Catching the Flu from the United States: Synchronisation and Transmission Mechanisms to the Euro Area

Catching the Flu from the United States

9780230_243231_01_prexvi.indd i

6/18/2010 7:06:02 PM

Also by Filippo di Mauro External Dimension and the Euro Area (Co-edited) Globalisation and Regionalism (Co-edited)

9780230_243231_01_prexvi.indd ii

6/18/2010 7:06:02 PM

Catching the Flu from the United States Synchronisation and Transmission Mechanisms to the Euro Area

Filippo di Mauro Stephane Dees and

Marco J. Lombardi

9780230_243231_01_prexvi.indd iii

6/18/2010 7:06:02 PM

© Filippo di Mauro, Stephane Dees and Marco J. Lombardi 2010 All rights reserved. No reproduction, copy or transmission of this publication may be made without written permission. No portion of this publication may be reproduced, copied or transmitted save with written permission or in accordance with the provisions of the Copyright, Designs and Patents Act 1988, or under the terms of any licence permitting limited copying issued by the Copyright Licensing Agency, Saffron House, 6-10 Kirby Street, London EC1N 8TS. Any person who does any unauthorized act in relation to this publication may be liable to criminal prosecution and civil claims for damages. The authors have asserted their rights to be identified as the authors of this work in accordance with the Copyright, Designs and Patents Act 1988. First published 2010 by PALGRAVE MACMILLAN Palgrave Macmillan in the UK is an imprint of Macmillan Publishers Limited, registered in England, company number 785998, of Houndmills, Basingstoke, Hampshire RG21 6XS. Palgrave Macmillan in the US is a division of St Martin’s Press LLC, 175 Fifth Avenue, New York, NY 10010. Palgrave Macmillan is the global academic imprint of the above companies and has companies and representatives throughout the world. Palgrave® and Macmillan® are registered trademarks in the United States, the United Kingdom, Europe and other countries. ISBN: 978–0–230–24323–1 hardback This book is printed on paper suitable for recycling and made from fully managed and sustained forest sources. Logging, pulping and manufacturing processes are expected to conform to the environmental regulations of the country of origin. A catalogue record for this book is available from the British Library. A catalog record for this book is available from the Library of Congress. 10 9 8 7 6 5 4 3 2 1 19 18 17 16 15 14 13 12 11 10 Printed and bound in Great Britain by CPI Antony Rowe, Chippenham and Eastbourne

9780230_243231_01_prexvi.indd iv

6/18/2010 7:06:03 PM

Contents List of Figures

vi

List of Tables

x

List of Boxes

xi

Acknowledgements

xii

Foreword

xiii

About the Authors

xv

1 Introduction

1

2 Business Cycle Synchronisation: Disentangling Global Trade and Financial Linkages

18

3 Business Cycle Synchronisation: The United States and the Euro Area

43

4 The United States and the Euro Area: What Do Structural Models Say About the Linkages?

61

5 The United States and the Euro Area: The Role of Financial Variables

97

6 Economic Interactions US-Euro Area Over the 2007–9 Financial Crisis: What Did We Learn?

144

7 The US-Euro Area Relationship in a Context of Possible Systemic Changes

189

8 Conclusion

216

Index

219

v

9780230_243231_01_prexvi.indd v

6/18/2010 7:06:03 PM

Figures 1.1 1.2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 4.1 4.2 4.3 4.4 4.5

US real GDP growth rates (detrended) and the common component of rest of the world growth rates (detrended) The effects of a US domestic demand shock (by 1 pp of GDP) on GDP of other countries/regions World GDP and world imports Evolution of the trade openness of the euro area, the United States and Japan Euro area trade volumes and real GDP Euro area net direct and portfolio investment flows Bank’s external claims in major countries Bilateral GDP correlation with the US and its determinants for selected countries Estimation framework Summary of estimation results Overall effects on business cycle correlation and breakdown by direct and indirect effects Smoothed GDP per capita growth rates Highly synchronised recessions Output gaps Average statistics for recessions Average statistics for recoveries Real GDP growth in the US and euro area Gap between US and euro area in per capita GDP Time-varying estimates of impact elasticity of real GDP to US and foreign real GDP Impulse response functions of a 1% shock to US real GDP growth on US and euro area real GDP growth Decomposition of euro area real GDP growth, 1999–2006 Impulse response of a positive unit (one standard error) global demand shock Impulse response of a positive unit (one standard error) global supply shock Impulse response of a positive unit (one standard error) US monetary policy shock Forecast error variance decomposition of global shocks in a Multi-Country New-Keynesian model

2 5 19 20 21 22 23 24 27 29 32 44 45 45 46 46 48 48 54 55 67 75 76 78 79

vi

9780230_243231_01_prexvi.indd vi

6/18/2010 7:06:03 PM

Figures

5.1 5.2 5.3 5.4a 5.4b 5.4c 5.4d 5.4e 5.4f 5.5 5.6 5.7 5.8 5.9a

5.9b 5.10 5.11 5.12 5.13 5.14 5.15 5.16 5.17 5.18 5.19 6.1 6.2

External assets and liabilities of euro area resident banks vis-à-vis the US US and euro area stock market indices US and euro area corporate bond yields MFI interest rates on new loans to non-financial corporations in the euro area and market interest rates MFI interest rates on new loans to households in the euro area and market interest rates Commercial and industrial loans interest rates in the United States and market interest rates Mortgage rates in the United States and market interest rates MFI interest rate spreads in the euro area Mortgage and loan interest rate spreads in the United States Structural shocks Historical decomposition of US GDP growth contributions Net financial spillovers Net real spillovers Ten-year government bond spreads of major euro area countries over German bonds (black) and US bond spreads of individual states over US treasury bonds (grey) (basis points): Levels of 29 June 2007 Median of US state spreads and euro area government bond spreads Rolling RMSE for different models GW test for conditional predictive ability Threshold variables US GDP: Impulse responses 3-year ahead Euro area GDP: Impulse responses 3-year ahead Transition probabilities US: Financial variables and recession episodes US: Recession probability using alternative indicators (in sample probit model (1957–2008)) US: Recession probability using alternative indicators (out-of-sample probit model (1957–2008)) Out-of-sample recession probabilities – comparison ProbVar/probit during the 2001 recession episode Real GDP per capita Unemployment rate

9780230_243231_01_prexvi.indd vii

vii

98 99 99 101 101 102 103 104 104 107 108 110 111

114 115 123 125 130 131 131 133 134 136 137 138 146 147

6/18/2010 7:06:03 PM

viii

6.3 6.4 6.5 6.6 6.7 6.8 6.9 6.10 6.11 6.12 6.13 6.14 6.15 6.16 6.17 6.18 6.19

6.20 6.21 6.22 6.23 6.24 6.25 6.26 6.27 6.28 6.29 6.30 6.31 6.32

Figures

GDP components – euro area GDP components – United States World trade, global activity and euro area exports Extra-euro area exports of goods Global PMI manufacturing Global PMI stocks of purchases and euro area industrial confidence Total exports and imports (intra plus extra; goods plus services) Extra-exports and intra-trade of goods Contributions to growth in extra-euro area export values of goods Contributions to growth in extra-euro area export values of goods Composition of world imports and euro area foreign demand Composition of EA and US exports to China US GDP forecasts made in 2008Q2 and 2008Q3 Forecasts of US and euro area GDPs as of 2009Q2 Forecasts for US and euro area GDPs as of 2009Q3 Forecasts of US and euro area GDPs as of 2009Q3 US GDP forecasts: Comparing results based on the stock market index and its volatility to those based on the stock market index and the corporate bond spread, as of 2009Q2 and 2009Q3 Industrial production index Cumulative deterioration in PMI between August 2007 and March 2009 Breakdown of value added in 2005 International openness Index of vertical integration Import elasticities of expenditure components Equity indices Changes in total external claims by instruments Consumer confidence Fiscal policy Monetary policy Evolution of housing investment share of economic activity in the euro area and US ‘Abnormal’ GDP growth contributions of residential investment

9780230_243231_01_prexvi.indd viii

148 148 149 150 151 152 153 154 155 156 157 157 159 160 161 163

164 166 167 167 168 169 171 172 173 174 174 175 176 177

6/18/2010 7:06:03 PM

Figures

6.33 6.34 6.35 6.36 6.37 7.1 7.2 7.3 7.4 7.5 7.6 7.7 7.8 7.9 7.10 7.11 7.12 7.13 7.14 7.15

‘Abnormal’ contributions to cyclical turning ‘Abnormal’ contributions to cyclical turning points in euro area countries Real GDP growth rates in the main euro area countries Cross-border holdings as a share of total holdings of short-term debt securities issued in the euro area Real effective exchange rates US Household net worth Correlation: Household savings and wealth Correlation: Household savings and wealth US personal saving rate Developments in world trade volumes during US recessions Imports and consumption Global trade with contributions by various countries House prices Residential investment Household saving Household indebtedness Comparison of potential output estimates for the United States Impact of the crisis on the level of US actual and potential GDP Simulated impact of a 4 pp gradual rise in the US saving rate Simulated impact of a 24% gradual rise in China’s domestic demand over 10 years

9780230_243231_01_prexvi.indd ix

ix

178 179 180 183 184 192 193 194 195 196 197 198 199 200 201 201 204 205 212 213

6/18/2010 7:06:03 PM

Tables 1.1

2.1 2.2 2.3 2.4 4.1 4.2 4.3 4.4 5.1 5.2 5.3 6.1 6.2 6.3 6.4

Variance decomposition of GDP growth based on a Factor-Structural VAR model: Common shocks, own-country shocks and spillovers Estimation results – direct effects Estimation results – indirect effects (whole sample) Estimation results (OECD sample) Channels to business cycle synchronisation Forecast-error-variance decomposition in NAWM Decomposition of conditional variances in Adjemian et al. (2008) Descriptive statistics for restricted coefficients Correlations among the estimated shocks Unconditional out-of-sample RMSE for the euro area GDP Diebold-Mariano tests for unconditional predictive ability at selected horizons Out-of-sample RMSE for the euro area, conditional on quarterly information Comparison of developments in GDP in the euro area and US Cumulated deviation from average quarterly growth Import intensities of expenditure components Annual growth rates of real GDP per-head

7 30 33 34 40 66 68 73 80 119 120 122 145 146 170 181

x

9780230_243231_01_prexvi.indd x

6/18/2010 7:06:03 PM

Boxes 5.1 5.2

5.3 6.1 7.1

Developments in euro area and US bank interest rates during the 2007–9 crisis Euro area sovereign bond spreads and intra-US state bond spreads during the 2007–9 crisis – are their developments comparable? Threshold VAR models A comparison of the forecasting performance of the stock market volatility vs the corporate bond spread Comparison of the external financing of households and non-financial corporations in the euro area and the United States and impacts of the 2007–9 crisis

100

113 127 164

206

xi

9780230_243231_01_prexvi.indd xi

6/18/2010 7:06:03 PM

Acknowledgements This work is the outcome of several months of analytical work within the External Developments Division of the ECB and puts together research carried out in collaboration with several staff members. We also thank Warwick McKibbin for his thoughtful comments. Concerning the general issue of transatlantic interactions and transmission channels, we had several illuminating discussions with Domenico Giannone, Michele Lenza and Lucrezia Reichlin. The chapter on the stylised facts is indeed heavily based on their research results, but also includes valuable contributions by Arthur Saint-Guilhem and Isabel Vansteenkiste. The chapter on the decomposition of transmission channels is based mainly on joint work with Nico Zorell, which we thank for his skilful contribution. The chapter on structural models is part of the research we have undertaken with Hashem Pesaran, Ron Smith and Vanessa Smith on GVAR models. The valuable contribution by Raphael Espinoza is also acknowledged. Concerning the chapter on financial variables, most of the research presented is co-authored by Fabio Fornari, and the part of the chapter covering probit models is indeed based on his research with Wolfgang Lemke. Alessandro Galesi also helped us, not only with his skilful research assistantship, but also with his econometric and programming expertise. We are also thankful to Ursel Baumann, Paul Hiebert, Luca Gattini and David Lodge for the study of the economic interactions of the US-euro area over the 2007–9 financial crisis. We are indebted to Tadios Tewolde, Stephanie Brown and Julia Fritz for having dealt with the technical and administrative details. We are finally thankful to Hans-Joachim Klöckers and Jürgen Stark for having supported and improved our work with their comments and guidance.

xii

9780230_243231_01_prexvi.indd xii

6/18/2010 7:06:03 PM

Foreword The recent financial crisis represented a formidable challenge for the euro area, as it exposed its economy to a massive externally originating shock. The euro area confirmed its role as an oasis of stability, but recent events have also demonstrated how external shocks can quickly propagate in the globalised world. The euro area is relatively closed compared with its constituent countries, yet the extent of the impact of externally originated shocks on its economy remains large, as recent developments have testified. The crisis also demonstrated the relevance of domestic developments in the United States for the global economy. This book explores from an empirical perspective the economic linkages between the United States and the euro area, with the aim of providing an understanding on how and to what extent economic developments in the United States transmit to the euro area. The issue is examined using different methodologies, and the various chapters of the book bring together up-to-date research results and stylised facts on the various channels of shock transmission, with a special emphasis on financial and trade linkages. As such, this work is very ambitious, as it tackles a number of complex issues ranging from trade and financial linkages to the identification of the shocks hitting the two economies and the interaction of the economies of the United States and the euro area during the recent financial crisis. More specifically, special attention has been devoted to the role of financial variables in explaining the interlinkages, which is in our view a topic of particular relevance in the aftermath of a financial crisis that has turned into one of the worst global economic downturns. In this respect, the book also devotes two chapters to the financial crisis, the global economic downturn and prospects for recovery, again examining the issue from the perspective of transatlantic linkages. The originality of the book lies in the use of a variety of empirical techniques in order to quantify the main linkages – trade, financial and expectations. Similarly, and perhaps more importantly, the book attempts at identifying the shocks that have historically been experienced in the global economy. This is critical in understanding the transmission mechanism of shocks between economies. The focus of the

xiii

9780230_243231_01_prexvi.indd xiii

6/18/2010 7:06:03 PM

xiv Foreword

book on the linkage between the US and the euro region contrasts with other research, which tends to focus on linkages between the US and a single economy or the G7 economies. Some of the results are worth emphasising: despite the emergence of new players in the global economic landscape, the United States still seems to play a prominent role in leading global economic developments, both in the positive and in the negative direction. More specifically, the transmission mechanism of shocks across the Atlantic has been particularly stable over time: empirical analysis reveals that the euro area business cycle seems to have lagged that of the United States by a few quarters. Hence, downturns in the US economy tend to be followed by a contraction of activity in the euro area; the euro area in turn benefits from a recovery of economic activity in the United States. Whether this is because the United States are themselves a source of shocks, or rather react more quickly than other economies to shocks of a global nature is still a subject of debate. The recent downturn confirmed the close ties between the two economies; such ties may also have been strengthened by the globalisation process, with trade and financial interlinkages having grown in importance with the globalisation process. Another key finding of the book concerning the role of financial variables is that they cannot systematically help in predicting real developments, but at times, when the economy is hit by shocks of financial nature, their role becomes important. Overall, the results presented in the book are very important for the conduct of monetary policy at the ECB, to the extent that developments in the United States affect the broad set of economic and financial indicators which are analysed under the economic analysis pillar of the ECB’s monetary policy strategy: as policy-makers, the ECB needs to be aware of the mechanism that governs the transmission of shocks hitting the US economy to the euro area. Against the background of the lack of global governance authorities, and the persistently relevant role of the United States for the global economy, the stability and growth of the US economy is a matter of global interest, and US policy-makers bear a great responsibility not only with respect to their fellow citizens, but also for the rest of the world.

9780230_243231_01_prexvi.indd xiv

6/18/2010 7:06:03 PM

About the Authors Filippo di Mauro has been Head of the External Developments Division of the European Central Bank since 1998. His division is in charge of international forecasting, medium-term FX analysis and the balance of payments of the euro area. Filippo started his career in 1984 at the Research Department of the Central Bank of Italy, following a short period consulting at the OECD. Prior to joining the ECB, he also held various economist positions at the IMF and the Asian Development Bank. He has widely published in the areas of international linkages, trade and competitiveness with a special focus on the euro area. Stephane Dees has been Principal Economist in the External Developments Division of the European Central Bank since 2001. He earned his Ph.D. in 1999 from the University of Bordeaux. He has previously worked for the National Institute for Economic and Social Research (NIESR) and the Centre d’Etudes Prospectives et d’Informations Internationales (CEPII). He has published articles in several academic journals such as the Journal of Money, Credit, and Banking, Journal of Economic Dynamics and Control, Journal of Applied Econometrics, Energy Journal, Open Economies Review and Journal of Policy Modeling. Marco Jacopo Lombardi works as an Economist in the External Developments Division of the European Central Bank since 2008. His tasks include monitoring and analysis of commodity markets and FX. Prior to joining the ECB, he had been Max Weber Fellow at the European University Institute and an Assistant Professor at the University of Pisa. Marco holds a Ph.D. in Applied Statistics and has published several articles in academic journals, including Journal of Applied Econometrics, Econometrics Journal and IEEE Transactions on Signal Processing.

xv

9780230_243231_01_prexvi.indd xv

6/18/2010 7:06:03 PM

9780230_243231_01_prexvi.indd xvi

6/18/2010 7:06:03 PM

1 Introduction

The fact that the United States are considered, de facto, as the ‘engine’ of the world economy has been a puzzle for at least a full generation of economists. How can it be that an economy that represents only less than a fourth of world GDP could still have such a role, particularly when large and fast-growing economic players, like China, are potently emerging in the global economy? The empirical regularity, however, is unequivocally strong, as all past US recessions have typically coincided with significant reductions in global growth (IMF, 2007). In this context, the current recession is no exception if one excludes some shortlived ‘decoupling’ at an earlier stage of the crisis. In looking for reasons for such a leading role, it is obvious that size cannot be the whole story. That engine role has rather to do with the whole set of interactions, which – departing from the US – tend to reverberate, amplified, to the rest of the world. Such interactions include all sort of transmission channels related to trade links, financial interconnections in globally interlinked markets, as well as other more ‘mood’ related variables such as confidence indicators of consumers and business operators. Understanding how shocks emanating in the US are transmitted abroad is the core of this book. The focus is on the euro area economy. First, because we want to fill the gap in an over concentration of the literature on the synchronisation of business cycles across G7 economies; second, because the current crisis appears to have burdened the euro area economy to an extent which seems inconsistent with its pre-crisis initial conditions relatively to the US. However, identifying the nature of the shock remains complex. In the present case, developments in the US may have acted as a trigger for vulnerabilities, which were common across countries and regions, including high levels of leverage and an under-pricing of risk. In this vein, the shock was actually global in 1

9780230_243231_02_cha01.indd 1

5/7/2010 1:06:29 PM

2

Catching the Flu from the United States

nature, with somewhat more negative implications for euro area activity than if the shock had been truly US-specific. In the following, we present some basic facts about the role of the United States in the world economy (Section 1.1), before quantifying the impact of a US shock on the rest of the world (Section 1.2). In Section 1.3, we attempt to define the nature of the shock by disentangling the average contributions to business cycles of common and idiosyncratic shocks as well as their spillovers. The subsequent chapters of the book are thereafter put into context. Section 1.4 reviews the various channels of international transmission of shocks and Section 1.5 focuses on the transatlantic linkages. An assessment of the 2007–9 financial crisis is proposed in Section 1.6, together with some lessons to be drawn.

1.1

The United States and the world economy

A quick glance at the data seems to provide support to the fact that the US economy acts as the ‘engine’ of the world economy. Figure 1.1 shows the relatively strong correlation of US real GDP growth and a common component derived from non-US economies’ growth rates. In addition

3.0 2.0 1.0 0.0 −1.0 −2.0 −3.0 −4.0 −5.0 1979

1982

1985

1988 US

1991

1994

1997

2000

2003

2006

Common components (Non-US)

Figure 1.1 US real GDP growth rates (detrended) and the common component of rest of the world growth rates (detrended) Note: Common component is the first principal component derived from a pool of 25 non-US economies. Last observation refers to 2006Q4. Source: Authors’ computation.

9780230_243231_02_cha01.indd 2

5/7/2010 1:06:29 PM

Introduction 3

to a large correlation (42%), it seems that in some periods, the US cycle tends to lead that of the rest of the world. Indeed, the correlation between US growth rates and the common component when lagged by one period increases to 50%. Only a number of periods can be identified when correlation was very low, implying that – in a sense – the US cycle is a good proxy for a global cycle. International business cycle linkages have been the subject of a large number of papers using a wide range of techniques going from calculating simple cross correlations to unobserved components and dynamic factor models. This literature aimed at uncovering the characteristics, as well as the degree of synchronisation, of economic activity fluctuations in industrialised countries. The earlier work – based on cross correlations of real activity growth rates – focused on analysing the degree of integration between industrial economies. In general, these studies found that although high, the correlation of growth rates among G7 economies has not increased over time, despite the increased integration of the industrial economies through more trade in goods and services and more global financial markets (see Doyle and Faust, 2002). This mainly results from the fact that these economies have already been integrated for a long time. Therefore, any further increase in economic and financial integration does not lead necessarily to a higher degree of synchronisation. However, when including emerging economies in the sample, our own work points to higher synchronisation of business cycles across countries (see Chapter 2). The process of globalisation has led to a sharp increase in the economic interlinkages across countries. Whereas the main fallout of this process has surely been positive for economic growth, such an increased interdependence of different economies has also implied that each economy has potentially become more vulnerable to external shocks than it was in the past. The 2007–9 economic downturn is paradigmatic in this respect: starting with a financial market turmoil limited to the United States, the crisis quickly extended to other financial markets and to the world real economy as well. Not only the euro area, but also emerging economies, which had recently been believed to be less prone to external shocks, have been heavily affected. In this respect, the occasionally mentioned ‘decoupling’ of the euro area and emerging economies from the US has been disproved by the most recent events, at least in its initial simplistic definition. On the other hand, identifying international linkages has become an even higher priority in order to ascertain, at the present juncture, the most likely recovery path across regions after the global economic slowdown.

9780230_243231_02_cha01.indd 3

5/7/2010 1:06:30 PM

4

Catching the Flu from the United States

The interaction mechanisms are, however, very complex and have manifested themselves in very different forms in recent history. Comprehending fully the nature and extent of such interactions is critical in order to gauge the impacts of the present global slowdown episode and it forms the main objective of this book. The focus is, however, mainly centred on the US economy – as the economy at the root of most of the global cycles in the last decades – and the euro area – as the second largest economic area in the world.

1.2 Measuring the transmission of US shocks to the rest of the world The external exposure of a country to its international environment is traditionally measured by the openness ratio and the geographical structure of its trade. For instance, given the trade distribution of euro area exports, the short-term impact on the euro area economy of demand shocks originating from the UK is likely to be larger than those from the US. Moreover, for countries that are relatively closed, the trade channel is expected to play a smaller role in the transmission of shocks, making the economy relatively immune to developments abroad. However, such measures of exposure may be misleading as they reflect only the direct effects of trade transmission. Beyond this, additional factors affect economies in an indirect manner. For example, positive demand shocks in the US will of course stimulate euro area economic activity through the bilateral trade channel. But higher import demand from the US also benefits the exports of other countries, which thereafter are expected to increase their imports and in particular their imports from the euro area, a so called ‘echo effect’. Dees and Vansteenkiste (2007) show that once the ‘echo effect’ is accounted for trade spillovers between a country and its international environment are likely to be amplified (see Figure 1.2). In the case of the euro area, for instance, the ‘echo effect’ is twice as large as the direct trade effect. The strong co-movements observed across countries’ real outputs are difficult to explain in terms of trade linkages alone. Indeed, while trade integration and trade flows have increased further globally, the impact of trade developments on regions’ GDP growth remains too limited to explain the correlation between US growth and the rest of the world’s GDP. New channels, such as financial or confidence channels, play an important role. In order to account for channels additional to trade, such as financial linkages, price adjustments and economic policy reactions,

9780230_243231_02_cha01.indd 4

5/7/2010 1:06:30 PM

Introduction 5 1.2 1 0.8 0.6 0.4 0.2 0 Euro area

Japan

Latin America

Direct trade effect

ODE

Emerging Rest of Asia Europe

Trade effect incl. echo effect

UK

US

GVAR

Figure 1.2 The effects of a US domestic demand shock (by 1 pp of GDP) on GDP of other countries/regions Source: Dees and Vansteenkiste (2007).

Dees and Vansteenkiste (2007) use a global Vector Autoregression (VAR) model (GVAR). The GVAR approach developed by Pesaran et al. (2004) consists of specifying and estimating a set of country-specific vector error-correcting models that are consistently combined to generate a global model that can be simultaneously solved for all the variables in the world economy. This approach addresses the problem of consistently modelling interdependencies among many economies through the construction of ‘foreign’ variables, which are included in each individual country model. Thus, each country model includes domestic variables plus variables obtained from the aggregation of data on the foreign economies using weights derived from trade statistics. Using the GVAR model estimated in Dees et al. (2007),1 Dees and Vansteenkiste (2007) show that, for most countries, output is seen to be more sensitive to a shock originating in the US when all channels are considered in addition to the purely trade-related ones. The point estimates are in most cases higher than the benchmark values for trade-related effects (Figure 1.2). For the euro area and Latin America, the effect of a US domestic demand shock is around 2.5 times that based on direct trade effects. It is slightly less (around 1.5) in the case of Japan and Other Developed Economies and it is above 5 for the rest of Europe. Interestingly, the GVAR-implied elasticity of the UK and emerging Asia is very close to the direct trade effect and below the total trade effect. For emerging Asia, this result might be due to different factors. First, the

9780230_243231_02_cha01.indd 5

5/7/2010 1:06:30 PM

6

Catching the Flu from the United States

Chinese economy, which represents almost half of the region’s output, has remained relatively immune to world economic developments and is increasingly providing a strong impetus to the region as a whole. Second, since the sample used to estimate the GVAR model includes the Asian crisis, it is likely that the results are influenced by this episode, which did not originate from a US shock. Finally, a large share of emerging Asian trade is related to processing trade, whose contribution to overall GDP is lower than traditional trade (when export demand for final Asian products decreases, Asian imports of the corresponding intermediate goods also decrease, leading to a broadly neutral impact on the net trade contribution to growth). Overall, Dees and Vansteenkiste (2007) find that a 1 percentage point (pp) positive shock in the United States would result in an increase in the GDP of the other regions in the world via the trade channel of between 0.1 to 0.5 percentage points, depending on the region considered. Including other channels this range would make this figure much higher, namely between 0.2 and 0.7 pp.

1.3 Co-movements in real GDP: Global shocks vs spillovers As shown in Figure 1.1, there are strong co-movements between the US and a measure of rest of the world activity. However, these co-movements do not necessarily reflect a pure transmission of US idiosyncratic shocks and might also represent impacts of shocks that are more global in nature. In order to understand the factors that could explain such co-movements, Dees and Vansteenkiste (2007) estimate a FactorStructural VAR model for the main countries and regions, following Stock and Watson (2005). They decompose the variance of the fourquarter ahead forecast error for GDP growth, by distinguishing three potential sources: unforeseen common shocks, unforeseen domestic shocks and spillover effects of unforeseen domestic shocks to other countries. They also consider developments during three different subperiods: 1979–88 (Table 1.1 – Panel A), 1989–98 (Table 1.1 – Panel B) and 1999–2006 (Table 1.1 – Panel C). Statistical tests confirm that there has been a structural break across the three sub-periods chosen. In this context, common shocks are defined as those that affect all countries within the same period. Country-specific shocks can lead to spillovers, but those spillovers are assumed to happen with at least a one-quarter lag. A potential disadvantage of this approach is that if an international shock affects several countries only with a lag, that effect may incorrectly be interpreted as a spillover.

9780230_243231_02_cha01.indd 6

5/7/2010 1:06:30 PM

Introduction 7

Table 1.1 Variance decomposition of GDP growth based on a Factor-Structural VAR model: Common shocks, own-country shocks and spillovers Fraction due to Country/ region

Error standard deviation

Panel A US Euro area UK Other Dev. Eco.

Spillovers

0.55 0.54 0.35 0.57

0.29 0.29 0.58 0.34

0.15 0.16 0.07 0.08

0.39 0.58 0.21 0.06

0.07 0.07 0.09 0.11

0.54 0.12 0.45 0.43

0.08 0.08 0.09 0.08

Period 1989–98 1.28 1.17 1.2 1.57

Panel C US Euro area UK Other Dev. Eco.

Idiosyncratic shocks

Period 1979–88 2.27 0.98 1.87 2.06

Panel B US Euro area UK Other Dev. Eco.

Common shocks

0.54 0.35 0.7 0.83 Period 1999–2006

1.01 0.9 0.71 0.9

0.38 0.8 0.46 0.49

Note: The entries in the second column are the standard deviation of the four-quarter ahead forecast errors in a given region.

Table 1.1 shows first that the standard deviation of the four-quarter ahead forecast errors has continuously decreased over time. This result is in line with the observed decrease in output growth volatility as reported in the literature (e.g. Stock and Watson, 2005) explained inter alia by weaker international shocks as well as key structural changes (such as new stock management methods and a more credible monetary policy). Table 1.1 also displays the fraction of the forecast error variance for GDP growth due to each of the three potential sources of shocks. In general, Dees and Vansteenkiste (2007) find that spillovers appear to play only a secondary role in determining regions’ GDP developments relative to idiosyncratic and common shocks. While most countries appear more sensitive to common shocks, idiosyncratic shocks become dominant in some periods, such as, for instance, for the UK in the 1979–88 period, for the euro area in the 1989–98 period or for the United States in the 1999–2006 period. This corresponds in most cases to country- or region-specific events (e.g. the

9780230_243231_02_cha01.indd 7

5/7/2010 1:06:30 PM

8

Catching the Flu from the United States

UK economic boom in the 1980s, German reunification and the ERM crisis for the euro area in the 1990s and the US housing market bubble in the last period in the latter). These results are in line with the literature on international business cycles. For instance, Kose et al. (2003, 2008) and IMF (2007) find that global factors generally play an important role in explaining business cycles, especially in industrial countries, while idiosyncratic factors have a larger role in emerging markets and developing countries. With the most recent data, however, regional shocks have become the most important factor in North America, Europe and Asia, where they explain more than 20% of the output fluctuations. The results suggest that the global factor has, on average, played a less important role in the later period and that regional factors have also become more important, especially in regions where trade and financial linkages have increased like in North America, Europe and Asia.2 Kose et al. (2008) show that, for the G7 countries, the common factor explains, on average, a larger fraction of output, consumption and investment volatility in the globalisation period (1986:3–2003:4) than it does in the Bretton Woods (BW) period (1960:1–1972:2). This result also questions the role of exchange rate regimes in business cycle synchronisation. This is studied by Bordo and Helbling (2003), who analyse business cycle synchronisation across 16 advanced economies over 120 years, using annual data that cover four distinct eras with different international monetary regimes: the classical Gold Standard (1880–1913), the interwar period (1920–38), the Bretton Woods regime of fixed but adjustable exchange rates (1948–72) and the modern period of managed floating among the major currency areas (1973–2001). They find that there is a secular trend towards increased synchronisation for much of the twentieth century and that it occurs across diverse exchange rate regimes. They also find that global (common) shocks are the dominant influence across all regimes. The increasing importance of global shocks, we posit, reflects the forces of globalisation, especially the integration of goods and services through international trade and the integration of financial markets. The results above are, however, based on approaches relying on low frequency data (quarterly or annual) and spillovers are defined as a shock originating in one country and affecting the other countries with at least a one-period delay. It might be difficult to clearly distinguish common shocks from spillovers, as some idiosyncratic shocks might take less than a quarter to propagate to other countries. Therefore,

9780230_243231_02_cha01.indd 8

5/7/2010 1:06:30 PM

Introduction 9

some common shocks might also be identified as fast-transmitting idiosyncratic shocks. It is therefore necessary to identify better the various channels of transmission in order to understand the synchronisation of business cycles.

1.4

Channels of transmission

The growing interlinkages across world economies may be due to various economic causes. First of all, we may suspect that the increasing relevance of external trade for advanced economies makes them more vulnerable to external shocks, either positive or negative: a reduction in US disposable income forces US citizens to consume less, and this also affects their demand for goods imported from the euro area. This is commonly referred to as the trade channel. Alternatively, cross-border capital and financial flows represent an increasingly important channel for the international transmission of shocks. This financial channel may also operate in a less direct fashion. Financial markets have become increasingly integrated, so that a tightening of financing conditions in one country has repercussions on other countries as well. This will of course have an impact on real activity in both countries. Chapter 2 will attempt to bring some evidence about the role of trade and financial linkages in international business cycle synchronisation. Drawing on the work by Imbs (2004, 2006), the proposed analysis will study the three determinants of business cycle co-movement: trade flows, financial links and cross-country differences in sectoral specialisation. The motivation for including the third explanatory variable alongside the two measures of economic integration is straightforward. Countries with similar patterns of sectoral specialisation are more likely to be hit by similar industry-specific shocks. This should make their business cycles more synchronised, all other things being the same. The interconnection of these various determinants is also complex. Chapter 2 will also therefore assess the total effects by decomposing direct and indirect effects. Indirect effects (e.g. the effect of trade integration on specialisation in production) could either amplify or counterbalance the direct effect. Hence the overall impact of trade integration has to be assessed by combining the direct and the indirect effects. The empirical evidence shows intuitive direct effects: economies with more intensive trade and financial ties move more closely together. Also,

9780230_243231_02_cha01.indd 9

5/7/2010 1:06:30 PM

10 Catching the Flu from the United States

similar patterns of sectoral specialisation lead to closer business cycle comovement. Regarding indirect effects, Chapter 2 reports that higher trade flows induce specialisation, which, in turn, reduces output correlations. While this effect diminishes the direct effect of trade on business cycle synchronisation, the indirect effects of financial links on specialisation reinforce the direct effects (i.e. they increase output correlations). The relationship between trade and financial integration appears ambivalent. Making allowances for indirect repercussions, the overall positive effect of financial integration on business cycle synchronisation is amplified significantly. By contrast, the positive direct effect of trade is diminished only marginally by indirect repercussions, so its overall effect remains positive. These results help in explaining how increasing linkages between the US and the euro area have fostered the high degree of business cycle synchronisation in the two areas. This co-movement can be partly explained by intensive bilateral relationships in both the real and financial spheres. The next chapters focus on the business cycle relationship between the two economic areas, retrieving historical patterns and measuring the degree of transmission of shocks from the United States to the euro area.

1.5 The US and the euro area linkages: Historical patterns and measures of the transmission of shocks Chapter 3 starts by presenting some stylised facts concerning the linkages and interactions of the US and euro area business cycles. The analysis starts first in the context of various VAR models, that is, time series analysis of the data, which tends to discover co-movements of variables without providing information on the nexus of causality which would connect them to one another. There is strong evidence that economic activity between the US and the euro area economies has been highly interrelated over the last 40 years. Comparisons with past cycles suggest some historical patterns that characterise the cyclical dynamics of both economic areas. First, activity in the US and euro area co-moves over lower frequencies while at a higher frequency, euro area cycles tend to lag those of the US. Second, despite strong co-movements, the dynamic adjustments are different between the two areas: the US economy tends to recover quickly after being hit by sharp cuts in demand, while the euro area countries have had historically milder downturns but slower rebounds. Third, US shocks have tended to become global over time, amplifying

9780230_243231_02_cha01.indd 10

5/7/2010 1:06:30 PM

Introduction 11

their transmission to the rest of the world in general and to the euro area in particular. The results in Chapter 3 are based on purely empirical approaches, aiming at assessing the degree of co-movement of economic activity across countries. Such approaches are however agnostic on the reasons why certain movements occur. Economic fluctuations are driven by a series of exogenous shocks. Demand, supply and monetary shocks are the main candidates to explain business cycle movements. We can also add the oil shocks or credit shocks that can play a large role in some episodes. In any case, this is a mixture of a small number of recurring shocks that drive economic fluctuations. Empirical research typically uses VAR models to understand the dynamics of a series of variables of interest. According to Sims (1980), VARs simply represent an atheoretical technique for describing how a set of historical data is generated by random innovations in the variables of interest. Therefore, the number of shocks is equal to the number of variables and there are often too many shocks to be economically meaningful. In a multi-country context, the problem is even more complex as the number of variables is multiplied by the number of countries. Therefore, it is difficult to disentangle not only the nature of the shock (demand, supply, monetary ... ) but also its source (idiosyncratic or foreign). The process used to recognise the nature and the source of shocks in empirical models is called identification. Identifying shocks is key, as their nature can alter the way the shock is transmitted to the rest of the world and require different policy actions. For instance, the transmission of a supply shock relates more to the diffusion of technology and might impact potential growth more than a demand shock, whose effects would be more immediate. These two shocks’ impacts on prices are also opposite, affecting their transmission across countries and through provoking different reactions in monetary policy. To identify US idiosyncratic shocks and distinguish between supply, demand or monetary ones, we need to estimate a structural model. The literature on estimated New Open Economy Macroeconomics modelling proposes a reduced-form model of open economies that includes DSGE features (see Gali and Monacelli, 2005). The inclusion of IS and Phillips curves allows the distinction between demand and supply shocks. Adding a Taylor rule and a real exchange rate equation takes into account the reaction of monetary policy and the role of terms of trade in the transmission of shocks. Large-scale dynamic general equilibrium models that model the real and financial linkages in the global economy which are New Keynesian in nature (including nominal rigidities, many

9780230_243231_02_cha01.indd 11

5/7/2010 1:06:30 PM

12

Catching the Flu from the United States

sectors and a rich array of countries, as well as explicit treatment of asset flows and asset prices) have been also been developed (see Bayoumi, 2004, or McKibbin and Stoeckel, 2009). However, these models remain largely calibrated and do not necessarily aim at being able to best replicate the data (see Kydland and Prescott, 1996, or Kydland, 1992).3 Chapter 4 first presents the results of the literature on estimating multi-country DSGE models. It shows, however, that such models feature international linkages that remain less strong than suggested by empirical models. To overcome this problem Chapter 4 proposes a framework that combines an empirical model (a Multi-Country New-Keynesian model) that both correctly accounts for international linkages and features New-Keynesian restrictions. While most estimated open-economy DSGE models reduce the rest of the world as a simple, exogenous block, the Multi-Country New-Keynesian approach accounts for the complex interlinkages across countries in a transparent and coherent framework and gives better ideas about the dynamics of the propagation of shocks. This framework therefore allows the simulation of structural shocks, while keeping cross-country linkages that fit the data. This approach, however, remains silent on the implications of asset holdings and asset price adjustment for cross-country business cycle spillover effects. The reason why only real variables are used in New-Keynesian macro models is that for many financial variables, the standard theory does not work, as reflected in the prevalence of ‘puzzles’, of which there are many. The consequence of these puzzles is that one cannot calibrate the standard models for financial variables, since there is no combination of plausible parameters which will allow the theory to match the data. However, we cannot ignore the role of asset prices as a source of shock and as a channel of transmission. In Chapter 5, we move to a closer scrutiny of the role of financial variables and interlinkages in the international transmission mechanism. After having presented some descriptive evidence of growing transatlantic linkages in the financial sector, we define the idea of financial shock, and discuss, by means of an index proposed recently by Diebold and Yilmaz (2009), to what extent financial spillovers can explain real co-movements. Building on these results, we try to assess whether financial variables can help in forecasting transatlantic GDP developments. In particular, we test whether financial variables, being forward-looking, are proven to embody independent information on agents’ expectations about the future prospects for real activity. Using VAR models linking US and euro area real and financial variables, it is shown that the information content of financial variables does not systematically improve the forecasts

9780230_243231_02_cha01.indd 12

5/7/2010 1:06:30 PM

Introduction 13

of real variables from an out-of-sample standpoint, even though increasing the forecast performance over selected sub-periods. The inability to improve a GDP forecast, however, does not imply any irrelevance for financial variables and linkages. Rather, it could suggest that, because of the presence of non-linearities, the link of financial variables to GDP cannot be captured in a linear VAR framework. Two non-linear frameworks are therefore proposed to analyse GDP developments: first, we check whether financial variables are able to anticipate turning points rather than the level of GDP by estimating recession probabilities using a probit model augmented by a large set of financial variables. The analysis shows that using cross-country frameworks that capture correlations across countries’ business cycle phases helps to quantify conditional probabilities of recession. Next, we employ a nonlinear threshold VAR to assess the impact of credit conditions in explaining co-movements and the impact of monetary policy across different regimes. Overall, over business cycle frequency, financial variables do not seem to play a significant role in explaining cross-country correlations in real output. However, in special episodes of financial stress – like in the present circumstances – financial variables play a role in explaining co-movements.

1.6

The 2007–9 financial crisis and lessons for the future

Chapter 6 finally concentrates on how the current recession episode fits into the empirical evidence analysed above. Underlying the possible differences during this episode provides insight on the likely profile of the forthcoming recovery. In particular, the chapter reviews the particularly severe downturn in activity and the unprecedented collapse in world trade. While, at first sight, the fall in trade flows appeared inconsistent with the economic activity downturn, a look at the sector composition of the downturn allows better understanding of developments in trade during the financial crisis. The emerging literature on the increased sensitivity of trade to GDP swings during the 2007–9 financial crisis focuses on the increased complexity in production (for a review, see Francois and Woerz, 2009). In the event of shocks with a large global content, increasingly complex international supply chains can magnify the impact on activity through a sharper contraction in trade, as goods are now manufactured via complex, international networks, so that countries have increasingly become nodes in international

9780230_243231_02_cha01.indd 13

5/7/2010 1:06:30 PM

14

Catching the Flu from the United States

supply chains. In the same vein, McKibbin and Stoeckel (2009), with a multi-country CGE model, provide an analysis of the financial crisis by looking at the diffusion of the shocks across sectors. They find that the drop for durables has been much higher than for non-durables. In addition, the bursting of the housing bubble is identified as being most responsible for the drop in consumption and imports, while the change in the assessment of risk was largely responsible for the drop in investment. After these general considerations about the crisis and its consequences at the global level, the chapter focuses on the impact of the crisis on the euro area. As seen above, history suggests that the euro area has generally had milder downturns but slower rebounds than the US. In the current downturn, however, euro area and US activity have both been strongly hit. Two main factors may have been decisive in explaining the relative intensity of the downturn in the euro area. First, ‘traditional’ structural differences between the euro area and the US – with the euro area having a larger industrial sector and stronger trade openness – may have played a critical role in the international transmission as the crisis had a particularly strong impact on global industrial production and trade. Second, the transmission channels have changed through time, intensifying the size of business cycle fluctuations and the persistence of shocks. This may be related both to closer trade linkages and financial markets becoming more intertwined. Confidence linkages across economies also appear to have become stronger over time. Looking forward, an important factor underlying the US adjustment – namely the need to embark on a long process of rebuilding lifecycle household wealth – may hint at a likely longer persistence of the ongoing correction as well as its greater amplitude in the US compared with the euro area. That would be contrary to historical experience, which suggests that the US has tended to rebound fairly strongly. On the other hand, there may be other aspects that could lead to a repetition of historical patterns in the recovery phase. The more rigid labour and product markets in the euro area might lengthen the time it takes for output to return to its potential and increase the cumulative output loss incurred over the cycle. The weak outlook for global trade and a delayed adjustment in the housing market could also be factors that may hinder a rapid, strong euro area recovery. Overall, the book examines international linkages from several different perspectives. The lesson we learn from the contributions, and in the light of the current episode of turmoil, is that world economies have

9780230_243231_02_cha01.indd 14

5/7/2010 1:06:30 PM

Introduction 15

become increasingly integrated, which makes the role of cooperation and coordination of policies ever more relevant. At odds with such conventional wisdom, Obstfeld and Rogoff (2002) suggested that the need for policy stance coordination would decrease with the increased level of international integration. The basic intuition is that integrated goods and financial markets provide a powerful riskpooling mechanism, leaving policy-makers free to focus on minimising the distortions that might restrict their respective domestic economies. However, such a result holds in Obstfeld and Rogoff’s framework ‘unless risk aversion is very high’. Outside ‘normal times’, international policy coordination becomes key again as, with increasing uncertainty, economies happen to be very sensitive to shocks propagated by tight international linkages. As shown by Chapter 5, some very important channels seem to be activated only in periods of stress. The recent financial crisis has taught us an important lesson: attitudes towards risk vary not only over time but they do so in waves. Phases of excessive risk taking can be followed by sudden reversals driven by abrupt global confidence shocks such as the one experienced in mid-September 2008. Joint interest rate cuts, like those on 13 September 2001 and on 8 October 2008, are examples of coordinated actions in a context of extraordinary uncertainty about the economic outlook (Trichet, 2008). Joint actions are therefore essential when there is a need to respond to a single shock – a shock that can be rapidly transmitted around the globe through financial linkages and effects on confidence. Focusing on the linkages between the US and the euro area, this book not only shows the importance of identifying shocks, their origin and nature in a timely fashion, but also underlines the need to monitor closely the vulnerabilities that could accelerate their propagation worldwide. This would also imply a heightened responsibility for the US in making sure – given its proven engine role in the world economy – that imbalances are not unduly transmitted to the rest of the world.

Notes 1. The GVAR model covers 33 countries, where 8 of the 11 countries that originally joined Stage Three of European Monetary Union on 1 January 1999 are grouped together, while the remaining 25 countries are modelled individually. For more details, see Dees et al. (2007). 2. On the role of globalisation and regionalism in the international linkages, see Dees et al. (2008). 3. On calibration in macro-modelling, see Canova and Ortega (2000).

9780230_243231_02_cha01.indd 15

5/7/2010 1:06:30 PM

16

Catching the Flu from the United States

References Bayoumi, T., 2004, ‘GEM A New International Macroeconomic Model’, International Monetary Fund Occasional Paper No 239. Bordo, M.D. and T. Helbling, 2003, ‘Have National Business Cycles Become More Synchronized?’ NBER Working Paper 10130. Canova, F. and E. Ortega, 2000, ‘Testing Calibrated General Equilibrium Models’, in R. Mariano, T. Schuermann and M. Weeks (eds), Simulation-Based Inference in Econometrics: Methods and Applications. Cambridge University Press. Dees, S., F. di Mauro and W.J. McKibbin, 2008, ‘International Linkages in the Context of Global and Regional Integration’, in F. di Mauro, S. Dees and W.J. McKibbin (eds), Globalisation, Regionalism and Economic Interdependence. Cambridge University Press. Dees, S., F. di Mauro, M.H. Pesaran and L.V. Smith, 2007, ‘Exploring the International Linkages of the Euro Area: A Global Var Analysis’, Journal of Applied Econometrics, 22 (1), 1–38. Dees, S. and I. Vansteenkiste, 2007, ‘The Transmission of US Cyclical Developments to the Rest of the World’, European Central Bank Working Paper No 798. Diebold, F.X. and K. Yilmaz, 2009, ‘Measuring Financial Asset Return and Volatility Spillovers, with Application to Global Equity Markets’, Economic Journal, 119, 158–71. Doyle, B. and J. Faust, 2002, ‘An Investigation of the Co-Movements among the Growth Rates of the G-7 Countries’, Federal Reserve Bulletin, 88 (10), 427–37. François, J. and J. Woerz, 2009, ‘Follow the Bouncing Ball – Trade and the Great Recession Redux’, in R. Baldwin (ed.), The Great Trade Collapse: Causes, Consequences and Prospects, VoxEU.org Ebook, http://www.voxeu.org/index. php?q=node/4297. Gali, J. and T. Monacelli, 2005, ‘Monetary Policy and Exchange Rate Volatility in a Small Open Economy’, Review of Economic Studies, 72 (3), 707–34. Imbs, J., 2004, ‘Trade, Finance, Specialization, and Synchronization’, The Review of Economics and Statistics, 86 (3), 723–34. Imbs, J., 2006, ‘The Real Effects of Financial Integration’, Journal of International Economics, 68, 296–324. International Monetary Fund, 2007, World Economic Outlook. Kose, Ayhan M., Christopher Otrok and Charles H. Whiteman, 2003, ‘International Business Cycles: World, Religion, and Country-Specific Factors’, The American Economic Review, 93 (4) (September), 1216–39. Kose, Ayhan M., Christopher Otrok and Charles H. Whiteman, 2008, ‘Understanding the Evolution of World Business Cycles’, Journal of International Economics, 75 (1), 110–30. Kydland, F., 1992, ‘On the Econometrics of World Business Cycles’, European Economic Review, 36, 476–82. Kydland, F. and E. Prescott, 1996. ‘The Computational Experiment: An Econometric Tool’, Journal of Economic Perspectives, 10 (1), 69–85. Kydland, F. and E. Prescott, 1982, ‘Time To Build and Aggregate Fluctuations’, Econometrica, 50, 1345–70.

9780230_243231_02_cha01.indd 16

5/7/2010 1:06:31 PM

Introduction 17 McKibbin, W. and A. Stoeckel, 2009, ‘The Potential Effects of the Global Financial Crisis on World Trade’, World Bank Policy Research Working paper 5134, World Bank: Washington DC. Obstfeld, M. and K. Rogoff, 2002, ‘Global Implications of Self-Oriented National Monetary Rules’, Quarterly Journal of Economics, 117 (2), 503–35. Pesaran, M.H., T. Schuermann and S.M. Weiner, 2004, ‘Modelling Regional Interdependencies Using a Global Errorcorrecting Macroeconometric Model’, Journal of Business and Economic Statistics, 22, 129–62. Trichet, J.-C., 2008, ‘International Interdependencies and Monetary Policy – A Policy Maker’s View’, Speech at the Fifth ECB Central Banking Conference, Frankfurt am Main, 14 November. Sims, C. A., 1980, ‘Macroeconomics and Reality’, Econometrica, 48, 1–47. Stock, J. H. and M. W. Watson, 2005, ‘Understanding Changes in International Business Cycle Dynamics’, Journal of the European Economic Association, 3, 968–1006.

9780230_243231_02_cha01.indd 17

5/7/2010 1:06:31 PM

2 Business Cycle Synchronisation: Disentangling Global Trade and Financial Linkages

2.1

Introduction

Trade and financial linkages are the arteries through which the lifeblood of the world economy circulates. Does this imply that trade and financial integration increasingly synchronise the ‘pulse rates’ of modern economies, that is, their business cycles? And how can we disentangle the roles played by trade and financial linkages? The relationship between economic integration and business cycle synchronisation has drastically gained importance of late, as the global economy witnessed a highly synchronised downturn in response to shock waves emanating from the US. It is widely held that this strong international co-movement can be partly explained by the high degree of economic integration of the world economy. After decades of globalisation, all major economies – including the US and the euro area – are tightly bound together by financial and trade linkages. Looking ahead, the link between economic integration and output co-movement will also affect the shape of the world economy after the crisis. In particular, it may partly determine whether emerging markets could decouple from conjunctural fluctuations in advanced economies, particularly in the US. Provided that the commitment to open markets by world leaders survives the crisis largely unscathed, decoupling in the midst of a globalising economy may prove illusory. Against this backdrop, we explore empirically whether economic integration fosters the co-movement of business cycles. Moreover, we disentangle the role played by financial and trade linkages in business cycle synchronisation. The remainder of this chapter is structured as follows. Section 2.2 presents stylised facts on the relationship between business cycle 18

9780230_243231_03_cha02.indd 18

5/11/2010 11:15:13 AM

Disentangling Global Trade and Financial Linkages

19

co-movement and linkages in trade and finance, with a special focus on the US and the euro area. Section 2.3 describes the empirical framework underlying our analysis, while Section 2.4 discusses the results. Section 2.5 concludes.

2.2

Stylised facts

Over the last decades, globalisation has drastically expanded the real and financial channels through which shocks can be transmitted across countries. Above all, the international exchange of goods and services has increased significantly, with world imports rising from around 20% of world GDP in the early 1980s to around 30% in 2008 (Figure 2.1). In parallel, financial markets have become more intertwined over time, as indicated by surging cross-country capital flows. Foreign direct investment (FDI) stocks, in particular, increased fivefold, from around 6% of world GDP in the early 1980s to 28% in 2007.1 These developments not only reflect an intensification of traditional forms of international transactions, but also sweeping changes in the organisation of production. In particular, firms are increasingly participating in global supply chains, as distance costs are plummeting. Such qualitative changes to cross-country linkages are likely to affect the synchronisation of business cycles, too.2

900

33

800

31

700

29

600

27

500

25

400

23

300

21

200

19

100

17

0 1980

15 1984

World GDP

1988

1992

1996

World imports

2000

2004

2008

Imports/GDP (rhs)

Figure 2.1 World GDP and world imports (left-hand axis: indices, 1980
100; right-hand axis: percentages; annual data) Note: Last observation refers to 2008. Source: IMF World Economic Outlook database.

9780230_243231_03_cha02.indd 19

5/11/2010 11:15:13 AM

20

Catching the Flu from the United States

2.2.1 Trade channel International trade in goods and services is the traditional channel through which shocks are transmitted from one economy to another. Different countries have a different degree of openness to external trade, and can therefore be more or less vulnerable to trade shocks. The euro area is significantly more open than either the United States or Japan. In fact, its openness in terms of the combined value of imports and exports of goods and services is equivalent (in 2008) to around 55% of its GDP, compared with around 36% and 32% for Japan and the United States, respectively. At any rate, the trade openness of leading world economies has been strongly increasing in the past decade (Figure 2.2). For a glance at the relevance of the trade on real activity in the euro area, we can look at the interaction between exports, imports, GDP and the net trade contribution to GDP over the last decade (Figure 2.3). Three points seem particularly relevant. First, euro area exports and imports tend to move closely together over the medium term. This might be related to the Feldstein-Horioka puzzle, stating that the robust correlation between saving and investment implies, prima facie, far from perfect capital mobility across countries (see Feldstein and Horioka, 100 90 80 70 (%)

60 50 40 30 20 10 0 1991 1993 1995 1997 1999 2001 2003 2005 2007 Euro area

US

Japan

Figure 2.2 Evolution of the trade openness of the euro area, the United States and Japan (percentage of GDP; annual data) Note: Trade openness is measured as exports plus imports of goods and services as a percentage of GDP. The data for the euro area includes intra and extra trade. Last observation refers to 2008. Source: IMF (World Economic Outlook).

9780230_243231_03_cha02.indd 20

5/11/2010 11:15:13 AM

Disentangling Global Trade and Financial Linkages 3.0 2.5 2.0 1.5 1.0 0.5 0.0 −0.5 −1.0 −1.5 −2.0 −2.5 −3.0 −3.5 −4.0

1996

1998

2000

Net trade

2002 EA exports

2004

2006

EA imports

2008

21 14 12 10 8 6 4 2 0 −2 −4 −6 −8 −10 −12 −14 −16 −18

GDP

Figure 2.3 Euro area trade volumes and real GDP (annual growth in percentages; net trade contribution in percentage points) Note: Exports and imports cover goods and services and include intra-euro area trade. Last observation refers to 2009Q3. Source: ECB computations based on Eurostat national accounts data.

1980). Nevertheless, the import-export correlation is not just a mirror of the saving-investment correlation (Bebczuk, 2008). In particular, the tendency of imports and exports to move together is possibly due to a higher share of the production processes being delocalised abroad in order to benefit from lower labour costs, therefore generating additional trade flows partly via an increase in imported intermediate inputs. Second, when considering the contribution of trade to GDP, one must take into account the overall impacts on the economy rather than just the basic net trade contribution, which appears to be rather small. Third, while the spillovers to euro area GDP of the relatively strong export growth appear to have been rather subdued, the dynamics of activity – though less volatile – shares some similarity with the trade ones. 2.2.2

Financial channel

Cross-border capital and financial flows represent an increasingly important channel for the international transmission of shocks. During the 1990s, cross-border portfolio financial flows increased in magnitude, stimulated by the liberalisation of financial markets and technological innovations that allowed investors to trade more easily on global markets. Moreover, global competition spurred merger and acquisition (M&A) activities between euro area and non-euro area companies, leading to a considerable increase in FDI.

9780230_243231_03_cha02.indd 21

5/11/2010 11:15:13 AM

22

Catching the Flu from the United States

Looking at its financial account, in net terms, the euro area was an importer of capital during 2002–3 and 2007, while being a net exporter since the beginning of 2008 (Figure 2.4). Foreign direct investments are not the only form of financial transmission: the financial channel may also operate in a less direct fashion. Financial markets have become increasingly integrated, so that a tightening of financing conditions in one country has therefore repercussions on other countries as well. This will of course have an impact on real activity in each country. Another important aspect of the financial channel is the role of international bank lending. In periods of financial stress, many banks respond by cutting lending or selling other assets to reduce the size of their balance sheet. This deleveraging process takes on a global dimension through the fall in international bank lending, amplifying the international propagation of financial turmoil. Globalisation has been an important feature in the banking sector. Banks’ external claims have shown a strong upward trend rising from USD 10 trillion in 1999 to about USD 35 trillion prior to the financial crisis in the second half of 2008 (Figure 2.5). As a response to capital shortages, several institutions

400 300 200 100 0 −100 −200 −300 2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

Net combined direct and Portfolio Investment Net direct investment Net portfolio investment

Figure 2.4 Euro area net direct and portfolio investment flows (EUR billions; 12-month cumulated data) Note: A positive (negative) number indicates a net inflow (outflow) into (out of) the euro area. Last observation refers to September 2009. Source: ECB.

9780230_243231_03_cha02.indd 22

5/11/2010 11:15:14 AM

Disentangling Global Trade and Financial Linkages

23

40,000 35,000 30,000 25,000 20,000 15,000 10,000 5000 0 1999

2001 Other US

2003 dev. co. EA

2005 offshore UK

2007

2009

Japan

Figure 2.5 Bank’s external claims in major countries (total amount outstanding – stock – in USD billions) Notes: Locational data, based on resident principle. EA includes intra-euro area lending. Last observation: 2009Q2. Sources: BIS and ECB calculations.

cut their external claims, leading to a steep decline in international bank lending. While Chapter 6 gives more details about the role of global deleveraging in the 2008–9 financial crisis, this episode of financial stress shows how a generalised weakness of the banking sector might have some impact on credit formation, in turn widening the negative impacts of a shock to the real economy. 2.2.3 Integration and business cycle synchronisation All in all, one would expect the business cycles of highly integrated economies to move more closely together. Figure 2.6 provides preliminary evidence for this conjecture (data description in Appendix 2.1). For the sake of simplicity, we focus on the bilateral GDP correlations of selected economies – including most euro area countries – with the US.3 First, it appears that countries trading intensively with the US co-move more with US GDP than others. Second, higher bilateral FDI-related linkages go hand in hand with higher GDP correlations. Third, there is a – rather weak – positive link between the bilateral stocks of portfolio

9780230_243231_03_cha02.indd 23

5/11/2010 11:15:14 AM

24 Catching the Flu from the United States Trade intensity (T)

(b)

Foreign direct investment

0.8

0.8

0.7

0.7

GDP correlation

GDP correlation

(a)

0.6 0.5 0.4 0.3 0.2

0.6 0.5 0.4 0.3 0.2

0

0.1

0.2

0.3

0.4

0

0.5

Portfolio investment

(d)

Sectoral specialisation (S)

0.8

0.8

0.7

0.7

GDP correlation

GDP correlation

(c)

100 200 300 400 500 600 FDI

Trade intensity

0.6 0.5 0.4 0.3 0.2

0.6 0.5 0.4 0.3 0.2

0

500 1000 Portfolio investment

1500

0

0.1 0.2 0.3 Specialisation

0.4

Figure 2.6 Bilateral GDP correlation with the US and its determinants for selected countries Note: This small sub-sample of 14 countries includes all euro area countries (except Greece, Finland, Ireland and Slovenia) as well as Japan and the UK.

capital and business cycle co-movement. Fourth, if the sectoral specialisation of a country differs significantly from that of the US, the bilateral GDP correlation is lower. Starting from these stylised facts, we will now investigate more rigorously the relationship between economic integration and business cycle synchronisation. To this end, we expand our sample to 56 emerging and advanced economies.

2.3 Literature review Trade and financial linkages play a significant role in the international transmission of shocks and in business cycle synchronisation. Empirical studies and theoretical predictions, however, give contradictory results.

9780230_243231_03_cha02.indd 24

5/11/2010 11:15:14 AM

Disentangling Global Trade and Financial Linkages

25

While empirical research has generally found that pairs of countries with relatively strong trade and financial linkages tend to have more highly correlated business cycles, the theoretical models cannot deliver results that are quantitatively consistent with such empirical findings. Theoretical models studying business cycle synchronisation are based on the standard international real business cycle model à la Backus et al. (1992). In a two-country open economy model with complete financial markets, these authors show that, in a world of fully integrated asset markets, high trade intensity is associated with lower business cycle correlations. Extending this model to account for vertical specialisation, Kose and Yi (2001) suggest that higher trade integration might lead to more or less synchronisation of cycles, depending on the nature of trade and the type of shocks hitting the economies. If higher trade linkages foster specialisation, then industry-specific shocks will mostly hit countries specialising in this industry, probably resulting in more idiosyncratic business cycles. On the contrary, if higher trade linkages increase intra-industry trade – implying in particular an increasing amount of vertical or fragmented trade – then the business cycle might be positively associated with stronger trade ties. Other theoretical models are also able to show that intense bilateral trade tends to accompany highly correlated business cycles (Canova and Dellas, 1993). While theoretical models can account to some extent for the positive relationship between trade linkages and business cycle synchronisation, the impacts of financial integration on output correlations remain unclear. On the one hand, increasing the ability to borrow and lend internationally fosters the transfer of resources across economies and can decrease output correlations. Backus et al. (1992) show that in a complete market model a positive technology shock in an economy attracts capital flows from the rest of the world into this economy, resulting in negatively correlated output. On the other hand, a model in which individuals have incomplete access to international risk sharing has opposite predictions, as Baxter and Crucini (1995) show. Another explanation for business cycle co-movement is similarity in industrial structure. In theory, similar production patterns should affect synchronisation positively, since two economies producing the same types of goods will then be subject to similar stochastic developments. Thus countries with similar production patterns will tend to have synchronised economic cycles. Empirically, higher trade integration increases cross-country output correlations (Clark and van Wincoop, 2001; Frankel and Rose, 1998). Also, most empirical studies show a positive relationship between financial

9780230_243231_03_cha02.indd 25

5/11/2010 11:15:15 AM

26 Catching the Flu from the United States

integration and business cycle synchronisation (see for instance Imbs, 2004, 2006). For Kalemli-Ozcan et al. (2009), however, the positive association between financial integration and business cycle synchronisation is mainly due to not accounting for the effects of country-pair factors and global shocks. Using rich panel data on banks’ international bilateral exposures over 30 years and 20 developed countries, they are able to account for these factors and find a negative relationship between financial integration and business cycle synchronisation. Finally, concerning the similarity in production structure, Kalemli-Ozcan et al. (2001), Bower and Guillemineau (2006) and Imbs (2004) all find that countries with a more specialised production structure exhibit output fluctuations that are less correlated with those of other countries. Overall, the links between trade and financial integration on business cycle synchronisation depend on the type of linkages one country has with another. Accounting for the impact of integration on specialisation is also an important aspect to look at. For instance, Kalemli-Ozcan et al. (2003) show that financial integration causes higher industrial specialisation. The production structure might in turn affect the way trade and financial integration affect output correlations. It is therefore important to consider all the linkages together. The methodology generally used in the literature to test for the relevance of trade and financial channels is the estimation of a single equation. The fact that there may be indirect effects going in opposite directions might account for the generally small impact found in studies using single equation regressions. For instance, Kose et al. (2003), using a single equation regression, find a positive effect of trade on business cycle synchronisation, but a non-significant effect of financial links on output (and consumption) co-movement. To address the possibility of conflicting indirect effects, Imbs (2004, 2006) estimates a system of simultaneous equations to take into account direct and indirect effects on the synchronisation of output. He finds that specialisation patterns have a sizable effect on business cycles. Most of this effect directly reflects differences in GDP per capita. Also, economic regions with strong financial links are found to be significantly more synchronised, even though they also tend to be more specialised.

2.4

Framework

If country pairs trade intensively and are tightly bound together by financial linkages, will they exhibit more synchronised business cycles? And how can we disentangle the roles played by trade and financial linkages?

9780230_243231_03_cha02.indd 26

5/11/2010 11:15:15 AM

Disentangling Global Trade and Financial Linkages

27

We assess these questions empirically, based on a broad cross section of 56 advanced and emerging countries. Compared to the existing literature, two innovations stand out:4 First, we also cover financial integration related to FDI, a salient feature of the most recent phase of globalisation. Second, our data sample includes several emerging economies, particularly in Central and Eastern Europe. Both contributions allow us to shed light on the repercussions of vertical integration on the co-movement of business cycles. Figure 2.7 gives an overview of the estimation framework. Our analysis focuses on three determinants of business cycle co-movement: trade flows, financial links and cross-country differences in sectoral specialisation. The motivation for including the third explanatory variable alongside the two measures of economic integration is straightforward. Countries with similar patterns of sectoral specialisation are more likely to be hit by similar industry-specific shocks. This should make their business cycles more synchronised, all other things being the same. As can be seen in Figure 2.7, our estimation framework captures a rich set of interactions between the determinants of business cycle synchronisation. This will allow us to disentangle the various channels through which economic integration affects cross-country output correlations. First, there are direct effects of trade, financial linkages and specialisation on business cycle synchronisation (marked by ‘a’ in Figure 2.7). For instance, closer trade linkages might increase output correlations, because a rise in country A’s activity will raise country B’s exports to A and, thereby, B’s production, resulting in a simultaneous rise in GDP.5 Second, we also take heed of indirect effects. For instance, higher trade flows may lead to more specialisation in production, which, in turn, can have an impact on output co-movement. This indirect effect of trade on business cycle synchronisation could either amplify or counterbalance

a

Business cycle synchronisation

a

a Financial linkages

Trade

Specialisation Figure 2.7

Estimation framework

9780230_243231_03_cha02.indd 27

5/11/2010 11:15:15 AM

28

Catching the Flu from the United States

the direct effect. Hence the overall impact of trade integration has to be assessed by combining the direct and the indirect effects. The same applies, of course, to financial linkages and specialisation. Put differently, the explanatory variables are not independent but rather interact with each other. Therefore, we simultaneously estimate the following system of four equations, following Imbs (2004, 2006): Direct effects:

i,j
0 1Ti,j 2 Fi,j 3Si,j 4 I l,i,j  l,i,j

(2.1)

Indirect effects: Trade links: Ti,j
ȕ0  ȕ1 Fi,j  ȕ2 Si,j  ȕ3 I 2,i,j  2,i,j

(2.2)

Financial links: Fi,j
 0 1 I 3,i,j   3,i,j

(2.3)

Specialisation: Si , j
0  1Ti , j  2 Fi , j  3 I 4,i , j   4,i , j

(2.4)

Each observation refers to one country pair (i,j). The coefficients of the first equation will give us the direct effects of, respectively, trade (T), financial linkages (F) and specialisation (S) on bilateral output correlation ( r ).6 The remaining equations (2.2)–(2.4) capture the interaction of the explanatory variables and, thereby, allow us to keep track of the indirect effects on business cycle synchronisation. More specifically, equation (2.2) shows how bilateral trade relationships are affected by financial linkages and the pattern of specialisation. Equation (2.3) gives the effects of trade and specialisation on financial linkages, while equation (2.4) looks at the impact of trade and finance on specialisation. Finally, additional exogenous variables are included in I1, I2, I3 and I4. To derive the overall (or ‘net’) effects of, respectively, trade and financial ties on business cycle synchronisation, one has to combine the direct and the indirect effects. To this end, we will ultimately use the

9780230_243231_03_cha02.indd 28

5/11/2010 11:15:15 AM

Disentangling Global Trade and Financial Linkages

29

parameters of equations (2.1)–(2.4), to study the overall ‘response’ of the whole system to increases in trade and financial linkages. As business cycle synchronisation and the three determinants considered here – trade, financial links and specialisation – form a system of simultaneous equations, simple ordinary-least-squares (OLS) estimation may lead to biased results. To account for this so-called endogeneity problem, we estimate the system (2.1)–(2.4) with three-stage-least-squares (3SLS). More details on this method are left for Appendix 2.1.

2.5 Empirical results In this section, we present the results of our empirical analysis based on a cross section of 56 advanced and emerging economies (which together form 964 country pairs).7 The following findings stand out: 1. Direct effects (Figure 2.8): In line with conventional wisdom, economies with more intensive trade ties move more closely together. Also, similar patterns of sectoral specialisation lead to closer business cycle co-movement. By contrast, it remains difficult to find a positive, significant relationship between bilateral financial linkages and output correlation. 2. Indirect effects (Figure 2.8): We find that higher trade flows induce specialisation, which, in turn, reduces output correlations. While this effect diminishes the direct effect of trade on business cycle synchronisation, the indirect effects of financial links on specialisation also increase output correlations. 3. Total effects: Making allowance for indirect repercussions, the positive effect of financial integration on business cycle synchronisation is amplified significantly. By contrast, the positive direct effect of Business cycle synchronisation

+

Non significant

− Trade −

Financial linkages

− + Specialisation

Figure 2.8

−

Summary of estimation results

9780230_243231_03_cha02.indd 29

5/11/2010 11:15:16 AM

30 Catching the Flu from the United States

trade is diminished only marginally by indirect repercussions so that its overall effect remains positive. Our estimation strategy consists of two steps. First, we estimate the system of equations (2.1)–(2.4) simultaneously to explore the direct and indirect effects of trade, financial links and specialisation on business cycle synchronisation. Second, we derive the overall effects, that is, the combination of direct and indirect effects. 2.5.1 Step 1: Estimation As described in the previous section, in a first step we estimate simultaneously the system of equations (2.1)–(2.4). This will give us (a) the direct effects of trade, financial integration and specialisation on business cycle synchronisation, as well as (b) the indirect effects stemming from the interaction of the three determinants. We take up each in turn. (a) Direct effects: The direct effects – as reported in Table 2.1 – are given by the estimated coefficients from equation (2.1). In a nutshell, the coefficients of trade and specialisation are significant with the expected sign. However, we cannot find any significant relationship between financial integration and output correlations. In more details, we find first that economies with intensive trade relationships move more closely together than other country pairs Table 2.1

Estimation results of equation (2.1) – direct effects

Measures of F → Right-hand side variables↓

FDI

Portf. total

Eq. (2.1) – Correlations ( ri,j ) T (a1) F (a2) S (a3) nb. obs.

0.05 (3.06) –0.02 (–1.38) –0.23 (–4.50) 853

0.04 (3.24) –0.01 (–1.20) –0.26 (–4.81) 964

Notes: All variables measured in logs, except r. Variables are averages over 1993–2007. All specifications perform 3SLS, with instruments detailed in Appendix 2.1. T-statistics in parentheses.

9780230_243231_03_cha02.indd 30

5/11/2010 11:15:16 AM

Disentangling Global Trade and Financial Linkages

31

(a1 . 0). This is in line with existing studies, for example, Frankel and Rose (1998), Clark and van Wincoop (2001) and Kose and Yi (2001). Intuitively, if countries A and B exchange goods on a large scale, a rise in country A’s activity will raise country B’s exports to A and, thereby, B’s production, resulting in a simultaneous rise in GDP. To be more specific, an increase in the trade intensity by 1% raises the GDP correlation by about 0.05, depending on the measure of financial integration used.9 Second, the direct financial channels (a2) refers in its narrowest definition to the financial integration between economies through crossborder capital and financial flows. While some studies have pointed out a positive relationship between financial integration and business cycle co-movements in the case of advanced economies (Imbs, 2004, 2006), this result runs against the predictions of a standard international business cycle model (Backus et al., 1992) and becomes challenged when it is extended to developing economies (Kose et al., 2003 or Garcia-Herrero and Ruiz, 2008). As shown in Table 2.1, we cannot find a positive, significant relationship between bilateral financial linkages and business cycle correlation. While the absence of direct link between financial linkages and output correlation is in contrast with previous empirical studies, this result might be partly due to the sample of countries chosen. In this respect, our sample is much larger than the one used by Imbs. For instance, Imbs (2004) uses a sample of 276 pairs and Imbs (2006) uses a maximum of 347 pairs. By contrast, our full sample comprises between 853 and 964 pairs, depending on the specification. Using a much broader sample seems to influence the results, especially when including countries with large differences in development levels. Garcia-Herrero and Ruiz (2008) also obtain results that are different from Imbs using a sample that includes many emerging economies. Moreover, Dees and Zorell (2010) show that the choice of financial instruments used to account for endogeneity issues among the dependent variables also explains part of the differences with the empirical literature. The financial instruments used in our estimations rely on measures of de jure restrictions on cross-border financial transactions, provided by Schindler (2009), while Imbs (2004, 2006) uses institutional variables that do not so much relate to cross-border financial transactions, but to the local legal framework in general. Testing for the sensitivity of the results to the financial instrument set, Dees and Zorell (2010) show that the choice of such instruments does matter, since cases with a positive, significant a2 can be found when using the same instruments as Imbs.

9780230_243231_03_cha02.indd 31

5/11/2010 11:15:16 AM

32 Catching the Flu from the United States

Finally, as expected, countries specialising in different sectors of production tend to have less synchronised business cycles (a3  0). This result is borne out, for instance, by Kalemli-Ozcan et al. (2001), Bower and Guillemineau (2006) and Imbs (2004, 2006). Countries specialising in different sectors of production are less likely to be hit by the same industry-specific shocks, which should reduce GDP correlations. More specifically, an increase in the measure of specialisation S diminishes output correlation by 0.23–0.26, depending on the measure of financial integration used in the estimation. (b) Indirect effects: Apart from the direct effects on business cycle synchronisation, there is a complex but intuitive interplay between the three determinants (Figure 2.9 and Table 2.2). Taking these indirect effects into account is important, as they may either reinforce or diminish the direct impacts. To start with, closer trade linkages foster specialisation in different sectors of production ( d1 0 ). In essence, international trade allows countries to specialise in sectors for which they have a comparative advantage. Since specialisation is related negatively with business cycle synchronisation, this indirect effect diminishes the direct effect of trade on output co-movement. At the same time, similarity in production structures (low S) is supportive to trade ( b2  0). This mainly reflects the fact that international trade is particularly intense between similar 0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0 −0.01 −0.02 −0.03

Whole

OECD

Goods market integration Direct effect

Whole

OECD

Financial market integration

Indirect effects

Overall effects

Figure 2.9 Overall effects on business cycle correlation and breakdown by direct and indirect effects (whole sample and OECD sample)

9780230_243231_03_cha02.indd 32

5/11/2010 11:15:16 AM

Disentangling Global Trade and Financial Linkages

33

Table 2.2 Estimation results of equations (2.2)–(2.4) – indirect effects (whole sample) Measures of F → Right-hand side variables↓

FDI

Portf. total

Eq. (2.2) – Trade (Ti,j) F (b1)

–0.07

–0.05

S (b2)

(–2.18) –0.23 (–1.55)

(–2.22) –0.37 (–2.59)

Eq. (2.4) – Specialisation ( Si,j ) T (d1) F (d2) Nb. obs.

0.14 (4.56) –0.20 (–8.42) 853

0.07 (3.65) –0.10 (–9.31) 964

Notes: All variables measured in logs, except r. Variables are averages over 1993– 2007. All specifications perform 3SLS, with instruments detailed in Appendix 2.1. T-statistics in parentheses.

countries – especially among advanced economies – due to the importance of intra-industry trade. Unlike trade, however, financial integration appears to make countries more similar in their production structures ( d2  0). As Imbs (2006) points out, theory provides no clear guidance regarding the expected sign of d2 . In any case, the negative impact of financial integration on specialisation creates a positive effect of financial linkages on business cycle synchronisation. Finally, financial integration has a negative impact on trade ( b1  0 ), thereby indirectly diminishing the positive effect of financial linkages on business cycle synchronisation. The negative impact is stronger if FDI rather than portfolio investment is used to measure financial integration. These results indicate that financial integration would be a substitute for trade integration, especially when it relates to production sharing. To check whether such conclusions hold with more restricted samples, we also estimate the system (2.1)–(2.4) based on OECD country pairs. As Table 2.3 shows, the results remain qualitatively similar. Interestingly, the negative relationship between financial linkages and specialisation is stronger for the OECD sample than for the whole sample. One explanation could be that financial integration is particularly important between

9780230_243231_03_cha02.indd 33

5/11/2010 11:15:16 AM

34

Catching the Flu from the United States

Table 2.3 sample)

Estimation results of equations (2.1)–(2.4) (OECD

Measures of F → Right-hand side variables ↓

FDI

Portf. total

Eq. (2.1) – Correlations ( ri,j ) T (a1) F (a2) S (a2)

F (b1) S (b2)

T (d1) F (d2) Nb. obs.

0.07 (3.12) 0.02 (1.03) –0.19 (–3.13)

0.07 (4.46) 0.02 (1.13) –0.17 (–2.50)

Eq. (2.2) – Trade ( Ti,j ) –0.03 –0.07 (–0.62) (–1.75) –0.34 –0.60 (–2.00) (–2.94) Eq. (2.4) – Specialisation ( Si,j ) 0.19 0.08 (4.43) (2.97) –0.23 –0.15 (–6.74) (–9.16) 416 421

Notes: All variables measured in logs, except r. Variables are averages over 1993–2007. All specifications perform 3SLS, with instruments detailed in Appendix 2.1. T-statistics in parentheses.

advanced countries, which engage in a division of labour in line with the rationale for intra-industry trade. Also, when F is measured by FDI, the value of d2 is – in absolute terms – higher than when it is measured by portfolio investment. This confirms the previous interpretation that sharing production processes increases the degree of similarity across countries, making them therefore more sensitive to common industryspecific shocks. 2.5.2 Step 2: Overall effects So far, we have studied the direct and indirect channels affecting business cycle synchronisation separately (as illustrated by Figure 2.9). Ultimately, however, we are interested in the overall (or ‘net’) effects of – respectively – trade and financial linkages on the co-movement of business cycles. To this end, we now compute the overall effects, using the estimated coefficients of the model.

9780230_243231_03_cha02.indd 34

5/11/2010 11:15:16 AM

Disentangling Global Trade and Financial Linkages

35

The system of equations Eq. (2.1)–(2.4) allows for a complex interplay of direct and indirect channels affecting business cycle synchronisation. Thus, the direct effects suggested by Eq. (1) could be offset by the indirect ones captured by the remaining equations. To derive the overall impacts of trade and financial integration as well as sectoral similarity, we combine the direct and indirect effects, using the results from the simultaneous estimation. Figure 2.9 illustrates the extent of the overall effects and its decomposition into direct and indirect influences when F is measured with portfolio investment. More details are available in Table 2.4, which reports the values for the indirect channels together with the overall effects. To start with, we have seen that the direct effect of trade on output correlations (a1) is overall found to be positive and significant. Given the specification of our system, indirect trade effects can only stem from interactions with sectoral specialisation S. We already know that trade integration tends to reduce the similarity in production structure (or increase specialisation). As specialisation in turn reduces output correlation, the indirect trade effect of trade on business cycle synchronisation (a3d1) tends to be negative, countervailing the direct impact. However, the overall effect (a1  a3d1) remains positive and significant. As already indicated, taking heed of indirect channels is particularly relevant for financial integration. While we have not been able to find positive, significant direct effects of financial linkages on output correlation, the indirect effects are large enough to change the overall assessment. The first indirect effect stems from interactions with trade integration. Since we have found that financial integration tends to reduce bilateral trade (and trade fosters output correlation), this indirect effect could diminish the impact of financial linkages on business cycle correlation. However, as shown in Table 2.4, this indirect effect is small and in most cases insignificant. The second indirect effect operates through sectoral specialisation. Our estimates show that financial integration between two countries makes them more similar in terms of sectoral production patterns. Sectoral similarity, in turn, tends to increase output correlation. Thus, the second indirect channel creates a positive link between financial integration and business cycle synchronisation. It is large and significant, so that the overall financial channel (a2  a1b1  a3g2) is clearly positive and significant. In a nutshell, we find that financial linkages do not foster output correlation directly, but indirectly, by increasing the similarity of the financially integrated economies. Imbs (2006) also reports cases where lifting financial restrictions lowers S, that is, where financial integration induces greater

9780230_243231_03_cha02.indd 35

5/11/2010 11:15:16 AM

36 Catching the Flu from the United States

similarity, which in turn increases the correlation of output. Estimating a similar system for Spain, Garcia-Herrero and Ruiz (2008) also find a positive indirect effect of financial linkages on output synchronisation via specialisation, although this indirect effect is not large enough to compensate the negative direct effect between financial linkages and GDP synchronisation. Finally, it seems worth noting that we define financial linkages in a very narrow sense, that is, in terms of bilateral asset holdings. Financial integration could be understood in a broader sense, for example, in terms of the mobility of financial flows rather than actual stocks or flows. Also, if financial linkages act through third countries or at the global level, this will not be adequately reflected by bilateral stock data. In this case, the tightening of financing conditions in one country, for instance, may even have repercussions on countries with a relatively low direct financial exposure to this country. In such a framework, financial market integration could also contribute to business cycle synchronisation by ‘globalising’ shocks rather than as a pure transmission channel.

2.6 Conclusion At the beginning of this chapter we noted – with only little exaggeration – that trade and financial linkages are the lifeblood of the world economy. We asked if this would imply that trade and financial integration synchronise the ‘pulse rates’ of modern economies, that is, their business cycles. In line with the literature, our empirical analysis provides an intuitive answer: Intensive trade and close financial ties clearly boost business cycle synchronisation. We also dissected the overall impact of, respectively, higher trade flows and financial integration into direct and indirect effects working through other variables. Notably, while the direct effects of financial integration are found insignificant, they become indirectly significant owing to the repercussions of lower specialisation in production induced by closer financial ties. The positive direct effects of trade integration, however, are slightly diminished by an opposing indirect effect resulting from higher specialisation. Our empirical analysis has direct implications for the linkages between the US and the euro area. Business cycles in the two areas are highly synchronised (see also Chapter 3). Our regression analysis would suggest that this co-movement could be partly explained by intensive bilateral relationships in both the real and financial spheres.

9780230_243231_03_cha02.indd 36

5/11/2010 11:15:16 AM

Disentangling Global Trade and Financial Linkages

37

Moreover, our analysis shows the importance of indirect effects, especially as regards the role of financial integration on similarity in production structure. From a euro area perspective, this is consistent with the fact that a significant part of euro area mergers and acquisitions moved away from manufacturing and towards telecommunications from the late 1990s. In particular, euro area firms have invested heavily in acquiring US technology companies associated with ICT, thereby internalising the knowledge capital of the US economy.10 Such financial integration between the two economic areas has fostered the catching up of euro area firms with their US counterparts, thereby increasing their similarity and probably strengthening their sensitivity to common shocks. Overall, this analysis supports the view that the high degree of economic integration between the two regions – and worldwide – at the eve of the global crisis has contributed to the staggering synchronicity of the downturn in late 2008 and early 2009. What do our findings imply for the medium-term prospects of the global economy? At the current juncture, in the midst of a severe economic crisis, fears of a protectionist spiral and de-globalisation haunt the world economy. Cross-border holdings of financial assets have dwindled amid a flight to safety and widespread repatriation of funds. Moreover, worrying signs of increasing protectionist pressures have emerged across the globe. Our empirical analysis suggests that a retreat of globalisation would weaken the synchronisation of business cycles in the future – in the US, the euro area and beyond. This could come on top of another source of de-synchronisation: The unwinding of global imbalances and deleveraging in some regions of the world (see Chapter 7). If, however, policy-makers avoid succumbing to economic nationalism and if investors resume prudential cross-border financial transactions, ‘Globalisation 2.0’ might somewhat counterbalance the side-effects on business cycle synchronisation exerted by the global readjustment process.

Appendix 2.1 List of countries Full sample (56 countries): Argentina, Australia, Austria, Belgium, Brazil, Bulgaria, Canada, Chile, Colombia, Costa Rica, Cyprus, Czech Republic, Denmark, Egypt, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, India, Indonesia, Ireland, Israel, Italy, Japan, Latvia,

9780230_243231_03_cha02.indd 37

5/11/2010 11:15:16 AM

38

Catching the Flu from the United States

Luxembourg, Malaysia, Malta, Mauritius, Mexico, Netherlands, New Zealand, Norway, Pakistan, Panama, Philippines, Poland, Portugal, Romania, Russia, Slovakia, South Africa, South Korea, Spain, Sweden, Switzerland, Thailand, Turkey, Ukraine, United Kingdom, United States, Uruguay and Venezuela. Data description The dependent variable in our regressions is the pairwise GDP correlation coefficient. GDP data (HP filtered), in PPP terms, cover the period 1993–2007 and are taken from the IMF’s WEO database. All endogenous explanatory variables, except for the specialisation index S, relate to economic integration, measured on a de facto basis. Starting with goods market integration, we follow Imbs (2006) and make use of Deardorff’s (1998) indicator:

Ti,j

1 T

¦ t

(EX ij,t  IMij, t ) NYWt NYi, t NYj,t

(2.5)

Here, EXij,t and IMij,t denote total merchandise exports and imports, respectively, from country i to country j. Furthermore, NYWt stands for world nominal output, while NYi,t and NYj,t denote nominal GDP in countries i and j, respectively. Data on bilateral goods trade are extracted from the IMF’s Direction of Trade Statistics, nominal GDP data from the IMF’s WEO database (all in US dollars). In the benchmark regression, we use data averaged over 1995–2007. Furthermore, we make use of several alternative measures of financial linkages. The first measure relies on bilateral FDI holdings, that is, the sum of country i’s direct investment stocks in country j and country j’s assets in country i. Information on FDI holdings (in US dollars) are taken from the OECD’s Foreign Direct Investment Statistics. Since most countries report inward as well as outward FDI holdings, we are able to expand the sample beyond the pairs formed by the 30 reporting OECD economies. We average observations over 2000–6. The second measure of financial linkages, in terms of portfolio investment, is defined analogously and based on the IMF’s CPIS database. For our estimation, we average available observations (expressed in US dollars) over the period 2001–6. Similarities in the production structure are captured by Si,j,

9780230_243231_03_cha02.indd 38

5/11/2010 11:15:16 AM

Disentangling Global Trade and Financial Linkages

Sij


1 ¦ T t

N

¦ n

39

| Sn,i Sn,j |

(2.6)

where Sn,i (Sn,j) is sector n’s share in total value added in country i (j). The total number of sectors is N. If i and j are completely symmetric, then Si,j
0. We use UNIDO data on sectoral gross value added (in 1990 US dollars) at the one-digit level, averaging observations over 2000–7.11 We turn now to a description of the exogenous variables and instruments used in our analysis. In the trade equation (2.2), I2 comprises standard gravity variables: the bilateral distance between the countries’ capitals and two dummy variables indicating, respectively, if the countries share a common border and if they were part of a single jurisdiction in the past. All these measures are provided by CEPII. Imbs and Wacziarg (2003) find that the patterns of specialisation depend on income per capita and that this relationship is non-monotonous. As countries become more affluent, they first diversify their production, only to specialise again when they pass a certain threshold. In line with Imbs (2004), we therefore include in I4 both the bilateral (log) product of and the difference between GDPs per capita. Both measures are assumed to be exogenous to S and are based on UN data for 1993–2007. The financial instruments are taken from a dataset by Schindler (2009), which, in turn, is based on the IMF’s Annual Report on Exchange Rate Arrangements and Restrictions (AREAER). Covering the period 1995– 2005, the dataset features several measures of financial restrictions. We make use of two indices reflecting, respectively, overall financial restrictions and restrictions to FDI. In addition, we construct a complementary index for restrictions on financial transactions other than FDI. For all three indices, we employ two different versions. The first version is a simple average over the rules applicable to financial inflows and outflows in both countries. In some cases, however, restrictions are not cumulative and only the stricter rule is binding. The second version takes this into account and averages only over the stricter set of rules, for example, the maximum of outward restrictions in country i on the one hand and inward restrictions in country j on the other hand. It should be noted that I2 and I4 are distinct sets of variables. This is necessary for the identification of the system (2.1)–(2.4), as shown by Imbs (2004). All in all, our full sample comprises 56 countries (see the list above), of which 27 are considered as emerging or developing economies and 29 as advanced economies. The latter group includes the US and 25

9780230_243231_03_cha02.indd 39

5/11/2010 11:15:17 AM

40 Catching the Flu from the United States

EU countries. Taking into account missing observations, we arrive at a maximum of 728 country pairs for the whole sample. Estimation method: Three-stage-least-squares (3SLS) As already indicated in the main text, the three-stage-least-squares (3SLS) method is used when endogenous variables are correlated with error terms and the error terms are correlated across equations. It is tantamount to the two-stage-least-squares (2SLS) approach followed by a Seemingly Unrelated Regression (SUR). More specifically, the 3SLS estimator is obtained by carrying out three steps. The first stage is identical to the 2SLS procedure: Instruments for the endogenous regressors are computed as the predicted values of an ordinary-least-squares regression of each endogenous regressor on all exogenous regressors. In the second stage, the 2SLS estimator for each equation is computed and the residuals are used to obtain an estimate of the covariance matrix of the error terms of the simultaneous equations model. In the third stage, the estimate of the covariance matrix of the error terms is used to calculate the generalised least-squares estimator and an estimate of its covariance matrix. As argued by Imbs (2004), this procedure is perfectly adapted to our needs since it combines the features of simultaneous equations procedures, while allowing for the possible endogeneity of some dependent variables. Details on the computation of overall effects Table 2.4

Channels to business cycle synchronisation

Sample

Direct effect (a1) Indirect effect (a31) Overall effects (a1  a3d1)

Whole

OECD

Trade channel 0.04*** –0.02*** 0.02*

0.07*** –0.01* 0.06***

Financial channel Direct effect (a2) –0.01 Indirect effect via trade (a1 b1) –0.00* 0.03*** Indirect effect via spec. (a3d2) Overall effects (a2  a1 b1  a3d2) 0.01***

0.02 –0.01* 0.03** 0.04***

Notes: The values are computed on the basis of the estimates reported in Tables 2.1, 2.2 and 2.3. ***/**/* denote significance at the 1%, 5%, and 10% levels, respectively. F is measured in portfolio investment terms, but the estimates based on FDI would give similar results.

9780230_243231_03_cha02.indd 40

5/11/2010 11:15:17 AM

Disentangling Global Trade and Financial Linkages

41

Notes 1. Sources: IMF WEO database (trade, GDP), UNCTAD World Investment Report 2008 (FDI). 2. For a theoretical model on the link between vertical specialisation and business cycle synchronisation see Burstein et al. (2008). 3. (Our subsequent regression analysis will cover a far larger sample, including more countries and all bilateral correlations.) 4. For closely related papers see, for instance, Garcia-Herrero and Ruiz (2008), Imbs (2004, 2006) and Abbott et al. (2008). 5. In a saving-investment perspective, if global savings are unchanged an investment-led increase in activity in country A should be counterbalanced by a decline in investment in the rest of the world (although not necessarily in country B). This effect might somewhat dampen the direct effects. 6. A detailed description of the data and measures used can be found in Appendix 2.1. 7. A country list can be found in Appendix 2.1. 8. We have also conducted the same estimation using several subcomponents of portfolio investment (equities, short-term debt and long-term debt). The results are broadly in line with those based on total portfolio holdings and therefore not reported here. 9. It should be noted that the trade intensity measure used in our estimation corrects for the size of the economies. Hence an increase in T by 1% is not tantamount to an increase in trade flows by 1%. 10. For further analysis of the knowledge-seeking motive behind euro area FDI to the US in the second half of the 1990s, see De Santis et al. (2004). 11. There are six broad sectors: agriculture, hunting, forestry, fishing (ISIC A-B); mining, manufacturing, utilities (ISIC C-E); construction (ISIC F); wholesale, retail trade, restaurants and hotels (ISIC G-H); transport, storage and communication (ISIC I); other activities (ISIC J-P).

References Abbott, A., Easam, J. and T. Xing, 2008, ‘Trade Integration and Business Cycle Convergence : Is The Relation Robust across Time and Space ?’, Scendinavian Journal of Economics, 110, 403–417 Backus, D.K., P. Kehoe and F.E. Kydland, 1992, ‘International Real Business Cycles’, Journal of Political Economy, 100 (4), 745–75. Baxter, M. and M.J. Crucini, 1995, ‘Business Cycles and the Asset Structure of Foreign Trade’, International Economic Review, 36 (4), 821–54. Bebczuk, R., 2008, ‘Imports-Exports Correlation: A New Puzzle?’, Banco Central de la República Argentina Working Paper 2008–33. Böwer, U. and C. Guillemineau, 2006, ‘Determinants of Business Cycle Synchronisation across Euro Area Countries’, European Central Bank Working Paper No. 587. Burstein, A., C. Kurz and L. Tesar, 2008, ‘Trade, Production Sharing, and the International Transmission of Business Cycles’, NBER Working Paper 13731.

9780230_243231_03_cha02.indd 41

5/11/2010 11:15:17 AM

42

Catching the Flu from the United States

Canova, F. and H. Dellas, 1993, ‘Trade Interdependence and the International Business Cycle’, Journal of International Economics, 34 (1–2), 23–47. Clark, T. and E. van Wincoop, 2001, ‘Borders and Business Cycles’, Journal of International Economics, 55, 59–85. De Santis, R., R. Anderton and A. Hijzen, 2004, ‘On the Determinants of Euro Area Fdi to the United States: The Knowledge-Capital – Tobin’s Q Framework’, European Central Bank Working Paper No. 329. Dees, S. and N. Zorell, 2010, ‘Business Cycle Synchronisation: Disentangling Trade and Financial Linkages’, European Central Bank Working Papers, forthcoming. Feldstein, M. and C. Horioka, 1980, ‘Domestic Saving and International Capital Flows’, Economic Journal, 90, 314–29. Frankel, J. and A. Rose, 1998, ‘The Endogeneity of the Optimum Currency Area Criteria’, Economic Journal, 108, 1009–25. Garcia-Herrero, A. and J.H. Ruiz, 2008, ‘Do Trade and Financial Linkages Foster Business Cycle Synchronization in a small economy?’, Moneday Credito, 226, 187–226. Imbs, Jean, 2004, ‘Trade, Finance, Specialization, and Synchronization’, The Review of Economics and Statistics, 86 (3), 723–34. Imbs, Jean, 2006, ‘The Real Effects of Financial Integration’, Journal of International Economics, 68, 296–324. Kalemli-Ozcan, S., B.E. Sørensen and O. Yosha, 2001, ‘Economic Integration, Industrial Specialization, and the Asymmetry of Macroeconomic Fluctuations’, Journal of International Economics, 55, 107–37. Kalemli-Ozcan, S., B.E. Sorensen and O. Yosha, 2003, ‘Risk Sharing and Industrial Specialization: Regional and International Evidence’, American Economic Review, 93, 903–18. Kalemli-Ozcan, S., E. Papaioannou and J.L. Peydro, 2009, ‘Financial Integration and Business Cycle synchronisation’, NBER Working Paper 14887. Kose, A., E. Prasad and M. Terrones, 2003, ‘How Does Globalization Affect the Synchronization of Business Cycles?’ American Economic Review Papers and Proceedings, 93 (2), 57–62. Kose, M.A. and K.-M. Yi, 2006, ‘Can the Standard International Business Cycle Model Explain the Relation between Trade and Comovement?’ Journal of International Economics, 68, 267–95. Schindler, M., 2009, ‘Measuring Financial Integration: A New Data Set’, IMF Staff Papers, 56, 222–38.

9780230_243231_03_cha02.indd 42

5/11/2010 11:15:17 AM

3 Business Cycle Synchronisation: The United States and the Euro Area

3.1

Introduction

Over the last 40 years economic activity between the US and the euro area has been highly interrelated to an extent which goes far beyond measurable trade, financial and other links. In this chapter we systematically examine such a link to discover that it is very solid over time although possibly not so across episodes, that is, during upturns and downturns. This chapter first shows some stylised facts about the co-movement in real GDP in the US and the euro area (Section 3.2). Section 3.3 shows that business cycle fluctuations in the US tend to lead those in the euro area, although the lead and lag structure also depends on the sign of the shock. Section 3.4 analyses how a change in US GDP is transmitted to the rest of the world and to what extent this transmission has changed over time. It shows that the role of the US in the global economy has changed over time. In particular, US shocks have tended to become more persistent in the most recent periods. Second-round and third partners’ effects also tend to make US cyclical developments more global, thus increasing the role of the United States in the world economy.

3.2

Stylised facts

Some historical patterns appear to characterise US-euro area cyclical dynamics. ●

Real GDP per capita in the US and the euro area strongly co-move over low frequencies, although euro area growth has tended to lag 43

9780230_243231_04_cha03.indd 43

5/20/2010 10:07:01 AM

44

Catching the Flu from the United States 4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 1973

1977

1981

1985

1989

1993

Euro area

1997

2001

2005

US

Figure 3.1 Smoothed GDP per capita growth rates (year on year percentage change, 5-year centred moving average). Note: Last observation is 2006Q2. Source: Eurostat and the US Bureau of Economic Analysis.

●

●

that of the US (see Figure 3.1). While Giannone et al. (2009) estimate the US cycle to lead the euro area one by around four quarters, Dees and Vansteenkiste (2007) find that US downturns are transmitted faster than upturns. The cyclical transmission between the US and the euro area lasts around two quarters for a downturn and seven quarters for an upturn.1 These estimates tend to hold on average over the past, whereas each recession was of course somewhat different in terms of lags, size and length.2 Recessions that originate from the US tend to be severe and are often accompanied by a synchronous and protracted global downturn (see Figure 3.2).3 The US economy recovers quickly after being hit by sharp cuts in demand, while the euro area countries have historically had milder downturns but slower rebounds: Duval et al. (2007) find that euro area countries4 suffer smaller initial impacts of common shocks on the output gap, but output gaps are more persistent than in the US. [Figure 3.3 confirms this pattern mostly for the early recessions in the sample shown.]

9780230_243231_04_cha03.indd 44

5/20/2010 10:07:01 AM

US-Euro Area Business Cycles Synchronisation

45

70 60 50 40

Ten concurrent recessions

30 20 10 0

1960

70

80

90

2000

08: Q4

Figure 3.2 Highly synchronised recessions (percentage of countries in recession; shaded areas denote US recession) Note: Sample over 1960Q1–2008Q4. Source: IMF.

4 2 0 −2 −4 −6 1973

1977

1981

1985

1989

1993

Euro area

1997

2001

2005

2009

US

Figure 3.3 Output gaps (output as percentage of estimated potential) Note: The euro area is for 12 countries only. The figure for 2009 is a European Commission forecast. Last observation is 2009Q4. Source: European Commission (AMECO).

Historical evidence shows also that the nature of the recessions is very relevant in order to ascertain the extent and patterns of the cycle co-movement. More specifically: ●

Recessions associated with financial crises as well as those associated with credit crunches and house price busts have typically been

9780230_243231_04_cha03.indd 45

5/20/2010 10:07:01 AM

46

Catching the Flu from the United States

Recessions

All recessions

Financial crises Highly synchronized recessions

Financial crises which are highly synchronized 0

1

2

3

Duration (quarters)

Figure 3.4

4

5

6

Output loss (% from peak)

Average statistics for recessions

Note: Sample over 1960Q1–2008Q4. Sources: IMF.

Recoveries

All recessions Financial crises Highly synchronized recessions Financial crises which are highly synchronized 0

1

2

Time until recovery to previous peak (quarters)

3

4

5

6

Output gain after four quarters (% from trough)

Figure 3.5 Average statistics for recoveries Note: Sample over 1960Q1–2008Q4. Sources: IMF.

severe and protracted, (see IMF, 2009, and Claessens et al., 2008). In particular, recessions associated with financial crises tend to generate a substantial rise in personal saving rates together with a decline in private consumption. This follows a reassessment of over-optimistic expectations and in the US the need to restore households’ balance

9780230_243231_04_cha03.indd 46

5/20/2010 10:07:02 AM

US-Euro Area Business Cycles Synchronisation

●

47

sheets. Against this background the subsequent upturns have typically been more muted, hampered by weak credit and persistence in the weakness of private demand. Economies hit by financial crises have typically benefited from relatively strong demand in the rest of the world. However, when financial crises originate in the United States, the recessions are deeper and longer – partly because these recessions tend to begin a global synchronous downturn. IMF (2009) shows that recessions that are associated with both financial crises and a global downturn have been usually severe and long lasting (Figures 3.4 and 3.5).

3.3 Leads and lags Economic activity in the United States and in the euro area has been co-moving over the last 40 years. Descriptive evidence suggests that the US leads the euro area business cycle and that fluctuations in the two economies evolve around a common trend. Since 1970 there have been systematic linkages between economic activity in the United States and in the euro area, with GDP in the euro area always lagging behind its US counterpart. Giannone and Reichlin (2006) show that – despite its simplicity – a parsimonious bivariate VAR model (BiVAR) for the two GDPs (United States and euro area) captures the main stylised facts of the linkages between both business cycles over the last 40 years and provides over the past robust and accurate out-of-sample predictions for the euro area GDP. A few key results emerge from this work. First, the US-euro area interaction is characterised by a few strong empirical regularities. First, real GDP per capita in the US and the euro area strongly co-move. Furthermore, business cycles fluctuations in the US lead those in the euro area (see Figure 3.6).5 Second, rates of real GDP per capita in the US and in the euro area share a common trend. The level of euro area real GDP per capita has been on average about 30% less than its US counterpart and the gap between the two areas is mean-reverting around such a value. Giannone and Reichlin (2005) confirm the finding that the US and the euro area share a common trend by formally testing and accepting the hypothesis of cointegration between the real GDP in the US and the euro area. In Giannone and Reichlin (2006) real GDP levels in the euro area and the US are considered in per capita terms and the gap between these two variables is found to be stationary. Figure 3.7 shows the gap between

9780230_243231_04_cha03.indd 47

5/20/2010 10:07:02 AM

48

Catching the Flu from the United States 4 3.5 3 2.5 2 1.5 1 0.5 0 1970

1976

1982

1988 US

1994

2000

2006

Euro area

Figure 3.6 Real GDP growth in the US and euro area (year on year percentage change, 5-year centred moving average) Note: The figure reports the 5-year centred moving averages of annual growth rates of GDP per head in the US and the euro area in the sample 1973–2004. Data source: OECD, National Accounts. Source: Giannone et al., 2009, Figure 9.

38 36 34 32 30 28 26 1970

1974

1978

1982

1986 Gap

Figure 3.7 points)

1990

1994

1998

2002

2006

Average

Gap between US and euro area in per capita GDP (percentage

Note: The figure reports the difference between the log-levels of GDP per head in the US and the euro area in the sample 1970–2006. Data source: OECD, National Accounts. Source: Giannone et al., 2009, Figure 7.

9780230_243231_04_cha03.indd 48

5/20/2010 10:07:02 AM

US-Euro Area Business Cycles Synchronisation

49

the US and euro area real per capita GDP (log) levels (solid line) and the average historical level of this gap (dashed line). These two stylised facts suggest that when GDP levels in the two economic areas are driven away from their long-run relationship, the adjustment dynamics are faster in the United States, while the euro area catches up more slowly. In other words, the euro area is expected to grow faster when it is below the equilibrium relation with the US and vice-versa. As a consequence, in this very stylised setting, the US-euro area gap could help predict the evolution of euro area GDP. In fact, ‘Granger causality’ tests evidence that the gap in the growth rates does not drive future US growth but helps predict growth in the euro area. Finally, recessions in the euro area tend to be milder but more protracted. This fact is explained by the finding in Giannone and Reichlin (2005) that the effect of common shocks is milder and more protracted in the euro area. Restricting the euro area business cycle to being dependent on the US one is however rather strong and ignores many idiosyncratic factors that also play a major role. Using a structural VAR that separately identifies common international shocks, Stock and Watson (2005) provide evidence of the emergence of two cyclically coherent groups, the euro area and English-speaking countries. This is consistent with the literature, which overall concludes that there appears to have been an emergence of at least one cyclically coherent group, the major countries in the euro area (Artis et al., 1997; Artis and Zhang, 1997, 1999; Carvalho and Harvey, 2002; Dalsgaard et al., 2002; Del Negro and Otrok 2003; Helbling and Bayoumi 2003; Luginbuhl and Koopman 2003), and possibly a second, English-speaking group, consisting of Canada, the UK and the US (Helbling and Bayoumi, 2003).

3.4

Asymmetry

Not only the overall sensitivity but also the speed at which developments in the US business cycle spill over to the rest of the world may influence euro area economic developments. To derive these results, Dees and Vansteenkiste (2007) make use of a non-linear Markov-switching model, similar to Philips (1991). In a nutshell, the analysis consists of estimating a two-region Markov-switching time series model for the US and euro area real GDP growth. Within the model, there are two possible states for each country, that is, a high and a low growth state. The correlation of business cycles across the two regions will then affect the probability that each region switches from

9780230_243231_04_cha03.indd 49

5/20/2010 10:07:03 AM

50 Catching the Flu from the United States

high to low growth regimes. While providing a historical insight into the degree and direction of business cycle synchronisation, it should be stressed, however, that such a model cannot attempt to explain the economic forces at work in the transmission of the business cycle. Rather, it attempts to characterise the behaviour of the economies. The starting point is the following two-country Markov-switching model for real GDP growth (y t): y t
nt  t

nt
m1s1t  m2s2t  m3s3t  m4s4t

(3.1)

where sit = 1 if state at date t is i and otherwise it equals 0. The equation moreover allows the error term, ␧t, to be vector-autocorrelated. In case of our two-country model, yt, nt, and ␧t are two-by-one vectors. In the model, there are two possible states for each country (a high and a low growth state); the four different combinations of these will be the four different states in the Markov process. In general, they can be described as follows. State 1: both countries are in high growth. State 2: the home country is in low growth while the foreign country is in high growth. State 3: the home country has high growth and the foreign has low. State 4: both countries have low growth. This convention gives the following values to the four vectors (whereby h represents the home and f the foreign country): ªmh º m1
« 1f » ; ¬ m1 ¼

ªmh º m2
« 2f » ; ¬ m1 ¼

ª mf º m1
« 1f » ; ¬ m2 ¼

ª mf º m1
« 2f » , ¬ m2 ¼

where m1 > m2 for both countries. In this model the correlation of business cycles across countries is measured through the nature of the transition matrix (see Philips, 1991). For the specific Markov-switching model used above, the transition matrix for the Markov process is a four-by-four matrix of probabilities, pij, where pij = Pr(st = j|st1 = i). These probabilities must then sum up to 1 over j for each i. In the business cycles of the two countries are truly independent, then the four-state transmission matrix will look like the following: h f h f h f h f ª 11 11 (1 11 )11 11 (1 11 ) (1 11 )(1 11 )º « » h f h f h f h f (1   )    (1   )(1   )  (1   ) 22 11 22 11 22 11 22 11 « » h f h f h f h f «  11 » (1  22 ) (1 11 )(1  22 ) 11  22 (1 11 ) 22 « » h f h f h f h f  22 (1  22 ) (1  22 ) 22  22 22 »¼ «¬(1  22 )(1  22 )

9780230_243231_04_cha03.indd 50

5/20/2010 10:07:03 AM

US-Euro Area Business Cycles Synchronisation

51

By contrast, if the two Markov processes are perfectly correlated, then they could be represented by a two-by-two transition matrix. In this case, the four-by-four matrix would look like the following: ª p11 «  « «  « ¬«(1  p22 )

0 0 (1  p11 )º    »»    » » 0 0 p22 ¼»

In this case, the values in the second and third row are irrelevant since these states never occur. The transition matrix can also allow for cases where one country leads the other. For example, suppose the foreign country is always in the same state that the home country was in last year. This would be the case where the home country leads the foreign one into and out of recessions. In this case, the transition matrix would look like the following: h h ª11 (1  11 ) 0 0 º « » 0 0 0 1 » « « 1 0 0 0 » « » h h 0 (1   22 )  22 «¬ 0 »¼

A similar matrix can be constructed for the case where the foreign country leads. It is also possible to allow for expected leads of longer than one period. The matrix below illustrates a case where the expected length of the home country’s lead into low growth is 1/(1  a) and its expected lead into high growth is 1/(1  b): h h ª  11 (1  11 ) 0 0 º « »  0 1  » « 0 «1   0  0 » « » h h 0 (1   22 )  22 «¬ 0 »¼

From the above it follows we can see that a great variety of cross-country business cycle transmissions can occur. One possibility that is rather difficult to model, however, is the case where two countries alternate their leads into and out of recessions. If the home country is a clear leader (as in the case illustrated above), one state will always be followed

9780230_243231_04_cha03.indd 51

5/20/2010 10:07:03 AM

52

Catching the Flu from the United States

by the same other state. However, if the two countries alternate in the lead, then this would not be the case. The result comes from the fact that the Markov process is first-order; only last period’s state is allowed to influence the determination of this period’s. For the case where the lead alternates, this is clearly not the case. If this occurs, it will be necessary to look not only at the transition matrix, but also at the smoothed probabilities that generate the estimates. This will show how the states actually evolve over time.6 The strategy of estimation consists in running the Markov-switching model using all possible transition matrices and choosing the specification which entails the highest likelihood ratio and then presents the most appropriate specification. The model is estimated for the US-euro area. We allow for two possible model setups namely: 1) the business cycles are uncorrelated or 2) the US leads the cycle of the other region.7 In the latter case, we can estimate by how many quarters the US leads the cycle of the euro area during an upturn and a downturn. Three main results emerge.8 First, the likelihood ratio test reveals that the scenario that the US and the euro area’s cycle are independent can be rejected in favour of the hypothesis that the US leads the cycle of the euro area. Second, we find that US downturns are transmitted faster to the euro area than upturns. It takes two quarters for a downturn in the US to transmit to the euro area whereas it takes six quarters for an upturn to spill over.

3.5

Evolution over time

The previous research looked at the relationship between the US and the euro area over a long period without considering any possible change over time. In order to verify whether the relationship between the two areas does indeed have a different time profile, Dees and Saint-Guilhem (2009) use a global VAR model (GVAR). The modelling strategy applied is the one introduced in Pesaran et al. (2004), which treats the foreign variables – ‘star variables’ – as weakly exogenous, in order to reduce the number of estimated parameters (the so called ‘curse of dimensionality’). In particular, at the estimation stage country-specific foreign variables are constructed using predetermined coefficients. The procedure has no effect on the estimation because for relatively small open economies it is possible to assume that such foreign variables are exogenous for the parameters of the conditional model. Country-specific VARX* models can be consistently combined to form a GVAR in which all the variables are endogenous.

9780230_243231_04_cha03.indd 52

5/20/2010 10:07:03 AM

US-Euro Area Business Cycles Synchronisation

53

Using such a modelling approach, Dees et al. (2007) are able to estimate a GVAR model for 26 economies with six variables (real output, inflation, exchange rate, a short interest rate, a long interest rate and real equity prices) over the period 1979–2003. Although the GVAR approach allows the estimation of high-dimension systems, the six-variable reduced-form model is too large to envisage time-varying estimations. In order to apply the GVAR modelling approach on a time-varying dimension and still have a workable framework, Dees and Saint-Guilhem (2009) reduce the dimension of the model by restricting the analysis to one variable only (real GDP). The transmission of changes in US economic activity to the rest of the world is then analysed by looking at GDP interactions over time. Moreover, owing to its important role in the global economy, the US is treated as a dominant economy, which implies some changes in the empirical modelling as detailed in Chudik (2008). Focusing on the transmission of US shocks to euro area real GDP during the period 1979–2006,9 Dees and Saint-Guilhem (2009) find a gradual change over time in the sensitivity of the euro area economy to US economic activity. To do so, they re-estimate the GVAR model on a moving window of ten years in order to estimate the time-varying contemporaneous effects of foreign output as well as the time-varying impulse response functions. 3.5.1 The changing role of foreign variables over time Figure 3.8 reports the time-varying estimates of the contemporaneous impacts of US and non-US foreign GDP on euro area domestic GDP, together with 95% bootstrap confidence intervals. The estimation has been realised using a ten-year rolling window, whereas the dates reported correspond to the end of the respective window. The graph starting date of 1989Q1 corresponds therefore to the estimation over 1979Q1–1989Q1, while the date 2006Q4 corresponds to estimation over 1996Q4–2006Q4. Starting with the United States, the impact elasticity with respect to euro area GDP has remained very close to the value resulting from the whole sample estimation (around 0.2), although with some gradual decrease in the most recent periods. For the euro area, the impact elasticity with respect to US GDP (0.1 on average) has varied over time (from 0 to 0.3) without displaying any particular trend. The most recent years, however, exhibit some decrease in the impact elasticity. The impact of non-US foreign GDP has continuously declined from the 1980s to the most recent years (from 0.6 to 0.2), featuring at the same time a great deal of uncertainty.

9780230_243231_04_cha03.indd 53

5/20/2010 10:07:04 AM

54

Catching the Flu from the United States 0.8 0.6 0.4 0.2 0 −0.2 −0.4 1989

1991

1993

1995

1997

US to US

1999

2001

2003

2005

US to Foreign

1.4 1.2 1 0.8 0.6 0.4 0.2 0 −0.2 −0.4 1989

1991

1993

1995

1997

Euro area to US

1999

2001

2003

2005

Euro area to Foreign

Figure 3.8 Time-varying estimates of impact elasticity of real GDP to US and foreign real GDP Note: Dotted lines indicate 95% bootstrap confidence intervals. Source: Dees and Saint-Guilhem (2009).

3.5.2 The changing dynamics of transmission of US cyclical developments over time Figure 3.9 shows the time-varying profiles of the Generalised Impulse Response functions of a 1% increase in US GDP. Similarly to the impact

9780230_243231_04_cha03.indd 54

5/20/2010 10:07:04 AM

US-Euro Area Business Cycles Synchronisation

55

Generalized IR of US GDP to shock on US GDP

Impulse response

1

0.5

0

−0.5 5 10

01Q1

15 20 Number of quarters after shock

04Q2

06Q4

98Q1 95Q1 92Q1 89Q1 Last quarter of sample estimate

Generalized IR of EA GDP to shock on US GDP

Impulse response

0.6 0.4 0.2 0 −0.2 5 10

01Q1

15 Number of quarters after shock

20

89Q1

04Q2

06Q4

98Q1 95Q1 92Q1 Last quarter of sample estimate

Figure 3.9 Impulse response functions of a 1% shock to US real GDP growth on US and euro real GDP growth Source: Dees and Saint-Guilhem (2009).

elasticities, the dates reported correspond to the end of a ten-year period. For the euro area, confirming the impact elasticities, the importance of US shocks seems to have decreased over time. However, it also seems

9780230_243231_04_cha03.indd 55

5/20/2010 10:07:04 AM

56 Catching the Flu from the United States

that the shocks take more time to fade away. While at the beginning of the sample, a US shock died out after a year, for the most recent periods the impacts remain above zero for more than two years. Another interesting observation concerns the shape of the responses. The profiles seem more volatile in the earlier periods while they tend to exhibit smoother dynamics at the end of the sample. Overall, the time-varying Generalised Impulse Response functions show that despite remaining overall rather large the US economy has – on impact – a more limited and declining impact on the rest of the world. This underlines the fact that US influence might have become more indirect (going more through third partners and creating snowball effects). No clear trend emerges from the time-varying analysis, although some slight decline in the responses might be noticed. More importantly, it seems that the impacts have become more persistent compared to earlier periods. This might indicate that the increasing trade and financial integration worldwide might have strengthened the transmission of shocks. The US shocks might have therefore become more global as they travel from the US to the rest of the world, reinforcing the persistence of their impacts. 3.5.3 Discussion of the results The empirical results have shown that the degree of transmission of US cyclical developments varies over time. Overall, it seems that changes in US GDP have had less impact during most recent periods than for earlier ones. The dynamics of the transmission seem to have become smoother, likely resulting from increasing trade and financial integration. Overall, while a shock originating from the US might have the same cumulated impacts, this shock appears to be transmitted in a more muted way. While there is some evidence that the impacts of US shocks have decreased over time, the persistence of such shocks seem to have increased in the most recent periods. The increase in persistence of US shocks, together with the increase in the impact elasticities of non-US foreign activity, emphasises the role of second-round and third partners’ effects. At the same time, however, the time-varying impact elasticities of domestic GDP with respect to foreign GDP have decreased in both the US and the euro area. The decrease of impact elasticities does not necessarily mean a decrease in the sensitivity of the two economies to their international environment. It might indicate instead that these economies face more idiosyncratic shocks and that foreign influences might take a few quarters before impacting on economic growth. The decrease over time

9780230_243231_04_cha03.indd 56

5/20/2010 10:07:05 AM

US-Euro Area Business Cycles Synchronisation

57

might indicate that during the sample period, while the US and euro area economies have become more integrated and more sensitive to their international environment, their respective patterns appear to be less synchronous than in the past. This result is also shown in Kose et al. (2008b).

3.6

Concluding remarks

The current economic recession in the United States has questioned the ability of the global economy to ‘decouple’ from US cyclical developments. While there were some signs of decoupling in the first quarters following the US downturn, they disappeared rapidly towards the end of 2008, when the crisis became more global and economic cycles turned out to be more synchronous across the world. While the increasing economic integration at the world level and the resulting emergence of large economic players, like China, are likely to have weakened the role of the US economy as a driver of global growth, the influence of the United States on other economies remains, however, larger than direct trade ties would suggest. Third-market effects, together with increased financial integration, tend to foster the international transmission of cyclical developments. This chapter has first shown that real GDP per capita in the US and the euro area strongly co-move and that business cycle fluctuations in the US lead those in the euro area. The lead and lag structure also depends on the sign of the shock. We find that US downturns are transmitted faster to the euro area than upturns. It takes two quarters for a downturn in the US to transmit to the euro area whereas it takes six quarters for an upturn to spill over. This chapter has lastly analysed how a change in US GDP is transmitted to the rest of the world and to what extent this transmission has changed over time. It has shown that the US economy remains dominant as a trading partner directly, as well as indirectly via other partners’ trade. Of course, the economies that trade a lot with the US are the most likely to be affected by US economic shocks. At the regional level, however, such effects tend to be diluted and the transmission of US cyclical developments seems to be somewhat dampened by regional integration. Moreover, while no clear trend seems to emerge, it seems that the role of the US in the global economy has changed over time. In the euro area case in particular, it seems that changes in US GDP have weaker impacts during the most recent periods than in earlier periods. However, the persistence of such shocks seems to have increased in the most recent periods. The increase in persistence of US shocks, together

9780230_243231_04_cha03.indd 57

5/20/2010 10:07:05 AM

58 Catching the Flu from the United States

with the increase in the impact elasticities of non-US foreign activity for some regions (emerging ones in particular), emphasises the role of second-round and third partners’ effects, making US cyclical developments more global.

Notes 1. In general, the paper suggests that it takes between one and three quarters for a downturn in the US to transmit to other economies, whereas it takes between two and ten quarters for an upturn to spill over. While the euro area is in the upper bound of these ranges, the transmission is much quicker to major US trade partners like Canada or Mexico. 2. According to the official business cycles dating by the NBER (for the US) and the CEPR (for the euro area), four recessions could be compared: in the 1970s, the start of the recession is 1973Q4 in the US and 1974Q3 in the euro area (three-quarter lag); in the 1980s, the recession starts in 1980Q1 in both economies; in the 1990s, the start of the recession is 1990Q3 in the US and 1992Q1 in the euro area (five-quarter lag); the current recession starts in 2007Q4 in the US and 2008Q1 in the euro area (one-quarter lag). 3. As shown by Figure 3.2, there were three other episodes of highly synchronised recessions in addition to the current cycle: 1975, 1980 and 1992. The 2001 recession was less synchronised, but also strongly affected the euro area. 4. For the euro area, the authors’ sample includes ten countries: Austria, Belgium, Germany, Finland, France, Ireland, Italy, the Netherlands, Portugal and Spain. 5. Figure 3.6 compares US and euro area real per capita GDP growth. Data are five-year, centred, moving averages, which filter out cycles of less than five years and, hence, isolate low frequency fluctuations of GDP growth rates. 6. In our case, an inspection of the smoothed probabilities revealed that in no case the lead and lagging country had switched. 7. Estimating the model allowing for the two cases enables us to verify if the model in which the US leads the cycle of the other region has a log-likelihood which is significantly higher than the model where the cycles are uncorrelated. If this is not the case, it is unclear whether the US indeed leads the cycle of the other region. 8. Complete results, together with results for other countries/regions, are available in Dees and Vansteenkiste (2007). 9. The original empirical analysis was conducted for 26 economies.

References Artis, Michael J., Zenon G. Kontolemis and Denise R. Osborn, 1997, ‘Business Cycles for G7 and European Countries’, Journal of Business, 70, 249–79. Artis, Michael J. and Wenda Zhang, 1997, ‘International Business Cycles and the ERM: Is there a European Business Cycle’, International Journal of Finance and Economics, 2, 1–16.

9780230_243231_04_cha03.indd 58

5/20/2010 10:07:05 AM

US-Euro Area Business Cycles Synchronisation

59

Artis, Michael J. and Wenda Zhang, 1999, ‘Further Evidence on the International Business Cycle and the ERM: Is there a European Business Cycle?’, Oxford Economic Papers, 51, 120–32. Carvalho, Vasco M. and Andrew C. Harvey, 2002, ‘Convergence and Cycles in the Euro-Zone’, Manuscript, University of Cambridge. Chudik, A., 2008, ‘GVAR Approach and the Dominance of the US Economy’, Manuscript, Faculty of Economics, University of Cambridge. Claessens, S., M.A. Kose and M.E. Terrones, 2008, ‘What Happens during Recessions, Crunches and Busts?’, IMF Working Paper WP/08/274. Dalsgaard, Thomas, Jorgen Elmeskov and Cyn-Young Park, 2002, ‘Ongoing Changes in the Business Cycle–Evidence and Causes’, OECD Economics Department Working Paper No. 315. Dees, S., F. di Mauro, M.H. Pesaran and L.V. Smith, 2007, ‘Exploring the International Linkages of the Euro Area: A Gvar Analysis’, Journal of Applied Econometrics, 22, 1–38. Dees, S. and A. Saint-Guilhem, 2009, ‘The Role of the United States in the Global Economy and Its Evolution over Time’, European Central Bank Working Paper No. 1034. Dees, S. and I. Vansteenkiste, 2007, ‘The Transmission of U.S. Cyclical Developments to the Rest of the World’, European Central Bank Working Paper No. 798. Del Negro, Marco and Christopher Otrok, 2003, ‘Time-Varying European Business Cycles’, Working Paper, University of Virginia. Duval, R., J. Elmeskov and L. Vogel, 2007, ‘Structural Policies and Economic Resilience to Shocks’, OECD Economics Department Working Papers, No. 567. Giannone, D. and L. Reichlin, 2005, ‘Euro Area and US Recessions, 1970–2003’, in ‘Euro Area Business Cycle: Stylized Facts and Measurement Issue’, in L. Reichlin (ed.). CEPR. Giannone, D. and L. Reichlin, 2006, ‘Trends and Cycles in the Euro Area: How Much Heterogeneity and Should We Worry about It?’, European Central Bank Working Paper No 595. Giannone, D., M. Lenza and L. Reichlin, 2009, ‘Business Cycles in the Euro Area’, European Central Bank Working Paper No. 1010. Helbling, Thomas F. and Tamim A. Bayoumi, 2003, ‘Are They All in the Same Boat? The 2000–2001 Growth Slowdown and the G-7 Business Cycle Linkages’, Manuscript, International Monetary Fund. IMF World Economic Outlook, April 2009, ‘From Recession to Recovery: How Soon and How Strong?’ Chapter 3. Kose, Ayhan M., Christopher Otrok and Charles H. Whiteman, 2008, ‘Understanding the Evolution of World Business Cycles’, Journal of International Economics, 75 (1), 110–30. Luginbuhl, Rob and Siem Jan Koopman, 2003, ‘Convergence in European GDP Series: A Multivariate Common Converging Trend-Cycle Decomposition’, Tinbergen Institute Discussion Paper 2003-031/4. Pesaran, M.H., T. Schuermann and S.M. Weiner, 2004, ‘Modelling Regional Interdependencies Using a Global Errorcorrecting Macroeconometric Model’, Journal of Business and Economic Statistics, 22, 129–62.

9780230_243231_04_cha03.indd 59

5/20/2010 10:07:05 AM

60

Catching the Flu from the United States

Phillips, K.L., 1991, ‘A Two-Country Model of Stochastic Output with Changes in Regime’, Journal of International Economics, 31, 121–42. Stock, J.H. and M. Watson, 2005, ‘Understanding Changes in International Business Cycle Dynamics’, Journal of the European Economic Association, 3, 968–1006.

9780230_243231_04_cha03.indd 60

5/20/2010 10:07:05 AM

4 The United States and the Euro Area: What Do Structural Models Say About the Linkages?

The results so far have been based on purely empirical approaches aimed at assessing the degree of co-movement of economic activity across the two sides of the Atlantic. Such approaches however are agnostic on the reasons why certain movements occurred, which hampers their capacity to ascertain the nature and extent of the transmission mechanism from the United States to the euro area. For instance, an equally sized shock to US activity would have a smaller global spillover if caused by a purely US-specific reason, for example, a real estate market boom-bust, than if caused by some global factor, for example, an oil price increase. Among purely domestic shocks also, impacts on the rest of the world will be different depending upon their nature, that is, whether they have a demand or supply origin. For instance, a supply shock – related to the diffusion of a specific technology – will tend to impact more on potential growth than a demand shock, whose effects would be more immediate both in the US and in the rest of the world. In order to shed light on the above, we need to avail ourselves of a structural approach, which would be able to disentangle to what extent economic activity movements have been generated by specific factors, and then use this information to ascertain the international transmission. Over the last decade, a considerable literature has emerged on ‘NewKeynesian DSGE models’, where the decisions made by households, firms and a policy-maker are interrelated and inter-temporal (see Clarida et al., 1999, for a review). These models also pay particular attention to the rigidities that exist in price and wage setting, often incorporating Calvo contracts, in which prices are reset only periodically and with 61

9780230_243231_05_cha04.indd 61

5/11/2010 4:57:41 PM

62 Catching the Flu from the United States

a fixed probability. A simple New-Keynesian DSGE model consists of three basic equations: 1) an ‘IS curve’ relating output to the real interest rate, 2) a Phillips curve relating inflation to output and 3) a ‘monetary policy rule’ relating the nominal interest rate to output and inflation. The IS curve is derived from the optimising behaviour of households, the Phillips curve is based on the profit-maximising pricing behaviour of monopolistically competitive firms and the policy rule is based on a policy-maker who optimises an objective function in terms of inflation and output. Price and wage rigidities have important implications for the dynamic properties of the DSGE models and their ability to fit the data. As shown for instance by Gali and Gertler (1999), Clarida et al. (1999), Smets and Wouters (2003), Favero and Rovelli (2003), Del Negro and Schorfheide (2006) and Christiano et al. (2003), the New-Keynesian DSGE models perform relatively well in explaining various episodes of historical macro experience and in forecasting. Most notably they represent useful tools for decomposing business cycle developments into exogenous shocks. Such models, however, particularly in an open-economy version, are extremely difficult to estimate and therefore they have been solved in general using calibrated coefficients, that is, drawing from average literature parameter estimations. Attempts to bring such models to the data have so far remained limited, although the increasing popularity of such frameworks within central banks is giving a boost to such efforts. In this chapter, after presenting an overview of this class of structural model – particularly in their open-economy/multi-country version (Section 4.1), we will focus on how they can be used to assess linkages between the United States and the euro area (Section 4.2). Overall, the main result is that this class of models features linkages that remain less strong than suggested by empirical models. The chapter will also present a framework (Section 4.3) that allows the simulation of structural shocks, although keeping cross-country linkages that fit the data. To do so, we will add DSGE-type restrictions to the empirical global model reviewed in Chapter 1 (global VAR, GVAR). Through the Multi-Country New-Keynesian model (MCNK model) we will be able to study the transmission of US supply, demand and monetary shocks to the rest of the world and, in particular, to the euro area. With respect to other estimated open-economy DSGE models (which usually reduce the rest of the world to a simple, exogenous block), the MCNK model would account for the complex interlinkages across countries thus providing a better evaluation of the propagation

9780230_243231_05_cha04.indd 62

5/11/2010 4:57:41 PM

What Do Structural Models Say About the Linkages?

63

of shocks. Section 4.4 will review the real-financial interactions in macro-finance models and will show how difficult it is to include financial variables in DSGE models and to model the transmission of financial shocks across economies.

4.1

An overview of open-economy DSGE models

In New-Keynesian DSGE models, behavioural equations are explicitly derived from inter-temporal optimisation of private sector agents under technological, budgetary and institutional constraints such as imperfections in factor, goods and financial markets. The models also feature real and nominal rigidities to account for the stickiness of the variables’ dynamics and to improve the model’s ability to capture the time series properties of the main macroeconomic data. Gali and Monacelli (2005) develop a small open-economy version of a DSGE model with Calvo-type staggered price setting that implies simple log-linearised equilibrium conditions for small open economies. This model can be reduced to a two-equation dynamic system for domestic inflation and the output gap, consisting of a New-Keynesian Phillips curve and a dynamic IS equation. Although identical to the sticky price model of a closed economy, the open-economy model features equilibrium conditions that include specific parameters like the degree of openness or the substitutability among domestic and foreign goods. The driving forces also include foreign output, which remain exogenous to the small open economy. As in its closed-economy counterpart, the two equations are complemented with a monetary policy rule, in order to close the model. In the simplified version of the Gali-Monacelli framework, Lubik and Schorfheide (2007) propose the following model. First, there is a forward-looking open-economy IS curve: y t  Et y t1 [
 (2   )(1
)]( Rt  Et t1 ) Z Zt  [
 (2   )(1
)]Et qt1 (2   )

1





Et y *t1

(4.1)

where, a is the import share and t is the inter-temporal elasticity of substitution. Endogenous variables are aggregate output, yt the CPI inflation rate πt and nominal interest rates Rt . qt are the terms of trade, y*t+1 is exogenous world output and zt is the growth rate of the level of technology. All real variables are expressed in terms of log-deviation from such a level of technology.

9780230_243231_05_cha04.indd 63

5/11/2010 4:57:41 PM

64

Catching the Flu from the United States

Second, optimal price setting of domestic firms leads to the openeconomy Phillips curve:

 t   t t1   t qt1  qt  

yt
 (2   )(1
)

(2   )(1
) y *t
[
 (2   )(1
)]

(4.2)

where k is a function of underlying structural parameters (labour supply, demand elasticity and degree of price stickiness). Third, an exchange rate equation assumes relative PPP to hold: et  t  (1  )qt   * t

(4.3)

where et is the nominal exchange rate and π*t is the world inflation rate. Finally, monetary policy follows an interest rate rule where the central bank adjusts its instrument in response to changes in nominal exchange rate. Rt  R Rt1 (1 R )[ 1pt  2 y t  3et ] tR

(4.4)

The open-economy DSGE model has been extended to a multi-country setting. Central banks and international institutions have also developed open-economy multi-country DSGE models, like the IMF’s Global Economy Model (Pesenti, 2008), the Federal Reserve Board’s SIGMA (Erceg et al., 2006), the Riksbank’s model (Adolfson et al., 2007) or the ECB’s New Area-Wide Model (Christoffel et al., 2008). More recently, the IMF has developed a reduced-form multi-country DSGE model (GPM, small quarterly Global Projection Model) that describes the joint determination of output, unemployment, inflation, a short-term interest rate and the exchange rate (Carabenciov et al., 2008). Efforts to bring small-scale open-economy DSGE models to the data have also been made (e.g. Bergin, 2004; Ghironi, 1999; Lubik and Schorfheide, 2005, 2007). In particular, a number of authors have estimated the structural parameters of DSGE models using Generalised Method of Moments (GMM) estimations of equilibrium relationships, full-information maximum likelihood (FIML) estimations or using Bayesian approaches, which combine the likelihood function with prior distributions for the parameters of the model, to form the posterior density function.1 For example, Smets and Wouters (2003) and Adolfson et al. (2007) estimate a medium-scale small open-economy model for the euro

9780230_243231_05_cha04.indd 64

5/11/2010 4:57:41 PM

What Do Structural Models Say About the Linkages?

65

area. Lubik and Schorfheide (2005), de Walque et al. (2005) and Adjemian et al. (2008) estimate two-country models using US and euro area data. Carabenciov et al. (2008) estimate a three-country DSGE model of the US, euro area and Japanese economies. No model with more countries has been estimated so far, although Carabenciov et al. (2008) indicate that small quarterly models of a number of countries (United States, the euro area, Japan, emerging Asia, oil-exporting countries, Canada, Russia, other industrialised countries, other major oil producers and the rest of the world) will be integrated into a future version of the GPM.

4.2

United States/euro area linkages in DSGE models

To study the role of foreign shocks in the euro area economy, the ECB has developed a micro-founded DSGE model of the euro area (the New Area-Wide Model – NAWM)2, which is designed for forecasting and policy analysis. The NAWM is neo-classical in nature and centred around the inter-temporal decisions of households and firms, which are maximising expected life-time utility and the expected stream of profits, respectively. As a result, forward-looking expectations play a key role in influencing the adjustment dynamics of both quantities and prices, and changes in supply-side factors have a pronounced impact even in the short run. At the same time, the NAWM includes a number of nominal and real frictions that have been identified as empirically important, such as sticky prices and wages (so some Keynesian features prevail in the short run), habit persistence in consumption and adjustment costs in investment. Moreover, it incorporates analogous frictions relevant in an open-economy setting, including local-currency pricing (giving rise to imperfect exchange-rate pass-through in the short run) and costs of adjusting trade flows. Employing Bayesian methods, the NAWM is estimated on 18 key macroeconomic variables, including real GDP, private consumption, total investment, government consumption, exports and imports, a number of deflators, employment and wages and the short-term nominal interest rate. In addition, in order to capture the impacts of external developments, the model includes specifications for the nominal effective exchange rate, euro area foreign demand, euro area competitors’ export prices as well as oil prices. In line with the number of variables, the estimation considers 18 structural shocks. These shocks are latent factors, with an economic interpretation, that help in typifying the sources of the observed fluctuations in the data.

9780230_243231_05_cha04.indd 65

5/11/2010 4:57:42 PM

66

Catching the Flu from the United States

The 18 structural shocks are grouped into five categories: technology, demand, markups, monetary policy and foreign.3 Table 4.1 provides details on the contributions of the NAWM’s structural shocks to the forecast error variances of euro area real GDP over short (one- and four-quarter) and medium-term (20- and 40-quarter) horizons. In the short run, the observed fluctuations in real GDP are primarily driven by domestic demand and foreign shocks. In the very short run, the two latter shocks explain, respectively, 33 and 29% of GDP fluctuations. Over the medium term, the contribution of these shocks gradually decreases and technology shocks become more important. At the 40-quarter horizon, they explain more than half of real GDP fluctuations. Foreign shocks only explain 10% of real GDP fluctuations. Figure 4.1 shows the decomposition of euro area real GDP growth over 1999–2006. The acceleration of real GDP growth in the first two years of EMU can be attributed largely to favourable markup and demand shocks, which offset the overall negative contribution of technology shocks. The NAWM suggests that the subsequent downturn starting in the second half of 2000 was triggered by adverse influences from abroad. For instance, the emergence of new competitors in euro area export markets eventually led to losses in export market shares. Moreover, the sharp deceleration of economic activity in the United States and its spillover to the rest of the world caused a pronounced fall in euro area foreign demand from 2001 onward. Finally, the subdued growth of real GDP over the period 2002–5 is largely explained by negative demand shocks. Throughout this period, monetary policy shocks (i.e. unanticipated deviations of the short-term nominal interest rate from the prescriptions of the estimated interest-rate rule) supported domestic demand and prevented a stronger slowing of real GDP growth. Since 2003 the overall contribution of foreign shocks has been rather

Table 4.1 Forecast-error-variance decomposition in NAWM Shocks Technology Demand Markups Monetary policy Foreign

1 quarter

4 quarters

20 quarters

40 quarters

0.21 0.33 0.07 0.1

0.27 0.26 0.13 0.08

0.49 0.18 0.17 0.04

0.51 0.13 0.13 0.03

0.29

0.23

0.12

0.1

Source: Christoffel et al. (2008).

9780230_243231_05_cha04.indd 66

5/11/2010 4:57:42 PM

What Do Structural Models Say About the Linkages?

67

Real GDP growth in deviation from mean

5.0 4.0 3.0 2.0 1.0 0.0 −1.0 −2.0 −3.0 −4.0 2000

2001

2002

Technology Monetary policy

2003 Demand Foreign

2004

2005

2006

Markups Real GDP growth (yoy)

Figure 4.1 Decomposition of euro area real GDP growth, 1999–2006 Source: Christoffel et al. (2008).

modest. This masks the fact that in 2003 the adverse impact of external risk-premium shocks (accounting for the marked appreciation of the euro) was largely offset by the unwinding of the previous shocks to export preferences. In contrast, the favourable developments in euro area foreign demand during the 2004–6 period have been largely compensated for by the continued appreciation of the euro and a renewed deterioration of foreign preferences for euro area exports. While being able to assess the contribution of foreign shocks to real GDP, the NAWM is unable to isolate the spillovers coming solely from the United States. To do so Adjemian et al. (2008) estimate a DSGE model for the US and the euro area. Following the closed-economy work of Smets and Wouters (2003), their model embodies a larger range of frictions and shocks that improve the model’s ability to capture the time series properties of the main macroeconomic data, including twoway interactions between the two areas. In this framework, Adjemian et al. (2008) find that the estimated model implies a relatively low transmission of domestic shocks from one country to the other. More specifically (see Table 4.2), foreign shocks contribute only 1% of aggregate fluctuations for both countries in the long term.4 De Walque et al. (2005) also report very moderate

9780230_243231_05_cha04.indd 67

5/11/2010 4:57:42 PM

68

Catching the Flu from the United States

Table 4.2 Decomposition of conditional variances in Adjemian et al. (2008) Euro area GDP Quarters shocks US shocks Euro area shocks Openeconomy shocks* Common shocks**

1

4

20

0.06 0.87

0.04 0.87

0.01 0.92

0.05

0.04

0.02

0.06

US GDP 40

1

4

20

40

0.01 0.95

0.81 0.05

0.85 0.04

0.9 0.01

0.92 0.01

0.01

0.01

0.12

0.08

0.03

0.03

0.06

0.03

0.02

0.04

0.06

0.05

Note: * Open-economy shocks refer to risk-premium shocks (i.e. shocks on UIP) and home bias shocks. ** Common shocks refers to common factors on productivity shocks, investment shocks, CPI markup shocks and monetary policy shocks, which were selected on the basis of their significance in explaining economic fluctuations. Source: Adjemian et al. (2008).

spillovers in their estimated models for the US and the euro area. At a below-two-year horizon, the contribution of foreign shocks in the estimated model is higher: the contribution of non-domestic shocks in the short-term (one to two years) is close to 13% for both regions, of which one fourth comes from the spillovers of domestic shocks of the foreign country. Overall, although DSGE models are useful for understanding the structural source of shocks, the contribution of foreign shocks remains unrealistically small and the spillovers from the US to the euro area remain marginal, especially when compared with the relationship measured by statistical tools like factor models (see Chapter 1) or GVAR models (see Chapter 3). As these tools remain silent on the structural nature of the shocks driving output fluctuations, in the next section we combine the GVAR modelling approach with DSGE restrictions.

4.3 A Multi-Country New-Keynesian model as a way forward This section presents a Multi-Country New-Keynesian (MCNK) model, developed by Dees et al. (2010). It is estimated for 33 countries, including eight members of the euro area, on quarterly data

9780230_243231_05_cha04.indd 68

5/11/2010 4:57:42 PM

What Do Structural Models Say About the Linkages?

69

from 1979Q1–2006Q4. The model is used to estimate the effect of specific supply, demand and monetary policy shocks. 4.3.1

Description of the MCNK model

Each country has an open-economy version of the standard threeequation forward-looking NK model, as presented in Section 4.1, with a Phillips curve determining inflation, an IS curve determining output and a Taylor rule determining interest rates, plus an equation for the real effective exchange rate. In accordance with the theoretical derivation of the standard DSGE model, all variables are measured as deviations from a steady state. These steady states are calculated explicitly as the long-horizon forecasts from the reduced form of an economic model rather than being approximated by some statistical procedure (see Appendix 4.1 for technical details). The model for each country is estimated by instrumental variables treating tradeweighted averages of foreign variables as weakly exogenous for estimation. The identification of the parameters in the equations of the country-specific models and the measurement of steady states is discussed in more detail in Dees et al. (2009) for the case of the Phillips curve. As a matter of convenience, we refer to this New-Keynesian model as structural, since under certain standard theoretical assumptions, its parameters can be related to ‘deeper’ parameters of technology and tastes, but we do not take a particular position on this interpretation. Within a country, shocks are often identified by assuming that the error in the Phillips curve is a supply shock, in the IS curve a demand shock and in the Taylor rule a monetary policy shock, and that these three shocks are orthogonal. We follow that convention here. While this gives identified supply, demand and monetary policy shocks for each country, we need to make assumptions about the correlation of shocks across countries. In the spirit of the closed-economy approach, we assume that there is no correlation between the different types of shocks across countries, so supply shocks in one country are uncorrelated with demand or monetary shocks in other countries. While it may be reasonable to assume that there is no correlation between different types of shocks, shocks of the same type are likely to be correlated across countries: for instance, supply shocks in all countries will be influenced by global factors that determine supply, such as technology. Similarly, in a globalised economy the influences on demand or monetary shocks in one country are likely to be weakly correlated with the influences on the corresponding shock in other countries. We do not restrict these cross-country correlations

9780230_243231_05_cha04.indd 69

5/11/2010 4:57:42 PM

70 Catching the Flu from the United States

between the same type of shock to zero, nor do we restrict the correlations between the errors in the exchange rates and the errors in other equations to the same figure, since floating exchange rates can respond rapidly to other shocks. The unrestricted correlations can be obtained from the covariance matrix of the estimated structural errors. The countries in the model are linked both directly, because foreign variables appear in the structural equations, and indirectly, because the shocks to the equations may be correlated. The US is treated differently because the use of the dollar as the numeraire for exchange rates means its nominal exchange rate is constant. However, the US real exchange rate varies with the US price level. Because of this, although a Phillips curve determining inflation is estimated for the US, the model is solved in terms of the US price level, in order to determine real exchange rates. The solved value for the US price level is then converted back to an inflation rate for the analysis of the effects of shocks. When the model is solved, the vector of all global variables can be written as a function of current and past structural shocks, letting us calculate impulse response functions and variance decompositions allowing for the possible correlations of supply (or other) shocks across countries. 4.3.2 Estimation of the MCNK model The framework used is a general one, which can be used for any NK type model applied to any number of countries, and the specific model estimated for illustration is chosen to be close to the standard threeequation closed-economy NK model. The country equations are similar to that used by the IMF’s small multi-country projection model, by Carabenciov et al. (2008), who consider three countries, though it would be difficult to generalise their structure. We have a stand~ , where ard Phillips curve, PC, determining the inflation deviation p π it ~ π it p itp i,t1 and p it is the log of the price level, of the form: i it   i i   ib i,t1  if Et1  i,t1  iy y it H i,st

(4.5)

There are no intercepts included in the equations since deviations from a steady state have mean zero. The IS curve determines the output gap, ˜yit, ~ as a function of its own lag; the real interest rate deviation, ~ r it Et1π i,t1 where interest rates and inflation are measured at quarterly rates; the real * effective exchange rate, ~ re it, and foreign output, yit . The equation is: i i  y *  y it iy y i,t1 ir (r it  Et1 ) ie re iy * it i,dt it i,t1

9780230_243231_05_cha04.indd 70

(4.6)

5/11/2010 4:57:42 PM

What Do Structural Models Say About the Linkages?

71

There is a standard Taylor rule, TR, determining the interest rate deviation, rit , (except in Saudi Arabia, for which there is no interest-rate data) of the form: i it  y  H r it  ir r i,t1  i  iy it i,mt

(4.7)

The (log) real effective exchange rate ~ re it is modelled as a first-order autoregression: i it   re i i,t1  re ie i,et

(4.8)

for i=1,2,..,N. There is no US exchange rate equation. The logarithm of the nominal exchange rate of country i against country j is the difference in their rates against the dollar, eijt = eit − e jt , and the effective N exchange rate is ∑ i = 0 wijeijt , using the trade weights introduced above. Thus the (log) real effective exchange rate is:

reit  ¦ j0 w ij(eit  eij)  ¦ j0 w ij( pij  pit) N

N

(eit  pit)  ¦ j0 w ij(ejt  pit)  epit  ep * it

(4.9)

N

* where epit = eit − pit and epit  ∑ j = 0 wij ( ejt  pjt ) . Deviations from a steady ~  ep ~* state can be defined accordingly with ~ re it  ep it it   it .5 The model is solved for the N+1 real exchange rates, ep ~ p ~ , For the US, the numeraire country, eot = 0 , and hence ep ot  ot where − p ot can be calculated from the inflation equation and an initial condition. There are 33 countries in the system and the total number of variables is 130 (three for the US and Saudi Arabia and four for each of the other 31 countries). Appendix 4.2 gives more details about the solution of the global rational expectation(RE) model. In this system, the error in PC (4.5) i,st is interpreted as a supply or cost shock, that in IS (4.6) i,dt a demand shock and that in TR (4.7) ε i,mt a monetary policy shock. These three types of shock are assumed to be uncorrelated with shocks of a different type but may be weakly correlated (in the sense of Chudik et al., 2009) with shocks of the same type in other countries. As discussed in 4.3.1, we allow for correlations between the three structural shocks and i,et, the errors in (4.8) the exchange rate equation.6 This system can be estimated for each country by instrumental variables, IV, subject to restrictions required by the theory. As instruments, following N

9780230_243231_05_cha04.indd 71

5/11/2010 4:57:43 PM

72 Catching the Flu from the United States

the argument in Dees et al. (2009), we use an intercept, the lagged values of the country-specific endogenous variables, the current values of the foreign variables and the oil price deviation. The equations were estimated for each country and, as is common when estimating equations for a large number of countries, there was considerable heterogeneity in the estimates, with those for Latin American countries often being outliers. There was evidence of misspecification in a number of cases, but since we wished to consider a tight specification that corresponds to the standard theory we did not conduct specification searches, for example, adding lags or other global variables, to try to improve the fit of the system. The estimates were restricted to ensure consistency with the theory and avoid non-uniqueness in the model solution. Because of the size of the RE system, the set of admissible parameter values consistent with a solution could not be determined analytically and the restrictions were motivated by the type of restrictions needed to get solutions in smaller systems (for example, that β ib + βif ≤ 0.99). An alternative would be an explicitly Bayesian procedure where the prior joint distribution is such that the posterior distribution has support only over the admissible values of the parameter space, that is, those that provide a solution. Unfortunately, the admissible values of the parameter space are not known. In addition, in some cases there are no strong prior distributions on the coefficients, for instance on the sign of the exchange rate coefficient in the IS curve, and it is not clear how one should adjust these to allow for differences between a very disparate group of countries. Because of the need to initialise the estimates of the steady states, the sample for all equations is t = 1980Q1–2006Q3, except for the Phillips curve in Argentina where it is 1990Q1–2006Q3. Table 4.3 gives descriptive statistics for the estimates of the restricted coefficients of the equations over all countries. More details on the restriction imposed and country estimates are available in Dees et al. (2010). 4.3.3 The effect of shocks There are a large number of possible simulations that could be considered, differing in the type of shock, the variable of interest rates, the structure of the covariance matrix and the structure of the equations in the system. We consider only the standard model, although one could also look at the effect of changing the degree of international linkage through the IS equation. We assume a bordered covariance matrix, where there are covariances between shocks of the same type but not between shocks of different types with the exception of exchange rates, which can respond to the other types of shocks through unrestricted covariances with them. Dees et al. (2010) also consider a block diagonal

9780230_243231_05_cha04.indd 72

5/11/2010 4:57:43 PM

What Do Structural Models Say About the Linkages?

73

Table 4.3 Descriptive statistics for restricted coefficients

Mean

Median

Standard Deviation

Phillips curve – Equation (4.5), N=33 bif

0.83

0.85

0.19

bib

0.09

0.10

0.17

biy

0.12

0.04

0.25

bib  bif

0.92

0.99

0.15

aiy

0.27

0.21

0.28

aie

0.02

0.00

0.12

IS curve – Equation (4.6), N=33

aiy*

0.79

0.75

0.47

air

–0.20

0.00

0.34

Taylor rule – Equation (4.7), N=32 gir

0.58

0.65

0.35

giy

0.04

0.04

0.11

gi

0.25

0.20

0.24

rie

0.67

Exchange rates – Equation (4.8), N=32 0.69

0.13

Note: The estimation sample for all equations is t = 1980Q1–2006Q3, except for the Phillips curve in Argentina where it is 1990Q1–2006Q3. N is the number of countries for which the equations were estimated.

covariance matrix, where exchange rate shocks are not correlated with the other shocks. There are a variety of other assumptions one could make about the structure of the covariance matrix. For supply and demand we consider global shocks. For instance, a global supply shock uses as , which has PPP GDP weights that add to one on the supply shocks of each of the N  1 countries, with zeros elsewhere. Much the same holds for demand shocks. For a monetary policy shock, we assume that the US is the leader and we consider a unit (one standard error) unexpected increase in US interest rates for one period, with simultaneous interest rate responses by all other countries as given by the covariance matrix. This can be interpreted as either an orthogonalised impulse response function, where the US is ordered first in the

9780230_243231_05_cha04.indd 73

5/11/2010 4:57:44 PM

74 Catching the Flu from the United States

interest rate block, or a generalised impulse response function; the two are identical.7 Notice that we are measuring the effects of an unexpected one-period shock not on the variables, but on their deviations from steady states. To examine the effects on the variables themselves, we would also need to consider the changes in the steady state. In response to these shocks the system returns to close to a steady state within five or six years. Notice that although there are only short lags in the system, of no more than one period, and strongly forward-looking behaviour in the Phillips curve, there are complicated dynamics and some slow adjustment to shocks. The largest eigenvalue of the system is 0.973, so adjustment can be slow. Many of the eigenvalues are complex, so adjustment often cycles back to zero. Inflation is a strongly forward-looking variable in this model, and so jumps as expectations adjust to a shock and interest rates respond strongly to inflation. We could examine the IRFS and FEVDs for all 33 countries in the model, but for brevity we present one set of graphs which include the US, the euro area, the UK, China, Japan and Canada. Further details are available in Dees et al. (2010). Appendix 4.3 gives technical details about impulse responses, variance decompositions and shock accounting in the MCNK model. Impulse response functions (IRFs) A global demand shock increases output, inflation and interest rates in all the countries graphed, although the size of the effect varies by country, for example, the effect on Chinese output and UK interest rates is quite small (Figure 4.2). The real effective exchange rate increases (the currency depreciates) in all the countries graphed. Output and interest rates are above steady state for 11–15 quarters, then go below, cycling back to steady state. The positive effect on inflation lasts for a somewhat shorter period. For all the shocks considered, the system returns to steady state within the ten years graphed, and usually within five or six years. A global supply shock causes inflation and interest rates to increase on impact, but then both go below steady state quite quickly, before slowly returning to steady state (Figure 4.3). The supply-side shock reduces output and increases exchange rates. A US monetary shock raises interest rates almost everywhere, but these then go below steady state to offset the shock within a couple of periods (Figure 4.4). This depresses inflation and output, which is consistent with the usual results, for example, Kim (2001). The effect on exchange rates differs between countries. The standard theory follows from the

9780230_243231_05_cha04.indd 74

5/11/2010 4:57:44 PM

What Do Structural Models Say About the Linkages? 0.07

75

Output

0.06 0.05 0.04 0.03 0.02 0.01 0 −0.01 −0.02 1

3

5

7

9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41

0.025

Inflation

0.02 0.015 0.01 0.005 0 −0.005 −0.01 −0.015

1

3

5

7

9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41

0.014

Interest rates

0.012 0.01 0.008 0.006 0.004 0.002 0 −0.002 −0.004

1

3

5

7

9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41

0.008

Exchange rates

0.007 0.006 0.005 0.004 0.003 0.002 0.001 0 1

3

5 US

7

9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 China

Japan

UK

Euro area

Canada

Figure 4.2 Impulse response of a positive unit (one standard error) global demand shock

9780230_243231_05_cha04.indd 75

5/11/2010 4:57:44 PM

76 Catching the Flu from the United States Output 0.02 0.01 0 −0.01 −0.02 −0.03 −0.04 −0.05 −0.06 1 0.025 0.02 0.015 0.01 0.005 0 −0.005 −0.01 −0.015 −0.02 −0.025 1

3

5

7

9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 Inflation

3

5

7

9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41

0.006

Interest rates

0.004 0.002 0 −0.002 −0.004 −0.006 −0.008 −0.01 −0.012

1

3

5

7

9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41

0.008

Exchange rates

0.007 0.006 0.005 0.004 0.003 0.002 0.001 0 1

3

5

US

Figure 4.3 shock

7

9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 China

Japan

UK

Euro area

Canada

Impulse response of a positive unit (one standard error) global supply

9780230_243231_05_cha04.indd 76

5/11/2010 4:57:44 PM

What Do Structural Models Say About the Linkages?

77

Dornbusch (1976) ‘overshooting’ result in a two-country model. A contractionary monetary shock of this sort should cause the exchange rate to appreciate immediately and then slowly return to equilibrium. There has been substantial dispute about whether the data are consistent with this theory, for example, Faust and Rogers (2003). In any event there is some question as to how this theory translates into a prediction for the real effective exchange rate in a multi-country world when other countries can respond immediately to the monetary policy shock and exchange rates also respond immediately to shocks. Among those countries shown in the graphs, the monetary shock causes the real effective exchange rate to increase (depreciate) in Japan and Canada; the UK exchange rate hardly changes and in the other countries it goes down. The shape the responses by output, inflation and interest rates, to the demand shock are very similar to those reported by Smets and Wouters (2007: 599, figure 2) for the US. The response of output and inflation to a monetary policy shock, as shown by their Figure 6, is also very similar. The major difference is that whereas in their model a monetary policy shock causes interest rates to go up then slowly return to zero, in our model the monetary shock, raises interest rates, but this is quickly offset as interest rates go below steady state to offset the fact that inflation has dropped below steady state. This is despite the fact that there is quite a lot of inertia in our Taylor rules, which have a coefficient of lagged interest rates which averages 0.58. The pattern of dynamic adjustments to the global supply shock is different from the standard closed-economy models because cross-variable feedbacks seem to operate at a faster pace: inflation and interest rates quickly move to offset the effects of the global supply shock. This faster rate of adjustment is very pronounced for the US monetary policy shock, which raises interest rates almost everywhere, but these then quite quickly go below their steady states to offset the effects of the shock within a couple of quarters. This depresses inflation and output. Forecast error variance decomposition (FEVDs) The FEVDs sum to close to unity, being a little below on impact and a little above after 12 periods on average (Figure 4.5). While there are differences across countries, in all of them supply and demand shocks account for most of the variation in all three variables in the long run, with monetary policy shocks and exchange rate shocks accounting for relatively little of the variation, and contributing similar amounts, about 8–10% each. Monetary policy shocks account for more of the variation in interest rates in Canada than in other countries, although even here

9780230_243231_05_cha04.indd 77

5/11/2010 4:57:44 PM

78

Catching the Flu from the United States Output

0.002 0.001 0 −0.001 −0.002 −0.003 −0.004 −0.005 −0.006 −0.007

1

3

5

7

9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41

0.0015

Inflation

0.001 0.0005 0 −0.0005 −0.001 −0.0015 −0.002 −0.0025 −0.003 −0.0035

1

3

5

7

9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 Interest rates

0.0025 0.002 0.0015 0.001 0.0005 0 −0.0005 −0.001 −0.0015

1

3

5

7

9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 Exchange rates

0.06 0.05 0.04 0.03 0.02 0.01 0 −0.01 −0.02

1

3

5 US

7

9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 China

Japan

UK

Euro area

Canada

Figure 4.4 Impulse response of a positive unit (one standard error) US monetary policy shock

9780230_243231_05_cha04.indd 78

5/11/2010 4:57:44 PM

What Do Structural Models Say About the Linkages? Inflation US

Output

0

3

6

9

12

0 1.2 1 0.8 0.6 0.4 0.2 0

Japan

1.2 1 0.8 0.6 0.4 0.2 0

UK

1.2 1 0.8 0.6 0.4 0.2 0

0

3

6

9

12

3

6

9

0

12

3

6

9

12

3

6

9

12

3

6

9

0

12

3

6

9

12

3

6

9

0

1.2 1 0.8 0.6 0.4 0.2 0

3

6

9

12

3

6

9

12

3

6

9

12

12

0

3

6

9

12

0

3

6

9

12

0

3

6

9

0

3

6

9

0

3

6

9

12

12

1.2 1 0.8 0.6 0.4 0.2 0

1.2 1 0.8 0.6 0.4 0.2 0 0

9

1.2 1 0.8 0.6 0.4 0.2 0 0

12

6

1.2 1 0.8 0.6 0.4 0.2 0

1.2 1 0.8 0.6 0.4 0.2 0 0

3

1.2 1 0.8 0.6 0.4 0.2 0

1.2 1 0.8 0.6 0.4 0.2 0 0

0 1.2 1 0.8 0.6 0.4 0.2 0

1.2 1 0.8 0.6 0.4 0.2 0 0

Canada

1.2 1 0.8 0.6 0.4 0.2 0

1.2 1 0.8 0.6 0.4 0.2 0

Euro area 1.2 1 0.8 0.6 0.4 0.2 0 China

Interest rates

1.2 1 0.8 0.6 0.4 0.2 0

1.2 1 0.8 0.6 0.4 0.2 0

79

0 Supply

3

6

Demand

9

12 MP

12

REER

Figure 4.5 Forecast error variance decomposition of global shocks in a MultiCountry New-Keynesian model (x-axis refers to quarters)

it is not a large proportion. On impact, supply shocks account for nearly all of the variation in inflation, but this drops rapidly and they only account for about half of the variation in the long run. Demand shocks account for most of the variation in output on impact, but again this drops quite rapidly. Smets and Wouters also find that monetary policy

9780230_243231_05_cha04.indd 79

5/11/2010 4:57:45 PM

80

Catching the Flu from the United States

shocks account for relatively little of the variation and this suggests that better monetary policy cannot account for the reduction in variance observed during the ‘Great Moderation’. The assumptions made above about the covariance matrix followed from the conventional theoretical approach, which assumes that supply, demand and monetary policy shocks are orthogonal. However, one can estimate the covariance matrix and examine the correlations among the estimated shocks. The correlation matrix is a symmetrical 130  130 matrix, so some summary is necessary. We define the shocks in a particular country, i, of type k  s,d,m,e in period t, i,k,t with correlations between countries and shocks given by ri,j,k,l where k,l  s,d,m,e with, for instance, ri,i,k,k  1. These can then be averaged over countries to give rk,l – the average correlations between shocks of a particular type. Notice that rk,k is not necessarily unity. This gives us a 4  4 matrix of average correlations between supply shocks, between supply shocks and demand shocks, and so on. The correlation set to zero turns out to be very low, especially between monetary policy and supply shocks and monetary policy and demand shocks. The correlation between demand and supply shocks is somewhat higher (around 16%). Concerning the correlations that were kept in the empirical analysis, we notice that supply shocks are relatively more correlated across countries (around 50%), than demand and monetary policy shocks (6 and 14% respectively). Finally, correlations with respect to exchange rate shocks are relatively low. Overall, this modelling exercise showed that it is possible to estimate, solve and simulate a forward-looking Multi-Country New-Keynesian model and use this to estimate the effects of identified supply, demand and monetary policy shocks. In constructing such a model it is necessary to be careful about the assumptions made about exchange rates, particularly the treatment of the numeraire, and about the assumed structure of the shock correlations. For all of the focus economies, the

Table 4.4 Correlations among the estimated shocks Supply Supply Demand Mon. pol. Exch. Rate

0.49 – – –

Demand 0.16 0.06 – –

Mon. pol. 0.04 0.06 0.14 –

Exch. rate 0.05 –0.01 –0.04 0.05

Note: Bold correlations are set to zero in the bordered covariance case.

9780230_243231_05_cha04.indd 80

5/11/2010 4:57:45 PM

What Do Structural Models Say About the Linkages?

81

qualitative effects of demand and supply shocks are as predicted by the theory, while monetary policy shocks are offset more quickly than is typically obtained in the literature. Global supply and demand shocks are the most important drivers of output, inflation and interest rates. By contrast, monetary or exchange rate shocks play a secondary role in the evolution of the world economy. Despite the uniformity of the specifications assumed across countries, there are major differences in the size of the effects of the shocks between countries.

4.4

Conclusions

In order to analyse the international transmission of shocks correctly, one needs to ascertain how they differ in nature and origin across different episodes. Since at any point in time actual shocks are multifaceted both in nature and origin, one needs to refer to structural models in order to identify them more precisely before proceeding to assess their impacts. Over the last decade, a considerable literature has emerged on ‘New-Keynesian DSGE models’, where the decisions made by households, firms and the policy-makers are interrelated and inter-temporal. As this class of model is increasingly used in central banks and international institutions, multi-country DSGE models have been estimated in order to improve the analysis of the international transmission of shocks. While efforts to bring DSGE models to the data have been made, the estimation of the multi-country versions remains however limited. Moreover, the quantification of international linkages resulting from their estimation happen to be unrealistically very small. This chapter has shown that this limitation could be overcome by estimating and simulating a forward-looking MCNK model using a similar framework to the GVAR. The MCNK has proven to be a promising way forward as the cross-country linkages appear to be much larger than those derived from estimated multi-country DSGE models, and results appear to be reasonable when compared across pre-identified shocks. The main limitation – common to both estimated multi-country DSGE models and the MCNK model – is the lack of financial linkages, which are as yet very hard to include: Appendix 4.4 reports on the progress made so far on this front. In the following chapter, however, the issue is pursued further by looking at financial variables (and their role in the international transmission of shocks) within a time series framework. It will show, however, that their role remains limited to periods of financial stress, giving a non-linear dimension to the interaction between real and financial variables.

9780230_243231_05_cha04.indd 81

5/11/2010 4:57:45 PM

82 Catching the Flu from the United States

Appendix 4.1 Deviations from steady states There are a variety of methods of measuring deviations from steady states, including taking deviations from ad hoc statistical measures like the Hodrick-Prescott filter. Fukac and Pagan (2009) discuss alternative methods. We follow Dees et al. (2009) and measure the steady states as the long-horizon forecasts from an economic model. Consider k  1 vector of endogenous variables, xt, decomposed into (permanent) steady states and deviations, it , xt  x Pt  x P

xt has deterministic and stochastic components: x Pt  x Pdt  x stP x Pdt   gt, . where μ and g are k  1 vectors of constants, t a deterministic trend. P The steady state (permanent-stochastic component) xst , is then defined as the ‘long-horizon forecast’ (net of the permanent-deterministic component) x Pst  lim Et (x t +h  x Pd,t +h ) lim Et [x t +h    g(t  h)] . h o

ho

This corresponds not only to the natural definition of a steady state to which the system is tending but also to a multivariate BeveridgeNelson (1981) decomposition. The economic model used to provide the long-horizon forecasts is a global VAR, GVAR, which takes account of unit roots and cointegration in the global economy (within as well as across economies). Dees et al. (2009) provide more detail on the GVAR and explain how it can be regarded as the reduced form of a structural model such as the MCNK considered here. For each country i  0, 1, 2, ... ,N, there is a VARX* model of the form: x it  h i0  hi1t  i1x i,t1 i2 x i,t2 C i0 x *it C i1x * i,t1  u it, i  0,1,.., N

and associated VECM, with the cointegrating restrictions

9780230_243231_05_cha04.indd 82

5/11/2010 4:57:45 PM

What Do Structural Models Say About the Linkages?

83

x it  ci0 i 'i [ zi,t1  i (t  1)] C i0 x*it Giz i,t1  u it

where zit  (xit' , x*'it )', ␣i is a ki  ri matrix of rank ri, and ␤i is a (ki  k*i ) matrix of rank ri. This allows for cointegration within x it and between x*it. Then we can stack the N  1 individual country models as a GVAR. GVARs have been very widely used in a variety of versions for a variety of purposes.8 The version used here to calculate the long-horizon forecasts is estimated over the same sample, 1979Q1–2006Q4, for the same 33 countries, explaining the same variables (output, inflation, short interest rates and exchange rates), with the addition of the price of oil, which is included as an endogenous variable in the US VARX*. These 131 endogenous variables are driven by 82 stochastic trends and 49 cointegrating relations. Weak exogeneity of the foreign variables for the individual VARX* equations is rejected only in 8.4% of the cases at 5% level. Other versions of the GVAR include financial variables, but these have been excluded for comparability with the MCNK model, which does not include them. The long-horizon forecasts from this GVAR provide estimates of the ~ P P steady states x t. The deviations from steady states xt  xt  xt are then uniquely identified and can be used as variables in the MCNK model. One advantage of using deviations from steady states is that they are I(0) by construction and so there is no danger of spurious regressions.

Appendix 4.2 Solution of the global RE model The system has three variables for the US and Saudi Arabia and four variables for the other 31 countries, a total of 130 variables. In solving the system we drop inflation from the US vector of variables and include the US price level as determining the real exchange rate ~ ~ epot   pot. To represent this we need to introduce a notation for the two different treatments of the US as reference country. For all ~ . ~ ~ ~ ~ countries i  0, 1, ... , N, let x  (it, yit, r it, epit). For the US this differs ~ ~ ~ ~ from the vector excluding inflation xot  ( yot, rot , epot) while for the ~ . ~ other countries the vectors are the same x it  x it. The system will be ~ . ~ solved in terms of x it but estimated and analysed in terms of x it which includes US inflation. ~ . ~ For the US we can relate the 4  1 vector x ot to the 3  1 vector xot by, x 0t  S00 x 0t  S01x 0,t1,

9780230_243231_05_cha04.indd 83

5/11/2010 4:57:46 PM

84 Catching the Flu from the United States

where § ¨ S00  ¨ ¨ ¨¨ ©

0 0 1 · § ¸ ¨ 1 0 0 ¸ , S01  ¨ ¨ 0 1 0 ¸ ¸¸ ¨¨ 0 0 1 ¹ ©

0 0 1 · ¸ 0 0 0 ¸ 0 0 0 ¸ ¸ 0 0 0 ¸¹

~ ~ ~ . .' ~ .' .' Putting all the countries together, the (k  1)  1 vector xt  (x 0t,x 1t,...,x Nt)' ~ ~' ~' ~' can be related to the k  1 vector x t  (x 0t,x 1t,...,x Nt)' by x t  S0 x t  S1x t1 ,

(4.10)

where  S00 S0    0

0   S01 0 , S1   Ik3   0 0

The country models can be written as * * A i0 x it  A i1x i ,t1  A i2E t1(x i ,t1 ) A i3x i ,t  A i4 x i ,t1 H it ,

(4.11)

~ .* * where x it are ki  1 vectors of foreign variables and expectations are taken with respect to a common global information set formed as the union intersection of the individual country information sets, ℑi,t1. ~ ~ .* .' ~ . *' ' Let z it  x it , x it then the N  1 models (4.11) can be written as

(

)

A iz0 z it  A iz z i,t1  A iz2 Et1( z i,t1 ) it for

i=0,1,2,...,N,

(4.12)

where, for instance, for i  1, 2, ... , N, A iz0  (A i0,  A i3). ~ ~ . . The variables z it are linked to the complete vector of variables, x it , through the identity z it  Wi x t ,

(4.13)

where the ‘link’ matrices Wi are defined in terms of the weights wij. For i  1, 2, ... , N, Wi is (ki k*i ) (k  1) , for i  0, W0 is (k0  1  k*0 )  (k  1). Using (4.13), equation (4.12) can be written as

9780230_243231_05_cha04.indd 84

5/11/2010 4:57:46 PM

What Do Structural Models Say About the Linkages?

A iz0 Wi x t  A iz1 Wi x t1  A iz2 WiEt (x t1 )H it ,

85

i=0,1,...,N

~ . and then stacked to yield the model for x it as A 0 x t  A1x t1  A 2 Et ( x t1 )H it,

(4.14)

where Aj is k  (k  1), for j  0, 1, 2 defined by  A 0zjW0   ε 0t  A W  ε  1zj 1  , ε t   1t  Aj  ⯗    ⯗ A W   ε   Nzj N  Nt ~ . ~ ~ Now recall that x t  S0 x tS1 x t1, so that (4.14) becomes i t S x i i i i i A 0 (S 0 x 1 t1 ) A 1 (S 0 x t1  S1 x t2 ) A 2 Et1 (S 0 x t1  S1 x t ) İ t ,

or i t H x i i i i H0 x 1 t1  H2 x t2  H3 Et1 ( x t1 ) H4 Et1 ( x t )H t

(4.15)

H0  A 0 S0 , H1  A1S0  A 0 S1 , H2  A1S1 , H3  A 2 S0 , H4  A 2 S1.

Note that H0 is a k  k non-singular matrix. Now multiplying (4.15) 1 by H0 i t F x i i i i x 1 t1  F2 x t2  F3 Et1 ( x t1 ) F4 Et1 ( x t ) u t , 1

(4.16)

1

where Fj = H 0 Hj, for j  1, 2, 3, 4, and ut = H0 t. § x t · Let ␨ t  ¨ ¸ then (4.16) can be written as © x t 1 ¹ ␨ t  A␨ t1 BEt1( ␨ t1 )t

where § F1 F2 · § F3 F4 · § ut · A ¨ ¸ , B¨ ¸ and t  ¨ ¸ I 0 0 0 © ¹ © ¹ ©0¹

9780230_243231_05_cha04.indd 85

5/11/2010 4:57:47 PM

86

Catching the Flu from the United States

The solution properties of this system, discussed in Binder and Pesaran (1995, 1997) depend on the roots of the quadratic matrix equation B 2    A  0 . There will be a globally consistent RE solution if there exists a real matrix solution such that all the eigenvalues of ⌽ and (I  B⌽) − 1B lie inside or on the unit circle and the solution is given by ␨t   ␨t 1 t ,

(4.17)

Partitioning ⌽ conformably to t (4.17) can be expressed as § x t · § 11 12 · § x t1 · § Ik ¨  ¸¨ ¸ ¨ ¸¨ 0 ¹ © x t2 ¹ © 0 © xt1 ¹ © Ik

0 · § ut · ¸¨ ¸, Ik ¹ © 0 ¹

~ so that the reduced form solution in terms of xt , is given by 1 i t  x i i x 11 t1  12 x t2  H0 H t

(4.18)

where «t  («0t, «1t, ... , «Nt,), and the structural shocks, «t can be estimated as i t  x i i H t  H0 ( x 11 t1  12 x t2 )

with E(«t«t)   , a k  k (130  130) matrix which can be obtained from the estimated structural shocks.9

Appendix 4.3 Impulse responses, variance decompositions and shock accounting in the MCNK model The effects of shocks is represented by impulse response functions, IRF, and forecast error variance decompositions, FEVD. In this case we need to extend the standard approach somewhat to calculate the IRF and FEVD. To measure the relative importance of the different types of shock in the global economy, it will be convenient to reorder the elements of the «t vector in terms of the different types of shocks as «t0 («st', «dt',«mt',«et')' where «st and «dt are the (N  1)  1 vectors of supply and demand shocks across all economies and «mt and «et are the N  1

9780230_243231_05_cha04.indd 86

5/11/2010 4:57:48 PM

What Do Structural Models Say About the Linkages?

87

vector of monetary policy shocks and shocks to real effective exchange rates. We can then write H0t  GH t ,

(4.19)

where G is a non-singular k  k matrix with elements 0 or 1. Correspondingly, 0  («t0 «t0)   ,. In calculating the IRF and FEVD the covariance matrix is usually orthogonalised to make it diagonal, by various identifying assumptions, whereas in our case while the within country covariance matrix of the three structural shocks is diagonal, the covariance matrix of the system 0 as a whole, ␧, is not. We discuss its structure below. The system is solved in terms of the k  1 vector ~ xt, which since ~ ~ ~ ~ ~ ~ ~ x0 t = (y 0 t, r 0 t, ep 0 t)( y 0t, r 0t, p 0t) does not include US inflation. To compute the effects of shocks on US inflation we need to work ~ . with the (k  1)  1 vector x t, which includes US inflation where ~ . ~ ~ ~ ~ ' x 0t = (π 0 t,y 0 t, r 0 t, ep 0 t) . This can be achieved by using (4.10). We can then consider the effect of particular shocks, say t  a 0t on a com~ . posite variable qt bx 0t. The k  1 vector a and the (k  1)  1 vector b are either appropriate selection vectors picking out a particular error or variable or weighting vectors either representing composite shocks, such as a global supply shock, or composite variables such as PPP GDP weighted averages of the variables for the eight euro area countries or effects on the real effective exchange rate. The IRFs provide the time ~ . profile of the response by qt  bx 0t to a unit shock to t  a 0t and the FEVDs provide the proportion of the variance of the n-step forecast errors of qt+n which is explained by conditioning on the shocks « 0jt, « 0j,t1,..., « 0j,tn, for j=1,...,k. Using (4.18) and (4.19), we get 1 1 0 i t  x i i x 11 t1  12 x t2  H0 G H t ,

~ and the time profile of x tn in terms of the lagged shocks can be written 0 0 0 0 i tn  D x i i x n1 t1  Dn2 x t2  C n H t  C n1H t1 ... C1H tn1  C n H tn ,

(4.20) 1

where Dn1 and Dn2 are functions of ⌽11 and ⌽12, Cj  Aj P0, P0  H0 G1, where the Aj can be derived recursively as A j 11A j1 12 A j2 , with A0  Ik Aj  0 for j<0.

9780230_243231_05_cha04.indd 87

5/11/2010 4:57:49 PM

88 Catching the Flu from the United States

~ and x ~ and using (4.10) we can express the Then, initialising on x t1 t2 ~ . time profile of x tn in terms of the shocks as i tn  D i  x  i x n1 t1  Dn2 x t2  Bn ε t0 Bn1ε t01 ... B1ε t0n1  Bn ε t0n ,

(4.21)

where  S D S D D n1 0 n1 1 n1,1  S D S D D , n2

0

n2

n1,2

1

and BA  S0CA  S1CA1

for

ℓ 1,2,....,n,

B 0  S 0 C 0. Notice that the Bl matrices ℓ  0, 1, 2, ... ,n are each of dimension (k  1)  k, transmitting the k structural shocks to the (k  1) elements ~ . of x tn. 0 As noted above, unlike the usual orthogonalised case,  is not diago0 ' nal. Using the reordering «t  («st', «dt',«mt',«et') and allowing for the correlation of shocks of the same type across countries, then for consistency with the standard theoretical treatment, we will calculate IRFs and FEVD for the case where the covariance matrix has the following bordered structure:  ss  0  0    0   es

0 dd 0 ed

0 0 mm em

se  de  . me  ee 

This imposes the restriction that the structural shocks of different types are uncorrelated, but there is no restriction on the correlation of the exchange rate shocks with the structural shocks. Notice that unlike the GVAR, where the inclusion of the vector of x*it variables removes commonalities and leaves the between-country correlations close to zero, here because of the limited role of foreign variables in the structural equations the structural errors may be correlated. These correlations can be calculated, thus we are able to evaluate the realism of the covariance matrix assumptions above.

9780230_243231_05_cha04.indd 88

5/11/2010 4:57:49 PM

What Do Structural Models Say About the Linkages?

89

~ . The generalised impulse response function for the effect on qt  b' x t of a one standard error shock to t  a' 0t is then

) (

(

'i i tn j  a' 0 a , ᑤ  g ( n ) = E b' x t

ᒑ −1 − E b x tn  ᑤ ᒑ −1

b Bn  '

g (n) =

0

a'  0 a ,

) (4.22)

, n = 0,1,2,

While we can identify the IRFs of, say, supply shocks as a group because they are othogonal to demand and monetary policy shocks, we cannot identify the supply shock in any particular country, because they are correlated with the supply shocks in other countries. We consider global supply or demand shocks. For instance, a global supply shock, uses as, which has PPP GDP weights, which add to one, on the supply shocks of each of the N + 1 countries, zeros elsewhere. Similarly for the demand shocks. For a monetary policy shock, it is more difficult to envisage a global shock. We consider instead a unit (one standard error) unexpected increase in US interest rates for one period, with the simultaneous interest rate responses by all other countries as given by the covariance matrix. This is of some interest since there has been a large VAR based literature considering the effect of a US monetary shock on the rest of the world, including Eichenbaum and Evans (1995) and Kim (2001). This shock can either be interpreted as an orthogonalised impulse response function, where the US is ordered first in the interest rate block or a generalised impluse response function, they are identical. Using (4.21) we can calculate the contributions to the variance of each variable or composite variable, that comes from global supply, demand and monetary policy shocks. In order to identify the effects of particular types of shocks we can partition the Bl in (4.21) conformably with the partitioning of «t0 («st', «dt',«mt',«et')', as BA (Bs,A , Bd,A , Bm,A , Be,A ). Then i tn  D  x i  i x n1 t1  Dn2 x t2 

n

¦ ¦B 

j shocks A 0

j ,nA

j ,t A ,

(4.23)

and calculate the variance decompositions, which will not sum to one because of the covariances with exchange rates.

9780230_243231_05_cha04.indd 89

5/11/2010 4:57:50 PM

90 Catching the Flu from the United States

The first four terms give the contributions to the variance from each of the four shocks; the following six terms arise from the covariances between the exchange rate shocks and the three structural shocks. Using the above FEVD, one can then estimate the importance of supply shocks, demand shocks or monetary policy shocks in the world economy for the explanations of output growth, inflation, interest rates and the real effective exchange rates for any of the individual variables or any given linear combinations of the variables. These proportions would not add up to unity, due to the non-zero correlations between the real effective exchange rates and the three structural shocks, unless we calculate the variance decomposition using the block diagonal covariance matrix. In this case, the covariance terms drop out and the terms do sum to unity.

Appendix 4.4 Financial puzzles in structural macro-finance models and recent research in financial components of DSGE models10 The existence of several ‘puzzles’ in the standard theory of finance has hampered the role of financial variables in DSGE macro models, so much so that DSGE models are currently mostly based on real variables, although research is ongoing to include financial variables. In this appendix, we report on such attempts. The usual macro-DSGE models tend to have no role for financial variables other than the short-term (policy) interest rate. However, the theory of finance is based on the consumer’s behaviour, which is key for DSGE models. As Weitzman (2007) puts it ‘In principle, consumption based representative agent models provide a complete answer to all macroeconomic asset pricing questions and give a unified theory integrating the economics of finance with the real economy.’ The interaction between real and financial variables had been a central issue in macroeconomics. The reason that only real variables are used to evaluate DSGE macro models is that for many financial variables, the standard theory does not work, or only works with such long and variable lags as to be irrelevant to the policy issues (Smith, 2009). The fact that the theory does not work is reflected in the prevalence of ‘puzzles’: 1. The ‘equity premium puzzle’ (Mehra and Prescott, 1985) results from the difficulty in reconciling the much higher return on equity stock compared to government bonds according to the standard economic theory, unless individuals have implausibly high risk aversion.

9780230_243231_05_cha04.indd 90

5/11/2010 4:57:51 PM

What Do Structural Models Say About the Linkages?

91

2. Changing the form of the utility function may solve part of this issue, but would not help in explaining why the real risk-free rate is so low (the ‘risk-free rate puzzle’; Weil, 1989). 3. The standard deviation of equity returns is also much too large relative to the standard deviation of any fundamental that may drive them, whether we look at the standard deviation of the growth rate or of the dividends (the ‘excess volatility puzzle’; Shiller, 1981). It would appear that very high, and quite implausible, rates of risk aversion would be required to match these three empirical features of financial markets. In general, expectational models work poorly in areas related to the link between macro and finance. Ad hoc risk premia are needed to explain the upward slope of the yield curve (Backus et al., 1989; Grossman et al., 1987) or the risk premium in the uncovered interest parity. Long-term interest rates and exchange rates are more volatile than theory would suggest (e.g. Gallmeyer et al., 2007). It is also quite hard to get evidence of interest rates and inflation adjusting to some natural real rate of interest as the Fisher relation would suggest. Finally, the inter-temporal theory of this account based on DSGE is particularly productive in throwing up quantity-related puzzles that mirror the price-related puzzles just discussed. For instance, there is a greater home bias in investment than efficient diversification would suggest and the Feldstein-Horioka correlation between savings and investment seems to be evidence against capital mobility. Also, with full risk-sharing, consumption at home relative to abroad should increase when the domestic consumption basket is relatively cheap and therefore there should be a strong correlation between relative domestic consumption and the real exchange rate, which does not appear in the data (the Backus-Smith puzzle, see Benigno, 2007). Further research in the financial components of DSGE models is also on its way. First, the macroeconomic models are now endogenously deriving portfolio diversification, something that was not yet possible in the DSGE literature because the models were linearised in the (non-stochastic) steady state, where the absence of risk prevents any discussion on the existence of different assets and risk premia. The current research agenda associated with Devereux and Sutherland (2006a, 2006b) aims at solving this technical issue by approximating the models at the second order (or even at higher orders), where risk matters.

9780230_243231_05_cha04.indd 91

5/11/2010 4:57:51 PM

92

Catching the Flu from the United States

Extensions that borrow from the asset pricing literature (including incomplete markets, agent heterogeneity, asymmetric information, market power, etc.) will eventually be required before the DSGE models provide a decent fit for financial data. There have been some advances in this direction, although market imperfections have often been added in a fairly simple way, but it remains very difficult to improve the match with the data in more than a few dimensions. For instance Gomes et al. (2003) are able to generate a higher equity premium in a DSGE model using financing constraints, but at the cost of creating a procyclical financing premium, a result at odds with the data. The literature has also been increasingly interested in the financing constraints since Bernanke et al. (1999) included a model of the financial accelerator in a DSGE model. A major argument for including financing constraints is that they help in modelling the other direction of causality, from financial shocks to real variables. It is therefore necessary to add some imperfections to the model to account for the effects of financial markets on the real economy. Goodfriend and McCallum (2007) thus develop a DSGE model with a banking sector, which produces a number of distinct interest rates, and show a major quantitative effect of the banking sector. This also reduces the puzzles over the risk-free rate and the equity premium. Christiano et al. (2003) add several financial frictions that force entrepreneurs either to source finance from their own wealth or to borrow from a representative bank, which is the centre of all financial intermediation. Although these models go in the right direction, little has been done to explain the importance of the financial situation of firms in determining the transmission of business cycles across economies. Recent attempts to close the gap have focused on the role of interdependence across countries through linkages among financial institutions, in addition to the trade linkages. Krugman (2008) suggests that traditional multi-country business cycle models lack a critical ‘international finance multiplier’, by which financial shocks in one country affect investment both in the country of origin and in the rest of the world through financial or balance sheet linkages. Devereux and Yetman (2009) develop a model of the international transmission of shocks through de-leveraging across financial institutions. In a macro model in which highly levered investors hold interconnected portfolios across countries, they show that the presence of binding leverage constraints introduces a powerful financial transmission channel, which results in a high correlation among macroeconomic aggregates during business cycle downturns, quite independent of the size of international trade linkages.

9780230_243231_05_cha04.indd 92

5/11/2010 4:57:51 PM

What Do Structural Models Say About the Linkages?

93

Notes 1. Much of the methodological debate surrounding the various estimation techniques is summarised in papers by Kydland and Prescott (1996) and Hansen and Heckman (1996). 2. For more details, see Christoffel et al. (2008). 3. The technology shock group comprises the permanent technology shock, the transitory technology shock and the investment-specific technology shock. The demand shock group includes the shocks to the domestic risk premium, government consumption and import demand, while the markup shock group consists of the wage markup, domestic price markup and export price markup shocks. The monetary policy shock is represented by the innovation in the model’s interest rate rule. Finally, the foreign shock group comprises the external risk-premium shock, the export preference shock, the import price markup shock and the shocks to the foreign variables, including foreign demand. 4. Note in addition that the role of foreign shocks in domestic fluctuations would even be lower without the estimated correlations with the UIP shocks. 5. Since the model explains the exchange rate and the forward rate (from domestic and foreign interest rates), it implicitly defines the uncovered interest parity risk premium. 6. We experimented with including the price of oil in the system, but identification of oil shocks raises a variety of difficult issues, for example, Kilian (2008) and Smith (2009), and this is left for further research. Oil prices are included in the information set used to form expectations and to calculate deviations from steady state. 7. In principle, one could examine the IRF for an exchange rate shock, by suitable choice of a, but it is not clear what the appropriate definition of an exchange rate shock should be, so we do not pursue this. 8. See for instance Dees et al. (2007). 9. Notice that the structural shocks cannot be obtained from the IV residuals, since Et −1 (x t +1 ) is not known until the model is solved. 10. We are grateful to Ron Smith and Raphael Espinoza for this appendix.

References Adjemian, S., M. Darracq Pariès and F. Smets, 2008, ‘A Quantitative Perspective on Optimal Monetary Policy Cooperation between the US and the Euro Area’, European Central Bank Working Paper No. 884. Adolfson, M., S. Laséen, J. Lindé, and M. Villani, 2007, ‘Bayesian Estimation of an Open Economy Dsge Model with Incomplete Pass-Through’, Journal of International Economics, 72 (2), 481–511. Backus, D., A. Gregory and S. Zin, 1989, ‘Risk Premiums in the Term Structure: Evidence from Artificial Economies’, Journal of Monetary Economics, 24. Benigno, P., 2007, ‘Portfolio Choices with Near Rational Agents: A Solution to Some International Finance Puzzles’, Nber Working Paper 13173. Bergin, P., 2004, ‘How Well Can the New Open Macroeconomics Explain the Exchange Rate and Current Account?’ NBER Working Paper No. 10356.

9780230_243231_05_cha04.indd 93

5/11/2010 4:57:51 PM

94 Catching the Flu from the United States Bernanke, B., M. Gertler and S. Gilchrist, 1999, ‘The Financial Accelerator in a Quantitative Business Cycle Framework’, in J.B. Taylor and M. Woodford (eds), Handbook of Macroeconomics, 1341–93. Canzoneri, M.B., R.E. Cumby and B.T. Diba, 2007, ‘Euler Equations and Money Market Interest Rates: A Challenge for Monetary Policy Models’, Journal of Monetary Economics, 54, 1863–81. Carabenciov, I., I. Ermolaev, C. Freeman, M. Juillard, O. Kamenik, D. Korshunov, D. Laxton and J. Laxton, 2008, ‘A Small Multi-Country Projection Model with Financial-Real Linkages and Oil Prices’, IMF Working Paper /08/280. Chudik, A., M.H. Pesaran and E. Tosetti, 2009, ‘Weak and Strong Cross-Section Dependence and Estimation of Large Panels’, CESifo Working Paper 2689. Christiano, L., R. Motto and M. Rostagno, 2003, ‘The Great Depression and the Friedman-Schwartz Hypothesis’, Journal of Money, Credit and Banking, 35 (6), 1119–1197. Christoffel, K., G. Coenen and A. Warne, 2008, ‘The New Area-Wide Model of the Euro Area: A Micro-Founded Open-Economy Model for Forecasting and Policy Analysis’, European Central Bank Working Paper No. 944. Clarida, R., J. Galí and M. Gertler, 1999, ‘The Science of Monetary Policy: A New Keynesian Perspective’, Journal of Economic Literature, 37 (4), 1661–1707. De Walque, G., F. Smets and R. Wouters, 2005, ‘An Open Economy DSGE Model Linking the Euro Area and the U.S. Economy’, Mimeo. Dees, S., F. di Mauro, M.H. Pesaran and L.V. Smith, 2007, ‘Exploring the International Linkages of the Euro Area: A Global Var Analysis’, Journal of Applied Econometrics, 22 (1), 1–38. Dees, S., M.H. Pesaran, L.V. Smith and R.P. Smith, 2009, ‘Identification of New Keynesian Phillips Curves from a Global Perspective’, Journal of Money Credit and Banking, 41, 1481–1502. Dees, S., M.H. Pesaran, L.V. Smith and R.P. Smith, 2010, ‘Supply, Demand and Monetary Policy Shocks in a Multi-Country New Keynesian Model’, forthcoming. Preliminary version presented at the ECB workshop on ‘International linkages and the macroeconomy: applications of GVAR modelling approaches’, 9 December 2009, Frankfurt am Main, Germany. Del Negro, M. and F. Schorfheide, 2006, ‘How Good Is What You’ve Got? DsgeVar as a Toolkit for Evaluating Dsge Models’, Economic Review, Federal Reserve Bank of Atlanta, 2, 21–37. Devereux, M.B. and A. Sutherland, 2006a, ‘Solving for Country Portfolios in Open Economy Macro Models’, mimeo. Devereux, M.B. and A. Sutherland, 2006b, ‘Solving for Country Portfolio Dynamics in Open Economy Macro Models’, mimeo. Devereux, M.B. and J. Yetman, 2009, ‘Financial Deleveraging and the International Transmission of Shocks’, Prepared for the Jmcb Board of Governors Conference Financial Markets and Monetary Policy, June 4–5. Dornbusch, R., 1976, ‘Expectations and Exchange Rate Dynamics’, Journal of Political Economy, 84, 1161–76. Erceg, J., L. Guerrieri and C. Gust, 2006, ‘SIGMA: A New Open Economy Model for Policy Analysis’, International Journal of Central Banking, 2 (1), 1–50.

9780230_243231_05_cha04.indd 94

5/11/2010 4:57:51 PM

What Do Structural Models Say About the Linkages?

95

Favero, C.A. and R. Rovelli, 2003, ‘Macroeconomic Stability and the Preferences of the Fed: A Formal Analysis, 1961–98’, Journal of Money, Credit and Banking, Blackwell Publishing, 35 (4), 545–56. Faust, J. and J.H. Rogers, 2003, ‘Monetary Policy’s Role in Exchange Rate Behaviour’, Journal of Monetary Economics, 50, 1403–24. Gali, J. and M. Gertler, 1999, ‘Inflation Dynamics: A Structural Econometric Analysis’, Journal of Monetary Economics, 44 (2), 195–222. Gali, J. and T. Monacelli, 2005, ‘Monetary Policy and Exchange Rate Volatility in a Small Open Economy’, Review of Economic Studies, 72 (3), 707–34. Gallmeyer, M.F., B. Hollifield, F. Palomino and S.E. Zin, 2007, ‘Arbitrage Free Bond Pricing with Dynamic Macroeconomic Models’, NBER Working Paper 13245. Ghironi, F., 1999, ‘Towards New Open Economy Macroeconometrics’, Federal Reserve Bank of New York Staff Reports 100. Gomes, J.F., A. Yaron and L. Zhang, 2003, ‘Asset Prices and Business Cycles with Costly External Finance’, Review of Economic Dynamics, 6 (4), 767–88. Goodfriend, M. and B.T. McCallum, 2007, ‘Banking and Interest Rates in Monetary Policy Analysis: A Quantitative Exploration’, NBER Working Paper 13207. Grossman, S., A. Melino and R. Shiller, 1987, ‘Estimating the Continuous Time Consumption Based Asset Pricing Model’, Journal of Business and Economic Statistics. Hansen, L.P. and J.J. Heckman, 1996, ‘The Empirical Foundations of Calibration’, Journal of Economic Perspectives, 10 (1), 87–104. Kilian, L., 2008, The Economic Effects of Energy Price Shocks', Journal of Economic Literature, 46, 871–909. Kim, S., 2001, ‘International Transmission of US Monetary Policy Shocks: Evidence from VAR’s’, Journal of Monetary Economics, 48, 339–72. Krugman, P., 2008, ‘The International Finance Multiplier’, mimeo, Princeton University. Kydland, F.E. and E.C. Prescott, 1996, ‘The Computational Experiment: An Econometric Tool’, Journal of Economic Perspectives, 10 (1), 69–85. Lubik, T.A. and F. Schorfheide, 2005, ‘A Bayesian Look at New Open Economy Macroeconomics’, NBER Macroeconomics Annual 20, 313–66. Lubik, T.A. and F. Schorfheide, 2007, ‘Do Central Banks Respond to Exchange Rate Movements? A Structural Investigation’, Journal of Monetary Economics, 54, 1069–87. Lucas, R.E., 1978, ‘Asset Prices in an Exchange Economy’, Econometrica, 46 (6), 1429–45. Mehra, R. and E.C. Prescott, 1985, ‘The Equity Premium: A Puzzle’, Journal of Monetary Economics, 15 (2), 145–61. Pesenti, P., 2008, ‘The Global Economy Model: Theoretical Framework’, IMF Staff Papers, 55 (2), 243–84. Shiller, Robert J., 1981, ‘Do Stock Prices Move Too Much to be Justified by Subsequent Changes in Dividends?’ American Economic Review, American Economic Association, 71 (3) (June), 421–36. Smets, F. and R. Wouters, 2003, ‘An Estimated Dynamic Stochastic General Equilibrium Model of the Euro Area’, Journal of the European Economic Association, 1 (5), 1123–75.

9780230_243231_05_cha04.indd 95

5/11/2010 4:57:51 PM

96

Catching the Flu from the United States

Smets, F. and R. Wouters, 2007, ‘Shocks and Frictions in US Business Cycles: A Bayesian DSGE Approach’, American Economic Review, 97, 586–606. Smith, R.P., 2009, ‘Real-Financial Interactions in Macro-Finance Models’, Quantitative and Qualitative Analysis in Social Sciences, 3 (1), 1–20. Weil, Philippe, 1989, ‘The Equity Premium Puzzle and the Risk-Free Rate Puzzle’, Journal of Monetary Economics, 24 (3), 401–21, Elsevier. Weil, Philippe, 1992, ‘Equilibrium Asset Prices with Undiversifiable Labor Income Risk’, Journal of Economic Dynamics and Control, 16, 769–90. Weitzman, M.L., 2007, ‘Subjective Expectations and Asset Return Puzzles’, American Economic Review, 97 (4), 1102–30.

9780230_243231_05_cha04.indd 96

5/11/2010 4:57:51 PM

5 The United States and the Euro Area: The Role of Financial Variables

As we have seen before, the traditional analysis of the transmission of shocks views the trade channel as the main source of spillovers: a slowdown in the US would decrease its imports, and the associated reduction of European exports would therefore lead Europe to a period of lower growth. However, this direct trade channel can hardly account for the extent of observed spillovers, particularly as the 2007–9 global financial turmoil’s impacts on the euro area are still far from being settled. Looking at the euro area, US imports represent around 15% of the former’s exports, and the euro area exports contribute only 10% of its GDP growth. The stylised fact that the euro area lags US business cycles by a few quarters, which was discussed in detail in Chapter 2, can therefore be hardly justified, on account of rather limited trade openness. Hence, the literature has concentrated on the financial sector as the possible missing element in the analysis of the channels of transmission, though, as we have seen in Chapter 4, incorporating financial variables in the analysis of international linkages is not a trivial matter. In this chapter, after providing reasoning and stylised facts on financial linkages, we undertake a more systematic attempt to incorporate financial variables as additional explanatory factors of business interaction between the US and the euro area. Domestic financing conditions of firms and individuals are relevant elements in explaining real developments in a given economic area. The reasoning behind this claim is twofold. First, tighter financial and credit conditions limit the potential for firms’ activity to expand, constraining their hiring and investment decisions (see Bloom, 2009; Bloom et al., 2008). Furthermore, they prevent credit-constrained 97

9780230_243231_06_cha05.indd 97

5/20/2010 10:39:20 AM

98 Catching the Flu from the United States

households from borrowing and so consuming during harsh times, thus restraining the possibility of smoothing consumption. As a result of financial globalisation, financing conditions in a major economy such as the United States become relevant beyond its borders because of the required equalisation of expected returns.1 The banking systems of the United States and the euro area have moved steadily towards an increasing interdependence: external assets and liabilities of euro area resident banks vis-à-vis US banks have more than quadrupled since 1997, and despite the recent deleveraging process still amount to about USD 1 trillion (Figure 5.1). Also stock market indices nowadays display a much tighter interdependence than was the case in the past: the five-year moving correlation of the Dow Jones Industrial index and the Eurostoxx (Figure 5.2) increased from roughly 0.8 during the early 1990s to 0.99 in 2009. Similar considerations hold if one looks at corporate bond yields (see Figure 5.3): here it appears that, after a period of weak co-movements at the beginning of the sample, correlation has markedly increased, especially since 2005. The literature has provided evidence of international financial transmission. Ehrmann et al. (2005) analyse the nature of financial integration and the transmission channels both within and between the United States and the euro area. Their results underline the importance of international spillovers, both within asset classes and across financial markets. Although the strongest international transmission of shocks

1200 1000 800 600 400 200 0 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Assets

Liabilities

Figure 5.1 External assets and liabilities of euro area resident banks vis-à-vis the US (amounts outstanding, in USD billions) Note: Latest observation: 2008Q1. Sources: BIS and ECB calculations.

9780230_243231_06_cha05.indd 98

5/20/2010 10:39:20 AM

The Role of Financial Variables 6,000

99

16,000 14,000

5,000

12,000 4,000

10,000

3,000

8,000 6,000

2,000

4,000 1,000

2,000

0 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 Euro Area

0

US

Figure 5.2 US and euro area stock market indices Note: Last observation is mid-December 2009. Source: Global financial data.

12 10 8 6 4 2 0 1990

1992

1994

1996

1998 US

2000

2002

2004

2006

2008

Germany

Figure 5.3 US and euro area corporate bond yields Note: Last observation is December 2009. Source: Global financial data.

9780230_243231_06_cha05.indd 99

5/20/2010 10:39:21 AM

100 Catching the Flu from the United States

takes place within asset classes, Ehrmann et al. (2005) find evidence that international cross-market spillovers are significant, both statistically and economically. For instance, shocks to US short-term interest rates exert a substantial influence on euro area bond yields and equity markets, and in fact explain as much as 10% of overall euro area bond market movements. Interestingly, the transmission of shocks also runs in the opposite direction as, in particular, the short-term interest rates of the euro area have a significant impact on US bond and equity markets. Overall, US financial markets explain on average more than 25% of euro area financial market movements in the period 1989–2004, whereas euro area markets account for 8% of the variance of US asset prices. Given the increasing financial co-movements and spillovers, one can envisage that the financial channel plays a relevant role in explaining transatlantic real developments. Before tackling the issue empirically, however, we will need first to define how to track financial shocks hitting an economy, and how to separate them from plain innovations to financial variables in a VAR setting. Box 5.1 Developments in euro area and US bank interest rates during the 2007–9 crisis The transmission across the Atlantic of bank interest rates is a particular aspect of the whole set of different financial linkages, but it has a specifically relevant role for monetary policy, given that in order to enable the monetary policy transmission to work effectively, it is crucial that changes in policy rates influence bank lending rates for households and corporations. In this box, we compare developments in the cost of lending from banks in the euro area with those in the United States, focusing both on levels of bank interest rates and their spreads with respect to market rates for both non-financial corporations and households. Looking at the components of the cost of bank financing in nominal terms and starting with the euro area, short-term MFI interest rates on loans to nonfinancial corporations with floating rates and an initial fixed-rate period of up to one year decreased by around 210 basis points between June 2007 and April 2009 in the euro area (see Figure 5.4a). Long-term MFI interest rates on loans to non-financial corporations with an initial fixed-rate period of over five years declined somewhat less, by 105 basis points in the same period. It should be noted, however, that the biggest share of these declines was recorded in recent months especially since October 2008, when bank interest rates in the euro area reached their historical maximum since January 2003. Thus, interest rates on loans declined almost in parallel with the key policy rate. Household MFI rates in the euro area for consumer credit purposes and for house purchases have followed a broadly similar trend as the interest rates for non-financial corporations. Short-term MFI interest rates on loans to households with floating rates and an initial fixed-rate period of up to one year declined by almost 150 basis points between June 2007 and April 2009

9780230_243231_06_cha05.indd 100

5/20/2010 10:39:21 AM

The Role of Financial Variables

(a)

101

6.5 6.0 5.5 5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 Jan-03 Oct-03

Jul-04

Short-term

Apr-05 Jan-06 Oct-06

Long-term

3m EURIBOR

Jul-07

Apr-08 Jan-09

5-yr gov. bond yield

Figure 5.4a MFI interest rates on new loans to non-financial corporations in the euro area and market interest rates (% per annum, monthly data)

(b) 6.5 6.0 5.5 5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 Jan-03 Oct-03 Jul-04 Apr-05 Jan-06 Oct-06 Jul-07 Apr-08 Jan-09 Short-term

Long-term

12m EURIBOR

10-yr gov. bond yield

Figure 5.4b MFI interest rates on new loans to households in the euro area and market interest rates (% per annum, monthly data) Note: The 5-year government bond yield is more comparable for the euro area longterm lending rates. Last observation is May 2009. Sources: ECB and Reuters.

9780230_243231_06_cha05.indd 101

5/20/2010 10:39:22 AM

102

Catching the Flu from the United States

(see Figure 5.4b). Long-term MFI interest rates on loans to households with an initial fixed-rate period of over five years declined only marginally, by 15 basis points between June 2007 and April 2009. In the United States average bank lending interest rates have declined since the FOMC started reducing its policy rate in September 2007. Since then, interest rates on commercial and industrial (C&I) loans have fallen to approximately 2% for short-term and 3.60% for long-term loans (2009Q1), declining more than 460 and 420 basis points respectively when compared with mid-2007 levels (see Figure 5.4c). Turning to mortgage rates in the United States, while the fixed rate on longterm conventional mortgages has declined substantially, interest rates on ‘jumbo’ mortgages (bigger loans ineligible for a Fannie Mae or Freddie Mac guarantee) have remained at higher levels as investors have retreated from this mortgage market segment since the start of the financial turmoil (see Figure 5.4d). Only recently have some declines in the jumbo mortgage rate been observed. All the figures above depict movements in bank interest rates and market interest rates. Developments in banks’ short-term funding costs are in normal times primarily affected by movements in the three-month EURIBOR in the case of the euro area and the three-month Libor in the case of the United States, while banks’ long-term lending rates normally reflect movements in government bond yields. Since June 2007, market interest rates have displayed slightly different trends in these economies. Short-term money market rates in the euro area have decreased by around 270 basis points for the three-month EURIBOR while five-year government bond yields declined by around 160 basis points between June 2007 and April 2009. The declines observed since October 2008, however account for the largest decreases in the overall period. In the United States the three-month Libor rate and the two-year government (c) 8.0 7.0 6.0 5.0 4.0 3.0 2.0 1.0 0.0 Mar-03 Dec-03 Sep-04 Jun-05 Mar-06 Dec-06 Sep-07 Jun-08 Mar-09 Short-term bank rate 2yr gov bond yield

Long-term bank rate 3m Libor rate

Figure 5.4c Commercial and industrial loans interest rates in the United States and market interest rates (% per annum, quarterly data)

9780230_243231_06_cha05.indd 102

5/20/2010 10:39:22 AM

The Role of Financial Variables

(d)

103

8 7 6 5 4 3 2 1 0 Mar-03 Dec-03 Sep-04 Jun-05 Mar-06 Dec-06 Sep-07 Jun-08 Mar-09 30 year jumbo mortgage rate USD 6M Libor

30 year fixed conventional mortgage rate 30 year Treasury bond

Figure 5.4d Mortgage rates in the United States and market interest rates (% per annum, monthly data) Notes: The loan rates are interest rates on commercial and industrial loans made by all commercial banks. For the US the 2-yr government bond yields matches better the duration of the loans. Last observation is March 2009. Sources: Federal Reserve: ‘Survey of Terms of Business Lending’, Haver and Reuters, Primary Mortgage Market Survey, Bankrate.com, Bloomberg.

bond yield declined by approximately 400 basis points (against a decline of 500 basis points in the federal funds policy rate) in the same period. These developments in euro area and US bank lending rates and the relative spreads against market rates provide useful evidence on the functioning of the interest rate pass-through mechanism in an environment of low interest rates. The evidence for both economies shows a somewhat weaker pass-through in lending rates to households than to non-financial corporations. For example, short-term interest rate spreads for non-financial corporations in the euro area have increased by around 65 basis points while interest rate spreads for households have increased for both short and long-term lending rates since June 2007 by between 125 and 140 basis points respectively (see Figure 5.4e). In the United States, the spread between US short-term C&I (Commercial and Industrial) loans and the three-month Libor rate has tightened by 50 basis points during the last easing cycle. Long-term interest rate spreads for non-financial corporations have tightened slightly in the US by 30 basis points relative to two-year government bond yields (see Figure 5.4f). As regards mortgage lending, the decline in US fixed conventional mortgage rates has been evident only since the announcement of the Federal Reserve’s purchases of mortgage-related assets in November 2008, while the passthrough to mortgage interest rates was very limited during the early phases of policy interest rate reductions (from September 2007 to November 2008).

9780230_243231_06_cha05.indd 103

5/20/2010 10:39:23 AM

104

Catching the Flu from the United States

(e) 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 Jan-03 Oct-03

Jul-04 Apr-05 Jan-06 Oct-06 Jul-07 Apr-08 Jan-09

Short-term non-financial corporations Short-term households

Figure 5.4e terly data)

Long-term non-financial corporations Long-term households

MFI interest rate spreads in the euro area (% per annum, quar-

(f) 5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 Mar-03 Dec-03 Sep-04 Jun-05 Mar-06 Dec-06 Sep-07 Jun-08 Mar-09 Spread: C&I short – 3m Libor Spread: C&I long – 2yr govt bond Spread: 30 year fixed mortgage rate – 30-yr govt bond

Figure 5.4f Mortgage and loan interest rate spreads in the United States (% per annum, monthly data) Note: Last observation is March 2009. Sources: ECB and Reuters. Primary Mortgage Market Survey, Federal Reserve, Bloomberg.

9780230_243231_06_cha05.indd 104

5/20/2010 10:39:23 AM

The Role of Financial Variables

5.1

105

What’s a financial shock?

Smooth and efficient access to credit financing is key to the functioning of any economic system. In a broad sense, examining financial variables should give an idea of how and to what extent firms and households have access to credit and/or can profitably invest their wealth. So, easier access to credit and better opportunities to raise capital should lead to economic growth. There are several financing channels which are relevant for this effect, and thus several financial variables that are worth monitoring, e.g. developments in stock markets, corporate bond spreads and bank lending rates. Yet, plainly, looking at the developments of the variables would be misleading, as they are all in principle influenced by real developments and, most importantly, by the monetary policy setting. From a policy-making perspective, it is therefore important to insulate endogenous developments in the financing sector from the impact of exogenous factors such as monetary policy and real shocks. We will term the former as ‘financial shocks’. Under this definition, turbulence in financial markets resulting from the burst of bubbles or the unwinding of imbalances, as well as abrupt changes in risk appetite and market liquidity that lead to portfolio reallocations, can be labelled as ‘financial shocks’. A device that enables us to catalogue the different types of shocks hitting the economy is therefore extremely relevant in that it allows the policy-maker to take the appropriate measures to stabilise the economy. In the specific case of a financial shock, the main task of policy-makers could be to restore confidence and stabilise financial and credit markets. Identifying financial shocks is, however, far from being straightforward. Confining our attention in the framework of structural VARs2, the main problem is that there is no consensus on the restrictions needed to identify shocks. Adopting a traditional Choleski ordering à la Sims (1980) has serious shortcomings not only for what concerns the sensitivity of the results to the ordering, but also because only short-run restrictions affecting the contemporaneous behaviour of the variables can be imposed. Instead, the identification of some types of shocks requires assumptions on the responses of the variables in the long run. For example, in identifying monetary policy shocks it is common in the literature to assume that they have no long-run impact on output (Walsh, 1993). Combining long- and short-run restrictions, as pioneered by Blanchard and Quah (1989), Galì (1992) and Gerlach and

9780230_243231_06_cha05.indd 105

5/20/2010 10:39:24 AM

106

Catching the Flu from the United States

Smets (1995), is a good strategy for incorporating a set of reasonable identifying assumptions. In the following, we will employ an identification scheme based on long- and short-run restrictions similar to that employed by Peersman (2005), which employs a VAR featuring four variables: oil prices, US gross domestic product, US consumer price inflation and the US short-term nominal interest rate. The structural shocks are identified as follows: an oil shock is assumed to have contemporaneous impact on all variables, a supply shock only has a long-run impact on aggregate output (implying that the long-run Phillips curve is vertical), a demand shock has only short-run impact on output and a monetary policy shock has no contemporaneous impact on output.3 We will use the same variables as in Peersman (2005), but complemented with an index of volatility in financial markets, which we use, as in Bloom (2009), to proxy the financial shock.4 For the financial shock, we use the identifying assumption that it has no short-run impact on GDP and inflation and no long-run impact on oil prices, inflation and the short-term rate.5 Shocks identified using the assumptions above are displayed in Figure 5.5. We can identify in particular the oil price shocks of 1973 and the (negative and then positive) shock associated to the recent rise and fall in oil prices in 2008. Looking at the most recent period, it also becomes apparent that the US economy was hit by an unusually long series of negative financial shocks. Monetary policy reacted promptly from the beginning of the recession, but failed to avoid the sequence of financial shocks to finally materialise into negative supply and demand shocks that led to the abrupt GDP contraction. Identifying and reconstructing a series of structural shocks is also relevant in that it allows us to perform an historical decomposition of the observed variables into their contributing elements, that is, to attribute to each structural shock its contribution to realised outcomes. Looking at the historical decomposition of US GDP according to the identification above (cf. Figure 5.6) reveals a mixed pattern for financial shocks, in the sense that it is hard to spot a systematic contribution: they seem to have played a role during the 2001 recessionary episode and, to a smaller extent, in 2008Q3.

5.2

Measuring financial spillovers

The results presented in the previous section relate to the identification of financial shocks in the United States and their impact on its GDP.

9780230_243231_06_cha05.indd 106

5/20/2010 10:39:24 AM

The Role of Financial Variables (a)

107

Oil 4 3 2 1 0 −1 −2 −3 −4 1965 1969 1973 1977 1981 1985 1989 1993 1997 2001 2005

(b)

Supply 4 3 2 1 0 −1 −2 −3 −4 1965 1969 1973 1977 1981 1985 1989 1993 1997 2001 2005

(c)

Demand 5 4 3 2 1 0 −1 −2 −3 −4 1965 1969 1973 1977 1981 1985 1989 1993 1997 2001 2005

(d)

Monetary 6 4 2 0 −2 −4 1965 1969 1973 1977 1981 1985 1989 1993 1997 2001 2005

Figure 5.5 Continued

9780230_243231_06_cha05.indd 107

5/20/2010 10:39:24 AM

108 Catching the Flu from the United States (e)

Financial 4 3 2 1 0 −1 −2 −3 1965 1969 1973 1977 1981 1985 1989 1993 1997 2001 2005

Figure 5.5 Structural shocks Note: The structural shocks are identified following the strategy detailed in Peersman (2005). The financial shock is identified by assuming that it has no short-run impact on GDP and inflation and no long-run impact on oil prices, inflation and the short-term rate. Last observation is December 2008. Source: ECB staff calculations.

0.015

0.02

0.01

0.015 0.01

0.005

0.005 0

0

−0.005

−0.005

−0.01

−0.01

−0.015

−0.015

−0.02 2000

−0.02 2001

2002

2003

2004

2005

2006

2007

2008

Supply

Figure 5.6

Historical decomposition of US GDP growth contributions

Note: Last observation is December 2008. Source: ECB staff calculations.

The relevant question for the purpose of this chapter, however, is how financial shocks transmit across the Atlantic and what role they play in the transmission mechanism of real developments. To assess this, as we have said, it is not sufficient to observe an increase in the correlation of financial variables: this indeed indicates higher financial integration, but does not tell us anything on the real impacts

9780230_243231_06_cha05.indd 108

5/20/2010 10:39:24 AM

The Role of Financial Variables

109

of financial links. What is needed instead is to show that higher financial links are also connected with higher real spillovers across different countries. In order to perform such analysis formally, we resort to the spillover index recently proposed by Diebold and Yilmaz (2009) and Yilmaz (2009). The index is built in the setting of a multi-country VAR model, comprising two variables per country, that is, industrial production as a proxy of real activity and stock market returns as synthetic proxy of financial variables.6 The index derives from the variance decomposition of the forecast errors made when using the model to predict industrial production. The underlying intuition is that if real spillovers across countries are negligible, what matters for an explanation of the growth profile of each individual country is its own, idiosyncratic-domestic factors. Put more formally, in the absence of spillovers, the variance-covariance matrix of forecast errors for industrial activity growth should be dominated by its diagonal elements; large off-diagonal elements would, however, indicate the presence of considerable spillover effects at the chosen horizon. The total spillover index is simply defined as the sum of the off-diagonal elements of the k-periods-ahead forecast error variance decomposition matrix divided by the sum of all of its elements. The index ranges therefore between zero and one, and represents the proportion of the k-period-ahead variance of the system that stems from non-idiosyncratic elements, that is, from spillover effects. A time-varying measure of spillover can be built by estimating the VAR on rolling samples and calculating the index for each of such samples. Directional net spillovers, that is, the difference between the amount of shocks transmitted and received by each country, can also be easily computed by isolating, in turn, the rows and the columns (i.e. the countries seen as either originators or receivers of shocks) of the forecast error variance decomposition matrix. We compute net financial spillovers in a VAR system including industrial production and stock market returns7 for four main economic areas: the United States, the euro area, Japan and the United Kingdom.8 Looking at net financial spillovers (Figure 5.7), the United States stands out as the main source of shocks, whereas the euro area has almost always been a strong net importer. Remarkably, the euro area started ‘importing’ financial spillovers again from mid-2005, after having enjoyed a relatively long period of ‘insulation’ around the turning of the century.

9780230_243231_06_cha05.indd 109

5/20/2010 10:39:25 AM

110

Catching the Flu from the United States

(a)

United States

80 60 40 20 0 −20 −40 −60 1965 1969 1973 1977 1981 1985 1989 1993 1997 2001 2005 (b)

Euro area

30 20 10 0 −10 −20 −30 −40 −50 −60 1965 1969 1973 1977 1981 1985 1989 1993 1997 2001 2005 Figure 5.7

Net financial spillovers

Note: The net financial spillovers are computed in a VAR system including industrial production and stock market returns for four main economic areas: the United States, the euro area, Japan and the United Kingdom. The net spillover is the difference between the spillover from a given country to the remaining three and the spillover from the three other countries to the selected country. Shaded areas are US recessions, as defined by the NBER. Last observation is February 2009. Source: ECB staff calculations.

This finding is in line with the general notion that developments in US financial markets quickly propagate to other markets. The relevant question here is, however, to what extent such financial spillovers can explain real developments in the two economic areas. To answer this question, we can compare the extent of real spillovers, as computed in a model excluding financial variables, with those obtained by including (and netting out) the impact of financial spillovers. Results (cf. Figure 5.8) suggest that, when the role of financial spillovers is neglected (the black area), United States appears to have been an exporter of real

9780230_243231_06_cha05.indd 110

5/20/2010 10:39:25 AM

The Role of Financial Variables (a)

111

United States

60 40 20 0 −20 −40 −60 1979

1983

1987

1991

(b)

1995

1999

2003

2007

1999

2003

2007

Euro area

60 40 20 0 −20 −40 −60 1979

1983

1987

1991

1995

Figure 5.8 Net real spillovers Note: Last observation is February 2009. Source: ECB staff calculations.

spillovers for the vast majority of the time since 1980, and the euro area appears to have been net importer, especially in the period following 2000. Once we net out financial spillovers (the grey area), real spillovers are much less stable and display much more time variability. Indeed, the United States appears to have been a net exporter of real spillovers during only some periods of time. During recent months, however, as the real impacts of the crisis intensified, the United States ceased to be an importer of real shocks and actually became a net exporter. The euro area has heavily imported real shocks during the last decade, and after a short-lived period of insulation in the months before the crisis erupted, it again became net importer in the latest months of the sample.

9780230_243231_06_cha05.indd 111

5/20/2010 10:39:26 AM

112

5.3

Catching the Flu from the United States

Do financial variables anticipate real developments?9

In the previous sections we have explained and documented how financial developments affect real variables and their impact on the transmission channel of spillovers across the Atlantic. As we anticipated in the introduction, there is, however, also a less structural explanation of the relevance of financial variables in explaining real developments: asset prices are determined in markets which are fundamentally forward looking. Equity prices incorporate expectations on firms’ profitability, which is linked to the future rate of growth of the economy. Hence, the correlation between depressed financial markets today and low growth in the near future could result from the forward-looking nature of financial markets, even in the absence of any causal link between financing conditions and growth. This argument is associated with an important branch of the literature analysing the usefulness of financial data in forecasting GDP cycles. While several authors have argued that financial variables do not consistently help in predicting business cycles (e.g. Stock and Watson, 2003), Ang et al. (2006), among others, show that in a model that takes into account expectations for GDP and no-arbitrage conditions, yields have a non-negligible role in forecasting growth. In this section we will tackle the issue of the information content of financial variables in anticipating real developments. To do so, we put ourselves in a forecasting perspective, and estimate a number of VAR models, constructed around the GDPs of three countries/regions (the US, the euro area and a seven-country aggregate called the Rest of the World) and extended to include selected financial variables and look at their performance in predicting real activity. 5.3.1 The data Our measure of real activity in the three economic areas is the seasonally adjusted quarterly real GDP observed between 1970Q1 and 2007Q4. The Rest of the World is an aggregation of seven countries (Australia, Canada, Denmark, New Zealand, Norway, Sweden and Switzerland) chosen to represent economies different enough that a shock affecting all of them can indeed be interpreted as ‘global’, as suggested by Bayoumi and Swiston (2007).10 We build the Rest of the World as a weighted average of the seven countries (with weights being the 1995 GDP expressed in US dollars – the results were similar with unweighted averages). Coming to financial variables, we collect stock market indices and dividend yields as well as ten-year and three-month yields for all of the

9780230_243231_06_cha05.indd 112

5/20/2010 10:39:26 AM

The Role of Financial Variables

113

countries in the sample. The dividend yield is employed to construct a measure of the disequilibrium between the stock market and the bond market, obtained from the cointegrating vector between the dividend yield and the Government bond yield (this relationship is sometimes referred to as the ‘Fed’ model).11 We use stock market returns to generate time-varying stock market volatilities as four-quarter backward-looking moving averages of their absolute values. The slope of the yield curve is the difference between the ten-year government bond and the threemonth T-Bill rates. For the euro area, the stock market is reconstructed from Global Financial Data while the slope of the yield curve relies on German data. For each country, all of the financial variables have been collected at a monthly frequency, between March 1970 and March 2008.12 These monthly financial variables are used primarily to assess the information content of the financial data at an intra-quarter level. To keep the size of the VAR within a reasonable limit, and thus reduce parameter uncertainty, we only include two financial variables per economic area in turn (i.e. our largest VAR therefore has nine variables).

Box 5.2 Euro area sovereign bond spreads and intra-US state bond spreads during the 2007–9 crisis – are their developments comparable? The 2007–9 financial turmoil led to a massive widening of financial spreads. This phenomenon did not affect only spreads and risk premia of assets issued by corporations, but also the yield differences of bonds issued by different governments. In the euro area, sovereign bond spreads (vis-à-vis German bonds) increased considerably and have been showing significant signs of abatement only since March 2009. However, such a differentiation of yields on public debt titles in a currency area is by no means exclusively a euro area phenomenon; the yield spreads of US state bonds (vis-à-vis US Treasury bonds) also showed remarkable increases during the crisis. Can developments on the other side of the Atlantic serve as a useful yardstick for comparison with euro area bond spread dynamics? This box explains why some caution is warranted in such a comparison and explores the developments for the two families of spreads in a synoptic fashion. At end-June 2007, that is, before the start of the turmoil, euro area government bond spreads were moderate (except in Malta, at that time not yet a euro area member state) while those of the US states were all deeply negative (see Figure 5.9a). Before explaining the striking sign difference, it is important to look somewhat closer at the construction of the two sets of bond spreads. Euro area bond spreads represent the difference of the ten-year bond yield issued by the respective member state and a ten-year bond issued by Germany. For the US states, in contrast, the state bond

9780230_243231_06_cha05.indd 113

5/20/2010 10:39:26 AM

114

Catching the Flu from the United States

80

80

60

60

40

40

20

20

0 MD SC OH NC CT GA MN NJ MI MA FL IL WI TX NY W CA NY CY SI IE DE FI NL FR AT ES BE SK PT GR IT MT

−20

0

−20

−40

−40

−60

−60

−80

−80

−100

−100

Figure 5.9a Ten-year government bond spreads of major euro area countries over German bonds (black) and US bond spreads of individual states over US treasury bonds (grey) (basis points): Levels of 29 June 2007 Sources: Datastream, Bloomberg, ECB calculations.

yields are derived from an index of several general obligation bonds issued by US state and local governments (such as municipalities) with approximately ten-year maturities, where all bonds in the index have the same average credit rating as the rating of the state. From the so-constructed state bond yield, the yield of a ten-year Treasury bond is subtracted. Besides this technical difference, there is an asymmetry regarding the relative position of the issuer of the respective benchmark bond to the issuers of the other bond. For the United States, the benchmark bond is issued by the Treasury, that is, the sovereign issuer, while the state bonds are issued by sub-sovereign entities. In the euro area, in contrast, the issuer of the benchmark bond (Germany) and the issuers of the other bonds are all sovereign states. The benchmark status of Germany among the set of countries arises because German bonds feature both the lowest perceived credit risk and very high liquidity.13 A second caveat to the cross-Atlantic comparison is the tax treatment of US state and local government bonds. Their interest payment is generally exempt from federal income taxes and – if the bonds are held by an investor resident in the state of issuance – state and local income taxes. As a result, interest paid on bonds issued by US states is usually smaller compared with fully taxable bonds. Accordingly, their spreads are usually negative vis-à-vis federal government debt: in the decade before the start of the current turmoil, the median ten-year yield on US state debt averaged about 260 basis points below comparable Treasuries. A similarly privileged tax treatment

9780230_243231_06_cha05.indd 114

5/20/2010 10:39:26 AM

The Role of Financial Variables

115

does not apply to euro area government bonds. Accordingly, even before the turmoil, bond spreads were typically positive. The latter two points suggest that comparisons of US and euro area government bond spreads should be conducted with some caution. In particular, a cross-Atlantic comparison of turmoil-related spread developments should rather focus on the dynamics of bond spreads than on their levels. Regarding the dynamics, sovereign and state bond spreads show a distinct co-movement for the time between July 2007 and end-2008 (see Figure 5.9b). The correlation of the medians of the two spread families amounts to 0.95 for the time between July 2007 and December 2008.14 In fact, since the current crisis is taking place globally, it may be expected that both euro area government spreads and US state spreads are driven by a common ‘global crisis factor’, capturing a deteriorating global macroeconomic outlook as well as the effect of changing investors’ risk aversion. In 2009Q1, the developments of the two families of spreads started to diverge markedly. While the US state bonds showed a strong decline in overall spread levels, euro area spreads edged further upward. The median spread on US states narrowed by approximately 100 basis points, which resulted from declining yields on state debt and rising yields on Treasuries. Over the same period, the median spread in the euro area countries increased slightly further.

300

160

250

140

200

120

150

100

100

80

50

60

0

40

−50

20

−100 Jul-07 Oct-07 Jan-08 Apr-08 Jul-08 Oct-08 Jan-09 Apr-09 Jul-09 Oct-09 Median US (5 day av)

0

Median EA (5 day av, right axis)

Figure 5.9b Median of US state spreads and euro area government bond spreads (basis points; 5-day averages) Note: Daily data, July 2007 to 08 December 2009. The respective medians are computed on the same two groups of states as in Figure 5.9a. Sources: Datastream, ECB calculations.

9780230_243231_06_cha05.indd 115

5/20/2010 10:39:27 AM

116

Catching the Flu from the United States

The decrease in US state spreads since the beginning of 2009 has been associated with the impact of President Obama’s announcement and enactment of a large-scale stimulus package containing approximately USD 150 billion relief to state governments, which covers a substantial share of their projected budget shortfalls over the next few years. As this represents a transfer of funds from the federal authority to state governments, the recent declines in US state spreads may partly reflect a perceived transfer of credit risk between the sovereign and sub-sovereign entities within the US federal structure.15 Taking a snapshot at mid-March 2009 (when euro area sovereign bond spreads were near their crisis peak) one observes that US state spreads have been generally more strongly affected than their euro area counterparts when compared to end-June 2007 levels. The exceptions are Greece and Ireland as well as Slovenia, Cyprus and Slovakia. The strong increases of US spreads led them all to stand at positive levels at that date. The most substantial increases were observed in areas particularly hard hit by the housing market downturn (California and Florida), but also in areas with a high concentration of economic sectors, which were at the heart of the current recession: the automobile industry (Michigan) and financial services (New York). All four states had fiscal balances which were weaker than the US average in the fiscal year 2008 and were projected to deteriorate further in the fiscal year 2009. Finally, from end-March 2009, both euro area sovereign bond spreads and US state bond spreads showed strong declines amid a general improvement in economic sentiment and globally decreasing risk aversion. These decreases came to a halt only in mid-May (euro area) or early June (United States), with spreads on both sides of the Atlantic still considerably exceeding their pre-crisis levels. To wrap up, during the financial turmoil, the spreads between euro area government bond yields and their German counterparts have been widening to levels not seen since in each respective country since it joined the EMU. Similar developments were observed in the United States, where spreads between state bond yields and Treasury bond yields were widening. The two sets of spreads were showing some marked co-movement until end-2008. After that, euro area sovereign bond spreads increased further, while US state bond spreads declined, which is probably associated with the announcements of large-scale support packages expected to transfer funds from the federal authority to the benefit of state governments. From endMarch 2009, both groups of spreads showed marked declines amid improving economic sentiment and decreasing risk aversion.

5.3.2 Model setup Results are based on VAR models with four lags.16 The VARs including GDP variables only are specified in levels with cointegration.17 Based on the trace statistic (Johansen, 1988) we find evidence of at most one cointegrating vector among the three economic areas. The specifications in level are mainly aimed at checking whether imposing cointegrating

9780230_243231_06_cha05.indd 116

5/20/2010 10:39:27 AM

The Role of Financial Variables

117

restrictions actually improves or worsens forecasting ability. Two VARs include measures of real activity only (GDP) while the remaining models also consider the three combinations of financial variables described before. The cointegration model is: Yt = A0 + A1Yt 2 1 1 A1Yt 2 2 1 A1Yt 2 3 1 A1Yt 2 4 1 C␭yt 2 1 1 ␧t ,

(5.1)

where the vector of endogenous variables Y is partitioned into two subsets as Y 5 [y fin]’ where y is a vector that includes the logarithm of the GDPs and fin is a vector that collects two financial variables selected in turn (out of three) for the different economic areas. The cointegrating vector l links the dynamics of the two or three GDPs while ␧t is a vector of error terms (whose dimensions go from a minimum of 2 3 1 to a maximum of 10 3 1) normally distributed with covariance matrix S. The models estimated can be classified as follows: 1. A two-country model of the US and euro area log GDP (BiVAR). Together with model 0 (the random walk for the GDP growth rate) this model constitutes the benchmark model against which the other specifications are tested. 2. A three-country model of the US, the euro area and the Rest of the World (ROW) log GDP in levels with cointegration (TriVar model). 3. A three-country model for the log GDP, in levels with cointegration, plus the stock market volatility and the slope (FiVAR1). 4. A three-country model including the log GDP, in levels with cointegration, plus the stock market index and the slope (FiVAR 2). 5. A three-country model for the log GDP plus the dividend yield, the bond yield and the slope (FiVAR3). 5.3.3 Unconditional forecasting evaluation The evaluation of predictive ability of the models is based on three main statistics: the Root Mean Square Error calculated through the outof-sample forecast errors between one and 12 quarters ahead; rolling RMSEs, calculated over moving windows of 12 quarters, which allow us to examine time variation in the predictive ability of the models, and the conditional predictive ability test proposed by Giacomini and White (2006).18 The rolling RMSEs should be seen as a complement to the results of the GW test, since they can be carried out on small windows of data while the GW test requires re-estimating the VAR on windows of fixed length, which cannot be done in practice on windows shorter than 48 quarters. The comparison of the RMSEs coming from

9780230_243231_06_cha05.indd 117

5/20/2010 10:39:27 AM

118

Catching the Flu from the United States

the fixed estimation windows also allows us to investigate the amount of instability displayed by the data. Table 5.1 shows the ‘unconditional predictive ability’ RMSE (i.e. the RMSE for euro area GDP growth rates between t 1 j and t 1 j 1 1, for j 51,2, ... ,11), coming from forecasts conditional on all information – GDP and financial data – as of time t. We highlight that, with the exception of the one-step-ahead forecast, over the vast majority of the chosen forecast horizons, model 1, i.e. the VAR that includes the two GDPs only, has the best performance (see the second row in in the table) and its performance is very close to that of the VAR which includes the three GDPs, as well as to the VAR where GDPs are complemented by the stock market volatility and the slope of the yield curve. Considering that these latter two models have many more parameters than the simple bivariate VAR, their performances should be taken as broadly similar. The random walk model is never a winning choice. Adding other combinations of financial variables in the VARs beyond the slope and the stock market volatility worsens the forecasts at horizons shorter than one year, while at longer horizons the forecasts tend to return to near the global minimum. Either the implicit cointegration relationships between GDPs that drive co-movements at longer horizons are not affected by the dynamics of financial variables or noisy information contained in financial variables is smoothed out at long horizons. Beyond the RMSEs, the Diebold and Mariano (1995) tests also suggest that financial variables provide no help in forecasting real activity. The Diebold-Mariano tests are reported in Table 5.2 and are obtained as the t-ratio on the coefficient of a regression of the difference between the absolute values of the forecast errors (for euro area GDP) from competing models on the constant itself. Entries in the table larger than 1.64 indicate that the model in the second row of the table is not outperformed by the model in the first column. All of the GDP forecasts upon which the tests are based rest on the parameters estimated on the full sample and are therefore unconditional in a broader sense than was implied by the RMSE, which was based on expanding window estimates. By using parameters based on a full-sample estimation in the construction of the Diebold-Mariano test we wish to control for parameter uncertainty, which should be mitigated in the broader 1970–2007 period. The random walk model has been eliminated from these comparisons as, unconditionally, we would have had the same GDP forecast for all the sample period (the average GDP growth rate) and we checked only the performance of the BiVAR and the TriVAR models relative to all of the remaining models at three horizons: four, eight and 12 quarters.

9780230_243231_06_cha05.indd 118

5/20/2010 10:39:27 AM

9780230_243231_06_cha05.indd 119

5/20/2010 10:39:27 AM

0.45 0.43 0.42 0.49 0.56 0.61

0.42 0.49 0.05 11.9

Panel A Random walk 2 GDP 3 GDP var slope sm slope dy slope

Panel B Minimum all Minimum fin Relative gap % gap

1

0.41 0.43 0.02 4.34

0.44 0.41 0.42 0.43 0.55 0.58

2

0.46 0.4 0.42 0.47 0.49 0.69

4

0.41 0.4 0.48 0.47 0.07 0.07 16.66 16.86

0.45 0.41 0.43 0.48 0.49 0.68

3

0.46 0.4 0.42 0.44 0.49 0.72

6

0.4 0.4 0.45 0.44 0.05 0.04 11.88 11.18

0.46 0.4 0.41 0.45 0.52 0.73

5

0.4 0.45 0.05 12.75

0.46 0.4 0.43 0.45 0.48 0.67

7

0.41 0.45 0.04 8.87

0.47 0.41 0.44 0.45 0.5 0.64

8

0.42 0.44 0.02 3.73

0.47 0.42 0.44 0.43 0.46 0.62

9

0.42 0.44 0.02 3.73

0.47 0.42 0.45 0.44 0.5 0.64

10

0.43 0.44 0.02 3.67

0.47 0.43 0.45 0.44 0.5 0.65

11

0.43 0.44 0.01 2.32

0.47 0.43 0.45 0.44 0.48 0.64

12

Note: The table shows the out-of-sample RMSE for a number of models between one and 12 quarters ahead. The forecasts are made as of time t conditional on the values of all the variables as of time t. The RMSEs are obtained by estimating the models on expanding windows, the first of which starts in 1970Q1 and ends in 1986Q4. One observation at a time is added in the rolling procedure. The last four lines of the table report the minimum RMSE across all models, the minimum RMSE across all models that also include variables other than just GDP, the relative gap between such two minima and the percentage gap.

RW BiVAR TriVAR FiVAR1 FiVAR 2 FiVAR 3

Forecast horizon

Table 5.1 Unconditional out-of-sample RMSE for the euro area GDP

120 Catching the Flu from the United States Table 5.2 Diebold-Mariano tests for unconditional predictive ability at selected horizons h=4 BiVAR vs.

TriVAR vs.

h=8

h=12

BiVAR vs.

TriVAR vs.

BiVAR vs.

TriVAR vs.

Euro area BiVAR TriVAR FiVAR1 FiVAR 2 FiVAR 3

1.52 0.81 0.94 3.47

20.20 20.02 2.78

0.94 0.81 1.69 3.14

0.12 1.38 2.90

0.59 20.44 0.45 1.30

21.12 20.15 1.18

20.33 20.46 20.93 0.60

0.36 20.85 0.67

20.93 20.34 0.92 1.52

20.25 1.11 1.59

21.03 20.89 0.36 2.82

20.85 0.47 2.85

United States BiVAR TriVAR FiVAR1 FiVAR 2 FiVAR 3

Note: The table reports the Diebold-Mariano tests for unconditional predictive ability between pairs of models. The test is the t-ratio on the constant in a regression of the differential between the absolute values of the forecast errors produced by competing models on the constant. The headings h 5 4, h 5 8 and h 5 12 denote the forecast horizons. Entries larger than 1.96 indicate that the models reported in the second row of the table (BiVAR and TriVAR) are not outperformed by the corresponding competing model in the first column.

In no comparisons would the VAR which included financial variables have been preferred to the BiVAR and the TriVAR at forecast horizons of four and eight quarters, while in most cases at the 12-quarter ahead horizon the test would be neutral between models with and without financial variables. 5.3.4 The timeliness of financial variables: Conditional forecasting GDP numbers (even taking into account flash and preliminary estimates) are often made public after a substantial delay. As a result, the literature dedicated to short-run forecasting often makes use of financial data as leading indicators, in addition to survey data (on retail sales, labour markets, and so on) to complement the flow of information used for current quarter forecasting (nowcasting) and next quarter forecasting. For instance, Giannone et al. (2005) analyse the information content of around 200 macroeconomic releases for the US economy and find that interest rates, among a number of macroeconomic releases, improve the nowcasting of GDP.

9780230_243231_06_cha05.indd 120

5/20/2010 10:39:28 AM

The Role of Financial Variables

121

We now put ourselves in the position of a short-term forecaster. We use our VAR models to forecast period t + 1 and beyond using financial data up to period t + 1, but the most recent GDP release ‘known’ by the model refers to period t only. The forecasts are generated as conditional forecasts, that is, we condition the VAR forecasts between t + 1 and t + 12 on financial data as of time t + 1 and GDP as of time t. The results are presented in Table 5.3, panels A and B. The third-tolast row in each panel reports the minimum RMSE achieved by the models, that is, the lowest RMSE of the unconditional forecasting exercise. Conditional forecasts are worth being employed if the RMSE that they produce is lower than the RMSE coming from unconditional forecasts, that is, if information dated t and t + 1 is richer than information dated t only. Looking at panel A, up to the eighth step ahead, forecasting performance deteriorates for all models with financial variables, by between 15 and 35%. Between the two- and the three-year ahead horizons, the RMSE gets close to the overall minimum in Table 5.1. Panel B reports the same information when knowledge about GDP is further restricted to time t 2 1 while financial variables are known for times t and t + 1. The picture does not change too much with respect to panel 1. Thus, it seems that the optimal forecast need not use the most recent quarterly financial data. Overall, our forecasting exercise shows that the more general ‘GDP models’ consistently beat the random walk over all the considered horizons for the euro area. In particular, models with two or three GDPs, also including the stock market volatility and the slope of the yield curve, consistently improve the forecasts. Adding other combinations of financial variables tends to worsen predictions (a result in line with the US literature, e.g. Stock and Watson, 2003) but the losses are small, especially at very long horizons (between one and a half and three years ahead) despite the increased uncertainty brought by the larger number of parameters to be estimated. It is, however, surprising that predictions worsen significantly in our conditional forecast exercise. We also remark that the forecasts produced by both model 1 (the two GDPs) and model 2 (the three GDPs) are extremely smooth throughout the sample, somewhat intuitively in contrast with the superior forecasting ability that they have displayed. On the contrary, the GDP forecasts from the models that include financial variables display much more time variability, and during certain periods they visually track very well the true values of the GDP growth. Therefore, the overall low forecasting power of models with financial variables actually hides periods where their predictive ability can be much better than that of simpler models.

9780230_243231_06_cha05.indd 121

5/20/2010 10:39:28 AM

9780230_243231_06_cha05.indd 122

5/20/2010 10:39:28 AM

Forecast horizon

1

2

3

4

5

sm slope dy slope Minimum Gain/loss over unco % gap

0.51 0.6 0.72 0.51 0.09 17.6

0.49 0.59 0.61 0.49 0.07 14.3

0.54 0.57 0.6 0.14 0.14 25.9

0.61

0.7

0.55 0.53 0.75 0.78 0.55 0.53 0.15 0.13 27.2 33.1

var slope sm slope dy slope Minimum Gain/loss over unco % gap

0.81 0.96 1.09 0.31 0.38 46.9

0.8 0.62 0.67 0.52 0.21 33.9

0.89 0.93 1.08 0.65 0.58 0.59 0.69 0.88 0.86 0.65 0.58 0.5 0.24 0.18 0.1 36.9 31 32.2

1.15 1.11 0.53 0.53 0.75 0.8 0.53 0.53 0.13 0.13 24.5 24.5

1.11 0.5 0.82 0.5 0.09 21.5

0.49 0.72 0.49 0.07 18

0.64

8

1.1 0.44 0.69 0.44 0.03 6.5

0.47 0.63 0.47 0.05 12.9

0.57

9

1.12 0.48 0.77 0.48 0.06 12.5

0.45 0.68 0.45 0.03 6.7

0.6

10

1.14 0.44 0.79 0.44 0.01 2.2

0.48 0.71 0.48 0.05 10.4

0.6

11

1.04 0.46 0.72 0.46 0.02 4.3

0.47 0.68 0.47 0.03 6.4

0.57

12

Note: The two panels of the table show the out-of-sample RMSE for a number of models between one and 12 quarters ahead. In Panel A, the forecast are made as of time t conditional on the values of the GDP as of time t and on the financial variables as of t 1 1. In Panel B, the forecast are made as of time t 2 1 conditional on the values of the GDP as of time t 2 1 and on the financial variables as of t and t 1 1. The RMSE are obtained by estimating the models on expanding windows, the first of which starts in 1970Q1 and ends in 1986Q4. One observation at a time is added to the previous sample in the rolling procedure. The last four lines of the table report the minimum RMSE across all models, the minimum RMSE across all models that also include variables other than just GDP, the relative gap between such two minima and their percentage gap.

FiVAR1 FiVAR 2 FiVAR 3

0.63

7

0.52 0.5 0.8 0.67 0.52 0.5 0.12 0.1 23.1 25.4

0.68

Panel B: Estimation up to t 2 1, forecasts conditional on time t and t 1 1 information

var slope

FiVAR1

FiVAR 2 FiVAR 3

6

Out-of-sample RMSE for the euro area, conditional on quarterly information

Panel A: Estimation up to t, forecasts conditional on time t 11 information

Table 5.3

The Role of Financial Variables

123

5.3.5 Conditional evaluation of forecasts To assess the predictive power of our models at different points in time we can as a first approximation resort to a rolling version of the RMSEs. The rolling RMSEs calculated over 12-quarter windows are reported for classes of models, at four-quarters-ahead horizons in Figure 5.10 for some selected individual models. They seem to contradict to some extent the picture given by the unconditional and conditional RMSEs calculated for the whole sample. Figure 5.10 suggests that the VARs which include the slope of the yield curve and the stock market volatility seem to be more successful. The rolling RMSE is not a formal test of model performance but has the advantage of being applicable to very short estimation windows and as such provides a hint about the ‘instantaneous’ predictive ability of a model. For a formal assessment of the predictive ability, the test proposed by Giacomini and White (2006) allows us to analyse out-of-sample predictive ability in realistic situations. The test generalises the widely employed

12

10

8

6

4

2

0 1990

1992

1994

1996

rw 3gdp+sr+slope Figure 5.10

1998

2000

2002

2004

3gdp+vol+slope

2gdp

3gdp+dy+slope

3gdp

2006

Rolling RMSE for different models (4-step ahead)

Note: Last observation is March 2007. Source: ECB Staff calculations.

9780230_243231_06_cha05.indd 123

5/20/2010 10:39:28 AM

124 Catching the Flu from the United States

Diebold-Mariano (1995) tests along two dimensions, namely the limiting properties and the conditional evaluation of the predictive ability. This latter feature allows us to answer, among other things, the questions ‘which forecasts will be better at a given point in time?’ or ‘which model among a set of competing models should one choose at a given point in time?’ The conditional predictive ability evaluation is particularly suitable for fine tuning the forecast evaluation to the underlying economic conditions. The test consists of regressions of the difference in the absolute forecast error from competing models over its lag and a constant. For each forecast series of the GDP in the euro area coming from a different model and for selected horizons (one, four and eight) we compute pairwise tests of equal conditional predictive ability of model ‘a’ against model ‘b’ through an absolute value loss function. Specifically, for lags t 5 1, 4 and 8 quarters we test the null hypothesis: ˆ t  GDP  GDP ˆ t |G » H 0 : E « GDPt  t  GDP t,ma tt t ,mb t ¬ ¼

E > ⌬Lt  t | Gt @

0,

where L is a loss function, ma stands for model ‘a’ and mb for model ‘b’. When the information set upon which the test is built is such that Gt 5 Ft then the test is a conditional test, whereas when Gt includes the full sample ([0,∞[) it is an unconditional test (the equivalent of the Diebold-Mariano test). The conditional test is a powerful tool for fine tuning the choice of a predictive model depending on the underlying environment. In simple terms, the choice of model ‘a’ over model ‘b’ is based on regressing DLt 1 t defined above on a constant and its first lag, and then preferring model ‘a’ to model ‘b’ if the predicted value of the error gap between the two models exceeds a given threshold. Of course, the above regression can be extended to include additional lags of the error gap as well as predetermined variables one may think of as being relevant in the determination of the performance of a given class of models (recession/expansion dummies, industrial production or GDP growth rates, and so on). All pairwise comparisons refer to the four-step ahead predictive ability only, for the period 1999–2007, but results are common to the other horizons and to further save space results are only presented for the ‘unconditional’ forecasts, that is, forecasts from the VARs as of time t conditional on the knowledge of all variables as of time t. Our results suggest that the random walk model would have been preferred in a significant fraction of the quarters only to model 1 (the two GDPs) and less frequently to model 2 (the three GDPs), but it would have been by far surpassed by nearly all the models with financial variables, both price- and non-price related. The model with the two

9780230_243231_06_cha05.indd 124

5/20/2010 10:39:29 AM

The Role of Financial Variables 1 vs 2

06

07 07 07

20

05

20 06

06

04

03

20

20

20

20

04

20

20

20

20

04

20

03

20

02

20

01

20

00

19

20

07

20

06

20

20

20

20

20

20

05

0.00 04

0.25

0.00 03

0.50

0.25

02

0.75

0.50

01

0.75

99

1 vs 9 1.00

00

05

1 vs 5 1.00

99

03

20

02

20

01

20

19

20

99

07

20

06

05

20

20

20

20

20

20

20

19

04

0.00 03

0.25

0.00 02

0.50

0.25 01

0.75

0.50

00

0.75

99

1.00

00

1 vs 8

1.00

20

05

1 vs 3

20

02

20

01

20

20

99

20

19

07

06

20

05

20

20

20

20

20

20

20

19

03

0.00 04

0.25

0.00 02

0.50

0.25 01

0.75

0.50

00

0.75

99

1.00

00

1 vs 51

1.00

19

125

Figure 5.11 GW test for conditional predictive ability Conditional choice: BiVAR model vs selected competing models (4-step ahead) Note: The panels in the figure show the results of the conditional model choice test proposed by Giacomini and White (2006). The test is presented as a series of zeros and ones, with one indicating that model 0 (the random walk assumption for GDP growth) would be chosen over the alternative model. The test builds on a regression of the differences between absolute out-of-sample forecast errors (4 quarters ahead) on their first four lags. The mean of the forecast errors is used as a threshold for the choice of model. The out-of-sample estimation is based on a moving window of 48 quarters. Model 1 is a BiVAR for US and euro area GDPs; model 2 adds the GDP of the rest of the world; models 3, 4 and 3l add the stock market volatility and the slope of the yield curve; models 5, 5l and 6 add instead the stock market return and the slope of the yield curve; models 7 and 7l add the dividend yield, the long-term nominal rate and the slope of the yield curve. Models 8 to 12 add to the three GDPs, respectively, the slope and the distance to default of a set of financial firms, the slope, the distance to default and its interquartile range for a set of financial firms, the Fama and French factors, the C&I loans, the C&I, real estate and consumer credit bank loans.

GDPs only (cf. Figure 5.11), which performs extremely well based on the in-sample and out-of-sample RMSE criteria, would have instead been frequently outperformed by a VAR which includes the stock market volatility and the yield curve slope beyond the three GDPs. The same occurs for the model with three GDPs only against the richer VAR model with stock market volatility and yield curve slope. It is interesting to note that models with financial variables performed best in 1999 and between 2001 and 2003 – periods in which the historical

9780230_243231_06_cha05.indd 125

5/20/2010 10:39:30 AM

126

Catching the Flu from the United States

decomposition attributed the largest revisions to the baseline forecast to shocks in financial variables. In a nutshell, although on average financial variables may not contribute to improving forecasts according to the RMSE metric, they seem to have conveyed useful information for the euro area GDP forecast in several episodes in the past.

5.4 Financial variables and the business cycle: Evidence from non-linear models We have argued in the previous section that, looking more in detail through the lens of conditional predictive ability tests, financial variables such as stock market returns, stock market volatility or the slope of the yield curve are not completely irrelevant in forecasting GDP but, rather, that their importance is rather time-varying, although marginal when one looks at the full sample. Indeed, given recent developments, too, it appears, at the least, puzzling that no forecasting content can be found in current financial indicators or current indicators of financial stress for the subsequent decline in real activity. What also contributes to making these findings even more puzzling is the strongly significant and large in-sample effect whereby, for example, a 1% shock to US equity index volatility leads to a 2% annualised drop of US GDP over a two-year horizon. How can this large effect vanish even at short horizons? One possibility for the absence of a systematic linkage in out-of-sample analyses is the implicit constraint imposed by the linear structure of the VAR. In fact business cycles, as well as financial turbulences, are strongly non-linear events. Recessions, for example, are deeper and more shortlived than expansions. In the same way, equity market volatility is very quick to spike up to then return to normal conditions. The non-linear data generating process could still be accommodated in sample by a linear VAR while it could be completely missed in the out-of-sample analysis. A simple non-linear VAR framework was proposed by Balke (2000), among others, as a remedy to pull out of the data the significant linkages between financial and real variables that one would expect and that some theoretical papers justify. The idea of a non-linear VAR is to incorporate in the model the possibility of two regimes: 1) a ‘calm/normal’ one where endogenous variables are linearly dependent from explanatory factors in light of historical regularity and 2) a ‘turbulent’ one where such relationships do not appear. Changes in credit conditions or financial volatility, for instance, could influence economic activity either directly or

9780230_243231_06_cha05.indd 126

5/20/2010 10:39:30 AM

The Role of Financial Variables

127

just by signalling ‘regime’ changes. For example higher Ted spreads could impact on activity either directly by generating tighter lending conditions or, indirectly, by signalling a ‘regime’ change in financial markets, particularly when specific ‘thresholds’ are exceeded. In order to account for a regime change, in this section we estimate a threshold VAR (TVAR) that comprises US and euro area GDP, as well as the US stock market return (or the US stock market volatility), the US Ted spread (used as a proxy for risk perceptions in the banking system) and the US short-term rate. The hypothesis to test is whether financial factors affect GDP differently when the Ted spread moves into the ‘turbulent regime’. To this aim the VAR allows values of the Ted to change the relationships among the endogenous variables (the coefficients of the VAR) as well as in the covariance matrix of the shocks. To anticipate, the estimation of the TVAR confirms that financial variables have a marked different impact on real developments depending on the regime, that is, on the past level of the Ted spread or the stock market volatility. In particular, the impact of financial factors on GDP is much larger within the ‘crisis regime’, that is, when the banking sector is suffering and confidence is lowest; interestingly, shocks to the Ted spread have a shorterlived – albeit stronger – impact on GDP during crises than in tranquil times, a pattern which is in accordance with the fact that recessions and crises are short-lived but severe.

Box 5.3 Threshold VAR models The threshold VAR (TVAR) is, as mentioned, a regime-switching model. The transition between one regime and the other (and, consequently, from a certain pattern of responses to another) is determined by the evolution of a variable selected as a threshold: when that variable crosses the threshold, it triggers the switching from one regime to the other. The TVAR is estimated by running the VAR across the two regimes using the same set of variables. Of course all typical VAR outputs, as impulse response functions, historical decompositions, forecasts and conditional forecasts have to be computed by simulation methods. This need stems from the fact that at each point in time after a shock has occurred or while forecasts are being generated one has to keep track of the state in which the TVAR is operating, as parameters, covariance matrix and bootstrapped residuals are potentially significantly different across regimes. The non-linearity shows up first of all in the different parameters and residuals covariance matrices across regimes, but also in the importance of the initial conditions and the size of the shocks: the behaviour of the model is not the same after a positive or a negative shock, in the same way as it differs after a small or a large shock,

9780230_243231_06_cha05.indd 127

5/20/2010 10:39:30 AM

128 Catching the Flu from the United States

thus mimicking our prior that economic activity should react to episodes of turbulence of negative surprises for individual economic releases. Abstracting for a moment from the determination of the threshold value, the TVAR can be written as:

( ≤ k) * (A

) )+ «

y t = i( zt − d > k) * A 0,1 + ∑ j =1 A1,p * y t − p + i( zt − d

p

+ ∑ j =1 A 2,p * y t − p p

0,2

t ,

(5.2)

where yt is the vector of endogenous variables; zt the threshold variable, itself one of the variables in y t; d is the delay with which the regime switch takes place; k is the threshold value; A01, A1p, A02, A 2p are the coefficient matrices in the two states and i is an indicator function. The estimation of the model at this point is simply obtained as two separate linear VARs across each of the two regimes, where the error term et is made up by two normal random variables each having zero mean but with variances S1 and S2. While the estimation can be easily obtained conditional on the knowledge of the threshold value, other pieces of information which are typical of the VAR have to be recovered in a more complex way. The simplest case is for the forecasts which have to be computed sequentially by conditioning on the forecast values of the threshold variable, that is, on whether the VAR is in ‘regime 1’ or ‘regime 2’. In a more complex fashion, impulse responses are obtained as differences between sets of ‘forecasts’ made with a given path of errors and with another path of errors where the first error for a given variable is shocked by a given amount. In the case of the reaction as of time t 1 k of the j-th variable to a shock at time t and of size h in the i-th variable this can be expressed as: IRFt 1 k[y(j)] 5 [Et yt 1 k(j)|(u1,u2,...ui,uN )] 2 [Et yt1k(j)|(u1,u2,...ui 1 h,uN )],

(5.3)

where ui is the error in the i-th variable of a VAR of size N and expectations are computed by averaging the IRF obtained across a given number of replications based on bootstrapping the reduced form errors ui (i 5 1, ... ,N). The probabilities of transition across regimes and an equivalent of the historical decomposition employed in a linear VAR also have to be computed by simulation. In the first case one employs the same ‘forecasts’ of the variables generated in the IRF calculation conditional on a given path of the error terms ui and referred to the threshold variables. Basically one has to work out through the simulations the percentage of times that the system finishes in a given regime, starting from a specified regime, after a given time has elapsed since the current time. Such probabilities can be calculated conditionally on a given path of errors as well as conditionally on the same path of errors but with the first one being shocked by plus or minus 1 or 2 standard deviations. Of course the probabilities of changing regime will be in principle extremely dependent on the initial condition as well as on the size of the shock that is imposed on the dynamic system.

9780230_243231_06_cha05.indd 128

5/20/2010 10:39:30 AM

The Role of Financial Variables

5.4.1

129

Data and VAR description

To assess the extent of non-linearity that potentially affects the linkages between credit/financial conditions and real developments in the United States and the euro area, we employ a TVAR including the following variables: the United States and the euro area GDP, the US federal funds rate as well as financial indicators reflecting, respectively, credit and financial market conditions, that is, the difference between US commercial paper yield and the three-month US Treasury bill rate (Ted spread) and stock market returns (for a formal description of the TVAR, see Box 5.3). As the financial variables employed (federal funds, stock market return and Ted spread) appear to be stationary between 1970 and 2008, we estimate our VAR using two lags and expressing economic activity as the (annualised) first differences of the two GDPs. Data are at quarterly frequencies and span the period between 1970Q1 and 2008Q4. The threshold variable is represented by a six-quarter moving average of the Ted spread, reflecting the idea that it should be reacting more quickly than loans or other credit indicators to a change in credit conditions. The moving average is used in order to smooth out spikes in the series which could give rise to frequent and unwanted swings in the regime. The value representing the threshold is selected via a grid search, based on the actual range of the Ted spread, by choosing the value that leads to a maximum in the likelihood function of the VAR. From the behaviour of the threshold variable, it is possible to identify four ‘credit crisis’ periods: the first two corresponding to the 1973 and 1980 recessions, the third associated with the downturn of 1990 and the last corresponding to the 2007–9 episode (see Figure 5.12). As mentioned, the TVAR is characterised by the presence of two regimes – turbulent and calm. Given the non-linear structure of the model, each regime will possess different impulse response functions, also as a function of the magnitude and the sign of the shocks. As is common in the literature (Boivin and Giannoni, 2007), the ordering of the variables depends inversely on their ‘speed’ of adjustment. Real variables are therefore ranked first and financial variables last. While among the financial variables Ted spreads are followed first by the stock market return or stock market volatility and last by the federal funds rate – the idea being that credit shocks tend to occur first and are then followed by stock market falls or increases in volatility, all of which are factors leading to an easing in monetary policy conditions.

9780230_243231_06_cha05.indd 129

5/20/2010 10:39:31 AM

130 Catching the Flu from the United States 3 2.5 2 1.5 1 0.5 0 1971 Figure 5.12

1975

1979

1983

1987

1991

1995

1999

2003

2007

Threshold variables (6-lag moving averages)

Note: Last observation is 2008Q4. Sources: Global Financial Data and ECB staff calculations.

5.4.2 Impulse responses across regimes Figure 5.13 shows the impulse response functions of US GDP to positive and negative shocks to US financial indicators. As impulse functions refer to a non-linear system we evaluate them for shocks of either one or two standard deviations and either positive or negative. The first point to highlight is that in the turbulent regime, that is, in presence of tensions in the credit market, the real economy is much more sensitive and reactive. It is important to observe, too, that monetary policy also has a different impact in the two regimes: quicker in the turbulent, shallower in the calm. This signals indeed that prompt monetary policy actions can be effective in restoring confidence and stabilising the credit market. The euro area GDP (Figure 5.14) seems to be rather unaffected by developments in the US Ted spread when the banking system is in a period of calm but the reaction becomes rather sharp (although just half of the corresponding reaction of US GDP) and extremely long-lived during turbulent phases. This finding corroborates the typical evidence of a contagion effect in financial prices and in addition indicates that credit market developments in the United States are more likely to affect real developments in the euro area when the US economy is facing a severe tightening of credit conditions. The opposite occurs in response to an unexpected decrease in the federal funds rate. The latter has a positive effect on euro area GDP, possibly resulting from the presence of a three– four-quarter lag between real developments in the two main economic

9780230_243231_06_cha05.indd 130

5/20/2010 10:39:31 AM

The Role of Financial Variables Ted spread

1.5

131

US federal funds rate

0.2

1.3 0.15

1.1 0.9

0.1

0.7 0.05

0.5 0.3

0

0.1 −0.1

−0.05 1

Figure 5.13

3

5

7

9

1

11

3

5

7

9 11

US GDP: Impulse responses 3-year ahead

Note: Responses of US GDP to a one standard deviation negative (expansionary) shock in US financial variables in calm (black line) and turbulent (dotted line) regime. Source: ECB staff calculations.

US Ted spread

0.7

0.05

US Federal funds rate

0.6 0

0.5 0.4

−0.05 0.3 0.2

−0.1

0.1 0 1 Figure 5.14

3

5

7

9 11 13

−0.15

1

3

5

7

9 11 13

Euro area GDP: Impulse responses 3-year ahead

Note: Responses of euro area GDP to a one standard deviation negative (expansionary) shock in US financial variables in calm (black line) and turbulent (dotted line) regime. Source: ECB staff calculations.

areas, while it becomes almost negligible during periods of turbulence. This may indicate that the transmission mechanism of monetary policy across the ocean becomes blurred during periods of financial stress. To some extent, this finding indicates that in turbulent periods real developments would benefit more from signs of improvement

9780230_243231_06_cha05.indd 131

5/20/2010 10:39:31 AM

132 Catching the Flu from the United States

in the banking sector and/or in confidence indicators rather than from the scale of monetary easing. Perhaps, this lack of effectiveness of the monetary policy within the model comes from the fact that the heightened volatility typically experienced by most asset classes in periods of stress distorts the transmission mechanism along the yield curve and makes the measurement of its real impact rather imprecise. As impulse responses are calculated by simulation inside the TVAR, the associated confidence intervals can be computed by exploiting the percentiles of the simulations. Although we do not report them in the figures for reasons of clarity, we confirmed that the impulses to US GDP coming from the Ted spread are extremely significant in both regimes and independently on the sign of the shock. In the turbulent regime, however, the significance tends to persist up to the fourth quarter after the shock, while the impact in the calm regime dies out in a couple of quarters. The GDP reaction to stock market shocks displays overall the same features, with negative and large equity returns inducing a much more prolonged negative impact on GDP. By contrast, the reaction of US GDP to innovations in the federal funds rate is rather small and almost negligible when compared to the impact originating from the financial variables. When the volatility drives the regime of the model, the Ted spread retains its significance while the reaction of US GDP to the volatility is very short-lived and rather noisy at longer horizons. In this case, the federal funds rate becomes more significant although it remains quite negligible. 5.4.3

Transition probabilities

One interesting feature of the nonlinear model is that it embodies implicitly information about the persistence of the various regimes as well as the probabilities for moving from one regime to the other. Empirically, the probabilities of moving from a given regime to another are calculated by simulating the VAR in each quarter a given number of times (100 in the application so far) each time for a fixed number of quarters ahead (eight). The probability of remaining in the turbulent regime is reported in Figure 5.15. Overall the turbulent regime is quite persistent, although less than the calm: four quarters after being in the turbulent regime, one has a 50% chance of being still there. The calm regime is more persistent: four quarters after being in the calm regime there is a 70% chance of being still there. Looking at how the probabilities change when the system is shocked may help to understand what changes in financial variables are required

9780230_243231_06_cha05.indd 132

5/20/2010 10:39:32 AM

The Role of Financial Variables

133

1 0.8 0.6 0.4 0.2 0 1

2

3 Baseline

Figure 5.15

4

5 + 2sd

6

7

8

− 2sd

Transition probabilities

Source: ECB staff calculations.

to bring back the system into the calm state. A two standard deviation shock to the Ted spread, approximately a 130 basis points rise or decline, has a very asymmetric impact. The shock increases only marginally the probabilities of remaining in the turbulent regime but a decline leads almost to nil in only two quarters the odds of seeing additional turbulences. This signals, again, that a recovery in the functioning of the banking system is key to abandoning the turbulent regime, and that policy interventions that substantiate in a negative shock to the Ted spread can lead to a quick return to the calm regime. 5.4.4 Probit models19 An additional possibility for fitting and forecasting the non-linear features of the business cycle is to model the expansion and recession phases through a probit. This approach is rather different from the TVAR presented before. The TVAR rests on the assumption that one of the endogenous variables forces the model to switch across regimes when it crosses a particular value. When one applies a probit to business cycle phases the aim is instead to fit the ‘exogenous’ historical pattern of recessions and expansions through some observable variables or combinations of variables. The final aim is not necessarily that of understanding how the economy behaves in one state rather in the other, but instead that of using the current values of the predictors to forecast the likelihood of recessions and expansions some quarters ahead of the current period. In the following, we use two main specifications, one that independently considers the predicting power of financial variables for the US, and a second, which takes also into consideration cross-border interdependencies.

9780230_243231_06_cha05.indd 133

5/20/2010 10:39:32 AM

134 Catching the Flu from the United States

Fitting US recession probability The use of probit regressions to fit business cycle phases has been popularised especially by Estrella (see Estrella 2005; Estrella and Hardouvelis, 1991; Estrella and Mishkin, 1998; Estrella et al., 2003). The main result of his papers is that the value of the yield curve slope today is able to anticipate the business cycle phase that will prevail some periods ahead, the most favourable horizon which one can exploit typically being one year. Figure 5.16 indeed depicts the typical anticipation of the yield curve slope, which becomes negative just before a recession begins. One of the main shortcomings of using one single indicator in forecasting economic conditions is that it can generate false signals, due to developments which affect the chosen indicator but which are uncorrelated to, or have low correlation with, subsequent recession probabilities. As an example, the use of financial variables to anticipate business cycle phases can be seriously misleading when one faces events as the Asian crisis in 1997, the stock market crash of October 1987 or the 9/11 attacks. Regarding the last event, we can recall that the US economy left

The Volatility of the US stock market 35 30 25 20 15 10 5 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 United States: The slope (line) and the corporate spread (dotted line) 5 4 3 2 1 0 −1 −2 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 Figure 5.16

US: Financial variables and recession episodes

Note: Vertical lines indicate recession episodes as signalled by the NBER. Source: ECB staff calculations.

9780230_243231_06_cha05.indd 134

5/20/2010 10:39:32 AM

The Role of Financial Variables

135

the recession only two months later (November 2001), while the burst in volatility caused by the attacks could possibly have led one to forecast the occurrence of a recession within the following year or so. It is for this reason that models with multiple indicators have been increasingly developed, as the false signals provided by one indicator are likely to be in contrast to, and therefore compensated by, the values of one or more of the other variables. Beyond the yield curve slope, which is typically measured via the tenyear less the three-month yield differential, typical choices for predictive variables are, among financial variables, the stock market return and the corporate bond spread. The corporate bonds spread, however, looks more like a coincident indicator, rising when the recession has already begun (although typically, the recession has as yet not been dated at that time). Recently, however, perhaps also as a consequence of the tensions in the banking system and due to the spike of volatility to historic records in all asset classes after the first signs of turbulence in summer 2007, the analysis has been broadened to consider more specific indicators. Among these, the stock market volatility seems to have been considered with particular attention. Bloom (2009) shows that a shock to volatility can indeed represent a shock to confidence. Volatility also has the same pattern as the yield curve, tending to fall just ahead of the beginning of a recession, to then peak near the latter’s end. Figure 5.17 shows the in-sample fit for US recessions between 1957 and September 2008. At each point in time the fitted probabilities are generated via the information set dated one year earlier. There are four specifications of a probit model, taken out of the even block examined in Fornari and Mele (2009): in the first, the only predictor is the lagged industrial production; in the second there is the slope, the corporate bond spread and the stock market return; in the third the slope and the short-term rate and the last one is a volatility block, with the stock market volatility together with the volatility of the yield curve slope. More specifically, the general equation has the following form: P(It1k 51)5 F(a 1 Sy51,N Sj51,5 b(y)j *xt-12-h(j)) 1 et1k ,

(5.4)

where F(x) is the cumulative normal density of x, It1k 5 0,1 is the expansion/recession indicator, N is the number of regressors considered (one, three, two and two, in the four blocks respectively), j the number of lags, in this case three, 12, 18, 24 and 36 months. The sample starts in January 1957 but the first fitted probabilities refer to 36 (maximum

9780230_243231_06_cha05.indd 135

5/20/2010 10:39:33 AM

136 Catching the Flu from the United States 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 57 59 61 63 65 67 69 71 73 75 77 79 81 83 85 87 89 91 93 95 97 99 01 03 05 07

Lagged industrial production Slope, short term rate

Slope, corporate bond spread, stock market return Stock market volatility, volatility of yield curve

Figure 5.17 US: Recession probability using alternative indicators (in sample probit model (1957–2008)) Note: Last fitted value is September 2008. Source: ECB staff calculations.

lag) 112 (forecast horizon) months, that is, four years after. It is striking that while future GDP growth cannot be predicted better than on basis of the lagged GDP itself, lagged industrial production makes a very bad job in predicting recessions (the black solid line). The estimated probabilities are not even higher overall in recessions than in expansions. The other models, despite originating some false signals, especially in the first few years of the sample and then especially in the first half of the 1980s, do quite a good job in matching recessions and expansions. The threshold probability for deciding whether a given period is a recession is actually 0.16, that is, the unconditional mean of the recession indicator between 1957 and 2008. When one takes an out-of-sample perspective, recessions and expansions become more complex to fit. Figure 5.18 below shows the out-of-sample performance of the four blocks of predictors considered before, with estimation based on moving windows of 360 months, that is, 30 years. As for the in-sample analysis, lagged industrial production does not help at all in forecasting recessions, while financial indicators retain some of their ability in fitting recessions, especially as concerns the episode that started in December 2007.

9780230_243231_06_cha05.indd 136

5/20/2010 10:39:33 AM

The Role of Financial Variables

137

Recession probability including cross-border interaction The analysis presented so far tries to fit business cycle phases in one country, the United States, with a combination of regressors that belong to that country only. However, the evidence is that business cycles are very interrelated across countries and financial variables react almost instantaneously to the intraday release of macroeconomic data (see also Kose et al., 2003, 2008). A way to allow for interdependencies across economic areas has been proposed by Fornari and Lemke (2009). They introduce a model, called ProbVar, that couples a VAR with a probit. The VAR has the advantage of being able to handle a potentially large number of indicators, as well as a large number of international common factors. To keep the problem within reasonable limits, not all of the variables that enter the VAR need to affect the recession probabilities via the probit relation, although their effect on the recession probabilities will be transmitted through their influence on the variables that enter the probit. The authors provide an application based on quarterly data for the United States, Germany and Japan, between 1960 and end-2007. They consider financial variables only (the yield curve slope, the corporate spread, the stock market return and the short-term rate), but allow for

1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

93 93 94 94 95 95 96 96 97 97 98 98 99 99 00 00 01 01 02 02 03 03 04 04 05 05 06 06 07 07 08 08

0

Lagged industrial production Slope, short term rate

Slope, corporate bond spread, stock market return Stock market volatility, volatility of yield curve

Figure 5.18 US: Recession probability using alternative indicators (out-of-sample probit model (1957–2008)) Note: Last fitted value is September 2008. Source: ECB staff calculations.

9780230_243231_06_cha05.indd 137

5/20/2010 10:39:33 AM

138 Catching the Flu from the United States

international factors in stock market returns and corporate spreads by taking a cross-sectional average of such variables across the countries belonging to the G7. At the four-quarter ahead horizon, the ProbVar does remarkably well in fitting recession phases for the United States, while success is less evident for Germany. It is worth noticing that the model has a remarkable fit at the one-quarter ahead horizon, which is extremely important for the aim of nowcasting recession phases, as the latter are typically identified by the dating committees and institutions long after their beginning. To see better the goodness of fit of the ProbVar, we show in more details the 2001 recession episode (Figure 5.19).

(a) United States, recession probs., conditional on 2000Q3 1 0.8 0.6 0.4 0.2 0 2000

2001

2002 ProbVar

2003

2004

Simple probits

(b) United States, recession probs., conditional on 2001Q3 1 0.8 0.6 0.4 0.2 0 2001

2002

2003 ProbVar

2004

2005

Simple probits

Figure 5.19 Out-of-sample recession probabilities – comparison ProbVar/probit during the 2001 recession episode Source: ECB staff calculations.

9780230_243231_06_cha05.indd 138

5/20/2010 10:39:33 AM

The Role of Financial Variables

139

The figure shows the out-of-sample recession probabilities based on information as of 2000Q3 (top chart), that is, one quarter before the beginning of the recession, and in 2001Q3, that is, the last recession quarter (bottom chart). Figure 5.19 also includes the same information but based on simple probits. Before the beginning of the recession, the ProbVar is much more informative about its duration than the simple probits are. At one quarter before the recession’s end, the two models seem to be equally able to detect the subsequent expansion, although the simple probits generate recession probabilities that return to the unconditional recession probability in a rather bumpy way and end up signalling a new recession at the beginning of 2004.

5.5

Conclusions

This chapter has provided an overview of the role of financial variables in explaining the real spillovers across the Atlantic. The issue was tackled from a different set of perspectives. First, we presented the theoretical reasons why financial developments should affect real activity, and we have tried to quantify the amount of financial shocks that have hit the US economy in the last years. Next, we tried to measure the amount of transatlantic real and financial spillovers by means of an appropriate index. Results indeed suggested that financial spillovers have increased in recent years, and that this explains the increasing real linkages between the United States and the euro area (cf. di Mauro et al., 2009). Subsequently, we considered the perspective of the information content of financial variables in anticipating real developments (cf. Espinoza et al., 2009). We found that financial variables do not appear to have a systematic information content, but they can indeed be relevant at certain points in time. Given the intuition that the relevance of financial variables can emerge during periods of financial turbulence, that is, when the economy is hit by financial shocks, we estimated a TVAR that allows for different linkages during calm and turbulent periods in financial markets (cf. Fornari et al., 2009). Results suggest that transatlantic linkages do indeed appear very different during the two phases, with the euro area reacting to US monetary policy and financing conditions (as measured by the Ted spread) only during turbulent phases. This result suggests somehow that the claim that financial variables do not help in predicting real developments should be played down, as results could be heavily dependent on the linear VAR structure adopted: indeed it could well be the case that financial variables have an information content in a non-linear framework. The last results we have

9780230_243231_06_cha05.indd 139

5/20/2010 10:39:34 AM

140

Catching the Flu from the United States

presented (based on Fornari and Lemke, 2009) relate to the adoption of a probit framework, in which financial variables are used to predict not the level of GDP, but rather its turning points. Overall, our results suggest the relevance of monitoring financial market developments for the purposes of monetary policy-making. The lesson we learn, also in the light of the recent recession, is that restoring confidence and the proper functioning of financial markets are necessary for good recovery prospects.

Notes 1. Of course due to the presence of currency-, country- and firm-specific risk premia, returns will not actually equalise instantaneously, but, to the extent that the investors perceive similarities between economies, risk premia are not expected to diverge too much. 2. Most of the general equilibrium models available in the literature feature only a very simplified financial sector, so the extent of information that can be extract by looking at the impact of a ‘financial shock’ in such setting is quite limited. 3. Note that this identification scheme relates to the structural VAR literature, which puts restrictions on the covariance matrix of the errors. This differs from the identification of shocks described in Chapter 4, where shocks are innovations of estimated structural equations. 4. A financial shock as defined above is related to financial market turbulences; hence stock market volatility seems a good indicator. Of course, having a broader set of indicators, each related to a different aspect of the financial channel, could improve the clarity of identification, although we remark that this would imply further data problems and estimation uncertainty. In order to keep the VAR size within a reasonable limit, we think it is preferable to employ only stock market volatility as a general measure of financial tensions. 5. There are very few results and no consensus in the literature about the definition and identification of financial shock. A newly proposed identification scheme such as ours should in principle be validated on historical data, so our results are to be taken with due caution. 6. Admittedly, a two-variable system is not capable of appropriately describing an economy, so in principle the model we use is at risk of misspecification. We remark, however, that the aim of the analysis is not to account completely for all sources of interrelations, nor to identify all relevant shocks hitting the economy, but rather to assess the amount of spillover effects that can be attributed to financial interlinkages. 7. The presence of two variables (the industrial production index and stock market returns) per country raises issues about using a generalised forecast error variance decomposition (as was done when only the industrial production indices were considered) and calls for a more structural identification where slow moving variables (industrial production) are placed ahead of financial variables or confidence indicators (see Boivin and Giannoni, 2007, for a similar identification in a different context).

9780230_243231_06_cha05.indd 140

5/20/2010 10:39:34 AM

The Role of Financial Variables

141

8. We have included Japan and the United Kingdom as a control, but given the focus of the book we will report results only for the United States and the euro area. We have limited our analysis to a small set of countries (although covering a large share of global GDP) in order to have access to a sufficiently large and meaningful time span. 9. This section is based on results presented in Espinoza et al. (2009), Fornari et al. (2009) and Fornari and Lemke (2009). 10. The euro area GDP series is obtained from the Euro Area Wide Model (see Fagan et al., 2001), the US GDP from the BEA national accounts while the GDP data used for the Rest of the World were taken from either the IFS of the IMF, the OECD or Global Financial Data. The GDP series for Canada was taken from the BIS, since the IMF and OECD data exhibited an unreasonable jump in 1995. The weights used for the construction of the Rest of the World series are the 1995 nominal GDPs in US dollars. All financial data come from the Global Financial Indicators database. 11. The stock market return is calculated as the quarter on quarter logarithmic change of the stock index but we also consider a four-quarter backwardlooking moving average of such returns, as smoother return series may help predictability by bringing financial volatility closer to the observed volatility of real GDP changes. 12. Monthly data are transformed in the same way as quarterly data, with moving averages being based on 12-month windows, matching the choice made for quarterly data. 13. There is no immediate solution for making the comparison more symmetric. There is no European federal entity issuing bonds which could take the role that Treasury bonds play in the United States. Alternatively, one may choose one of the individual US state bonds as reference entity to increase the symmetry of comparison. However, such an approach would come with considerable arbitrariness, and would require that the respective benchmark state maintains low perceived credit risk for a long period of time. 14. Even when considering daily changes, the correlation amounts to 0.3. 15. Although the improvement in US spreads started somewhat earlier (late December 2008) than the actual enactment of the stimulus package (17 February 2009), the two developments appear to be time-consistent as the announcement of the stimulus package went through a process spanning several months. As early as the beginning of January 2009, it was increasingly clear that federal aid would be offered to state and local governments as part of the forthcoming package. 16. While standard lag choice tests (AIC, SIC) suggest that one lag captures sufficiently well the dynamics of the variables, these tests are well known to underestimate the true dependence structure of the data. Based on the likelihood ratio test, and also considering that we are working with quarterly variables, we decided to fix the lag length at four. Furthermore, the slope of the yield curve and the stock market volatility predict business cycles at rather long horizons, typically 12–24 months, so the choice of a short lag would automatically limit the measured predictability. 17. We do not investigate fully fledged Vector Error Correction models, as we do not want to enter a discussion on the dimension of the cointegration

9780230_243231_06_cha05.indd 141

5/20/2010 10:39:34 AM

142

Catching the Flu from the United States

space for financial variables. Furthermore, financial variables are rather synchronised across economic areas (the literature has also evidenced the presence of risks when forcing cointegration in models with nearly integrated variables, such as interest rates, see Mitchell, 2000). To an extent, the major misspecification deriving from not considering cointegration among economic variables will affect the stationary models, as all models in levels will to some extent accommodate the long-run relationship among the variables. 18. While the first two tests rely on out-of-sample forecast errors coming from expanding window estimation, the third is built from a fixed window recursive estimation of the VAR models. 19. We are particularly indebted to Fabio Fornari (ECB) for having provided most of the material and research for this subsection.

References Ang, A., M. Piazzesi and M. Wei, 2006, ‘What Does the Yield Curve Tell Us about GDP Growth?’ Journal of Econometrics, 131, 359–403. Balke, N.S., 2000, ‘Credit and Economic Activity: Credit Regimes and Nonlinear Propagation of Shocks’, Review of Economics and Statistics, 82, 344–9. Bayoumi, T. and A.J. Swiston, 2007, ‘Foreign Entanglements: Estimating the Source and Size of Spillovers Across Industrial Countries’, IMF Working Paper No. 07/182. Blanchard, O. and D. Quah, 1989, ‘The Dynamic Effects of Aggregate Demand and Supply Disturbances’, American Economic Review, 79, 655–73. Bloom, N., 2009, ‘The Impact of Uncertainty Shocks’, Econometrica, 77 (3), 623–85. Bloom, N., M. Floetotto and N. Jaimovich, 2009, Really Uncertain Business Cycles, Stanford University, mimeo. Boivin, J. and M. Giannoni, 2008, ‘Global Forces and Monetary Policy Effectiveness’, in J. Galì and M. Gertler (eds), International Dimensions of Monetary Policy, Chicago: University of Chicago Press, NBER Working Papers 13736. Di Mauro, F., F. Fornari and M.J. Lombardi, 2009, ‘Measuring International Spillovers’, mimeo. Diebold, F.X. and R.S. Mariano, 1995, ‘Comparing Predictive Accuracy’, Journal of Business and Economic Statistics, 13, 253–63. Diebold, F.X. and K. Yilmaz, 2009, ‘Measuring Financial Asset Return and Volatility Spillovers, with Application to Global Equity Markets’, Economic Journal, 119, 158–71. Ehrmann, M., Fratzscher and R. Rigobon, 2005, ‘Stocks, Bonds, Money Markets and Exchange Rates. Measuring International Financial Transmission’, ECB Working Paper, No. 452. Espinoza, R.A., F. Fornari and M.J. Lombardi, 2009, ‘The Role of Financial Variables in Predicting Economic Activity’, ECB Working Paper No. 1108. Estrella, A., 2005, ‘Why Does the Yield Curve Predict Output and Inflation?’ Economic Journal, Royal Economic Society, 115 (505), 722–44.

9780230_243231_06_cha05.indd 142

5/20/2010 10:39:34 AM

The Role of Financial Variables

143

Estrella, A. and Mishkin, F.S., 1998, ‘Is There a Role for Monetary Aggregates in the Conduct of Monetary Policy?’ Journal of Monetary Economics, 40 (2), 279–304, Elsevier. Estrella, A., A.P. Rodrigues and S. Schich, 2003, ‘How Stable is the Predictive Power of the Yield Curve? Evidence from Germany and the United States’, The Review of Economics and Statistics, 85 (3), 629–44, MIT Press. Fagan, G., J. Henry and R. Mestre, 2001, ‘An Area-wide Model (AWM) for the Euro Area’, ECB Working Paper No. 42. Fornari, F. and A. Mele, 2009, ‘Financial Volatility and Economic Activity’, London School of Economics, mimeo. Fornari, F., A. Galesi and M.J. Lombardi, 2009, ‘Financial Fragility and the Business Cycle: A Nonlinear VAR Analysis’, mimeo. Fornari, F. and W. Lemke, 2009, ‘Financial Variables and Recession Probabilities’, ECB Working Paper, forthcoming. Galì, J., 1992, ‘How Well Does the Is-Lm Model Fit Postwar US Data?’ Quarterly Journal of Economics 107(2), 709–38. Gerlach, S. and F. Smets, 1995, ‘The Monetary Transmission Mechanism: Evidence from the G7 Countries’, CEPR Discussion Paper 1219. Giacomini, R. and H. White, 2006, ‘Tests of Conditional Predictive Ability’, Econometrica, 74, 1545–78. Giannone, D., L. Reichlin and D. Small, 2005, ‘Nowcasting GDP and Inflation: The Real Time Informational Content of Macroeconomic Data Releases’, CEPR Discussion Paper No. 5178. Johansen, Soren, 1988, ‘Statistical Analysis of Cointegration Vectors’, Journal of Economic Dynamics and Control, 12 (2–3), 231–54, Elsevier. Kose, M.A., C. Otrok, and C.H. Whiteman, 2003, ‘International Business Cycles: World, Region, and Country-Specific Factors’, American Economic Review, 93, 1216–39. Kose, M.A, C. Otrok, and C.H. Whiteman, 2008, ‘Understanding the Evolution of World Business Cycles’, Journal of International Economics, 75 (1), 110–30, Elsevier. Mitchell, J., 2000, ‘The Importance of Long Run Structure for Impulse Response Analysis in VAR Models’, NIESR Discussion Papers 172, National Institute of Economic and Social Research. Peersman, G., 2005, ‘What Caused the Early Millennium Slowdown? Evidence Based on Vector Autoregressions’, Journal of Applied Econometrics, 20, 185–207. Sims, C., 1980, ‘Macroeconomics and Reality’, Econometrica, 48, 1–48. Stock, J.H. and M.W. Watson, 2003, ‘Forecasting Output and Inflation: The Role of Asset Prices’, Journal of Economic Literature, 41, 788–829. Walsh, C.E., 1993, ‘What Caused the 1990–1991 Recession?’ Economic Review, Federal Reserve Bank of San Francisco, 33–48. Yilmaz, K., 2009, ‘International Business Cycle Spillovers’, UNPUBLISHED Working Paper.

9780230_243231_06_cha05.indd 143

5/20/2010 10:39:34 AM

6 Economic Interactions US-Euro Area Over the 2007–9 Financial Crisis: What Did We Learn?

6.1

Introduction

Despite the long history of interdependence of global cycle developments, prior to the onset of the most recent global downturn in late 2008, there was a widely held view that the euro area – and more generally the global economy – could ‘decouple’ from the US. Somehow, the consensus was that, given its plethora of domestic problems, the US would remain the most affected country: in this ‘new world’, unlike the past, the US shocks would not necessarily become global, nor would the euro area economy necessarily be affected with a lag. Experience to date in this downturn has unfortunately refuted the validity of this argument. The recession in the US turned itself into one of the most damaging and synchronised global recessions in history. The euro area in this context was also severely hit, and even more strongly than historical relations would have suggested. The purpose of this chapter is to assess the most recent episode in the context of the empirical patterns connecting the US and Euro area economies examined so far in this book. The aim is to ascertain to what extent the ongoing downturn may have differed from longer-term empirical patterns. The questions we want to address are therefore: is this episode – apart being one of the most severe – really different from the past as regards the reactions of the euro area economy to a US shock? Are there specific factors – of either cyclical or structural nature – this time around that may have caused a departure from previous history? These questions – rather backward-looking nature – will be put into a more forward-looking perspective in Chapter 7. In this chapter, we will examine in particular three aspects not treated before: first, if and to what extent the real spillover of the financial 144

9780230_243231_07_cha06.indd 144

5/11/2010 5:02:58 PM

US-Euro Area Interactions over the 2007–9 Crisis

145

crisis in the US to the euro area could have been anticipated; second, whether more globally synchronized real estate adjutsments may have helped to spread the downturn more widely; third, how the existence of the European Monetary Union may have modified traditional patterns of transmission between the US and the euro area.

6.2

The current downturn 2008–9: Stylised facts

Against the background of the empirical patterns observed in past economic cycles (see Chapter 2), this section looks at economic developments in the two areas during the current cycle, particularly in the context of the global trade downturn. 6.2.1 Stylised facts In line with other recessions associated with financial crises, this downturn has been deep; activity in the euro area and the US fell sharply during 2008 and the first half of 2009. It appears that euro area activity has been hit quite severely, although the precise extent relative to the US depends on the time span and metric used. In particular, the contraction in absolute GDP was relatively strong in the euro area (see Table 6.1). Using however cumulated deviations from average growth since the beginning of 2008, results have been relatively similar in both economies (see Table 6.2).1 The transmission of the downturn from the US to the euro area has been relatively fast during the current downturn. At first, the Table 6.1 Comparison of developments in GDP in the euro area and US 2007 Q4

2008 Q1

Q2

Q3

2009 Q4

Q1

Q2

Q3

0.7 0.5

–0.8 –1.7

–2.6 –4.2

–3.9 –4.6

–4.1 –4.5

-0.1 -0.4

–1.6 –1.8

–1.6 –2.5

–0.6 –0.6

–0.3 –0.2

Year-on-year percentage change US Euro area

2.3 2.2

2.5 2.2

2.1 1.5

Quarter-on-quarter percentage change US Euro area

0.0 0.4

0.2 0.7

0.7 –0.3

Note: Actual data until 2009Q2. Source: Eurostat, Bureau of Economic Analysis and ECB staff calculations.

9780230_243231_07_cha06.indd 145

5/11/2010 5:02:58 PM

146

Catching the Flu from the United States

Table 6.2 Cumulated deviation from average quarterly growth (pp) 2008Q1–2008Q4

2008Q1–2009Q2

–3.6 –3.7

–7.1 –7.8

US Euro area

Note: Average taken over the period 1996–2008 (0.7% in the US and 0.5% in the euro area). Sources: Eurostat, Bureau of Economic Analysis and ECB staff calculations.

6 4 2 0 −2 −4 −6 1996

1999

2002 Euro area

2005

2008

US

Figure 6.1 Real GDP per capita (year-on-year change, %) Note: Last observation is 2009Q3. Source: Eurostat and the US Bureau of Economic Analysis.

moderation in GDP growth in the US (from early 2006) preceded that in the euro area (mid-2006); see Figure 6.1. Since the onset of the financial turbulence in July 2007, the cyclical patterns have become very synchronous, however. The NBER dates the start of the US recession at December 2007, while according to the CEPR, the euro area entered recession just one month later, in January 2008.2 In contrast to output, adjustments in the labour markets have been less synchronous, with a later and slower adjustment in unemployment in the euro area compared with the US (Figure 6.2).

9780230_243231_07_cha06.indd 146

5/11/2010 5:02:58 PM

US-Euro Area Interactions over the 2007–9 Crisis

147

12.0 10.0 8.0 6.0 4.0 2.0 0.0 1995

1998

2001

Euro area

2004

2007

United States

Figure 6.2 Unemployment rate (percentage of labour force) Note: Last observation is November 2009 for euro area and July 2009 for US. Standardised unemployment rate (euro area), civilian unemployment (US). Source: Eurostat and the US Bureau of Labour Statistics.

It is also worth noting that the composition of changes in (absolute) GDP in recent quarters has been different between the two regions (Figures 6.3 and 6.4). The downturn in the US has seen a stronger slowdown in consumption than the euro area. Both economies experienced a contraction (in yearon-year terms) in investment at the end of 2008. But the US has seen a longer and more sustained decline in capital formation over the past two years, driven by the sharp falls in residential investment. In addition, as global growth declined, the euro area suffered from a sharp reversal in the net trade position, whereas growth in the US has been boosted by trade. 6.2.2

The global trade collapse

As the financial crisis intensified in the autumn of 2008 and confidence collapsed, the strong world trade growth of the last decade – mostly fuelled by higher integration of emerging economies and delocalization in production processes – reversed sharply with trade back to 2005 levels in a matter of two quarters. The size of the drop and its synchronisation across the globe (with over 90% of countries reporting declines of more than 5%) was unprecedented in postwar history (Figure 6.5). Given its open nature, the euro area was particularly affected, with exports of goods declining by about a fifth between September 2008 and February 2009. Like in the rest of the world, the contraction was

9780230_243231_07_cha06.indd 147

5/11/2010 5:02:59 PM

148

Catching the Flu from the United States 4 2 0 −2 −4 −6 −8

2007

2008 Consumption Inventories

2009 Investment Net trade

Government GDP

Figure 6.3 GDP components – euro area (contributions to year-on-year change in GDP, pp) Note: Last observation is 2009Q3. Source: Eurostat.

4 2 0 −2 −4 −6

2007

2008 Consumption Inventories

2009 Investment Net trade

Government GDP

Figure 6.4 GDP components – United States (contributions to year-on-year change in GDP, pp) Note: Last observation is 2009Q3. Source: US Bureau of Economic Analysis.

9780230_243231_07_cha06.indd 148

5/11/2010 5:02:59 PM

US-Euro Area Interactions over the 2007–9 Crisis

149

105

100

95

90

85

80

75 2005

2006

2007

2008

2009

Euro area exports World trade World industrial production Figure 6.5 World trade, global activity and euro area exports (volume indices: 2008Q3=100; 3-month moving average; monthly data) Notes: Euro area exports include intra- and extra-area exports. Vertical line represents September 2008 (i.e. Lehmann). Latest observation is September 2009. Sources: CPB Netherlands Bureau for Economic Policy Analysis.

particularly strong for intermediate and capital goods, rather than consumer goods (Figure 6.6), and this fact is critical to understand the underlining factors. While world output levels fell by a cumulative 3% in 2008Q4 and 2009Q1, trade volumes fell by about five times more. Not surprisingly, standard export equations have failed fully to account for this downturn in trade, which seems to be partly explained by the following factors.3 First, the fact that the manufacturing sector was hit more severely partly reflects both its relatively high dependence on capital and problems in access to credit following the intensification of the financial crisis. At the same time, activity in the services sector was much less affected (Figure 6.7). As the share of services in GDP is much higher than that of trade, trade volumes declined significantly more than GDP levels. Moreover, the recession and initial responses led to a shift

9780230_243231_07_cha06.indd 149

5/11/2010 5:02:59 PM

150 Catching the Flu from the United States 105

100

95

90

85

80

75

70 2005

2006

2007 Total trade Intermediate goods

2008

2009

Capital goods Consumer goods

Figure 6.6 Extra-euro area exports of goods (volume indices: Q3 2008 = 100; SA; 3-month moving average) Note: Latest observation is July 2009. Source: Eurostat, ECB staff.

away from the higher trade-intensive GDP expenditure components, such as investment, to other less-trade intensive components such as government consumption. The fall in exports also exacerbated the decline in trade as these are highly import-intensive (see also the following section).4 The financial crisis and the simultaneous confidence shock led firms abruptly to reduce their inventories and their investment in capital goods (Figure 6.8). As a result, trade in intermediate and capital goods declined much more than the trade of consumer goods.5 Following the severe contraction at the turn of the year, world trade stabilized by the end of 2009 benefiting strongly from the impact of temporary factors such as the fiscal stimulus plans, which also boosted demand for durable goods, cars in particular. The supportive rebound in the inventory cycle following the renewed growth of manufacturing

9780230_243231_07_cha06.indd 150

5/11/2010 5:03:00 PM

US-Euro Area Interactions over the 2007–9 Crisis

151

65 Services 60 55 50 45 40 Manufacturing 35 30 25 2004

2005

2006

2007

2008

2009

Figure 6.7 Global PMI manufacturing (diffusion index; monthly data) Note: Last observation is November 2009. Source: Markit.

activity and the gradual reactivation of global supply chains has also contributed to the recovery. The stabilisation in world trade was also paralleled by improving conditions in trade finance, as evidenced by the September 2009 IMF-BAFT survey. Euro area trade flows have also improved (Figure 6.9). As shown in Figure 6.10, this recovery was particularly pronounced in intra-euro area trade flows,6 which– as described in further detail below – seems to have benefited from strong growth in both consumption goods exports (possibly related to car scrappage schemes) and trade in intermediate goods (probably helped by the rebound in the inventory cycle). In geographical terms, the recovery in global trade flows was mainly led by Asia, particularly by a strong rebound in Chinese demand, probably largely related to the fiscal stimulus. Extra-euro area exports have also benefited from increased demand from Asia (Figure 6.11), which has partly compensated for the still subdued demand from most advanced economies. In terms of product decomposition, the improvements have been more pronounced for intermediate goods (Figure 6.12), partly due

9780230_243231_07_cha06.indd 151

5/11/2010 5:03:00 PM

152 Catching the Flu from the United States 4

10

2

0

0

−10 Stocks of purchases

−2

−20

−4

−30 EA industrial confidence (rhs)

−6

−8 2004

2005

2006

2007

2008

2009

−40

−50

Figure 6.8 Global PMI stocks of purchases and euro area industrial confidence (lhs: Mean-adjusted diffusion index; rhs: percentage balance) Note: Last observation is October 2009. Sources: European Commission, Markit.

to the turn in the inventory cycle outside the euro area and the aforementioned impact of the fiscal stimulus in China. Exports of cars and petrol have posted some increases lately for both internal and external trade, possibly supported by public car scrapping schemes. However, these overall developments hide some heterogeneity among euro area member countries. For instance, in Germany total exports were boosted in the third quarter of 2009 by consumption (namely cars) and intermediate goods, while capital goods had not yet reached positive territory. In France the recovery in exports is being led by intermediate exports, but also capital goods. In summary, a notable part of the recent global and euro area trade recovery seems to be driven by temporary factors ranging from the fiscal stimulus and car scrappage schemes to the rebound in the inventory cycle. At the same time, compared with the pick up in world trade and foreign demand, euro area exports recovered at a slower pace. This was

9780230_243231_07_cha06.indd 152

5/11/2010 5:03:00 PM

US-Euro Area Interactions over the 2007–9 Crisis

153

Sep-08

6 4

Imports

2 0 −2

Exports

−4 −6 −8 −10 −12 −14 2007

2008

2009

Figure 6.9 Total exports and imports (intra plus extra; goods plus services) (quarter-on-quarter percentage change; chained linked; SA) Note: Last observation is 2009Q3. Sources: National Accounts trade volumes data. Eurostat and ECB staff calculations.

partly related to differences in the geographical orientation and product composition of euro area exports compared to those of world imports. In particular, the recovery in world trade mostly reflects strong increases in import volumes in Asia, particularly by China and the Asian NICS, which constitute a smaller percentage of euro area exports (Figure 6.13). For instance, imports in China recorded the strongest increases for primary goods and mineral fuels. Imports in manufacturing and machinery, i.e. in products and sectors that euro area exporters are specialised in, however, grew much less. While euro area exports were supported in particular by higher demand for cars, the US may also have benefited from an increasing demand in iron and non-ferrous ores (representing around 7% of US exports to China, see Figure 6.14).

9780230_243231_07_cha06.indd 153

5/11/2010 5:03:00 PM

154

Catching the Flu from the United States

10

Sep-08

5 Extra-exports

0

−5

−10

Intra-trade

−15

−20 2007

2008

2009

Figure 6.10 Extra-exports and intra-trade of goods (3-month on 3-month percentage change; values; SA) Note: Last observation is 2009Q3. Sources: Eurostat and ECB staff calculations.

6.3

Financial indicators and the economic downturn

The large role of trade notwithstanding, the reaction of financial variables has certainly magnified the transmission of the global demand contraction. The turbulence and the associated losses experienced by financial markets worldwide since approximately the summer of 2007 have spurred research as well as policy debate about the possibility of using the forward-looking nature of financial prices to forecast economic activity.7 Against this background, and as a follow-up to what we discussed in the previous chapter, we may wonder whether considering the massive financial turmoil could have helped in anticipating the extent of the real downturn, and to what extent the stabilisation of financial markets could lead to an improvement in GDP forecasts.

9780230_243231_07_cha06.indd 154

5/11/2010 5:03:01 PM

US-Euro Area Interactions over the 2007–9 Crisis

155

10

Asia

5 US

0

−5

−10

Non-EA EU MS Total

−15

−20 Jan-08

Other

Jun-08

Nov-08

Apr-09

Sep-09

Figure 6.11 Contributions to growth in extra-euro area export values of goods (3-month on 3-month percentage change) Note: Last observation is September 2009. Sources: Eurostat and ECB staff calculations.

In this section, we will focus on the recession that started in the United States in the last quarter of 2007 and look at predictions for US and euro area GDP coming from VAR models that include either GDPs alone or GDPs together with financial variables. More specifically, we will examine a subset of the models studied in Espinoza et al. (2009) and presented in the previous chapter, with the aim of finding out whether looking at financial variables could have helped in detecting the first signs of the recession, its depth and its – possibly – close end.8 The three developments are evaluated by looking at forecasts produced in different quarters, that is, between 2008Q2 and 2009Q3, by two VAR whose variables are, respectively: 1) the US, the euro area and the Rest of the World GDP, and 2) these three GDP plus the stock market indices and the stock market volatilities. The forecasting ability of VAR models including financial indicators has been thoroughly investigated in the previous chapter (see also Espinoza et al., 2009), with the key finding that there is a clear dichotomy between the in-sample and out-of-sample predictions, and that

9780230_243231_07_cha06.indd 155

5/11/2010 5:03:01 PM

156

Catching the Flu from the United States

during particular phases of financial turbulence (for example between 1999 and 2002), VAR models including financial indicators would have been preferable to models including past GDP only. This finding, which to some extent suggests that the relation between real and financial variables may be somewhat unstable, partly reconciles the presence of a strong in-sample significance with an absence of out-of-sample predictive power. All VARs were estimated on quarterly data starting in 1970Q1; the last available observation refers to 2009Q2 for GDP and to 2009Q3 for financial variables. The Rest of the World is an aggregation of seven small countries, different from each another (Australia, Canada, Denmark, New Zealand, Norway, Sweden and Switzerland); the stock market indices are broad indices taken from Global Financial Indicators; the stock market volatility is a four-quarter moving average of absolute stock market returns.9

10

5 Capital goods 0

−5 Intermediate goods −10

Total

−15

−20 Jan-08

Cars & petrol

Jun-08

Nov-08

Apr-09

Sep-09

Figure 6.12 Contributions to growth in extra-euro area export values of goods (3-month on 3-month percentage change) Note: Last observation is July 2009. Sources: Eurostat and ECB staff calculations.

9780230_243231_07_cha06.indd 156

5/11/2010 5:03:01 PM

US-Euro Area Interactions over the 2007–9 Crisis 100

157

Other EMEs Emerging Asia

90

Other developed economies

80

Other EU member stat

70

Japan

60

United States

50 40 30 20 10 0 Figure 6.13 Composition of world (excl. euro area) imports and euro area foreign demand (percent; annual data for 2008) Source: IMF and ECB staff calculations.

EURO AREA

Specialized machines Cars, vehicle comp. Engines Aeronautics Electrical app. Machine tools

Iron, non ferrous metals/ores Iron Steel, tubes Precision instruments Electronic comp. Other

UNITED STATES

Aeronautics

Computer equip.

Electronic comp.

Engines

Agricultural. Prod.

Specialized machines

Plastic articles

Cars, vehicles comp.

Precision instr.

Other

Iron, non ferrous metals/ores

Figure 6.14 Composition of EA and US exports to China (shares in %, 20032007 averages) Source: Chelem database.

9780230_243231_07_cha06.indd 157

5/11/2010 5:03:01 PM

158 Catching the Flu from the United States

The estimation of the VAR systems is carried out for all quarters between 2008Q2 and 2009Q3 (the last date for which GDP is available being 2009Q2), with data going back to 1970Q1. In each quarter various forecasts are generated: 1) traditional GDP forecasts for the subsequent 12 quarters based on GDP information up to t⫺1 (the previous quarter), and 2) conditional forecasts of the GDP based on GDP information up to quarter t⫺1 (the previous quarter) and on the current quarter values of the financial variables. Results show that standing in 2008Q2 and in 2008Q3 negative realisations for US GDP in the subsequent few quarters could have been predicted, although their overall depth and duration would have been missed. The GDP forecasts made in 2009Q2 and in 2009Q3 evidence that models with financial variables tend to exhibit some overreaction to previous swings in asset prices, at least when one compares them to forecasts coming from VAR models that only consider past GDP.

6.3.1

Detecting the US recession and its depth

The first experiment aims at assessing whether financial variables would have helped in detecting the recession in the United States standing in 2008Q2 and 2008Q3 (i.e. before it was officially announced by the NBER, in December 2008). Figure 6.15 reports the actual growth rates of the US GDP together with forecasts for it made in these two quarters based both on past GDP alone (information for which was limited to, respectively, 2008Q1 and 2008Q2), as well as on GDP considered together with financial variables, which were instead known also for the actual quarters (i.e. 2008Q2 and 2008Q3). Comparing the two black straight lines (forecasts made in 2008Q2, GDP known up to 2008Q1, financial variables known up to 2008Q2) and the two dotted lines (forecasts made in 2008Q3, GDP known up to 2008Q2, financial variables known up to 2008Q3) shows that some indications about the forthcoming slowdown in activity would have been achieved, although the extent of the output contraction in 2008Q4 and 2009Q1 would have been by and large missed. 6.3.2 US financial shocks and real impacts on the euro area The second experiment looks at the extent in which the stabilization of US financial variables influence the recovery phase in the euro area. The third tries to assess the impact of a possible new round of financial turbulences.

9780230_243231_07_cha06.indd 158

5/11/2010 5:03:02 PM

US-Euro Area Interactions over the 2007–9 Crisis

159

0.015 0.010 0.005

a

0.000

b

−0.005 −0.010 −0.015 −0.020 2007

2008

2009

2010

Actual US GDP Based on GDP up to 2008Q1 Based on GDP up to 2008Q1 and financial variables up to 2008Q2 Based on GDP up to 2008Q2 Based on GDP up to 2008Q2 and financial variables up to 2008Q3

Figure 6.15 US GDP forecasts made in 2008Q2 and 2008Q3 (quarter-on-quarter change, percentage) Sources: ECB staff calculations and Global Financial Data.

To answer these questions, we will employ the VAR framework detailed above and look at forecasts produced between 2008Q2 and 2009Q3 by two VAR whose variables are, respectively: 1) the US, the euro area and the Rest of the World GDP, and 2) these three GDPs plus the stock market indices and stock market volatilities. For the second experiment we examine how the forecasts produced using financial indicators differ from those based on GDP only. Standing at the end of 2009Q2, when the first tentative signs of bottoming-out had appeared, Figure 6.16 shows the 12-quarter ahead growth profiles for US and euro area GDPs. For both economic areas, the fall in financial prices recorded in the aftermath of the Lehman’s collapse led the VAR including financial information to provide a much gloomier assessment of future growth, relative to models looking only at past GDP. Taking an ex-post standpoint, both types of models, however, provided a lower forecast than the actual GDP realised. It is interesting to note that while for the United States the Consensus forecast (black dotted line) is quite close to the forecasts based on past GDP alone, this is very far from true concerning the euro area,

9780230_243231_07_cha06.indd 159

5/11/2010 5:03:02 PM

160

Catching the Flu from the United States

0.015 0.01 0.005

Conditions in financial market represented a drag at least in the short

0 −0.005

All models consistently more pessimistic about the actual re-bound in 2009Q2

−0.01 −0.015

Latest available GDP figure

United States −0.02 2007 2008

2009

2010

2011

2012

Actual Based on GDP up to 2009Q1 and financial variables up to and including 2009Q2 Based on both GDP and financial variables up to 2009Q1 Based on GDP only, up to 2009Q1 Consensus as of 2009Q1 0.015 0.01 0.005 0 −0.005 −0.01 Latest available GDP figure

−0.015

Growth path for euro area significantly lagging behind the US path. Consistent across models.

−0.02 −0.025

Euro area −0.03 2007

2008

2009

2010

2011

2012

Actual Based on GDP only, up to 2009Q1 Based on GDP up to 2009Q1 and financial variables up to and including 2009Q2 Based on both GDP and financial variables up to 2009Q1 Consensus as of 2009Q1

Figure 6.16 Forecasts of US and euro area GDPs as of 2009Q2 (quarter-onquarter change, percentage) Sources: ECB staff calculations, Global Financial Data and Consensus Economics.

where GDP and financial variables, alone or considered jointly, seem to provide a consistently more negative assessment about growth prospects. Standing instead at the end of 2009Q3 (see Figure 6.17) provides the opposite result, that is, the improvements in financial conditions recorded throughout 2009 (see the grey dotted line relative to the black

9780230_243231_07_cha06.indd 160

5/11/2010 5:03:03 PM

US-Euro Area Interactions over the 2007–9 Crisis

161

dotted line) led the model to a stronger growth profile relative to what predictable based on past GDP alone. It is worthwhile to notice, however, that both types of models predict that the rebound in US activity will be followed by some moderation. In addition, from the beginning of 2009 especially, growth

0.02

United States While considering GDP dynamics only reveals lower growth profile

0.015 0.01

Financial markets point to initial optimism in GDP growth to then slide somewhat

0.005 0 −0.005

Latest available GDP figure

−0.01 −0.015 −0.02

2007

2008

2009

2010

2011

Actual Based on GDP up to 2009Q2 and financial variables up to and including 2009Q3 Based on both GDP and financial variables up to 2009Q2 Based on GDP only, up to 2009Q2 0.015

Euro area

In EA, financial markets optimism does not show up in higher GDP, as it is dominated by lagging behaviour relative to US.

0.01 0.005 0 −0.005

Latest available GDP figure

−0.01 −0.015 −0.02 −0.025 −0.03

Mar-07

Mar-08

Mar-09

Mar-10

Mar-11

Mar-12

Actual Based on GDP up to 2009Q2 and financial variables up to and including 2009Q3 Based on GDP only, up to 2009Q2 Based on both GDP and financial variables up to 2009Q2

Figure 6.17 Forecasts for US and euro area GDPs as of 2009Q3 (quarter-onquarter change, %) Sources: ECB staff calculations, Global Financial Data and Consensus Economics.

9780230_243231_07_cha06.indd 161

5/11/2010 5:03:03 PM

162

Catching the Flu from the United States

in the euro area is forecast to lose momentum relative to the United States. This feature is also evidenced in the Consensus forecast produced as of 2009Q3 (not reported), in contrast with what emerged from the Consensus forecast dated March 2009. The lagging behaviour of euro area growth relative to the United States occurs notwithstanding the positive effect which in principle should have been played by the recovery in financial markets in the first three quarters of 2009. The latter effect may have possibly been dominated, within the model, by the historical strength of the lag existing between real developments in the two areas. To understand how much future growth can be affected by a new deterioration in financial market conditions (third experiment), we compute conditional forecasts resting on the hypothesis of a 20% drop of the S&P500 index and a rise of 50% in stock market volatility, in 2009Q4, relative to the values prevailing as of end-September 2009. The size of this scenario is not particularly large when one looks at recent swings in financial prices and is such that the two variables would still remain far from the extremes reached in the aftermath of Lehman’s collapse, although it still represents a much worse outcome relative to the market-based expectations prevailing as of September 2009. The conditional forecasts under such an envisaged scenario of worsening financial conditions have been reported in Figure 6.18 as grey dotted lines. The GDP path reveals the sensitivity that, at the stock market index and volatility levels prevailing as of end-September 2009 and given the GDP figures in 2009Q2 and before, still characterises the real-financial linkages in the two main economic areas. In addition, possibly always on account of the strong lagging behaviour of the euro area relative to the United States, the adverse scenario would have a larger and more prolonged effect on growth in the euro area than in the United States. To wrap up, our results show that standing in 2008Q2 and in 2008Q3 negative realisations for US GDP in the subsequent few quarters could have been predicted, although their overall depth and duration would have been missed. The GDP forecasts made in 2009Q2 and in 2009Q3 evidence that models with financial variables tend to exhibit some overreaction to previous swings in asset prices, at least when one compares them to forecasts coming from VAR models that only consider past GDP. Also, financial variables offer the possibility to design risk-based scenarios in a rather simple way. In this respect, the results of a simulation carried out under the assumption

9780230_243231_07_cha06.indd 162

5/11/2010 5:03:03 PM

US-Euro Area Interactions over the 2007–9 Crisis

0.015

163

United States

0.01 0.005 0 −0.005 Latest available GDP figure

−0.01

Worse financial conditions seriously hamper the outlook. The scenario envisages a stock market drop by 20% and a rise in volatility by 50% relative to their levels at end-September 2009.

−0.015 −0.02 2007

2008

2009

2010

2011

Actual Scenario of worse financial conditions 0.015

Euro area

0.01 0.005 0 −0.005

Latest available GDP figure

−0.01 −0.015

Relative to the US, worse growth outlook under the scenario envisaging a stock market drop by 20% and a rise in volatility by 50%, relative to their levels at end-September 2009.

−0.02 −0.025 −0.03 2007

2008

2009

2010

2011

2012

Actual Scenario of worse financial conditions

Figure 6.18 Forecasts of US and euro area GDPs as of 2009Q3 under a scenario of worse financial conditions (quarter-on- quarter change, %) Sources: ECB staff calculations, Global Financial Data and Consensus Economics.

of a new deterioration in financial conditions seem to be sensible and encourage further investigation.

6.4 Why was the slowdown relatively more intense in the euro area than in the United States? That a shock originating in the United States – out of domestic excesses and financial market related distortions – could have such a strong impact across the world and particularly on the euro area, was

9780230_243231_07_cha06.indd 163

5/11/2010 5:03:03 PM

164

Catching the Flu from the United States

Box 6.1 A comparison of the forecasting performance of the stock market volatility vs the corporate bond spread To supplement results of Section 6.3, in this box we compare the GDP forecasts based on models that include the stock market index and its time-varying volatility to the same forecasts based on a model which includes the stock market index and the corporate bond spread. The latter has been a traditional regressor in forecasting business cycle phases, jointly with the slope of the yield curve, as measured by the difference between the ten-year and the three-month rates. The corporate bond spread is the differential between the Baa-rated industrial bond yield with a ten-year maturity and the corresponding Government bond yield. For the euro area, German data have been used, due to their availability over the very long sample employed in the econometric application, that is, 1970–2009. In the end, as the stock market index is always included as a regressor, the comparison involves the information contained in the stock market volatility versus the corporate bond spread. The comparison between the two information sets was carried out for the US GDP only and relative to the forecasts made in 2009Q2 and in 2009Q3. Results are reported in the Figure 6.19. As regards 2009Q2, the stock market volatility and the corporate bond spread provided approximately the same results in the very short term, while the corporate spread was anticipating a much shallower growth profile starting from the beginning of 2010. As of 2009Q3, however, the two variables were providing a very consistent picture for the US growth profile. 0.015

Point at which forecast was made

0.01 0.005 0 −0.005 −0.01

Corporate bond spread provides a similar information as stock market volatility in the very short term but current values lead to less sustained output profile since beginning of 2010

−0.015 −0.02 −0.025

United States 2007

2008

2009

2010

2011

Actual Based on stock market index and its volatility as of 2009Q2 Based on stock market index and the corporate bond spread as of 2009Q2

Figure 6.19 US GDP forecasts: Comparing results based on the stock market index and its volatility to those based on the stock market index and the corporate bond spread, as of 2009Q2 and 2009Q3 (quarter-on-quarter change, percentage) Sources: ECB staff calculations, Global Financial Data and Consensus Economics.

9780230_243231_07_cha06.indd 164

5/11/2010 5:03:03 PM

US-Euro Area Interactions over the 2007–9 Crisis

0.02

Volatility-based information returns in line with corporate bond spread

Point at which forecast was made

0.015

165

0.01 0.005 0 −0.005 −0.01 −0.015 −0.02

United States 2007

2008

2009

2010

2011

2012

Actual Based on stock market index and the corporate bond spread as of 2009Q2 Based on stock market index and its volatility as of 2009Q2

Figure 6.19

Continued

Overall the brief evidence provided in this box suggests that while the stock market volatility and the corporate bond spread provide a consistent picture for future GDP developments, differences across quarters would require a more formal assessment of the forecasting power as well as the quality of the signals in terms of future growth.

certainly unexpected at first. The reason that it happened, however, is less puzzling now, as the shock has been clearly global in nature, and as the transmission channels, from mainly financial have become more trade related. This section accordingly elaborates such factors in order to establish why the intensity of the euro area slowdown was larger than in the US. In so doing, the objective is to draw lessons on possible future patterns of the respective business cycles going forward. 6.4.1 Structural factors Strong ‘traditional’ differences exist in the structure of the euro area and US economies, thereby explaining the relative macroeconomic performance in both economic areas in this downturn. One aspect is the larger share of manufacturing in output along with trade openness in the euro area. In particular, there has been an abnormally strong correction in industrial sector activity (see Figure 6.20) in the current recession. Survey data suggest that since the onset of financial turmoil in August 2007, the manufacturing sector has fared worse than services (see Figure 6.21) – a development which is not unusual given its traditionally higher cyclical volatility compared with the services sector.

9780230_243231_07_cha06.indd 165

5/11/2010 5:03:04 PM

166

Catching the Flu from the United States 15 10 5 0 −5 −10 −15 −20 1986

1991

1996 Euro area

Figure 6.20

2001

2006

United States

Industrial production index (annual percentage change)

Note: Last observation is September 2009. Sources: Eurostat and Federal Reserve.

The combination of housing market corrections, household’s balance sheet adjustments and difficulties in short-term financing has contributed to the collapse of the demand for durable goods (particularly in the automobile sector), which constitutes a large share of manufacturing output. The worldwide fall in confidence and the inventory adjustment process have further aggravated the dynamics of the downturn, leading to a severe and synchronous contraction in industrial production. The sharp slowdown in industrial activity has implied that countries with a stronger manufacturing base have tended to be harder hit in the ongoing slowdown. Indeed, the euro area has a relatively larger manufacturing sector share in overall value added than the US (see Figure 6.22). This difference is partly counterbalanced by a slightly smaller share of financial intermediation services as a percentage of total value added in the euro area – a sector that has also been strongly impaired in the current recession. The sectoral dynamics described above also have a counterpart in the strong trade adjustment in the current downturn. Despite increasing services content over recent years, trade is predominantly concentrated in goods, thereby implying a stronger impact on industrial production

9780230_243231_07_cha06.indd 166

5/11/2010 5:03:04 PM

US-Euro Area Interactions over the 2007–9 Crisis

167

0 −5 −10 −15 −20 −25 −30 −35 −40 −45

Services

Manufacturing

Euro area

United States

Figure 6.21 Cumulative deterioration in PMI between August 2007 and March 2009 (percentage change) Note: Last observation in December 2009. Sources: Markit, US ISM.

(a) Euro area, percentage of total value added 6% 6%

(b) US, percentage of total value added 8% 5%

19% 14% 4% 65%

Other services Construction Other Figure 6.22

5%

68%

Financial Intermediation Total Manufacturing

Breakdown of value added in 2005

Notes: Other services include: Real estate, rental and business activities, wholesale and retail trade, hotels and restaurants, transport storage and communication, community social and personal services. Other includes: Agriculture, hunting, forestry and fishing, mining and quarrying, electricity, gas and water supply. Source: EU KLEMS database – March 2008.

9780230_243231_07_cha06.indd 167

5/11/2010 5:03:04 PM

168

Catching the Flu from the United States

compared with service sector activity. The euro area also has a higher degree of openness than the US – defined not only in terms of trade, but also in terms of capital flows (both foreign direct and portfolio investment), as well as trade barriers (see Figure 6.23) and actual exposure to the external environment.10 The global nature of the slowdown has had a strong effect on trade, which has hit countries with high degree of external openness (and a strong manufacturing base) particularly hard. Consistent with this, euro area export growth began to moderate in 2006 and has recently become a large drag on growth, more strongly so than in the US. 6.4.2 Transmission channels A second reason for the stronger impact on activity in the euro area is that the transmission channels have changed over time, intensifying the size of business cycle fluctuations and the persistence of shocks. While this does not necessarily explain the relative strength of the euro area reaction, it might explain the relative speed of the transmission of shocks on euro area activity.

85 80 75 70 65 60 55 50 1970

1975

1980

1985

1990

Euro area

1995

2000

2005

USA

Figure 6.23 International openness (summary measure of ‘economic globalisation’, index) Notes: Summary economic globalisation measure obtained on the basis of a weighted measure constructed using actual flows (trade flows, foreign direct investment flows and stocks, portfolio investment and income payments to foreign nationals) as well as hidden import barriers (mean tariff rate, taxes on international trade and capital account restrictions). Euro area obtained using 2007 GDP weights. Source: KOF Swiss Economic Institute.

9780230_243231_07_cha06.indd 168

5/11/2010 5:03:04 PM

US-Euro Area Interactions over the 2007–9 Crisis 32

169

16 14

28

12 10

(%)

24

8 20

6 4

16

2 0

12 1980

1985

1990

1995

2000

2005

Index of World Vertical Supply Integration (right)* World Trade/World GDP (left) Figure 6.24

Index of vertical integration (lhs: percentage; rhs: index, 1975=1)

Source: Amador and Cabral (2009).

First, as mentioned in Chapter 3, Dees and Saint-Guilhem (2009) find that US cyclical developments may have become more global over time.11 While a change in US GDP has weaker contemporaneous impacts on activity in many countries during most recent periods than for earlier periods, shocks originating in the US have become increasingly persistent over time. This, together with an increase in the impact elasticity of non-US foreign activity for some regions (emerging in particular), emphasises the role of second-round and third partners’ effects, making US cyclical developments more global. In the face of shocks with a large global element, increasingly complex international supply chains can magnify the impact on activity via a sharper contraction in trade. Second, there is reason to believe that the globalisation process, and particularly the higher integration of industrial activity across borders, has changed the relationship of trade versus economic activity, thus possibly increasing trade fluctuations for a relatively more open economic area such as the euro area. Amador and Cabral (2009) show such a process by constructing an index of vertical supply integration, which is highly correlated with trade (Figure 6.23), as complex international networks require numerous production stages across borders.

9780230_243231_07_cha06.indd 169

5/11/2010 5:03:05 PM

170 Catching the Flu from the United States

As a result, for instance, Freund (2009) estimates that the elasticity of world trade to world income has increased over recent decades, from 1.9 in the 1960s to 3.7 in the current decade. In addition, the response of trade is estimated to be even higher during global downturns, at around 4.7. But the overall elasticity of trade to income depends also on the relative importance of expenditure categories at high import intensities. For this crisis, Anderton and Tewolde (2009) show that a structural imports function which captures the different and time-varying import-intensities of the components of total final expenditure – consumption, investment, government expenditure, exports, etc – can partly explain the recent sharp decline in global imports of goods and services. Table 6.3 shows a range of estimates of the import intensities of expenditure components based on various specifications.12 While the average total demand elasticity of imports is between 1.5 and 1.7, the individual elasticities can vary significantly, with private consumption (con) and government expenditures (gov) being lower than the average, and investment and exports higher. Also, as the share of high elasticity components in total final expenditure – and particularly exports – has increased over recent years, this has also contributed to raising the overall effective elasticity of trade to change in demand. The high and rising import intensity of exports also reflects and captures the rapid growth in ‘vertical specialisation’, suggesting that widespread global production chains may have amplified the downturn in world trade and partly explaining its high degree of synchronisation across the globe. The results are also consistent with the stylised facts that the contraction in global trade was especially pronounced in intermediate and capital goods.

Table 6.3 Import intensities of expenditure components conex govex invest expgs

[1.3–1.7] [1.0–1.3] [1.6–1.8] [1.8–1.9]

average

[1.5–1.7]

Note: Range of estimates based on various specifications. Source: ECB staff.

9780230_243231_07_cha06.indd 170

5/11/2010 5:03:05 PM

US-Euro Area Interactions over the 2007–9 Crisis

171

The panel estimates also find that stockbuilding, business confidence and credit conditions also played a role in the sharp contraction in global imports. Third, as seen in Chapter 5, financial markets have also become more intertwined over time, as indicated by an increase in the gross volume and diversity in the composition of capital flows and the increasing correlation of equity markets (Figure 6.25).13 In the current episode in particular, financial innovation has allowed the securitisation of US mortgages into asset-backed securities, which were partly sold to international investors, without the ratings of these products fully reflecting the inherent risks. This allowed US-specific risks to be spread globally, leading to a particularly strong transmission of US mortgage-related credit problems to the rest of the world.14 Fourth, globalisation appears to have been an important feature in the banking sector, as illustrated by the strong increase in bank’s external claims (Chapter 2). As a response to the financial crisis, financial institutions cut lending and sold other assets to reduce the size of

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 conex

govex Start period

invert

expgs

End period

Figure 6.25 Import elasticities of expenditure components (%, start period: 1995; end period: 2008) Note: Expenditure component elasticity refers to import intensity weighted by the share of the various components in total final expenditure. Source: ECB staff.

9780230_243231_07_cha06.indd 171

5/11/2010 5:03:05 PM

172

Catching the Flu from the United States 2500 2000 1500 1000 500 0 1980

1984

1988

1992

1996

United States Figure 6.26

2000

2004

2008

Euro Area

Equity indices (stock market price indices, base = 1 January 1980)

Note: Last observation: 21 December 2009. Source: Datastream.

their balance sheets (deleveraging). International bank lending started falling sharply in 2008Q4, as external claims fell by about 7% of the total amount outstanding (Figure 6.26). The composition is also worth noting: BIS banks withdrew mostly international loans while maintaining – broadly unchanged – their holdings of debt securities and other positions. The global dimension of this deleveraging makes it an important channel for propagation of the financial crisis worldwide. Finally, confidence linkages across economies may also have become stronger over time: the correlation between euro area and US consumer confidence has increased from 0.68 for the period 1985–99 to 0.90 in the more recent period (2000 to April 2009). In this downturn in particular, confidence indicators have been highly synchronised across countries (see Figure 6.27 for the US and the euro area), which could to some extent reflect a faster spreading of worldwide news across the globe compared with the past, with negative news abroad more strongly affecting domestic confidence.15 6.4.3

Policy responses

Both fiscal and monetary policy responses to the downturn have also differed between the two areas, partly elicited by the timing of the slowdown in each. First, in 2008, the fiscal deficit in the US increased by slightly more than 3 pp relative to the previous year, whereas it rose by 1.3 pp in the euro area (Figure 6.28). For 2009 and 2010, the cumulative fiscal impulse over the two years, as reflected by the

9780230_243231_07_cha06.indd 172

5/11/2010 5:03:05 PM

US-Euro Area Interactions over the 2007–9 Crisis 2500 2000 1500 1000 500 0 −500 −1000 −1500 −2000 −2500 1999

2001

Debt securities

2003 Loans

2005

2007

Other positions

2009

173

2500 2000 1500 1000 500 0 −500 −1000 −1500 −2000 −2500

Total positions

Figure 6.27 Changes in total external claims by instruments (evaluated at constant exchange rates, USD billions) Note: Locational data, based on resident principle. Last observation: 2009Q2. Sources: BIS and ECB calculations.

change in the overall deficit ratio, is estimated at 5.9 pp for the U.S. and 4.6 pp for the euro area, on the basis of the Commission spring forecast. For the US, this assumes a total of 2.4 pp bank rescue measures that are counted as expenditure by the Commission but will not affect aggregate demand.16 The increase in the deficit ratios is driven by the operation of automatic stabilisers and non-automatic policy changes, including the impact of specific fiscal stimulus packages. While a quantification of the relative impact of these channels is difficult (not least because it requires real-time output gap estimates), rough estimates suggest that the operation of automatic stabilisers is more important in the euro area, whereas discretionary measures play a larger role in the US. Second, the timing: the monetary policy easing by the Federal Reserve had already started in September 2007, partly reflecting earlier signs of a moderation in US economic activity (Figure 6.29), whereas there were persisting inflationary pressures in the euro area. Lags in the effects of these policies imply that, in part, their effects could be concentrated in the coming quarters and thus would influence the timing and shape of the respective recoveries.

6.5

What is the role played by real estate cycles?17

One of the key triggers of the current global downturn was a correction in housing markets, involving rapid adjustments in house prices

9780230_243231_07_cha06.indd 173

5/11/2010 5:03:06 PM

174

Catching the Flu from the United States 3 2 1 0 −1 −2 −3 −4 1985

1990

1995 Euro area

Figure 6.28 indices)

2000

2005

United States

Consumer confidence (consumer confidence surveys, standardised

Note: Last observation is November 2009. Sources: EU Commission and the Conference Board.

1 0 −1 −2 −3 −4 −5 −6

2008

2009 Euro Area

2010

2011

US

Figure 6.29 Fiscal policy (general government balance, annual change, percentage point of GDP) Sources: European Commission, 2009 Autumn Forecast.

and housing activity not only in the US and the euro area, but also in several other developed economies. For the US, in particular, Leamer (2007) had already argued that residential investment offers the best early warning sign of an oncoming recession of any expenditure component of GDP. This section assesses whether a similar assertion can be made for the euro area as well on the basis of his methodology, particularly for the most recent episode.

9780230_243231_07_cha06.indd 174

5/11/2010 5:03:06 PM

US-Euro Area Interactions over the 2007–9 Crisis

175

7 6 5 4 3 2 1 0 1999

2001

2003 US

2005

2007

2009

Euro area

Figure 6.30 Monetary policy (US and euro area key central bank policy rates, (%, daily data) Note: Latest observation is 18 December 2009. Source: Bloomberg.

6.5.1 Some stylised facts Historically, housing investment in the euro area and in the US recorded similar dynamics with strong declines and subsequent recoveries around recessions As a share of GDP, housing investment has been lower and higher in the US compared with the euro area, particularly in the recent period (see Figure 6.31). 6.5.2 Measuring ‘abnormal contributions’ of housing investment to business cycles a) Procedure ●

While the contribution of housing investment to overall economic activity appears to have been generally weak and declining in the euro area and even weaker on average in the US over the last four decades, it is possible that housing activity has nonetheless played a role in driving economic cycles, and in particular in triggering cyclical turning points. This is the hypothesis successfully tested by Leamer (2007) for the US and which we have also extended to the euro area. Following his strategy, Gattini and Hiebert (2009) calculate ‘abnormal’ contributions of housing investment to GDP growth and we verify whether such contributions are particularly large at the time the cycles turn up or down. The procedure is based on three steps: First, ‘normal’ contributions to GDP growth are computed by applying

9780230_243231_07_cha06.indd 175

5/11/2010 5:03:06 PM

176

Catching the Flu from the United States 12 10 8 6 4 2 1970

1975

1980

1985

1990 EA

1995

2000

2005

US

Figure 6.31 Evolution of housing investment share of economic activity in the euro area and US (percentage, nominal variables) Note: Data prior to 1995 for the euro area were obtained on the basis of national sources for the five biggest countries. Sources: Eurostat, AWM database, and BEA (US).

●

●

a filter to the actual contribution of residential investment to quarterly real GDP growth. This step is designed to obtain estimates of the long-run trend of the real housing investment contribution to real GDP growth. To this end, an HP filter is applied to the real housing investment contribution to real GDP growth, with smoothing parameter ␭ set in such a way as to replicate the profile of the series for the US obtained on the basis of a kernel regression smoother used in Leamer (2007) – implying a very high value of 25,000 (with a very smooth resulting series).18 The results indicate a low and generally positive contribution of residential investment to GDP growth in the euro area, compared with a more marked trend in US housing investment contributions. Second, ‘abnormal’ contributions are obtained as a difference between actual and ‘normal’ contributions. Cumulating the results (which are centred around zero by construction) we obtain a fairly subdued profile for the euro area as a whole compared with the US and UK (see Figure 6.31). Third, around each recession, data related to four quarters preceding a peak and eight quarters afterwards are collected and normalised (by subtracting the value at the GDP cycle peak). Peaks and troughs in activity are obtained from the CEPR business cycle

9780230_243231_07_cha06.indd 176

5/11/2010 5:03:07 PM

US-Euro Area Interactions over the 2007–9 Crisis (a)

6

177

Euro area, the US and the UK

4 2 0 −2 −4 −6 −8 1973

1977

1981

1985

Euro area (b)

6

1989

1993

1997

United States

2001

2005

United Kingdom

‘Big 5’ euro area countries

4 2 0 −2 −4 −6 −8 1973

1977

Germany

1981

1985

France

1989 Italy

1993

1997 Spain

2001

2005 Netherlands

Figure 6.32 ‘Abnormal’ GDP growth contributions of residential investment (percentage, cumulated) Note: Cumulated sum of the difference between actual and ‘normal’ GDP contributions of housing investments. Last observation is 2008Q2. Source: ECB calculations based on Eurostat data.

9780230_243231_07_cha06.indd 177

5/11/2010 5:03:07 PM

178 Catching the Flu from the United States

dating committee for the euro area19 and from the IMF for the other countries.20 The resulting cumulative abnormal contribution to GDP growth can be interpreted as normal (a flat line), abnormal weakness or abnormally declining strength (if the line is declining), or abnormal strength or abnormally subsiding weakness (if the line is rising). b) Results For the euro area, the Leamer hypothesis tends to hold only for the current recession and – even more strongly – for the 1970s episode (see Figure 6.33). This contrasts starkly with the US and the UK where abnormal residential investment weakness has anticipated all recessions since the 1970s. Important differentiations emerge however across major euro area countries (Figure 6.34). On the other hand in some countries (such as Spain since the late 1970s, the Netherlands in general and Germany in the early 1970s and early 1990s) housing investments appear to have

Euro Area 2.5 1.5 0.5 −0.5

1974

1980

1992 2008

−1.5 −2.5

2.5

2.5 2 1.5 1 0.5 0 −0.5 −1 −1.5 −2 −2.5

US

1973 1980 1981 1990 2001 2007

UK

1.5 0.5 −0.5

1973 1974 1979 1990 2008

−1.5 −2.5 Figure 6.33 ‘Abnormal’ contributions to cyclical turning (percentage points; quarter-on-quarter annualised changes) (four quarters prior to pre-recession peak) Sources: OECD, ECB, Eurostat.

9780230_243231_07_cha06.indd 178

5/11/2010 5:03:07 PM

US-Euro Area Interactions over the 2007–9 Crisis DE

4.5

4.5

3.5

3.5

2.5

2.5

1.5

1.5

0.5

0.5

−0.5

−0.5

1974 1981 1992 2002 2008

179

FR

1974 1981 1992 2002 2008

−1.5

−1.5

−2.5

−2.5

ES

4.5

IT 4.5

3.5

3.5

2.5

2.5

1.5

1.5

0.5

0.5

−0.5 −1.5

1974 1978 1980 1992 2008 −0.5 −1.5

−2.5

−2.5

1974 1981 1992 2001 2008

Figure 6.34 ‘Abnormal’ contributions to cyclical turning points in euro area countries (percentage points; quarter-on-quarter annualised changes) (four quarters prior to the pre-recession peak in activity) Sources: OECD, ECB, Eurostat.

strongly contributed to changes in cycles, similarly to the US and UK. On the other, in France and Italy, housing appears to have made limited abnormally strong contributions to recessions. While the ongoing correction in housing markets should allow some normalisation, the outlook for the housing sector appears gloomy.21 Historical experience across advanced countries suggests that the contraction phase of real house-price cycles tends to last for about five years, with the length and scope of the correction being larger when combined with banking crises.22 Of particular concern in this context is that the magnitude of the housing boom in the last years has exceeded what had been observed in past cycles, suggesting that the ongoing corrections might be even more severe and protracted than before. The ongoing correction in this sector also implies that, contrary to the recovery experienced earlier this decade, construction investment is unlikely to provide a positive contribution to GDP growth for some time.

9780230_243231_07_cha06.indd 179

5/11/2010 5:03:07 PM

180

Catching the Flu from the United States

6.6 Has the EMU helped the euro area weather the financial turmoil? The previous section has shown how country-specific developments can be different within the euro area. In line with their housing market booms, countries like Spain or the Netherlands were growing faster than the euro area as a whole prior to the financial crisis. The euro area recession also masks some country divergences. While the timing of the economic downturn in 2008Q4 and 2009Q1 was relatively synchronised across euro area countries, the extent of the output loss was more severe in Italy and Germany compared with the area as a whole (Figure 6.35). More specifically, some differentials in real output growth in the euro area, as in any currency union, are natural, for instance those related to catching-up effects occurring when a country’s income level is catching up with those of other more developed countries in the monetary union. Other differentials may present some challenges if protracted, particularly if they reflect inappropriate national economic policies, structural rigidities or malfunctioning adjustment mechanisms. After

8 6 4 2 0 −2 −4 −6 −8 2000

2001

2002

2003

Belgium Italy

2004

2005

Germany Netherlands

2006

2007

Spain Euro area

2008

2009

France

Figure 6.35 Real GDP growth rates in the main euro area countries (year-onyear growth rates, %) Note: Latest observation is September 2009. Source: European Commission.

9780230_243231_07_cha06.indd 180

5/11/2010 5:03:08 PM

US-Euro Area Interactions over the 2007–9 Crisis

181

reviewing some features of the cross-euro area countries’ business cycles, this section assess to what extent the European Monetary Union may have modified pre-existing patterns of transmission of external shocks and in particular, to what extent the latter has facilitated or aggravated their absorption. To measure how EMU could have influenced the individual business cycle performance, Table 6.4 reports average growth and its variance before and after the inception of EMU. The table also includes a test on whether the numbers are significantly different across periods (Giannone et al., 2009). This test is constructed by comparing the measure computed using the observed post-EMU data and the distribution of the measures obtained by using block bootstrap over the preEMU period. For most countries, the average rate of growth has been lower during the EMU period. However, the difference is not significant, except for Austria and Italy. The same is true for the variance,

Table 6.4 Annual growth rates of real GDP per-head

Countries

Euro area Germany France Italy Spain Netherlands Greece Belgium Portugal Austria Finland Ireland Luxembourg

Average growth fate Pre EMU 2.24 2.21 2.07 2.35 2.4 2.03 1.71 2.2 3.04 2.5 2.35 3.85 3.00

EMU 1.59 1.3 1.5 0.92** 2.38 1.68 3.8 1.8 1.07 1.66* 2.99 4.69 3.76

Variance growth rate Pre EMU 2.3 2.64 2.7 3.96 4.62 2.36 12.29 3.29 14.03 3.01 9.57 7.9 11.48

EMU 1.27 1.58 1.05 2.13 1.3 2.47 0.28*** 1.37 2.68 1.3 1.56 5.39 4.42

Notes: The table reports 1) the average real GDP per capita growth rate and 2) the variance of the growth rate of the euro area and the 12 countries under study. One (two, three) star(s) indicates that the EMU values are significantly different from those in the pre-EMU period at 10 (5, 1) % confidence level. The test is constructed by comparing the measure computed using the observed post-EMU data and the distribution of the measures obtained by using block bootstrap over the pre-EMU period. Statistical significance has been assessed by using block bootstrap, with blocks of two years length. Source: Giannone et al. (2009).

9780230_243231_07_cha06.indd 181

5/11/2010 5:03:08 PM

182

Catching the Flu from the United States

which has decreased everywhere, but significantly only for Greece (it should be recalled that numbers for Greece are not very reliable). Giannone et al. (2009) show that asymmetries are very small for countries with similar levels of development and larger for countries with low GDP per capita relative to the euro area. Asymmetries have declined over time as an effect of lower output volatility in the early 1980s (the ‘Great Moderation’). Since asymmetries have changed very little as a consequence of EMU, the costs of business cycle heterogeneity associated with it have been small. Moreover, it is shown that some idiosyncrasies are definitely present in generating euro area countries’ business cycles and, in general, they have not decreased over time. However, they remain confined to the experience of small countries, both before and after the introduction of the common currency. Given the uncertainty, any statement on the real effect of EMU in these countries is likely to be ill founded. In a monetary union such as the euro area, with a single currency and a single monetary policy, the main adjustment mechanism – in the absence of a high degree of labour mobility and fiscal transfers across countries – is the cross-border trade or competitiveness channel.23 Available evidence shows that the competitiveness channel appears, as a result of structural rigidities and a lack of full implementation of the Single Market, to require a relatively long period to work through in the euro area. This implies that in response to asymmetric shocks, larger price and inflation differentials, as well as higher regional unemployment, may be observed than would be the case with a higher degree of cross-border competition and economic integration. At the same time, the euro has made an important contribution to economic and financial integration within Europe. The single currency has enhanced the ability to take advantage of the single market, and trade and capital ties between euro area countries have grown significantly. For example, cross-border trade in goods and services in the euro area has increased by 10 percentage points as a share of GDP since the introduction of the single currency. Trade between individual euro area countries now accounts for about half of their total imports and exports. At the same time, trade with countries outside the euro area has also developed very dynamically. The euro has also promoted competition, price transparency and price convergence. It has lowered transaction costs and eliminated exchange rate risk. The euro has boosted direct investment within the euro area. Mergers and acquisitions in the euro area have also risen noticeably (Figure 6.36).

9780230_243231_07_cha06.indd 182

5/11/2010 5:03:08 PM

US-Euro Area Interactions over the 2007–9 Crisis

183

Very few counterfactual experiments are available as regards the costs and benefits of EMU membership. Using a GVAR model, Dubois et al. (2009) show that most euro area countries exhibit output gains thanks to EMU membership. This is particularly so for a group of small opened economies, for whom the benefits of the single currency have possibly exceeded the costs. This outcome could be attributed to the credibility gains for monetary policy (translated into lower interest rates) and the increase in intra-euro area trade allowed by the single currency. Conversely, there is some evidence of EMU-related output loss for large countries like France, Germany or Italy, although this result appear less clear cut. Pesaran et al. (2007) also use the GVAR model to provide a counterfactual analysis as regards the entry of the UK and Sweden in the EMU. They show that the UK entry into the euro area would probably have reduced UK output in the short term, but raised it in the medium term. UK entry would have also probably caused euro area GDP to be higher in the medium term. A similar analysis for Sweden shows that entry to the euro in 1999 would probably have increased Swedish output. Beyond the effects of EMU in terms of integration and economic performance, the adoption of the single currency has also provided a

14 12

(%)

10 8 6 4 2 0 Intra-euro area 2001

2002

Extra-euro area 2003

2004

2005

2006

2007

Figure 6.36 Cross-border holdings as a share of total holdings of short-term debt securities issued in the euro area (percentage) Sources: BIS, IMF and ECB calculations.

9780230_243231_07_cha06.indd 183

5/11/2010 5:03:08 PM

184

Catching the Flu from the United States

shield from financial asset volatility. The benefits of the euro in the current crisis should not be underestimated. Actually, without the euro, the euro area countries would be facing potentially large fluctuations between Europe’s national currencies, in addition to the financial market turbulence, the problems in the financial sector and the economic downturn. The currencies of some countries outside the euro area have experienced considerable fluctuations during the current crisis (Figure 6.37). Both the Swedish krona and the pound sterling have declined substantially in value since September 2008 and some of the most recent countries to join the EU have seen even larger fluctuations in their currencies’ purchasing power. This is an additional source of uncertainty for these countries. The current turmoil has also confirmed an advantage the euro has, namely that, in stormy seas, it is better to be aboard a large ship than a small boat (Tumpel-Gugerell, 2009). The euro has had a very important stabilising effect in difficult times.

6.7 Conclusion Overall, against both 1) the prevailing view prior to intensification of the crisis, that the downturn in the euro area would be far less 110 100 90 80 70 60 Sep-08

Nov-08

Jan-09

Euro area UK Sweden Figure 6.37 data)

Mar-09

May-09

Switzerland Norway Poland

Jul-09

Sep-09

Nov-09

Czech Republic Romania Hungary

Real effective exchange rates (index: September 2008=100; monthly

Note: Latest observation is November 2009. Sources: ECB, OECD and ECB calculations.

9780230_243231_07_cha06.indd 184

5/11/2010 5:03:08 PM

US-Euro Area Interactions over the 2007–9 Crisis

185

pronounced than in the US, and 2) what historical episodes would suggest, the current downturn in euro area activity occurred quickly and was deeper than expected. The existence of financial shocks of a global nature, thereafter amplified by a world trade collapse, has led to more synchronised cycles across countries than suggested by historical patterns. Against this background, the next chapter uses the results of the backward-looking analysis to provide hints on possible medium- to long-term prospects for the global economy, focusing most notably on the US-euro area relationships.

Notes 1. When making comparisons, it is important to remember that the US has a higher potential growth rate, hence ideally the cumulated output losses relative to potential output should be compared. As in the current environment, estimates of potential output are highly uncertain and the output loss here is simply approximated by the sum of deviations from average growth, which is a rough approximation of the difference in the cyclical component of output during this downturn. This measure ignores, however, the fact that potential growth could also have changed during the course of the crisis. 2. For the euro area, the CEPR reckons that calling the start of this recession was difficult, as GDP peaked in 2008Q1, while employment continued to rise into 2008Q2. However, the CEPR also based its assessment on employment figures for the full year as well as on monthly figures for industrial production. Note also that euro area GDP growth in 2008Q1 was assessed to have been partly boosted by construction output because of the unusually mild weather conditions. 3. There are indeed a number of econometric trade models that, despite their relative high demand elasticities, predict, based on the observed fall in output, a less pronounced decline in trade flows in 2008Q4. This is the case for the models estimated by the OECD (see Pain et al., 2005) and the ECB (Fagan et al., 2001), which relate trade flows to demand (foreign and domestic) and relative prices. 4. In addition, high quality products were affected more than low quality goods, leading to a sharper decline in trade values. As pointed by Berthou and Emlinger (2009), the role of product quality matters, especially during recessions, with high quality imports being more responsive to income changes. This also explains why world trade values fell faster during the crisis than world trade volumes. 5. McKibbin and Stoeckel (2009) claim that the shocks observed in financial markets can be used to generate the severe economic contraction in global trade and production experienced in 2009. In particular, the distinction between the production and trade of durable goods and non-durable goods plays a key role in explaining the much larger contraction in trade than GDP experienced by many economies. 6. Based on the latest trade data in values, as the intra-extra trade breakdown for 2009Q3 is not yet available in volume terms.

9780230_243231_07_cha06.indd 185

5/11/2010 5:03:08 PM

186

Catching the Flu from the United States

7. As also stressed in the previous chapter, this approach to the use of financial variables in forecasting gives them a rather passive role as determinants of the business cycle, as the possible way they can convey information about activity relates to the fact that asset prices are determined in a forwardlooking manner. Yet financial markets can also have a more active role in affecting real activity, through their impact on the balance sheets of firms and banks, that is, limiting access to credit for the former and the willingness to lend for the latter. In the setup of this note, which is based on reduced-form VAR model, the two roles are of course not distinguishable. 8. It is key to highlight that the forecasting exercises reported in this note involve only linear models. Another branch of literature focuses instead on non-linear models, namely logit or probit regressions, aimed at forecasting the business cycle phase, that is, expansion or contraction. For recent applications see Bellego and Ferrara (2009) and Fornari and Lemke (2009). 9. Other types of financial variables have been employed in Espinoza et al. (2009), among them the Ted spread, the distance to default of a number of financial firms and C&I loans, all of which provide overall similar forecasting results. 10. Being a sum of very open economies, the euro area economy has a degree of ‘effective’ openness, which is actually larger than aggregated euro area statistics would suggest (see Anderton and di Mauro, 2007). 11. See Dees and Saint-Guilhem (2009). 12. The import intensities of the expenditure components reported in Table 6.3 are based on estimates which adjust for the weights of the individual components in total final expenditure, hence – unlike elasticities – they reflect the relative importance of imports for each component uncontaminated by their differing weights in total final expenditure. 13. See Bourguignon et al. (2002). 14. See Hoffmann and Nitschka (2008). 15. Evidence of a sharp rise in the use of the Internet supports the argument of the faster spread of worldwide news across the globe: world Internet usage increased by around 340% between 2000 and 2008, while world population growth has only been 10%, indicating a sharp rise in the internet penetration. See http://www.internetworldstats.com/stats.htm. 16. The headline numbers of the US general government deficit in Figure 6.17 includes bank rescue measures; the figure therefore exaggerates the differences in policies in the euro area vs. US. 17. We are very grateful to our ECB colleagues, Paul Hiebert and Luca Gattini for having provided the material used in this section. 18. As indicated in Leamer (2007), several methods exist for filtering the signal from the noise in these data, including the kernel regression smoother, exponential smoothing, moving averages, the Kalman filter and the Hodrick-Prescott filter. He adopted the kernel regression smoother given a desire not to track the short-term movements in the data too closely in order to uncover a long-term moving trend – approximated by a high value for the smoothing parameter then running an HP filter. 19. This methodology was applied to all business cycle turning points since 1970, implying four to five recessions for each of the five biggest euro area countries – which amounts to 22 country-specific recessions.

9780230_243231_07_cha06.indd 186

5/11/2010 5:03:08 PM

US-Euro Area Interactions over the 2007–9 Crisis

187

20. These values are determined using a simplified Bry-Boschan (1971) dating algorithm. 21. See for instance, ECFIN Economic Brief, Issue 4 (July 2009). 22. See for instance, Girouard et al. (2006) or Reinhard and Rogoff (2009). 23. The competitiveness channel works as follows: following a wage shock, or a shock that drives a country’s output above its potential, domestic inflationary pressures – on wages and other domestic costs – will give rise to a deterioration in external competitiveness. This will, in turn, gradually reduce foreign demand for the country’s exports, such that lower external demand will tend to restore output to its potential level and to dampen previous inflationary pressures.

References Amador, J. and S. Cabral, 2009, ‘Vertical Specialisation across the World: A Relative Measure’, North American Journal of Economics and Finance, 20, 267–280. Anderton, R. and F. di Mauro, 2007, ‘The External Dimension of the Euro Area’, Cambridge University Press. Anderton, R. and T. Tewolde, 2009, ‘Turmoil, Global Trade and the Internationalisation of Production’, Paper Presented at Conference ‘The Global Financial Crisis’, University of Nottingham, China, 10–11 November 2010. Bellégo, C. and L. Ferrara, 2009, ‘Forecasting Euro Area Recessions Using TimeVarying Binary Response Models for Financial Variables’, Banque de France, mimeo. Berthou, Antoine and Charlotte Emlinger, 2009, ‘Crisis, Trade Collapse, and the Decrease of Import Prices’, Mimeo. Bourguignon, F., D. Coyle, R. Fernàndez, F. Giavazzi, D. Marin, K. O’Rourke, R. Portes, P. Seabright, A. Venables, T. Verdier and L.A. Winters, 2002, ‘Making Sense of Globalization: A Guide to the Economic Issues’, CEPR Policy Paper No. 8. Bry, G. and C. Boschan, 1971, ‘Cyclical Analysis of Time Series: Selected Procedures and Computer Programs’, NBER Technical Paper No. 20. Dees, S. and A. Saint-Guilhem, 2009, ‘The Role of the United States in the Global Economy and Its Evolution over Time’, ECB Working Paper No. 1034. Dubois, E., J. Héricourt and V. Mignon, 2009, ‘What If the Euro Had Never Been Launched? A Counterfactual Analysis of the Macroeconomic Impact of Euro Membership’, Economics Bulletin, 29 (3), 2252–66. Espinoza, R., F. Fornari and M. Lombardi, 2009, ‘The Predictive Power of Financial Variables for Economic Activity’, ECB Working Paper No. 1108. Fagan, G., J. Henry and R. Mestre, 2001, ‘An Area Wide Model (Awm) for the Euro Area’, ECB Working Paper No. 42. Fornari, F. and W. Lemke, 2010, ‘Predicting Recession Probabilities with Financial Variables over Multiple Horizons’, ECB Working Paper Series, forthcoming. Freund, Caroline, 2009, ‘The Trade Response to Global Downturns. Historical Evidence’, World Bank Working Papers No. 5015. Gattini, L. and P. Hiebert, 2009, ‘Has Residential Investment Made an Unusually Strong Contribution to Euro Area Activity’, ECB mimeo. Giannone, D., M. Lenza and L. Reichlin, 2009, ‘Business Cycles in the Euro Area’, ECB Working Paper No. 1010.

9780230_243231_07_cha06.indd 187

5/11/2010 5:03:08 PM

188

Catching the Flu from the United States

Girouard, N., M. Kennedy, P. van den Noord and C. André, 2006, ‘Recent House Price Developments: The Role of Fundamentals’, OECD Working Paper No. 475. Hoffmann, M. and T. Nitschka, 2008, ‘Securitization of Mortgage Debt, Asset Prices and International Risk Sharing’, Institute for Empirical Research in Economics Working Paper No. 376. Leamer, E.E., 2007, ‘Housing is the Business Cycle’, NBER Working Paper No. 13428. McKibbin, W. and A. Stoeckel, 2009, ‘The Potential Impact of the Global Financial Crisis on World trade’, World Bank. Melander, A., M. Orellana, D. Paternoster and M. Stierle, 2009, ‘At a Turning Point? Assessing the First Positive Signals for the Euro-Area Economy’, ECFIN Economic Briefs, Issue 4, July. Pain, N., A. Mourougane, F. Sédillot and L. Le Fouler, 2005, ‘The New OECD International Trade Model’, OECD Economics Department Working Paper No. 440. Pesaran, M., L. Hashem, Vanessa Smith and Ron P. Smith, 2007, ‘What If the UK or Sweden Had Joined the Euro in 1999? An Empirical Evaluation Using a Global VAR’, International Journal of Finance and Economics, 12 (1), 55–87. Reinhart, C. and K. Rogoff, 2009, ‘The Aftermath of Financial Crises’, American Economic Association Meetings in San Francisco, January. Tumpel-Gugerell, G., 2009, ‘Monetary Policy Challenges in Light of the Current Financial Market Developments’, Speech, Alpbach Talks Vienna, 15 June.

9780230_243231_07_cha06.indd 188

5/11/2010 5:03:09 PM

7 The US-Euro Area Relationship in a Context of Possible Systemic Changes

7.1

Introduction

Following the most severe and synchronised economic downturn since the Great Depression, the global economy entered into a phase of recovery, though somewhat shallow. For some 15 years, we had experienced an exceptionally long period of strong and stable growth in the world economy. But this high rate of growth was based in part on the emergence of large and unsustainable global imbalances. In principle, current account imbalances can be desirable, if they channel funds across the world to their most productive use. But in recent years, these imbalances have been a symptom of economic distortions: in a number of countries, asset-price bubbles supported a fall in private saving rates to very low levels, while household debt levels rose and consumer expenditure boomed. As a counterpart, other regions have pursued export-oriented growth strategies, implemented in part via managed exchange rate regimes that have kept the value of their currencies artificially low. These policies led to large current account surpluses and implied a vast accumulation of foreign exchange reserves with potentially high opportunity costs. Managed exchange rate regimes may have contributed to hampering necessary domestic adjustments by providing incentives for domestic savings and distorting the allocation of resources towards export-oriented industries. The build-up of global imbalances was exacerbated by abundant financial market liquidity and excessive risk taking. The latter reflected the apparently benign macroeconomic environment, as well as financial 189

9780230_243231_08_cha07.indd 189

5/11/2010 11:51:37 AM

190

Catching the Flu from the United States

sector innovation, which allowed the development of complex financial products whose risks turned out to be severely underestimated. At the same time, the institutional framework did not keep pace with the speed of innovation, giving rise to financial sector vulnerabilities. While global imbalances did not trigger per se the crisis – at least in the way it had been foreseen in most common risk scenarios – they certainly contributed to causing the vulnerabilities that eventually translated in an unprecedented global downturn. The financial crisis has opened up a new era of global economic interactions, as there are reasons to believe that a return to the pre-crisis growth model is both unlikely in the near term and undesirable. These changes are certainly relevant to assessing the relationships between the US and the euro area in the future. Two important aspects are likely to influence the global growth going forward, at least in the first years of the post-crisis. First, on the demand side, households in the United States, as well as in other countries that experienced debt-fuelled consumption booms, will be unable to return to the spending behaviour they exhibited prior to the crisis, at least in the medium term. Massive falls in the values of both financial and real estate assets imply that current ratios of household debt to personal income are not sustainable and higher personal savings rates will be required in the years to come. Retrenchment by debt constrained consumers has already begun and is likely to continue, as it will take a long time to rebuild the wealth destroyed during the recession. This crisis has made consumers aware of the tail risks of the economic environment. As a result, it is likely that – at least in the short to medium term – a structural shift in their behaviour will take place, which could result in a sustained reduction in the ratio of consumption to GDP. As for the supply side, the crisis may lead to a prolonged reduction in the level of potential output. This may occur via depressed capital accumulation, held back by a higher cost of capital and credit restrictions. The crisis may also have negative repercussions for labour markets. Structural unemployment may increase as the downsizing of some sectors (for instance, the financial, construction or car sectors in some regions) requires a reallocation of labour. People who are outside the labour market could lose skills, which could hamper their employability. The financial crisis may also leave its mark on productivity. On the downside, the impairment of the financial sector could hamper the channelling of funds to the most productive processes and R&D spending could be scaled back. Aggregate productivity could also rise,

9780230_243231_08_cha07.indd 190

5/11/2010 11:51:37 AM

Relationship US-Euro Area and Systemic Changes

191

however, as the recession creates incentives to restructure and improve efficiency, and removes inefficient firms from the market. This chapter will review how the above factors may affect the global economy and most notably the relationship between the US and the euro area economies. Section 7.2 will address how weaker consumption growth in countries that had experienced asset-price and debt-fuelled consumption booms might affect the world economic prospects and the euro area outlook. Section 7.3 will assess how potential growth may be affected, particularly in economies that have been hit most strongly by the financial crisis. Section 7.4 will show how these changes might affect the issue of global imbalances. Section 7.5 will conclude.

7.2 The role of the US economy as ‘consumer of last resort’ will be challenged going forward Because US households entered the current recession with an overhang of debt, one important aspect that will shape the US recovery is the need for them to repair their balance sheets and reduce the excessive debt burden. This will require a rise in the US personal saving rate, which is likely to have a negative impact on US consumption and growth in the medium term. 7.2.1 Why US should consumption weaken in the future?1 Over the last three decades, the ratio of US household debt to personal disposable income (PDI) has increased at a rapid pace, roughly doubling from about 65% in the early 1980s to a peak of 131% in 2008Q1. The growth of the debt ratio over this period has been associated with a process of financial innovation (e.g. the rise in loan securitisations and improvements in credit scoring techniques), which, combined, with the liberalisation of mortgage markets, eased borrowing constraints for households. From 2000 onward, the rise in the household debt ratio accelerated further relative to its previous trend, fuelled by a period of ‘easy credit’ which resulted in a mortgage lending boom and ultimately contributed to the US housing market bubble. The rapid growth of US household debt over the span of several decades, which allowed consumption to grow faster than income, was accompanied by a pronounced erosion of personal savings. The personal saving rate (relative to PDI) declined from levels of around 10% in the first half of the 1980s to an average of less than 2% over 2005–7. The rise in the debt-to-income ratio might have been a significant

9780230_243231_08_cha07.indd 191

5/11/2010 11:51:38 AM

192

Catching the Flu from the United States

contributing factor in explaining this long-term decline, as empirical findings suggest that the enhanced ability to borrow relaxes liquidity constraints on households and reduces their need for precautionary savings. The trends observed in US household borrowing and saving patterns have proved to be unsustainable, and balance sheet adjustments appear unavoidable for a number of reasons. First, the rise in household debt led to an increase in the debt service burden to record high levels by historical standards. This in turn resulted in an acceleration of defaults on mortgage and consumer loans. Second, the downturn in US housing markets from 2006 caused strong erosion in household real-estaterelated assets, thus lowering the value of collateral available to back up mortgage debt (roughly three-quarters of US household liabilities). A third and related factor has been the dramatic decline in US household wealth relative to PDI from 630% in 2007H1 to 481% in 2009H1, which was driven by the decline in both house and equity prices (see Figure 7.1). Empirical evidence strongly suggests that lower levels of wealth are associated with a tendency to save more out of disposable income (see Figure 7.2). As a result, since the beginning of the current economic and financial crisis we have witnessed a reversal of previous trends in borrowing and 700

650 US household net worth 600

550 Average 1980–2009Q3 500

450

400 1980 Figure 7.1

1983

1986

1989

1992

1995

1998

2001

2004

2007

US Household net worth (as a percentage of PDI)

Note: Last observation refers to 2009Q2. Source: Federal Reserve Board.

9780230_243231_08_cha07.indd 192

5/11/2010 11:51:38 AM

Relationship US-Euro Area and Systemic Changes

193

14

US personal saving rate

12 10 8 6 4 2 0 400

2009Q3 450

500

550

600

650

US household net worth Figure 7.2 Correlation: Household savings and wealth (as a percentage of PDI) Note: Observations from 1980Q1–2009Q2. Sources: Federal Reserve Board, BEA.

saving. The personal saving rate picked up from its previous lows to about 4% in 2009. At the same time, households made some progress towards deleveraging, with their debt-to-income ratio declining from a high of 131% in 2008Q1 to 126% in 2009Q2. This reduction is likely to have reflected repayment of debt through higher savings, as well as debt being written off due to higher default rates. The contraction in credit growth and the rise in the personal saving rate implied a sharp retrenchment in US consumer spending, contributing to the deep economic recession that started in December 2007. The recovery profile of US consumption and growth therefore crucially hinges upon the extent and timing of future deleveraging of household balance sheets. Although retrenchment by consumers has already begun, with some apparent repayment of household debt, this is likely to be a long-term process which may possibly lead to a structural shift in US consumers’ behaviour and a sustained reduction in the ratio of consumption to GDP. Our baseline projections envisage a further rise in the US personal saving rate to above 5% by the end of the projection horizon at end-2011. This implies a sustained period of subdued consumption growth which – given its share of about 70% of GDP – is likely to lead to a muted recovery by historical standards. Our calculations, based on a simple model of the relationship between household savings and debt, illustrate that the debt-to-income ratio is

9780230_243231_08_cha07.indd 193

5/11/2010 11:51:38 AM

194

Catching the Flu from the United States

likely to remain well above its long-term trend (approximated by the period 1980–2000) even at the end of 2011 (see Figure 7.3). This implies that the adjustment of US balance sheets is likely to be a medium-term process extending beyond the next several years. In order to gauge the timing and extent of the possible balance sheet consolidation, we extend the horizon until the end of 2019 and consider two new scenarios. Under scenario 1, the US households saving rate is extended over the medium term at the same level at which we anticipate this variable to be at the end of 2011 (see Figure 7.4). Such a scenario leads to a very gradual correction of the household debt ratio, with convergence to its long-term trend only in 2018. Under an alternative scenario 2, US households progressively increase their saving rate to levels approaching almost 10% by the end of 2019 (in line with the levels observed in the early 1980s). Even such a dramatic shift in saving patterns leads to a return of the debt-to-income ratio to trend only by early 2016. Several tentative conclusions could be drawn: 1) US household balance sheet adjustment is likely to require a long-term horizon to be completed (possibly up to ten years), suggesting the risk of a prolonged period of subdued consumption and a sluggish economic recovery in the US. 2) A scenario in which US households progressively increase their saving rate in order to converge faster to historical trends in their debt ratios (scenario 2 versus scenario 1) is likely to subtract mechanically

150 140 Scenario 1

130 120 110

Household debt

100

Scenario 2

90

Trend 1980–2000

80 70 60 Mar-80

Sep-87

Mar-95

Sep-02

Mar-10

Sep-17

Figure 7.3 Correlation: Household savings and wealth (as a percentage of personal disposable income; quarterly data) Notes: Last observation refers to 2009Q2. Projections up to Q4 2019. Sources: Federal Reserve Board, ECB staff.

9780230_243231_08_cha07.indd 194

5/11/2010 11:51:38 AM

Relationship US-Euro Area and Systemic Changes

195

14 12 10

US personal saving rate Scenario 2

8 6 4 Scenario 1

2 0 Mar-80

Mar-87

Mar-94

Mar-01

Mar-08

Mar-15

Figure 7.4 US personal saving rate (as a percentage of personal disposable income; quarterly data) Notes: Last observation refers to 2009Q2. Projections up to 2019Q4. Sources: BEA, ECB staff.

about 0.5 percentage points from annual PCE growth over the medium term. Accounting for the feedback effects between income and consumption, the negative impact is likely to be much larger (-1.2 pp per annum). 3) The retrenchment by the US consumer is likely to weigh on the recovery of global growth and trade due to a weaker US imports profile. 7.2.2 What will be the impacts on global growth and global trade? Looking further ahead, the recovery in world trade, while subject to a high degree of uncertainty, could be slow and protracted partly depending on how globalisation factors develop in the post-crisis world, and will also hinge heavily upon the extent of the recovery in world demand as well as its expenditure composition. It may be the case that global supply chains may help to amplify the trade recovery via the same mechanisms which seemed to exacerbate the trade contraction. Moreover, the recent fall in trade followed a period of exceptionally strong growth in trade flows, and the latter may not necessarily be repeated if some of the past trends in globalisation are reversed. Some commentators assess the global economic downturn as marking the change of the global economic model that resulted in strong global growth in the 2003–8 period, possibly even translating into a temporary break with past

9780230_243231_08_cha07.indd 195

5/11/2010 11:51:38 AM

196

Catching the Flu from the United States

globalisation trends. It remains, however, difficult at this juncture to assess how the factors that fostered trade growth in the recent past will be affected by the current global financial crisis.2 In particular, it is still too early to assess the impact of the financial crisis on the level of potential output for most economies, which will affect the level of future global demand. Equally importantly, the medium-term outlook for trade is likely to remain strongly linked to the future growth profile of the United States, as past recession episodes have shown. From an historical perspective, the link between US business cycles and global economic developments remains strong, and US recessions have often been associated with downturns in world trade. Although this relationship does not necessarily imply any causality, it provides a useful benchmark for gauging current developments. On average, US real imports contract a few quarters after the start of a recession and, in most cases, world trade follows a similar pattern with some lags (Figure 7.5). The recession of the early 1980s in particular shows significant similarities with the current cycle.3 The impacts of that recession on world trade were particularly strong, with trade remaining below its pre-recession levels for the next four years, although it should be noted that the depth of the current trade

30 25 Outset of US recession

20 15 10 5 0 −5 −10 −15 −20

−6

−4

−2

0

Average

2

4

6

8

Current recession

10

12

14

16

Early 1980s

Figure 7.5 Developments in world trade volumes during US recessions (index: Q0=100; quarterly data) Notes: Quarter ‘0’ refers to the start of the US recession (2007Q4). Average is based on US recessions since 1957. Source: IMF, CPB and ECB staff calculations.

9780230_243231_08_cha07.indd 196

5/11/2010 11:51:39 AM

Relationship US-Euro Area and Systemic Changes

197

downturn exceeds that of the early 1980s. Using the latter episode as a rough benchmark, it seems legitimate to expect some further weakness in world trade for a period of time. Against this background, one important element defining the profile for world trade in the medium term will be the likely decreasing role of the US as the ‘consumer of last resort’, due to the need to restore balance sheets and rebuild savings in the US. Given the HouthakkerMagee asymmetry – whereby the US income elasticity for imports is significantly greater than the foreign income elasticity for US exports, the expected contraction in US consumption may imply a particularly sharp impact on US imports (see Figure 7.6). Retrenchment by US consumers is therefore likely to weigh on the recovery of global growth and trade. Although some retrenchment is already underway, this is likely to be a long-term process which may possibly lead to a structural shift in US consumers’ behaviour and a sustained reduction in the ratio of US consumption to GDP. In order to gauge the timing and extent of the possible balance sheet consolidation, we consider the impacts of the two previous scenarios on world trade. Retrenchment by the US consumer might impact the recovery of world trade directly, through lower US import demand, and indirectly, with spillover effects to other countries, as the shortfall in US trading partners’

72

16 15

71

14 13

70

12 69

11 10

68

9 67 1995

8 1998

2001

2004

2007

Ratio of consumption to income (LH axis) Ratio of goods imports to income (RH axis) Figure 7.6 Imports and consumption (US nominal import and consumption shares in income, %) Note: Last observation is 2009. Source: AMECO database.

9780230_243231_08_cha07.indd 197

5/11/2010 11:51:39 AM

198

Catching the Flu from the United States

5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 2019 United States

Other advanced economies

Other emerging economies

Scenario 1

China Exponential trend

Scenario 2 with spillover effects Figure 7.7

Global trade with contributions by various countries (1994Q4 = 1)

Source: ECB staff calculations.

exports would result in lower GDP growth and further suppress world imports. Figure 7.7 shows world imports projections based on these two scenarios. If we assume only the direct impact of a reduction in US imports, the impact on the world trade profile will be relatively limited. If, however, we account for the spillover effects among countries, the impact on world trade is much larger, with world imports projected to be 15% lower than under scenario 1. In summary, a further retrenchment by US consumers as a result of a more pronounced deleveraging would significantly dampen the recovery in world trade and result in an even lower a more subdued profile for world trade than currently envisaged. As a final point, it is also important to mention the risk of protectionism as an additional matter of concern. So far, cases of rising protectionist measures have been sporadic, as most government have resisted the temptation to bow to political pressure and support domestic demand. The Great Depression showed that intensifying protectionist measures can delay recovery in world trade and aggravate the length and extent of the downturn. However, should the recovery falter, at the time fiscal and monetory stimulus is withdrawn, temptation to resort to protectionist measures – both through imposition of tariff or more subtle nontariff barriers – could be hard to resist. The results of such protectionist policies however could dampen further the outlook for world trade and should therefore be strongly resisted.

9780230_243231_08_cha07.indd 198

5/11/2010 11:51:39 AM

Relationship US-Euro Area and Systemic Changes

7.2.3

199

Will the euro area be affected by such structural changes?

This prospect of long-term subdued global trade notwithstanding, going forward the euro area could benefit from a ‘healthier’ set of fundamentals than the United States. At the same time, there are a number of factors that could prevent the euro area from a quick and strong rebound. a) Factors favouring a healthier euro area recovery First, one should recall that – at present – there is a larger need for adjustment in the US housing market relative to the euro area on the whole. In the US, financial innovation over past years has broadened US households’ access to credit markets and this, together with persistently low interest rates, has led to a real estate bubble (Figure 7.8). Credit problems in the subprime segment of the US mortgage market triggered a sharp adjustment in the real estate market. House price inflation turned negative at end-2006 and an oversupply of houses on the market led to a sharp contraction in residential investment (Figure 7.9).

20 15 10 5 0 −5 −10 −15 −20 1997

2000

2003

2006

2009

Euro area US (FHFA house price index) US (S&P/Case-Shiller national index) Figure 7.8 House prices (year-on-year change, %) Note: Last observation is 2008Q4 for the euro area, and 2009Q3 for S&P/Case Shiller index and FHFA house price index. Sources: ECB, Office of Federal Housing Enterprise Oversight and S&P.

9780230_243231_08_cha07.indd 199

5/11/2010 11:51:40 AM

200 Catching the Flu from the United States 20 10 0 −10 −20 −30 1996

1999

2002 Euro area

2005

2008

US

Figure 7.9 Residential investment (real residential investment, year-on-year change, %) Notes: Last observation is 2009Q3. Gross fixed capital formation: housing (euro area). Real private residential investment (US). Sources: Eurostat, US Bureau of Economic Analysis.

While the euro area housing market has also seen some house price overvaluation, and household balance sheets have also been stretched somewhat, the average extent of household leverage pales in comparison with that of the United States.4 One metric of the required correction involves the household saving rate, which declined by much less in the euro area over the last decade or so and, as a result, has also seen less of a marked increase since the onset of the crisis (Figure 7.10).5 Moreover, the adjustment lags that of the United States, with the housing market only having started to become a drag on activity (via residential investment and negative wealth effects) in 2008. Second, partly as a result of the housing market situation, US household balance sheets deteriorated significantly. US households built up large amounts of debt (Figure 7.11) and increases in housing and financial wealth fuelled consumption and led to a sharp fall in the personal saving rate (Figure 7.10). The extent of past debt accumulation and the contraction in wealth experienced by US households suggests that these will need to embark on a long process of rebuilding savings in order to repair their balance sheets. This adjustment is likely to be protracted over the next few years given that households tend to smooth consumption while rebuilding life-cycle savings and will lead to a muted recovery

9780230_243231_08_cha07.indd 200

5/11/2010 11:51:40 AM

Relationship US-Euro Area and Systemic Changes 30

14

25

12 10

20

8

15

6

10

4

5

2

0 1980

0 1985

1990

1995

Euro area (lhs)

Figure 7.10

201

2000

2005

US (rhs)

Household saving (household saving rates, percentage of Yd)

Note: Last observation is 2009Q3. Sources: Eurostat, US Bureau of Economic Analysis.

120 100 80 60 40 20 0 1981

1987

1993 Euro area

1999

2005

US

Figure 7.11 Household indebtedness (mortgage) (mortgage debt as a percentage of disposable income) Notes: Last observation is 2009Q3. Loans for house purchasing (euro area), Mortgage debt (US). Sources: ECB, US Federal Reserve Board.

in US personal consumption expenditures.6 The situation in the euro area remains healthier in this respect, both in terms of households’ mortgage-related indebtedness and saving rate. All in all, although the ongoing housing market adjustment in the euro area will weigh on overall economic activity as redundant resources are reabsorbed elsewhere in the economy, balance sheet factors and low

9780230_243231_08_cha07.indd 201

5/11/2010 11:51:41 AM

202 Catching the Flu from the United States

wealth effects could dampen the negative impacts on activity going forward compared with the US. In this respect, the fact that the euro area economies work under a single monetary policy would certainly have a net positive impact on the absorption possibilities of the area as a whole. While we have shown that some country-specific factors tend to solidify persisting country performance differences, they remain mostly confined to small countries and dare only marginally related to the introduction of the common currency. On the contrary, the euro ensures a solid shield from asset market volatility, over and above the absence of exchange rate fluctuation. b) Factors hampering a healthier euro area recovery ‘Traditional’ structural factors will also tend to slow down the euro area recovery following the US upturn. Most notably, rigidities in the euro area labour market have led to a lagging reaction within it, which may also hinder a fast recovery in euro area domestic demand. In general, countries with growth-friendly structural policy settings have tended to recover relatively quickly from a downturn. Rigid labour and product markets lengthen the time it takes for output to return to potential following a shock, and increase the cumulative output loss incurred over the cycle. By contrast, growth-friendly structural policy settings, while possibly amplifying the initial impact of negative shocks, generally support the rebound in economic growth, by allowing for a more rapid adjustment in prices and wages. This in turn enables a more rapid recovery in aggregate demand, reducing the cumulative losses in output and employment that might otherwise arise. Therefore, differences in the structural policy setting not only explain the different growth potential in Europe compared with the United States, but they also determine the dynamics of the recovery. Historically, owing to lower rigidities in the labour and product markets, the US economy recovers more quickly than those of European countries, although the downturns tend to be milder in Europe.7 This might therefore have significant implications going forward, as recovery in the euro area might take longer than in the US and the cumulative output loss might be larger.

7.3 The impact of the financial crisis on potential output8 In addition to the demand-side aspects discussed above, the financial crisis is expected to have had a substantial influence on the supply side, which can strongly influence the outlook for US as well as

9780230_243231_08_cha07.indd 202

5/11/2010 11:51:41 AM

Relationship US-Euro Area and Systemic Changes

203

world economic activity. The experiences of other countries that have undergone financial crises suggest that the level of potential output tends to be substantially and persistently lower relative to its pre-crisis trend over the medium term. While potential output growth tends to return to its pre-crisis rate eventually for most economies, this can take several years. A fall in potential output after financial crises can occur via a number of channels. Regarding labour input, the structural unemployment rate may increase as the crisis results in the downsizing of some sectors, such as the financial and construction sectors in the US, resulting in a need for a reallocation of labour across sectors. People who are outside the labour market for an extended period may also experience deterioration of skills, which could hamper their future employability. The impact on labour force participation is ambiguous, with negative employment prospects discouraging workers from participating in the labour force, while second-income earners may enter the labour force to compensate for a loss in income and wealth. Capital accumulation is slowed over time by weaker investment, driven by the higher cost of capital, credit restrictions and higher uncertainty regarding economic prospects. Moreover, financial crises can also lead to the scrapping of capital, thus quickly eroding part of the capital stock. Total factor productivity can also be influenced by financial crises, although the direction of the effect is ambiguous: on the one hand, the impairment of the financial sector may lead to a sub-optimal allocation of funds, in particular, if productive investment projects are not undertaken due to a lack of access to funding. Moreover, productivity may also suffer due to a reduction in R&D spending during the recession. On the other hand, the financial crises may remove inefficient firms from the market, thus raising aggregate productivity and creating incentives to restructure and improve efficiency. The historical evidence suggests that TFP declines significantly during the crisis and moves closer to the pre-crisis trend after around seven years, see Abiad et al. (2009). This section starts by assessing the impacts of the 2007–9 crisis on US potential output. Next, we will review to what extent the euro area might differ to the US case. 7.3.1 Supply-side considerations: lower US potential output Potential output in the US is likely to be substantially lower over the medium term, following the adjustment of the economy to the financial crisis. Figure 7.12 compares our own estimates with the ones of a

9780230_243231_08_cha07.indd 203

5/11/2010 11:51:41 AM

204

Catching the Flu from the United States

3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 2000–6 2007 IMF

2008 OECD

2009 CBO

2010

2011

ECB

Figure 7.12 Comparison of potential output estimates for the United States (annual percentage change) Note: OECD estimates are not available for 2011. Sources: IMF, OECD and ECB staff.

number of institutions. Potential output is estimated to grow by 1.0, 1.2 and 1.6% in 2009, 2010 and 2011, respectively, representing a considerable slowdown relative to the 2000–6 average of 2.5%. Beyond 2011, we would expect potential output gradually to return to pre-crisis growth rates. While these estimates are highly uncertain, they are broadly in line with those of other organisations (Figure 7.12). The impact of the financial crisis on the level of actual and potential GDP is depicted in Figure 7.13, where losses in actual output relative to its pre-crisis trend are estimated to amount to 12% and losses in potential output relative to its pre-crisis trend to 4.7% by 2011.9 7.3.2 Comparison of the US and the euro area and the role of structural factors The impact of the crisis on potential output might differ across countries owing to structural differences. First, the impact of the crisis on the labour input might depend on the degree of labour market frictions. Due to frictions in the labour market (e.g. wage rigidities and constraints to labour mobility) and the long adjustment lags, sharp increases in actual unemployment will bring about a temporary but persistent increase in structural unemployment, which would only return to its original level when industry reallocation of labour has been completed. As shown by Furceri and Mourougane (2009), countries where Employment Protection Legislation (EPL) is high are likely to

9780230_243231_08_cha07.indd 204

5/11/2010 11:51:41 AM

Relationship US-Euro Area and Systemic Changes

205

15,000 14,500 14,000 13,500 13,000 12,500 12,000 11,500 11,000 10,500 10,000 1999

2002 Potential output

Figure 7.13

2005

2008

Actual output

2011 Pre-crisis trend

Impact of the crisis on the level of US actual and potential GDP

Note: Actual data up to 2009Q3, projections up to 2011Q4. Sources: BEA and ECB staff.

experience a marked rise in structural unemployment in the short and medium terms while low-EPL countries will see very little change. In particular, the effect of a crisis on structural unemployment is found to be very large in high-EPL countries in situations of extremely severe downturns (around a five percentage point increase after five years). These findings appear to be mostly driven by the stringency of EPL for permanent contracts. The average replacement ratio and the index of product market regulation are also found to matter. In all cases, the impact of crises on structural unemployment appears to be significant only for countries with a more rigid economy than the average across OECD countries. The 2007–9 crisis has been diversified across countries as regards the downsizing of some sectors, such as the financial, construction and automotive sectors, after their disproportionate expansion during the boom. Besides the significant downsizing of some sectors, the crisis might have depressed investment over a protracted period due to the fact that the supply of credit becomes more limited leading to tighter

9780230_243231_08_cha07.indd 205

5/11/2010 11:51:41 AM

206 Catching the Flu from the United States

lending standards and higher effective costs of borrowing, as well as to lower current and expected profits. In this respect, the nature of the financing matters a lot. Box 7.1 compares the external financing of households and non-financial corporations in the euro area and the United States and presents a comparative analysis of the importance of bank loans for the euro area and the United States. Structural differences might also lead to different adjustment processes. For instance, a higher trend in potential growth rates may materialise in the longer term if the readjustment after the crisis leads to the elimination of low productivity sectors and the strengthening of high productivity sectors in the economy. A key condition for this is the existence of flexible markets.

Box 7.1 Comparison of the external financing of households and nonfinancial corporations in the euro area and the United States and impacts of the 2007–9 crisis The external financing structure of households and non-financial corporations is connected with the structure of the financial system, namely the importance of bank financing versus market-based financing. Overall, capital markets related to the private sector are somewhat larger in the United States than those in the euro area, amounting to 375% of GDP in 2007, compared with 311% of GDP in the euro area. The euro area has traditionally had a largely bank-based financial system, with loans to the private sector that remain on banks’ balance sheets amounting to 145% of GDP in 2007. By contrast, in the United States’ bank lending (with loans being originated and held by banks) to the private sector is much smaller and amounted to 63% of GDP in 2007 (see ECB, 2009). In line with the largely bank-based financial system in the euro area, the external financing of euro area non-financial corporations consists to a large extent of loans originated and held by banks. In the United States, loans from non-bank providers and market-based external financing are considerably more important than in the euro area. The importance of non-bank loans reflects the larger role of securitisation and syndicated loan activity compared with the euro area. At the same time, however, the largely bank-based system in the euro area has also changed, with a rapid growth of securitisation markets and a surge in syndicated lending activity. This has led to a stronger market orientation on the part of the euro area banking sector. Prior to the financial crisis, the worldwide move from a traditional ‘originate to hold’ model to an ‘originate to distribute’ model was accompanied by an insufficient pricing of risk and increased reliance on complex credit instruments, which proved fragile under stress. During the financial crisis, in contrast to the US situation, the central role of banks in financing and the smaller scale of securitisation activities in

9780230_243231_08_cha07.indd 206

5/11/2010 11:51:42 AM

Relationship US-Euro Area and Systemic Changes

207

the euro area have limited the capacity for re-intermediation and possible substitution into bank lending of non-bank and market-based funding. This was particularly problematic for non-financial corporations in the euro area, which are relatively more vulnerable to any excessive tightening of bank lending standards and more sensitive to changes in financial conditions, given that they have relied on net borrowing to finance investment to a greater extent than their US counterparts. Beyond the typical credit cycle, the financial crisis has also brought attention to the extraordinary credit supply constraints linked to banks’ weaknesses. The financial crisis led to considerable losses, impairing bank’s profitability and eroding their capital bases. Banks have also been coming under significant pressure from bank regulators and investors to rebuild their capital. Under these circumstances, banks have become increasingly reluctant to lend. Such impairments in bank lending practices are likely to imply additional quantitative restrictions in the allocation of credit in the economy. The decline in loans to NFCs in 2008–9 can largely be explained by mostly demand-related fundamentals in the euro area. While this is also true for the US in the period 2008–9Q1, such factors are not sufficient to explain the contraction in 2009Q2–Q3. Empirical evidence, however, suggests that ‘pure’ supply-side factors have also played a role in explaining the contraction in activity or loans to NFCs in the euro area and US. Using a VAR model including GDP, inflation, measures of loan supply and demand as well as the monetary policy rate, Ciccarelli et al. (2009) find that the tightening of credit supply to businesses (particularly due to bank balance sheet problems) does help to explain changes in GDP growth during the crisis period (2007Q3 to 2009Q2) in the euro area and, to a lesser extent, in the US.10 In addition, according to Hempell and Kok Sørensen (2009), after controlling for demand-side factors, loan growth to euro area NFCs was negatively affected by supply-side constraints (including the banks’ cost of capital, disruptions to their access to wholesale funding and their liquidity positions). Looking ahead, while the slowdown in credit growth, which occurred in the aftermath of the crisis, is not unusual given the severe decline in activity, a weaker turnaround of the credit cycle than in previous recessions can be seen as a risk for the strength of the recovery. In particular, the recent sharp increases in loan loss provisions by large euro area banks bode ill for the credit quality of loan books and suggest that euro area banks can expect intensifying financial distress from the corporate sector. Similar risks hold for the US, in particular stemming from potential losses resulting from the commercial real estate sector, where write-downs have risen particularly strongly. Thus, while economic indicators have improved, there may be some concern that the banking system may not be strong enough to finance a sustained recovery in spending. Moreover, the strong cumulated tightening of credit standards may have had limited effects on loan dynamics in an environment of weak demand, but supply constraints may become binding as economic conditions normalise. This could be a major issue in the euro

9780230_243231_08_cha07.indd 207

5/11/2010 11:51:42 AM

208 Catching the Flu from the United States area, where banks dominate the provision of finance for NFCs. Clearly, the upsurge in debt securities issuance, notably by the euro area corporate sector, suggests that companies have started to seek alternative ways to secure financing. In the US, the general use of more diversified sources of external financing could be of help in this context.

7.4 Implications of the financial crisis for global imbalances While the problems in the US subprime markets can be seen as the starting point of the financial crisis, and the collapse of Lehman Brothers on 15 September 2008 as the main trigger of its recent intensification, the origins of the crisis can be traced back to the build-up of global imbalances during the 1990s and the first years of this century. One may debate whether the cause has been the excess consumption of the US, or the savings glut in emerging Asia with the US acting as a kind of ‘consumer of last resort’. Although this may now be a largely historical debate, there is a substantial consensus on the main trends that have led to the current imbalances and will have to be faced in the future. First, there has been the globalisation of labour. Emerging Asia has boosted the global labour supply enormously through its integration into the world economy during the 1990s. This higher labour supply reduced production costs at the aggregate level and resulted in the improved comparative advantage of Asian economies. Also, emerging Asia, and China in particular, decided to keep exchange rates fixed visà-vis the US dollar and to accumulate trade surpluses instead of fostering domestic demand. This happened at the same time as the United States accumulated huge current account deficits. The mutually beneficial and dependent relationship between the US economy and the emerging economies has been referred to as Bretton Woods II system (following Dooley et al., 2003). This arrangement has allowed China and other emerging economies to pursue an export-led development strategy and run persistent current account surpluses. These economies have accumulated exchange rate reserves with policies that more or less explicitly targeted the US dollar. The official flows have prevented any disorderly depreciation of the US dollar and allowed US interest rates to remain at a lower level than otherwise would have been the case. This accumulation of US liabilities by emerging economies helps to explain how the US, and by extension the rest of the developed world, could maintain cheap credit, high growth and low

9780230_243231_08_cha07.indd 208

5/11/2010 11:51:42 AM

Relationship US-Euro Area and Systemic Changes

209

inflation. China alone has supplied the United States with almost USD 2,000 billion in low-cost debt. It is important, however, to reflect on whether this focus on the symmetry between Chinese trade surpluses and US trade deficits may have not somewhat distracted our attention from the possibly most important source of the global imbalances, namely excess consumption in the United States. It is indeed somewhat uncomfortable to associate the emergence of economies that have allowed a large number of people to come out of poverty with the lack of savings of a wealthy society. Although the sustainability of these global imbalances has been subject to debate in academic and policy fora, until the advent of the current financial turmoil the main concern remained focused on why these ‘anomalies’ persisted and how they could be resolved. In this context, some reflection should be given to the role of asset prices in the emergence of excess consumption and, in particular, the inability to detect bubbles in the recent past. This inability to identify asset-price bubbles has created a wrong equilibrium worldwide with a level of global consumption higher than allowed by productivity developments. This explains why the correction affects all countries in the world and has led to a real adjustment, which might be longer and deeper than we could have expected in more normal circumstances. It has to be noted, too, that the role given to the exchange rate as a signal of imbalances and as a tool to solve them might have been somewhat excessive. Indeed, prior to the current crisis, the US dollar had depreciated and the US current account deficit was declining, thus sending wrong signals regarding the strength and resilience of the global environment. Global imbalances have been associated with an increase in excess liquidity at the global level, fuelled by very low interest rates. Against this background, financial institutions in many advanced economies have become much more highly leveraged. This phenomenon was amplified by the resulting underpricing of financial risk and unsustainable growth in prices across a range of financial assets. Low global interest rates and subdued financial market volatility reinforced investor confidence and facilitated the development of financial products to reward risk-seeking behaviour in new ways. Further, the period of high and stable growth was also characterised by an increase in commodity prices. Commodity prices have increased dramatically over the past few years on the back of robust demand from emerging Asia to fuel its rapid industrialisation. Higher commodity prices reinforced the mechanisms of global imbalances as they led to

9780230_243231_08_cha07.indd 209

5/11/2010 11:51:42 AM

210

Catching the Flu from the United States

a further deterioration in the trade balances of commodity importers (the United States in particular). Also, the surpluses among commodity exporters were partly recycled in the accumulation of reserve assets. It is undeniable that these tendencies provided the basis for an exceptionally long period of high growth. Excess global liquidity and a low interest rate environment contributed to macroeconomic conditions characterised by remarkably high and stable growth, low inflation and reduced financial market volatility. However, as imbalances between different regions became more and more pronounced, the risks in the external environment eventually materialised, thus contributing to the 2007–9 turmoil. A return to sustainable world growth rates requires a paradigm shift in economic policies towards medium-term and stability-oriented macroeconomic policies. This will help to limit booms and busts in the future. A rebalancing of demand across regions is crucial in this respect. While the financial crisis has triggered some rebalancing, developments in the immediate aftermath of the crisis have been rather cyclical in nature. For instance, lower oil prices relative to pre-crisis levels have led to a contraction in the surpluses of oil-exporting countries and improved the current account balances of oil-importing countries. Contractions in domestic demand in the United States and other industrialised economies severely affected by the crisis have reduced their current account deficits, while the surpluses of some Asian economies have shrunk as a consequence of large stimulus packages that have increased domestic demand. But structural changes in policies are required to achieve a permanent reduction in imbalances. Higher national savings are needed in the United States and other deficit countries, achieved by a strong commitment to fiscal consolidation, in conjunction with increased household savings. A rebalancing also requires stronger domestic demand elsewhere. One way to achieve this is via reforms to boost domestic demand in emerging Asian economies, most notably in China, by developing the financial systems, as well as by raising spending on social insurance and health care with the aim of reducing households’ motive for precautionary savings. To quantify the extent of global rebalancing needed, a scenario of a rise in the US saving rate of 4 pp, matching scenario 2 discussed in Section 7.2, is simulated using NiGEM. In parallel, we also calibrate by how much, for example, Chinese domestic demand would need to rise to fill the gap in world trade resulting from higher US savings. The simulations refer to a medium-term horizon (ten years) and are based

9780230_243231_08_cha07.indd 210

5/11/2010 11:51:42 AM

Relationship US-Euro Area and Systemic Changes

211

on the assumption of exogenous oil prices, short-term interest rates and exchange rates. In line with scenario 2 discussed in Section 7.2, we assume that US households progressively increase their saving rate to almost 10% over the next ten years, reflecting a deleveraging process that would bring the debt-to-income ratio back to its trend. Simulations indicate that such a process would reduce US GDP relative to the baseline by more than 7% over the next ten years, or around 0.7 pp per annum, see Figure 7.14a. The impact on US consumption would be even larger (–1.2 pp per annum), reflecting the feedback effects of higher savings on income that, in turn, affect consumption. The decrease in US demand resulting from increased savings would lead to a fall in US imports of 23% relative to baseline, implying weaker import growth by 2.1 pp on average per annum over the next ten years. Figure 7.14b shows the implied effect on the world and the euro area, indicating a reduction in world GDP of 3.6% relative to baseline after ten years, in world imports of 7% and in euro area exports of 6.8% over the same period. This is equivalent to a reduction in world GDP growth of 0.4 pp on average per annum, and in both world import growth and euro area export growth of 0.7 pp on average per annum. Such a process of US household balance sheet repair would thus have significant economic implications, not only for the US, but also for global activity and trade and therefore the euro area, unless other regions in the world take up an increasing role in global demand. One region that has the potential to strengthen its domestic demand, which could eventually turn into a new engine of global growth, is emerging Asia, and most notably China. To explore such a possibility, we use NiGEM to quantify by how much Chinese domestic demand would need to rise in order to make up for the shortfall in world trade that would arise due to higher US private savings by the end of the ten-year horizon. Figure 7.15 presents the results: the level of domestic demand in China (used here as a proxy for emerging Asia) would need to increase by a total of 24% relative to baseline, implying higher growth in domestic demand of 2.4 pp per year. This would imply a rise in GDP growth in China of 0.9 pp per annum on average. An important caveat to this result is, of course, that such an increase in demand would represent a structural break in the historical relationships that describe the period over which the model has been estimated. While it would be challenging for China (or the rest of emerging Asia) to achieve such a strong and self-sustained rise in domestic demand in the near term, it is not implausible for such a scenario to occur over

9780230_243231_08_cha07.indd 211

5/11/2010 11:51:42 AM

212

Catching the Flu from the United States (a)

Quarters 1

5

9

13

17

21

25

29

33

37

Deviation from baseline %

10 5 0 −5 −10 −15 −20 −25 US savings rate (pp.)

US GDP (%)

US consumption (%)

US imports (%)

(b)

Quarters 1

5

9

13

17

21

25

29

33

37

Deviation from baseline %

0 −1 −2 −3 −4 −5 −6 −7 −8 World GDP (%)

EA exports (%)

World imports (%)

Figure 7.14 Simulated impact of a 4 pp gradual rise in the US saving rate. (a) Impact on US activity and trade (deviation from baseline). (b) Impact on activity and trade in the world and the euro area (deviation from baseline) Source: ECB staff.

9780230_243231_08_cha07.indd 212

5/11/2010 11:51:42 AM

Relationship US-Euro Area and Systemic Changes

213

Deviation from baseline

30 25 20 15 10 5 0 1

5

9

13

17

21

25

29

33

37

Quarters China domestic demand (%) China GDP (%)

China imports (%) World imports (%)

Figure 7.15 Simulated impact of a 24% gradual rise in China’s domestic demand over 10 years (deviation from baseline) Source: ECB staff calculations.

longer horizons, which allow for the possibility of structural reforms being implemented. Such domestic demand adjustment, however, would translate only gradually into increased imports.11 This reflects the currently low import content of domestic consumption in emerging Asia (notably China), which would take time to move closer to the levels of advanced economies. Moreover, the composition of consumer goods imports in China is very different from the goods imported by US households. According to the IMF, when measured by an import similarity index for more than 300 types of consumer goods, China’s imported consumer goods basket overlaps that of advanced economies by only about 35%. To sum up, should the offsetting of the shortfall in US demand and its significant global spillovers be achieved through rebalancing of global demand towards emerging Asia, this would take time and require substantial structural adjustments and more flexible exchange rates in these economies. Greater exchange rate flexibility in some countries that conduct managed floating foreign exchange policies is therefore a further necessary step towards achieving more balanced global growth. Under more flexible exchange rates, the price of currencies would be more strongly determined by market mechanisms and would thus minimise economic distortions.

9780230_243231_08_cha07.indd 213

5/11/2010 11:51:42 AM

214

Catching the Flu from the United States

Meanwhile, although a reduction in net capital flows across countries may occur as a result of smaller imbalances, it is necessary to ensure that gross cross-border capital flows continue to take place as they allow a transfer of knowledge and technology, thus raising productivity on a global scale.

7.5

Conclusion

Taking stock of US demand- and supply-side prospects and their global implications suggests that going forward, it is unlikely that global activity and trade could soon attain the strength recorded in the past decade. Such a conclusion is also supported by the fact that current economic recovery is to a large extent supported by government spending, which tends to have low import content and will need to be unwound over the medium term. As advanced and developing economies adjust for precrisis excesses, it is plausible to expect that global activity and trade will re-equilibrate at a lower but potentially more sustainable level, unless a rather rapid rebalancing of world demand can be achieved. The prospects of this are, however, rather slim given the need for sharp changes in policies and implementation of structural changes.

Notes 1. We are very grateful to Ursel Baumann, Ferdinand Fichtner and Georgi Krustev for having provided the material used in this section. 2. For Pascal Lamy, Head of the World Trade Organization, it is hard to say if the current global economic crisis will spell the end of ‘globalization’ (on 24 June 2009). For Robert Zoellick, President of the World Bank, ‘trade was one of the first sectors decimated by contracting consumption in high-income countries. But it will also be one of the first sectors to recover once the global economy picks up steam’ (on 6 July 2009). For Dominique Strauss-Kahn, IMF Managing Director, ‘trade can start up quite quickly, as quickly as growth, once the recovery starts in developed countries’ (6 July 2009). 3. Both are characterised by significant adverse impacts on consumption, residential investment and imports. In addition, the recession in the early 1980s was associated with difficulties in the banking sector, as in the present cycle. They do, however, differ as regards the role of monetary policy, which was at that time contractionary, and further impaired the recovery in economic activity. 4. This holds for the euro area as a whole, although there are great disparities across euro area economies, with some countries possibly requiring a correction of at least a similar magnitude to the United States. 5. The IMF World Economic Outlook (2009) indicates that the cumulative effects of declines in housing and financial wealth on the US household

9780230_243231_08_cha07.indd 214

5/11/2010 11:51:43 AM

Relationship US-Euro Area and Systemic Changes

6. 7. 8. 9.

10.

11.

215

saving rate could be in the range of 3.75–11.5 percentage points – a rise which would have substantial implications for private consumption and, by extension, economic growth. See Glick and Lansing (2009). See for instance Duval et al. (2007). We are very grateful to Ursel Baumann, Ferdinand Fichtner and Georgi Krustev for having provided the material used in this section. For actual output, the pre-crisis trend is calculated as the linear trend of the actual output series during a seven-year pre-crisis period that ends three years before the onset of the crisis, defined as 2007Q4. This definition follows that by Abiad et al. (2009). For the US, their findings suggest that restrictions in credit supply in the form of mortgages have been more important in explaining movements in GDP growth than the supply of loans to NFCs. See International Monetary Fund (2009).

References Abiad, A., R. Balakrishnan, P. Koeva Brooks, D. Leigh and I. Tytell, 2009, ‘What’s the Damage? Medium-term Output Dynamics After Banking Crises’, IMF Working Paper WP/09/245. Ciccarelli, M., A. Maddaloni and J. Peydró, 2009, ‘Trusting the Bankers: A New Look at the Credit Channel and Hints for the Crisis’, ECB mimeo. Dooley, M.P., D. Folkerts-Landau and P. Garber, 2003, ‘An Essay on the Revived Bretton Woods System’, NBER Working Paper Series 9971, September. Duval, R., J. Elmeskov and L. Vogel, 2007, ‘Structural Policies and Economic Resilience to Shocks’, OECD Economics Department Working Papers No. 567. European Central Bank, 2009, ‘The External Financing of Households and NonFinancing Corporation: A Comparison of the Euro Area and the United States’, Monthly Bulletin, April, 69–84. Furceri, D. and A. Mourougane, 2009, ‘How do Institutions Affect Structural Unemployment in Times of Crisis?’ OECD Economics Department Working Paper No. 730. Glick, R. and K. Lansing, 2009, ‘US Household Deleveraging and Future Consumption Growth’, FRBSF Economic Letters No. 2009-16. Hempell, H. and C. Kok Sørensen, 2009, ‘The Impact of Supply Constraints on Bank Lending in the Euro Area – Crisis Induced Crunching?’ ECB Mimeo. International Monetary Fund, 2009, ‘Regional and Economic Outlook Asia and Pacific: Building a Sustained Recovery’, October.

9780230_243231_08_cha07.indd 215

5/11/2010 11:51:43 AM

8 Conclusion

The 2007–9 financial crisis has had severe impacts, bringing the global economy to its deepest and most widespread recession since World War 2. The euro area economy has not been spared: it has been in recession since the second quarter of 2008, losing in one year around 5% of its output. One important characteristic of that recession has been its broad-based nature – not only across countries, but also across demand components affecting consumption, investment and world trade to a massive extent. Initially the shock appeared to be US-specific, and to some extent limited to the housing market. However, developments there may have acted as a trigger for vulnerabilities which were common across countries and regions, including high levels of leverage and an under-pricing of risk. In this vein, the shock could actually be thought of as global in nature since its beginning – although not widely recognised as such at the outset – with somewhat more negative implications for euro area activity than if the shock had been truly US-specific. The book has examined extensively the transatlantic shock propagation mechanism; we have stressed in particular the historical regularity that the US business cycle tends to lead that of the euro area. As seen in Chapter 3, history suggests that the euro area economy has generally had milder downturns but slower rebounds than the US. The 2007–9 crisis, however, hit euro area and US activity severely, with the former actually more affected according to some indicators. Three main factors – of course not exclusively (oil prices being an obvious omitted factor) – may have been decisive in explaining the relatively high intensity of the downturn in the euro area. First, ‘traditional’ structural differences between the euro area and the US – for example, the euro area having a larger industrial sector 216

9780230_243231_09_cha08.indd 216

5/7/2010 1:17:48 PM

Conclusion 217

and stronger trade openness – may have played a critical role in the international transmission as the crisis hit global industrial production and trade in particular. This difference is partly counterbalanced by the euro area having a slightly smaller share of financial intermediation services in total value added – a sector that has also been strongly impaired in that recession. Second, as also suggested in Chapter 2, closer global trade linkages, as well as the increasing intertwining of financial markets, may have intensified the size of business cycle fluctuations and the persistence of shocks. Higher confidence linkages across economies could also have contributed to this. Third, macroeconomic policy responses to the downturn have also differed, partly prompted by the timing of the slowdown in both areas. The US discretionary fiscal expansion has been stronger than the euro area’s – a difference partially offset, though, by weaker automatic stabilisers – and the monetary policy response has also differed between the two economic areas. To sum up, the 2007–9 financial crisis has taught us a number of important lessons. First, attitude toward risk not only vary over time, but also change in waves. Phases of excessive risk taking can be followed by sudden reversals driven by abrupt global confidence shocks such as the one experienced in mid-September 2008. Joined interest rate cuts, like those on 13 September 2001 and on 8 October 2008, are examples of coordinated actions in a context of extraordinary uncertainity about the economic outlook (Trichet, 2008). Joint actions are therefore essential when there is a need to respond to a single shock – a shock that can be rapidly transmitted around the globe through financial linkages and confidence effects. Second, – as examined in Chapter 5 – the growing financial interlinkages may have increased real cycles synchronization. Our main finding is that, although financial variables do not seem to systematically anticipate real developments, they can at times be relevant, especially when the shock hitting the economy is of financial nature. Hence, during periods of financial turmoil, restoring the proper functioning of financial markets appears to be key, and hence coordinated policies should be welcome. Looking forward, an important factor underlying the US adjustment – namely the need to embark on a long process of rebuilding life-cycle household wealth, may hint at a likely longer persistence of the needed correction as well as its greater amplitude in the US compared with the euro area. That would be contrary to historical experience, which suggests that the US has tended to rebound fairly strongly. On the other

9780230_243231_09_cha08.indd 217

5/7/2010 1:17:48 PM

218 Catching the Flu from the United States

hand, more rigid labour and product markets in the euro area might lengthen the time it takes for output to return to its potential and thus increase the cumulative output loss incurred over the cycle. The weak medium-term outlook for global trade and a delayed adjustment in the housing market could also be factors that – this time around – may hinder a rapid and strong euro area recovery. In a nutshell, the message of the book is that world economies have become increasingly integrated, which makes increasingly relevant the role of monitoring of external developments, as well as cooperation and coordination of policies. More specifically, we showed not only the importance of identifying shocks, their origin and nature in a timely fashion, but we also underlined the need to monitor closely vulnerabilities that could accelerate their propagation worldwide. In this context, given its leading role in driving global economic cycles, the United States has a special responsibility to ensure that no negative spillover is unduly transmitted to the rest of the world. At the same time, having shown that the US is confirmed as the engine of the world economy – despite the dramatic changes occurring in the last few years in the balance between major economies – requires also for the rest of the world a heightened need to agree on a sufficiently binding level of global surveillance.

References Trichet, J.-C., 2008, ‘International Interdependencies and Monetary Policy – A Policy Maker’s View’, Speech at the Fifth ECB Central Banking Conference, Frankfurt am Main, 14 November.

9780230_243231_09_cha08.indd 218

5/7/2010 1:17:48 PM

Index

Asian crisis, 6, 134 Bi-variate VAR (BiVAR) model, 47, 117, 118, 119, 120, 125 Bretton Woods, 8, 208, 215 business cycle synchronisation, 8, 9, 10, 18, 23, 24, 25, 26, 27, 28, 29, 30, 32, 33, 34, 35, 36, 37, 40, 41, 42, 50 business cycle co-movement, 9, 24, 25, 29, 31 international, 9, 42, 58, 59, 60, 143 channel confidence, 4 financial, 9, 19, 21, 22, 26, 31, 35, 40, 100, 140 trade, 4, 6, 9, 20, 40, 97 confidence, 1, 4, 14, 15, 105, 127, 130, 132, 135, 140, 147, 150, 152, 166, 171, 172, 174, 181, 209, 217, 218 confidence linkages, 14, 172, 217 cross-border capital flows, 9, 21, 214 Diebold-Mariano test, 118, 120, 124 Dynamic Stochastic General Equilibrium (DSGE) model, 11, 12, 61, 62, 63, 64, 65, 67, 68, 69, 81, 90, 91, 92, 93, 94, 96 multi-country DSGE model, 12, 64, 81

Generalised Impulse Response Functions, 54, 56, 74, 89 Giacomini and White test (GW), 117, 123, 125, 143 global imbalances, 37, 189, 190, 191, 208, 209 global supply chains, 19, 151, 195 global trade collapse, 147 Great Depression, 94, 189, 198 Great Moderation, 80, 182 GVAR (Global vector autoregressive), 5, 6, 15, 52, 53, 59, 62, 68, 81, 82, 83, 88, 94, 183 housing investment, 175, 176, 177, 178 market, 8, 14, 116, 166, 173, 179, 180, 191, 192, 199, 200, 201, 216, 218

echo effect, 4, 5 equity premium puzzle, 90, 96 ERM crisis, 8 Economic and Monetary Union (EMU), 66, 116, 180, 181, 182, 183 excess volatility puzzle, 91 Factor structural VAR model, 6, 7 Feldstein-Horioka, 20, 91

financial spillovers 12, 106, 109, 110, 111, 139 financing constraints, 92 Financial VAR model (FiVAR), 117, 119, 120, 122 Forecast error variance decomposition (FEVD), 74, 77, 86, 87, 88, 90 Foreign direct investment (FDI), 19, 21, 22, 23, 24, 27, 30, 33, 34, 38, 39, 40, 41, 42, 168

impulse response functions (IRFs), 53, 54, 56, 70, 73, 74, 86, 87, 88, 89, 93, 127, 128, 129, 130 integration economic integration, 9, 18, 24, 27, 37, 38, 42, 57, 182 financial integration, 3, 10, 16, 18, 25, 26, 27, 29, 30, 31, 32, 33, 35, 36, 37, 42, 56, 57, 98, 108, 182 vertical integration, 27, 169 international risk sharing, 25, 188

219

9780230_243231_10_ind.indd 219

5/7/2010 4:51:59 PM

220

Index

IS curve, 62, 63, 69, 70, 72, 73 Lehman Brothers, 149, 159, 162, 208 market imperfections, 92 Markov-switching model, 49, 50, 52 merger and acquisition (M&A), 21, 37, 182 Multi-country New-Keynesian (MCNK) model, 62, 68, 69, 70, 74, 81, 82, 83, 86 National Institute Global Econometric Model (NiGEM), 210, 211 New Area Wide Model (NAWM), 65, 66, 67 New Keynesian model, 12, 62, 68, 69, 79, 80, 94 Phillips curve, 63, 94 New Open Economy Macroeconomics, 11, 95 non-linear VAR, 126 openness, 4, 14, 20, 63, 97, 165, 168, 186, 217 Phillips curve, 11, 62, 63, 64, 69, 70, 72, 73, 74, 94, 106 portfolio investment, 22, 24, 33, 34, 35, 38, 40, 41, 168 potential output, 185, 190, 196, 202, 203, 204, 205 Probit, 13, 133, 134, 135, 136, 137, 138, 139, 140, 186 ProbVar model, 137, 138, 139 real estate cycle, 173 recession probability, 134, 136, 137, 139 risk-free rate, 91, 92, 96 second-round effect, 43, 56, 58, 169

9780230_243231_10_ind.indd 220

sectoral specialisation, 9, 10, 24, 27, 29, 35 shock common, 6, 7, 8, 9, 37, 44, 49, 68 country-specific, 6 demand, 4, 5, 11, 61, 66, 69, 71, 73, 74, 75, 77, 79, 80, 81, 86, 89, 90, 93, 106 exogenous, 11, 62 external, 3, 9, 181 financial, 12, 63, 92, 100, 105, 106, 108, 139, 140, 185 idiosyncratic, 2, 6, 7, 8, 9, 11, 56 monetary policy, 66, 68, 69, 71, 73, 77, 78, 80, 81, 87, 89, 90, 93, 94, 95, 96, 105, 106, 109 structural, 12, 62, 65, 66, 70, 71, 86, 87, 88, 90, 93, 106, 108 supply, 11, 61, 69, 73, 74, 76, 77, 79, 80, 81, 87, 89, 90 spillover, 2, 4, 6, 7, 8, 12, 16, 21, 61, 66, 67, 68, 97, 98, 100, 106, 109, 110, 111, 112, 139, 140, 142, 143, 144, 197, 198, 213, 218 structural VAR, 6, 7, 49, 105, 140 Taylor rule, 11, 69, 71, 73, 77 three-stage-least square (3SLS), 29, 30, 33, 34, 40 Threshold VAR (TVAR), 13, 127, 128, 129, 132, 133, 139 third-market effect, 57 transmission channels, 1, 14, 36, 92, 98, 112, 165, 168 transmission of shocks, 2, 4, 9, 10, 11, 21, 24, 56, 81, 92, 94, 97, 98, 100, 168 TriVAR model, 117, 118, 119, 120 variance decomposition, 7, 66, 70, 74, 77, 79, 86, 89, 90, 109, 140 vertical integration, 27, 169 vertical specialisation, 25, 41, 170, 187

5/7/2010 4:51:59 PM