, 220, 399; net, 41, 46, 72, 404; target, 44, 81–2, 128, 133, 156, 413–14, 416, 419, 426 realization time, 46–8, 51 recursive dynamic model, 61, 161, 219 recursive solution procedure, 69 regional firms, 23, 26, 29, 92, 422, 425–7 regional household, 8, 21, 23–4, 26–9, 37, 39–40, 69, 72, 91–4, 105, 117–18, 121–2, 129, 134, 159, 181, 225, 327, 343–4, 372, 379, 387–8, 390–2, 395, 422, 424–7 regional trade agreement, 235, 237, 245, 312 regional wealth, 8, 27, 65, 71, 91, 382, 394 regression model, 111 Reimer, 8, 293, 310, 312, 410 research and development (R&D), 340 resource price, 269 Rigidity: of allocation of wealth by regional household (RIGWQH), 72, 121; of source of funding of enterprises (RIGWQ F), 72, 121, 129
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
436
Index
rigidity parameters, 32, 92–5, 97, 112–16, 118, 129–30; recommendations for setting, 114 Rimmer, 211, 232, 292, 310, 373 Roberts, 313, 340 Rodrik, 315, 322, 340, 341 Roland, 315, 319, 340 Rose, 289, 310–11 rules-based economy, 206 RunDynam program, 173 secondary education, 145 sensitivity analysis, 276, 336 services: capital, 38; cross border supply of, 215, 228; direct trade in, 209, 215, 263; trade in, 7, 209–10, 215, 217, 219–20, 222, 244, 263; trade liberalization, 210, 244, 261 set information, 178–9 shareholder equity, 23 Shi, 8, 342, 344, 366, 374–5, 410–11 Shiells, 245, 267, 312–13, 340 Shock: additional (ashock), 52, 188; basecase, 188; numeraire, 166; policy, 4, 13, 158, 164, 166, 169, 183, 185, 187, 188, 191, 407; productivity, 216–17, 220, 226, 229; total (tshock), 66 Shoemaker, 7, 269, 409 Short run dynamics, 65 simulation: baseline, 160, 238, 249–1, 270, 276, 291, 297–8, 300, 305, 327, 342, 357, 361, 365; demand-side, 297; dynamic, 16, 43, 162, 166–8, 199–200; partial equilibrium, 300, 320; policy, 136, 160, 164–5, 185, 188–92, 197, 200, 210, 222, 279, 283, 325–6 skilled labor: 137–8, 140, 144–7, 160, 273, 326; projections, 138, 146–7 Sohngen, 291–2, 295–7, 310–11 Sokoloff, 314, 340 Solow-Swan, 343 solution: file, 193, 200; method, 15, 42, 184, 186; procedure, 17–19, 66, 69, 270–1 South Asia, 99, 102, 109, 111, 116, 130, 142, 144–5, 147–8, 152, 225–7, 231, 254, 261, 266, 293, 295–6, 299–304, 309, 339, 350–2, 355–6, 358, 360, 362–3, 365, 367, 382, 385–6, 400 speed of convergence, 4, 72, 81–4, 117 Spinanger, 213–15, 228, 232–3, 249, 267 standard closure, 152, 155
steady growth equilibrium, 392 steady-state equilibrium, 407 stock-flow dynamics, 69 structural adjustment, 161, 166 Strutt, 149–50, 157, 173, 232, 267, 289 Tang, 207–8, 233 target rate of return, 44, 81–2, 128, 133, 156, 413–14, 416, 419, 426 tariff: and quotas, 152, 210, 217, 228; rate, 149–51, 212–13, 226, 273, 324; revenue, 149–50, 209 tax: income, 38, 205; revenue replacement, 166 technological progress in agriculture, 282 terms of trade, 97, 159, 162, 170–1, 225, 227, 261, 263, 326–7 tertiary education, 145 TFP growth, 7–8, 279–80, 282–3, 285, 288, 310, 409 time: base, 15, 16, 46, 51; continuous, 16; current, 20, 46, 50; dependent variables, 166; expectation, 46, 48; future, 45–6, 133; realization, 46–8, 51; treatment of, 5, 14–15; as a variable (TIME), 20, 46 total equity, 23, 26, 37, 39 total factor productivity (TFP), 7, 183, 214, 216–17, 270 total shock (tshock), 66 trade: balance, 223–4, 246, 250–1, 254, 260, 301–3, 407; data, 316; in services, 7, 209–10, 215, 217, 219–20, 222, 244, 263; and transport, 226, 244; volume effects, 254 traded commodities, 180 transport margins, 323 treatment: of saving, 58, 69, 379; of time, 5, 14–15 Trefler, 357 Truong, 23, 70, 149, 157, 209, 268, 276, 289 Tyers, 8, 342, 344–5, 348, 351–3, 356, 360–2, 366, 368, 373–5, 381, 399, 405, 410–11 UNCTAD, 239 United Nations (UN), 239, 342, 350–2, 362–3, 375 unskilled labor, 4, 137, 140, 146–7, 152–3, 162, 164, 187, 247, 273, 288, 296–7, 326; projections, 147 updated data, 30, 167, 193, 197–9 updated policy data, 165–6
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923
Index Uruguay Round (UR), 3, 98, 107, 113, 148–50, 157, 213, 232, 238–9, 247–8, 267–8, 273, 289, 326, 339; agreement, 150, 247 utility function, 353, 379, 381 validation exercise, 410 Venables, 315–16, 341 Ventura, 10, 97, 118–19, 135, 405, 411 ViewHAR, 179, 191, 195, 198 Walmsley, 7, 9–10, 71, 120, 136–7, 157–8, 173, 205, 207–8, 213, 233, 235, 313, 340, 342, 379, 406, 408–10, 413, 421 Walton, 72–3, 119
437
Wang, 207, 209, 233 wealth: accumulation, 14–15, 18, 65, 70; in equity, 26, 29, 424–7 wealth-to-income ratio, 386–99, 401, 410 Weber, 357, 359, 373, 375 welfare: analysis, 6, 22, 225; decomposition, 158–9, 161–2, 164–5, 168, 173, 199, 200 Wenping, 234 Western Europe, 144, 146, 151–3, 226–7, 244, 254, 260, 301–3, 305–6, 309, 350–1, 355, 358, 361–5, 367–9, 371 Wilcoxen, 381, 405 Yang, 207, 234, 365, 374
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923