Asymmetry and Aggregation in the EU

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Asymmetry and Aggregation in the EU

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Also by David G. Mayes TOWARDS A NEW FRAMEWORK FOR FINANCIAL REGULATION (Co-authored) DESIGNING CENTRAL BANKS (Co-authored) THE MICROFOUNDATIONS OF ECONOMIC SUCCESS: Lessons from Estonia PROBLEMS OF GOVERNANCE IN THE EUROPEAN UNION: Migration, Monetary Integration, Socio-Economic Change and Trade THE VARIETIES OF LINGUISTIC, RELIGIOUS AND GEOGRAPHICAL IDENTITIES IN EUROPE: Essays on the Problem of European Integration DEPOSIT INSURANCE (Co-authored) OPEN MARKET OPERATIONS (Co-authored) PROSPECTS FOR FINANCIAL MARKETS THE STRUCTURE OF FINANCIAL REGULATION (Co-authored) ADJUSTING TO EMU (Co-authored) NEW ZEALAND AND EUROPE: Connections and Comparisons (Co-authored) WHO PAYS FOR BANK INSOLVENCY? (Co-authored) SOCIAL EXCLUSION IN EUROPEAN WELFARE STATES (Co-authored) SOCIAL EXCLUSION AND EUROPEAN POLICY (Co-authored) THE EVOLUTION OF THE SINGLE EUROPEAN MARKET (Co-authored) SOURCES OF PRODUCTIVITY GROWTH (Co-authored) THE SINGLE MARKET PROGRAMME AS A STIMULUS TO CHANGE: Comparisons between Britain and Germany (Co-authored) INEFFICIENCY IN INDUSTRY (Co-authored) THE EVOLUTION OF RULES FOR A SINGLE EUROPEAN MARKET: VOL 1 Industry and Finance VOL 2 Rules Democracy and the Environment VOL 3 Social and International Issues FOREIGN DIRECT INVESTMENT AND TRANSITION: The Case of the Visegrad Countries (Co-authored) THE EXTERNAL IMPLICATIONS OF EUROPEAN INTEGRATION (Co-authored) PUBLIC INTEREST AND MARKET PRESSURES: Problems for the 1992 Programme (Co-authored) ACHIEVING MONETARY UNION (Co-authored) A NEW STRATEGY FOR SOCIAL AND ECONOMIC COHESION AFTER 1992 (Co-authored) THE EUROPEAN CHALLENGE: Industry’s Response to the 1992 Programme (Co-authored) A STRATEGY FOR THE ECU (Co-authored) SHARPBENDERS: The Secrets of Unleashing Corporate Potential (Co-authored) INTEGRATION AND EUROPEAN INDUSTRY (Co-authored) THE EXCHANGE RATE ENVIRONMENT (Co-authored) MODERN PORTFOLIO THEORY AND FINANCIAL INSTITUTIONS (Co-authored) APPLICATIONS OF ECONOMETRICS THE PROPERTY BOOM INTRODUCTORY ECONOMIC STATISTICS (Co-authored)

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

IMPROVING BANKING SUPERVISION (Co-authored)

Asymmetry and Aggregation in the EU David G. Mayes and

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Matti Virén

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

© David G. Mayes and Matti Virén 2011 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 2011 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 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 978-0-230-53808-5

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 catalogue record for this book is available from the Library of Congress. 10 20

9 19

8 18

7 17

6 16

5 15

4 14

3 13

2 12

1 11

Printed and bound in Great Britain by CPI Antony Rowe, Chippenham and Eastbourne

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Palgrave Macmillan in the US is a division of St Martin’s Press LLC, 175 Fifth Avenue, New York, NY 10010.

Contents List of Figures

vi

List of Tables

viii

List of Abbreviations

x

Acknowledgements

xi

Preface

xii

The Nature of Asymmetry

1

2

Estimation and Aggregation Concerns

10

3

Aggregate Supply and Demand in an Open Economy

42

4

The Phillips Curve

77

5

Regional and Sectoral Concerns

115

6

Output, Unemployment and the Labour Market: The Okun Curve

134

7

Asymmetry and the Role of the Public Sector

158

8

Monetary Policy

174

9

Fiscal Responses

204

References

219

Index

232

v

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

1

List of Figures Illustration of a simple threshold model with cyclical data A comparison of a rigid and smooth threshold An effect of smoothing on a two threshold model Illustration of the minimum conditions Friedman’s plucking model Kernel densities of output growth with the EU27 data for 1990–2008 2.7 Implications of the convexity of the Phillips curve 2.8 An effect of heterogeneity of estimation results 2.9 Kalman filter estimates of threshold model coefficients 2.10 Distribution of coefficients 2.11 Comparison of cross-section variance of GDP growth rates and the squared growth rate of EU27 GDP 3.1 Estimated country-specific interest rate effects from an IS curve 3.2 Sectoral coefficients of the real interest rate 3.3 Coefficients of the real exchange rate 3.4 Median of key macro variables before and after the EMU 3.5 Similarities in (a) growth and (b) inflation performance in the euro area and the US 3.6 Scatter plot between change rates of house and stock prices 3.7 The growth rate of (real) house and stock prices (medians) 3.8 Median values of output growth and output gap 3.9 Comparison of GDP growth rates and inflation over 15 EU countries 4.1 Indicators of output (medians) 4.2 Comparison of inflation rates (medians) 4.3 Expected inflation 4.4 Coefficient of unemployment in a New Keynesian Phillips curve 4.5 Estimates of simple nonlinear Phillips curve for the EU for the pre-euro period 4.6 Linearity and nonlinearity in the Phillips curve and the setting of policy 4.7 Real time OECD output gap estimates 1994–2002 vi

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

19 21 22 25 27 28 30 32 34 37 41 52 53 53 54 56 58 59 62 69 86 87 90 91 92 96 102

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

2.1 2.2 2.3 2.4 2.5 2.6

List of Figures vii

5.6 5.7 6.1 6.2 6.3 6.4 6.5 6.6 7.1

7.2 8.1 8.2 8.3 8.4 8.5 8.6 8.7 8.8 9.1 9.2 9.3

Real time HP filtered output gaps Real time and revised HP filtered output gaps Comparison of unemployment dispersion measures Sectoral output growth dispersion indicators The extent of change Evolution of sectoral shares (median values) Comparison of income (per capita) and unemployment dispersion Impulse responses of the VAR model Finnish data on regional unemployment rates Convergence of unemployment in the EU Comparison of income (per capita) and unemployment dispersion Indicators of output Country-specific coefficients of a threshold model for the Okun curve Variance of the unemployment rate across countries Variance of GDP growth across countries Long-run effect of a 1 per cent increase in public consumption on government surplus/GDP with and without policy coordination Comparison of expansive fiscal policy effects in the euro area Interest rate forecast for the euro period An FCI for the euro area An FCI for Germany An FCI for the UK An FCI for the euro area with an increased weight on the real exchange rate The real interest rate, real exchange rate and asset prices in the euro area The impact of house prices on a country by country basis Confidence and house prices An EU average of the impulse responses of d/y* to growth shocks Median of fiscal variables before and after the EMU Change in the responsiveness to the debt ratio

107 108 121 125 125 126 128 130 131 146 147 148 149 149 150 170

171 193 194 195 195 196 197 198 202 210 212 215

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

4.8 4.9 5.1 5.2 5.3 5.4 5.5

List of Tables LR test results for a threshold model Mean values of ˆc1 and cˆ2 from the simulated data Summary of Granger and Lee findings Estimates of the ratio of the real interest rate to the real exchange rate effect from Mayes and Virén (2000) 3.2 OLS estimation results for the 1987:1–1997:4 period 3.3 IS curves (basic specification) 3.4 Estimation of an extended IS curve 3.5 Basic IS curve specification with different lags 3.6 Comparison of different IS curve specifications 3.7 Estimation of the IS curve with the output gap variable 3.8 Asymmetry in the IS curve 3.9 The effect of house and stock prices: An update 3.10 Euler equations for output using the Consensus Forecast data 3.11 Estimation of a ‘consumption function’ 4.1 Phillips curves 4.2 GMM estimates of a New Keynesian Phillips curve 4.3 When did the EMU show up? 4.4 Estimates of a Phillips curve with the OECD forecast data 4.5 Correlations and test for unbiasedness 4.6 Estimates of restricted Hybrid model with HP filtered output gap, 1977–2006 4.7 Estimates of restricted Hybrid model, 1994–2006 4.8 Estimates of the hybrid model, GMM, 1977–2006 5.1 Phillips curves from regional data 5.2 Backward-looking Phillips curve with regional data 5.3 Estimates using unemployment rather than the output gap 5.4 Estimates of a nonlinear Phillips curve with output dispersion variables 5.5 Sectoral shares of output over countries and time (1991–2008) 5.6 Estimates of a VAR model with the unemployment dispersion variable 5.7 Finnish regional data results viii

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

35 38 39 47 48 49 50 60 61 63 66 67 71 74 84 88 89 90 104 110 111 112 120 122 123 124 127 129 132

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

2.1 2.2 2.3 3.1

List of Tables ix

7.2 7.3 8.1 8.2 8.3 8.4 8.5 8.6 8.7 8.8 8.9 9.1 9.2

The Okun curve Individual country estimates of a nonlinear Okun curve Threshold model estimation results G/Y as the threshold variable Estimation results with panel data A summary of the public consumption simulation Estimates of the Taylor rule Autocorrelation of inflation Threshold model estimates for the Taylor rule Simultaneous system estimation from panel data Reaction function estimates Corridor reaction functions Impact of house and stock prices on interest rates Estimates of alternative interest rate equations The effect of house and stock prices: An update Selected country-specific estimates of equation (9.1) Evidence of changing fiscal behaviour

139 142 162 164 169 179 181 181 183 184 185 190 192 200 207 214 Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

6.1 6.2 7.1

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

ARIMA BDS BEPG CPI DSGE EEA ECB ECOFIN EDP EMU EMS EU ERM FCI FE FRB GDP GLS GMM HP IV LM LR LS MCI NiGEM NUTS OECD OLS REH SGP SUR STR VAR

Autoregressive Integrated Moving Average Brock Dechert Scheinkman Broad Economic Policy Guidelines Consumer Price Index Dynamic Stochastic General Equilibrium European Economic Area European Central Bank European Council of Finance Ministers Excessive Deficit Procedure Economic and Monetary Union European Monetary System European Union Exchange Rate Mechanism Financial Conditions Index Fixed Effects Federal Reserve Board Gross Domestic Product Generalised Least Squares Generalised Method of Moments Hodrick–Prescott Instrument Variable Lagrange Multiplier Likelihood Ratio Least Squares Monetary Conditions Index National Institute Global Econometric Model Nomenclature of Territorial Units of Statistics Organisation for Economic Cooperation and Development Ordinary Least Squares Rational Expectations Hypothesis Stability and Growth Pact Seemingly Unrelated Regression Smooth Transition Regression Vector Autoregression

x

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

List of Abbreviations

Many people have contributed to our understanding of the problems discussed in this book over the last 25 years, too many to name them all. We should however mention Maritta Paloviita who has participated in joint research with both of us, as we use some of the results from that work. Pentti Saikkonen from the University of Helsinki provided useful advice on modelling issues, Jan Fidrmuc, Prasanna Gai, Alfred Guender, Mark Holmes, Dimitri Margaritis, Alberto Montagnoli and Brian Silverstone all provided very helpful advice on an earlier draft of this book at a workshop at the University of Auckland and we thank the Andrew Shonfield Association for providing the finance for the occasion. All of the joint research in the book was undertaken while we were both at the Bank of Finland, Mayes as Advisor to the Board and Virén as Research Consultant. Juha Tarkka as Head of the Research Department in the early years did much to encourage the work. However, many of our colleagues participated in seminars and conferences in which the work was presented along the way. We have pillaged our previous publications in putting the book together, but most of the results have been reworked on a common dataset that brings the analysis up to the end of 2008. Our thanks go to a string of helpful research assistants who have put the data together. In particular, we thank Heli Tikkunen and Tarja Yrjöla from the Bank of Finland. We are grateful to Janet Mayes to making the corrections to the final typescript, which got rather battered with two authors on opposite sides of the world. Through much of the period when we were working the downside of experience was not prominent and interest was polite rather than enthusiastic. With the global financial crisis the relevance of our work has become only too obvious and we regret that we were not more influential earlier.

xi

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Acknowledgements

This book has been a long time in the making. We began writing separately on asymmetry over 20 years ago and jointly in the late 1990s. We had also been approaching the topic from different directions. Our joint interest arose from concern over Finland’s transition to the euro area and how euro area monetary policy might be put together for the area as a whole. Since there was the potential for considerable variation among the member states over where they lie in the economic cycle and over the value of the parameters of key economic relationships it appeared to us that there was likely to be a great deal of difference between aggregating the data and estimating euro wide relationships and aggregating the results from estimating the impact on individual countries. Little prior work had been done and the preference at the time was to estimate a new area-wide model. We thought the consequences should at least be explored. As the work developed we realised that the extent of asymmetry in the euro economies was quite complex, relating not just to the behaviour of aggregate output but to the labour market, inflation and above all to policy itself both with regard to the fiscal stance and to monetary policy. This indeed became clear. However, our work coincided with the period of the ‘great moderation’ so that the consequences became less obvious. With the global financial crises the extent of the asymmetry and indeed the differences among the euro area countries has become very obvious, to the extent that it has placed strains on the area and required the building up of a major fund to help the worst affected countries through their difficulties. Our analysis suggests that by not aggregating the asymmetric relationships monetary policy may tend to underestimate the lack of downward pressure from the countries in relative difficulty. It is noticeable that over much of the first decade of the euro area, that inflation turned out to be a little higher than wanted. Perhaps asymmetry made a contribution to this. Just the same asymmetries exist outside the euro area but there the exchange rate can vary to cope with asymmetric shocks and asymmetric responses. This is not to suggest that membership of the euro area was a mistake for some of the countries that are most different from the general movement, but that they need an adequate cushion to cope with the harsh adjustment. Among the euro countries, Finland xii

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Preface

made among the greatest preparations, with firm fiscal consolidation over the course of a decade and with the setting up of buffer funds for both unemployment and pensions. Now the shock has come and Finland has been one of the most affected countries as a result of the importance of investment goods in its export markets. Nevertheless, the country appears to be getting through such a sharp recession without a threat to long-run sustainability and without an unreasonable burden on future generations. In retrospect the provision of the Stability and Growth Pact both to encourage fiscal consolidation and to try to guard against the attempt to be overconfident in expansions so that contractions build up too great budgetary deficits has proved to be very wise. It is a pity that it was not followed more enthusiastically in some cases. Work on asymmetry will be given a considerable boost by the crisis, not simply because its importance has become clear but because we now have many more observations on how people behave when there is a serious adverse shock and financial uncertainty. The extent of the difference is clear, with the temporary closure of some financial markets. While most of our work was initially published before the crisis our compiling it now may seem a bit after the event but we hope that the results will be of assistance in enabling policy to be sufficiently asymmetric in future that the difficulties are better offset, particularly in good times when the problems are built up. DAVID G MAYES MATTI VIRÉN

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Preface xiii

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

This page intentionally left blank

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

1

The global financial crisis has been a vivid reminder of how asymmetric economic behaviour is. Recessions in an economy do not have the same pattern as expansions. Expansions are less sharp and last longer. In part they reflect preferences. Macroeconomic policy seeks to encourage and prolong expansions but seeks to make recessions as shallow and short as possible. We thus see asymmetry in both economic behaviour and in economic policy. While the asymmetry in the macroeconomy is very obvious the same asymmetry can be found in microeconomic and sectoral behaviour as well. A well-known example comes from consumption. When incomes rise, all but the poor spend much of the increase but save the rest. The proportions vary according to whether they expect this increase to be one off or enduring. However, if incomes fall by the same amount people resist seeing their consumption fall, particularly if the shock is expected to be temporary. In the longer term consumption will fall as the ability to dissave or borrow is inhibited but the pattern of behaviour is clearly asymmetric. More trivially there are many actions that are not reversible because of experience. However much of economic analysis largely ignores the existence of this asymmetry and hence most models are symmetric in character. Such asymmetries are of limited importance when discussing marginal changes but are much more important when the shifts are large, as recently, or where they very obviously represent a regime change. The failure to anticipate the consequences of the events that led to the present crisis have generated a burst of enthusiasm in investigating this phenomenon. We have been working in this area for many years and have taken this opportunity to put together a range of these ideas in a single volume. 1

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

The Nature of Asymmetry

In this book we look at various aspects of asymmetry in the European economy and explore the consequences for policy. Our reason for picking Europe is two-fold. First, trivially, because this is the region on which we have done most work but, second, because it offers an extra feature that makes the policy consequences more interesting. In any event macroeconomic policy applied to a country as a whole averages out over the needs of the households, firms and regions within it. But in the European Union (EU) and in the euro area in particular, a single monetary policy is run for a large and relatively disparate set of countries. Averaging across countries would not matter very much if the disparities were relatively small. But if they are large and the asymmetries are important then this can have important consequences for policy. Take a simple Phillips curve for example. When there are strong demand pressures, say with a clearly positive output gap, changes in demand will have a considerable effect on inflation. The same changes in demand, when the output gap is the same size but negative, may have almost no impact on inflation. Simple arithmetic averaging could, therefore, give a misleading implication for the single policy. The positive and the negative do not simply cancel out. In the EU cross-border fiscal transfers to help offset the unequal impact of shocks and other policies are trivial. In individual countries they are extensive, hence asymmetry has greater consequences for the EU than elsewhere, particularly in the euro area where changes in the nominal exchange rate among the members is not possible. In the chapters that follow we begin by setting out what we understand by asymmetry and then how it might be measured before exploring it in detail in a number of areas relating to inflation, unemployment, growth, fiscal and monetary policy. We build up the analysis into a small model of the economy that we can use for analysing the policy problems. We do not pretend to offer a complete treatment and one area that is ripe for further examination is the behaviour of firms, particularly entry and exit. New firms face a steep learning curve and their productivity improves rapidly as a result. However, that improvement needs to be rapid and many new firms fail quite early in their lives as they are unable to reach the profitable cost structures of their more established competitors before the financing runs out. Declining firms on the other hand tend to decline rather slowly and can continue in business for a considerable length of time. Unlike new firms which have had to make all their physical and productive investment, whether in their staff or processes, up front and have to bear the costs of this before the sales income starts, mature firms have long written off this initial investment and can let their equipment

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

2 Asymmetry and Aggregation in the EU

The Nature of Asymmetry 3

run down and not retrain their staff. A substantial proportion of their costs are sunk – their experience, their customer and supplier networks etc., with little value in an alternative use. While such actions may merely be postponing the inevitable they can nevertheless be drawn out and there is always the hope of new invention or major new customer to turn things round. Even insolvency is usually put off as creditors hope they can avoid losses. Some of these issues are explored in Mayes (1986) but they require a very different dataset from the macroeconomic information that we use in the present study, so we have to leave them for the future.

Asymmetry and its causes

In our discussion thus far we have mainly been concerned to point out the nature of the asymmetry that we observe. Interest tends to focus on the economy as a whole so let us begin there. Thus Keynes (1936) was initially concerned simply to point out the depth, steepness and duration of the up and down phases of economic cycles were clearly different. The approach of Friedman (1964) goes a little further by describing a ‘plucking’ model. Economies tend to run at full capacity or potential output, where these two are defined as some sort of level of maximal output or output growth beyond which signs of overheating emerge, whether in terms of a sharp rise in costs, breakdown of machinery, or exhaustion of staff, or in terms of inflation. These are often expressed as ‘natural rates’. Operating below this is not only a waste of resources but an unplanned event, generated by adverse shocks. Attempting to run faster than potential is hence bound to be followed by some element of slow down, whereas firms can quickly return to potential once demand picks up, provided the downturn has been short-lived and staff have not been laid off or machinery run down. Hence in this view one can return swiftly to normal like a plucked string. However there will be some overshooting with the initially rapid return to normality tending to go too far. Although the vibration of the strings may not be a good analogy for what tends to happen thereafter. However, these represent observations rather than explanations. As Caballero and Hammour (1994) remark, one of the aspects of asymmetry over the cycle can be seen in the operation of the labour market. In a downturn, job destruction in the sense of people losing their jobs, whether or not the company itself goes out of business, tends to be much more rapid than the process of job creation once the upturn

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

1.1

begins. Indeed when the former level of output is regained it is likely to be at a noticeably lower level of employment. It is only possible to regain the same levels of employment after a much longer lag. Here explanations can be quite straightforward. It is the less productive staff and less productive firms where the job losses are concentrated. Furthermore, making people redundant is an unpleasant and indeed costly business that is often contested. Hence initially it is avoided and contraction achieved through ‘natural wastage’ and not hiring new staff. Once it is embarked upon it is done to fullest plausible extent as a second round would simply add to the costs and the loss of morale. Some of the redundancies can always be rescinded. Once the economy starts to recover only some of the jobs will be in the firms that have contracted others will be in new firms and in the production of new goods and services. New firms tend to be more productive than existing firms just as exiting firms tend to be less productive (Mayes, 1996). Given their extra costs, new firms have to have a competitive edge. In any case the recovery is likely to require new investment and most new equipment is more productive than that which it replaces. Hence fewer people, although probably more highly skilled, are required. Such investment may well not apply simply to the new production but to the whole of the firm’s output, thus emphasising the contrast. In what follows we look at asymmetry in three key areas of the economy, aggregate supply and demand (Chapter 3), the relation between the extent of pressure in the aggregate economy and in employment and inflation – the Phillips curve – (Chapters 4 and 5) and the relation between output and unemployment – the Okun curve (Chapter 6). In the case of both the Phillips curve and the Okun curve we look at behaviour at the sectoral and regional level. One particular feature worthy of note is that the greater the variance or range of unemployment the greater the impact on inflation. Between them these three areas cover the aggregate asymmetry we have just described. However, this is only part of the story, as the response of policy to this underlying asymmetry in the economy is itself asymmetric. Depending on its nature this asymmetric response can reduce or exacerbate the asymmetry in behaviour. Take the case of monetary policy. In writing their assessment of the Greenspan years, Blinder and Reis (2005) characterised the ‘Greenspan Standard’ as a cautious approach to excess demand pressure and rises in asset prices in particular, provided they did not provide a threat to overall inflation, but a vigorous response to serious downturns to avoid the threat of damaging debt-deflation spirals and the danger of approach-

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

4 Asymmetry and Aggregation in the EU

ing the zero bound for nominal interest rates after which different and probably less effective monetary policy tools have to be used. Up until the boom and bust of the present crisis this approach received considerable support because it is very difficult to decide what constitutes an asset price bubble or an unsustainable increase in the rate of economic growth. No central banker wants to be responsible for halting a sustainable improvement in economic performance so the likely response is sceptical caution, falling short of outright halting of the process. To continue the reference to Alan Greenspan – his famous remark about irrational exuberance in the US stock market was made in December 1996 well before the peak of the dot.com boom in 2000 (Greenspan, 1996). With the benefit of hindsight his caution proved justified but markets ignored the advice and the Federal Reserve did not follow it through with any firm action. We look at this from the point of view of monetary policy in Europe, where there was no such espousal of an asymmetric approach and very little focus on the behaviour of the stock market, yet an exploration of the revealed asymmetry in the setting of monetary policy reveals just the same pattern as in the United States, albeit not so strongly (Chapter 8). Monetary policy does appear to respond a little to both stock and house prices and in an asymmetric manner. However, there may be co-incident factors in recent years as the European Central Bank (ECB) monetary policy also appears to be influenced by monetary policy in the United States, perhaps to reduce the extent of the exchange rate fluctuations that would otherwise emerge if no regard were paid to it. The asymmetry is more complex than simply tending to fight the threat of deflation more vigorously than that of excess inflation. Small fluctuations round the ECB’s objective are not fought with as much vigour as large deviations, irrespective of sign. While asymmetry in monetary policy is of considerable current interest because of its possible contribution to the present crisis, asymmetry in fiscal policy is also marked and in two respects. The first, which we consider in Chapter 7, is simply that increasing the size of the public sector appears to be counter-productive beyond a certain point. At lower ratios of public expenditure to Gross Domestic Product (GDP), increasing the size of the public sector tends to increase the rate of growth, especially when it is focused on infrastructural investment. At higher ratios, as are prevailing at present, the reverse is true and reductions in the size of the public sector are likely to increase the growth rate. We do not speculate on the causes but in combination with the asymmetry in labour markets discussed earlier it is clear that the modern tendency towards

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

The Nature of Asymmetry 5

active labour market policies rather than public sector employment in downturns makes a lot of sense. The second issue, addressed in Chapter 9, relates to the SGP (Stability and Growth Pact) among the EU countries and the asymmetry in the treatment of fiscal deficits and surpluses. The SGP is obviously asymmetric, for although it sets a long-term target of debt reduction, it explicitly rules against deficits greater than 3 per cent of GDP, except in hard times such as the present, but has no limit to surpluses. In this chapter we show that this asymmetry is clearly warranted by the asymmetric tendency of policy in the opposite direction. Across the period as a whole, the asymmetric problem has been that governments tend to be overoptimistic about how much they can cut taxation in the up phase of the cycle rather than taking the opportunity for consolidation. Behaviour changed clearly in the run up to the Economic and Monetary Union (EMU) and thereafter. Before that there was also some asymmetry in expenditure but nowadays expenditures tend to be simply counter-cyclical and expenditure reduction was the main tool used to get the member states to conform to the Maastricht criteria and the SGP requirements. We do not consider whether this has changed recently. We explore whether there have been changes in asymmetry as a result of EMU, or to be more literal coincident with the run up to EMU in each of the chapters. Indeed we test more widely for regime changes, particularly since structural breaks are often found around 1985 and 1992. While there are clear differences with respect to policy we also find changes in other sectors of the economy, particularly in regard to the determination of inflation. Before ending these introductory remarks it is worth pointing out that there is no accepted definition of asymmetry and as a result people use the word in a variety of contexts. One which is particularly confusing in the study of Europe is that many people use the word asymmetry when they simply wish to point out that the EU countries are different. Not only is that not the focus of our study but we find that we can pool EU data (with appropriate fixed effects) in almost all cases. There are some outliers and we explore where countries differ from the general run of behaviour.

1.2

Estimating asymmetry and its impact

Perhaps the easiest way to consider these differences in behaviour is just to assume that the coefficients in models will be different in the two

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

6 Asymmetry and Aggregation in the EU

circumstances of growth and recession. This would give us regime switching models, see Holmes and Silverstone (2005). Indeed one might want to argue that there are at least three phases in the business cycle, the downturn, the sharp recovery and the more normal period of growth. But this is not the only way to consider the problem. First of all it may only be some parts of the relationship that are different in the two circumstances, others may remain the same. An alternative is that the distribution of shocks that an economy faces may not always be the same and in recessionary times the economy is subject to downward shocks; see Kim and Nelson (1999), for example. This is a less attractive approach as it has no behavioural component to it. An alternative with a behavioural component to it is one that assumes that normal behaviour is not subject to asymmetry but that the asymmetry occurs in the response to shocks (Enders and Siklos, 2001). In output terms the response to an upward shock will be small (in price terms it will be large) but the response to a downward shock is much larger but eroded quickly (in price terms there will be little impact from a downward shock). An advantage of this model is that it will react in the same manner to further shocks, thus generating primarily a response through inflation on the upside and one through recession on the downside. In this book we primarily use the first approach, considering which variables might be likely to be subject to behavioural change under the two regimes. However, we also consider asymmetry in what is effectively the error correction mechanism. We discuss the methodology used in Chapter 2. One important issue is how the switch between regimes takes place. It could be approached in three obvious ways. One is simply to define a switchover point in terms of points in the cycle or other threshold values of relevant variables, such as the sign of the output gap for example. A second approach would be to look empirically for the changeover point on each occasion or perhaps to use a Markov switching model. Lastly one could assume that the change does not take the form of a simple switch but is progressive, in the form of a smooth transition from one regime to the other. We predominately use the first model in part because this simple approach illustrates the existence of asymmetry very clearly. We do however search for switching points rather than assume that purely arithmetic points, such as whether the economy is growing or contracting, are the only reasonable choice. We also consider the smooth transition model (Granger and Teräsvirta, 1993) in case this is a more realistic representation than the simple model. In some respects we are more inclined to the Sargent (2001) approach which assumes that the changeover between

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

The Nature of Asymmetry 7

8 Asymmetry and Aggregation in the EU

regimes is different depending upon whether the economy is entering downside or the upside regime. His suggestion is that the switch from good times regime to the recessionary regime is sharp, indeed this loss of confidence is what characterises the sharp decline. On the other hand, as the economy recovers, the move towards the good times regime is more hesitant and progressive. This is in itself would be an explanation of the shape of the economic cycle. Whether inflation and unemployment are subject to the same shape of switch between regimes is a more open question. Sudden bursts of inflation are for example rather more common than sharp declines.

A final remark

Chapter 2 also develops our concern with the problems of aggregation in the EU. Hence before we tackle any of the individual sectors we explore the degree to which the EU economies are different in the major characteristics of the macroeconomy (growth, inflation, unemployment, sectoral composition etc). We also explore the degree to which there has been convergence in the economies as the process of European integration has increased, in particular looking at the changes associated with the advent of EMU. Whether it is because of the deepening integration is difficult to say but there is a clear reduction in dispersion of macroeconomic performance over the period in all respects. The nature of the convergence is however relatively complex. Inflation rates have converged because of a general trend in the developed world towards inflation focused monetary policy. Thus the convergence outside the EU in the rest of the Organisation for Economic Cooperation and Development (OECD) has been similar. Indeed in some cases, such as Ireland, membership of EMU has resulted in an increase in inflation as the common monetary policy has been aimed at the euro area’s overall inflation and not that of the more dynamic parts. Prior to membership Ireland was controlling its own inflation, and indeed all countries needed to do so if they wished to qualify for EMU under the Maastricht criteria in the first place. In the case of unemployment, countries with higher levels of unemployment, such as Finland, were able to bring their unemployment down through faster growth. In some respects therefore there is a trade off between convergence in levels and convergence in rates of change. Convergence in levels necessarily results in diverging in rates of change and vice versa. Our analysis is timely in the sense that the present crisis has illustrated all too vividly how sharply behaviour can change in a recession,

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

1.3

The Nature of Asymmetry 9

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

particularly with the closure of wholesale financial markets when confidence was lost and spreads jumped sharply. However, the analysis is also in a sense premature as the data generated during this period will be the most relevant, given that behaviour over the previous 30 years has been mild by comparison. No doubt we will rework our analysis as the data become available but that must wait for another book when the shape of this cycle is clear.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

2

In this chapter we explore first the problems of estimating asymmetry before going on to consider the extent of the aggregation problems that we face in dealing with a group of countries rather than with aggregate data for the EU as a whole. However we begin by summarising the model employed. The individual equations and their justification are developed in subsequent chapters.

2.1

The model in outline

In order to examine asymmetries in macroeconomic behaviour in the EU/EEA we use a simple and very conventional four equation model of the economy, consisting of an IS curve, a Phillips curve, an Okun curve and a monetary policy reaction function that we have employed earlier (Mayes and Virén, 2005).1 We augment this to include fiscal policy, showing how expenditure, revenue and the net position are affected by the business cycle, debt and interest rates. 2.1.1

The IS curve

Following Duguay (1994), Goodhart and Hofmann (2000), the IS curve is of the form ∇yt = a0 + a1∇yt–1 + a2∇yt–2 + a3rrt–i + a4ret–j + a5∇y*t–k + εt

(2.1)

where ∇y is the deviation of output y from its Hodrick–Prescott (HP) filtered trend, rr is the real three-month interest rate (i.e. the nominal 1

A three equation version, omitting the labour market has received considerable attention – Cho and Moreno (2006). 10

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Estimation and Aggregation Concerns

rate of interest r less the annual rate of consumer price inflation p), re the real exchange rate (in logs), ∇y* the deviation of OECD output from its HP trend (lag lengths i, j and k typically vary from 1 to 3 quarters in estimation)2 and εt is the error term. Because the construction of output gap is a controversial issue we have also carried out all empirical analysis using the growth rate of output instead (to be denoted as ∆y). As can be seen from the more detailed exposition in Chapter 3, these two measures are far from identical in describing cyclical movements and it is no surprise that the results are somewhat sensitive to the choice of output measure. Fortunately, none of the results is crucially dependent of this choice. The IS curve represents a quite conventional open-economy demand relationship in which real interest rate and real exchange rate represent the main ingredients. The open-economy nature of the equation is further re-enforced by the foreign export demand variables. In addition, we include two wealth (shock) variables, the rate of change of real house prices (hp) and the rate of change of real stock prices (sp). Then the estimating equation reads: ∇yt = a0 + a1∇yt–1 + a2∇yt–2 + a3rrt–i + a4ret–j + a5∇y*t–k + a6hpt + a7spt + εt

(2.1′)

where lag lengths i and j may be different and may also differ over countries. The basic equation is completely backward-looking but we also estimate the model as a hybrid Euler equation for total output which takes the form: ∇yt = a0 + a1∇yt–1 + a2∇eyt+1 + a3rrt–i + a4ret–j + a5∇y*t–k + a6hpt + a7spt + εt

(2.1″)

Estimation of (2.1″) is obviously more complicated because we have to deal with expectations. Then we must either have some (survey) data on expectations or we have to estimate (2.1″) with the joint hypothesis of Rational Expectations which necessitates the use of the

2 As for the real exchange rate, we have used two alternative measures: the real exchange rate in terms of the US dollar and the real effective exchange rate against major currencies.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Estimation and Aggregation Concerns 11

12 Asymmetry and Aggregation in the EU

Generalised Method of Moments (GMM).3 Equation (2.1′) is useful also in the sense that it allows to us compute the co-called Financial Conditions Index (FCI). Technically, it can be derived from (2.1′) in the following way: FCIt = (1/aˆ3)(aˆ3rrt–i + aˆ4ret–j + aˆ6hp + aˆ7sp)

(2.2)

where the carets denote estimated values. This index thus reflects the weight of all the financial variables in the model on the real output.4 2.1.2

The Phillips curve

The starting point is the following standard expectations augmented Phillips curve: (2.3)

where ∆p is (actual) inflation, ∆pe is expected inflation, p* is the foreign price level (in domestic currency) and un is unemployment. However, in common with many authors (Galí and Gertler, 1999) we typically use ∇y instead of the unemployment rate, un, to represent the pressure on the economy. This then represents a form of the New Keynesian Phillips curve on the grounds that the output gap may move in step with marginal cost.5 However, in the form set out in (2.3) the curvilinear property of the relationship is largely lost so we augment it to show two facets either side of a threshold, where ∇y+ denotes the values of the output gap that exceed the threshold value τ (if ∇y > τ, ∇y+ = ∇y while if ∇y ≤ τ, ∇y+ = 0). Accordingly ∇y– denotes the remaining values

3 The issue of expectations will be dealt with later on but some comments already merit note here. From the Rational Expectations Hypothesis (REH) point of view ‘actual values’ of future (say, period t+1) inflation deviate from expected inflation by a stochastic disturbance. In a sense, the disturbance is a measurement error. Also survey data probably include a measurement error (in terms of the ‘true’ expected inflation) but this error is not the same as the ‘forecast’ error which is related the REH. This (latter) error ought to be orthogonal to the information set that is used in forming expectations. The corresponding orthogonality conditions can be utilised with the GMM estimator and then no survey data are required. The possible measurement error with the survey data could, in principle, be handled with the Instrument Variable (IV) estimator but it is far from obvious which instruments should be used. 4 We also solve the full model for these effects, which is a more complete procedure than the simple FCI used in the literature. 5 This model is used by Goodhart and Hofmann (2005) successfully in examining both the euro area and the US.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

∆pt = b0 + b1∆pt–1 + b2∆pet+1 + b3∆p*t + b4unt + ηt

Estimation and Aggregation Concerns 13

of ∇y.6 As an alternative, we use the so-called Smooth Transition Regression (STR) model to account for possible nonlinearity of the output effect (see the following section for details). In both cases this allows the impact of output to be different in booms and busts, with the expectation that the effect is stronger when the output gap is positive than when it is negative. In other words b4 > b5 in (2.4).7 (2.4)

When estimating the Phillips curve (2.4) we are usually interested in the parameter values of future and past values of inflation. Even though that is also important from our point of view, the main emphasis here is related to the potentially nonlinear role of the output (marginal cost) variable. Thus, (2.4) is estimated with the various threshold model techniques (see below). Because the Phillips curve is crucially dependent on inflation expectations, there are various problems in estimating the equation (already in the linear form). The conventional solution is to use the GMM estimator and so do we in the current context. As alternatives, we use some survey data and also some (more old-fashioned) backward-looking specifications. As for additional variables, we also experiment with consumption (VAT) taxes. In testing the nonlinearity hypothesis, we mainly use crosscountry panel data which allows for exploiting cross-country differences in labour market conditions (unemployment). More interesting applications can be developed, however, by making use of regional data from the EU. These data include a lot of more variation because intra-country differences in unemployment can be taken into account. For practical reasons, the estimating equation is fitted into crosscountry data but we include intra-country unemployment dispersion variables as additional regressors into (2.3) which now reads: ∆pt = b0 + b1∆pt–1 + b2∆pet+1 + b3∆p*t + b4un+t + b5un–t + b6dispt + ηt

(2.5)

where the disp variable simply reflects either the range or the standard deviation of unemployment rates over regions in a country i. Ideally, 6 Obviously we could have more than two regimes (facets) for ∇y but since we have only limited numbers of observations we use this simple specification (which has been widely used elsewhere, see Yates (1998) for instance). 7 There is some analogy between the Phillips curve and the wage curves and the price equations. Especially with the latter, we have a lot of evidence that (with the micro data) prices tend to increase faster than they fall (for a useful summary, see e.g. Peltzman (2000).

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

∆pt = b0 + b1∆pt–1 + b2∆pet–1 + b3∆pt* + b4∇y+t + b5∇y–t + ηt

14 Asymmetry and Aggregation in the EU

we ought to use regional data for all variables but, unfortunately, regional data are available only for unemployment and (to some extent) output, but not for prices or inflation, or other macro variables such as taxes, exchange rates and so on. 2.1.3

The Okun curve

In its simplest form the Okun curve can be expressed as: ∆Ut = c0 + c1∆yt + εt

(2.6)

where U denotes the unemployment rate and y output. In the empirical analysis we allow population shocks (measured by the working-age population pop) to affect unemployment and introduce simple dynamic adjustment in the form of an error-correction model to derive the following estimating equation: (2.6′)

which reflects an error-correction format.8 Here ∆y is the growth rate in GDP (alternatively, the output gap, ∇y, is used), pop the population of working age and EC the error-correction term (that is lagged by one period) and τ a threshold value for the asymmetry. Prachowny (1993) inter alia argues that some scaling of the labour variable in (2.6) is required so we have also included population of working age in our formulation. Once again we use a threshold approach to the relationship to allow at least some approximation to a curvilinear relationship. Equation (2.6) is estimated using a two-step procedure. First a long-run relationship between employment, output and working-age population is defined and on the basis of this relationship an error-correction term EC is derived. This term is used in the Okun curve specification to take care of long-run relationship between these (quite clearly) nonstationary time series. Thus, the Okun curve is not a simple bivariate relationship but it takes into account possible population shocks and cyclical movements in output. In our mind, the Okun curve is the most obvious candidate for nonlinear economic relationships. If there is any sort of wage-stickiness one might expect that output (demand) changes translate in a different way to changes in employment. Thus, the unemployment effects ought to be larger in depression than in boom periods.

8 The EC term is derived from the level form terms of U, y and pop. In the panel setting, country-specific fixed effect terms are also introduced. See Kiander and Virén (2001) for details of the derivation of this model.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

∆U = c0 + c1∆y+(τ) + c2∆y–(τ) + c3 ∆pop + c4EC–1 + τt

Estimation and Aggregation Concerns 15

2.1.4

The Taylor rule

Finally, we include a monetary reaction function in the form of a Taylor rule rt = θrt–1 + (1 – θ){d0 + d1(πt – πo) + d2(∇yt – ∇yo)}

(2.7)

where the parameter θ permits an element of interest rate smoothing. π is the rate of inflation (which equals ∆p) and π0 its target value. Similarly ∇y0 is the target value of output gap which obviously can be zero (Huang et al., 2001). Rearranging terms leads to the following estimating equation: rt = ρ0 + ρ1rt–1 + ρ2πt + ρ3∇yt + ut

(2.7′)

– allowing for a nonlinear structure in terms of the inflation effects – including the asset prices (more precisely, the change rates of house and stock prices) as additional regressors in the estimated Taylor rule equation. Thus, the final estimating version of the Taylor rule reads: rt = ρ0 + ρ1rt–1 + ρ2π+t + ρ3π–t + ρ4∇yt + ρ5HPt + ρ6SPt + ut

(2.7″)

where HP and SP denote the change rates of nominal house and stock prices. π+ and π– denote inflation rates above and below some critical level (e.g. 2 per cent). Here we have also experimented two thresholds to allow for corridor-type behaviour: interest rates react only to sufficiently large deviations from the target level. The Taylor rules, in the same ways as most economic relationships, could be specified in many different ways, e.g. by using expected inflation instead of actual inflation but here we stick into the most common ways of specifying this and the other three equations. Some differences arise already because of the choice of the inflation variable; we may use either the GDP or consumption deflator, or the CPI (national, or harmonised CPI) in deriving the inflation data. For reasons of robustness checking, we have, in fact, used all these indicators. Equation (2.7′) could be interpreted as some sort of ‘activist’ policy rules where the central bank not only pays attention to price stability (and cyclical situation) but also react to developments in asset prices (state of financial markets). This asset price reaction might just follow

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

where ρ0 represent the sum of the constant terms. In addition to this standard linear model for interest rates, inflation and output we have estimated the policy rule (2.7) with two extensions:

16 Asymmetry and Aggregation in the EU

from the idea that asset prices convey information on future (consumer) price changes but from a more ‘activist’ point of view reactions to asset prices could be motives by a desire to avoid the harmful consequences of asset market bubbles. If that is the case, one might think that the central bank only reacts to substantial deviations from equilibrium levels, that is, to very large cumulative changes in prices. This interpretation would suggest that also the asset price effect is probably nonlinear. As pointed out in Chapter 1, the question whether to react to asset prices is quite controversial. Irrespective of the policy conclusions, it would be useful to test whether in the past, central banks have indeed reacted to financial market developments or not. 2.1.5

The full model

2.1.6

The fiscal sector

We take a straightforward approach to estimating the evolution of public sector balances, revenues and expenditures. (d/y–)t = f0 + f1(d/y–)t–1 + f2t + f3∆y–t + f4∆y+t + f5rt + f6(D/y)t–1 + νt

(2.8)

–

where d is the fiscal measure, y is trend GDP, t is a time trend, r is the nominal interest rate and D government debt. Here the purpose is to test the symmetry of the response in upturns and downturns in the economy. This equation is not fully integrated with the rest of the model, in part because fiscal decision-making tends to follow an annual cycle and hence we do not have exactly matching data.

2.2

Estimation of the model

As mentioned in Chapter 1 there are at least two obvious ways for handling asymmetry in estimation. The first is simply to assume that there are different regimes and that some or all of the parameters differ across regimes. This is the approach of Holmes and Silverstone (2005) to the Okun curve for example. We also need to decide whether there is a simple switch from one regime to the other or whether the transition is progressive. One alternative is to assume that the underlying relationship does not change but it is the response to shocks which varies, dependent either upon whether they are positive or negative, or

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

This set of equations determines inflation, output, unemployment and the rate of interest. Foreign prices, foreign output and the exchange rate are treated as exogenous to the system. The same is true with asset prices, taxes and the labour force.

upon the type of shock. We use both these approaches but we concentrate on a simplified version of the first, namely to see whether the parameter of key variables is different in the two regimes. This approach is often described as using threshold models. The regimes are determined on the basis of some transition (threshold) variable. The existence of this transition variable is important in distinguishing threshold models from other switching regime models where identification of regimes is based on properties of the model, an issue we return to at the end of this section. One may question the choice of threshold models because within the set of nonlinear models there are various alternatives. The reason becomes clearer if we briefly consider some obvious alternatives but we should stress the fact that we do not only want to test whether there are asymmetries but we want also to have a clear estimate of their nature and magnitude. For this purpose, the threshold model is ideal because the model gives an exact characterisation of regimes as well as the difference in regimes. Thus, such constructions as bilinear models are not very useful.9 This reasoning also means that we are not as such interested in testing the existence of asymmetries although, of course, such testing is no doubt useful and gives guidance for specification and application of nonlinear models.10 The critical values of the transition variable are estimated from the data using some grid search procedure (that minimises the residual sum of squares). We have also made quite extensive use of some intuitively obvious threshold values. Thus, for instance, in the case of the output gap we have used the value of zero, which, in the case of Hodrick–Prescott filter, also comes quite close to the sample average values. Also with the growth rates of output and real asset prices using the zero value as a benchmark makes the interpretation of results easier. There is also a practical reason for using this kind of a priori threshold values: the estimation of the threshold values in the context of e.g. the so-called smooth transition regression models has been found to be notoriously difficult and it is has been found very difficult to judge the accuracy of the estimates.

9 In fact, there is a correspondence even between threshold models and bilinear models (cf. Priesley 1988 for details). 10 Here might refer to Verbrugge (1997, 1998) which shows (using the nonparametric triples test of Randles) that there are clear asymmetric features in many macroeconomic series, most notably in price series. Moreover, the existence of asymmetries varies a lot across countries.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Estimation and Aggregation Concerns 17

18 Asymmetry and Aggregation in the EU

The basic threshold model can be expressed as 

α11yt–1 + α12xt + u1t, if st < s0 and

 

α21yt–1 + α22xt + u2t, if st ≥ s0

(2.9)

with Var(ui) = δ2i. Here s is the transition (threshold) variable and s0 the (fixed) threshold value. The transition variable is here indexed for period t but obviously it can have a lag.11 Usually, the lag length is also unknown which obviously complicates the problem. In our exercise the lag length is always zero. Statisticians very often use a version (2.9) which only includes autoregressive terms on the right-hand side. By definition they are Threshold Autoregressive (TAR) models. In these models, the transition variable is in most cases some lagged value of y. For obvious reasons, the emphasis in TAR models is the dynamics of the y variable. These models are quite popular in ecological applications but not so much in economics except for applications in finance, of course, where univariate models could be considered on a priori grounds, or in error-correction processes (Enders and Siklos, 2001). Within the set of threshold autoregressive models, we may distinguish a set of momentum threshold autoregressive models (M-TAR) in which the transition variable is the difference of the dependent variable. With such models (see, e.g. Enders and Granger (1998)), we may test whether the time series are more persistent with positive growth and vice versa. Enders and Granger’s (1998) application with interest rates gives, in fact, strong support to this notion suggesting, more precisely, that interest rates respond to the deviation from the equilibrium level only when the discrepancy is positive. It is easy to find a straightforward application to a threshold model from economics. Take for instance a simple model inflation where inflation is positive with positive values of ouput gap (or unemployment below the nature rate of unemployment) but zero with negative values (Figure 2.1). Then the output would be the transition variable and zero the corresponding threshold value. Obviously, there could be leads or lags in the transition variable and nonzero threshold values but the main concern would be a difference in the relationship between inflation and output over the cycle. When using specification (2.8) we assume that all parameters and the error term are different for these two regimes (obviously the number of regimes can be more than two although we mostly deal with the 11

This is not to say that dynamic adjustment towards equilibrium is unimportant. Results in Holmes and Maghrebi (2005), for example, suggest that, with interest rates at least, the speed of adjustment is larger with increasing rates.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

yt = 

Estimation and Aggregation Concerns 19 Figure 2.1

Illustration of a simple threshold model with cyclical data

3 gap inflation

2

1

0 11

21

31

41

51

–2

two-regime case). In other words the whole data generating process is different. It could, however, be that the regimes are not that much different and only part of the parameters vary over regimes. Thus, we could arrive at the following simple case: yt = α1yt–1 + α21x–t + α22x+t + u1t , where xt = x–t if st < s0 and xt = x+t if st ≥ s0

(2.10)

In this specification, nonlinearity would only apply to the slope of x, not the dynamics of the model (here the lagged y term) or the error term. In other words, the stochastic structure would be the same for the both regimes (in particular, the variance of the error term u would be invariant over regimes and hence δ21= δ22).12 Equally well, we could imagine that nonlinearity would apply to the ‘speed of adjustment’, in (2.4), the α1 parameter.13 Accordingly, if we had an error-correction model, the coefficient of the error-correction term would be a possible candidate for regimes shifts. In an error-correction set-up, we would, in fact, 12

A common alternative to (2.10) is to estimate yt = a1yt + a2xt + a3Dxt + u1t , where D = 1 when st ≥ s0 and 0 otherwise. 13 There is also an extensive literature of cases where the condition variance of the error terms is subject to nonlinear dynamics in the ARCH/GARCH framework. Here focus, however, entirely on coefficients of the behavioural/mean) equations.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

–1

20 Asymmetry and Aggregation in the EU

have several alternative ways of expressing the regime-sensitiveness: both the long-run coefficients and the short-term impact effects plus the speed of adjustment parameter(s). In specifying the model, we have not, in general, used the error-correction structure.14 Hence, the choice of nonlinear elements in the estimating equations is more straightforward. In a typical case, with the Phillips curve, we focus on the output variable, i.e. the impact of output (or, real marginal costs) on current inflation. This is not say that we firmly believe that the past of future (expected) inflation only affects in a linear way. Rather it is the case that testing all features of nonlinear elements in the basic equations is beyond the scope of our discussion. Although we mainly deal with the two-regime case, more complicated specifications may make sense. In more concrete terms, we have found some evidence on a corridor type adjustment, especially in the case of policy (Taylor) rules. Thus the model would be of the following general structure:

(2.11)

It could be, for instance, that α22 is zero because small deviations from target values of interest rates do not lead to any policy reactions. Only if the values of s deviate ‘enough’ could such a reaction occur. In the case of policy reaction functions such behaviour could be rationalised by e.g. measurement errors: if the policy-maker knows that the transition variable (say, inflation) includes some measurement error, immediate reactions to small deviations from the target value would not be optimal policy. Alternatively, one could defend the ‘corridor’ specification by adjustment costs. If changing interest rates (in the case of a Taylor rule) required some discrete costs, policy reaction would be delayed until the deviation from the target value is large enough. There could also be some ‘political economy’ or ‘psychological’ reasons. One very interesting feature of threshold (autoregressive) models is that they can give rise to limit-cycle behaviour. In this respect they come close to features of nonlinear differential equations. Thus, if we switch off the stochastic term the threshold autoregressive model may have a solution that has an asymptotic periodic form. 14

In the literature, two classical (Canadian) datasets have been used quite extensively, one dealing with lynx and the other with mink and muskrat. In these examples, the main issue was the dynamics of these animal populations, which showed quite complicated asymmetric lag structures. See Tong (1983) for details. For more recent ‘economic’ applications see Kavler et al. (2008) and the references cited therein.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

 α11yt–1 + α12xt + u1t, if st < rL  yt =  α21yt–1 + α22xt + u2t, if sL ≤ st ≤ sU   α31yt–1 + α32xt + u3t, if st ≥ sU.

Estimation and Aggregation Concerns 21

Although the basic threshold model is intuitively appealing it has some questionable properties. Above all, one may find it difficult to accept the idea that the regime change represents a discrete jump. A small change in the transition variable (in the neighbourhood of the threshold value) may change all parameters completely. An obvious remedy is to introduce some smoothness into the transition of the parameters. That is easy with the simplified version of the threshold model (where the threshold only affects some of the parameters). Thus, we could arrive at the following specification: yt = α1yt–1 + α2xt + [α3 + α4G(s)]xt + ut

(2.12)

where the indicator function G would be defined for the interval {0, 1}. In practice, G could be either an exponential or a logistic function. In the latter case, which is more frequently used later on, we may write G(r) in the following way: G = 1/{1 + exp[–γ(st – s0)]}

where γ is the smoothness parameter which defines the shape of the threshold. By setting the parameter very high we end up with a switching regression (threshold) model as can be seen from Figure 2.2. By contrast, if γ equals to zero, G = 1⁄2, and we end with a linear regression model. Figure 2.2

A comparison of a rigid and smooth threshold G 1

0.5 γ=1 γ = 10

1

St

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

(2.13)

22 Asymmetry and Aggregation in the EU

If we have more than one threshold (the ‘corridor’) we can simply generalise G accordingly. Thus, in the case of two thresholds, we may express the logistic G in the following way: G = 1/{1 + exp[–γ(st – s0)(s – s1)]}

With a sufficiently large value of the smoothness parameter we can derive a corridor type nonlinearity for the effect of x on y (see Figure 2.3 for an illustration). The figure is drawn in such a way that the threshold effect is not symmetric: there is a ‘corridor’ but the effect of x on y differs depending on which side of the ‘corridor’ we are. We could imagine that, for instance, the central bank does not react to small deviations from the inflation target but it reacts to high inflation and low inflation (deflation) differently (for more details, see e.g. Granger and Teräsvirta (1993)). Estimation of the smooth transition regression model is in principle straightforward: it can be done with maximum likelihood or nonlinear least squares (which here represent suitable quasi-maximun likelihood estimation). It is only that some computational problems may arise, in particular with the smoothness parameter γ in the context of smooth transition models. One may also find some problems in testing the sequence of thresholds but may bypass these things and only refer to Figure 2.3

An effect of smoothing on a two threshold model G 1

0.5 γ=1 γ = 10

–1

1

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

St

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

(2.14)

Estimation and Aggregation Concerns 23

Granger and Teräsvirta (1993), Hansen (1999a, b, 2000) and Luukkonen et al. (1988)). In the case of switching regression threshold (2.10) testing is reasonably straightforward because we can just test the parameter restriction α21 = α22 with the conventional Wald test. With smooth transition models things are more complicated because if γ = 0, the linear model (H0) is not identified. Hence, LR, Wald and LM tests cannot be used. One way to circumvent the problem is to linearise the transition function G which in the logistic case gives us the following testing equation: (2.15)

so that the test boils down in testing the parameter restriction β3 = 0 (cf. Luukkonen et al., 1998). When estimating the model, we face the additional problems of a panel data setting which is the typical environment for estimating the reported equations. Estimating dynamic equations from panel data is not the easiest exercise in econometrics but when we add nonlinearity things can be quite complicated. We discuss these problems more in the context of reporting the results but here a couple of most compelling problems should be mentioned, at least. When using the threshold model, we typically have to assume that the threshold parameter is constant over time and over countries or regions. The later assumption is not, of course, completely innocent. Countries might well differ in terms of the explicit inflation target, or the critical value of the state of product markets. In countries where markets function ‘well’, wage or price stickiness could be of less importance and hence inflation or employment reactions show larger ‘tolerance’, which in turn would show up in larger values of output gap, for instance. Basically, the problem can be solved by using individual country data, and so is in fact done in the subsequent analysis. Another weak point is the dynamics of the model that basically boils down in the speed of adjustment parameter. We have good reasons to believe that these parameters differ across countries and estimating equations from single country data confirms this presumption. Still, most of the reported results represent specifications where the dynamics is assumed to be similar across countries. This is motivated by our desire to concentrate on the short-run effects. Only when we find really substantial differences in speeds of adjustment we allow these parameters to vary over the cross-sections (countries). Even now, the conventional way of estimating above-mentioned linear equations (2.9)–(2.11) is to use least squares with fixed cross-section

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

yt = β0 + β1yt–1 + β2xt + β3stxt + u1t

effects. Efficiency can be increased by using the Generalised Least Squares (GLS) (to account for different error variance) and the Seemingly Unrelated Regression (SUR) estimator to account for correlated error terms. There are, however, several obvious problems in estimating the models. First of all, the data may not be completely stationary. More serious problems arise however, because of the dynamics. Most of the models include a lagged dependent variable (or an error-correction term, see the Okun curve) which, in the panel setting, makes least squares a rather poor estimator. An obvious choice is to move the use of the Generalised Method of Moments (GMM) estimator which in practice means using the Arellano-Bond estimator. Although, that ought to be the right thing we know that the estimator is often very sensitive to instruments and other features of estimation. Thus, in this way, we have to look at a wide range of estimates with different estimators and, as usual, with different data samples and variable measures. With these kinds of data, one cannot disregard the endogeneity problem, either. Take for instance the Phillips curve and the Okun curve. In a sense, all variables are endogenous and, moreover, the expected future values are also random (unless we have some survey data for them). In an IS curve, interest rates drive output and in the Taylor rule, output determines the interest rate. In this kind of setting, least squares seem a quite poor alternative but it is still used as a basic alternative. The reason is simple: we need some benchmark values that are reasonably robust and computationally well-behaving for various comparisons, especially when we move to nonlinear specifications. As pointed out earlier, nonlinear threshold models are not the way of allowing a nonlinear structure in the particular transmission mechanism. Earlier, in the 1970s and 1980s, a lot of work was done using socalled disequilibrium models which took the form: q 1 = a 0 + a 1P + a 2Y + u 1 q 2 = b 0 + b 1P + b 2X + u 2 q = min(q1, q2)

(2.16)

where superscripts denote possible regimes which could, for instance, be interpreted as excess demand and excess supply regimes so that the realised values were determined on the basis of the minimum condition (see e.g. Quandt (1988)). The identification of regimes is obviously more difficult than in the case of threshold models because the choice of regimes cannot be based on any observable data. The minimum condition (‘short-side rule’) provides the necessary information for each regime but still there are serious difficulties in identification of

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

24 Asymmetry and Aggregation in the EU

Estimation and Aggregation Concerns 25

regimes. The minimum condition in (2.16) will fail if there are some frictions in the market and the markets are not efficient. Then, we have to adopt a less stringent condition q < min(q1,q2) which is obviously less informative in identification of regimes because we do not know how far actual transactions are from intersection of (notational) demand and supply curves. This (so-called Muelbauer-Hajivassiliou) condition with a conventional supply-demand model is illustrated in Figure 2.4. In addition to these conceptual issues, it is probably fair to say that the disequilibrium models (of type equation (2.16)) have been notoriously sensitive to the estimation procedure (cf. e.g. Stenius and Virén, 1984). Moreover, the disequilibrium models have been notoriously sensitive to the estimation procedure. In our case, this kind of disequilibrium specification is not very attractive because in general we do have some idea how the regimes are Illustration of the minimum conditions

S

E(Q)

D

Q The left-hand side of the supply and demand axes stand for the fixed minimum condition (bold lines) while E(Q) illustrates the Muelbauer-Hajivassiliou minimum condition for heterogeneous markets.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Figure 2.4

26 Asymmetry and Aggregation in the EU

M=

p11 1–p11 1–p22

(2.17)

p22

Denote the path of the (unobserved) two-state first order Markov process that would take values 0 or 1 by st and, accordingly the observed time series that depends on st, by xt. Then the density of xt conditional to st would be: F(xt|st, β) = (2πσ)–1⁄2 exp[–(xt – us)2/2σ2]. On other words xt would be uncorrelated white noise with switching mean. The states could be characterised as ‘high’ and ‘low’ which would be useful in characterising e.g. business cycle fluctuations and inflation dynamics. If we were just testing the existence of regime changes, this approach would be appealing. We do, however, want to show that regime changes appear to follow some specific pattern that can be easily traced from observable variables (most often from the cyclical situation). That is why we here concentrate on the threshold models. Even so, it is worth pointing out that the menu of ‘traditional’ nonlinear models is quite rich. Take for instance, the Brechling’s (1973) estimating equation which he used in testing the so-called nonlinear aggregation hypothesis. This equation was derived from the following model: wt = bg(Ut) + 2bg″(Ut)Var(Ut) + k0pt + a′C(wt – k0pet)

(2.18)

where g corresponds to some nonlinear function (of the unemployment rate) Var(U) being the variance of cross-sectional unemployment

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

related to observable indicators (transition variable) so that identification of regimes can be done with the help of those data. Moreover, in our case the regimes usually differ from each other only in terms of the parameter values – not in terms of the variables or the functional form. Needless to say, in such a case the minimum condition may not make sense at all. Thus, rather than the disequilibrium model, we could used other econometric tools like the state-space models (see Hamilton (2008) for general reference). In Hamilton’s regime-switching models the big difference vis-à-vis the threshold models is the fact that the regime switches occur with latent (unobserved) states and not with observed states as in the threshold models. In the model, time-series dynamics are governed by a finite-dimensional parameter vector that switches depending on which of the two unobservable states is realised with the state of transitions governed by a first-order Markov process. Thus, in the simplest case, we would have a transition probability matrix which could be of the following form:

rates. b, k, a and C reflect the underlying microparameters. Brechling used various nonlinear functions (such as log(U) or Uθ) in deriving the final estimating form. The problem with this kind of functional forms is the fact that while they allow for testing linearity they do not really provide much help in economic interpretation of the results. Moreover, they typically lead to quite complex forms which create their own computational problems. Some of the nonlinear structures may be difficult to express with explicit functional forms. Take for instance the plucking model which was originally introduced by Friedman (1993). Friedman proposed a model of business fluctuations in which output cannot exceed a ceiling level, but will, from time to time, be plucked downward by recession. The model implied that business fluctuations are asymmetric, that recessions have only a temporary effect on output and that recessions are duration dependent while expansions are not. This behaviour is illustrated in Figure 2.5 below. This model could perhaps be expressed by means of the minimum condition of the type q = min(capacity, Q(t)) where Q represent the level of output obtainable in the absence of any capacity constraints (as a consequence of, say, demand disturbances). Even so, it is not

Figure 2.5

Friedman’s plucking model

REAL OUTPUT

BUST

BUST

BOOM (RECOVERY)

BOOM (RECOVERY) BOOM (RECOVERY) BUST

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Estimation and Aggregation Concerns 27

28 Asymmetry and Aggregation in the EU

immediately clear how this can be transformed to an estimating (and testing) equation. Instead one may simply test whether the cyclical movements of, say, GDP are symmetric by using various time-series testing procedures. Here we may refer to Diebold and Rudebusch (1990) who examine the basic business cycle summary statistics and find that with historical data Figure 2.6

Kernel densities of output growth with the EU27 data for 1990–2008

2.0 G*(G<0)

1.0

0.5

0.0 –12

–10

–8

–6

–4

–2

0

2

0.20 G*(G>0)

Dens ity

0.16

0.12

0.08

0.04

0.00 –2

0

2

4

6

8

10

12

14

16

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

18

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

De nsity

1.5

from the mid-1800s to the 1930s the mean duration of expansions was 31.5 months and mean duration of contractions 23.5 months. For the post-war data, the differences were even much larger. Thus with the US, the mean duration of expansions were 49.9 months and the duration of contractions 10.7 months only. This suggests that output growth may indeed follow some nonlinear pattern which would be roughly consistent with the plucking model. A look at the more recent EU data suggests that the distributional properties of positive and negative output growth values are indeed very different which again suggests that they be generated by different data generating mechanism (Figure 2.6). Assume for a while that output growth followed the plucking model. This would have powerful implications in terms of policy. Policy effectiveness would crucially depend on the state of economy vis-à-vis the capacity constraint. If the economy were close the constraint expansionary policy would be very much inflationary. Only if the timing of policy would just match the cyclical downturn might one alleviate the severity of eventual depression. As for the Phillips curve, it is hard to see that it could be linear in terms of the observable levels of output. Even if output were to grow very slowly there could be strong demand pressures and high rate of price and wage inflation. On the contrary, high growth rates of output would not necessary lead to immediate inflation pressures. As for a decrease of output, immediate price and wage effect might not materialise because of lagged demand pressures and wage and price stickiness. Obviously, the plucking model would change the interpretation of output data from the point of view of excess demand. Now low growth numbers of output follow either from low demand growth or from binding capacity constraints. The implications in terms of price pressures would be different indeed even if the Phillips curves were completely linear. If the underlying Phillips curve were nonlinear even some perverse results could emerge. It is very difficult to find remedy for this problem; perhaps the only thing that can be done is to use alternative proxies for cyclical situation (output growth, output gap, the unemployment rate and the deviation of the unemployment rate from the natural rate).

2.3

Aggregation

One issue which is highly relevant in the context of the EU is aggregation. Basically, the question is: does it make a difference if we consider the data for each individual country or just look at the average

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Estimation and Aggregation Concerns 29

30 Asymmetry and Aggregation in the EU

values of the whole euro area. From the point of view of economic theory, we know that aggregation matters. The case of a nonlinear Phillips curve (Figure 2.7) is usually presented as an illustration of this effect: with heterogeneous labour markets, an uneven distribution of unemployment leads to higher equilibrium unemployment than with well-functioning labour markets where unemployment rates are more or less equal (this is the basic idea in the so-called nonlinear aggregation hypothesis developed by Brechling (1973) which is discussed in more detail in Chapter 5).15 But Figure 2.7

Implications of the convexity of the Phillips curve

1

α

A

E(u) 0

u

u1 u2

u*

–1

B ∏

Note: The figure reflects the convexity of the Phillips curve ππ′. Assuming that the Phillips curve is of the form π = f(u – u*) + µ we may use Jensen’s equality in deriving the properties of zero-inflation equilibrium of the system. That is, E{f(u – u*) ≥ f[E(u – u*)]}; now knowing that f(u – u*) is decreasing with respect to u – u*, we can show that when E(π) = 0 so that E[f(u – u*)] = 0 (E(µ) = 0 by definition), E(u) – u* ≥ 0.

15

The nonlinear aggregation hypothesis basically originates from Lipsey’s (1960) dispersion hypothesis.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

∏

what are implications in terms of empirical analysis? It turns out that aggregation is of nontrivial importance both in terms of estimation and interpretation of the estimated results. In the European context, aggregation is surely an issue: we have common monetary policy which is based on aggregate (or average) values of the euro area variables but other policies mainly belong to national governments. Moreover, countries seem to differ a lot in terms of their level of income, structure of production and economic policy record (for details, see section 2.4). In this environment, there is an obvious conflict of interest because the governments of individual countries evaluate the operations of the Central Bank from their own perspective and it may well look that the policy pursued is not consistent with the economic environment of some individual country. The common monetary policy does not only manifest itself in ‘common’ monetary policy decisions but also in the way the whole policy analysis is carried out. A concrete example of this is the macro model for the euro area economy, more specifically the Area-Wide Model of the ECB (see Christoffel et al., 2008) which only uses the data from euro area aggregates or averages. Thus, cross-country differences, or dispersion in the relevant variables, do not matter.16 At least implicitly, this is based on assumption that the basic features of member countries are sufficiently similar that policy transmission can be treated as being roughly the same for all countries – from the perspective of modelling. Obviously countries and regions are different but the differences are not equally important. The fact that individual coefficients and elasticities are different obviously leads to severe estimation and interpretation problems. Take for instance the case illustrated in Figure 2.8. If countries differ a lot in terms of basic parameters it would be difficult to reach any affirmative conclusion of the nature of the eventual relationships. Moreover, the estimation results may be quite sensitive to details of the estimation procedure. Haque et al. (1999) shows that the conventional fixed effects model may produce highly misleading

16

It is obviously a complex matter to decide upon the extent to which the Governing Council of the ECB takes decisions based on aggregate rather than disaggregate information. The key issue here however is that some form of aggregation whether before or after estimation needs to take place and in practice substantial use is made of aggregated information.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Estimation and Aggregation Concerns 31

32 Asymmetry and Aggregation in the EU Figure 2.8

An effect of heterogeneity of estimation results

Y

country 1

country 3

X

results. In particular, the highly significant but nonlinear effects can be detected for some variables even though the basic structure is perfectly linear.17 Obviously, there is a simple remedy to this problem: that is a scrutiny of the results for individual micro units (countries). If results with the individual country data differ substantially from the pooled cross-country data (with common coefficients) one has to reconsider the estimation procedure with the pooled data. In what follows in this book, we do indeed estimate all specifications with individual country data even though most results are presented in the form of pooled cross-country data. Aggregation is also an important issue in the context of markets in disequilibrium. Then if we have some frictions which prevent markets from

17

The heterogeneity problem becomes serious if the slope parameter(s) are not fixed by somehow systematically relating them to background variables (see Haque et al. (1999) for details). Not surprisingly, the problems become more serious with dynamic (panel) models.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

country 2

clearing and, at the same time, prevent free movement of resources from market to market (i.e. markets are no more efficient), we end up with the case where, at the same time, we have (sub) markets with excess demand and excess supply (Figure 2.4). That in turn implies that transactions will be smaller than what the aggregated supply and demand curve would suggest. Again, we would have problems in identifying the true (nonlinear) data generating mechanism if we just had aggregate data (see Quandt, 1988 for more details). But things would be at least equally complicated when the underlying micro relationships were nonlinear. Then, even simple aggregation of data would produce great problems in identification of the true data generating process. This is shown by Granger in several papers (cf. e.g. Granger (1987, 1988)).18 Not only would nonlinearity make it difficult to assess the impact of different (policy) variables on the aggregate level but it would also make the interpretation of data generating mechanism very tedious. If the policy-makers use only aggregated data they could get the false impression that the data generating mechanism is linear even if the opposite would indeed be the truth. Hence, a crucial question is whether aggregation itself affects the properties of the data. If for instance, the country data generating processes were indeed nonlinear would this property disappear, or increase, or stay the same as we aggregate over countries? Obviously we could arrive at nonlinearities at the aggregate level even if the underlying models at the country level were linear if the random shocks were asymmetric (in a suitable way). This is an issue from which we know simply too little to make even rough conclusions (see however, Folk et al. (2004) and Jones et al. (1996)). Granger and Lee (1993, 1999) present some Monte Carlo evidence on impact on aggregation appearance of nonlinearity. They use some conventional nonlinear models to generate the data and apply several tests (the Neural Network Test, the Tsay test, the RESET test and the Dynamic Information Matrix test) to detect nonlinearities from series that had been aggregated in different ways. In general, the tests could detect nonlinearities only in a fraction

18

Granger (1988) shows that if the data were generated by an ARIMA model family (and the random terms were at most correlated contemporaneously) the aggregated series would still have the same structure. If there were some common factors, identification of micro/macro structures would become very difficult (even impossible) but the most severe problem would occur when the micro relationships were nonlinear. The properties of the aggregate data would grossly deviate from the micro data.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Estimation and Aggregation Concerns 33

34 Asymmetry and Aggregation in the EU

of (aggregated) series which obviously can reflect the low power of the tests or the fact that aggregation indeed destroys the nonlinear features of the data generating mechanisms. A similar result emerges in the case of temporal aggregation. Moreover, with temporal aggregation the loss of power seems to increase with the extent of aggregation. To investigate the importance of this problem, we have carried out some Monte Carlo simulations where the work-horse is the following simple threshold model: yt = ayt–1 + b1xtxt–1≤0 + b2xtxt–1>0 + ut

(2.19)

where u is a white noise stochastic term that is cross-sectionally either uncorrelated or correlated. The threshold model is, of course, a rather specific alternative of nonlinear models but we know that it provides a Figure 2.9

Kalman filter estimates of threshold model coefficients

coefficient

0.5 0.4 0.3 0.2 0.1 0.0 –0.1 –0.2 –3

–2

–1

0

1

2

3

individual x 8

coefficient

4 0 –4 –8 –12 –16 –0.8 –0.6 –0.4 –0.2 0.0 0.2 average x

0.4

0.6

The data are simulated with sample size of 100.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

0.6

reasonable good approximation of different nonlinear models. To test nonlinearity, we have used both the BDS test and a Wald test to test whether a threshold model structure can still be detected in the aggregated time series. The number of cross-sections has been either 15 or 27 reflecting the number of countries in the euro area (or the ‘old’ EU) or the countries of the ‘current’ EU. A summary of findings is reported in Tables 2.1 and 2.2 and Figure 2.10. Figure 2.9 illustrates the effects of aggregation in a way a bit similar to that which Priestley (1998) uses for the identification of parameters of state dependent processes. This involves generating data with model (2.19) using parameters on row 1 in Table 2.1 and estimating a simple Kalman Filter model of the type yt = a0yt–1 + atxt + e1t, where at = at–1+ e2t and graphing the (smoothed) estimates of a against variable xt–1 with a single micro (country) data series and with aggregated data from 27 micro units. Quite clearly, the nonlinear structure can be detected from the micro data but no longer from the aggregated data (here we do not need the results for the original micro data – in practically all cases the hypothesis of linearity (= equal coefficients) is rejected). With the aggregated data it looks like there is no relationship between y and x. In Table 2.1, the numbers displayed are percentage rejection frequencies of a hypothesis c1 = c2 which implies a linear model (symmetric behaviour above and below the threshold). In the Table, the 6th model is a linear model (so, the data are produced with a linear model and a nonlinear threshold model is estimated from the corresponding aggregated data). The results suggest that, if the number of micro (cross-section) units is 15, aggregation does not completely remove nonlinear structure form the Table 2.1

LR test results for a threshold model

Simulated model 1: c0 = 0.15, c1 = –0.10, c2 = 0.50 2: c0 = 0.15, c1 = –0.10, c2 = 0.10 3: c0 = 0.15, c1 = –0.50, c2 = 0.50 4: c0 = 0.50, c1 = –0.50, c2 = 0.50 5: c0 = 0.50, c1 = –0.10, c2 = 0.50 6: c0 = 0.15, c1 = 1.00, c2 = 1.00

n = 15, 5%

n = 15, 1%

n = 27, 5%

n = 27, 1%

43.3 36.2 37.0 41.0 35.8 7.0

29.8 25.0 22.6 28.4 24.2 2.8

28.0 22.0 21.4 22.6 24.2 4.6

17.6 12.4 15.0 12.6 15.2 1.6

The data (for different ‘countries’) were generated with a threshold model yt = c0yt–1 + c1(xt | xt–1<0) + c2(xt | xt–1 ≥ 0) where x is a N(0,1) random variable. From the aggregated data a similar threshold model (also including constant terms) is estimated and LR test statistics is computed for the parameter restriction c1 = c2. The sample size is 100 and the number of replications 500. 5% and 1% indicate the size of the LR test.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Estimation and Aggregation Concerns 35

data, with 27 micro units fewer traces of nonlinearity are found in the data. If the data were from a linear model a threshold model could be detected only with (roughly) the same frequency as the LR test size. Basically the same result emerges if we use (2.15) as the testing equation and test the significance of the additional β3stxt-term using the t-test.19 Then, for instance, rejection frequency for a linear model with data that are derived from a linear model (with n = 15) turned out to be 5.2 per cent (2.4 per cent). Alternatively, we may write the fixed threshold switching threshold model in the form yt = c′0yt–1 + c′1xt + c′2(xtxt–1>0) + u′t, and test the hypothesis c2 = 0. Simulation results with these two alternatives were practically identical.20 The computed BDS tests point in the same direction. That is, if the tests are computed for the aggregated time series, the 5 per cent critical value of BDS(2) test statistics is exceeded in roughly 10 per cent of cases (e.g., in 8.8 per cent of the cases with model 1 and n = 27). Thus, it seems that a nonlinear structure that is generated by a threshold model cannot be captured any more after aggregation of micro units. To get some more flavour of the simulation results, a set of Kernel densities of the coefficient estimates of the threshold models are displayed in Figure 2.10 (abbreviation c1-1-15 denotes coefficient c1 with model 1 in Table 2.1 and with 15 cross-section units in aggregation). Similarly, the coefficients from a linear model (model 6 in Table 2.1) are shown on the right-hand side of Figure 2.10. We see that the estimated coefficients c1 and c2 in Figure 2.10 deviate a lot from the values that are used in the data generating mechanism. The general tendency is a convergence towards the average values of the coefficients. Thus, one might assume that if the number of cross-

19

The rejection frequencies with the alternative tests come so close to the LR test that we conclude that the results may not only reflect the particular LR test procedure used compiling Table 2.1. 20 By contrast, the results tuned out to be quite sensitive in terms of correlation of the random terms of the micro units in the same way as in Granger and Lee (1993). Also if we assume that the aggregate data are not a simple sum of the micro units but there is some additional random term (say, measurement error) the test results change a lot, i.e. the rejection frequencies fall. If for instance with model 1 in Table 2.1, the variance of the random term is the same as the variance sum of micro units, the rejection on frequencies fall to one-third of those in Table 2.1 (to 15.6 and 7.0 per cent respectively).

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

36 Asymmetry and Aggregation in the EU

Estimation and Aggregation Concerns 37 Figure 2.10

Distribution of coefficients c1_LIN_15

c1_1_15 10

0.8

Density

Density

8 6

0.6

0.4

4 0.2

2 0 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40

0.0 –2

–1

c2_1_15

0

1

2

3

4

c2_LIN_15

12

0.8

8

Density

Density

10

6

0.6

0.4

4 2

0.0 –2

–1

0

1

2

3

2

3

2

3

c1_1_27 10

c2_lin_27 0.7 0.6 Density

Density

8 6 4

0.5 0.4 0.3

2

0.2 0.1

0 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35

0.0 –2

c2_1_27

–1

10

0

1

c2_lin_27 0.8

6 4

Density

Density

8 0.6 0.4

2 0.2 0 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40

0.0 –2

–1

0

1

Note: On the left-hand side of the figure the graphs are Kernel densities for parameters c1 and c2 for model 1 shown in Table 2.1, the number of micro-units (countries) being either 15 or 27. The right-hand side shows comparable densities for the same two parameters in a linear model, labelled 6 in the table.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

0.2

0 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40

38 Asymmetry and Aggregation in the EU

Table 2.2

Mean values of ˆc1 and ˆc2 from the simulated data

Model 1: c0 = 0.15, c1 = –0.10, c2 = 0.50 2: c0 = 0.15, c1 = –0.10, c2 = 0.10 3: c0 = 0.15, c1 = –0.50, c2 = 0.50 4: c0 = 0.50, c1 = –0.50, c2 = 0.50 5: c0 = 0.50, c1 = –0.10, c2 = 0.50 6: c0 = 0.15, c1 = 1.00, c2 = 1.00

n = 15, ˆc1

n = 15, ˆc2

n = 27, ˆc1

n = 27, ˆc2

0.153 –0.015 –0.089 –0.079 0.154 1.001

0.248 0.017 0.085 0.086 0.249 0.960

0.161 –0.013 –0.064 –0.060 0.164 0.969

0.239 0.011 0.055 0.058 0.237 0.978

21

This seems indeed to be the case. Thus, for instance with model 3, the rejection frequencies of the linear model go down to 9.2 per cent (with the 5 per cent test size) and 5.6 per cent (with the 1 per cent test size) when the number of cross-sections is increased to 100 (notice that in model 3, the comparative values for n = 27 are 21.4. and 15.0). As for the ‘other end’, it seems that effect of aggregation becomes apparent quite quickly when number of micro units is increased from 1 or 2. For instance with simulation model 1 in Table 2.1, the rejection rate with 5 per cent level is 0.90 when n = 3 but when the number of micro units is increased to 5, it falls to 0.63 per cent. 22 The analysis of Granger and Lee (1999) is much more extensive in the sense that they scrutinise a wide range of nonlinear models although some of the nonlinear data generating mechanisms are not very realistic from an economic point of view. Moreover, they just test the generated times series with various nonlinearity tests and do not attempt to discover the underlying nature of nonlinearities.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

sections increases the aggregate data no longer show any signs of nonlinearities.21 Notice that the threshold model performs reasonably well in the sense that it does not find thresholds from the data that has been constructed by a linear model which also show up in the fact that the coefficients c1 and c2 turn out to be practically identical (Figure 2.10). The mean values of cˆ1 and cˆ2 are reported in Table 2.2. Although the effect of aggregation on the series which have been generated by a threshold model seems quite clear we have to acknowledge that this result may not be generalised to other nonlinear models. Thus, for instance, if the data were generated by quadratic polynomials the power of different test procedures with the aggregate data can be quite different. However, the results of Granger and Lee (1999) provide some guidance in this matter. To illustrate the results of Granger and Lee (1999), we might here reproduce some of their main findings for 20 cross-sections and a sample size of 200 (cf. Table 2.3).22 In this example, the data were derived using a

Estimation and Aggregation Concerns 39 Table 2.3

Summary of Granger and Lee findings

Type of test

Neural Network Test Tsay test Dynamic Information Matrix Test Reset Test

No aggregation

Uncorrelated errors

Equal variance

Highly correlated errors

780 52

61 71

187 56

501 68

46 59

46 70

71 58

85 76

simple threshold model of type (2.19). In the case of ‘no aggregation’ the data are thus reflecting this nonlinear data generating process. The Granger and Lee findings are somewhat puzzling. Only in the case of Neural Network tests, is there a clear difference between the original series and the aggregated series (in the case of the original series, the first column just shows the power of the test in detecting the nonlinear structure in the data). The other three tests are not helpful at all: they do not detect nonlinearity in the original series and they do not discriminate between the original and aggregated time series. The performance of tests is somewhat better in terms of other nonlinear alternatives (like the bilinear model) but even then it appears that detecting the (right sort of) nonlinearity is not a settled issue.23 The problem is, of course, the fact that we do not have analytical results of the size of the aggregation effect. Granger (1987) and Granger and Lee (1999) give some examples of the implications of aggregation in the case of nonlinear time-series structures but for obvious reasons, we cannot produce some general results. That is why we also experiment with some alternative nonlinear threshold model structures. In practice, this boils down in using different coefficient vectors for equation (2.19). One has to acknowledge that aggregation does not only affect the interpretation of the aggregate data but also the possibilities of getting information from important micro features of the data. Putting it very simply, if the data for individual countries mainly reflect country 23 Obviously, the results may reflect some specific features of the model and testing procedure. Thus, for instance, Granger and Lee (1999) use very different parameter values for the two regimes.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Numbers are frequencies of rejection at the 5% level out of 1000 replications. The cross-section error-terms are derived as follows: uit = εt + ηit, where the variances of ε and η range from 0 to 1. With uncorrelated errors εt = 0, while with highly correlated errors, Var(ε) = 0.9 and Var(η) = 0.1.

specific features, aggregate data may be quite impotent in reflecting these features. This point is nicely illustrated by Granger (1987) with the following simple model. Assume that the country relationship are of the form: yit = axit + b(xit)2 + εit where xit = µt + x′it where µ is the common factor (mean) for all N countries and x′ the individual country dispersion (with zero mean). For our purposes, we could assume, for instance, that y is inflation and x output gap. To simplify notation, write S(y) = ∑yi and S(x) = Nµ + ∑x′i which reduces to Nµ with large N. Now Granger (1987) shows that if µ ≠ 0, i.e. there is a common factor in the data, S(y) = aS(x) + b[S(x)]2 + S(ε), the original nonlinear form is maintained at the macro level. But by contrast, if µ = 0, so that there is no common factor, S(y) = aS(x) + bS(x2) + S(ε). Notice that now the squared term reflects the variance of the micro (individual country) terms which are not readily observable. Moreover, S(x2) is probably not related so [S(x)]2.24 Thus, the ‘true’ nonlinear relationship cannot be estimated from the aggregate data unless the common factors are dominant enough. Quite clearly, one has to be careful in interpreting the properties of the aggregate EU/euro area series from the viewpoint of individual countries. Aggregate data may simply hide important features of the underling micro series and, moreover, the connection between micro and macro may be quite weak. This fact is illustrated again with the EU27 data; see Figure 2.11. As one can see, cross-country differences in output growth do not nicely coincide with the aggregate output growth, rather the relationship is almost nonexistent. Thus, the aggregate data may not reveal nonlinear features that are incorporated in individual country data. Here we have focused on cross-sectional aggregation only and it is certainly one of the main concerns from the policy point of view. Even so, temporal aggregation also matters (as shown by Granger and Lee (1993)). In practice, we have only annual data on some fiscal variables which effectively hide possible short-term nonlinear features of the data. Needless to say, we can arrive at a situation where both crosssectional and temporal aggregation affect the data at the same time

24

This can also be seen from the data. Thus, with the EU27 data, correlation between the cross-section variance of GDP growth rates and squared GDP growth rate for the whole EU27 turned out to be 0.04 only (N = 52). For details, see Figure 2.11.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

40 Asymmetry and Aggregation in the EU

Estimation and Aggregation Concerns 41 Figure 2.11 Comparison of cross-section variance of GDP growth rates and the squared growth rate of EU27 GDP 24 Cross-section variance squared growth rate 20

16

12

8

4

96

97

98

99

00

01

02

03

04

05

06

07

08

The sample period is 1976.1–2008.4, the number of countries is 25 (for data reasons, Romania and Bulgaria are not included).

giving an impression that relationships are perfectly linear even though the opposite is true. What looks simple on an average may be much more complicated at the grass-root level as the recent fiscal problems in the European Union have shown.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

0

3

In order to explain macroeconomic behaviour in the euro area we use the simple and very conventional four equation model of the economy, consisting of an IS curve, a Phillips curve, and Okun curve and a monetary policy reaction function that we have employed earlier (Mayes and Virén, 2005) and described in outline in the previous chapter.1 This enables us to explore the overall level of activity, inflation, unemployment and monetary policy in an open economy framework. We do not attempt to look at the determination of the exchangerate or the components of the balance of payments, nor do we consider wealth acquisition. Our model is thus incomplete. However, we are not seeking to substitute for Dynamic Stochastic General Equilibrium (DSGE) or other system approaches on this scale, we simply wish to explore the problems of asymmetry and aggregation and the concerns they pose for macroeconomic policy particularly in the EU, where single policies are implemented for member states that are quite heterogeneous. In this chapter we deal just with the IS curve. As set out in Chapter 2, we explore a conventional IS curve of the form: ∇yt = a0 + a1∇yt –1 + a2∇yt–2 + a3rrt–i + a4ret –j + a5∇y*t –k+ εt ,

(3.1)

where ∇y is the deviation of output y from its Hodrick–Prescott filtered trend, rr is the real three-month interest rate (i.e. the nominal

1 A three equation version, omitting the labour market has received considerable attention – Cho and Moreno (2006). However, in our view, including the labour market is essential, as it is one of the core areas of asymmetry in the macroeconomy, behaving in a manner that is clearly different from the asymmetry in aggregate activity.

42

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Aggregate Supply and Demand in an Open Economy

rate of interest r less the annual rate of consumer price inflation p), re the real exchange rate with the US dollar (in logs) and ∇y* the deviation of OECD output from its HP trend (lag lengths i, j and k typically vary from 1 to 3 quarters in estimation). Because the construction of output gaps is a controversial issue, we have also carried out all the empirical analysis using the growth rate of output as well (to be denoted as ∆y). As can be seen from the subsequent graphs, these two measures are far from identical in describing cyclical movements and it is no surprise that the results are somewhat sensitive to the choice of output measure. Fortunately, none of the results is crucially dependent of this choice. The IS curve represents a quite conventional open-economy demand relationship in which the real interest rate and the real exchange rate represent the main ingredients. The open-economy nature of the equation is further re-enforced by the inclusion of foreign export demand variables. In addition, in section 3.5 we include two wealth (shock) variables, the rate of change of real house prices (hp) and the rate of change of real stock prices (sp). Then the estimating equation reads: ∇yt = a0 + a1∇yt–1 + a2∇yt–2 + a3rrt–i + a4ret–j + a5∇y*t–k + a6hpt + a7spt + εt

(3.2)

where lag lengths i and j may be different and may also differ over countries. The basic equation is completely backward-looking but we also estimate the model in a hybrid Euler equation for total output which takes the form: ∇yt = a0 + a1∇yt–1 + a2∇eyt+1 + a3rrt–i + a4ret–j + a5∇y*t–k + a6hpt + a7 spt + εt

(3.3)

where ∇eyt+1 is the expected output gap. While we are persuaded that incorporating forward-looking variables is essential for a clear understanding of economic behaviour, we do not take a stand on this and consider both hybrid and purely backward-looking models as well. Estimation of (3.3) is obviously more complicated because we have to deal with expectations. Then we must either have some direct observations of expectations, say, from survey data, or we have to estimate (3.3) with the joint hypothesis of the process of expectation formation, such as, Rational Expectations, which necessitates

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Aggregate Supply and Demand in an Open Economy 43

44 Asymmetry and Aggregation in the EU

the use of the Generalised Method of Moments (GMM). Equation (3.2) is useful also in the sense that it allows us to compute what is called a Financial Conditions Index (FCI), as explained in Chapter 8. It is simply the weighted sum of the effects from each of the financial variables in the equation can be derived from (3.2) in the following way: (3.4)

where ^ denotes estimated values. In estimating (3.1), or its different versions, the main emphasis is estimation of the coefficients of the policy variables, rr, or more generally rr, re, hp and sp. The IS curve itself is not of itself our main interest. Thus, our main concern is not so much to test different versions of the IS curve against each other as to see whether the role of the policy variables is robust over different specifications. Obviously, the interpretation of the old-fashioned backward-looking IS curve and the Euler equation are quite different. Thus, with the Euler equation, the coefficient of the interest rate reflects intertemporal substitution rather than the impulse response of the policy rate. In the context of the Euler equation, the interpretation of the real exchange rate and the asset prices is also a bit tedious e.g. because we cannot really distinguish anticipated and unanticipated parts of these variables/shocks.

3.1

Monetary conditions

Our primary concern in this chapter is more with aggregation than asymmetry. As we have just noted, there is a lot of variation in the openness of the EU countries not just generally but to each other and to the euro area. Monetary policy in the euro area is set with regard to future inflation in the area as a whole. While the impact of monetary policy on the real exchange rate and the impact of the real exchange rate on inflation at the aggregate level will be taken into account, the impact on the individual countries will be very different. In Mayes and Virén (2000b) we explored this in some detail using data from before the start of the euro area. Our primary concern was to consider the relative importance of the influence of the real exchange rate and the real interest rate in affecting output, i.e. â3/â4 = λ in terms of equations (3.1) to (3.4). Thus for example if λ = 2, a one percentage point increase in the real

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

FCIt = (1/â3)(â3rrt–i + â4ret–j + â6hpt + â7spt)

interest rate has the same impact as a 2 per cent increase in the real exchange rate. The weighted pressure of both the real interest rate and the real exchange rate on aggregate demand – weighted by their coefficients as in the FCI shown in (3.4) – is often referred to as a Monetary Conditions Index (MCI). Such indexes are no longer in fashion, as it has become much easier to talk about the range of factors affecting inflation in a coherent manner using a model driven framework over the last decade and such simplified measures are no longer required.2 We obtained estimates from a variety of sources including our own data set (reproduced here as Table 3.1) which suggested that most estimates lay in the range 2–6 – our estimate for the euro area as a whole was 3.5 in the light of the evidence. This is considerably smaller than the run of estimates for the US and Japan, which were of the order of 8–10, yet the openness to trade, as conventionally measured, was much more similar to the euro area. The estimates were quite sensitive and varied considerably both for each individual country and across countries. The estimates in column [4] used data from 1972Q2–1997Q4 but we felt that this was bound to cover a regime change as it was not until 1987 that the Exchange Rate Mechanism (ERM) of the European Monetary System (EMS) became sufficiently robust that it could continue for five years without realignment. One might therefore only regard the period after 1987 as belonging to a single regime.3 The results for this shorter period are shown in column [5]. (Columns [6] to [9] show other specifications, respectively adding a price variable, using the growth in GDP rather than the output gap, SUR (Seemingly Unrelated

2 In the second half of the 1990s both the Bank of Canada and the Reserve Bank of New Zealand used MCIs to help explain the bite that the ‘monetary’ variables policy can affect in the short run are likely to have on inflation. However, the exchange rate in particular is affected by a wide range of factors, in addition to policy and changes in it indicate that the central bank might want to reappraise its forecast and the setting of interest rates. Since central banks only change interest rates at predetermined meeting dates except in emergency, such an indicator was thought helpful to markets in their efforts to anticipate policy moves. There are many pitfalls in using such indices (Eika et al., 1996) particularly since it is only nominal values that are available continuously, deflators only being computed at monthly or even quarterly intervals. 3 We searched empirically for structural breaks in the data; while there was some evidence for an earlier break in 1985 and a problem with the breakdown of the ERM in 1992, 1987 appeared the most satisfactory date.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Aggregate Supply and Demand in an Open Economy 45

Regression) estimates instead of Ordinary Least Squares (OLS) and adding a long interest rate (bond rate) as suggested by Kennedy and van Riet (1995) among others. The individual country estimates are relatively poorly determined, see Table 3.2 for the case of the shorter data period, but there is little evidence that the exchange rate was anything other than relatively important. Country differences in the main did not appear to be affected by size or openness. The pooled estimates using the whole of the data in a single regression – on the assumption that the coefficients were the same across all the countries were much better determined, with λ in the range 2 to 3. The λ ratios lie mainly in the range 1 to 8 with the Belgian and Portuguese results being implausibly high. Calculating the weighted average value (GDP weights) for the euro area gives a value of approximately 3.5 which is towards the more open end of the spectrum and more than twice as open as using the ratio of trade to GDP would lead one to expect. We therefore conclude that, even in the euro area, movements in the real exchange rate are an important means of adjustment. The drawback is that neither the timing nor distribution of that impact is expected to be very even. The impact tends to be relatively rapid. Mayes and Virén (2000b) estimate from impulse responses that the bulk of the impact is completed within two years. However the size range of the impact across the member states is roughly a factor of 2. This clearly provides somewhat of a problem for a single monetary policy, especially if the parts of the economy most in need of relative stimulation are those that are least responsive. Nevertheless, there is a tendency to forget that inside national economies both sectors and regions are also affected very differently by changes in both interest and exchange rates and so we address this below. We now, of course have not only the benefit of another 11 years of data but can also see whether the formation of the euro area seems to have had any impact on the relative importance of the two variables. However, other factors have also affected the structure of the international system so we need to adopt a difference in differences approach if we are to have any hope of isolating the impact of the euro area. The results for the IS curve in Tables 3.3 and 3.4 using the whole dataset show quite clearly that our simple model with world output, the real interest rate and the real exchange rate can explain the movements of output over time quite well. As in most of our analysis, the dataset is formed of quarterly data for the EU15 less Luxembourg but plus Norway for

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

46 Asymmetry and Aggregation in the EU

2.9

1.5 8.1

Italy

2.3(3), 4.3(6) 2.3(1), 4(2), 2.3(3), 4.3(5), 2.7(6) 10(2), 4(3), 8.8(5), 7.9(6) 2(1) 3(1), 1.4(6) 6.4(4), 1.7(6) 10(2), 9(3), 39(5), 10.1(6)

1.5(3), 2.5(4), 4.2(6) 3.4(1), 1.5(3), 0.5(4), 2.1(6) 3(2), 4(3), 14.4(4), 5(5), 2.9(6)

3.7(4), 0.8(6)

3(2), 4(3), 6.6(4), 6(5), 4.1(6)

2.5, 4(2), 4(3), 2.6(4), 4.2(5), 2.3(6)

3.3(4) 0.4(6) 1.9(6) 2.5(6) 3(2), 4(3), 3.4(4), 2.1(5), 3.5(6)

[3]

Other3

12.8 –4.6 –2.1 5.2 0.8

17.7

29.7

14.3 60.1 3.6 8.9 19.2

[4]

long

short

2.0

2.3 11.6 0.8 1.2 1.5

7.8

3.6

2.4 60.6 8.3 3.1 2.5

[5]

sample

1.5

2.3 5.6 0.8 1.3 0.3

8.4

2.5

3.1 89.9 9.6 3.2 1.9

[6]

short + prices

1.9

2.8 2.3 0.6 0.7 1.5

2.4

4.9

1.3 6.4 –13.3 3.3 2.5

[7]

short ∆GDP

1.1

1.2 14.8 0.5 1.2 1.1

4.3

3.4

1.2 6.2 14.4 3.4 3.4

[8]

short SURE

–1.7

3.4 3.6 3.7 0.8 0.9

13.1

2.7

14.1 2.2 –4.1 4.3 2.1

[9]

short + bonds

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

2

Drawn from Dornbusch et al. (1998) Table 10. Drawn from Banque de France (1996) Table 1. Numbers in parenthesis are drawn from Peeters (1998) Table 5. Peeters also provides estimates from EUROMON: Belgium, 6.7; France, 3.5; Germany, 9.0; Italy, 5.7; Netherlands, 8.1 and UK, 3.0. 3 Drawn from Ericsson et al. (1997) Table 1. Numbers in parenthesis denote sources of estimates as follows: (1) central banks, (2) IMF, (3) OECD, (4) Deutsche Bank, (5) Goldman Sachs and (6) JP Morgan.

1

Ireland Australia Canada Japan New Zealand Norway Switzerland US

6.2 (4.6)

1.3

1.4

Germany

Netherlands Portugal Spain Sweden UK

3.0 (2.0) 4.0 (6.1) 0.1 (1.8) (3.0)

2.1

[2]

[1] (1.5)

NIGEM2

Other estimates Dornb1

Our estimates

Estimates of the ratio of the real interest rate to the real exchange rate effect from Mayes and Virén (2000)

Austria Belgium Denmark Finland France

Table 3.1

47

48 Asymmetry and Aggregation in the EU OLS estimation results for the 1987:1–1997:4 period

Name lags

∇yt–1

oecdt–k

R2 (SEE) DW

λ

Austria 2,2,2

0.729 –0.095 (6.39) (0.74)

–0.021 (0.91)

0.009 (0.65)

0.338 (0.65)

0.660 1.90 (0.58)

2.4

Belgium 4,3,2

1.145 –0.457 (6.60) (2.85)

–0.046 (1.28)

0.001 (0.08)

0.334 (1.55)

0.882 1.81 41.7 (0.40)

Denmark 1,3,1

0.105 (1.02)

–0.152 (1.77)

0.018 (1.51)

0.065 (1.12)

0.261 1.96 (0.84)

8.3

Finland 3,2,2

0.773 –0.158 (5.36) (1.00)

–0.152 (2.36)

0.048 (3.90)

0.406 (0.83)

0.881 1.99 (1.19)

3.1

France 4,2,2

0.960 –0.274 (6.64) (1.56)

–0.069 (2.06)

0.027 (1.99)

0.305 (1.35)

0.871 1.94 (0.44)

2.5

Germany 7.3.2

0.545 (2.73)

0.181 (1.23)

–0.072 (0.87)

0.020 (0.84)

0.123 (0.57)

0.911 1.53 (0.81)

3.6

Ireland 1,3,2

0.970 –0.298 (7.18) (2.63)

–0.056 (1.73)

0.028 (1.56)

0.788 (3.09)

0.867 1.81 (0.72)

2.0

Italy 3,2,1

0.701 (8.88)

–0.095 (1.90)

0.012 (1.43)

0.332 (2.35)

0.767 1.97 (0.50)

7.8

Netherlands 1.077 –0.381 1,2,2 (11.41) (4.72)

–0.037 (1.53)

0.016 (1.76)

0.259 (2.34)

0.824 1.86 (0.39)

2.3

Portugal 3,1,1

0.135 (1.30)

–0.081 (2.87)

0.007 (1.24)

0.747 (3.74)

0.901 1.95 11.6 (0.48)

1.518 –0.595 (15.98) (6.53)

–0.008 (1.16)

0.009 (3.01)

0.115 (1.85)

0.982 1.44 (0.18)

0.8

0.226 (1.70)

–0.065 (5.21)

0.052 (4.05)

0.604 (2.30)

0.809 2.33 (0.77)

1.2

0.981 –0.175 (10.50) (1.86)

–0.033 (1.84)

0.022 (2.96)

0.262 (4.34)

0.950 1.83 (0.40)

1.5

Spain 1,2,1 Sweden 1,2,2 UK 1,1,1

0.447 (3.47)

0.537 (5.21)

∇yt–2

rrt–i

ret–j

The dependent variable ∇y is the output gap constructed by the HP filter. rr is the real interest rate, re the real exchange rate with respect to US dollar. oecd denotes the output gap for OECD GDP. λ is the ratio between interest rate and exchange rate elasticities. The numbers below the country names give the lag length for rr, re, oecd (in this order), respectively. The data are quarterly and cover the period 1987:1–1997:4. For the UK, the US output gap is used instead of the OECD output gap. The German equation includes a level and one period dummies for the unification period (1991:1–1997:4). All estimates are OLS estimates t statistics in parenthesis.

the period from the beginning of 1971 through to the end of 2008. However, there are some changes in the performance of this basic equation over time, as can be seen if we compare estimates for different sub-periods.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 3.2

Table 3.3

IS curves (basic specification)

Data

re/100

rs/100

World output

lag

1971–2008 0.002 GAP (1.51)

–0.018 (3.06)

0.388 (12.65)

0.697 (34.61)

1971–2008 0.043 g (4.09)

–0.027 (2.83)

0.354 (11.47)

0.705 (32.05)

1971–2008 0.029 g (3.43)

–0.019 (2.12)

0.345 (13.85)

1999–2008 0.064 g (2.89)

–0.077 (1.49)

1971–2008 0.041 g (3.92)

R2/SEE

DW Method

0.720 0.0079

1.98

OLS

0.047 (2.46)

0.737 0.0122

2.02

OLS

0.719 (40.09)

0.034 (2.29)

0.737 0.122

2.05

GLS

0.680 (8.61)

0.494 (7.21)

0.058 (0.32)

0.741 0.0105

2.07

OLS

–0.027 (2.80)

0.369 (12.67)

0.717 (35.42)

0.750 0.0123

1.93

OLS

1999–2008 0.056 g (3.90)

–0.084 (1.53)

0.734 (90.71)

0.483 (7.41)

0.736 0.0105

2.09

OLS

1971–2008 0.010 g1 (0.13)

–0.002 (0.36)

0.489 (6.87)

0.734 (90.71)

0.463 0.074

1.96

SUR

1971–2008 0.048 g2 (1.33)

–0.088 (2.39)

0.489 (6.87)

0.734 (90.71)

0.651 0.035

1.71

SUR

1971–2008 0.0054 g3 (1.10)

–0.255 (5.05)

0.489 (6.87)

0.734 (90.71)

0.707 0.047

1.70

SUR

1971–2008 0.070 g4 (2.30)

–0.080 (2.59)

0.489 (6.87)

0.734 (90.71)

0.719 0.029

1.45

SUR

1971–2008 0.029 g5 (0.80)

–0.186 (5.02)

0.489 (6.87)

0.734 (90.71)

0.647 0.035

1.72

SUR

1971–2008 0.024 g6 (0.84)

–0.080 (2.67)

0.489 (6.87)

0.734 (90.71)

0.736 0.028

1.29

SUR

rl-rs

GAP indicates that the dependent variable was the output gap while g denotes that it was the rate of output growth. The last six equations represent sectoral output growth. g1 denotes agriculture and fishery, g2 manufacturing, g3 construction, g4 trade, g5 financial and real estate services and g6 production of public services. The standard error of estimate (SEE) has been multiplied by 100 for the equations using aggregate economy data. The sectoral equations have been estimated using the restriction that the coefficients for lagged output growth and world output growth are equal. re is the real exchange rate, rs the real short rate of interest and rl the real long rate.

The interesting result from the point of view of the previous discussion is that λ for the whole period or recent years has risen noticeably to values close to the external trade share.

3.2

The advent of the euro area

In general, the equation performs better for the most recent (post1998) sample than for the whole sample; at least if we focus on the role

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Aggregate Supply and Demand in an Open Economy 49

50 Asymmetry and Aggregation in the EU

data

Estimation of an extended IS curve lag

world

rr

re

hp

sp

R2/SEE

DW/J

1971–08, ∆y

0.679 0.374 (27.54) (11.38)

–0.006 (0.55)

0.018 (4.08)

0.685 0.138

1.99

OLS

1987–08, ∆y

0.678 (22.51)

0.381 (8.20)

–0.038 (3.20)

0.030 (5.35)

0.739 0.114

2.19

OLS

1987–08, ∆y

0.711 0.409 (39.08) (13.51)

–0.028 (3.29)

0.025 (6.80)

0.737 0.113

2.25

GLS

1979–08, ∇y

0.669 0.416 (23.86) (10.56)

–0.003 (0.35)

0.001 0.024 0.001 (0.37) (7.85) (0.67)

0.715 0.068

2.23

OLS

1979–08, ∇y

0.243 (9.56)

0.747 (6.91)

–0.047 (1.65)

0.036 0.047 –0.003 (1.90) (9.63) (4.85)

.. 0.0076 7.31

GMM

1979–08, ∆y

0.614 (25.06)

0.324 (7.77)

–0.007 (0.60)

0.032 0.044 0.007 (6.73) (8.87) (4.75)

0.798 2.14 0.0097

OLS

1979–08, ∆y

0.410 (16.31)

0.185 (0.88)

–0.040 (3.92)

0.149 0.077 –0.008 (1.49) (7.42) (0.99)

.. 0.0124 8.93

GMM

1979–08, ∆y

0.565 (13.61)

0.231 (3.57)

–0.008 (0.44)

0.030 0.056 0.010 (4.68) (7.77) (3.80)

0.758 1.77 0.0101

OLS gap>0

1979–08, ∆y

0.605 (18.09)

0.412 (7.47)

–0.005 (0.33)

0.041 0.036 0.007 (5.56) (4.74) (3.94)

0.828 1.68 0.0088

OLS gap<0

1979–08, ∆y

0.544 (16.46)

0.370 (6.95)

–0.004 (0.32)

0.026 0.040 0.005 (4.22) (5.38) (2.50)

0.763 2.12 0.0096

OLS hp>0

1979–08, ∆y

0.643 (17.89)

0.306 (4.52)

–0.001 (0.04)

0.045 0.072 0.009 (5.42) (4.40) (4.13)

0.804 1.85 0.0092

OLS hp<0

Variables are defined as before with hp and sp denoting house prices and stock prices respectively.

of real interest rate and real exchange rate. There can be many reasons for this difference. The real exchange rate is of course deficient, because it only takes the US exchange rate into account.4 The degree to which member states have transactions with the non-US dollar foreign exchange environment varies and in some cases is quite substantial, since for the non-euro countries that includes the euro area. Even within the euro area, the real exchange rate (i.e. relative prices) can vary, which will 4

Other currencies, particularly sterling play an important role in some countries so focusing purely on the US dollar may be misleading. In Mayes and Virén (1998) we show that in the case of Finland, where both sterling and the Swedish krona account for significant proportions of trade, using a trade-weighted index does alter the numerical value of the coefficient noticeably. However, the qualitative impact, which is the focus of our discussion here, is small. The Irish Republic is the only country where the dollar is clearly not the most important external currency. Similarly over this period, although most of the countries were participating in the Exchange Rate Mechanism of the European Monetary System, exchange rates with respect to each other also changed, particularly around 1992. We show that adding the Deutsche Mark (DM) exchange rate adds very little to the overall explanation but results in poorly determined coefficients and perverse effects in four cases.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 3.4

still affect aggregate output. As for interest rates, omitted variables, such as, financial market liberalisation, financial innovations and the changing role of bank credit could have affected the relevant impact multipliers. Nevertheless the principal different between the two time periods is that in the euro area era both real interest rates and real exchange rates appear to have a much more substantial effect on output. In some respects this supports other findings (see Chapter 8) which suggest that as a result of the ‘great moderation’ and their increased credibility, central banks have been able to have their desired impact on the economy with only very limited changes in their instruments. However, neither variable is a direct instrument, certainly not the exchange rate, so a wider explanation is required – perhaps related to increasing openness and to the increasing depth of financial markets, which improves the effectiveness of the transmission mechanism.5 As might be expected, the impact of foreign GDP on euro area output has increased as time has passed. Although shares of intra-euro area trade in total trade also increased during the period, the increase in overall trade is considerably greater. The same result is reflected in the real exchange rate effect, whose value has risen as the economies became more open, especially since the 1980s. As with the earlier data, we have also explored whether longer interest rates have an additional role to play. This is again the case, although here we have set this up in the form of the term structure. Long rates are generally not so directly affected by monetary policy. The picture for the individual countries is, however, now somewhat different in the euro area period (Figure 3.1). Finland, Ireland and Sweden stand out as having rather stronger interest rate effects. It is tempting simply to ascribe this to the faster rates of growth in these countries. Finland and Ireland experienced the same nominal rates of interest as members of the euro area but with higher growth rates than many of their partners. However, we are discussing real interest rates and here the two countries move in different directions. Finland in general experienced a little less inflation than the euro area as a whole, while Ireland was one of the most rapidly inflating countries.

5 We are of course only looking here at the impact on output. We require here at least the joint impact through the Phillips curve if we are to see the consequences for inflation. Since that impact is asymmetric it is more difficult to generalise (see Chapter 4).

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Aggregate Supply and Demand in an Open Economy 51

52 Asymmetry and Aggregation in the EU Figure 3.1

Estimated country-specific interest rate effects from an IS curve

0.1 0.05 0 –0.05 –0.1 –0.15 –0.2

The estimation sample is 1976–2008

The UK stands out in the opposite direction as having a perverse coefficient – perhaps a result of the movements in its exchange rate relative to the euro. While the results are sensitive to the period chosen we can certainly conclude that there is variation across the different countries. The F statistic for the hypothesis that the interest rate coefficient is the same for all countries is F(13,1621) = 2.27 (p = 0.0038). However the impact also varies according to industry. Construction is far more affected by interest rates than most industries while agriculture is least affected (Figure 3.2). We would see similar high impacts for investment goods industries and other heavily pro-cyclical sectors. There is somewhat less diversity in the case of the exchange rate (Figure 3.3) as these days most sectors are exposed to foreign prices even if, as in the case of the construction sector, the impact tends to be more on inputs than on output directly. Once again agriculture is only relatively lightly affected, not because it is not traded but because farmers have to absorb the result of the variations in prices. Most agricultural products have to be sold fresh and the decisions over how much to grow or raise have been made some time beforehand and variation in output is more difficult. We need to be somewhat cautious in interpreting these findings as the data period is short and the variables of interest have had a very specific pattern over the decade of monetary union (Figure 3.4).

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

–0.25

Aggregate Supply and Demand in an Open Economy 53 Figure 3.2

Sectoral coefficients of the real interest rate

0 –0.05 –0.1 rr: ols

–0.15

rr: gls –0.2

rr: sur

–0.25 –0.3

rr is the real interest; ols, gls and sur denote the estimation technique, respectively, ordinary least squares, generalised least squares and the seemingly unrelated regression technique.

Figure 3.3

Coefficients of the real exchange rate

0.016 0.014 0.012 0.01 0.008 fx: ols

0.006 0.004

fx: gls

0.002

fx: sur

0 re

r tu

m

u an

fa

n

io

c

ri

ag

g

in

tu

l cu

r

t uc

st

n co

s ce

de

tra

e

i rv

.s

fin

ic bl

s.

pu

fx is the real dollar exchange rate; ols, gls and sur denote the estimation technique, respectively, ordinary least squares, generalised least squares and the seemingly unrelated regression technique.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

–0.35

54 Asymmetry and Aggregation in the EU

The period has been characterised by a clear economic cycle. Whether unemployment or output gaps are used as an indicator, an initial period of growth was followed by a substantial downturn – a growth recession rather than an actual recession – which was relatively proFigure 3.4

Median of key macro variables before and after the EMU Output growth

Unemployment

8

10 9

6

8 7

4

6 5

2

4 3

0

2 1 1975 1980 1985 1990 1995 2000 2005

1975 1980 1985 1990 1995 2000 2005 Short and long interest rates

Inflation 14

18 16

12

14 10

12

8

10

6

8 6

4

4 2

2

0

0 1975 1980 1985 1990 1995 2000 2005

1975 1980 1985 1990 1995 2000 2005

Median of real interest rates

Median of log real exchange rates

10 0.6

8 6

0.4

4 0.2

2 0

0

–2 –.0.2

–4 –6 1975 1980 1985 1990 1995 2000 2005

1975 1980 1985 1990 1995 2000 2005

Interest rates: ---- long __ short

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

–2

tracted. Only more recently has growth reached what were traditionally average values. Beyond the data period this short-lived period of growth has not merely been arrested but the euro economy has gone into recession with sharp declines in many member states in 2009, although most seem to be emerging again on the basis of preliminary information in 2010.6 Over the period as a whole inflation has remained low and stable (and hence more predictable), a sharp contrast to the prolonged and steady falls in the years before Stage 3 of EMU started. The only exception was the rapid rise on the back of commodity price increases at the end of the data period in 2008. However, this has been rapidly rectified by the recession. Real interest rates continued to decline, reaching negative territory before the recent increases. As we discuss below, a major feature of this reduction in the median is a reduction in the skew. Some countries had low and stable inflation throughout. What characterises the 1990s is the decline in inflation in the more inflationary countries to the levels prevailing in the least inflationary. This of course was precisely what the Maastricht convergence criteria required: convergence to within 1.5 per cent of the average of the three lowest inflation rates among the EU member states7 (Art. 121(1) of the Treaty on European Union). However, other variables have remained quite volatile. The exchange rate for example depreciated by 20 per cent before reversing its loss entirely and appreciating to around 20 per cent above its starting value, a total switch of 40 per cent. The real fluctuations have been rather more limited. Fiscal policy has also shown a clear change. While the prolonged consolidation that preceded the start of Stage 3 continued in the early years it was reversed hand-in-hand with the economic downturn (this is discussed in more detail in Chapter 9, see Figure 9.2). However, it did not backtrack very far and improved again towards the end of the period before the present crisis called several member states’ strategies into question. It is important also to recall that the experience of the United States has been similar over the period (Figure 3.5).

6 Pessimists are raising the spectre of a double dip recession but it is too early to judge. 7 The exact words are ‘best-performing’ and subsequent practice has seen states with deflation excluded from the calculation.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Aggregate Supply and Demand in an Open Economy 55

56 Asymmetry and Aggregation in the EU Figure 3.5 Similarities in (a) growth and (b) inflation performance in the euro area and the US 5

7

GDP growth

4

Inflation

6

3

EU

5

2 4 1 3

EU

US

0

US

2

–1 –2

1 88

90

92

94

96

98

00

02

04

06

88

90

92

94

96

98

00

02

04

06

3.3 The role of asset prices in the asymmetry of the euro economy and monetary policy It has long been accepted that asset prices and house prices in particular, have an important role to play in fluctuations in the economy. Moreover, the present crisis has heavily reinforced the importance of understanding this relationship. Altissimo et al. (2005) provide a helpful survey and conclude that with some small exceptions for investment in residential property the effect comes almost entirely through consumption.8 However, the present crisis has dramatically increased the focus on linkages through the financial system. The importance of the debt-deflation spiral (King, 1994) has always been well known but in recent years it has been only the Japanese experience which has exposed its reality. In the Nordic crises, for example, although the impact on GDP in Finland was deeper than in the 1929 depression, the countries recovered quickly and state intervention brought a halt to the downward spiral and drew a line under the losses in the banking system so that it could recover.9

8

They play down the credit channel, discussed in Bernanke and Gertler (1995). While asset management companies may have been still realising impaired assets a decade later, the need for precipitous sales ended. While house prices fell by about half from their peak, they began to recover by 1993. Even at their lowest they were 10 per cent above their 1983 values in real terms.

9

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Hence there is a strong incentive not to argue post hoc propter hoc and to suggest that some of the changes may be simply due to the particular pattern of the data.

Aggregate Supply and Demand in an Open Economy 57

Asset prices are also clearly related to the inflationary process, both as part of the transmission mechanism of monetary policy and as indicators of future inflationary pressure (Goodhart and Hofmann, 2000). It is also clear that their role in the process is asymmetric over the course of the economic cycle. This asymmetry is expected to be different for stock market prices and house prices, the two most widely available asset prices. House prices also clearly influence consumers’ expenditure, as housing provides the least cost route for consumers to obtain loans, through a mortgage on the property, thus enabling them to consume out of their wealth.10 Stock prices affect a limited number of households directly but business activity more directly. The impact of asset prices on economic growth

In this section we expand the analysis by using equations (3.2) and (3.3) which incorporate both stock prices and house prices.11 For details of the data, see Appendix 3.1. Real interest rates and real exchange rates enter the estimating equations with a lag (Table 3.4). Thus, there is no obvious simultaneity problem with them. In the case of house and stock prices such a problem can exist. It is only that the data do not strongly support the notion that the exogeneity assumption is violated.12 It is obvious, even before we start, that house prices and stock prices are likely to play different roles as they show little correlation (Figure 3.6). If we set this out in the time dimension, using medians, the difference in

10

Mayes (1979) suggests that the asymmetry in the house price cycle in the UK stemmed from a complex interaction of the constraints on production, prudential constraints on housing finance and a strong upward dynamic in the housing market. 11 The functional form is dictated by the fact that the level form data are nonstationary while the transformed variables in (3.3) are, in general, stationary (see Appendix 3.1 for the panel unit root tests). Thus, the hypothesis of unit root can be rejected in all cases except for real exchange rates. With (the change rate of) real house prices the hypothesis can be rejected with individual unit roots but not in the case of a common unit root assumption. For theoretical reasons, it is difficult to take the real exchange rate result very literally and, therefore, we do not take differences of re. 12 Computing the Hausman-Wu test statistic for hp and sp gives the value 2.73 (0.067) which suggests that the violation of the exogeneity assumption is not very severe. A similar result is obtained if hp and sp are lagged by one period. Then the estimates and the explanatory power remain practically unchanged. We also computed the differencing test statistics for all equations. They showed some problems with the IS curve that includes both house and stock prices. That could be explained by the (in)stability properties that are illustrated in Table 3.8.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

3.3.1

58 Asymmetry and Aggregation in the EU Figure 3.6

Scatter plot between change rates of house and stock prices 150

100*DLOG(SP, 0, 4)-INF

100

50

0

–100 –40

–20

0

20

40

HP-INF

pattern is clear (Figure 3.7) – house price data are only available from the beginning of 1979 in our sample. Stock prices are much more volatile and show some peaks and troughs not reflected in house price data, the most obvious of which is the fall associated with the collapse of the dotcom boom in 2000. However, it is not necessarily stock prices that have the greater signalling power. The recent downturn was presaged much earlier by house prices than stock prices. This is unrelated to the fact that the problems emerged first in the sub-prime housing finance market in the US, simply that house prices were more readily affected by the tightening monetary policy in Europe in the face of rising CPI inflation. It is immediately apparent (from Table 3.4) that both house prices and stock prices have a clear impact on output growth, with the effect being stronger in the case of house prices. The results are robust to differing lag lengths (Table 3.5) and the other coefficients have plausible signs and size. A one percentage point increase in interest rates has a similar effect on output growth to a 2–3 percentage point change in the real exchange rate. This is a slightly stronger exchange rate effect than we found using a shorter data period (Mayes and Virén, 2002a).

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

–50

Aggregate Supply and Demand in an Open Economy 59 The growth rate of (real) house and stock prices (medians)

20

50

16

40

12

30

8

20

4

10

0

0

–4

–10

–8

–20

–12

–30

< -- real house prices real stock prices -->

–16

–40 1980

1985

1990

1995

2000

2005

Note that the data period has been extended slightly to include more recent observations.

If, however, we difference the model to enable us to use the ArellanoBond GMM panel estimator, the results become a little less satisfactory (see the last column in Table 3.5). Both the interest rate and stock price terms become insignificant. It is not allowing for the simultaneous relationships through GMM which creates the problem. Indeed the GMM results are more plausible than their least squares counterparts. Our estimation period has been chosen by the maximum length of the data series available, rather than by any clear choice based on the existence of a single regime. Extending the model back to 1970 (while omitting the asset price terms) gives some problems with the exchange rate effect (Table 3.6, column 2), as does omitting the fixed effects (columns 3 and 4). Restricting the sample just to the euro area period (column 6) suggests that the interest rate has become less important. This is usual for a very credible regime (Blinder and Solow, 1973). With inflation rates approximately on target throughout the estimation period it is not really surprising if inflation has been relatively unimportant. Similarly it is not surprising to see that the stock price effect looks weak, since there was a substantial fall and recovery in most stock markets in that period, without any substantial effect on

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Figure 3.7

60 Asymmetry and Aggregation in the EU Basic IS curve specification with different lags 1

2

3

0.871 (19.92)

0.820 (17.77)

0.835 (17.83)

re

0.023 (8.07)

0.027 (8.90)

rr

–0.055 (2.91)

hp sp

∆yw

5

6

7

0.320 (7.01)

0.309 (8.35)

0.756 (9.57)

0.418 (3.04)

0.026 (8.69)

0.010 (4.85)

0.006 (3.69)

0.021 (2.96)

0.016 (3.32)

–0.074 (4.30)

–0.055 (3.37)

–0.032 (2.69)

–0.021 (2.13)

–0.028 (1.00)

–0.005 (0.09)

0.100 (12.97)

0.094 (12.67)

0.096 (12.75)

0.035 (6.13)

0.023 (5.31)

0.081 (5.54)

0.068 (3.39)

0.006 (3.02)

0.009 (4.53)

0.008 (4.96)

0.008 (5.41)

0.007 (6.24)

–0.002 (0.62)

–0.008 (0.85)

0.630 0.679 (21.43) (33.72)

–0.248 (5.43)

0.360 (7.21)

y–1 R2 SEE DW Estimator Panel Lags

0.623 0.0138 0.638 LS CFE 0,0

0.632 0.0136 0.644 LS CFE 2,4

0.629 0.0137 0.643 LS CFE 2,2

4

0.802 0.0100 2.127 LS CFE 2,2

0.800 0.0099 2.223 GLS CFE 2,2

0.191 0.1039 2.095 LS Dif 2,2

.. 0.0125 .. GMM Dif 2,2

The dependent variable is the growth rate of GDP, denoted by ∆y. Number of observations is 1037 (with first differences, the number is 1022). Numbers in parentheses are corrected t-ratios. Lags denote the fixed lags of re and rr, respectively. CFE denotes the inclusion of fixed effects, Dif indicates that the data are differenced, LS denotes ordinary least squares and GLS, generalised least squares, while GMM denotes Generalised Method of Moments (Arellano-Bond) estimator. Then the J-statistic has the value of 9.28 that is far from significant with the instrument rank of 15. If one tests the presence of fixed effects one can typically reject the hypothesis that these effects are identically equal to zero. Thus, e.g. in the case of equation (4) above the value of the F-test statistic is 7.80 which is significant at all conventional levels.

output.13 In part this reflects the offsetting monetary policy. However, to some extent this can be circumvented by including policy in the model as we go on to do and in part the endogeneity will be accounted for in the GMM estimates. A glance at Figure 3.8, suggests that the results obtained from using the output gap instead of output growth will be fairly similar as the two series 13

Shortening the estimation period to just eight years so that we incorporate only one business cycle is likely to lead to data specific problems. Even with the 28 years for our main estimation the period is somewhat shorter than might be ideal for purely statistical purposes but extending the data period also increases the chance of encompassing a regime change.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 3.5

Aggregate Supply and Demand in an Open Economy 61 Comparison of different IS curve specifications 1

2

3

4

5

6

∆y

0.397 (8.93)

0.367 (10.48)

0.224 (7.63)

0.217 (8.12)

0.794 (9.29)

0.600 (6.33)

re

0.009 (4.00)

–0.001 (0.47)

0.001 (0.31)

0.001 (0.27)

0.021 (3.37)

0.017 (3.45)

rr

–0.059 (4.89)

–0.016 (1.13)

–0.023 (2.00)

–0.021 (1.93)

–0.043 (1.70)

–0.041 (1.38)

hp

0.028 (5.10)

0.018 (3.74)

0.060 (3.37)

0.044 (3.70)

sp

0.008 (5.18)

0.007 (4.36)

–0.002 (0.67)

0.004 (1.00)

0.759 (28.03)

0.760 (34.82)

–0.085 (1.83)

0.447 (7.34)

∆y–1

0.683 (24.05)

R2 SEE DW Estimator Panel Lags N

0.787 0.0104 2.107 LS CFE 2,2 1037

0.692 (20.36) 0.695 0.0139 1.927 LS CFE 2,2 1682

0.781 0.0105 2.219 LS None 2,2 1037

0.549 0.0105 .. LAD None 2,2 1037

0.096 0.0106 .. LAD None, dif 2,2 1037

0.791 0.0089 2.063 LS CFE 2,2 449

Variables and other labels defined as in Table 3.5. The dependent variable is ∆y. LAD denotes the least absolute deviations estimator. None denotes that no fixed or random effects are included, dif that the data (all variables) are differenced. If house and stock prices are not included, the sample size would increase considerably (i.e. from 1037 to 1682). Equation 6 is estimated from the sample of the EMU period 1999Q1–2006Q4.

have been moving quite closely together. However, this is not quite the case (Table 3.7). The stock market coefficient has a tendency to show a perverse sign and is significantly so at the 5 per cent level in the last two columns. The results are conventional if we take just the period of the euro area’s existence (columns 4 and 5). Nevertheless, whichever specifications we look at it is very difficult to suggest that housing prices are not clearly related to the growth rate and the run of results suggests that stock prices also are likely to have an effect, albeit clearly weaker. 3.3.2

The effect of asymmetry

Thus far all our results consider a symmetric approach, assuming that it does not matter whether the economy is in the expansionary or contractionary phases of the growth cycle. Both economic theory stretching back to Keynes (1936) and beyond and previous empirical results

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 3.6

62 Asymmetry and Aggregation in the EU Figure 3.8

Median values of output growth and output gap

8

6

4

2

0

–4 1970

1975

1980

1985 output gap

1990

1995

2000

2005

output growth

(Mayes and Virén, 2000b) suggest that such symmetry is unlikely and we find the same to be true here. The economic cycle itself is asymmetric with recessions tending to be sharper, shorter and shallower than expansions, at least in recent years for most European countries in our sample14 if the Finnish crisis of the 1990s is excluded.15 On the whole the asymmetry in the cycle is attributed, not so much to asymmetry in the shocks which assail economies, although this is the case if wars are included, but to asymmetries in behaviour. Although negative shocks tend to be transitory and positive shocks permanent (Nadal De Simone and Clarke, 2007). Many sources have been identified, in labour markets,

14 Verbrugge (1998) provides a helpful exposition of the nature of asymmetry in the main macroeconomic variables in 22 countries, including most of those in our sample. 15 The crises in the other Nordic countries round the same period, although traumatic, did not involve major falls in GDP. Finland’s recession was however deeper than in 1929.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

–2

Aggregate Supply and Demand in an Open Economy 63 Estimation of the IS curve with the output gap variable 1

2

3

0.457 (10.66)

447 (10.31)

0.466 (11.12)

re

0.003 (2.18)

0.002 (1.33)

rr

–0.010 (1.14)

–0.004 (0.41)

hp

0.023 (5.97)

sp

–0.001 (0.23)

0.002

0.003

0.641 (19.59)

0.601 (17.96)

0.663 (22.20)

∇yw

∇y–1 R2 SEE DW Estimator Panel Lags N

0.716 0.0070 2.199 LS CFE 2,2 1037

0.587 0.0103 1.929 LS CFE 2,2 1682

6

7

0.683 0.513 (8.13) (10.34)

0.797 (9.67)

0.699 (5.06)

0.004 (2.43)

0.022 (5.37)

0.014 (5.83)

0.007 (1.49)

0.015 (2.90)

–0.030 (3.39)

–0.010 (4.03)

–0.034 (2.89)

–0.052 (2.58)

–0.080 (1.18)

0.022 (2.97)

0.018 (3.94)

0.038 (4.09)

0.037 (2.07)

–0.005 (0.78)

–0.002 (1.77)

(2.31)

(2.01)

0.263 0.495 (4.04) (13.64)

–0.250 (4.84)

0.138 (2.44)

0.704 0.0071 2.163 LS CFE 2,2 1037

4

0.650 0.0058 1.923 LS CFE 2,2 449

5

0.618 0.0055 1.916 GLS CFE 2,2 449

0.167 .. 0.0074 0.0080 2.091 .. LS GMM Dif Dif 2,2 2,2 1022 1022

Variables and other labels as defined in Table 3.5. The dependent variable is the output gap, denoted by ∇y. Equations in the two last columns (4–5) are estimated from the sample of the EMU period 1999Q1–2006Q4. The value of the J-statistic is 10.74 which is not significant with the instrument rank of 15.

in productivity (Artis et al., 1999), in exit and entry (Chetty and Heckman, 1986; Baldwin and Krugman, 1989). The asymmetries in real behaviour and in inflation, while closely related, are different (Dupasquier and Ricketts, 1998, explore this for Canada, for example). Both fiscal policy and monetary policy have asymmetric elements to them (Mayes and Virén, 2004, 2005). Given this rich background, there are several ways in which we could introduce asymmetry. Their appropriateness depends on the specification of the model and the extent of the data we have to hand. One approach is simply to follow the framework of Sims and Zha (2006) and assume that there is a regime switch that corresponds to the up and down phases of the cycle. This would imply that we simply estimate two different models depending upon the phase. These could perhaps explain the phenomenon that Keynes noted that recessions tended to be shorter and sharper than expansions. A second possibility is to assume that there is more than one

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 3.7

equilibrium, as in Sargent (2001) for example; where in one case the economy is dominated by optimistic expectations and in the other by pessimistic expectations – shocks driving them from one to the other. There is some attraction in this approach in the context of asset prices. One way of explaining the bull and bear phases of the stock market would be to use expectations in this manner. As forward-looking prices they will be heavily affected by changes in expectations. A further possibility would be to consider the difference in constraints that appear in the up and down phases by using a form of Friedman’s (1968, 1993) plucking model, applied in Nadal De Simone and Clarke (2007) and Kim and Nelson (1999) for example. Here the assumption is that there is some maximal rate of growth determined by capacity and underlying technologies but that shocks drive the economy below that attainable level hence there is different behaviour when the economy is recovering from a shock from when it is running close to capacity. The model therefore finds that negative shocks tend to be temporary whereas positive shocks are more likely to be permanent, both driving the economy upwards and leading to clearly different behavioural responses. Housing (property) cycles might fit quite neatly into this framework as there are strong capacity constraints limiting the rate of expansion, with considerable lags involved. Moreover, given the interaction with financial markets, the up and down phases are characterised by rather different behaviour. When the market starts to go down people are inhibited from selling as otherwise they might realise collateral prices that are relatively low compared to the loans used to purchase. Indeed in some cases equity can become negative. This generates a complex interaction between prices and quantities. From the point of view of economic growth it is new construction that matters (in net terms at any rate) whereas prices reflect both the existing stock and new construction and are heavily dominated by the former. All these various models explain why we should expect different behaviour over the cycle and between them suggest two general ways in which we might represent them. The first is simply to suggest that the coefficients are different in the two phases. The second is to assume that there is a single equilibrium but that adjustment to it varies according to the phase of the cycle. Thus, for example, the reaction to a downward shock may be more rapid than to a positive shock which leads to an extended period above the longer-term equilibrium; see Enders and Siklos (2001) for example. We have explored this in Huang et al. (2001) in the case of monetary policy. Moreover, the switch between regimes may be a smooth transition with coefficients changing gradually over a number of periods, rather than an immediate switch from one to the other.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

64 Asymmetry and Aggregation in the EU

This gives us a considerable problem in choosing the best representation as the adjustment in behaviour will be spread across a number of equations in the model. Since we are limiting our main focus to the IS curve and the behaviour of monetary policy, we have opted for a straight forward approach, which is a version of the first group described above, namely to assume that the coefficients in the model are different in the two phases. To do this we introduce asymmetry through a threshold model (Tong, 1983; Teräsvirta and Granger, 1993) as discussed in Chapter 2. This means that we allow the variables of interest: the real exchange rate, real interest rate, house prices and stock prices to have different values if the economy is contracting from when it is expanding (Table 3.8). It is immediately apparent that all the variables have clearly different effects in expansions compared to contractions, with the exception of stock prices.16 The nature of the effect is interesting as all variables except foreign growth have a greater impact in an upturn than in a downturn. One possible way of thinking about this is to suggest that in expansions there will always be an element of capacity constraints that do not apply in a downturn. Thus there is some restraint in the way in which the economy can respond to a change in foreign demand. Interest rates and the exchange rate could be expected to have the same characteristics in some sort of real equivalent of the Phillips curve, where policy becomes less effective when the economy is relatively slack. Clearly we can produce arguments for other forms of asymmetry. For example, that producers will struggle to retain markets even if they are making short-run losses, because it will be much more expensive to try to enter a market having exited, as many contacts will be suspicious about the continuity of future supply. In Table 3.9 we consider a different form the asymmetry might take. In Table 3.8 we defined the cycle in terms of the growth of GDP. We can also consider it in terms of the direction of change of asset prices. This gives a much more direct representation of the change in expectations. We look in particular at the role of the real interest rate as representing the main monetary policy variable. If house or stock prices

16

The coefficients are jointly different in the two phases as indicated by a Wald test. In addition to the switching regression threshold model we have examined the results using a Smooth Transition Regression (STR) model where a logistic function is used to transform the transition variable. See Teräsvirta and Granger (1993) for details. Because the results with STR model were almost identical with the switching regression threshold models we do not report them here.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Aggregate Supply and Demand in an Open Economy 65

66 Asymmetry and Aggregation in the EU Asymmetry in the IS curve 1

2

3

4

∆yW

0.366 (6.30)

0.260 (3.47)

0.365 (6.94)

0.280 (5.11)

re

0.013 (4.35)

0.038 (1.80)

0.010 (4.41)

0.016 (0.55)

rr

–0.030 (1.86)

–0.060 (2.93)

–0.029 (1.98)

–0.042 (2.75)

hp

0.018 (2.36)

0.051 (5.86)

0.011 (1.88)

0.032 (5.22)

sp

0.010 (5.19)

0.009 (3.24)

0.010 (6.02)

0.007 (3.52)

0.636 (17.80)

0.575 (10.08)

0.659 (25.81)

0.615 (16.18)

∆y–1 R2 SEE DW Estimator Panel Lags sample N

0.830 0.0090 1.611 LS CFE 2,2 gap≤0 562

0.762 0.0106 1.973 LS CFE 2,4 gap>0 475

0.830 0.0089 1.621 GLS CFE 2,2 gap≤0 562

0.757 0.0104 1.798 GLS CFE 2,4 gap>0 475

The dependent variable is output growth. Notation is the same as in Table 3.5. Using the Chow test; parameter equality can be rejected (Thus, in the case of equations 1 and 2, F(21,106) = 3.31)).

are falling the real interest rate has a much more limited effect on output than when they are rising. This may help explain why in Chapter 8 on monetary policy, we find that interest rates change more vigorously in the down phase of the cycle. However, the coefficients are not well determined in the case of rising prices. Since we are looking here at European monetary policy this has nothing to do with any ‘Greenspan effect’. There has not been any suggestion that European countries have responded to house and stock market prices in the same explicit manner as has been developed in the US. What we see here, however, is a justification in Europe for just such an asymmetric policy response to asset price movements. Of course we have to take both the asymmetry in the asset price movements themselves as well as in interest rates to judge the policy

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 3.8

Aggregate Supply and Demand in an Open Economy 67

Dep var → ∆yw

The effect of house and stock prices: An update ∆y

∆y

∆y

0.281 (6.80)

0.318 (8.02)

0.289 (7.65)

gap

gap

gap

0.417 (10.38)

0.423 (10.70)

0.342 (10.76)

0.615 (2.97)

0.078 (0.61)

0.213 (1.66)

0.145 (1.08)

∇yw re/100

0.888 (4.13)

0.668 (3.35)

rr/100

–0.027 (2.99)

–0.037 (3.13)

–0.011 (1.29)

–0.012 (1.53)

hp

0.036 (7.49)

0.021 (4.38)

0.023 (7.14)

0.013 (3.28)

rr|hp<0

–0.061 (2.58)

–0.063 (3.87)

rr|hp≥0

0.015 (0.15)

–0.026 (1.89)

rr*hp

0.210 (1.76)

sp

0.009 (5.77)

0.005 (3.78)

0.231 (2.92) 0.001 (0.79)

rr|sp<0

–0.063 (3.47)

–0.002 (0.17)

rr|sp>0

0.001 (0.95)

0.017 (1.08)

rr*sp

0.120 (2.28)

lagged dep.var R2 SEE DW Wald

0.645 (25.49) 0.803 0.0100 2.12

0.675 (27.43) 0.800 0.0098 2.16 14.07 (0.000)

0.636 (26.13) 0.975 0.0098 2.13 5.11 (0.006)

–0.022 (0.62) 0.669 (23.71) 0.716 0.0069 2.23

0.676 0.659 (23.91) (24.41) 0.723 0.718 0.0066 0.0067 2.20 2.23 1.21 4.27 (0.299) (0.014)

hp and sp denote growth rates of real house and stock prices. rr is the short-term (three month) real interest rate. In interest rate (Taylor rule) equations hp and sp are, however, nominal change rates. The data cover the period 1979q1–2008q3. In the third equation, the hypothesis that the two multiplicative terms are identically equal to zero can be rejected by F test (F(2,1089) = 3.96).

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 3.9

response. More rapid responses on the downside could simply represent the steepness of the decline not asymmetry in policy. What is also noticeable from Table 3.9 is that the single regime that we apply to the rest of the model when using asset prices as the threshold looks very similar to the up phase model in the growth threshold for foreign growth, the real exchange rate and the lag but like the down phase model for stock prices and house prices. This suggests a much more complex asymmetry and one focused very firmly on the asset price variables. First of all it does not seem to matter much whether we use stock prices or house prices as the threshold. This is rather surprising as Figures 3.6 and 3.7 indicate that the two of them have not moved particularly closely together. Secondly, since the influence of stock prices is very similar in the up and down phases, this implies that the major concern in Europe is house prices, perhaps reflecting the smaller role of stock market funding in much of Europe outside the UK. We therefore deliberately return to this in our discussion of the consumption function, as it is here that housing wealth may have its main effect on GDP. Part of the explanation is that rising prices have been more prevalent than falling ones in most countries – Germany being the most obvious exception. Thus while the first four columns split the sample roughly in half, the split on asset prices is less equal. We have also applied our smooth transition model to (3.3). ∇yt = 0.660∇yt–1 + 0.416∇yWt + 0.114ret – 0.747rrt + 0.021hpt + 0.001spt + (22.80) (10.68) (0.88) (2.86) (3.39) (0.49) 1.470rrt (1/(1(1+exp(–hpt))) + 0.003rrt (1/(1(1+exp(–spt))) (2.81) (0.20) R2 = 0.718, SEE = 0.0067, DW = 2.23, θ1 = 0.006, θ2 = 0.900

(3.5)

∇yt = 0.659∇yt–1 + 0.412∇yWt + 0.144ret – 0.7821rrt + 0.014hpt + 0.001spt + (22.04) (10.59) (1.11) (2.94) (3.26) (0.98) 1.550rrt (1/(1(1+exp(–hpt – τhp))) – 0.014rrt (1/(1(1+exp(–spt – τsp))) (2.91) (1.04) R2 = 0.718, SEE = 0.0067, DW = 2.27, θ1 = 0.006, θ2 = 0.900, τhp = 0.2, τsp = 8.0

(3.6)

The first of these equations (3.5) applies the smooth transition model without a trend and the second (3.6) with trends in both house prices and stock prices. We can see that house and stock prices affect the transmission mechanism of interest rates. The interest rate effect is

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

68 Asymmetry and Aggregation in the EU

Aggregate Supply and Demand in an Open Economy 69

Figure 3.9

Comparison of GDP growth rates and inflation over 15 EU countries

80

GDP growth 1987–1998

70

60

GDP growth 1999–2006

50

60 40

50 40

30

30

20

20 10

10 0

0 –5.0 –2.5

0.0

2.5

5.0

240

7.5 10.0 12.5

Inflation 1987–1998

200

–2 100

0

2

4

6

8

10

12

Inflation 1999–2006

80

160 60 120 40 80 20

40

0

0 –4

–2

0

2

4

6

–1.25 0.00 1.25 2.50 3.75 5.00 6.25 7.50

All values are annual percentage growth rates.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

stronger when house and stock prices are falling. In fact, when house and stock prices increase very considerably, interest rates may become largely impotent as a policy variable. This is clearly of importance in the light of the experience of the present crisis. Writing shortly before the present crisis, though published somewhat later, we made the unfortunate remark in Mayes and Virén (2009) that as there had not been a serious recession in the data period that it was rather difficult to estimate the degree of asymmetry in the IS curve. In a few years’ time that will no longer be the case but too late for us to re-estimate for this book. Some of the problems of aggregating very different countries may be falling in importance in that growth rates and inflation rates have been converging (Figure 3.9). In the early period there was very considerable variation in the parameters of the IS curve across the member states as shown in Table 3.2, both in terms of lag structures and the values of the coefficients. The impact of a 100 basis point interest rate increase, after allowing for the

lag structures, varies from 0.5 to 3.8 per cent of GDP with the bulk of the estimates falling in the range 1.0 to 2.2 per cent. Thus, if the problem to be corrected by policy lay in low response countries in the euro area, other, more responsive, member states would bear a greater proportion of the adjustment if there were an equal change in the interest rate across the whole area.17 The problem posed for monetary policy by these ‘asymmetric’ differences even in the linear IS curve is magnified when the rest of our model is added. Estimating the effect of any particular setting of monetary conditions on inflationary pressure in the euro area involves not just the IS curve but the link from the output gap through to inflation.18 If the economic cycles of the member states are not in phase then the individual output gaps will be relevant in assessing the likely bite of monetary policy. In such a case it would be inappropriate to estimate an IS curve using aggregate data for the euro. Instead separate IS curves should be estimated at the disaggregated level and then aggregated.19 This is particularly important if the short-run Phillips curve is not linear and positive output gaps have a much stronger impact on increasing inflation than negative gaps have on decreasing it, as we show in the next chapter.

3.4

The forward-looking model

The analysis thus far is entirely backward-looking. When we include expectations the picture changes (Table 3.10). As noted earlier, our preference is to use measures of expectations rather than to assume rational expectations and insert actual future values in the estimation. One of the most widely available and widely used forward-looking variables

17 The data period for estimation is prior to the operation of the ECB so using it to draw inferences about the operation of monetary policy under Stage 3 of EMU implies some strong assumptions about the invariance of behaviour. However, it would require implausibly large changes for the problem we illustrate to disappear rapidly. 18 It requires at least a ‘Phillips curve’ relating price inflation to the output gap. If the Phillips curve uses unemployment as the determining variable then an Okun curve is required as well to provide the link between output and (un)employment. 19 It is not of course self-evident that it is the member state level that is appropriate for the disaggregation. It should really be regions in which behaviour is fairly homogeneous. (Dupasquier et al. (1997) demonstrate that in some cases there is more variation between some Canadian provinces than there is between Canada and the US.) Commodity price shocks may have regional rather than national impacts. However, the data to hand are on a member state basis.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

70 Asymmetry and Aggregation in the EU

Aggregate Supply and Demand in an Open Economy 71 Table 3.10 Eq

∆yt–1

1

0.460 (6.41) 0.412 (6.89) 0.467 (7.85) 0.409 (6.95) 0.422 (9.04)

2 3 4 5

Euler equations for output using the Consensus Forecast data ∆yt+1

∆ywt

0.485 0.278 (4.31) (0.58) 0.453 0.066 (6.89) (0.16) 0.485 0.048 (3.66) 0.455 0.136 (3.22) 0.481 0.097 (6.35)

rr

re

hp

sp

R2/SEE

DW

Method

–0.061 (0.21) –0.080 (1.22) –0.036 (0.67) –0.074 (1.24) 0.041 (1.62)

–0.010 (0.73) –0.000 (0.03) –0.003 (1.62) –0.002 (1.25) 0.003 (2.88)

0.056 (2.74) –0.049 (2.71) 0.0059 (3.24) 0.048 (2.68) 0.065 (8.35)

0.028 (7.25) 0.027 (7.41) 0.028 (5.89) 0.027 (7.37) 0.025 (8.73)

0.611 0.0127 0.302 0.0126 0.610 0.0127 0.602 0.0126 0.607 0.0126

1.88

OLS

1.80

GLS

1.90

OLS

1.79

GLS

1.90

SUR

is Consensus Forecasts. They are compiled by Consensus Economics, an organisation that polls a wide range of reputable agencies to obtain their forecasts and hence represents a good estimate of the mean value of what is expected. They are not of course necessarily the best forecasts in the sense of being the best informed, or those with the best track record for accuracy. However, we are concerned not with an accurate forecast – we know the recorded value – but with a good representation of what people thought was likely to happen. The consensus in this sense represents that. The first point to note is that the weight on the forward variable is much stronger than the weight on the lag. Secondly, this equation (3.3) also includes stock prices and house prices. These two variables between them represent a wealth effect. Both clearly have a positive effect on output, although the influence from stock prices seems clearer, which is unusual. If we consider the more recent data, shown in the last row, it appears that the importance of house and stock prices is increasing. Clearly prices are an imperfect measure of wealth but one that is widely used (see Chapter 8).

3.5

Consumers expenditure

By focusing initially on aggregate demand to explore the relationship, we have compounded a number of different routes through which asset prices could be having their effect on economic activity. We therefore look explicitly at the most obvious area where we should expect to see an

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

In equations (3)–(5) the sum of output coefficients is restricted to be one.

72 Asymmetry and Aggregation in the EU

influence from asset prices namely in the consumption function. We use a generalised form (3.7)

where ∆cq is the growth rate of real consumers’ expenditure, ζ is an error term and all other variables are defined as before. Clearly a properly specified function would use disposable income and wealth not GDP and asset prices as these are proxies but nevertheless they enable us to explore both the influence of asset prices and whether consumption is an area where asymmetry appears to important, as is shown in Table 3.10. Altissimo et al. (2005) give a clear review of the literature on the wealth effect in the consumption and look at experience in trying to estimate the relationship, particularly for European countries. A further review and new estimates is to be found in Labhard et al. (2005). We are in good company in proxying wealth by asset prices (Ludwig and Sløk, 2002). The alternative of using incompatible definitions or omitting many of the countries is not very attractive. Furthermore house prices can have an effect on consumption by a variety of routes in addition to wealth. The simplest is that they affect borrowing constraints. Indeed without this effect it is not so clear why a change in house prices should affect consumption as having a house is a route to consuming housing services. When house prices rise so do implicit rentals (Campbell and Cocco, 2007). The results are fairly similar to those expected. Real interest rates do not seem to be very important in the euro area period.20 Both house prices and stock prices have an effect but significance levels are rather variable.21 The effect from stock prices is small but that from house prices noticeable. This is the expected way round, as housing wealth is held by a much larger range of consumers than financial wealth that tends to be concentrated in the hands of the rich, whose (marginal) propensity to consume is lower (Carroll, 2004). It also conforms to the empirical results in Case et al. (2005) and Catte et al. (2004), although Ludwig and Sløk (2002) obtain a larger coefficient for stock prices than housing prices. Slacalek (2006) suggests that on the basis of a sample of 16 OECD countries that each extra unit of wealth leads to a 0.03 increase in

20

This result reflects the dominance of the continental European countries in the sample. The interest rate effect is clearly stronger for the UK (and also for Finland) (Labhard et al., 2005). 21 While our work focuses on macroeconomic data, there are cross-section studies that also find clear evidence of an effect on consumption from housing wealth – see for example Campbell and Cocco (2007) and Disney et al. (2007) for the UK.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

∆cqit = δ0 + δ1∆yit + δ2 rrit + δ3hpit + δ4spit + δ5cqit–1 + ζit

consumption.22 Our long-run stock price estimates are roughly of this order of magnitude, although of course this makes no allowance for new wealth creation, only the revaluation effect. However, our stock price effect is only around one-third of this. Our results fall between Slacalek’s estimates and those of Labhart et al. (2005) who use a subset of 11 of Slacalek’s 16 countries. The degree of persistence illustrated in Table 3.11, at around 0.6, is the same as Slacalek finds. The evidence on asymmetry is rather thinner. It is only in the case of column 7, where stock prices are used as the threshold variable, that the Wald test suggests that the coefficients above and below the threshold are different at the 5 per cent level. The coefficients themselves are different in each case and appear to tell a plausible story. Consumption is less affected by interest rates when asset prices are falling (or below their trend rate of growth as we explore for house prices in column 6). Consumption responds more to changes in ‘income’ when growth is positive or the output gap is positive. We were expecting a stronger effect here as there is considerable evidence that people are reluctant to see their consumption fall in the short run when their incomes fall but are happy to take a proportion of any rise in the form of consumption (Duesenberry, 1949). This result is quite striking in Disney et al.’s (2007) study of the UK, where a surprise rise in house prices gets translated into small fall in saving (and hence rise in consumption) but a surprise fall in house prices leads to an even higher fall in saving – thus showing notable asymmetry. Our results are not as clear cut as those of Labhard et al. (2005) who show both that the relationship between wealth and consumption is nonlinear – in that large changes in wealth have a less than proportionate effect on consumption than small changes – and that it is asymmetric, with consumption falling less when wealth falls than it rises when wealth rises.23 We were expecting that house prices would be a clear indicator of the share of liquidity constrained households. This would provide an observable distinction between periods when the market is moving ahead normally and those when house prices fall, with people facing negative 22

Slacalek (2006) includes the US, Australia, Canada and Japan to our sample but excludes Greece, Norway and Portugal. A second feature is that wealth effects are much larger in the Anglo-Saxon countries than in continental Europe. This may imply that the UK sits a little uneasily in our sample. 23 Labhart et al.’s panel of 11 OECD countries covers eight of the 15 in our sample plus the US, Japan and Canada. Most reassuring from our point of view is their finding that, in their panel estimation, the hypothesis that the long-run marginal propensity to consume from wealth is the same across the countries cannot be rejected and that its magnitude of a little over 6 per cent is slightly larger but consistent with ours.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Aggregate Supply and Demand in an Open Economy 73

74 Asymmetry and Aggregation in the EU Estimation of a ‘consumption function’ 1 ∆y ∆y|x≤0

0.278 (8.08)

2

–0.030 (2.00)

4

0.233 (3.85)

∆y|x>0 rr

3

–0.022 (0.73)

0.270 (7.91) 0.263 (7.75) 0.329 (9.18) –0.030 (1.99)

rr|x>0 0.015 (2.40) sp 0.004 (2.22) ∆cq–1 0.604 (19.05) R2 0.787 SEE 0.0095 DW 1.953 Estimator LS Panel CFE Period 79–07 x .. N 843

0.023 (1.87) 0.001 (0.45) 0.603 (11.60) 0.793 0.0088 1.989 LS CFE 99–07 .. 450

0.015 (2.34) 0.002 (1.61) 0.598 (18.87) 0.791 0.0094 1.980 LS CFE 79–07 gap 843

6 0.270 (8.04)

7 0.264 (7.82)

0.309 (7.45) 0.250 (6.93) –0.026 (1.63)

rr|x≤0

hp

5

0.021 (2.65) 0.004 (2.23) 0.600 (18.76) 0.789 0.0095 1.954 LS CFE 79–07 hp 843

–0.034 (1.46) –0.028 (1.70) 0.014 (1.97) 0.004 (2.22) 0.604 (19.02) 0.788 0.0095 1.952 LS CFE 79–07 hp 843

–0.042 –0.058 (2.19) (3.01) –0.015 –0.003 (0.70) (0.20) 0.011 0.014 (1.51) (2.19) 0.003 0.001 (2.24) (0.69) 0.604 0.610 (19.00) (19.25) 0.788 0.790 0.0094 0.0094 1.951 1.969 LS LS CFE CFE 79–07 79–07 hp<3.3 sp 843 843

Notation as in previous tables. The dependent variable is the growth rate of private consumption ∆cq. Here, the real interest rate rr appears without a lag. With the threshold models, the threshold value of the threshold variable x is zero except for equation (6) where it is 3.3 per cent. Although the coefficient estimates in threshold estimation appear to be different the hypothesis that the coefficients are equal cannot be rejected at the 5 per cent level of significance. Only with equation 7 is the Wald test statistic significant (F(1,823) = 6.49 (0.011)).

equity problems and related constraints on the willingness to sell. Those who cannot realise their investments would then face liquidity constraints that would feed through into consumption. Consumption smoothing across the cycle requires the effective operation of financial markets, inter alia (Morduch, 1995) so the liquidity constraint should be asymmetric. Contrary to what one might expect, however, when house prices are falling consumption changes more with income than when they are rising. Possibly this is because the usual experience might be a fall in income in these conditions. This is of course in addition to the direct effect of the

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 3.11

change in wealth as indicated by asset prices. However, perhaps the easiest way to look at this is that if asset prices and hence wealth is rising that can be a generator of increased consumption and income does not have such an important role to play. The nature of the impact of asset prices on the economy is clear from a consumption function, where they can be used to proxy the effect of wealth (Ludwig and Sløk, 2002). The marginal propensity to consume from wealth that can be derived is of the order of 0.06, which is in the middle of the range of estimates available for the OECD countries. The bulk of this effect comes from house prices and not stock prices, although the effect varies across the European countries, driven to a large extent by the relative importance of stock markets in company finance and the extent of direct ownership of housing (Maclennan et al., 1999). It is thus clear that house prices play an important role in the economic cycle and inflation in our sample of European countries, the EU15 plus Norway and minus Greece, and that their impact varies over the course of the cycle. Our results are, of course subject to a range of measurement, econometric and theoretical provisos. House price data typically show variations in definition across countries. Although we have experimented with a number of variants to test the robustness of our results, to quite some extent they will be dependent on the specification we have chosen. While our use of a panel of 15 countries enables us to derive estimates in a way which would be difficult for any individual country, our assumption of parameter constancy (after allowing for fixed effects) is clearly a very strong one even though it seems to be statistically consistent with the data.24 The routes of influence, particularly through the monetary policy transmission mechanism, are complicated and can be illustrated to an extent by simulation (see Mayes and Virén, 2005, Fig. 6, for example).25 However, our model is rather too simplified to give it justice. What is clear is that there are distinct channels of influence from asset prices in addition to those from the exchange rate and the direct influence of interest rates through aggregate demand, even though the lag structures are likely to be much more complex than we can allow.

24

Labhart et al. (2005) also have problems in testing for parameter constancy. With the current data, if we allow both house and stock prices to depend on real interest rates, the combined effect of interest rate on GDP becomes almost twice as large as with a single equation model (4) in Table 3.1 (the short-term GDP effects of a 1 per cent increase in real interest rate turned out to be –0.31 and –0.16 per cent, respectively). Thus, from the point of view of monetary policy, it is not trivial how the aggregate demand relationship is assumed to function. 25

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Aggregate Supply and Demand in an Open Economy 75

76 Asymmetry and Aggregation in the EU

Appendix 3.1 Panel unit root tests the main variables of the model

GDP growth, ∆y output gap, ∇y real interest rate, rr real exchange rate, re real house prices, hp real stock prices, sp

Levin-Lin-Chu ‘t’

Pesaran-Shinn ‘W’

2.345 (0.095) 4.699 (0.000) 2.690 (0.004) 0.230 (0.409) 0.787 (0.215) 2.791 (0.003)

10.221 (0.000) 15.132 (0.000) 5.235 (0.000) 1.126 (0.130) 3.002 (0.013) 5.268 (0.000)

Inside parentheses are the marginal significance levels. The number of cross-sections is 16.

∆y = GDP growth rate is the four-quarter growth rate of Gross Domestic Product. ∇y is the output gap that is derived from the GDP using the Hodrick–Prescott filter. rr is the real ex-post interest rate that is derived as a difference between nominal short-term (three month) interest rate and the four-quarter change rate of GDP deflator. re is the real exchange rate that is derived from the nominal Euro/USD exchange rate and the GDP deflators of the USA and the home country. hp is the four-quarter change rate of real house prices that are derived from nominal house prices and GDP deflator. Similarly, sp (the change rate of real stock prices) are derived from nominal stock prices indexes and the GDP deflator. The data source for ∆y, ∇y, rr, re and sp is the OECD Main Economic Indicators data bank. House prices come from various national data sources (a more detailed list of those is available upon request from the authors).

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Data sources

4

The Phillips curve provides the key link between the real economy and inflation and lies at the heart of our analysis. Although there are various asymmetries involved in the relationship the most obvious facet is that the relationship is referred to as a curve. Although in many models it is estimated as a linear relationship in part because of the difficulties that Phillips himself encountered in the original estimation (Phillips, 1958). Indeed it is only partly an accident of history, with the collapse of the long-run regularity and its replacement with a short-run expectations augmented curve (Phelps, 1967), that it has frequently been estimated as a straight line.1 In Phillips’ original specification, the rate of change of money wages is related to the rate of unemployment and described as a regularity without any formal attempt at a firm theoretical basis. Since then, in addition to including price expectations following Phelps (1967) and Friedman (1968) and later rational expectations (Lucas, 1976), the tendency has been to replace wage inflation with price inflation and frequently to use the output gap instead of unemployment as a measure of the pressure in the economy. We deliberately do not take a stand on what is the most appropriate specification. In Paloviita and Mayes (2006), we have examined a number of common versions of the models including expectations from the ‘new classical’ model using last period’s expectations to the ‘pure’ New

1 The discussion of the Phillips curve remains contentious. Gordon (1997) maintains that it is ‘resolutely linear’ in the US while Stiglitz (1984) suggests that it could have the opposite curvature with firms being more reluctant to raise rather than lower prices. Yates (1998) offers a helpful classification of the main different factors that could lead to nonlinearity.

77

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

The Phillips Curve

Keynesian model, which uses this period’s expectation of future inflation. They all have the feature that expectations matter. A purely backwardlooking approach to the explanation of inflation does not conform to the data as well. Here therefore we use the well-known hybrid New Keynesian specification of the Phillips curve as a representation of the prevailing view, which allows some effect for both inflation expectations and lagged inflation. This has the advantage of making the results directly comparable with a wide range of evidence already available. There are however some major disadvantages in using an output gap as an explanatory variable, as it, like expectations, is also an unobservable variable that needs to be estimated. We therefore also show results using unemployment as in the original Phillips curve. Many measures of output gaps suffer from the ‘end point’ problem, which means that they rely on unknown future data. We therefore explore the approach of using ‘real time’ data in the form of survey evidence and forecasts made at the time in order to provide a more accurate view of what people thought at the time was likely in the future. It is only well after the event that we can form a clear view of whether the trend from which the gap is measured has itself changed. In the short run there is considerable scope for confusing gaps with changes in trend. Chapter 5 includes further work on the Phillips curve looking at more disaggregated information both at the sectoral and regional levels.

4.1

The specification of the Phillips curve

The original New Keynesian specification is:

πt = λ

∞

∑ βkEt{mct+k}

k=0

(4.1)

where Et is the expectations operator conditional on information available in period t, πt denotes the period t inflation rate, defined as the rate of change of prices from period t–1 to period t, suggesting inflation is equal to the discounted stream of future real marginal costs, mc. As it is usually impossible to obtain a direct estimate of these costs, empirical studies commonly use the output gap as a proxy, although labour costs are also used (Galí and Gertler, 1999; Sbordone, 2002). These variables are assumed to capture changes in real marginal costs associated with variation in excess demand in the economy. Under certain assumptions about technology, preferences and the structure of labour markets we can link

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

78 Asymmetry and Aggregation in the EU

The Phillips Curve 79

the output to real marginal costs within a local neighbourhood of the steady state of log real marginal costs according to mct = δyˆt (Fuhrer and Moore, 1995; Roberts, 1998) where yˆt denotes the period t excess demand and hence by substitution we obtain:

πt = βEt {πt+1} + κ yˆt,

(4.2)

for the New Keynesian relationship, where κ = λδ. Many empirical studies suggest that the purely forward-looking New Keynesian Phillips curve is unable to capture the persistence of inflation and hence it has typically been modified by incorporating some backward-looking element. The Hybrid model of Galí and Gertler (1999) is the main example: (4.3)

where 0 ≤ θ ≤ 1.2 We use this Hybrid model as our starting point. Before we go any further we need to sort out what is meant by asymmetry in this context, as there is no commonly accepted definition.3 Sorting out nonlinearity is perhaps a simpler task as we take it here to refer to relationships that are curvilinear or have different parameter values over different ranges, rather than exhibiting discontinuities or chaotic behaviour. In the European context the most common use of the word ‘asymmetric’ merely means ‘different’. The simplest example comes in the concept of asymmetric shocks, which are just shocks that affect one part of the economy rather than another. Secondly, asymmetry is commonly used to refer to relationships where there are omitted variables or even omitted secondary equations. Gaiotti and Generale (2001) and Loupias et al. (2001) in showing that there is a credit channel for monetary policy describe this additional feature as ‘asymmetry’ in the monetary transmission mechanism.

2 In Paloviita and Mayes (2006) we have also explored using further lags of prices (Galí et al., 2001; Jondeau and Le Bihan, 2003) and the labour share of income rather than the output gap. 3 Much of the traditional treatment of asymmetry (Keynes, 1936; Diebold and Rudebusch, 1999) is concerned with the shape of the business cycle. Three characteristics of asymmetry in shape can be identified: deepness – do recessions tend to be deeper than booms are high (compared to some trend or sustainable growth path); length – do expansions tend to last longer than recessions and steepness – does the decline occur more rapidly than the recovery. See the discussion in Chapter 1.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

πt = θEt {πt+1} + (1 – θ)πt+1 + φ yˆt

80 Asymmetry and Aggregation in the EU

A simple example for the Phillips curve, is to suggest that inflation responds differently depending upon whether the output gap is positive or negative. We can easily respecify the relationship in first difference form (or first difference of logarithms) and show that for example the impact of changes in GDP on unemployment depends on whether the economy is growing or contracting.4 We thus begin by applying the form of threshold asymmetry described in Chapter 2 to the hybrid Phillips curve shown in (4.3)5 (4.4)

where ∇y+ denotes the values of the output gap that exceed the threshold value and accordingly ∇y– denotes the remaining values of ∇y, πe is expected inflation for ease of notation. By having these two facets unlike much of the rest of the literature we take at least one step towards admitting that the Phillips curve is a curve.6 All reported estimates have been derived using a panel data and restricting the key parameters to be the same for all countries and periods (although, with some exceptions). All equations have been estimated with Least Squares (LS) and Generalised Least Squares (GLS). Because all equations include lagged dependent variables (either directly or through the error-correction terms) we have also used the ArellanoBond version of the Generalised Method of Moments Estimators (GMM). To illustrate the robustness of results, we present results from all estimators although space prevents a complete report.

4 Corrado and Holly (2003) try to estimate a general hyperbolic functional form for the Phillips curve. In practice, they end up by estimating two thresholds. Their results for the UK and the US suggest that the Phillips curve is steeper for larger positive output gaps than it is for larger negative gaps, while in the middle, for small positive and negative gaps, the curve is fairly flat. 5 See Clark and Laxton (1997) for a brief review and an alternative approach. 6 Obviously we could have more than two regimes (facets) for ∇y but since we have only limited numbers of observations we use this simple specification (which has been widely used elsewhere, see Yates (1998) for instance). Alternatively we could smooth the once-and-for-all regime shift in the threshold model by using the so-called Smooth Transition Regression model (STR) (Granger and Teräsvirta, 1993), also used by Teräsvirta and Eliasson (1998) and set out in Chapter 2. The lack of sufficiently long time series also made this alternative less appealing. Introducing a quadratic term in the output gap would also be a straightforward way of incorporating nonlinearity.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

πt = a0 + a1πt–1 + a2πe + a3∇y+ + a4∇y– + u

The Phillips Curve 81

Asymmetry in the Phillips curve

There is no shortage of estimates of Phillips curves in recent years. Estimating the conventional New Keynesian hybrid equation gives the result that forward- and backward-looking inflation expectations are equally important and that a perverse sign is achieved for the output gap. In part this stems from trying to estimate the curve as if it were a straight line. As soon as even a two part linear specification is permitted the sign is corrected and the slopes for positive and negative output gaps are clearly different. In particular, when the output gap is negative and the economy is characterised by slack capacity the Phillips curve is nearly horizontal. This flattening confirms the flattening that has been observed more generally as inflation rates fall. However, it is important to see that this feature does not apply to positive output gaps. In the case of the Phillips curve differences between the euro area and the rest of the EU do matter. Elsewhere inflation is not so responsive, in part perhaps because the economies are more open to the world outside the euro area. Although the output gap-based New Keynesian Phillips curve under rational expectations has often been used in monetary policy analysis, the empirical validity of the model has not been firmly established. Inflation seems to be poorly captured and the output gap is often insignificant or has a wrong, negative, sign. To represent the data better, the model has typically been modified, using the lagged inflation rate and alternative measures of real economic activity. The real marginal costs implied by the New Keynesian theory are difficult to measure empirically. This has led to a debate in applied work over whether the output gap or labour costs is the appropriate measure of cyclical inflationary pressure. The output gap has been criticised for having problems in measurement and it may not move proportionally with real marginal costs due to the failure to account for labour market frictions. Moreover, the relation between unobservable firm-level marginal costs and observable aggregate marginal costs appears to be problematic and the estimation results seem to be highly sensitive to the specification of labour costs (Lindé, 2002). Empirical studies have raised questions about the adequacy of the underlying theory. The theoretical model under rational expectations does not imply any lags of inflation. Such lags have been interpreted as representing agents who only look backwards when setting prices. The poor empirical fit may thus be associated with the possible inaccuracy of the rational expectations hypothesis assumption.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

4.2

For example, Fuhrer and Moore (1995) have shown with US data that the New Keynesian model under the rational expectations hypothesis without lags cannot capture inflation dynamics. While in a more recent study by Gagnon and Khan (2001) inflation responds better to anticipated movements in labour costs than to the output gap. When the output gap is used in that study, the estimated slope of the New Keynesian Phillips curve is negative for the euro area and the US but positive for Canada. When lagged inflation is added to the output gapbased model, the slope of the curve is positive for the euro area and Canada but negative (wrong) for the US. In most of the output gapbased estimation by Gagnon and Khan, the estimated coefficient on output gap itself is insignificant. As a result of looking at a wide range of estimates, Galí and Gertler (1999) conclude that the New Keynesian Phillips curve provides a reasonable approximation to inflation dynamics in the US, when the labour income share is used to measure real marginal costs. Although backward-looking price setting is statistically significant in their study, it is not quantitatively important. Galí et al. (2001) favour the New Keynesian Phillips curve conditional on real unit labour costs instead of the output gap both for the euro area and the US. While there seems to be some backward-looking in inflation dynamics, they find that the New Keynesian Phillips curve fits the euro area data. Others have a similar point of view – Sbordone (2002) argues also that the New Keynesian Phillips curve performs better when real labour costs are used. By contrast, Rudd and Whelan (2002) argue that the empirical fit of the New Keynesian model is not improved when using real labour cost instead of the output gap. There is rather less evidence outside the US. Benigno and Lopez-Salido (2002) have compared inflation dynamics in five major EMU countries. When studying France, Germany, Italy, the Netherlands and Spain they provide evidence on heterogeneity in price changes across the countries. This study suggests that inflation has a dominant forward-looking component only in Germany and there is a significant backward-looking component in inflation processes in the other four countries. Unfortunately there is very little work done with specifications other than rational expectations but Roberts (1997, 1998) has analysed inflation dynamics in the US with the New Keynesian specification by using survey estimates of inflation expectations instead of rational expectations assumption. He finds evidence that inflation expectations are not rational, which appears to be connected to the poor empirical fit of the New Keynesian theory. However, the results are highly dependent on inflation

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

82 Asymmetry and Aggregation in the EU

The Phillips Curve 83

surveys, which have been criticised as being unreliable estimates of inflation expectations. We find by contrast that if model driven rational expectations are replaced by OECD forecasts of inflation it is possible to get reasonable estimates of the New Keynesian relationship, for the euro area, when only using the forward-looking term, i.e. without resort to lagged inflation terms. Once the hybrid model is used then surveys also appear to work satisfactorily.

Estimates of the Phillips curve

We begin by considering forward-looking expectations and estimate the New Keynesian hybrid Phillips curve using the Generalised Method of Moments in a dynamic panel framework (Arellano and Bond, 1991). The results (Table 4.1) clearly indicate that the role of the lagged inflation term (inflation persistence) diminishes over time and is relatively unimportant for the 1999–2006 period. By contrast the role of the output gap becomes more prominent. Contrary to the 1987–1998 period, the coefficient is clearly significant. Thus, for the EMU period, the New Keynesian hybrid Phillips curve works reasonably well; it is only that the sum of the inflation coefficients fall short of one, which suggests that there are some problems in the modelling inflation expectations with the REH assumption under the GMM orthogonality restrictions. The data set used here is the same as in the previous chapter and includes all the EU15 countries except Luxembourg but includes Norway. The data are quarterly, running from the beginning of 1971 through to the end of 2008. Thus they include the high inflation period of the 1970s, the progressive reduction in inflation in the 1980s and early 1990s, the period of the ‘great moderation’ of low and stable inflation and the blip in inflation at the end that accompanied the boom before the present recession. The rapid decline of inflation in 2009 and the descent into deflation in some cases lie outside the data period. In the main the coefficients on the output gap/GDP growth rate/ unemployment are small, positive and plausible. It is also clear that when the output gap is negative or unemployment is above its trend rate that the response coefficient is lower then when it is positive (below trend for unemployment, thus emphasising the curvature of the Phillips curve as expected). However, in common with other authors we do have some examples of negative coefficients despite the presence of a lagged inflation term. Unlike some of our other estimates, lagged inflation tends to

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

4.3

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

0.119 (2.87) 0.007 (3.80) 0.001 (0.02) –0.167 (11.26) 0.001 (13.83) 0.006 (0.64) 0.007 (0.25) 0.080 (5.58) 0.031 (4.75)

1987–2008 DEF, GAP 1987–2008 CPI, GAP 1999–2008 CPI, GAP 1971–2008 CPI, GAP 1987–2008 CPI, UNC 1987–2008 CPI, GAP 1971–2008 CPI, GAP 1971–2008 CPI, GAP 1971–2008 CPI, g 1999–2008 CPI, GAP 1971–08 CPI, GAP 1971–08 CPI, UNC 1971–08 CPI, UNC 0.016 (0.44) 0.035 (1.38) –0.214 (4.34) 0.149* (1.93)

Output y<0

0.062 (1.82) 0.138 (4.96) –0.066 (1.45)

Output y≥0

0.507 (45.21) 0.524 (45.21) 0.935 (107.5) 0.933 (122.4) 0.916 (59.54) 0.932 (127.8) 0.929 (120.6) 0.929 (120.5)

Lag 0.701 (35.70) 0.895 (34.92) 0.794 (18.16) 0.942 (28.15) 0.859 (34.80) 0.490 (10.37) 0.506 (12.67)

Forward

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

–0.214 (4.34)

Output

Phillips curves

Data

Table 4.1

0.028 (7.71) 0.029 (7.90) 0.027 (6.42) 0.028 (6.44) 0.030 (6.80) 0.030 (6.80)

Import p

0.007 (5.25) 0.003 (1.79) 0.009 (4.57) 0.005 (3.20) 0.006 (3.84) 0.006 (3.85)

Max-min

.. 0.0021 .. 0.0061 .. 0.0060 .. 0.0065 .. 0.0154 0.972 0.0033 .. 0.0056 0.973 0.0050 0.806 0.0060 0.973 0.0098 0.970 0.0052 0.970 0.0053 0.970 0.0053

R2/SEE

15.42 .. 12.78 2.78 – .. 11.27 1.69 – 1.16 – 1.69 – 1.67 – 1.64 – 1.64

.. 10.59 .. 10.45 .. 10.80

DW/J

OLS

OLS

OLS

OLS

GLS

GMM, Dif GLS

GMM, Dif GMM, Dif GMM, Dif GMM. Dif GMM, Dif IV

Method

84

–0.210 (4.65)

1971–08 CPI, UNC 1999–08 CPI, GAP EMU 99–08, CPI, GAP NEMU 99–08, CPI, GAP 1990–2007 CPI, GAP, C 1990–2007 CPI, GAP, C 1990–2007 CPI, GAP, C 1999–2007 CPI, GAP, C

0.156* (2.12) 0.015 (0.44)

Output y<0

0.062 (1.82)

Output y≥0 0.929 (126.5) 0.916 (59.59) 0.967 (96.61) 0.967 (47.14) 0.592 (6.06) 0.678 (12.96) 0.608 (18.67) 0.624 (7.38)

Lag

0.302 (3.55) 0.238 (4.73) 0.291 (9.04) 0.413 (3.74)

Forward 0.031 (8.43) 0.026 (6.42) 0.036 (7.17) 0.030 (3.04)

Import p 0.006 (3.65) 0.009 (4.57)

Max-min

0.814 0.0041 0.829 0.0043 0.723 0.0042 0.599 0.0084 0.594 0.0084 0.598 0.0084 0.441 0.0073

0.972

R2/SEE

OLS OLS

1.44

OLS GLS SUR OLS

1.80 1.94 1.82 1.97

GLS

GLS

Method 1.63 0.0053 1.89 – 1.94

DW/J

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

UNC refers to an unemployment gap (the unemployment rate – Hodrick–Prescott trend unemployment rate). In the threshold model, the high unemployment rate variable comes first. With the J-statistic, the degrees of freedom are 11 (thus, none of the statistics is significant at conventional levels of significance). The last four lines (marked with C) make use of the Consensus Forecast data for next year’s inflation. In this case the data are annual, and the forecasts are made in June. * indicates the coefficient of UNC{1/[1+exp(–UNC)]}. EMU restricts estimation to the countries that are members of the euro area and NEMU to those that are not.

0.035 (1.52) 0.043 (1.11) 0.360 (5.26) 0.367 (7.14) 0.359 (13.78) 0.361 (4.69)

Output

Phillips curves – continued

Data

Table 4.1

85

have a similar weight to expected inflation and there appears to be little change over time. We deliberately show a wide range of specifications using different estimation methods and time periods so that the robustness of our results is clear. We consider both the CPI and the GDP deflator as measures of inflation and the output gap, GDP growth rate and unemployment rate compared to trend as measures of demand pressure. The results are quite stable, whether the monetary union or whole dataset are used and are similar for both the euro area countries and the nonmembers. However, a quick look at the data behind the three demand variables (Figure 4.1) suggests that their properties are likely to be relatively different. The inflation rates on the other hand are relatively similar (Figure 4.2). The early part of Table 4.1 effectively uses a rational expectations approach by including the actual forward value as the inflation expectation. The last four rows use Consensus Forecasts. This has a noticeable affect. The output gap coefficients become larger and clearly differ from zero at even the 1 per cent significance level. The forward weight on inflation falls somewhat. Of course it is open for debate whether

Figure 4.1

Indicators of output (medians)

0.100 0.075 0.050 0.025 0.000 –0.025 –0.050 –0.075 –0.100 1970

growth gap un gap 1975

1980

1985

1990

1995

2000

2005

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

86 Asymmetry and Aggregation in the EU

The Phillips Curve 87 Figure 4.2

Comparison of inflation rates (medians)

16 14 12 10 8 6

2

GDP deflator CPI

0 1970

1975

1980

1985

1990

1995

2000

2005

Consensus Forecasts are actually more accurate estimators of expectations. The standard errors of estimate are a little higher. We have also tried augmenting the regression and including two further variables that might help improve the explanation. The first is to include import prices. Import prices can contribute to inflation at different times from the pressure of demand and their importance is clear from the estimates. We also consider whether the economy-wide measure of unemployment/demand pressure is appropriate and hence look at the distribution of unemployment ∆p = b0 + b1∆pt–1 + b2∆pet+1 + b3un+t + b4un–t + b5mt + b6dispt + ut

(4.5)

where the disp variable reflects either the range or the standard deviation of unemployment rates over regions in a country i. In Table 4.1 only the range is considered, labelled ‘max – min’ for clarity. m denotes import prices. The choice of a zero output gap as the point round which to split the data is somewhat arbitrary. Although by construction of the output gap variable this will be a split around the mean value. Other splitting points might perform better empirically but a search over the range

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

4

reveals that the errors are minimised at a value of –0.005 for the output gap. Both these measures work well as explanations. Clearly there will be some complex inter-relationships as import prices themselves will be affected by the exchange rate. The importance of the spread of unemployment is explored in more depth in Chapter 5 but the concern here is simply that neither unemployment gaps nor output gaps that relate to the whole economy are appropriate indicators of inflationary pressure on their own. Buxton and Mayes (1986) argue on the basis of evidence from the UK that even when collective agreements across the economy are not strong, it is labour shortages in the tightest markets, whether geographical or sectoral, that tend to have the influence on the overall rate of inflation. It also raises a wider issue of whether considering average values in the setting of monetary policy is adequate, thus emphasising the issue of aggregation that we raised in Chapter 2. However, in earlier work (Paloviita, and Mayes, 2005; Paloviita and Virén, 2005) we have found that the rational expectations formation is a rather strong assumption and that if instead we use OECD forecasts or Consensus Economics survey data, we get a better determined equation and a much larger forward-looking weight, more in line with what is expected from the New Keynesian model.7 This is again true Table 4.2

GMM estimates of a New Keynesian Phillips curve ∆4p–1

∆4p+1

gap

SEE

J(6)

1975–1998

0.533 (65.81)

0.430 (9.64)

0.003 (0.16)

0.0127

9.49

1987–2006

0.423 (186.74)

0.422 (75.24)

0.035 (3.35)

0.0136

13.66

1987–1998

0.502 (17.61)

0.370 (20.30)

0.067 (0.78)

0.0130

11.08

1999–2006

0.267 (65.08)

0.397 (59.15)

0.114 (5.39)

0.0135

11.95

All estimates are Arellano-Bond GMM estimates with current and lagged values of import prices as additional instruments (in addition to the lagged values of the right-hand-side variables). First differences are used to take into account the cross-section fixed effects. Estimates are based on quarterly OECD data. None of the values of the J test are significant at conventional levels of significance. The data come from EU15.

7 We have used both the June and December published OECD forecasts for the following year but the results are fairly similar in character.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

88 Asymmetry and Aggregation in the EU

here (Table 4.2). The model becomes roughly equally balanced between forward-looking and backward-looking but the output gap becomes significant and correctly signed once we restrict estimation to the monetary union era. However, simply assuming that that monetary union can best be represented by the period from 1999 onwards does not seem the best explanation. Taking the starting date backwards in time progressively (Table 4.3) suggests that the change in behaviour occurs around 1996. This break-point is more or less the same for the two data sets used in estimating the New Keynesian Phillips curve. Thus it is when the member states were trying to converge under Stage 2 of EMU that their behaviour changed and this change has continued thereafter. Prior 1995 or so, the New Keynesian Philips curves perform rather badly in the sense that, according to the coefficient estimates, inflation seemed to be more or less unrelated to the output gap (or other cyclical variables). Along with the EMU, the theory-consistent role of the output gap experienced a new come-back (Tables 4.2 and 4.4). This is not so much because of the output gap variable itself but because of the new role of inflation expectations. Before the EMU there was no genuine monetary-policy-anchored European view of future inflation developments. Nevertheless, there could be some more technical reasons for the observed pattern of results. After the early 1990s both inflation and inflation expectations have been stationary which makes estimation of Philips curves much easier, although there are no guarantees that the estimates do not represent some spurious correlations. For the data of the 1970s and 1980s inflation and inflation expectations seemed to have some trend while the output gap variable is ‘by construction’ a stationary variable (Baxter, 1994). Table 4.3

When did the EMU show up?

Quarterly data with GGM estimates Starting year 1999 1998 1997 1996 1995 1994 1993

Annual data with OECD forecasts

Coefficient of ∆p–1

Coefficient of gap

Coefficient of ∆p–1

Coefficient of gap

0.265 (65.08) 0.209 (28.37) 0.284 (29.30) 0.293 (21.65) 0.294 (7.15) 0.342 (13.76) 0.326 (8.87)

0.114 (5.39) 0.055 (2.08) 0.097 (7.32) 0.065 (3.42) 0.095 (2.09) 0.058 (1.38) 0.011 (0.15)

0.345 (3.65) 0.420 (4.56) 0.397 (4.83) 0.373 (4.47) 0.352 (4.31) 0.361 (5.26) 0.384 (6.09)

0.119 (2.03) 0.130 (2.20) 0.130 (2.29) 0.147 (2.76) 0.108 (1.89) 0.091 (1.76) 0.077 (1.58)

Selected parameter estimates for equation 4 in Tables 4.1 and 4.2. In all cases, the last period is 2006(Q4).

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

The Phillips Curve 89

90 Asymmetry and Aggregation in the EU Figure 4.3

Expected inflation

1.2

1.0

1.1

0.9

1.0

0.8

0.9

0.7

0.8 0.6

0.7

0.5

0.6 0.5

0.4

0.4

0.3

0.3

0.2 1985

1990

1995

2000

2005

Coefficient of expected inflation (June)

Table 4.4

1975

1980

1985

1990

1995

2000

2005

Coefficient of expected inflation (Dec.)

Estimates of a Phillips curve with the OECD forecast data

F1, 1980–1998 OLS F1, 1999–2006 OLS F1, 1980–1998 OLS F1, 1999–2006 OLS F1, 1980–1998 GLS F1, 1999–2006 GLS F1, 1980–1998 GMM-AB F1, 1999–2006 GMM-AB F2, 1976–1998 OLS F2, 1999–2006 OLS

∆p–1

∆pe+1

gap

R2/SEE

0.380 (7.45) 0.347 (3.55) 0.453 (8.98) 0.345 (3.65) 0.414 (10.73) 0.402 (5.34) 0.318 (4.92) 0.223 (2.74) 0.424 (8.21) 0.244 (3.16)

0.684 (11.53) 0.649 (5.78) 0.547

–0.002 (0.03) 0.121 (1.86) 0.023 (0.51) 0.119 (2.03) 0.042 (1.20) 0.118 (2.31) –0.072 (1.08) 0.228 (6.36) –0.26 (0.55) 0.088 (1.90)

0.944 1.376 0.600 0.835 0.938 1.441 0.600 0.831 0.949 1.165 0.702 0.831 .. 2.016 .. 1.187 0.933 1.566 0.706 0.716

0.655 0.629 (13.01) 0.604 (6.87) 0.707 (9.27) 0.622 (2.07) 0.618 (9.94) 0.762 (9.67)

DW/Jstatistic 2.287 1.760 2.250 1.761 2.167 1.935 .. 43.80 .. 25.49 2.493 1.909

F1 denotes the inflation forecast from the June forecast and F2 the inflation forecast from the December forecast. OLS denotes panel least squares estimates (no fixed effects) and GLS generalised panel least squares (with cross-section weights) estimates. In the GMM Arellano-Bond estimation, the set of additional instruments include both the lagged values of the right-hand side variables and lagged values of F2. The instrument rank with the J-test is 12. Thus, both J-statistics are significant although the one with the EMU sample has a marginal significance level over 1 per cent. In the equations on lines 3 and 4, the sum of inflation variable coefficients is set to one. The data are annual and consist of the EMU countries only.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

1980

One interesting feature is that the year 1999 creates some problems for all expectations-oriented Phillips curves (Figure 4.3). Thus, in the New Keynesian Phillips curve, the expectations channel appears to be temporary out of use for 1999 but start being operative since that. Price developments in 1999 were largely independent of the future inflation expectation. This could be interpreted in two ways: either there has been a lot of noise in inflation in 1999 (due to adoption of the euro) or there has been a lot of uncertainty in terms of future monetary policy and inflation regime. Perhaps this is all a coincidence but that would be surprising. Part of the problem in considering asymmetry is that the data period itself since the start of monetary union is short. To quite some extent the ‘flattening of the Phillips Curve’ represents the state of the economic cycle and a steeper segment started to be revealed as the recovery developed in the second half of the 2000s. Expectations formation certainly seems to have changed and people have become more forwardlooking. At the same time the distribution of behaviour among the various euro area countries has become smaller. Thus Europe is looking more like a single country than a group of different countries, even though there are still some striking differences. However the extent of the similarity should not be exaggerated. If we estimate the Phillips curve for the individual countries (Figure 4.4)

Figure 4.4

Coefficient of unemployment in a New Keynesian Phillips curve

0,1 0,05 0 –0,05 –0,1 –0,15 –0,2

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

The Phillips Curve 91

92 Asymmetry and Aggregation in the EU

Figure 4.5 Estimates of simple nonlinear Phillips curve for the EU for the pre-euro period –0.4

–0.2

0

0.2

0.4

0.5 Sweden

Italy Austria Denmark France Belgium UK Germany Netherlands Finland Portugal Spain

y>0

y<0

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

we get some outliers in the coefficients on the unemployment variable. Austria is quite striking and Italy even has a perverse coefficient. Among the more normal range the UK exhibits the largest effect but the range of the remainder is more than a factor of two, suggesting that what is done for a euro area-wide purpose may have effects that are somewhat inappropriate if the areas with the main problems do not lie close to the average. However, these results are without any regard to asymmetry. It is immediately clear from Figure 4.5 that at an individual country level, with the exception of Spain and Finland, the results conform to the expected asymmetry whichever estimation method is used. GLS and SUR make the picture rather clearer yet do not weaken the overall explanatory power. In each case the positive output gap shows a clearly positive relationship, while the negative output gap does not

appear to exert any significant influence on inflation either upwards or downwards.8 We now have a striking implication for policy. When the output gap is negative this will exert very little downward influence in its own right on inflation. Attempts to run the economy in an overexpansionary manner will on the other hand have substantial and quite rapid effects on inflation. There is therefore a strong incentive to avoid inflationary pressures taking hold. With this asymmetric model the costs of pursuing a price level as opposed to an inflation target could be considerable. If the actual relationship should be a curve and that there is unlikely to be any sharp regime shift around the zero gap then this model will tend to underestimate the importance of the output gap for small negative values and overestimate it for small positive values.9 Values nearer the original single line will tend to be most appropriate. At large negative and positive gaps the mis-estimation will be the other way round. The line will overestimate the importance of large negative output gaps and underestimate the importance of large positive gaps, possibly exponentially so, depending on the shape of the curve, as limits are likely to be approached in both dimensions. Countries with positive output gaps should have a much more important influence on monetary policy than those with negative gaps. Or turning the argument round, if policy is set symmetrically it will tend to have an inflationary bias (see Clark et al., 1996 for a clear description).

8 Estimation of specifications like (4.4) is quite straightforward but testing for the threshold is much more complicated, even though we treat the threshold value as a nuisance parameter (see Hansen (1999b) for details. In particular, in the case of heteroscedasticity, the conventional percentage points of the F distribution can be quite misleading. The choice of a zero output gap as the point round which to split the data is somewhat arbitrary, although by construction of the output gap variable this will be a split around the mean value. A grid search revealed that this value was only trivially different from the error minimising result. 9 Pyyhtiä (1999) using a similar model but with fewer countries and semi-annual data (without lags) obtains similar results for the pooled model. When the individual countries are estimated separately the pattern of the coefficients is similar in all cases, with positive gaps having a greater effect than negative gaps. Only in the case of Germany does the coefficient for the negative gap approach significance but the positive gap coefficients are not particularly strong except in the case of Italy. However, Pyyhtiä’s main focus is on a curvilinear specification, using a quadratic representation of the output gap. Adding the quadratic term improves the explanation for five out of the seven countries in the sample but the findings are relatively weak even in the pooled case. Mayes and Virén (2000b) also show examples of more explicitly curvilinear relations.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

The Phillips Curve 93

94 Asymmetry and Aggregation in the EU

• the shifts along the curves are expected to be substantial • the nonlinearity is considerable • the different countries have very different output gaps (their cycles are not well coordinated) • the individual country relationships are very different from each other. The asymmetries in the Phillips curve that we have explored appear to be primarily cyclical in character. Our analysis does not offer much scope for a discussion of the causes of asymmetry. In their tests of causes of asymmetry in the Phillips curve Dupasquier and Ricketts (1998) are able to isolate some evidence for the hypotheses of costly adjustment, capacity constraints and misperception (of aggregate and relative price shocks). The nominal wage resistance hypothesis was not obviously sustained, a result consistent with Yates (1998). Although to some extent these causes should be separable the results from their joint inclusion were not well determined. Eliasson’s (1999) finding that the Phillips curve, using unemployment and not an output gap as the determining variable, shows different sources of nonlinearity in Sweden and Australia is also helpful. In the Swedish case it is the rate of change of inflation expectations that is important, while for Australia it is the rate of change of

10 This is simplistic because the component economies interact, see Virén (2000b) for example.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

In general the contribution of large negative output gaps to holding inflation down will be overestimated and the contribution of high positive gaps to driving inflation up will be underestimated. However, this is assuming that there is a common relationship, which applies to all of the euro economies. There is considerable evidence that there are important differences in the transmission mechanism across the member states. Thus it is necessary to add not just results from different points on a nonlinear relationship but from different nonlinear relations. We thus need to consider where each of the countries is on its own curve and add together the change in inflation that would stem from the impact of the single monetary policy on each country’s output gap and then aggregate.10 From a practical policy point of view the use of a single linear relationship will only generate significant errors, if

unemployment.11 The former case will have particularly important implications for the conduct of monetary policy. Moreover the fact that the sources of nonlinearity differ for these two countries and are not found in the case of the US in contrast to Laxton et al. (1999) emphasises the potential problem of aggregation that we have outlined for the euro area. Yates (1998) questions the existence of downward rigidity in nominal prices and wages12 and hence one might wish to attribute our observed relationship to a different source. Yates himself points out that the shape may reflect the reaction of the authorities.13 There is certainly little reason to think that the euro area has suffered from a ‘lower bound’ problem in the period we are looking at.14 In general it tends to be assumed that the problems in the euro area will balance out because the gainers will offset the losers and indeed, in theory, could compensate them. However, if the behaviour of the economy is decidedly nonlinear then the problem of balance is no longer so straightforward and policy for the ‘average’ may turn out to be rather different from what would have been applied in each of the member states independently. As Laxton et al. (1995) point out if the Phillips curve is indeed a curve rather than a straight line, policy will have to be run tighter on average to achieve any given price stability target. If we consider the implications for policy at a national rather than euro area level, then countries with positive output gaps should have a much more important influence on monetary policy than those with negative gaps. Or turning the argument round, if policy is set symmetrically it will tend to have an inflationary bias (see Clark et al. (1996) for a clear description). Using the very simplified example shown in Figure 4.6 it is very obvious how ignoring asymmetry and aggregation problems could have an unfortunate effect for policy. Assume first of

11

Buxton and Mayes (1986) also made this finding for the importance of the rate of change of unemployment in the case of the UK. 12 Real rigidities can, however, be overcome even in the case of downward nominal rigidity through changes in the exchange rate. 13 Yates (1998) generally finds that the Phillips curves he estimates are not very well determined even before trying to add evidence of nonlinearity. While the nonlinearity may have the expected sign in many of his estimates it is not normally significant despite specification searches. 14 These problems are extensively discussed in the papers presented at the conference on ‘Monetary Policy in a Low Inflation Environment’, Federal Reserve Bank of Boston, October 18–20, 1999.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

The Phillips Curve 95

96 Asymmetry and Aggregation in the EU Figure 4.6 policy

Linearity and nonlinearity in the Phillips curve and the setting of

inflation

A

C

∆p1

target

B

D

gap

output gap/unemployment

all that the relationship between inflation and the output gap is as shown by the curve in the Figure. Then simple arithmetic aggregation of forecasts of the output gap for two countries/regions/industries, which generate two expected outcomes, one at A and the other at B, will give a result such as ‘gap’ shown on the horizontal axis (even if weights are used). Assuming the relationship is a straight line will result in forecast inflation being ∆p1 rather than the appropriate value ∆p2. Under an inflation targeting regime this will tend to mean that the policy response will be rather harsher under the assumption of a linear relationship than it should be. Indeed, in the case illustrated, the correct policy decision would be to ease while the actual decision, wrongly assuming linearity, would be to tighten. We have chosen the deliberately simplified case where both A and B are on the linear as well as the curvilinear relationships. In general the contribution of large negative output gaps to holding inflation down will be overestimated and the contribution of high positive gaps to driving inflation up will be underestimated. However, this is assuming that there is a common relationship, which applies all of the euro economies. There is considerable evidence that there are important

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

∆p 2

The Phillips Curve 97

differences in the transmission mechanism across the member states. As a result one is not only trying to add inputs where the relationships are nonlinear but where the nonlinearity itself varies. The nature of the nonlinearities is not surprising: in good times inflation is much more sensitive to output than in bad times. That could be a reflection of old-fashioned downward price (wage) rigidity although we are obviously not working with the price (wage) level but with the change rates of these variables.

Forms of real time information15

We noted earlier that it is important to ensure that the data used in estimation provide a realistic representation of the information that people had to hand in taking decisions. This is usually referred to as real time data, although this expression is sometimes also used for information that is continuously updated, such as some market data. Thus far we have simply used the latest OECD data available at the time of estimation, which incorporates all the revisions that become available since the preliminary estimates became available. We have therefore reconsidered some of our estimation and replaced the data by their real time equivalents. The first of the three sources of real time information that we explore relates directly to expectations. A common approach is to assume rational expectations and try to model expectations directly from the model. Rational expectations are normally expressed, however, in terms of the most up-to-date information. A construct based on the information available at the time could be made ‘model consistent’ but strictly rational expectations would imply that they were ‘correct’ not just that they conformed to a specific less-revised dataset. In any case, not only does the rational expectations assumption impose substantial problems for estimation (Rudd and Whelan, 2001) but it perpetuates the problem of handling an unobservable (two actually, since the output gap is also unobservable). We therefore consider a less ambitious assumption and employ a direct measure of expectations as a means of getting at what people thought at the time from the information available to them. We

15

These sections on the use of real time information relate to work done with Maritta Paloviita, whose essential contribution is gratefully acknowledged. The real time data set was put together by Heli Tikkunen and is available from Deutsche Bundesbank.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

4.4

use published OECD forecasts (Pyyhtiä, 1999). These forecasts were generally available at the time pricing decisions were taken. While there is no particular reason to suppose that the OECD represented general beliefs, such forecasts were widely discussed and respected. More importantly from the point of view of our analysis, they are produced by a coherent methodology that is applied to each of the euro area countries and evolves only slowly across time. There is nothing similar available with such a coverage. Even so with only annual data stretching over the period 1977–2003, this is a very limited sample to operate on. We have therefore chosen to pool the data and estimate the model in panel form, which gives us a maximum around 300 observations, depending on the exact specification.16 By using direct measures of inflation expectations, we can avoid the problem faced by many previous studies of inflation dynamics, of having to test dual hypotheses, about the specification of the Phillips curve and the formation of expectations, at the same time.17 Thus, in our study we can allow for the possibility that the expectations themselves may adjust slowly or move for spurious reasons. A simple form of explanation would be to use one of the specifications of least squares or other learning processes (Evans and Honkapohja, 2002). The OECD forecasts are likely to be more reliable proxies for inflation expectations than some survey estimates that have been used, as they are based on systematic monitoring of economic developments and econometric models. The OECD’s forecasts are produced twice a year and published in June and December. The June forecasts are normally for the current and the next calendar year, while a second future year has been added in December, in recent years. They cover, inter alia, inflation in both the GDP and consumer price deflators. OECD’s database is quarterly, so it would be possible to compile semi-annual series for all the variables and estimate the models on that frequency. One can also interpolate the series of forecasts and hence estimate the models at quarterly

16

Not all series are of equal length and the availability for particular countries varies slightly. It is, however, the forecast information that starts in 1977 for ten countries in the euro area. For Luxembourg, the forecasts are available since 1982 and for Portugal since 1980. We can and do go back earlier to 1960 with the historical series published by the OECD since 1977, particular in the case of real GDP, when estimating output gaps. 17 Roberts (1997, 1998) and Adam and Padula (2003) provide similar studies with survey-based expectations for the US.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

98 Asymmetry and Aggregation in the EU

frequency.18 However, we stick with the annual information. The timing of the forecasts raises the first question about what real time constitutes. Pricing decisions that affect both deflators will be taking place during much of every working day and probably outside them as well. The annual outcome is the result of a mass of decisions spread, unevenly over the year. There are some important elements of bunching in the early part of the year, both with administered prices and wage-setting in many euro area countries over the period. This might argue that the December forecasts were more typical of the information available. On the other hand the June forecast coincides with the publication of the first estimates of the outcome for the previous year, so perhaps this has more merit. We explore both but focus on the December forecasts because they look slightly further ahead than those in June.19 One advantage of using OECD forecasts is that the self-same publication the OECD Economic Outlook, produces compatible data series for the history of the variables in the model and estimates of their current value.20 Since we are dealing with estimates made in December for the current year, they still contain an element of forecasting. This emphasises a general problem in estimation in that reliable official estimates may only be available with a considerable lag. The first published vintage of the data for a particular year is not really ‘real time’, as it appears

18

Normally interpolation is done with some reluctance because of the effect it has on the dynamics of the relationship. In this case it might actually be desirable because the OECD forecasts are only a proxy and some smoothing of their impact might be appropriate. We only take them as representative of a more general view, not that their publication constitutes ‘news’ on which behaviour would change. 19 These differences in horizon and information base pose problems for a semiannual approach. Not only will the timeliness of the published information available alternate between the June and December OECD estimates but the length of the forecast horizon will also vary by six months. 20 Prior to 1985 (1983 for France, Germany and Italy) Economic Outlook did not contain estimates of inflation two periods earlier. These real time estimates are needed for the instrument set. We therefore used the nearest estimate in time published in the OECD National Accounts. The decision over which year’s estimates to use was based on the degree of correlation between the National Accounts and the Economic Outlook estimates in the years from 1985 onwards where we had both sets of estimates. This was done country by country, as the lag in information provided to the OECD by national statistical authorities varies. For five countries the current year National Accounts were used and the next year’s for the remainder. While this muddies the definition of real time, the effect is likely to be small.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

The Phillips Curve 99

after the decisions have taken.21 We explore using the OECD’s own ‘real time’ estimates of the output gap published in Economic Outlook as well, so that the entire model is expressed only in terms of the information actually used at the time price setting decisions are made. But they are only available for a few years, so it is also necessary to estimate the gap using relatively robust methods to represent what could been done over the period with the real time data and the techniques then available. The second element of real time information we consider relates to the data set used in constructing the output gap. If we use up-to-date information we actually know with the benefit of hindsight what happened to output in subsequent periods and hence can avoid the well known end point problem. However, at the time people cannot avoid the problem. They have to make judgements about how appropriate trend values should be estimated and as Orphanides (2001) has shown this can help explain some large policy errors. HP filters are particularly subject to this difficulty and it would be very helpful if we could use a different form of estimation, say, the production function approach that the OECD uses. It is arguable (Neiss and Nelson, 2002; Robinson et al., 2003; Orphanides and van Norden, 2002) that estimating the output gap will dominate the problems that people faced at the time from having to use real time data. However, using more sophisticated methods would not replicate what people might reasonably have done at the time.22 It is particularly unfortunate therefore that these potentially less contaminated estimates of the output gap by the OECD only stretch as far back as 1994. We are therefore compelled to use the HP filter or similar rather deficient methods if we want to consider the whole data period. Although we can use the full extent of the OECD output forecasts available in calculating the filter, we need to use real time data in estimating the output gaps as they would have been seen at the time.23

21

This would not be such a problem with a backward-looking specification or higher frequency model, if data are published quickly. As it is, the current year ‘estimate’ will be based on initial published data for the first part of the year, estimates of related and indicator variables for the middle part of the year and forecasts combining backward and forward-looking information for the last few months of the year. 22 Using ‘one-sided’ filters may reduce the problem. 23 There is clearly a trade off here between considering robust methods of estimation using real time data that might have been more in line with contemporary estimates and using more reliable estimates. The difference between the two may help to explain policy errors.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

100 Asymmetry and Aggregation in the EU

Lastly in estimation, we apply real time data in the GMM estimation process. Here the question of what data set should be used is more contentious. GMM is a statistical technique. Appropriate instruments need to be predetermined and correlated with the variables they seek to explain but uncorrelated with the error term. Using GMM does not per se involve the question of what information was available at the time. It might however, seem more logical to use a common data set so that the instruments are also drawn from the data that were available at the time. By using more up-to-date information in one part of the estimation process than in another we may introduce spurious correlations. The issue of simultaneity would be represented by using the same real time data set even if it might be statistically easier to handle it with different information. In this sort of context one of the functions of GMM is to help clear up an ‘errors in variables’ problem. If we assume that the final estimates are more accurate then inevitably the real time estimates must include an error. We conduct some limited tests for bias to see if we can get a prima facie indication. We thus have quite a complex database that contains a series for each variable in the model every year. Since we have used 26 issues of the OECD Economic Outlook annually from 1977 to 2002 we have 26 sets of series on each variable.24 Thus the real time data for a variable x in period t consist of series running from the first year recorded, 1960 in most cases, through to t + 2, i.e t x1960 + τ , τ = 0, …, t + 2; t = 1977, … , 2002. The observations from 1960 to t – 1 will be published ‘data’, t will be an ‘estimate’25 and t + 1 and t + 2, are forecasts, all published by the OECD in December of year t. These then have to be placed into the appropriate series for estimation. Real time forecasts made in year t for year t + 1 are thus denoted t xt + 1 , real time lagged values are t xt – l , where l is the lag, and forecasts made last year for this year are t – 1 xt . Thus there is always a contrast between real time and the most recently published estimates. However, for the last data point, 2002, the most recent data have not as yet been revised. Since many of the main revisions occur early in the first year or two, we could end the real time data earlier by eliminating the most recent observations if we

24

We have a 27th set of series from the December 2003 Economic Outlook, which is the source for our most recent revised data. The last complete year is thus 2002 as 2003 was not yet over in December. Hence the 2003 real time estimates cannot be used as they have no ‘actual’ value against which they can be compared. 25 They are all of course estimates in the sense that we never know the true values.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

The Phillips Curve 101

wished to increase the potential difference between the last real time observation and the most recently published revised estimate. How many years we should omit in this way is fairly arbitrary unless we could reach a point where the data are not further revised. Since that involves knowledge of what the statisticians at the OECD might do in future revisions, which they themselves do not know, there can be no ‘right’ answer. The more periods we omit the poorer our explanation of the Phillips curve is likely to be. There is thus a trade off. We can gain some insight over the appropriate choice from the pattern of previous data revision by the OECD. There are typically two sorts of revisions to the OECD data.26 In the first few periods there may be fairly substantial revisions and then at less frequent intervals there are comprehensive revisions to the series over quite a long time period, usually coinciding with rebasing, particularly for constant prices. This second type of revision tends to shift the series as a whole rather than simply individual observations. This difference is important in context of the Phillips curve, as variables are expressed either in rates of change or compared to some form of ‘trend’. Shifting a series may have little effect on rates of change but it can alter Figure 4.7

Real time OECD output gap estimates 1994–2002 Spain

6 4 2 0 –2 –4 –6

2002 1997

2001 1996

2000 1995

1999 1994

02 20

20 00

19 98

19 96

19 94

19 92

19 90

19 88

19 86

19 84

19 82

19 80

–8

1998

Note: Each line in the graph shows the estimates of the output gap published in OECD Economic Outlook in December.

26 See Paloviita and Mayes (2006) for details on the impact of revisions for each of the four largest euro area economies.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

102 Asymmetry and Aggregation in the EU

the complexion of deviations from trend, particularly where there are nonlinearities or asymmetries. It is noticeable that the revisions have typically been greatest round the turning points. Since turning points are also associated with forecast errors, this has the potential for even larger real time discrepancies. It is also observable that there can be noticeable changes even ten years or more after the event. The second issue can matter much more for the output gap, as it is a derived measure and not just a published series. The case of Spain shown in Figure 4.7 is particularly striking. While the shape of the output gap does not change a lot, where it is pitched can. The revision between 1999 and 2000 is particularly large but its greatest effect is not on the immediate period but on the estimates of the fairly recent past. The profile of the data has been shifted rather than individual observations. To some extent this represents an upward revision of the underlying rate of growth. It is fairly obvious that the inflation series are not subject to substantial revision, Table 4.5.27 The real time series typically show correlation coefficients of 0.95 or better with both the revised estimates and with the forecasts.28 In the case of the output gap however, we are looking at markedly different series, with correlation coefficients between 0.58 and 0.77 for the real time and revised series. Not surprisingly the two estimation methods are somewhat more closely correlated for the same data, 0.86 to 0.88, but still less than the correlations between real time and revised price series. We have also checked to see whether the discrepancies appear to be biased (Table 4.5c). Simple Wald tests comparing the real time and revised estimates suggest there are consistent differences between the two series in most cases. The clear exception is the real time GDP deflator. The nature of the discrepancy varies from case to case. The real time HP filter estimate of the output gap is on average about half of one percentage point below the estimates from the most recent data. We had anticipated that the end point problem would bias its absolute value towards zero, not this asymmetric bias. Real time consumer price inflation tends to underestimate the revised series. The OECD’s output

27

We explore both the GDP deflator and the private consumption deflator. The hypothesis of no correlation is rejected at least at the 5 per cent level in all cases except that between the revised OECD and real time HP-filtered estimates of the output gap in Table 4.5b where the probability is slightly above 5 per cent.

28

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

The Phillips Curve 103

104 Asymmetry and Aggregation in the EU Table 4.5

Correlations and test for unbiasedness

GDP deflator

Revised

Forecast

Real time estimate

1 0.953 0.976

0.953 1 0.963

0.976 0.963 1

Revised

Forecast

Real time estimate

Revised Forecast Real time estimate

1 0.955 0.991

0.955 1 0.951

0.991 0.951 1

Output gap

Real time HP filtered

Revised HP filtered

Revised OECD estimate

1 0.604 0.577

0.604 1 0.859

0.577 0.859 1

Revised Forecast Real time estimate CP deflator*

Real time HP filtered Revised HP filtered Revised OECD estimate

*In all the tables CP denotes private consumption.

(b) Output gap correlations 1994–2002 Output gap

Real time HP filtered Revised HP filtered Real time OECD estimate Revised OECD estimate

Real time HP filtered

Revised HP filtered

Real time OECD estimate

Revised OECD estimate

1 0.679 0.873 0.627

0.679 1 0.746 0.881

0.873 0.746 1 0.769

0.627 0.881 0.769 1

continues on facing page

gap estimates, using the production function approach, have an average value nearly 0.4 of a percentage point lower and are poorly correlated.29 There is one correlated item in the revisions. Since real GDP is deflated nominal GDP and the GDP deflator is one of the inflation measures we use in the study, revisions to real GDP could come from one or both

29

We have checked the data for stationarity, as its absence would affect the validity of the inferences. The downward trend in inflation in the first part of the period for many of the countries poses an obvious problem.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

(a) Correlations 1977–2002

The Phillips Curve 105 Table 4.5

Correlations and test for unbiasedness – continued

(c) Unbiasedness 1977–2002 (1994–2002 for OECD output gap) Rows show estimates from equations of the form, xt = a + bxt* where x* is the variable shown in column 1 and x is the most recent revised data for the same variable. The Wald test is of the joint hypothesis a = 0 and b = 1, which is asymptotically distributed as χ2(2) under the null. Wald test

Real time GDP deflator Real time CP deflator Real time HP filtered output gap Real time OECD output gap GDP deflator forecast CP deflator forecast

a

s.e.

b

s.e.

0.204 11.231 17.243

0.903 0.004 0.0002

0.038 0.176 0.496

0.106 0.999 0.013 0.063 0.975 0.008 0.122 1.038 0.083

4.137

0.126

0.365

0.185 1.133 0.096

11.751 10.537

0.003 0.005

0.007 0.104

0.149 1.043 0.019 0.145 1.029 0.018

In all Tables, all euro area countries are included: Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain but the sample starts for Portugal only in 1980 and Luxembourg in 1982, so for the full data period it is slightly unbalanced. ‘Revised’ indicates data as published in December 2003 OECD Economic Outlook, ‘forecast’ is forecast published in December OECD Economic Outlook of each year for the following calendar year, ‘real time’ indicates estimates of current year published in each December OECD Economic Outlook, with the exception of the HP-filtered output gap which is computed separately for each year by the authors from the entire real GDP series published in the December OECD Economic Outlook, including all past years, the estimate of the current year and the forecasts of the next two years. Real time estimates of lagged values, used in subsequent Tables are each drawn from the same December issue of OECD Economic Outlook as the current year estimate. They are not the series of current year real time estimates with lags applied. Prior to 1985 (1983 for France, Germany and Italy) the real time estimate of inflation two periods earlier is not available from Economic Outlook and is obtained from the issue of OECD National Accounts for each country in each year which most closely matches the December Economic Outlook.

of two sources. Nominal GDP and/or the GDP deflator may have been revised. Thus there will tend to be some inverse correlation between revisions of real GDP and the GDP deflator. The change to the output gap, which is derived from the GDP series will be at one remove. Since the output gap for a single year is not dependent on GDP in just one year, it is not possible to go on to argue that revisions in the output gap and in the GDP deflator are therefore also likely to be correlated but it remains a possibility. Insofar as such correlations do exist they can affect the extent of the change in the estimates from using real time data.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Chi-square Probability

106 Asymmetry and Aggregation in the EU

Real time estimation

We face the normal problems in constructing the output gap and use an HP filter, not because it is obviously best but because it was the most widely used approach and does not involve further data series.30 We follow the common procedure of using forecast values of real GDP to construct the filter in order to reduce the impact of the end point problem. This only has to be done once with the revised data set. However, if we want real time output gaps we have to construct them from each data set in turn. Thus in period t, computing the output gap entails using the December year t OECD Economic Outlook to provide the most up-to-date estimates of real GDP in previous years, the estimate of year t and the forecasts of year t + 1, and t + 2 where it is available. All these estimates of the year t output gap, one from each December’s Economic Outlook, have to be transcribed into the single output gap series for estimation. When using the OECD’s own published estimates of the output gap, which use a production function and not an HP filter, they are treated just the same way as the most recent and real time series for the inflation variable.31 The evolution of each individual computation of the real time output gaps is illustrated in Figure 4.8 for Italy, as this shows the largest revisions of the four main euro area countries.32 The first real time gap is thus computed for 1977, the beginning of our forecast sample, using the December 1977 vintage data including its forecasts. This line has its end point in 1978. There is then a new line superimposed for each succeeding year, all of them stretching back to 1960, which is our origin year for the data. The most heavily revised observations are the forecasts for the two years ahead, which are not used in estimating the Phillips curve itself but give the full flavour of how sensitive output gaps are to the end point

30

As Rünstler (2002) has shown for the euro area, Orphanides and van Norden (2002) for the US and Cayen and van Norden (2002) for Canada, Nelson and Nikolov (2001) for the UK and Gruen et al. (2002) for Australia, measures of the output gap can vary widely according to the method used. Furthermore, as the output gap is a generated regressor it could present problems for inference (Pagan, 1984) despite the use of GMM. 31 The OECD has only published its own estimates of the output gap since 1994, hence the need to present the correlations of these with other measures separately in Table 4.5b. 32 Clearly in forming a judgement it was necessary to explore the patterns for all countries in the sample and not just for the four largest, although their experience will dominate the aggregate result.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

4.5

The Phillips Curve 107 Figure 4.8

Real time HP filtered output gaps Italy

4 3 2 1 0 –1 –2 –3

GIT77 GIT84

GIT78 GIT85

GIT79 GIT86

GIT80 GIT87

GIT81 GIT88

GIT82 GIT89

GIT83 GIT90

GIT91 GIT98

GIT92 GIT99

GIT93 GIT00

GIT94 GIT01

GIT95 GIT02

GIT96

GIT97

GITnn stands for ‘gap’ for Italy in year nn, which runs from 1977 to 2002.

problem.33 The extent of the revision in the output gaps used is clearer if all of the other observations are removed from the chart and only the sequence of real time gaps, without the history are shown, as in Figure 4.9, by comparison with the HP filtered gaps estimated using the most recent, December 2003, data. The deflators tend to show quite negligible differences by comparison, as can be anticipated from the high correlations in Table 4.5a. Our main results focus on the Hybrid model as this gives a more comprehensive opportunity to consider how forward-looking expectation formation appears to be. Here we constrain the coefficients on the backward and forward-looking expectation terms to sum to unity, giving us (4.3) for estimation. The restriction is not confirmed by the data but the impact is quite small.34 We used two inflation measures 33

The problem might be reduced by using a one-sided filter but, as the real time gaps produced by the OECD using the production function method show, real time and revised gaps vary considerably, whatever the method used in real time. 34 The sum of the unrestricted coefficients when real time data are used throughout is slightly greater than unity (1.05 and 1.07 for the GDP and CP deflators) and the result is an increase in the relative importance of the forward-looking component. The output gap coefficients become positive but insignificantly so.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

19 60 19 62 19 64 19 66 19 68 19 70 19 72 19 74 19 76 19 78 19 80 19 82 19 84 19 86 19 88 19 90 19 92 19 94 19 96 19 98 20 00 20 02 20 04

–4

108 Asymmetry and Aggregation in the EU Figure 4.9

Real time and revised HP filtered output gaps Italy

5 4 3 2 1 0 –1 –2 –3 20 01

19 97 19 99

19 95

Revised

in estimation: the annual changes of the GDP deflator and the private consumption deflator, because both measures are widely used in the existing literature. Although the two series are strongly correlated, they show noticeable differences in estimation. Despite the rather wide range of results shown for individual countries in Paloviita and Mayes (2005) the restrictions entailed in pooling the data are not rejected for the GDP deflator.35 In the case of the private consumption deflator the pooling restrictions are not rejected for ten countries, including the six largest. Only Austria and Finland fail to meet the criterion and they represent only 5 per cent of euro area GDP between them. The impact of omitting these two countries from the estimation is small so only the results using the full data set are shown.

35

In pooling we are treating all countries as if they were equally important. However, for some purposes it would be more relevant to weight the countries by their economic size so we have also used weighted regression to approximate what might apply across the euro area and in Paloviita and Mayes (2005) we estimated Phillips curves at the euro area level of aggregation. In this case much of the euro area information is synthetic. Not only did stage 3 of EMU not start until 1999 with the common currency only in the last year of our sample but in the early years it was not even in prospect. Five of the subsequent members were not even in the EU in that period. Using mean GDP weights on the real time data has the effect of reducing the forward-looking weight by around 10 per cent and removing (substantially reducing) the negative output gap coefficient for the GDP (CP) deflator equation.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Real time

19 93

19 89 19 91

19 87

19 85

19 83

19 79 19 81

19 77

–4

It is immediately obvious from Table 4.6, using the maximum data set available, that the balance of expectations formation falls slightly in the forward-looking direction.36 The successive rows, 1–4 and 5–8, show the effect of adding more real time information, for each of the GDP deflator (GDP) and consumers’ expenditure deflator (CP) measures of inflation. Rows 1 and 5, which provide the starting point, with just the OECD forecasts included as the measure of expectations can be contrasted with rows 9 and 11 which show the effect of estimating the model using the actual outcome the following year, on the basis of the most recently revised data (December 2003 Economic Outlook).37 The difference is surprisingly small despite the relatively low accuracy of the forecasts recorded in Table 4.5a. Adding the real time estimate of lagged inflation makes relatively little difference but using our constructed real time estimate of the output gap with an HP filter leads to the well-known problem discussed above of obtaining a wrongsigned coefficient (Galí and Gertler, 1999). Given the rather poor determination of the output gap coefficients in any case, this should perhaps be no surprise. Expressing current inflation in real time terms, which is also an OECD forecast in that it is the estimate of the current year published in December but in effect based on only two quarters official estimates, increases the forward-looking weight considerably. In the consumers’ expenditure case the forward-looking weight is now twice the backward-looking weight. As each item of real time information is added to the picture so the forward-looking component increases in importance. To some extent price setters appear to be able to take account of information that was not in the currently published data but was incorporated in the revised information after the event. As we noted, it is unfortunate that we have to estimate a rather crude real time measure of the output gap. Constructing some more elaborate multivariate estimate using real time data would increase the scale of the exercise substantially. While the OECD itself has computed estimates of the output gap using the production function method, these are only available in real time, i.e. published in Economic Outlook, since 1994. The result is a heavily truncated sample of only 99 observations

36

In Table 4.5b we have used OECD inflation forecasts since 1977 with the exception of Luxembourg and Portugal, where forecasts are only available from 1982 and 1980, respectively. This gives a total of 304 observations and not the 312 that would stem from a full balanced panel. 37 Rows 10 and 12 show estimates using a second lag on inflation and a lag on the output gap as instruments.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

The Phillips Curve 109

110 Asymmetry and Aggregation in the EU

Model

θ

s.e.

ψ

s.e.

DW

SEE

R2

1 GDP, exp

0.557

0.03

0.014

0.03

2.39

1.459

0.934

2 GDP, exp, plag-realt

0.551

0.04

0.018

0.04

2.04

1.465

0.933

3 GDP, exp, plag-realt realtgap

0.560

0.03

–0.028

0.06

2.04

1.465

0.933

4 realtGDP, exp, plag-realt realtgap

0.602

0.02

–0.098

0.03

2.13

1.107

0.960

5 CP, exp

0.567

0.02

0.070

0.03

1.95

1.100

0.962

6 CP, exp, plag-realt

0.584

0.03

0.064

0.03

1.74

1.167

0.957

7 CP, exp, plag-realt realtgap

0.613

0.03

–0.044

0.04

1.67

1.176

0.957

8 realtCP, exp, plag-realt realtgap

0.672

0.02

–0.138

0.03

1.84

1.146

0.960

0.927

9 GDP, plead

0.527

0.02

0.007

0.02

2.99

1.530

10 GDP, plead, 2sls

0.496

0.06

0.109

0.03

2.95

1.550

11 CP, plead,

0.517

0.01

0.011

0.02

2.43

1.190

12 CP, plead, 2sls

0.522

0.04

0.116

0.04

2.36

1.220

0.956

GDP: GDP deflator; CP: private consumption deflator. The following notation explains which series have been used in the model – exp: OECD forecast of inflation; plag-realt: real time inflation for previous year; plead: most recent estimate of inflation in next year; plag: most recent estimate of inflation in previous year; realtgap: real time output gap estimates; realtinstr: real time instruments in GMM; realtGDP, real time GDP deflator; realtCP: real time estimate of private consumption deflator.

(Table 4.7). They have calculated output gaps using that method back to the beginning of our sample period but that uses revised data. In this case the weights are slightly different with the forward-looking element in the consumers’ expenditure deflator case being only a little above half while the GDP deflator sample gives a weight of two-thirds and above. Both are notably higher than what is observed if we use the most recent revised data. This is, of course, not a matched comparison as

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 4.6 Estimates of restricted Hybrid model with HP filtered output gap, 1977–2006 (LS with Newey-West correction, 304 observations)

The Phillips Curve 111 Table 4.7 Estimates of restricted Hybrid model, 1994–2006 (LS with NeweyWest correction, 99 observations)

(a) using OECD estimates of the output gap 1 realtGDP, exp, plag-realt, realtgap 2 realtCP, exp, plag-realt, realtgap 3 GDP, plead, 2sls 4 CP, plead, 2sls (b) using HP filtered output gap 5 realtGDP, exp, plag-realt realtgap 6 realtCP, exp, plag-realt realtgap

θ

s.e.

ψ

s.e.

DW

SEE

R2

0.733

0.04 –0.020

0.02

2.41

0.589 0.896

0.582

0.04

0.071

0.02

2.08

0.547 0.902

0.466 0.475

0.06 0.03

0.088 0.086

0.02 0.01

3.13 2.60

1.084 0.594

0.710

0.02

0.035

0.03

2.41

0.589 0.896

0.585

0.03

0.142

0.03

2.08

0.535 0.906

See notes to Table 4.6.

the sample in Table 4.6 is much longer. However, if we use the shorter sample with the HP-filtered estimates (Table 4.7b), the forward-looking weight is very similar to those when the OECD output gap estimates are used. The output gap coefficients are also positive. There is therefore some difference in behaviour in the two data periods. Inflation has been clearly lower since 1994 and hence in some senses easier to predict. However, it has also become more persistent, so it is not immediately obvious what the effect of this would be on the resultant estimates. Nevertheless it remains that real time data are able if anything to explain inflation a little better and have a noticeably larger forward-looking element in the explanation, in no case lower than the backward-looking weight. We now move on to consider the use of real time data for the instruments in GMM estimation. This aspect can be examined using a database that contains real time variables, not just for current values but also lagged information that was available at the time. When real time information is used in the expectations variable, it is logical to choose a common data set so that the instruments are also what were available at the time instead of final variables. As Orphanides (2001) points out, decision-makers have to use noisy data without knowing what the noise is. If we use instruments without the noise then they may be

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Model

112 Asymmetry and Aggregation in the EU Table 4.8 Estimates of the hybrid model, GMM, 1977–2006 (304 observations) Model

θ

s.e.

ψ

s.e.

J stat

p value

DW

SEE

1

GDP, plead

0.50

0.06

0.108

0.033

0.000

0.850

2.949 1.549

2

CP, plead

0.46

0.05

0.125

0.035

0.013

0.048

2.390 1.219

3

realtGDP, exp, 0.30 plag-realt, realt-instr, realtgap

0.10

0.216

0.094

0.003

0.321

1.922 1.379

4

realtCP, exp, plag-realt realt-instr, realtgap

0.12

0.359

0.138

0.008

0.118

1.680 1.570

0.23

See notes to Table 4.6. Instruments: second lag of inflation and two lags of the output gap (revised data, rows 1 and 2, real time, rows 3 and 4). The standard errors of the estimated parameters were modified using a Bartlett kernel with fixed bandwidth (without prewhitening). In all cases, the Hansen test (J statistic) of the overidentifying restrictions of the model was used (Hansen, 1982).

correlated with the errors. They will also not be so well correlated with the omitted relationship, such as the setting of monetary policy. In the Hybrid model, Table 4.8, using two lags of the output gap and the second lag on inflation as instruments, the immediate effect is to reduce the forward-looking weight considerably and steepen the slope of the Phillips curve.38 In commenting on an earlier version of the chapter, Orphanides argued that this is exactly what one would expect if monetary policy is also forward-looking and is captured by expectations. Since there is some doubt whether the normalisation used is the most suitable (Søndergaard, 2003) this may help explain the higher backward-looking weight, although Jondeau and Le Bihan (2003) argue that the bias in using GMM may be in the opposite direction. In any case there will be a degree of persistence observed in the data even if the decision-making process itself is entirely forward-looking (Goodfriend and King, 2001). Hence, it is important not to misinterpret the implications of empirical lags as suggesting that a less forward-looking monetary policy should be employed. Overall, what is clear is that using real time instruments does have an effect. It would be inappropriate to ignore the appropriate choice

38

This is a considerably smaller instrument set than used in Søndergaard (2003) or Galí and Gertler (1999) both in terms of range of variables and number of lags.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Row

of real time as opposed to most recent data for instruments as it has a material effect on the estimates. A number of conclusions can be drawn on the basis of this analysis. First of all, using real time information on expectations, in the form of forecasts – in this case those published by the OECD – does seem to act as an improvement over some simple adaptive or rational expectations approaches in estimating Phillips curves for the euro area countries in the period since the mid-1970s. Second, using real time data in the model offers a marginal improvement to the explanation, although the principal means of estimating the output gap used, namely, an HP filter creates difficulties of its own, not least through the end point problem. If we tackle the problem by using the OECD’s own estimates of the output gap, which are available for only a short period, the results are very similar to those when the HP filter is applied. However, this period is characterised by low inflation so this may reflect the time period rather than the method of estimating the output gap. The most striking result, however, is that using real time data increases the apparent forward-looking weight as indicated by the Hybrid model. This confirms the results found for other countries, Orphanides (2001) for the United States and Huang et al. (2001) for New Zealand for example. In real time people do try to take into account other information about what is happening and likely to occur, which is not in the currently published statistics. After the event those statistics themselves can be revised as some of that extra information is revealed and any inconsistencies in the series become apparent. Thus using revised data the forwardlooking element will be reduced. Using real time instruments in GMM estimation instead of instruments based on revised data also has a clear impact on the structure of the Phillips curve. It lowers the weight on forward-looking expectations but gives higher and clearly positive output gap coefficients. In real time decision-makers have to use noisy information without knowing what the noise is. If the instruments are revised to omit the noise they will be correlated with the errors and not so well correlated with omitted relationships like monetary policy that also have to rely on the noisy data. The use of real time information in the Phillips curve confirms that the timing of expectations formation matters in inflation dynamics and that the euro area inflation process is not purely forward-looking. Similar to the experience for the US obtained by Adam and Padula (2003), where real time data from surveys are used to measure expectations, the

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

The Phillips Curve 113

114 Asymmetry and Aggregation in the EU

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Hybrid model with a lagged inflation term is needed to model inflation dynamics properly. Although the estimation results are sensitive to the choice of the forcing variable and the output gaps, based on HP filtering, suffer from end point problems, we can say that the use of real time information makes a noticeable difference when explaining inflation dynamics. Revisions in this data set, even in the price series, are sufficient to matter.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

5

In the previous chapter on the Phillips curve we indicated that some of the problem came from the focus on aggregate unemployment without regard to its distribution over the regions of Europe. In this chapter we tackle that issue directly and whether the dispersion of output growth or unemployment rates within a country has a direct effect on the determination of inflation for the country as a whole in the context of a nonlinear relationship. This discussion is by no means the first in this area. Nonlinearity of the Phillips curve has been tested in numerous analyses (see Laxton et al. (1995), Laxton et al. (1999) and Linzert (2005) among others). The whole issue itself is also quite old. Lipsey (1960) remarks (p. 19) ‘If one wishes to predict the rate of change of money wage rates, it is necessary to know not only the level of unemployment but also its distribution between the various markets of the economy.’ (emphasis in original). While Lipsey does not attempt any estimates, Archibald (1969) offers some for the UK, where the variance of both regional and industry unemployment are shown to have a positive effect on wage inflation. Extending this to the US gives more problematic results with quarterly data.1 However, it is Brechling (1973) who introduced the nonlinear aggregation hypothesis which basically formulates the problem and suggests ways of testing the proposition. He also carried out some empirical tests with the US data. The results of the tests were somewhat disappointing from the hypothesis’ point of view

1 Somewhat later, when quarterly data were available for an adequate period for the UK, Buxton and Mayes (1986) show that it is unemployment in the tightest labour markets that plays a key role in determining inflation and that both slacker labour market regions and the long-term unemployed have much more limited effect in the Phillips curve.

115

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Regional and Sectoral Concerns

and, maybe, therefore explain why the aggregation case has not been revisited with any intensity since.2 If Phillips curves are nonlinear and different regions/sectors are at different points on them, then aggregation of the regional/sectoral results will give different implications for the application of a macroeconomic policy aimed at affecting inflation than one estimated from aggregate data for the whole country or area. Furthermore a policy aimed at reducing the heterogeneity of labour markets, as is the case with European integration, will reduce the sacrifice ratio (unemployment cost) of lowering inflation. If on the other hand it is the regions or sectors with the tightest labour markets that have a disproportionate impact on inflation for the area as a whole then addressing the shortage of labour in those regions, say, through the encouragement of migration, would be an appropriate complement to policies such as monetary policy that do not discriminate in the same way.3 Why then are Phillips or wage curves nonlinear? Luckily or unfortunately, there are several explanations for the regularities observed thus far (Mayes and Virén, 2002b). On the one side the simplest is that unemployment is bounded even if the level of participation in the labour market is itself endogenous. As the bound is approached so inflation is likely to take off. A second common feature in explanations is the key role of labour market institutions. Thus, one may refer to downward rigidities of nominal wages, which themselves can be explained in various ways. One may also refer to asymmetries in employment adjustment – for instance to the apparent asymmetry of hiring (training) and lay-off costs. Given the fact that asymmetries appear to be particularly typical of estimated Okun curves, this explanation may be not completely irrelevant (Silverstone and Harris, 2001).4 Asymmetries do not only appear in behavioural equations but may also be present in policy rules. There the issues can be quite complex:

2 Brechling’s (1973) contribution is also noteworthy in that it offers a clear explanation of why it is that the regions/sectors with the tightest labour markets should lead in the determination of wages in the country as a whole. 3 Monetary policy itself has an impact that varies across sectors (Mayes and Virén, 2002b) but there is no guarantee that this variation would be appropriate for the treatment of the distributed inflationary pressure. 4 A satisfactory explanation thus requires a more substantial model, including at least an Okun curve, an aggregate supply function and a policy reaction function if the nonlinearities are to be taken into account in the appropriate relationships and not simply captured in the limited specification of single equation (Mayes and Virén, 2005).

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

116 Asymmetry and Aggregation in the EU

policy rules may just respond to underlying perceived asymmetries in, say, wage and employment equations. But policy rules can also be genuinely asymmetric. The relevant loss functions can simply be asymmetric, Brainard-uncertainty type constraints in policy behaviour may make certain type of policy actions less desirable, or there may be some institutional constraints in policy, for instance in terms of legislation on welfare systems (see e.g. Schaling (1999) and Tambakis (1999) for more thorough analysis and evaluation of policy implications).5 Many analyses in this area are largely empirical and concentrate on finding evidence on asymmetries/nonlinearities. Most analyses just use aggregate time series data from single countries, which makes the analyses straightforward and maybe also less subject to measurement errors. The problem with this kind of data, however, is that it is difficult to trace back the origins of asymmetries. It is also possible that with small sample sizes some outlier observations may create results which look like nonlinear relationships. With a larger set of data on countries, sectors and regions, empirical findings may be better determined and more widely applicable. We therefore use the much wider data set, drawn from the EU countries that we have employed in the previous chapters and also make use of a range of alternative estimating equations and models to test for the robustness of the results.

5.1

Regional dispersion

The problem for aggregation from asymmetry and nonlinearity applies to some extent at whatever spatial level we choose to measure activity. Indeed regional data within countries will help show the extent of structural change and the degree of mismatch in behaviour across sectors and the economy. We therefore test the hypothesis that the greater the range/variance of regional unemployment at any given level of average unemployment then the more inflationary will be the impact as the low unemployment regions will contribute to inflationary pressure for the EU

5

In Mayes and Virén (2005) we suggest that monetary policy reactions in the euro area appear not only to be asymmetric in that the threat of deflation is fought even more vigorously than the threat of inflation but that the reaction function is nonlinear, with the harshness of the reaction increasing more than proportionately with the distance from price stability.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Regional and Sectoral Concerns 117

118 Asymmetry and Aggregation in the EU

as a whole. This can be regarded as an extension to the Lilien index (Lilien, 1982; Mayes and Silverstone, 1998). (5.1)

where ei is the rate of growth of employment in region or sector i and E is the growth of employment in the area as a whole, wi being the weight, the share of employment in that region in the total. Lilien’s hypothesis is that the greater the dispersion of growth rates in employment the higher is likely to be the unemployment rate. This reflects the idea that it is costly to retrain or move labour. Purely macroeconomic statistics will cover up the consequences of this. If growth is not evenly spread then the more rapidly growing regions will not be as successful in reducing unemployment elsewhere as the less rapidly growing regions are at creating unemployment. The variance of unemployment acts as a measure of the mismatch across the EU. However, it has also been argued that it is the pool of suitably qualified unemployed in the areas of the main demand for labour that are most important in determining inflation. Those with less relevant qualifications or unable to take a job offer quickly will be less relevant, thus generating an asymmetric departure from the simple Phillips curve. The effect of the range of regional unemployment on inflation may be even more extreme. For the case of the UK Buxton and Mayes (1986) showed that the region with the lowest unemployment (the South East) had a highly disproportionate impact on wage inflation for the country as a whole. More than that it appeared to be short-term unemployment that had the effect. Those unemployed for a year or more appeared to be effectively out of the labour market from the point of view of affecting the inflationary process. We therefore return to the estimating equation for the Phillips curve discussed in the previous chapter: ∆pt = b0 + b1∆pt–1 + b2∆pet+1 + b3un+t + b4un –t + b5dispt + ut

(5.2)

where the disp variable simply reflects either the range or the standard deviation of unemployment rates over regions in a country i. Ideally, we ought to use regional data for all variables but, unfortunately, regional data are available only for unemployment and (to some extent) output, but not for prices or inflation, or other macro variables such as taxes, exchange rates and so on. The regional unemployment data are obtained from the Eurostat Regio database. At the NUTS3 level the EU has some 251 regions for

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

L = [∑iwi(ei – E)2 ]

our subset of countries.6 There are, however, some gaps in the data, which cuts the sample size by some 10 per cent. The major drawback, however, is that the data are only available annually. We can see from Table 5.1 that the hypothesis is borne out whichever of the two unemployment variance measures is used. Variance in unemployment across regions has a positive effect on inflation. It is also clear that the individual member states react differently. Inserting fixed effects into the equation improves the fit of the basic Phillips curve considerably, increasing the (negative) impact of average unemployment substantially while also increasing both the size and the significance of the positive impact of the spread. Using the threshold model also shows up a striking difference between periods of positive and negative output gaps. The coefficient where the output gap is above the threshold is double that when a constant value is imposed for all output gap levels. For values below the threshold the coefficient is zero. Slack demand has no impact on inflation. Not surprisingly when this dichotomy is applied dispersion has a somewhat weaker effect on inflation. (Unemployment is bounded so when overall rates are lower the distinction between the average and tightest markets becomes smaller. It is not surprising that the two measures of dispersion give similar results as they tend to move together, as is clear from Figure 5.1, although the standard deviation is the more volatile of the two. In earlier results, Table 5.2, unemployment was used instead of the output gap. As there is no obvious a priori value for the threshold in this case, we use Maximum Likelihood to estimate the threshold and the parameters of (5.2) jointly. This gives a value of 10.8 per cent for the threshold, somewhat higher than the average value of unemployment of 8 per cent for the estimation period. The difference in the two unemployment coefficients is not substantial (column 6) but it is significant at the 1 per cent level. The results follow the expected convexity with the effect of unemployment on inflation being greater at lower levels of unemployment and weaker at higher levels. Table 5.3 offers a further comparison of output gaps and unemployment gaps in an adaptive expectations framework, exploring both the CPI and the GDP deflator as the dependent variable. Where the output gap is negative there is little influence on inflation. Similarly, when unemployment is below its threshold level it too shows no impact on

6

The Irish Republic also had to be excluded through lack of data.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Regional and Sectoral Concerns 119

0.372 (4.84) 0.343 (6.60) 0.363 (10.37) –0.160 (3.23) 0.337 (4.22) 0.364 (5.14)

1992–2007 sd 1992–2007 sd 1992–2007 sd 1992–2007 (y = ∆un), sd 1992–2007 sd/un 1992–2007 sd/sd* 1992–2007 sd/un 1992–2007 EMU, sd/un 1992–2007 Non-EMU, sd/un 1992–2007 range/un 1992–2007 range/un 1995–2006 cv of income pc

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

–0.003 (0.34)

–0.006 (0.62)

Output y<0

0.771 (5.98)

0.735 (5.74)

Output y ≥0 0.682 (12.59) 0.712 (15.00) 0.657 (15.82) 0.754 (15.94) 0.662 (14.25) 0.698 (13.49) 0.682 (14.78) 0.681 (10.59) 0.572 (4.46) 0.673 (12.81) 0.686 (15.04) 0.666 (12.64)

Lag 0.172 (2.51) 0.142 (2.78) 0.202 (3.39) 0.095 (2.11) 0.162 (2.54) 0.080 (1.27) 0.108 (1.99) 0.089 (1.20) 0.063 (0.56) 0.181 (2.47) 0.118 (2.16) 0.108 (1.94)

Forward

1.988 (3.91)

0.079 (2.43) 0.082 (2.79) 0.084 (3.94) 0.061 (2.56) 1.035 (3.79) 0.427 (3.97) 0.508 (1.88) 0.427 (4.20) 2.370 (1.97)

sd

0.223 (2.47) 0.110 (1.27)

Max-min

0.711 0.6678 0.711 0.6678 0.710 0.6679 0.858 0.6789 0.725 0.6659 0.731 0.6452 0.755 0.6168 0.764 0.6317 0.400 0.6764 0.702 0.6728 0.745 0.6279 0.727 0.6714

R2/SEE

OLS

2.21

2.19 17.71 2.28

2.08

OLS

OLS

OLS

OLS

OLS

2.04

2.04

SUR

2.00

OLS

SUR

2.02

OLS

GLS

2.11

2.18 17.86 1.97

OLS

Method 2.07

DW/Wald

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Consensus (June) data are used for the inflation expectations. Inflation measured by the change rate of the CPI. Except for the fourth equation, output is measured by the output gap. Both Wald test statistics (for the restriction that the coefficients of y < 0 and y ≥ 0 are equal) clearly exceed conventional levels of significance.

sd and range = max-min are the unemployment dispersion variables, cv of income pc is the (per capita) income dispersion variable. sd* denotes the sample average of sd.

0.462 (5.50)

0.367 (4.42) 0.311 (3.11) 0.368 (4.62)

Output

Phillips curves from regional data

Data

Table 5.1

120

Regional and Sectoral Concerns 121 Comparison of unemployment dispersion measures

9

2.6

8

2.4

7

2.2

6

2.0

5

1.8

4

1.6

3

1.4 86

88

90

92

range

-->

84

94

96

98

00

02

04

06

standard dev -->

inflation. Both variables thus give a clear indication of the flatness of the Phillips curve below the threshold. One of the disadvantages of using unemployment data is the strong downward trend over much of the period which means that unemployment is not such a helpful variable in defining demand pressure. One alternative would be to derive some measures of the non-inflation augmenting rate of unemployment (NAIRU) or other concept that attempted to set out the neutral value of unemployment round which one could describe periods of pressure. Since these are themselves controversial we stuck with the output gap.

5.2

Sectoral dispersion

There is a second way in which we can approach dispersion, namely through sectors, as we showed in the previous chapter. If rather than dealing simply with aggregate output and output gaps we compute

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Figure 5.1

0.281 (2.01) 0.192 (1.59) 0.254 (4.47)

GDP deflator, SD GDP deflator SD GDP deflator, SD

0.106 (0.79)

0.237 (1.32)

0.763 (8.21)

0.545 (8.94)

0.507 (3.75)

u–u* |u–u*>0

0.759 (60.71)

0.758 (31.67)

0.764 (45.37)

0.773 (36.51)

0.765 (23.55)

0.768 (8.21)

–795 (33.30)

0.798 (34.45)

0.793 (33.21)

Lagged inflation

0.092 (10.10)

0.099 (7.42)

0.043 (4.70)

0.038 (2.75)

0.059 (2.87)

0.083 (7.27)

0.088 (6.01)

0.087 (6.24)

0.085 (8.02)

Import prices

0$.063 (3.24)

0.071 (2.29

0.157 (4.38)

0.149 (4.00)

0.167 (4.15)

0.042 (2.11)

0.063 (1.99)

0.033 (3.68)

0.117 (3.95)

SD/ range

0.867 0.6789

0.867 0.6788

0.733 0.0105

0.731 0.0105

0.735 0.0106

0.867 0.9591

0.868 0.6856

0.849 0.7285

0.852 0.7233

R 2/ SEE

2.10

2.09

2.13

2.29

2.41

2.09

2.16

2.09

2.08

DW

SUR

OLS

SUR

GLS

OLS

SUR

OLS

OLS

OLS

Method

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Notation is the same as in Table 5.1. u–u* denotes the deviation of the unemployment rate from the ‘natural’ level. Notice that with the threshold model, the value of the output gap (unemployment gap) variable above the threshold is obtained by adding the coefficients of y and y|y > 0. With the unemployment variable, the coefficient turns out to be almost zero when unemployment is large. The rate of inflation is computed using either the Consumer Price Index or the GDP deflator. The sample period is 1983–2007.

–0.545 (13.35)

–0.107 (1.76)

CPI, SD

0.718 (3.94)

CPI, SD

–0.088 (0.81)

CPI, SD

u–u*

–0.550 (6.12)

0.363 (5.91)

CPI, range

y|y>0

CPI, SD

0.356 (5.72)

Output y

Backward-looking Phillips curve with regional data

CPI, SD

Data

Table 5.2

122

Regional and Sectoral Concerns 123 Estimates using unemployment rather than the output gap (1) ∆p

e

(2)

0.655 0.649 (12.42) (10.17)

(3)

(4)

0.513 (12.72)

0.488 (11.77)

(5)

(6) 0.522 (13.76)

∆p–1

0.254 (5.72)

0.214 (3.92)

0.191 (5.26)

0.143 (3.66)

0.567 (23.89)

0.187 (5.31)

∆m

0.058 (6.56)

0.056 (5.43)

0.063 (10.14)

0.068 (9.25)

0.085 (13.19)

0.065 (10.54)

–0.053 (3.36)

–0.003 (0.23)

–256 (10.06)

–248 (12.84)

–290 (11.82)

–306/–0.260 (11.66/12.00)

U Umax-Umin

0.068 (4.81)

Usd

0.147 (6.47) 0.103 (2.32)

t

0.154 (6.86)

0.130 (5.91)

0.192 (2.93)

–0.016 (1.82)

–0.001 (0.45)

–0.112 (10.06)

–0.108 (9.05)

–0.112 (9.25)

–0.110 (10.57)

R2

0.868

0.866

0.914

0.918

0.885

0.918

SEE

0.963

1.073

0.816

0.878

0.938

0.797

DW

1.526

1.289

1.800

1.590

1.928

1.822

Dummies

No

No

yes

yes

Yes

Yes

Obs

153

143

153

143

153

153

All estimates are SUR estimates. ∆ p e denotes expected inflation (OECD forecasts), m import prices, ∆p is inflation in consumption prices, U the aggregate unemployment rate, Umax-Umin the range of regional unemployment rates, Usd the corresponding standard deviation and t time trend. Column (6) is estimated using a threshold model specification and allowing the coefficient of the unemployment rate to vary depending on whether the rate is below (first coefficient) or above the 10.8 per cent (second coefficient) threshold. The hypothesis that the coefficients are equal can be rejected with marginal probability of 0.0013 per cent using the F test.

gaps for each of the four industries separately,7 for all four sectors the impact on inflation is higher when there is a positive output gap (see the last four rows of Table 5.4). In each case the positive segment coefficient is clearly significantly different from zero. In the case of agriculture and construction the impact is relatively limited. The negative segment coefficients are close to zero and poorly determined, with the exception of services where there is a moderate affect. The sectoral distribution of any excess supply thus has an effect on the overall

7

Inflation also relates to the sectoral prices.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 5.3

124 Asymmetry and Aggregation in the EU

Output dispersion

Estimates of a nonlinear Phillips curve with output dispersion

pm

p–1

∆tax

Dispersion

y

y<0

R 2/ SEE

y≥0

DW/ (Wald)

GAP sd(un)

0.099 0.769 (7.97) (42.09)

0.156 (1.47)

0.082 (4.43)

0.181 (5.85)

0.896 0.768

2.11

∆Un sd(un)

0.106 0.770 (6.61) (28.56)

0.038 (0.29)

0.081 (3.37)

–0.132 (2.27)

0.792 0.826

1.87

GAP sd(gi)

0.157 0.783 (10.24) (46.12)

0.092 (1.25)

0.057 (2.69)

0.262 (8.81)

0.932 1.079

2.13

GAP sd(gi)

0.158 0.771 (10.61) (42.03)

0.074 (1.16)

0.025 (1.12)

g sd(gi)

0.160 0.772 (9.48) (39.12)

0.061 (0.89)

0.045 (1.74)

g sd(gi)

0.149 0.786 (9.25) (41.21)

0.064 (0.92)

0.034 (1.45)

0.102 (1.99)

0.369 (7.09)

0.055 (1.90)

0.933 2.11 1.072 (0.005) 0.904 1.181

–0.081 (2.03)

0.145 (4.72)

1.81

0.916 2.02 1.135 (0.000)

sd(un) is the regional unemployment rate dispersion variable while sd(gi) corresponds to the sectoral output growth dispersion (standard deviation). The transition variable in equation 6 is also the output gap. With regional unemployment dispersion, the number of data points is 176 and with the sectoral output growth dispersion variable 298. The sample period is 1974–2004.

outcome. Since shocks have differential effects across sectors we would expect this to have differential effects on inflation and hence on monetary policy. Note here that this was not a forward-looking version of the Phillips curve and contained both import price and consumption tax variables to take account of other ‘external’ sources of inflation. pt = β11u –t + β12u+t + β2sdt + β3pt–1 + β4pm + β5∆taxt + vt

(5.3)

where pm = import prices tax = VAT rate (the main rate) Clearly to quite some extent this is illustrating what we know already as these asymmetric impacts would be picked up by other aspects of macroeconomic models. Sectoral shocks would have differing effects on the exchange rate or import prices for example. Nevertheless these results make it clear that neglecting the distribution of the impact below the EU level could have misleading implications for policy, whether the neglect was national, regional or sectoral. Even within smaller countries the distributional differences still matter.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 5.4 variables

Regional and Sectoral Concerns 125 Figure 5.2

Sectoral output growth dispersion indicators

0.24 Standard deviation max-min 0.20

0.16

0.12

0.08

0.00 1980 Figure 5.3

1985

1990

1995

2000

2005

The extent of change

0.5 Pub MIN 0.4

MAX

Man

0.3

Ser

0.2 Con 0.1

Agr

0.0 1

2

3

4

5

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

0.04

126 Asymmetry and Aggregation in the EU Figure 5.4

Evolution of sectoral shares (median values)

0.30 6

5

0.25

4

0.20 2 0.15

0.10

3

0.05

0.00 1980

1985

1990

agriculture construction financial services

1995

2000

2005

manufacturing trade public services

Sectoral growth has been far more volatile than overall output growth (Figure 5.2) and sectoral shares have changed in a marked manner over time (Figures 5.3 and 5.4 and Table 5.5). Within the economy the relative importance of financial and other private services have grown whilst the share of manufacturing has declined. As a result it is relatively difficult to treat the problem from only one specific direction. Early in the period, shortages in manufacturing would have been the pinch points in the economy. By the end of the period the crucial shortages could just as well have been in the services sector, altering both the structure of influence and the structure of wages and salaries. Our results seem to be a little more robust to the finding of asymmetry and nonlinearity than some other recent studies. In their work on asymmetry and nonlinearity in the Phillips curve, Laxton

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

1

Regional and Sectoral Concerns 127

Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis Observations

Sectoral shares of output over countries and time (1991–2008) 1

2

3

4

5

0.025 0.026 0.070 0.003 0.012 0.809 3.733

0.220 0.215 0.371 0.103 0.043 0.456 4.741

0.063 0.060 0.120 0.035 0.014 0.758 3.390

0.211 0.210 0.280 0.155 0.026 0.040 2.223

0.249 0.238 0.484 0.162 0.057 1.672 6.728

1140

1140

1140

1140

1140

6 0.232 0.232 0.313 0.132 0.033 –0.736 3.650 1140

et al.8 find that while the evidence supports the existence of convex relationships between inflation and unemployment in an expectations augmented specification, the convexity is not strong over the policy relevant range and the evidence relatively weak.9 Indeed they conclude (Laxton et al., 1999: 43) ‘standard empirical techniques are not likely to be capable of providing a reliable answer on the functional form’. However, in no case is the convex relationship rejected by the data. They use both the regime change model we employ and a continuous curve and consider the US, UK and Canadian economies. McDonald (1997) and Razzak (1997) find similar relationships for the Australian and New Zealand economies. Inside the euro area the convexity will have a particular effect if the various member states are out of phase in their economic cycle or have been subject to asymmetric shocks that require structural adjustment that may be slow to come if there is substantial hysteresis in the economy. The economies that are suffering a negative output gap will be doing less to bring inflation down than the economies with the positive output gaps are providing upward pressure. Therefore in general the more asynchronous the euro area turns out to be the tighter monetary policy will need to be compared with any given growth rate for the area as a whole. If cycles are asymmetric in the sense that it tends to be more difficult to get out of recessions then the problem will be exacerbated.10 8

Laxton et al. (1993); Laxton et al. (1995); Clark et al. (1996); Debelle and Laxton (1996); Clark and Laxton (1997); Laxton et al. (1999). 9 The authors use both piecewise linear and curvilinear specifications. 10 The Phillips curve is also asymmetric in a different sense in the Ball (1993), Mayes and Chapple (1995) discussion of the ‘sacrifice ratio’. Here the gains in terms of extra output when the output gap is positive are more than offset by the losses when a negative output gap has to open to return inflation to its previous level. In this case the relationship is not merely a curve but its shape depends upon whether the output gap is falling or rising.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 5.5

128 Asymmetry and Aggregation in the EU

Further facets of dispersion

Finally, we also use the dispersion (coefficient of variation) of per capita income in different regions as an additional regressor and it is no surprise to find it behaves more or less in the same way as unemployment dispersion. This obviously reflects the fact low income regions are also characterised by high unemployment rates. However Figure 5.5 suggests that per capita income is likely to work less well as its variance is more limited. Moreover, since the behaviour is clearly different it may represent an additional rather than a substitute explanation. Taken together, these results have powerful policy implications: to achieve the goal of low inflation not only requires monetary discipline but also structural policies which tend to diminish the difference between extreme values of economic development within economies. Achieving larger mobility of factors of production would obviously contribute to this aim. One way out of the puzzle is to use a more general dynamic specification. In this case, an obvious alternative is a VAR model which includes the relevant variables (except for the dispersion variable) as ‘endogenous’ and the dispersion variable as predetermined variable. Identification is

Figure 5.5

Comparison of income (per capita) and unemployment dispersion

0.24

2.4

0.22

2.2

0.20

2.0

0.18

1.8

<--Income Unemployment - ->

0.16 95

96

97

98

99

1.6 00

01

02

03

04

05

06

Income dispersion is measured by the coefficient of variation of income per head in different regions and unemployment dispersion by the standard deviation of unemployment rates. The values are medians of country values.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

5.3

Regional and Sectoral Concerns 129 Estimates of a VAR model with the unemployment dispersion

Lag

u

pe

p

u–1

1.312 (17.66)

–0.162 (2.62)

–0.076 (1.10)

u–2

–0.412 (5.80)

0.178 (3.02)

0.083 (1.26)

pe–1

0.045 (0.38)

0.595 (6.07)

0.755 (6.87)

pe–2

0.175 (1.28)

0.020 (0.17)

–0.090 (0.70)

p–1

–0.057 (0.52)

–0.054 (0.58)

0.207 (2.01)

p–2

–0.023 (0.25)

0.010 (0.12)

0.060 (0.69)

sd

0.099

0.010

0.052

(2.07)

(0.25)

(1.17)

C

0.035 (0.17)

0.461 (2.60)

–0.087 (0.44)

R2

0.962

0.672

0.772

SEE

0.889

0.656

0.825

Note: The VAR model is estimates from EU cross-country data which after adjustments include 150 observations. pe corresponds to OECD June inflation forecast for the following year. The dispersion variable sd is considered exogenous. The sample period is 1983–2003.

based on the conventional Cholesky decomposition with the variable ordering (unemployment, expected inflation and actual inflation). This sort model is indeed estimated and the results are reported in Table 5.6 and Figure 5.6 for the impulse responses functions. A useful property of this analysis is the possibility of handling inflation expectations as genuinely endogenous variables, so that inflation expectations both affect actual inflation and unemployment, and shocks in all other variables affect inflation expectations. In a single-equation model with the Rational Expectations Hypothesis orthogonality restrictions, inflation expectations typically just reflect lagged inflation and output terms. Although the estimation procedure does not violate the basic idea of rational expectations it is quite far from the more sophisticated and advanced ways of constructing forecasts that are used in practice and hence is likely to contain less information than in the OECD forecasts used in Table 5.1.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 5.6 variable

130 Asymmetry and Aggregation in the EU Figure 5.6

Impulse responses of the VAR model Response to Cholesky One S.D. Innovations ± 2 S.E.

Response of UN to UN

Response of UN to CPES2(1)

Response of UN to PCP

1.6

1.6

1.6

1.2

1.2

1.2

0.8

0.8

0.8

0.4

0.4

0.4

0.0

0.0

0.0

–0.4

–0.4

–0.4

2

3

4

5

6

7

8

9

10

1

Response of CPES2(1) to UN

2

3

4

5

6

7

8

9

10

1

Response of CPES2(1) to CPES2(1) 1.0

1.0

0.8

0.8

0.8

0.6

0.6

0.6

0.4

0.4

0.4

0.2

0.2

0.2

0.0

0.0

0.0

–0.2

–0.2

–0.2

–0.4 1

2

3

4

5

6

7

8

9

10

2

3

4

5

6

7

8

9

10

1

0.8

0.4

0.4

0.4

0.0

0.0

0.0

–0.4 3

4

5

6

7

8

9

10

5

6

7

8

9

10

3

4

5

6

7

8

9

10

9

10

Response of PCP to PCP

0.8

2

2

Response of PCP to CPES2(1)

0.8

1

4

–0.4 1

Response of PCP to UN

–0.4

3

Response of CPES2(1) to PCP

1.0

–0.4

2

–0.4 1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

In short, the VAR results show that unemployment dispersion has a positive impact on all other variables: unemployment as well as actual and expected inflation (cf. Table 5.6). In this sense, dispersion is a bad thing. As for the impulse response functions, expected inflation tends to increase unemployment while actual inflation tends to lower it. This could be interpreted from the point of view of standard supply-side behaviour in which unexpected inflation has a positive effect on output and a negative effect on unemployment. It could also reflect the operation of a forward-looking monetary policy, which responds to expected inflation but has to live with the results of any contemporaneous errors.

5.4

An illustration with Finnish regional data

One of the problems with the foregoing analysis is that data availability have constrained it to be annual. This makes the dynamics much less interesting and may help to obscure some of the origin and nature of

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

1

the nonlinearity. At a national level on the other hand we may not only find quarterly data on unemployment but also monthly data. We have therefore explored the Phillips curve with dispersion both in regional unemployment and sectoral data for Finland over the period 1971M1– 2009M5. Finland is particularly interesting because it contains a complete financial crisis starting in late 1992 and in employment terms continuing right until the present crisis, although output had returned to (and indeed surpassed) its previous trend by the end of the 1990s. Thus there are clear periods of excess pressure and very considerable slack in the labour market to take into account. Furthermore the picture across Finland is very mixed with northern and eastern Finland being hit much harder by the crisis than much of the rest of country. Indeed the crisis itself prompted a move to the major centres, particularly Helsinki, so there was time for quantitative adjustment. Figure 5.7 shows that despite levels of unemployment rising markedly after 1992, the range and standard deviation become only slightly higher than was prevalent before the pre-crisis boom in the late 1980s. The early consequences of the present crisis are also clear. Although a more rapid recovery of output and indeed a smaller fall are expected this time round, the forecast for unemployment is pessimistic. Figure 5.7

Finnish data on regional unemployment rates

16 SD RANGE

14 12 10 8 6 4 2 0 70

75

80

85

90

95

00

05

10

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Regional and Sectoral Concerns 131

132 Asymmetry and Aggregation in the EU

We estimate two versions of the Phillips curve for these data, first of all the standard hybrid New Keynesian model as in (5.2). ∆12p = 0.505∆12 p–1 + 0.495∆12 pe+1 – 0.010UNC + 0.040SD (15.30) (14.96) (1.33) (0.24) R2 = 0.994, SEE = 0.344, J = 0.007, N = 460. and secondly a simplified version of the adaptive expectations model. Assume adaptive expectations in the form ∆pt+1 – ∆pt = θ(∆pt – ∆pte)

(5.4)

and the Phillips curve of the form ∆pt = a∆pte – but

(5.5)

which is a form of the expectations-augmented model. Eliminating pte gives us: ∆pt+1 – ∆pt = – (bθ/a)u – (1–a)(θ/a)∆pt which is the form we estimate in Table 5.7. Table 5.7

Equation

Finnish regional data results

u–u*

u–u* |u–u*>0

∆p

SD/ range

SEAS D3

R2/ SEE

DW

1 SD

–0.061 (1.84)

–0.505 (12.64)

0.063 (12.64)

0.002 (2.78)

0.256 0.0050

2.29

2 range

–0.063 (1.86)

–0.495 (12.42)

0.017 (6.59)

0.002 (2.82)

0.258 0.0050

2.31

3 SD

–0.201 (2.58)

–0.525 (12.73)

0.045 (4.33)

0.002 (2.81)

0.269 0.0049

2.27

0.285 (2.47)

Estimating equation ∆2pt+1 = a1*(ut–ut*) + a2pt + a3SDt + a4∑SEASit.

where p = log(CPI), ∆12p = annual inflation, and ∆2p second difference of p. SD = standard deviation of regional unemployment rates, RANGE is the range of Finnish regional unemployment rates GMM estimates for New-Keynesian hybrid Phillips curve. What is striking about these results is not just the importance of the dispersion of unemployment but that the influence of unemployment varies so clearly across the threshold. In each case the unemployment

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

(5.6)

Regional and Sectoral Concerns 133

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

term is correctly signed and the dispersion term contributes to a closer explanation. The evidence for the importance of dispersion seems overwhelming and hence the treatment of Phillips curves without regard either to the dispersion of unemployment or to the dispersion in sectoral performance are likely to be misleading for decisions on aggregate policy. Clearly one way to deal with this is to consider the problem at a more disaggregate level and then aggregate taking into account the differences among the nonlinear estimating equations. However, simply by looking at measures of dispersion it is possible to perceive a more nuanced picture. Increases in dispersion in both unemployment and sectoral dimensions increase inflationary pressure. Insofar as there is going to be increasing integration in the EU then these pressure from dispersion will become progressively less important. But if there is going to be further substantial structural change this may offset at least part of the lessening impact through convergence.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

6

After the Phillips curve, the Okun curve – the relationship between output and unemployment – is probably the most obviously nonlinear. In the growth phase of the economic cycle unemployment tends to fall steadily until the point when labour shortages are reached. In the downturn, even if it only represents slow growth, unemployment tends to rise disproportionately but with a lag. This inter-relationship, which is reflected in fluctuations in labour (and total factor) productivity, lies at the heart of the real business cycle literature and it has proved difficult to provide a totally convincing explanation that covers the whole cycle, the shocks that are experienced and the leads and lags that affect behaviour. Hiring and firing have costs. Hence firms try to take a longer-term view of their labour needs. In periods of rapid expansion they prefer to get the workforce to work longer hours and seek means of increasing productivity, albeit temporarily, if they think that the growth may be exceptional. Similarly in a downturn, if the pause is expected to be relatively short-lived, they will attempt to hold onto their labour force and reduce costs by cutting back on hours worked and limiting pay increases. One of the characteristics of the present recession in Europe, unlike the United States, is that unemployment has risen by less than what might otherwise have been expected from previous, milder fluctuations. In part, this is probably due to the concentration of the problems in the financial sector but it no doubt also represents a general feeling that demand will bounce back, particularly in the investment goods sectors, as soon as the banking problems are sorted out. Clearly the problems vary by individual country, depending not only on the structure of their industry but also on the size of their banking problems and whether they had been subject to a housing boom in the period leading up to the crisis. In any event the 134

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Output, Unemployment and the Labour Market: The Okun Curve

crisis is not over yet and there are fears that there will be a double dip recession as the emergency measures both to save the banking system and to bolster demand temporarily are unwound. This has been exacerbated by the realisation that some projected debt levels, particularly in Europe, are unsustainable and hence need to be addressed sooner rather than later. Such a second contraction would be much more likely to result in more substantial falls in employment as optimism declines and the ability to bridge the temporary slowdown falls. Such speculations lie beyond the scope of this chapter as our data are predominantly for the 30-year period before the crisis, during which fluctuations in most western European countries were more limited and characterised more by oil or other external shocks. There is one exception, the Nordic countries, which experienced a previous financial crisis at the beginning of the 1990s. However, this is a single episode, concentrated in just three countries in our data set, and, hence, is unlikely to be sufficient to characterise the estimates for the data as a whole. In the rest of this chapter, we therefore begin by setting out the characteristics of the Okun curve and how we treat it, before going on to look at the behaviour of Okun curves both for our panel of countries and for them individually. We conclude with a wider consideration of the problems that the asymmetries in behaviour that our analysis reveals pose for economic policy, particularly in the light of the lack of success of the Lisbon Agenda to increase the rate of growth and the new problems that the crisis has added to those of ageing and competition from China inter alia.

6.1

The specification and estimation of Okun curves

The Okun curve has been the subject to several recent studies (Attfield and Silverstone, 1998; Harris and Silverstone, 1999a, b; Kaufman, 1988; Moosa, 1997, Palley, 1993, Prachowny, 1993 and Weber, 1995, for example) so we have plenty of experience to draw upon. There is clear experience of asymmetry. For example, Harris and Silverstone (1999b) find asymmetry of some form for Australia, Japan, New Zealand, the UK, US and West Germany over the period 1978 to 1999.1 However, the finding is not universal and they cannot reject the null hypothesis of symmetry for Canada over the same period. Perhaps more interesting is

1

Silvapulle et al. (2004) and Cuaresma (2003) also find asymmetry for the US over the period since the war.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Output, Unemployment and the Labour Market: The Okun Curve 135

136 Asymmetry and Aggregation in the EU

the much more detailed treatment of the relationship by Haltiwanger and Schuh (1999), who introduce sector specific factors to help explain the lack of symmetry. We deal with the aggregated relationship first. The Okun curve (Okun, 1962) is normally expressed as the relationship between the change in unemployment and the percentage change in real output in the economy (6.1)

However, it is also argued (see Prachowny (1993) for example) that some scaling of the labour variable is required so in our formulation we have also included population of working age, POP.2 The curve may therefore offer some additional insight into the nonlinear operation of the labour market to augment the Phillips curve results of the previous section. Both employment and unemployment appear to respond in an asymmetric manner to demand shocks. We focus on the Harris and Silverstone approach, as rather than estimating a curve or piece-wise linear function they build the asymmetry into the error correction mechanism, assuming that there are different correction paths depending upon whether real output is above or below its trend value. In effect, therefore this gives us three different ways of handling the asymmetry. The first, following Kim and Nelson (1999) is to assume that although the function itself is linear, we should treat potential output more in the form of a frontier, very much along the lines of frontier production functions (Aigner et al., 1977; Mayes et al., 1994; Mayes, 1996). This provides a direct extension to Prachowny’s (1993) production function basis for the Okun curve. Here the errors in the relationship can be decomposed into a symmetric term e and a nonsymmetric term v, which permits a longer tail of values when the economy is operating inside the frontier. Thus in the case of (6.1) the error term u in the estimated relationship ∆U = c0 + c1∆Y/Y + u

(6.2)

would be composed u = v + e, with e ~ N (0, σ2e) and v ~ M (µ,σ2v) where M is a nonsymmetric distribution.3

2

Prachowny (1993) employs a rather more elaborate transformation, incorporating the Okun curve into a production function and considering linear discrepancies of the factors of production from their equilibrium values as well. 3 In line with the early frontier production function Kim and Nelson (1999) assume that M is half Normal, Mayes et al. (1994) also consider the more general case of a truncated normal.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

∆U = c0 + c1∆Y/Y

Output, Unemployment and the Labour Market: The Okun Curve 137

The second approach, used by Harris and Silverstone (1999a, b), is to estimate the cointegrating relationship in (6.2) and assume that the error correction mechanism ε ε = U – cˆ0 – cˆ1 y

(6.3)

where ^ denotes an estimate, can be divided into ε where ε < 0 and ε+ ≥ 0. The coefficients in the equations explaining ε– and ε+ are then not constrained to be equal in the adjustment process. Our approach and that of Laxton et al. (1999) and Pyyhtiä (1999) is to treat the relationship itself as being nonlinear and hence we use the more conventional threshold model in terms of output growth –

(6.4)

In (6.4) y is the growth rate in GDP,4 pop the population of working age and ε the error correction term defined in (6.3) (lagged one period). t is a threshold value for the asymmetry. There are a number of routes to determining this threshold. One would be simply to use a simple form of output gap, although since our data are annual this would entail a fairly trivial definition of potential output, such as the mean rate of growth over the sample period. We show the effects of setting t equal zero so that we distinguish actual recessions from other behaviour and determining the maximum likelihood value for t. In the second case the outcomes tend to be near the mean. Each of the three approaches gives a somewhat different flavour to the problem but it is possible to take the study of asymmetry further as demonstrated by Holmes and Siverstone (2005), who suggest that the asymmetry has two facets. First of all, the response depends upon whether output is rising or falling but, secondly, it also depends upon whether the economy is above or below trend. This can be considered in terms of output or unemployment. It is really only with the United States that there is enough data available to estimate this more complex relationship, as over 40 years of data and five cycles are available for study in the period since the early 1960s (not counting the present downturn). What Holmes and Silverstone find is that unemployment is more sensitive to changes in output in periods when activity is below trend than when it is above. Taken simply this might imply that when the

4 We also use the output gap as the measure of output in some regressions as a test of the robustness of output growth.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

∆U = c0 + c1y+ + c2y– + c3 ∆pop + c4ε–1 + ut

labour market is under pressure output growth has to be largely driven by productivity if it is to succeed. Correspondingly if there is a downturn in activity firms will tend to hoard labour, at least initially, because of the difficulty of recruiting again in such an environment. Taken together therefore this gives four phases over the course of the cycle. As activity begins to fall away from the peak, unemployment will increase and at faster rate than it fell per unit of output in the period running up to the peak. Once output falls below trend the rise in unemployment for each given fall in output will increase. Once the bottom of the cycle has passed and output begins to pick up then unemployment will begin to fall but again by less than it rose in the latter stages of the downturn per unit of output change. Once output rises enough that it moves above trend levels, unemployment will fall more slowly for each increase in output until eventually the economy peaks and the process is repeated. Holmes and Silverstone (2005) suggest that this double asymmetry may explain the jobless growth of the recovery after 2001 as the US economy was rapidly back over trend levels of output. Koenders and Rogerson (2005) offer a somewhat different explanation based on the length of the previous period of expansion – the longer the period of expansion then the smaller the employment growth thereafter. However, the Holmes and Silverstone explanation sounds the more plausible, as it relates to the state of the labour market. Such findings would also help explain why, in the EU, unemployment has not risen as steeply as one might have expected in the present recession. It is however more difficult to explain the US experience where unemployment has risen sharply. Perhaps a more complex view of the dynamics is required which takes expectations into account as well.

6.2

Estimates of the Okun curve

Our principal results are shown in Table 6.1. As expected there is a clear relationship between output, population and the level of unemployment across our sample of countries. There is also a trend in the data. As expected, increases in output decrease unemployment while increases in population increase it. The same applies if we look just at the changes in the relationship as in the first seven rows in the table. This seems largely invariant to estimation method or to the choice of time period. The relationship is clearly asymmetric with unemployment responding noticeably less to changes in output above the threshold than it does below, whether we chose to distinguish between positive and negative

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

138 Asymmetry and Aggregation in the EU

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

0.156 (2.62) 0.334 (5.75) –0.424 (1.62) 0.258 (5.51) 0.302 (4.98) 0.303 (4.41) 0.271 (2.42) 0.260 (2.76) 0.314 (3.30)

1971–2008 1999–2008 1999–2008 1971–2008 1971–2008 1971–2008 1971–2008 1971–2008 1971–2008

–0.143 (6.64)

–0.182 (12.53)

–0.205 (12.53)

–0.158 (13.23)

Output

–0.128 (3.03)

–0.217 (4.51)

–0.238 (1.84)

–0.302 (6.21)

–0.171 (15.05)

–0.248 (6.71)

Output x<0

–0.172 (3.78)

–0.125 (3.82)

–0.146 (2.98)

–0.082 (3.72)

–0.132 (8.00)

–0.145 (11.80)

Output x≥0

–0.089 (3.99)

–0.081 (3.78)

–0.086 (3.88)

Lagged EC-term

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

0.283 (4.63)

Population

The Okun curve

1971–2008

Data sample

Table 6.1

0.164 1.212

0.169 1.206

0.165 1.213

0.135 1.238

0.131 1.239

0.135 1.236

0.092 1.689

0.111 1.278

0.138 1.249

0.130 1.239

R2/SEE

0.36 11.97

0.37 15.71

0.36 0.70

0.38 21.2

0.39 9.16

0.35 5.71

0.53

0.32

0.39

0.34

DW F(1,1759)

GLS

GLS

GLS

GLS

GLS

GLS

OLS

GLS

OLS

GLS

Method

g GAP=0

g g= 0.048

g g=0

GAP GAP=0

g GAP=0

g g=0

g

g

g

g

Output/x

139

0.070 (2.57) 0.088 (3.03) 0.089 (3.08) 0.287 (2.12) 0.172 (1.09)

1971–2008 ∆1un 1971–2008 ∆1un 1971–2008 ∆1un 1971–2008 ∆1un 1971–2008 ∆1un

–0.088 (6.20)

–0.093 (2.07)

Output

0.056(*) (3.16)

0.061(*) (0.80)

–0.072 (2.02)

–0.034 (2.53)

–0.090 (3.39)

–0.071 (2.12)

Output x<0

–0.012 (1.40)

–0.068 (2.80)

–0.032 (2.32)

–0.041 (2.09)

Output x≥0

–0.029 (2.93)

–0.030 (3.61)

–0.015 (2.80)

–0.012 (2.40)

–0.013 (2.83)

–0.010 (1.71)

Lagged EC-term

0.083 0.569

0.039 0.580

0.030 0.580

0.031 0.578

0.030 0.579

0.28 0.580

R2/SEE

1.85

1.78

1.80 2.25

1.80 6.25

1.79 9.28

1.79 1.25

DW F(1,1759)

OLS

OLS

GLS

GLS

GLS

GLS

Method

g UNC=0

GAP GAP=0

GAP GAP=0

g GAP=0

g g=0.012

g g=0

Output/x

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

All equations include fixed cross-section effects. x indicates the transition variable. The dependent variable is ∆4(un). In the case of the four last equations, it is ∆1(un). Accordingly, the growth rate of output (and population) is also expressed as quarterly log differences. The EC term is derived from the following long-run relationship: un = fixed effects + 0.300log(pop) – 0.160log(GDP) + 0.107trend. Nonzero values of the threshold parameters represent optimal values in the sense of minimum sum of squares. (*) indicates the coefficient of gap*{1/[1+exp(–θx)]} in a Smooth Transition Regression. Here, θGAP = 68.0 and θUNC = 2.5.

0.080 (2.97)

Population

The Okun curve – continued

1971–2008 ∆1un

Data sample

Table 6.1

140

growth, positive and negative output gaps or seek the threshold that minimises the overall errors.5 The last two rows relate to our smooth transition model, which works rather better when the transition relates to unemployment levels rather than the output gap. In earlier work, reported in Table 6.2 we have used both a longer data series and a wider range of countries than Harris and Silverstone (1999b). While we did experiment with a split error correction term it appeared that incorporating the asymmetry into the coefficients of the equation was a rather better determined approach. Different speeds of adjustment alone had lower explanatory power and added little when the output coefficient spilt was already present. In part this may be due simply to the use of annual rather than quarterly data. The table shows estimates of simple Okun curves for 21 countries from 1961 to 1997.6 Only in the case of the UK and Japan do we find that there seems to be little relation between output and unemployment when using a linear formulation of (4).7 However, the relationship for New Zealand is weak (a very different result from that found in Harris and Silverstone, 1999a). Once we introduce the asymmetry, most countries produce the positive and negative segments with different slopes and show the expected asymmetry very clearly. If we separate out the data according to whether or not the economy is in recession,8 columns 1 and 2 in Table 6.2, in 14 of the 21 cases the coefficients are larger when the growth rate is negative. In other words unemployment rises more when the economy contracts than it falls when the economy expands. This fits with our expectations about hysteresis. However, the differences are not in general significant. Of the seven cases that do not conform to this pattern, Finland shows no asymmetry, the US shows the reverse asymmetry but with appropriately signed coefficients, while Greece, Italy and Japan have perversely signed coefficients for the negative segment. However, in each case the likelihood ratio test does not lead us to reject the symmetric relationship. Symmetry is also rejected in the case of the UK but here the negative segment also has a perverse coefficient. If on the other hand we split the relationship at the point which maximises the likelihood function then only five cases show coefficients where the effect on unemployment is smaller (less negative) below the

5

There is one exception when the output gap is used as the threshold. Switzerland was eliminated from our initial sample owing to data difficulties in the 1960s. 7 Moosa (1997) also gets a low value for Japan. 8 I.e. if GDP falls. 6

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Output, Unemployment and the Labour Market: The Okun Curve 141

142 Asymmetry and Aggregation in the EU Individual country estimates of a nonlinear Okun curve y+(0)

y–(0)

y+(c)

y–(c)

F

Australia

–0.021 4.04)

–0.049 (0.41)

–0.045 (3.42)

–0.002 (0.09)

10.26

Austria

–0.039 (3.76)

–0.512 (–0.047)

–0.050 (5.03)

–0.075 (3.93)

14.53

Belgium

–0.026 (2.33)

–0.125 (2.85)

–0.038 (4.42)

–0.070 (4.66)

18.29

Canada

–0.038 (4.22)

–0.068 (2.62)

–0.040 (5.78)

–0.071 (4.93)

15.23

Denmark

–0.022 (1.03)

–0.451 (3.36)

–0.030 (1.48)

–0.392 (3.52)

18.45

Finland

–0.071 (5.29)

–0.070 (3.07)

–0.066 (6.15)

–0.079 (6.28)

16.82

France

–0.019 (1.15)

–0.050 (0.43)

–0.028 (2.00)

–0.080 (2.48)

15.75

Germany

–0.096 (4.56)

–0.135 (0.93)

Greece

–0.023 (3.03)

0.024 (0.67)

–0.027 (3.58)

0.038 (1.19)

21.67

Iceland

–0.072 (4.84)

–0.119 (2.81)

–0.076 (5.76)

–0.121 (3.35)

15.67

Ireland

–0.019 (2.17)

–0.088 (0.35)

–0.025 (3.31)

–0.050 (2.86)

5.20

Italy

–0.026 (2.27)

0.021 (0.34)

–0.019 (1.81)

–0.043 (2.82)

14.05

Japan

–0.007 (1.16)

0.013 (0.18)

–0.008 (1.35)

0.020 (2.02)

10.33

Netherlands

–0.023 (0.95)

–0.182 (1.22)

–0.048 (2.73)

–0.123 (4.07)

112.86

New Zealand

–0.075 (1.21)

–0.036 (0.22)

–0.086 (1.78)

0.025 (0.33)

24.38

Norway

–0.043 (2.14)

–0.185 (0.49)

–0.059 (3.00)

–0.094 (2.95)

6.79

Portugal

–0.044 (2.47)

–0.250 (0.71)

–0.055 (3.19)

–0.094 (3.60)

16.11

Spain

–0.019 (3.49)

–0.062 (0.79)

–0.026 (5.09)

–0.013 (1.61)

29.24

Sweden

–0.064 (2.72)

–0.122 (1.92)

–0.062 (3.65)

–0.110 (5.05)

13.11

–

–

–

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 6.2

Output, Unemployment and the Labour Market: The Okun Curve 143 Table 6.2

Individual country estimates of a nonlinear Okun curve – continued y+(0)

y–(0)

y+(c)

y–(c)

F

UK

–0.032 (1.50)

0.095 (1.83)

–0.031 (1.51)

0.102 (1.97)

21.08

USA

–0.067 (7.39)

–0.036 (0.98)

–0.061 (9.57)

–0.081 (5.24)

9.39

threshold (columns 3 and 4 in the table). Three of the countries from the perverse split at zero and in this group, but the US, Japan and Italy no longer show perversity in this case but Australia and Spain now do. (We were unable to produce estimates for Germany because of the overwhelming effect of unification.) Only in the case of the UK was the coefficient for the negative segment significantly different from zero at the 5 per cent level and here the threshold value, at –0.53 per cent, was very much out of line with the rest of the sample. Most thresholds lay in the range 2.3 to 4.3 per cent and in all cases the restriction that the two GDP coefficients be equal was rejected. Harris and Silverstone (1999b) also encounter the problem of perversity but only on a limited scale and their estimates are well determined. They find that Canada does not show asymmetry, New Zealand, the US and Germany show no adjustment when the error correction term is negative – and hence clear asymmetry – while Australia and Japan have larger effects for the negative segment, i.e. reverse asymmetry. They do not suffer from unexpected signs. Thus their more limited sample demonstrates asymmetry on a rather similar scale to our own but with somewhat different country characteristics. The differences may simply reflect that the asymmetric process investigated is not the same. However, use of these aggregate models in some senses only provides a description of the stylised facts and not an explanation of why the asymmetry may be occurring. This becomes clearer at the disaggregated level. In discussing regional disaggregation of the Phillips curve in Chapter 5, we suggested that it was the tightest labour markets that contributed to

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Numbers inside parentheses are t-ratios. y +(0) and y –(0) denote estimates with zero threshold and y+(c) and y–(c) estimates with nonzero (estimated) threshold value. The parameters are derived from the following estimating equation ∆ut = a0 +a1∆y+t + a2∆y–t + a3∆popt + a4ε–t–1+et, where u denotes the (log) number of unemployed, y the growth rate of output, pop the (log) working-age population, ε an error-correction term in terms of u, pop and time trend and e the error term. In the nonlinear case, y is replaced by y+ and y– so that y+ corresponds to positive values of y and y– of negative values. F is the F(1,31) test for the equality of the coefficients of y+ and y– in the case of nonzero threshold. Estimates are based on annual OECD data for 1961–1997.

inflation and hence that we needed to consider the spread of unemployment across markets and not just the level in order to understand the nature of the problem. In the case of the Okun curve Haltiwanger and Schuh (1999) demonstrate that it is necessary to understand the dynamics of the labour market at the plant level to get an appreciation of asymmetry. They show that a further term should be added to our formulation of the Okun curve, used in Table 6.2, which reflects the degree of ‘job reallocation’9 both within and between sectors. For all of the five different measures they use there is a clearly significantly positive effect on unemployment from increased rates of job reallocation. However, Haltiwanger and Schuh (1999) go even further and estimate determinants of job reallocation. Here not surprisingly it is downturns in the overall economy that help, including the lagged influence of monetary policy. Relative price shocks also provide an explanation so supply as well as demand shocks have a role to play. The problem also shows considerable persistence. Thus in downturns unemployment is more than symmetrically large than in upturns and takes longer to fall than it did to rise. Taking the Phillips curve, Okun curve and IS curve results together gives us a somewhat better insight into the nature and causes of both asymmetry and nonlinearity in macroeconomic behaviour. Although, of course, some of the picture is clearly still omitted. It is clear that the variations across regions in labour markets and across sectors in product markets lead to important deviations in aggregate behaviour. When combined with the different national and sectoral responses to monetary policy, whether through the exchange rate or interest rates, this permits substantial departures from linearity. The asymmetries in the Phillips curve that we have explored appear to be primarily cyclical in character. The asymmetries in the Okun curve, on the other hand are more complex, reflecting not just cyclical factors but the degree of sectoral and regional mismatch in the operation of the labour market. There is thus not just a nonlinear underlying relationship but asymmetric departures from it. As the average level of unemployment falls so the scope for regional and sectoral disparities also falls as there is a lower bound. It seems likely therefore that there is more than one source of asymmetry. The structural mismatch in the labour market appears to be an additional cause to the traditional Phillips curve result.

9

We describe this as ‘churning’ in Mayes (1996).

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

144 Asymmetry and Aggregation in the EU

The asymmetries are likely to interact. The asymmetric nominal rigidities implicit in the Phillips curve are likely to contribute to the asymmetric labour demand effects revealed in the Okun curve. Downward rigidities in prices and wages would tend to increase the variance of unemployment. The different sectoral responses to monetary policy will be a reflection of this. Asymmetric shocks will interact with the nonlinear responses and asymmetric processes themselves. When combined with the policy reaction this generates a considerable identification problem (as explained by Blinder and Solow (1973) in the case of fiscal policy and Haldane and Quah (1999) for monetary policy). In their tests of causes of asymmetry in the Phillips curve Dupasquier and Ricketts (1998) are able to isolate some evidence for the hypotheses of costly adjustment, capacity constraints and misperception (of aggregate and relative price shocks). The nominal wage resistance hypothesis was not obviously sustained, a result consistent with Yates (1998). Although to some extent these causes should be separable the results from their joint inclusion were not well determined. Eliasson’s (1999) finding that the Phillips curve, using unemployment not an output gap as the determining variable, shows different sources of nonlinearity in Sweden and Australia is also helpful. In the Swedish case it is the rate of change of inflation expectations that is important, while for Australia it is the rate of change of unemployment.10 The former case will have particularly important implications for the conduct of monetary policy. Moreover the fact that the sources of nonlinearity differ for these two countries and are not found in the case of the US emphasises the potential problem of aggregation that we have outlined for the euro area. We primarily focus on asymmetries stemming from the behaviour of the labour market. Rapid downturns in the economy appear to have more than proportionate downward effects on unemployment, partly because of mismatch between the sectors and regions where the jobs and unemployed lie. This effect is likely to be greater in the EU where labour mobility is lower than in the US where the phenomenon is already clear. A slower response to adverse shocks makes recovery phases longer and unemployment persistent. However, these forms of asymmetry have a rather different impact on the inflationary process. The straightforward asymmetry, inherent in the convexity of the Phillips curve, is that excess demand in product or

10

Buxton and Mayes (1986) also made this finding for the importance of the rate of change of unemployment in the case of the UK.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Output, Unemployment and the Labour Market: The Okun Curve 145

labour markets has a significant upward effect on inflation while deficient demand has little or no effect on lowering inflation. The process is however more complex as the dynamics suggest that big differences between sectors and regions distort the picture. It is the existence of tightness in parts of the labour market that affects overall inflation and average unemployment and by analogy probably tightness in sectors of the product market that tends to intensify the inflationary pressure. Thus our findings indicate that in each example we have considered, ignoring the disaggregated problem will tend to result in misleading policy conclusions. The asymmetry is not restricted to demand shocks, as supply shocks, particularly through the exchange rate and foreign sector, can have sharply differing impacts both across the member states of the EU and across the sectors of industry. The traditional implication for policy set out in Laxton et al. (1995) is that monetary policy will need to be set somewhat more restrictively than is implied by linear symmetric models. However, it is also likely that any ‘new economy’ effects, where faster noninflationary and higher unemployment growth develops, may occur in the areas of high demand and relative labour shortage (Oliner and Sichel, 2000). Hence the implications of the asymmetric effects, observed in data from the past may need to be rethought if major sectors in the economy are undergoing structural change in their responsiveness and flexibility.

6.3

Some wider concerns

The EU has continued to struggle with the problem of unemployment and until recently has found it quite difficult to obtain reductions (Figure 6.1). It is perhaps no surprise therefore that the relationship

Figure 6.1

Convergence of unemployment in the EU 60

50 Unemployment 1987–1998

Unemployment 1987–2006 50

40

40 30 30 20 20 10

10 0

0 2

4

6

8

10

12

14

16

18

2

4

6

8

10

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

12

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

146 Asymmetry and Aggregation in the EU

between output and unemployment has been largely unchanged by monetary union (Table 6.1). One feature which does come through, however, is that the theoretical curvilinear aspect to the relationship does hold. The impact of changes in output on unemployment is much greater in the down phase of the cycle than it is in the up phase, and if anything this dichotomy has strengthened since the start of stage 3. Thus the increase in output following a decline needs to be three times as large to restore unemployment to its previous level. This is a strong effect and reflects the difficulty the EU has faced in reducing unemployment. In this regard there seems to be little difference between the EU as a whole and the euro area, which is perhaps a little surprising as the employment record of the UK in the last ten years has been considerably better than the EU average. It is only in the years since 1998 we cannot estimate the effect of population shocks effectively (Table 6.1). In most countries, the working-age population has stayed relatively constant and labour supply has been more affected by pension policies as well as changes in the educational system. However, the effect of output on unemployment has been more or less the same. The nonlinear nature of this relationship can most clearly be seen from Figure 6.2. The effect of

Figure 6.2

Comparison of income (per capita) and unemployment dispersion

0.24

2.4

0.22

2.2

0.20

2.0

0.18

1.8

0.16

<--Income Unemployment -->

1.6

95 96 97 98 99 00 01 02 03 04 05 06

Income dispersion is measured by the coefficient of variation of income per head in different regions and unemployment dispersion by the standard deviation of unemployment rates. The values are medians of country values.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Output, Unemployment and the Labour Market: The Okun Curve 147

output on employment is far larger in bad times than in good times. Thus, deep recessions show up in layoffs and a halt in recruiting activities. In good times, output growth lowers unemployment but the effect is much smaller. This may also reflect the fact that in good times, labour supply increases which partly nullifies the output effect. The result is rather robust with respect to different threshold model specifications but there is some ambiguity in terms of variables, most notably with respect to the output variable. Thus, with the output gap, the threshold effect is less apparent than with GDP growth. Comparing the movements of output gap and output growth (Figure 6.3) gives some hints of possible explanations (output growth appears to be more volatile and it also appears to lead output gap with a couple of quarters). It is also possible to consider the individual country coefficients from applying our quarterly model in Table 6.1. As with the annual data most countries follow the pattern of stronger responses of when output is below the threshold than above, with Ireland and Luxembourg being the only counter-examples, although Italy, Portugal and Germany show little difference between the two regimes. Given that these countries are spread over all four of the normally used models of welfare capitalism Figure 6.3

Indicators of output

0.100 0.075 0.050 0.025 0.000 –0.025 –0.050 –0.075 –0.100 1970

growth gap un gap 1975

1980

1985

1990

1995

2000

2005

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

148 Asymmetry and Aggregation in the EU

Output, Unemployment and the Labour Market: The Okun Curve 149 Figure 6.4

Country-specific coefficients of a threshold model for the Okun curve

0.2

0.0

–0.2

–0.4

–0.6

–0.8 GAP<0

GAP>0

Figure 6.5

Variance of the unemployment rate across countries

24

20

16

12

8

4

0 1970

1975

1980

1985

1990

1995

2000

2005

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

–1.0

150 Asymmetry and Aggregation in the EU Figure 6.6

Variance of GDP growth across countries

24

20

16

12

8

0 1970

1975

1980

1985

1990

1995

2000

2005

(Sapir, 2006) it is very difficult to say what sorts of institutional environment is particularly susceptible to this form of asymmetry. All appear to exhibit it. It is not immediately apparent whether this similarity can be attributed to other features of the data. The variance of unemployment rates across Europe has diminished rapidly in recent years (Figure 6.5) but for much of the data period it was considerable. One can see the effect of the Nordic crises on both unemployment and the variance of GDP (Figure 6.6).

6.4 Long-term forces and their asymmetric impact on the labour market Many of the long-term processes affecting the EU are external and cannot be reversed by EU policy. Tackling unemployment is thus a major problem and the reverses of the present crisis have undone the slow progress made over the last decade and quite possibly longer in many countries. Two features of this economic change stand out

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

4

in the EU context. First, part of the longer-term underlying pressure comes from an increase in the pace of innovation as global competition increases. Second, part of it comes from a reduction in the barriers between countries, which tends to concentrate production, as economies of scale can be exploited. There are thus forces that change both the nature of what is produced and the location of where many goods and services are produced. Although, the exact nature of that process of relocation and whether or not it encourages concentration and specialisation has been heavily debated (see Fujita et al. (1999) and the new economic geography literature, for example). While it is possible that some of these changes have a one-off step impact on the euro countries, leading to a change in structure that can then be sustainable in the future, much of the discussion is in terms of a change to dynamic behaviour. Patterns of production and the products themselves are not likely to be so enduring in the future. In this sense therefore the corresponding solution does not involve just a one-off burst of effort, after which the system can return to normal, but a continuing change in behaviour. The immediate impact of these pressures comes from the impact on employment and employability. It is not just a matter of the change in the nature and location of production but of the change in the skills required to produce the new goods and services. To become reemployed many people may require both extensive retraining and need to move somewhere else. These are heavy and difficult demands upon them. Not only will their incomes be cut in the short run but years of experience will become irrelevant and their earning capacity in the future may be sharply reduced. If a move in location is required all the investment in the previous location, not just in economic but in social terms will tend to be lost. If a family has been in a particular location for a long time, the ability to call on friends to help, the trust with local traders, the network of contacts will be very considerable. So a difficult choice may be offered between a decline in standing, income and wealth at home and a rather smaller decline in income and wealth but much greater initial reduction in social inclusion in a new location. Even without considering all the other incentives and difficulties these factors alone suggest that it may be more difficult to get people back into employment than it was to push them out of it in the first place. Hence the process of change will not be symmetric. The process of recovery may be more difficult to achieve than just a reversal of the

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Output, Unemployment and the Labour Market: The Okun Curve 151

decline in demand that led to downturn in activity. Ironically therefore large favourable shocks may also be difficult to accommodate even when there is substantial unemployment as the resources may not be available either with the right skills or in the right place. In any case technical progress will tend to mean that the passage of time alone is sufficient to ensure that less people are required to produce any given level of output. Thus even if retraining and mobility are highly successful mechanisms for re-employment it would take substantial periods of economic growth to be able to drive aggregate employment back to the levels it had before the shakeout. For a variety of reasons unemployment rose in Europe during the 1970s, 1980s and 1990s to above 10 per cent and participation rates similarly declined.11 Although there has been clear progress in reversing the position up till the present crisis, the prospects for a rapid improvement in most regions are poor. An increase in the underlying sustainable rate of growth of half or even 1 per cent a year would still be required for at least a decade just as set out in the Lisbon Strategy, which was itself not successful. Although success is relative and without the Lisbon Strategy outcomes might have been even worse. Most of this discussion has been held in the context of the labour market, although there is not a one to one mapping from constraints on the rate of growth to employment. Factors affecting the level of unemployment have been clearly set out in Nickell (1998), Blanchard (1998) inter alia. The big factors affecting the differences in the levels of unemployment across the OECD countries have been the replacement rate (ratio of income out of employment to that in employment), the length of time for which benefits can be received and trade union density and coverage, although these latter two can be offset by cooperation between unions and employers. The tax rate has a limited positive relationship with unemployment and the change in the rate

11

Muffels et al. (2002), OECD Employment Outlook. In individual countries the pattern has differed with female participation rates increasing in some countries, particularly when part-time employment is taken into account.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

152 Asymmetry and Aggregation in the EU

of inflation a small negative one.12 In time series studies Nickell and Bell (1995) find that real interest rates are positively associated with unemployment but this will tend to be more of a cyclical phenomenon, as no one disputes that there is a clear short-run relationship between monetary policy and unemployment. Whether one prefers to think in terms of the natural rate or the NAIRU in this context the fundamental determining factors are the same and largely not related to macroeconomic policy but to the characteristics of the economic system: bargaining institutions, legislative impediments to free movement of labour etc. (In Kilponen et al. (2000) we show that the independence of the central bank also has an impact on unemployment – more independence is associated with lower unemployment other things being equal, contrary to the common prejudice that central banks seek to increase unemployment.)13 However, increasing the flexibility of the labour force also costs public money, where it is not just a matter of legislation, and the provision of incentives to earn incomes rather than receive benefits may affect the amount of revenue that can be raised. There is thus an interaction between macroeconomic policy and the structural determinants of unemployment. The revenues used to provide extensive reskilling have to raised somewhere. Even if they are privately financed they will have indirect implications for public revenues. The interaction is much more obvious in the case of redistribution of income and wealth through the tax and benefit system. Here the process of redistribution itself directly affects people’s incentives. Those earning the incomes that are taxed get a reduced incentive to earn extra money, while those receiving the benefits may also get a reduced incentive to earn. Thus while their wellbeing may be increased simply through receiving the benefits their incentive to escape from unemployment or exclusion may be reduced. Hence, while the initial redistribution may have a strong positive effect on welfare because it takes people out of extreme poverty the more extensive it becomes the greater chance that the overall impact on incomes and wealth may be negative. This applies more generally to the intervention of the public sector in raising revenue and spending the proceeds. According to the traditional Laffer curve, while

12 13

Daveri and Tabellini (2000) take a harsher view. Mayes and Silverstone (1998) set out a framework for relating.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Output, Unemployment and the Labour Market: The Okun Curve 153

there will be positive benefits from the initial public spending (financed by taxation) the benefits diminish with the tax rate and ultimately turn negative. There is thus a potential conflict between the benefits from doing more about unemployment and the aggregate benefit to the economy insofar as this involves public spending. This is particularly clear for Nordic countries, which appear to have used public employment deliberately as a means of trying to reduce unemployment in the economy as a whole (Agell et al., 1997).14 In long-run equilibrium the growth of the economy will follow the size of the labour force and the underlying rate of growth of productivity. The problem in the interim is to decide the extent to which the euro economy has been shocked away from that long-run path, as opposed to moving to higher long-run levels of unemployment as a result of increasing rigidities in the economy. The problem for economic analysis is that it is very difficult to distinguish between any increase in the long-run sustainable level of unemployment and a slow acting return to a low equilibrium – what has been labelled the problem of hysteresis. If it takes 20 years for the economy to return to previous levels of unemployment after the underlying shocks, starting as far back as the first oil crisis in 1974, it is not surprising if high levels of unemployment are still with us, even if most of the news since the mid-1990s had been largely favourable until the present crisis. Indeed, in trying to draw up any long-run projections of what might happen to unemployment and the level of incomes per head, it is necessary to consider the extent of the favourable shocks of recent years as well. Favourable shocks will also have temporary effects, although the temporary nature will be rather different. Looking at the asymmetry between losing one job and gaining another from this point of view, one will still argue that it is difficult for the economy to respond to the favourable shock, as people need to be trained, but easier to let them go if the stimulus disappears. Thus the more favourable experience of the period since the mid1990s may be due to the slow return towards the long-run trend after the period of unfavourable shocks, overlaid with the beneficial effect of recent favourable moves – the development of the single market, Stage 3 of EMU etc. At some stage both these sources of faster growth will come to an end. The question is at what level of unemployment that will be.

14

See Koskela and Virén (2000) and Mayes and Virén (2001).

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

154 Asymmetry and Aggregation in the EU

If in the intervening years nothing has been done that raises the determinants of the underlying level of unemployment then ipso facto the EU can return to the more favourable experience of the 1960s – eventually. This has not been primarily a discussion about productivity. The question at issue here is, given the rate of growth of productivity, what will be the level of unemployment. The two cannot of course be separated in any easy manner. The rate of productivity growth will be due to exogenous and endogenous factors. The extent to which the rate of innovation and hence the basis for productivity growth is demand driven, supply driven or purely a random walk is still debated (Scherer, 2001). While there are fairly predictable fluctuations over the economic cycle productivity growth rates have tended to be fairly stable over quite long periods (Gordon, 2000). The 1970s appear to have been a period of relatively slow productivity growth and the jury is still out as to whether there are good grounds for thinking that there has been an improvement since the mid1990s as a result of the ‘new economy’. Faster growth rates do not necessarily have an impact on unemployment but they do necessarily have an impact on wealth. Thus faster productivity growth will be largely unambiguously good news for being able to do more about social exclusion and the ability to redistribute incomes to those most in need. Unless of course such redistribution itself reduces that rate of growth back to its previous levels. Whether faster growth is associated with higher or lower unemployment will depend upon just the same institutional factors about how the labour market institutions work and the degree of match between the demand and the supply. For short periods of time the growth can occur through a better utilisation of either capital or labour but as an enduring process it will depend on their rate of growth and on the rate of growth of their productivity. Although there is broad agreement among policy-makers on the most fruitful directions for change, it is much more difficult to predict how big an impact recommended changes will have on growth or employment. The IMF has hazarded some suggestions in some of its Article 4 assessments of OECD countries. Typically regulatory, institutional and other structural factors are assumed fixed in macroeconomic models, so such models can tell us little in the present beyond exploring the initial impact of shocks to the system. The results in Nickell (1998), discussed above, suggest the size of the reductions in unemployment that might be achieved by reducing relative benefits and their duration, weakening the power of or increasing the cooperation with the trade unions, increasing active labour market policies and perhaps reducing the overall rate of taxation. But these

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Output, Unemployment and the Labour Market: The Okun Curve 155

ignore collateral effects on social exclusion such as a reduction in job security and that the gains may not filter through to some of the most disadvantaged groups (as Mayes and Silverstone (1988) show with the New Zealand reforms). Moreover, the quantitative impact of the increased investment under the Lisbon strategy, particularly in knowledge-based skills and other aspects of human capital remains speculative. There is a third macroeconomic factor added by the process of Economic and Monetary Union. Namely that the EU has chosen not to be able to make as full use of some of the mechanisms of more focused macroeconomic policies in the solution, than was the case in the past. By having a single monetary policy individual member states cannot allow their exchange rates to move as part of the adjustment process, as was strikingly the case in the adjustments after 1992. Similarly, the Stability and Growth Pact limits the degree to which counter-cyclical fiscal policy can be used. (Although Artis and Buti (2000) argue that in practice there is rarely likely to be any constraint on prudent behaviour.) Furthermore, the rules of operation of the single market ban on the use of state subsidies and other actions that would discriminate in favour of one member state’s firms at the expense of the others’. In the same way member states cannot discriminate in favour of employees (or the unemployed) according to whether the come from that or another member state. Typically macroeconomic models do not try to endogenise the NAIRU or natural rate simply because that involves regulatory, institutional or other microeconomic factors. Thus we can run simulations with the EDGE model of the euro area (Kortelainen and Mayes, 2001), which show how rapidly a reduction in the NAIRU would be translated into an actual fall in unemployment, depending upon how quickly the authorities and the private sector recognise that the change has happened. This, however, is rather uninformative in the current context, as it does not tell us how the fall in the NAIRU has been generated. (The NAIRU is a more difficult beast to deal with in these circumstances since, unlike the natural rate, it also includes the effects of the lagged adjustment of behaviour. The NAIRU thus restricts the speed at which one can return to the natural rate.) We need to return to the discussion of the underlying determinants. If the results in Nickell (1998) could be translated into recommendations for action then it would be possible to have a substantial impact on unemployment by reducing relative benefits and their duration, weakening the power of or increasing the cooperation with the trade unions, increasing active labour market policies and perhaps reducing the overall rate of taxation. (There is other evidence that reducing corporate taxation tends to attract investment but this is relative to other countries.)

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

156 Asymmetry and Aggregation in the EU

It seems therefore relatively unlikely that the current National Action Plans in the EU’s employment strategy will be sufficient to reduce the problems of unemployment in many of the member states to the sorts of levels that are desired. With unemployment persistently at higher levels it seems unlikely that social exclusion will also be amenable to reduction to acceptable levels. More will therefore have to be done but largely in areas which are politically difficult as the case of Greece illustrates. However, most of the emphasis in the discussion has not gone on the level of unemployment, which is our principal focus here, but onto the issue of stabilisation. The principal concern is that the SGP, by capping deficit spending, limits the ability of countries to respond to adverse shocks, thereby exacerbating the consequence of the loss of the exchange rate as a weapon for individual euro countries. As mentioned earlier we think that the evidence for this is rather limited as it is not clear that these limits do anything other than require what a prudent government would have done in any case (Artis and Buti, 2000). The balance of fiscal policy is not a very potent weapon for reaping fundamental changes in unemployment. Indeed, if, as in the case of the buffer funds developed in Finland, it is a means of increasing the replacement rate, it may serve to stabilise employment but make the task of getting the unemployed into employment more difficult. Taken together therefore we present a somewhat mixed message. There are some reasons for thinking that the cyclical impact of EMU will not be as disadvantageous as some people have worried. However, on the other side of the scale there are good reasons for thinking that the euro area will continue to be rather asymmetric in its responses to policy over the coming years making adjustment relatively difficult and perpetuating some of the differences among countries, particularly in terms of unemployment. Furthermore we showed in Chapter 5 that having a wider regional spread in unemployment exacerbates the impact of any given level of unemployment on inflation. Nevertheless despite these difficulties and differences there could be a substantial payoff from macroeconomic policy coordination. The message that some greater degree of fiscal federalism may be helpful but at lower overall tax rates is likely to be one that will not be well received among the governments of the member states. The present crisis has largely brought an end to discussions of the longer term as governments struggle to cope with the present. It is too soon to say whether the consequence of the crisis is a substantial pause in the process of recovery or a significant downgrade in the possibilities for longer-term growth.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Output, Unemployment and the Labour Market: The Okun Curve 157

7

We consider two contrasting issues relating to the role of the public sector in this chapter. The first relates to the role of the public sector in the economy and the symmetric pressures that fall upon it. The second relates to the EU as a whole and a problem of ‘aggregation’. In this second case, it is argued that a degree of coordination among the EU countries may ease the fiscal problems of each of the member states because of the spill over benefits. There is a third issue, namely, asymmetries to the balance of fiscal policy, which are reflected in the highly asymmetric nature of the Stability and Growth Pact in the EU that guards against excessive deficits without any matching restraints on surpluses. However, this is such a significant topic that it has its own chapter (Chapter 9). So we begin with the asymmetric pressure on fiscal policy. We have already noted that asymmetries in the process of growth and the generation of inflation pose asymmetric pressures on fiscal policy, encouraging relatively strong action in downturns both to head off the heavy re-employment costs of any ‘unnecessary’ job losses and discouraging the creation of further inflationary pressure when the output gap is positive. A natural reaction to this might be to try to increase the role of the public sector in combating unemployment but to maintain fiscal prudence by raising or at least not cutting taxes. Such an action would encounter the next facet of asymmetry we consider in this chapter. Downturns in the economy tend to shake out the less efficient. In most industries this implies firms with relatively low labour productivity. Hence this process of shake out in itself tends to improve average productivity but add substantially to unemployment. When the recovery comes, new entrants tend to be relatively efficient, as they not only have to have an ability to match and quite probably undercut existing firms if they are to 158

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Asymmetry and the Role of the Public Sector

overcome the barrier of being less well-known but they have to finance all the other start up costs, which would tend to add to the overall cost. Thus they will be employing fewer people for any given output, once the economy has returned to its previous level or trend in output. Unemployment thus tends to lag economic recovery. This generates a considerable political difficulty as the unemployed and their adult families are voters. Not only that, but unemployment is regarded as socially undesirable and a waste of resources. There are thus strong pressures to get people back into work. Governments have responded to this in a number of respects but most strongly through what are described as active labour market policies. Thus, irrespective of where countries fall in the traditional categorisation of welfare regimes (Esping-Andersen, 1990; Sapir, 2006) there is now much more emphasis on encouraging people to get jobs through providing better information on vacancies, reducing the costs of job search, providing or subsidising retraining etc., rather than simply offering income support. However, public employment is itself also a response. To some extent this is involuntary as more staff are needed to handle the increased numbers of payments. Moreover, as was vividly illustrated in the present crisis, governments may feel forced to takeover failing firms in crucial sectors, whether essential services or services such as banking where one failure might lead to the progressive collapse of the sector. While it is easy to get drawn into increased public spending in the circumstances of a crisis it can be very difficult to run the expenditure down again when the economy picks up, except where this is automatic. Norway and Sweden both took major stakes in banks in the Nordic crises of the early 1990s and much of this ownership was still in place when the present crisis struck, despite many years of prosperous growth when the shares could have been unloaded very profitably.1 What tends to result therefore is a ratchet process, where problems or the desire to introduce new services increase the size of the public sector. However, when circumstances improve or services might be provided by the private sector the incentives to act are no longer so strong. Tax increases can be imposed in periods of difficulty but much in the way of tax increases comes naturally through inflation and from increases in income. With a progressive system more people progress to the higher rates and corporate taxation is highly nonlinear with respect to the 1 The Norwegian government has a 30 per cent holding in DNB-nor and Sweden a substantial stake in Nordea as a result of its ownership of Nordbanken, one of the two original partners in Nordea.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Asymmetry and the Role of the Public Sector 159

growth rate, for example. The UK over the last decade is a case in point where the public sector’s share of GDP has increased from 35 per cent to 45 per cent. Changes in government also provide a regime switch that can result in a reduction in both taxation and expenditure but for substantial parts of the post-war years the role of the government increased once the wartime economy had been unwound. Clearly this process is bounded – literally of course by ending up in a completely socialist state where the private sector has been eliminated – but more straightforwardly because diminishing returns set in and even in the most enthusiastically social democratic countries there is a limit to how far the electors want to see private activity diminished. Similarly, at low levels of expenditure state intervention is so obviously beneficial not simply to increasing equity but to improving the rate of growth through key infrastructure investment which the private sector cannot organise. Our focus is therefore on the range between the extremes where the economic case is debated and political views differ so changes in regime are possible.

7.1 Limits to increasing the public sector share in the economy It is a feature of many macro-models of advanced economies that increasing public sector consumption will reduce the overall level of economic activity in the economy, primarily because it will increase the share of lower productivity activity in the economy; see the Bank of Finland’s EDGE model of the euro area for example (Kortelainen and Mayes, 2001). This is of course a contentious finding (see Koskela and Virén (2000) for a short survey of the literature) as both positive and negative effects have been found in empirical studies. In this section, following Koskela and Virén, we argue that the relationship is nonlinear with a positive effect of increased public sector employment on overall activity at low levels of public sector employment and a negative effect at high levels. Increased public expenditure on the physical, technical and human capital infrastructure will tend in particular to be of aggregate benefit, as the private sector tends to deliver suboptimal quantities when unaided. The assumption of the negative effect in Kortelainen and Mayes (2001) reflects the finding that the euro area as a whole is in the negative section of the relationship. Hence the policy implication is that a switch to greater public spending as a route out of current difficulties may well not be beneficial. Indeed a cut may be desirable for some member states.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

160 Asymmetry and Aggregation in the EU

Asymmetry and the Role of the Public Sector 161

∆log Dt = α + β∆log Lg,t–1 + γ∆log Dt–1 + ut

(7.1)

into two groups according to whether they are on/above or below a particular threshold for the share of the public sector in GDP. Here we consider government consumption, G, as a share of GDP at current prices, Y, (G/Y) as the threshold variable, although Koskela and Virén also consider labour shares and consumption shares. However, the results in those two cases are similar so we do not pursue them here. The hypothesis is that the parameter β differs according to whether G/Y is above or below the threshold value (G/ˆY ) ∆log Dt = α + β1∆log Lg,t–1 + γ∆log Dt–1 + et

if

G/Y ≤ (G/ˆY )

(7.2a)

∆log Dtt = α + βz∆log Lg,t–1 + γ∆log Dt–1 + et

if

G/Y > (G/ˆY )

(7.2b)

We expect to find β1 > 0 > β2. It is clear from Table 7.1 that the coefficient of β1 is positive except for a couple of cases and in all cases larger than the coefficient β2. β2 is negative except for Australia, France, Italy and Norway and in these cases the β2 coefficients are not significant. The estimation and test procedures made use of a GAUSS routine, available on Bruce Hansen’s homepage: http://www.ssc.wisc.edu/, which searches for the value that gives the smallest residual variance. According to the estimation results the threshold value of the public sector (output) size varies between 10 and 30 per cent. Koskela and Virén (2000) report that the threshold value is higher with the public sector share of total

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

The implications for using extra public spending to attempt to solve problems of unemployment thus face a prima facie difficulty. If this spending is not abnormally productive it will tend to result in a net loss to society, which may in itself tend to increase unemployment. Indeed the general wealth of the economy might be increased by a shift towards less public spending. It would depend on the nature of the redistributive process both in incomes and employment whether this diminished or increased unemployment, social exclusion or some other measure of social deprivation – a point not amenable to macroeconomic analysis. Using annual data for 22 OECD countries for the period 1960–96, it is possible to show (Table 7.1) first of all that a nonlinear model can represent the relationship between public sector output and overall output and secondly that over the relevant range the relationship is negative. The equation fitted is a simple application of the Granger and Teräsvirta (1993) threshold model (see Chapter 2). Denoting private sector output by D and public sector employment by Lg, we can divide the observations on the variables in (7.1)

162 Asymmetry and Aggregation in the EU Threshold model estimation results G/Y as the threshold variable ˆβ1

ˆβ2

SEE/DW

FHO

FHT

LM

Australia

0.365 (1.74)

0.049 (0.51)

0.025 (2.069)

18.9 (0.051)

9.7 (0.016)

2.14 (0.154)

Austria

0.580 (1.71)

–0.568 (1.69)

0.019 (1.759)

20.4 (0.046)

11.52 (0.003)

0.39 (0.538)

Belgium

0.690 (2.34)

–0.119 (0.48)

0.023 (2.159)

36.7 (0.000)

7.5 (0.119)

0.120 (0.283)

Canada

0.370 (1.57)

–0.751 (1.26)

0.027 (1.714)

4.9 (0.865)

4.3 (0.663)

2.86 (0.104)

Denmark

0.113 (0.80)

–0.700 (2.86)

0.024 (1.833)

19.2 (0.046)

7.2 (0.167)

1.63 (0.212)

Finland

0.458 (1.68)

–1.144 (2.25)

0.032 (1.648)

10.6 (0.308)

3.6 (0.876)

1.25 (0.274)

France

1.417 (3.23)

0.121 (0.25)

0.017 (1.961)

12.9 (0.270)

8.7 (0.028)

0.002 (0.966)

Germany

–0.063 (0.80)

–1.537 (3.64)

0.023 (1.767)

14.6 (0.138)

6.3 (0.283)

0.98 (0.331)

Greece

0.933 (1.98)

–0.354 (1.39)

0.031 (1.734)

26.6 (0.007)

11.6 (0.003)

0.16 (0.696)

Iceland

0.138 (0.61)

–1.021 (1.61)

0.040 (1.813)

5.25 (0.862)

3.8 (0.830)

0.95 (0.338)

Ireland

–0.109 (0.44)

–0.941 (1.89)

0.029 (1.947)

7.1 (0.697)

7.2 (0.177)

0.04 (0.845)

Italy

1.278 (3.28)

0.293 (0.99)

0.022 (1.785)

12.7 (0.221)

5.9 (0.415)

0.89 (0.354)

Japan

1.325 (2.16)

–0.880 (2.55)

0.024 (2.366)

24.3 (0.024)

6.8 (0.237)

2.74 (0.108)

Netherlands

0.156 (0.67)

–1.617 (4.37)

0.013 (1.868)

24.6 (0.040)

6.3 (0.210)

0.27 (0.605)

New Zealand

0.418 (1.06)

–0.697 (1.69)

0.037 (2.047)

15.8 (0.129)

6.1 (0.360)

0.04 (0.853)

Norway

0.448 (1.54)

0.159 (1.01)

0.019 (1.642)

7.6 (0.663)

7.3 (0.131)

5.05 (0.033)

Portugal

0.169 (1.37)

–0.153 (1.23)

0.032 (2.076)

7.7 (0.521)

3.5 (0.935)

0.28 (0.603)

Spain

0.186 (1.14)

–0.172 (1.57)

0.020 (2.272)

17.6 (0.096)

5.8 (0.359)

0.31 (0.584)

Sweden

0.330 (1.90)

–0.123 (0.88)

0.022 (1.673)

12.8 (0.222)

7.7 (0.117)

3.99 (0.055)

Country

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 7.1

Asymmetry and the Role of the Public Sector 163 Table 7.1 Threshold model estimation results G/Y as the threshold variable – continued ˆβ1

ˆβ2

SEE/DW

FHO

FHT

LM

Switzerland

0.325 (1.06)

–0.904 (2.05)

0.022 (1.407)

15.2 (0.106)

5.4 (0.449)

7.51 (0.010)

UK

0.628 (1.74)

–0.131 (1.04)

0.024 (1.488)

7.9 (0.636)

4.6 (0.681)

12.41 (0.002)

USA

0.551 (1.62)

–0.008 (0.03)

0.024 (1.594)

8.8 (0.491)

3.9 (0.876)

10.72 (0.003)

Country

consumption and lower with the public sector share of total employment. With all threshold variables, the threshold values are very similar for all countries suggesting that there is some invariance across countries. The results are of course rather tentative owing to the very small sample size for each country. This can to some extent be eased by pooling the data and estimating the panel of 22 countries by seemingly unrelated regression estimation (SURE), as is shown in Table 7.2. In addition to linear and threshold models Koskela and Virén also estimated a multiplicative specification of the following form: ∆log Dt = α + β∆log Lg,t–1 + γ∆log Dt–1 + φHt • ∆Lg,t–1 + et

(7.3)

where H denotes the threshold variable (G/Y). According to this specification the public employment effect depends on the interaction term Ht • ∆Lgt–1 and thus on the size of the government sector. According to our hypothesis φ should be negative. Using this specification we can compute the critical (or, in a sense ‘threshold’) value of this variable at which public sector employment growth has zero effect on private sector output growth. The results with panel data conform with the results from individual country data. With a linear model there is no relationship between public sector employment and private sector output, while with the threshold model quite a clear relationship is obtained. There is a similar relationship using the multiplicative specification (7.3) in which the public sector

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Numbers inside parentheses below the coefficient estimates are t-ratios. SEE is the standard error of estimate and DW the Durbin-Watson test statistic (which here suffers from the bias caused by lagged dependent variable). FHO denotes the LM (F) test for no threshold and FHT the corresponding test for threshold allowing for heteroscedastic errors. Numbers inside parentheses below the F statistics are bootstrap probability values. Finally, LM denotes a LM test for first-order autocorrelation of residuals (corresponding marginal significance levels are inside parentheses). When computing this LM test we have utilised Chan (1993), in which it is shown that the threshold parameter is superconsistent and can thus be treated as a known parameter.

164 Asymmetry and Aggregation in the EU Table 7.2

Estimation results with panel data β/β1

Linear G/Y Threshold Eq (7.3) with H = G/Y

–0.020 (0.12) 0.131 (2.64) 0.404 (5.55)

β2

γ

–0.058 (1.35)

0.294 (8.59) 0.325 (9.35) 0.281 (8.35)

φ

SEE/R2

ˆ H –

–2.460 (5.85)

0.028 0.171 0.027 0.185 0.028 0.186

0.157 0.164

Data sources for Tables 9 and 10: Y Gross Domestic Product at current or constant 1990 prices, OECD National Accounts, CD-ROM, OECD, Paris. G Public consumption or public sector (i.e. producers of government services) production, both at current or constant 1990 prices, OECD National Accounts, CD-ROM, OECD, Paris. Lg Public sector employment (thousands of persons). Employment in the Public Sector, OECD 1982, Paris; OECD National Accounts, CD-ROM, OECD, Paris; and some national sources. Data available from Virén upon request. Lp Private sector employment (thousands of persons). (Data source as Lg.)

employment effect depends on the size of the public sector. When the size of the public sector increases, the employment effect diminishes and, after some critical value, becomes negative. The implied critical values are, in fact, quite close to the average threshold values in the context of threshold model estimation. There is other evidence to support these findings. Karras (1996) has estimated the optimal government size by exploring the role of public services in the production process. As the theoretical framework he takes the analysis by Barro (1990), according to which government services are optimally provided when their marginal product equals unity (the so-called ‘Barro rule’). He finds for a data set of 118 countries over the period 1960 to 1985 that in some cases government services are over-provided, in some cases under-provided and in many cases optimally provided. The optimal government size in the Barro sense is 23 per cent (±2 per cent) for the average country, which number, however, masks important differences across regions. Interestingly, this number is not very far away from the value of the thresholds shown in Tables 7.1 and 7.2. Our analysis is less ambitious in the sense that we do not study the welfare issues. The key conclusion from this section

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

All estimates are SUR estimates with panel data consisting of 736 data points. All equations also include country intercepts, which are not reported. The threshold models (columns 2–4) are estimated using the average values of the threshold variable from the single country models. With the multiplicative model (the last three set of estimates) the threshold values are derived from the estimates of β and φ.

Asymmetry and the Role of the Public Sector 165

in the context of current European macroeconomic policy is that it supports the current inclination to try to cut taxation of labour as part of the Lisbon Strategy and to increase growth and employment.

Stabilisation and policy coordination

Our second concern in this chapter is with the scope for improved outcomes that might stem from fiscal policy coordination under EMU and under the SGP in particular. If the ability to coordinate is increased then this may help offset some of the disadvantages from the inability to run an independent macroeconomic policy. (It is of course always debatable the extent to which there was scope for independent action by the smaller countries in the previous regime, as in the main they had to follow the German lead because their economies were so integrated.) The EU does not attempt fiscal coordination in a strict sense of the word – there are no directives to the member states telling them how fiscal policy is to be set as part of some annual ‘plan’ – but there is what the European Commission (2002) describes as ‘weak coordination’ through the Broad Economic Policy Guidelines (BEPG). Second there is a set of rules on how budgetary balances may be set, laid out in the SGP (described by the European Commission (2002) as ‘strong coordination’). The formulation of the BEPG is a complex annual process, orchestrated by the Commission, aimed at trying to ensure that the macroeconomic policies of the member states contribute to the overall goal of sustainable noninflationary growth that achieves full employment. Much of what is involved relates to structural policies, wage developments and labour market reform – the Cardiff, Cologne and Luxembourg ‘processes’ – but also involves the application of the SGP. While the BEPG have no legal force and rely on peer pressure for their achievement, the SGP does have some coercive powers, although despite breaches no penalties have as yet been imposed. The SGP has two main sides to it. The first is to try to ensure that the member states all achieve a strong and sustainable budgetary position. This involves progress each year towards having a low debt ratio. While the end point has not been defined the process involves trying to remain in surplus or in balance through the course of the cycle. In order to qualify for monetary union the member states were supposed to have a debt ratio of less than 60 per cent of GDP or be making sustainable progress to achieving that. While that state was somewhat liberally interpreted in 1998 when the original membership of the monetary union was decided it has nevertheless remained at the heart of the Commission’s

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

7.2

longer-term predictions. The second side to the SGP is the Excessive Deficit Procedure (EDP), which is designed to prevent deficits in any particular year exceeding 3 per cent GDP except in cases of severe economic difficulty, as applied to several member states in the present crisis. Like the BEPG the EDP is essentially forward-looking. If a country looks, in the view of the Commission, that it is going to run an excess deficit, then it has to take steps to try to avoid it. If these steps are not taken and an excess deficit appears, ultimately the member state has to make a noninterest-bearing deposit from which the remaining states benefit. This can ultimately be converted into a fine if action is not taken in a period of two years. The detail, as set out in European Commission (2002) as amended, need not concern us here. In Chapter 9 we look at the impact that this asymmetric EDP has in policy. In this chapter our concern is with coordination. Coordination in the SGP framework is largely a matter of the appropriate design of the system (Virén, 2000b). It is not realistic to think of negotiated decisions that would lead to one country following an expansionary policy in order to help offset a deflationary shock to another. This does not of course involve fiscal federalism, as this is not part of the current EU arrangements except in rather indirect manner through the structural funds. This is an area where the EU differs clearly from other countries and federal arrangements. Others have found the substantial ‘automatic’ transfer of resources from the ‘gainers’ to the ‘losers’ when shocks hit, on a scale not contemplated by the EU, to be a necessary part of the attack on social exclusion. The absence of such mechanisms in the EU has been a persistent source of criticism (see, for example, Feldstein, 1997). Politically, it is pretty clear that a substantial system of inter-regional transfers similar to those that apply in the US, Germany, Canada or other mature fiscal federations is implausible for the foreseeable future. This in itself constrains what might be possible in social policy, because with a budget capped at just 1.27 per cent of GNP, the EU level cannot aspire to engage in the forms of equalisation and redistribution that the economic theories of fiscal federalism would prescribe (see, notably, Oates, 1999). Yet it should not be overlooked that within member states, these mechanisms are, typically already well developed: Southern England, for example, manifestly transfers resources to the ‘North’, while in Italy the geographical transfer is North to South; Germany transfers from West to East, Ireland from East to West. Again, coordination can help to maximise the impact of such mechanisms.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

166 Asymmetry and Aggregation in the EU

Asymmetry and the Role of the Public Sector 167

1. The cyclical behaviour of the economies and the nature of shocks must be similar. 2. Countries must have similar prerequisites for policy actions. 3. The tax and transfer systems and the budgetary process must be similar so as to provide reasonably similar automatic stabilisers. 4. Forecasts and the assessment of the current situations must be sufficiently accurate. 5. Effects of fiscal policy actions must be reasonably similar and predictable. 6. The effectiveness of coordinated policy actions must be much larger than uncoordinated actions. 7. Different countries must share the same policy view (in terms of the instruments and objectives of policy). 8. Policy commitments must be enforceable in different countries. These requirements are all straightforward in nature. If problems are uncorrelated then joint action is less likely to be valuable. If countries do not behave in a fairly similar manner then having relatively uniform prespecified responses is unlikely to constitute an optimal policy. If we do not know what the impact of policy is going to be on the economies then it is much more difficult to decide what to do. Perhaps

2 The normal definition of automatic stabilisation relates to the fact that the tax and benefit system, widely defined, is contra-cyclical in nature. As the economy slows, tax revenues slow more than proportionately and unemployment starts to rise generating increased welfare payments and activity measures to try to get people back into work. Thus the budgetary position worsens on both the revenue and expenditure sides of the account. The reverse happens in an expansion and a ‘sustainable’ fiscal system should be able to go through the cycle without the need to change tax rate or expenditure rules.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

The key element in this discussion will therefore be the ‘automatic’ stabilisers.2 Automatic in inverted commas because this includes the normal response of governments, which as we have noted is often asymmetric, differing between upturns and downturns – we explore this asymmetry in detail in Chapter 9. We should not regard coordination through automatic stabilisation in this sense as being necessarily a favourable response (Blanchard, 2000) as this reaction is appropriate to demand shocks. Supply shocks can require quite the opposite response. Fiscal policy coordination in the sense we are describing has certain requirements for it to take full effect:

the most important element that has to be sorted out is a reasonably accurate decomposition of the key variables into their ‘cyclical’ and structural components (Brandner et al., 1998). Lastly the incentive structures must be adequate. If there is little to be gained from coordination but substantial costs (both economic and political) in precommitting to do so then coordination is less likely. Similarly if there are no adequate penalties for reneging the incentive to free ride on the system will be substantial. Given that is known, again countries will not cooperate. To assess the importance of policy coordination for policy effectiveness we use the NiGEM multicountry model to compare the effects of different fiscal policy actions in the single country setting and in the case of collective policy action.3 In the simulations (see Table 7.3) public consumption was first increased in all EU countries in an uncoordinated way (i.e. country-by-country). Then it was increased in all EMU countries at the same time and by the same amount (1 per cent).4 In all cases the coordinated fiscal expansion produces almost twice as much an increase in output as an uncoordinated fiscal expansion. (In Table 7.3 the insertion of the letter c in the variable name shows the results of the coordinated action, with the exact definitions of the variables shown in the footnote to the Table.) As expected we have the result that in uncoordinated actions small countries are able to achieve relatively little (mainly because of import leakage). The multiplier values (the last two columns in Table 7.3) reveal that in an uncoordinated case fiscal policy effects for the small countries are mainly only around 0.5. For large countries, the values exceed unity but not by very much. The average value for all countries is 0.72 (with four lags) and 0.63 (with eight lags), 0.85 being the average maximum value. In the case of coordinated policies, there is not much difference

3 In evaluating the effects of fiscal policy, an obvious analytical framework is provided by (structural) vector autoregression (VAR) models (see Blanchard and Perotti (1999), Dalsgaard and De Serres (1999) and Virén (2000a)). Because we concentrate here on the policy coordination problem, structural multicountry models are, however, more convenient. The model vintage used was 2002. 4 The share of public consumption in GDP differs somewhat across EU countries, and so the corresponding GDP effects also differ. The differences in the public consumption/GDP ratio are after all not so large as the following 1998 values indicate: Austria 18.7 per cent, Belgium 21.1 per cent, Denmark 25.5 per cent, Finland 21.4 per cent, France 24.2 per cent, Germany 19.0 per cent, Greece 14.8 per cent, Ireland 13.4 per cent, Italy 18.8 per cent, Luxembourg 14.0 per cent, Netherlands 13.6 per cent, Portugal 20.2 per cent, Spain 15.8 per cent, Sweden 25.9 per cent and UK 18.2 per cent.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

168 Asymmetry and Aggregation in the EU

Asymmetry and the Role of the Public Sector 169 Table 7.3

A summary of the public consumption simulation y4

y8

yc4

yc8

ymax

ycmax

def

defc

ym

ymc

Austria Belgium Finland France Germany Ireland Italy Netherlands Portugal Spain

0.059 0.099 0.124 0.273 0.224 0.065 0.147 0.107 0.092 0.166

0.042 0.074 0.151 0.261 0.156 0.054 0.128 0.090 0.076 0.159

0.162 0.233 0.175 0.333 0.304 0.232 0.208 0.211 0.156 0.246

0.143 0.208 0.228 0.332 0.224 0.189 0.189 0.195 0.157 0.274

0.107 0.113 0.159 0.274 0.299 0.066 0.156 0.121 0.116 0.175

0.279 0.239 0.268 0.339 0.374 0.233 0.212 0.219 0.241 0.274

–0.154 –0.220 –0.117 –0.168 –0.167 –0.127 –0.146 –0.230 –0.185 –0.157

–0.075 –0.107 –0.050 –0.144 –0.130 –0.079 –0.102 –0.144 –0.144 –0.109

0.574 0.536 0.741 1.130 1.574 0.488 0.829 0.891 0.574 1.109

1.489 1.131 1.251 1.398 1.967 1.740 1.128 1.612 1.193 1.732

Average

0.136 0.119 0.226

0.214

0.159

0.268

–0.167 –0.108 0.845 1.464

between small and large countries. Thus, the average value is 1.25 (with four lags) and 1.17 (with eight lags), 1.46 being again the average maximum value. This represents an improvement for all countries but a major one for the smaller countries. The multiplier values (in the coordination case) are, in fact, quite close to the values obtained by Cohen and Follette (1999) with the US FRB/US macroeconomic model.5 On the other hand, they are a bit higher than the SVAR values obtained by Blanchard and Perotti (1999), which are about one. The multiplier values in the uncoordinated case are, of course very low (suggesting that the marginal propensity to spend out of income is very low and the income elasticity of imports is very high) but also in the case of coordinated fiscal policies the multipliers are not terribly high although they obviously still facilitate fiscal policies. Note also that in the case of uncoordinated policies, the output effect diminishes more rapidly than in the case of coordinated policies. 5

The Cohen and Follette (1999) value with US data (with four lags) was 1.23 which may be compared with our average EMU10 value of 1.25. When the tax rates were set to zero in the FRB/US model the multiplier increased to 1.35 which indicates how much (or, in fact, little) automatic stabilisers will affect on the multiplier. An interesting thing is that the multiplier value of 1.25 implies a relatively low value of the marginal propensity to consume. Assuming the average tax rate to be 0.4 we end up with a marginal propensity to consume to be about 0.3 only (or, 0.4 if we account for imports).

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

y4 (y8) denotes the output effect of an uncoordinated increase in public consumption (by 1 per cent) after four (eight) quarters, y4c and y8c denote the corresponding values in a case where all countries increase public consumption by the same amount, ymax and ycmax denote the maximum values of y over 20 quarters and ym and ymc the corresponding multiplier values for an increase in public consumption by 1 per cent of GDP. Def and defc denote the deficit effects of an increase in public consumption computed after 20 quarters.

170 Asymmetry and Aggregation in the EU

N et he rla nd Po s rtu ga l Sp ai n

G er m an y Ire la nd Ita ly

Be lg iu m Fi nl an d Fr an ce

Au

st ria

Figure 7.1 Long-run effect of a 1 per cent increase in public consumption on government surplus/GDP with and without policy coordination

0

–0,05

–0,1

–0,15

–0,2 coordination

The effect of an increase in public consumption on government deficits is almost equally clear (see Figure 7.1). Deficits increase but because output also increases the effect on the deficit/GDP ratio differs from the pure deficit effect. The values for various countries are surprisingly different, reflecting the differences in the output effects. In other respects, it is rather difficult to say why the country results are so different (the size of the country and the size of the public sector do not seem to explain the size of the output and deficit effects). In these short-run simulations it is perhaps reasonable to ignore the long-term solvency constraint but, not surprisingly, imposing the solvency condition makes a lot of difference, particularly in the long run (when the additional taxes start to have an effect). Thus, the GDP effect almost completely vanishes and the effect on deficits is also quite marginal. If countries increase public consumption and balance the budget in the long run by raising taxes, the long-run output effect is simply zero or even negative.6 Gains from coordination seem to be much larger for small 6

The importance of the solvency condition obviously depends on the level of debt in the country concerned. Given the fact that indebtedness still varies a great deal among the EU countries, we again face an aggregation problem in pursuing EU fiscal policies (see Mayes and Virén (2001) for more about this problem in terms of monetary policy).

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

uncoordination –0,25

countries while the impact of the solvency requirement depends mainly on the size and nature of the fiscal policy effect. So far, we have considered public consumption only but the picture for direct taxes is very similar. Coordination makes a lot of difference in terms of output effects but the results are less clear for the deficit/ GDP ratio. The problem stems from the output effects. When taxes are increased, output and income decrease, which eliminates part of tax revenues and – ceteris paribus – increases the deficit/GDP ratio because of lower output. If taxes are increased (by 1 per cent) in all EMU member countries at the same time, Finland’s GDP would fall by almost 0.5 per cent and that would also lead to a smaller surplus/GDP ratio. The long-run effect of direct taxes (on output) is noticeably larger than the effect of public consumption. This mainly reflects the larger GDP share of taxes compared with public consumption.7 The dynamics of the effects are, however, quite different, as can be seen from Figure 7.2, which illustrates the effects for the whole EMU area. The effect of public consumption diminishes over time while the tax effect shows no signs of a diminished impact. Figure 7.2

Comparison of expansive fiscal policy effects in the euro area

0,4 0,35 0,3 0,25 0,2 0,15 0,1

public consumption taxes

0,05 0 2000Q1

2001Q1

2002Q1

2003Q1

2004Q1

7 In Finland, for instance, the share of public consumption in GDP was 20.7 per cent in the first quarter of 2000 while the share of direct taxes was 27.1 per cent.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Asymmetry and the Role of the Public Sector 171

When dealing with fiscal policy simulation, an obvious question is what happens to interest rates. The answer provided by the NiGEM model is ‘not very much’. Thus, imposing the inflation targeting assumption for monetary policy produces only a five basis point increase in long rates in the case of coordinated policies. In the case of uncoordinated policies, the result is practically zero (for instance, in the case of Finland, just onetenth of a basis point). The NiGEM model, like most other models, generates the somewhat odd result that interest rates have a strong impact on deficits while deficits have only a very marginal effect on interest rates.8 This latter result is obviously in sharp contrast with all theorising on credibility and peso effects (but not necessarily with empirical evidence; see e.g. Alesina et al. (1992)). The model result only reflects the direct crowding out effect and does not account for direct expectations and portfolio effects. That is clearly a weakness of the model (and of all similar models). The weakness may also be quite crucial with regard to the assessment of policy coordination effects within EU. The implication of these results is interesting. On the one hand it shows that it is the small countries that have most to gain from policy coordination. However, one can reverse the argument and point out that the others have the least to lose if it is small countries that do not coordinate well. Historically coordination among the EU countries has been fairly weak (Virén, 2000b) except among the countries tracking the deutschemark. There will therefore have to be quite a considerable change in behaviour if this is to occur in future. The SGP has only a limited effect on this as limiting the size of deficits is only part of the problem. Indeed it is only when fiscal policy is not coordinated that this is likely to be a problem as such anomalies occur mainly when small countries experience asymmetric shocks. However, in the early steps of fiscal coordination through European Council of Finance Ministers (ECOFIN) under the SGP the member states, particularly those involved in the euro group, have sought to go a little further and recommend general stances for fiscal policy compared with the cycle (relating to the timing of tax cuts, for example). The BEPG are readily criticised for having no compulsion but in many respects this misses the point. It is simply that on the one hand the member states are becoming steadily more concerned with each other’s 8 As with all such models they are regularly updated, often changing their characteristics markedly. Using earlier or later vintages of the model would no doubt change all the magnitudes but our concern here is with the generalised outcome. The benefits of ‘coordination’ mainly accrue to the smaller countries.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

172 Asymmetry and Aggregation in the EU

Asymmetry and the Role of the Public Sector 173

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

policy while on the other they are becoming more closely linked. Thus, even if overt reasoned policy coordination is in short supply, there is likely to be increasing coordination simply by result. Even though many of the processes for coordination in areas such as employment are through the even looser Open Method of Coordination (Hodson and Maher, 2001) nevertheless there has been considerable policy borrowing and a convergence of some areas, particularly in active labour market polices for example (Bienkowski et al., 2008; Sapir, 2006).

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

8

Monetary policy is probably the most obvious area of policy where behaviour is likely to be asymmetric or at any rate nonlinear. We can see this from four directions. The first is that since the Phillips curve is nonlinear as set out in Chapter 4, monetary policy acting through the IS curve, Chapter 3, will have a different effect depending on whether the output gap is positive or negative. If the output gap is positive then inflation is quite sensitive to the operation of policy. If the output gap is negative then policy is much less effective. Hence, we can expect that movements in interest rates will be more substantial in the face of the threat of deflation than in the case of excess inflation. Clearly the same applies to any other asymmetries in the operation of the economy, presuming that is that monetary policy is attempting to help hold the economy on a smooth track of low inflation and make a contribution to stable sustainable economic growth. Monetary policy needs to be asymmetric to try to offset some of the asymmetries in the economy. There is a second reason for the wish to act more vigorously in the down phase of the cycle, namely that except in special circumstances, nominal interest rates cannot be negative – the zero bound problem. While it is real interest rates that have the effect on the economy, in severe downturns, not only is output likely to fall but so are prices. Thus a zero nominal interest rate becomes a positive real rate, just at a time when the monetary authority might want to see negative rates. In the present crisis both the US and the UK have attempted to get round the zero bound, as did the Japanese in the 1990s by expanding the money supply so as to create the expectation of future inflation. It is still being debated whether this is effective but, whatever the conclusion, the effect is clearly far from symmetric. 174

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Monetary Policy

Thirdly, monetary policy targets themselves tend to have a nonlinear element to them. In some cases they are also asymmetric by construction. However, the asymmetry that is often observed tends to be by result rather than by intention (Mayes, 2007). In most countries, whether inflation targeting or not, the principal aim is to keep inflation below some sort of target level. The target may be unstated as in the US or explicit as in the euro area (less than but close to 2 per cent over the medium term). Small fluctuations below this are acceptable but higher inflation and the threat of deflation are fought with vigour. Where there is an explicit inflation target as in the case of the UK, Canada, Australia, New Zealand, Norway, Japan, Czech Republic, Poland and Sweden among many others, then there is either a target band or central value with a symmetric tolerance band round it. Within the target band outcomes are acceptable but outside they are not. While the transition may be smooth and not show a sharp change in policy once the band is expected to be breached there is a clear asymmetry in policy between being inside and outside the target inflation band. In what follows we explore whether monetary policy can be described by two regimes – normal times and when breaching the edge of the band is threatened – or three – normal times, facing excess inflation and facing deflation. We find evidence for three. Lastly there is the problem of asset prices, which have come to the fore strikingly again in the present crisis. Asset prices perform in a highly volatile manner but their effect on inflation is not symmetric between rapid rises and falls. The effect differs between share prices and house prices. It is not clear whether this concern should be described as monetary stability or financial stability as it clearly has an element of both. The area is still hotly debated. The traditional view is that asset prices are an important information variable for monetary and that they play an important role in the shape of the economic cycle and the inflationary process. The contrary view is that asset prices should themselves be a target of policy. Some see this as a part of monetary policy where asset prices should be included in the target rather than having simply some measure of consumer prices (Goodhart, 2001). Others regard asset prices as a target in their own right (Cecchetti et al., 2000). In the latter case, if the objective is to maintain financial stability rather than monetary stability per se then it will be necessary to have an additional instrument of policy that can operate on one or other side of banks’ balance sheets. We only deal with the traditional monetary policy view, although in our analysis we cast some light on the role of asset prices in the economic cycle and the implications this has for monetary policy.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Monetary Policy 175

Perhaps the most important side of this asymmetry in policy is what has been labelled the ‘Greenspan standard’ by Blinder and Reis (2005). The argue that Alan Greespan’s policy when Chairman of the Federal Reserve was to lean against what appeared to be asset price bubbles and advise people of the risks they were taking, but not to burst them, as it was both very difficult to decide what prices are justifiable and important to avoid being responsible for an unnecessary downturn. It was better in this view to stand ready to act very swiftly and forcefully when the bubble burst to stop the economy entering a downward debt-deflation spiral. Thus, while policy would seek to avoid bubbles and particularly recessions, it was more geared to handling the problem of volatility than eliminating it. It is clear that the same view was held by Greenspan’s successor Ben Bernanke. However, the recent crisis is likely to have changed minds and the balance of effort will move strongly towards avoiding future such drastic cycles. Nevertheless the bulk of this effort will come through work on financial stability rather than from monetary policy. It remains to be seen whether US monetary policy will be more cautious when asset prices next start to rise rapidly. There is no evidence in the rhetoric from the Eurosystem that they have a similar asymmetric approach to asset prices. However, our research and work by Taylor (2009) for example, suggest that policy settings during recent years could be consistent with an implicit view of that form. The rest of this chapter is structured in two parts to cover this whole range of issues. The first deals with the issue of whether monetary policy appears to have been asymmetric while the second considers the role of asset prices in creating asymmetry in the economy and in monetary policy.

8.1

Asymmetry in monetary policy

As explained in earlier chapters, we have set up a simple and very conventional model of the economy, consisting of an IS curve, a Phillips curve, and Okun curve, where these three relationships have the general form ∇yt = a0 + a1∇yt–1 + a2∇yt–2 + a3rrt–i + a4re t–j + a5∇y* t–k

(8.1)

where ∇y is the deviation of output y from its Hodrick–Prescott filtered trend, rr is the real three-month interest rate (i.e. the nominal rate of interest r less the annual rate of consumer price inflation ∇p), re the real exchange rate with the US dollar (in logs) and ∇y* the deviation of

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

176 Asymmetry and Aggregation in the EU

Monetary Policy 177

OECD output from its HP trend. (Lag lengths i, j and k typically vary from one to three quarters in estimation) (see Chapter 3). ∆p = b0 + b1∆pt–1 + b2∆pe + b3∆p* + b4u

(8.2)

pe is expected inflation, p* is the foreign price (in domestic currency) and u is the deviation of unemployment from its trend1 (see Chapter 4) and (8.3)

where pop is the population of working age (see Chapter 6). One of the difficulties about measuring the three foregoing relationships is that in practice the observations that we have are ‘policy inclusive’. Over the period covered by our data, governments have sought to stabilise the economic cycle with some combination of monetary and fiscal policy, partly through ‘automatic stabilisers’ and partly through discretionary action on each occasion. The immediate consequence is that our estimates of these equations may be biased, although we have used estimation methods that should take omitted endogeneity into account. Laxton et al. (1993) argue, for example, that this omission tend to reduce our ability to observe the curvature of the Phillips curve. Not only does the policy reaction reduce the variance but it makes the impact of the underlying relationship appear smaller. However, the impact of policy could be even more distorting if policy is itself not symmetric or linear, as it may not get picked up by the instruments used in the estimation. Economists typically express loss functions in quadratic terms implying that policy will respond more than proportionately as expected outcomes deviate from their targets. However, they tend to make them symmetric (Taylor, 1993). It is perhaps a little more realistic to consider the ‘opportunistic’ approach to policy (Orphanides and Wilcox, 1996) where ‘favourable’ outcomes such as more rapid recoveries, balance of payments improvements etc. than expected are accepted and not offset, whereas less favourable outcomes stimulate further policy responses.2 A more general asymmetric loss function is used in Koskela and Virén (1990) and Virén (1993) drawing on the work of

1

We used a more complex lag structure in estimation. Monetary authorities may seek to offset the asymmetries in the inflationary process, while governments may be more concerned to combat high unemployment or take advantage of periods of higher growth (the ‘inflation bias’ discussed clearly in Walsh (1995) inter alia). 2

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

∆U = c0 + c1 ∆y + c2 ∆pop

Waud (1970) and Hosomatsu (1970). This also applies to the threshold model approach that we use in this book. However, here we experiment by introducing a policy reaction function directly into the model. It is difficult to decide on a form for the monetary policy reaction function as the EU countries were following different regimes during the period since 1970 to which our data refer. The Bundesbank used a form of enhanced money targeting (Issing et al., 2001), many of the other central banks were targeting the exchange rate first of all within the snake and then the ERM, while others including the UK, Sweden, Finland and Spain have had periods of inflation targeting in the years since the early 1990s. However, as Collins and Siklos (2002) demonstrate, a simple Taylor rule where interest rate smoothing is included provides a reasonable representation of the behaviour of most modern regimes including the US, despite the fact that their ostensible objectives are different. It even embraces the ‘speed limit’ interpretation of US policy (Walsh, 2001; Woodford, 2001), although for some small open economies it might make sense to include the exchange rate. What is particularly interesting is that even though monetary policy is firmly forward-looking in the eyes of central banks including forecasts of inflation and the output gap it does not alter the performance markedly. A simple form for such a Taylor rule would be rt = ρrt–1 + (1 – ρ)[d0 + d1(∆p – ∆pT )t + d2∇yt ]

(8.4)

where the parameter ρ permits an element of interest rate smoothing and ∆pT is the target for inflation (Huang et al., 2001). (∆p – ∆pT ) is expressed as π, inflation, in much of what follows. Where the target remains constant this acts simply as a change in numeraire but the impact is more complicated. Table 8.1 offers a comprehensive set of results for Taylor rules of the form of (8.4), both in Europe and the US for the period between 1971 and the end of 2008, using our dataset drawn from OECD data, relating to the ‘EU 15’ – i.e. the EU between the 1995 enlargement and the 2004 enlargement (less Luxembourg and plus Norway) and in selected cases to the US. It is immediately clear that whether the GDP deflator or the CPI are used a Taylor rule is a reasonable representation of behaviour, irrespective of the estimation method and detailed specification. The only occasion where there does not appear to be the expected relationship is if we look at the euro area countries, during the period that the euro area was operating. If we look at the non-euro area EU countries during the same period then there is still a significant

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

178 Asymmetry and Aggregation in the EU

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

6.147 (50.55) 5.621 (45.25) 5.698 (54.75) 0.482 (6.21) 0.390 (6.99) 0.320 (4.48) 0.357 (5.35) 0.516 (6.60) 1.061 (5.12) 0.461 (6.69) 0.513 (4.67) 0.258 (1.70) 0.210 (1.12) 1.227 (2.36)

1971–2008 DEF, GAP 1971–2008 CPI, GAP 1971–2008 CPI, GAP 1971–2008 CPI, GAP 1971–2008 CPI, GAP 1991–2008 CPI-H, GAP 1987–2008 CPI, GAP 1971–2008 CPI, GAP 1999–2008 CPI, GAP EMU 1999–2008 CPI, GAP NEMU 1999–2008 CPI, GAP US 1971–2008 CPI, GAP US 1971–2008 CPI, GAP US 1999–2008 CPI, GAP –0.109 (0.63)

0.045 (1.58) 0.015 (0.66) 0.127 (3.11) 0.123 (2.42)

0.477 (19.57) 0.539 (20.30) 0.562 (27.56) 0.067 (5.76) 0.067 (6.46) 0.057 (5.61) 0.100 (4.61)

Inflation

0.182 (1.8)

–0.008 (0.11)

Inflation π<π0

0.129 (2.4)

0.057 (6.01)

Inflation π≥π0 0.163 (2.74) 0.163 (2.75) 0.271 (5.79) 0.185 (9.69) 0.184 (9.78) 0.082 (3.99) 0.115 (5.46) 0.186 (9.71) –0.057 (1.67) 0.161 (8.13) 0.236 (7.34) 0.291 (4.96) 0.291 (4.91) 0.638 (3.73)

Output

0.901 (62.50) 0.910 (85.56) 0.918 (72.76) 0.903 (58.82) 0.907 (83.37) 0.810 (24.08) 0.845 (37.19) 0.786 (19.72) 0.878 (25.15) 0.879 (25.03) 0.703 (8.32)

0.497

Lagged r

0.458 2.901 0.497 2.878 0.21 2.871 0.917 1.166 0.917 1.169 0.960 0.726 0.918 0.868 0.917 1.166 0.736 0.576 0.906 0.295 0.948 0.283 0.922 0.997 0.923 0.999 0.946 0.451

R2/SEE

OLS, FE

0.20

GLS, FE GLS, FE

1.87 1.86

OLS, FE OLS

1.45 1.84

0.97

OLS

OLS

OLS, FE

1.05

1.84

GLS, FE

1.57

GLS, FE

GLS, FE

1.81

1.81

OLS

1.79

GLS, FE

OLS, FE

Method 0.22

DW

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Robust standard errors are inside parentheses. The inflation threshold (that minimises the sum of squares) is 2.4 per cent p.a. CPI-H denotes the harmonised CPI. EMU denotes the set of EMU countries and NEMU the set of non-EMU EU countries.

Constant

Estimates of the Taylor rule

Data

Table 8.1

179

relationship with both the output gap and the inflation deviation from the target. However, the problem is clearly not simply to do with the euro area as the US has exactly the same characteristics. There too, there seems to be little impact from inflation on monetary policy during that period from 1999 to 2008. One obvious response to this would be to suggest that this reflected the ‘great moderation’. Although there were threats to price stability, particularly from the dotcom boom and then later on with the rise in commodity prices which preceded the present downturn, actual inflation remained largely under control. At the same time, however, monetary policy was very active, moving from high levels to historically low levels in the early 2000s and then back up again until the present crisis when interests rates fell even more rapidly and further. The non-EMU countries, with the exception of Denmark were inflation targeting during that period, and as small open economies may have been more susceptible to the normal pressures. As pointed out very clearly by Taylor (2009) the US and to a lesser extent the euro area departed very substantially from a Taylor rule in the years 2002–2004, running a much looser policy. This he claims is one of the main reasons for the emergence of the sub-prime crisis as credit had never been so easy. Thus this period might best be viewed as an aberration in those two areas. The lack of relationship is clearly not because inflation is no longer the focus of policy but perhaps simply because policy has been so successful. If there had been more variance in prices during the period then something other than a general response to the economic cycle would have been needed. In effect we face the same sort of identification problem which was acknowledged already a long time ago by Blinder and Solow (1973). This suggests that the nature and performance of monetary policy cannot easily be measured with a single equation model. Rather a complete general equilibrium model is required. A second possible explanation is that because policy tends to be forwardlooking with respect to inflation, that this simple formulation of the Taylor rule using current inflation misses the link because a lead should have been used. Huang et al. (2001) show that a short lead does improve the fit of a Taylor rule in New Zealand but the differences are not large. Inflation has become less persistent (Table 8.2) but GDP growth has retained roughly the same autocorrelation structure. If we extend the EMU period backwards progressively from 1999 to 1987 the coefficients change but only slowly. There is no obvious break point. EMU may therefore have reinforced a trend rather than constituting a major change in its own right.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

180 Asymmetry and Aggregation in the EU

Monetary Policy 181 Table 8.2

Autocorrelation of inflation

Sample period 1970Q1–2006Q4 1987Q1–2006Q4 1999Q1–2006Q4

Data points

Quarter to quarter

Over 4 quarters

2070 1146 474

0.606 (20.78) 0.139 (3.32) –0.256 (2.03)

0.961 (114.45) 0.888 (41.88) 0.499 (5.62)

However, all this analysis is undertaken on the basis that there is no asymmetry in monetary policy. When this is admitted we do find some differences in the reactions to inflation depending upon the level of inflation. This is shown both for the EU and for the US using a single switching model (Table 8.1). The implication is that the central banks have reacted more affirmatively to high inflation than to low inflation. This shows up both in using a simple regime switching Threshold model and a Smooth Transition Regression model. We have also experimented with a model that includes two thresholds; in a sense a ‘corridor’ where reactions are different for low, mid-range and high inflation. In a panel setting, this kind analysis is somewhat tedious because one might expect that, to some extent at least, the threshold parameter values are countryspecific. The set of results are reported in Table 8.3 do suggest that some form of corridor might be relevant in modelling the policy response for the EU. For some deflationary values, the response coefficient looks quite large but it cannot be estimated very accurately (the t ratio does not allow us to reject the hypothesis that the coefficient is zero). The coefficient is much smaller in the neighbourhood of 2 per cent while for values larger than 4.5 per cent, the reaction is again larger even though Table 8.3

Threshold model estimates for the Taylor rule

Linear corridor model – corridor runs between –0.6 per cent and 4.5 per cent rt = 0.814rt–1 + 0.084GAPt + 0.242(pt|pt <–0.6) + 0.059(pt|–0.6≤pt ≤4.5) + 0.096(pt|pt >4.5) (35.82) (3.13) (1.10) (2.21) (5.91) R2 = 0.942, SEE = 1.0199, DW = 1.85. Continuous model with the same corridor rt = 0.814rt–1 + 0.084GAPt + 0.058pt + 0.37pt*(1/1+exp(–θ(pt –pL)(pt –pH))) (35.76) (3.13) (2.22) (1.86) R2 = 0.942, SEE = 1.0199, DW = 1.85. pL = –0.006 and pH = 0.045, θ = 16.4. Where L and H denote the lower and higher sides of the corridor.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

The data are derived from all EU countries; corrected (White) t-ratios are in parentheses.

the long-run coefficient still falls short of unity. A similar result emerges with the Smooth Transition Regression model. The fact that the inflation coefficients are so low obviously reflects the fact that the 1970s are included in the data set. Starting the sample from, say, 1987 gives more reasonable values for the coefficient although it is hard to say that the inflation coefficient would have been clearly above one. This presents an important problem as it implies that policy did not follow the Taylor principle, which requires the coefficient to exceed unity if policy is to be effective. The corridor chosen is rather broader than that which any of the central banks have claimed to be using. The corridor size has not been determined by looking at the expressed policy intentions but simply by looking at the data. This is the pair of values that maximises the likelihood. There are no sharp peaks in the relationship so moving the corridor boundaries somewhat has only a limited effect on either the coefficients or the fit of the relationship. Reducing the width of the corridor to say 0 to 2 per cent does however have a substantial effect on the value of the coefficient on the range above the corridor, removing much of the difference in policy. Changing it to 1–3 per cent has a similar effect on the lower boundary. To get clear differences we need a wide corridor so that only really extreme values are included outside the bounds. To some extent this is a function of having only a limited data period and hence perhaps asking too much of the information available. In view of the high level of endogeneity we have also estimated the Taylor rule explicitly as part of the whole system of four equations (8.1)– (8.4) (see Table 8.4).3 There, Taylor rule coefficients are shown in the last three rows. The coefficients from the other equations are shown in stacked form in the earlier rows of the Table. Columns 1, 3 and 5 use the deviation of unemployment from its Hodrick–Prescott filtered trend in the Phillips curve, while the other three columns show the results for the output gap. The results are similar to those obtained when estimating the equations separately. The asymmetry appears to be concentrated in the Phillips curve (asymmetry due to unemployment dispersion in the Okun curve could not be included with these quarterly data) when using this full sample. However, if we confine the estimation to the period after the ERM crisis (cols. 5 and 6) then there is asymmetry in the Okun curve as well. Including the years when the euro area was in operation does not appear to have a major effect. We have used rolling regressions to test for

3

Note that this section refers to earlier results.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

182 Asymmetry and Aggregation in the EU

Monetary Policy 183 Simultaneous system estimation from panel data SUR y0 ∇y–1 ∇yworld,–1 rs re u0 u–1 y– y+ ∆pop–1 p0 ∆p–1 ∆p–2 ∆m ∆m–1 x– x+ RS–1 ∆p ∇y period def. of x λ Wald b1 = b2 Wald c5 = c6

SUR

SUR

SUR

SUR

SUR

0.031 0.031 0.033 0.030 0.072 0.071 (4.16) (4.12) (5.16) (4.93) (15.15) (15.18) 0.759 0.757 0.767 0.764 0.714 0.714 (52.30) (52.40) (58.87) (57.49) (113.09) (111.56) 0.181 0.188 0.195 0.206 0.184 0.183 (6.91) (7.26) (8.17) (8.74) (13.33) (13.42) –0.191 –0.195 –0.156 –0.156 –0.299 –0.299 (5.58) (5.72) (5.78) (5.85) (39.13) (38.08) –0.034 –0.032 –0.035 –0.033 –0.077 –0.078 (4.00) (3.95) (5.02) (4.78) (14.87) (14.91) –0.013 –0.014 –0.014 –0.014 –0.008 –0.010 (1.66) (1.70) (1.70) (1.64) (1.30) (1.51) 0.743 0.740 0.743 0.742 0.677 0.677 (81.46) (80.53) (90.76) (89.06) (94.32) (91.00) –0.169 –0.171 –0.173 –0.174 –0.205 –0.206 (19.11) (18.87) (21.32) (20.82) (31.90) (30.19) –0.147 –0.148 –0.151 –0.151 –0.131 –0.130 (16.00) (15.98) (18.33) (18.13) (13.33) (12.58) 0.038 0.031 0.022 0.017 0.084 0.085 (2.17) (1.79) (1.40) (1.15) (3.63) (3.61) 0.003 0.002 0.003 0.003 0.003 0.003 (11.52) (9.20) (12.69) (9.93) (26.88) (25.33) 0.254 0.261 0.263 0.271 0.097 0.085 (11.79) (11.51) (13.17) (12.69) (9.84) (8.24) 0.313 0.329 0.318 0.337 0.268 0.259 (15.12) (15.14) (16.45) (16.32) (27.56) (25.51) 0.019 0.015 0.033 0.028 0.028 0.027 (6.66) (5.13) (11.02) (8.90) (19.59) (18.03) 0.018 0.018 0.027 0.028 0.023 0.018 (5.72) (6.09) (8.84) (9.13) (15.52) (11.45) –0.115 –0.002 –0.111 0.019 –0.072 –0.017 (4.66) (0.06) (5.01) (1.28) (5.36) (3.12) 0.052 0.105 –0.082 0.103 –0.012 0.041 (2.25) (7.27) (4.00) (7.78) (1.66) (5.09) 0.853 0.853 0.834 0.833 0.772 0.771 (89.20) (88.56) (84.84) (82.65) (252.43) (238.45) 0.212 0.212 0.222 0.219 0.390 0.391 (8.27) (8.12) (8.68) (8.37) (66.80) (65.89) 0.214 0.212 0.230 0.227 0.159 0.159 (15.77) (15.45) (17.61) (17.01) (41.77) (42.13) 1985–2001 1985–2001 1985–1998 1985–1998 1993–2001 1993–2001 ∇u ∇y ∇u ∇y ∇u ∇y 5.70 6.02 4.42 4.70 3.86 3.85 2.32 (0.128) 2.31 (0.124) 2.66 (0.103) 2.88 (0.089) 30.28 (0.000) 27.90 (0.000) 2.74 (0.098) 17.92 (0.000) 0.74 (0.389) 13.13 (0.003) 11.54 (0.001) 25.34 (0.000)

y0, u0 and p0 denote the constant terms of IS, Okun and Phillips curves, respectively. In the Taylor rule (last three rows of estimates), the intercept r0 was allowed to vary from country to country. Number of observations 720.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 8.4

other sources of instability. It is immediately apparent that the results are dependent on the data period chosen. The problem lies with the breakdown of the ERM in 1992. If estimation is restricted to the post-1992 period (cols. 5 and 6) then we observe the expected result. The weight on inflation is about twice that on the output gap and there is a large element of smoothing in policy. If 1992 is included in the data period then the weights are equal. However, these results involve a symmetric reaction function. It is clear from Table 8.5 that the reaction function is itself asymmetric. The authorities appear to have responded more vigorously when inflation has been above 2 per cent a year than when it is below it.4 This asymmetry also seems to apply to the output gap. The interest rate response has been clearly stronger when output has been above trend than when it was below it. We wondered whether this asymmetry was in fact somewhat misleading as the Eurosystem’s target for price stability is for inflation not exceeding 2 per cent over the medium term. Thus, if this were followed we would expect to see disproportionate reactions to inflation above 2 per Table 8.5

Reaction function estimates

Rt–1

0.771 (238.15)

∆pt

0.391 (65.89)

0.863 (63.54)

∆pt|∆pt<.005

0.281* (2.30)

∆pt|∆pt>.005

0.164 (3.30)

∇yt

0.159 (42.13)

∇yt|∇yt<0

0.112* (2.65)

∇yt|∇yt>0

0.381 (5.72)

The Wald test result for the equality of the two respective coefficients is 927 (0.009), Thus, the linear model is rejected at the 1 per cent significance level with the chi-square distribution. All estimates are derived form the whole system of equations. Data period is 1993–2001. 4

Shown at the quarterly rate of 0.005 in the table.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

184 Asymmetry and Aggregation in the EU

Monetary Policy 185 Table 8.6

Corridor reaction functions

∆pt| ∆pt<0

0.602 (3.63)

∆pt| 0<∆pt <.01

0.153 (3.32)

∆pt| ∆pt >.01

0.230 (6.60)

∇yt| ∇yt <–.02

0.249 (9.32)

∇yt| –.02<∇yt<.02

0.074 (3.80)

∇yt| ∇yt >.02

0.147 (0.41)

cent and to deflation. Rather than impose our own view of where the different regimes should lie we searched for the maximum likelihood estimates for rounded intervals. Here it appears that deflation is tackled even more vigorously than inflation above the target range (Table 8.6). The lowest weight is for inflation in the range 0 to 4 per cent a year. This somewhat wider range for milder action than that implied by the Eurosystem target is probably accounted for by the fact that most of the data period is before the ECB was set up. A similar set of results is obtained for the output gap, with larger coefficients outside a corridor of 2 per cent either side of zero. However, it was not possible to obtain significant coefficients for the output gap above the corridor. Indeed trying to include the output gap poses considerable convergence problems for the model. Taking the Phillips curve, Okun curve, IS curve and monetary reaction function results together gives us a somewhat better insight into the nature and causes of both asymmetry and nonlinearity in macroeconomic behaviour. Although, of course, some of the picture is clearly still omitted. It is clear that the variations across regions in labour markets and across sectors in product markets lead to important deviations in aggregate behaviour. When combined with the different national and sectoral responses to monetary policy, whether through the exchange rate or interest rates, this permits substantial departures from linearity. The asymmetries in the Phillips curve that we have explored appear to be primarily cyclical in character. The asymmetries in the Okun curve, on the other hand are more complex, reflecting not just cyclical factors

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Data period is 1993–2001, maximum likelihood, reaction functions only.

but the degree of sectoral and regional mismatch in the operation of the labour market. There is thus not just a nonlinear underlying relationship but asymmetric departures from it. As the average level of unemployment falls so the scope for regional and sectoral disparities also falls as there is a lower bound. It seems likely therefore that there is more than one source of asymmetry. The structural mismatch in the labour market appears to be an additional cause to the traditional Phillips curve result. The asymmetries are likely to interact. The asymmetric nominal rigidities implicit in the Phillips curve are likely to contribute to the asymmetric labour demand effects revealed in the Okun curve. Downward rigidities in prices and wages would tend to increase the variance of unemployment. The different sectoral responses to monetary policy will be a reflection of this. Asymmetric shocks will interact with the nonlinear responses and asymmetric processes themselves. When combined with the policy reaction this generates a considerable identification problem (as explained by Blinder and Solow (1973) in the case of fiscal policy and Haldane and Quah (1999) for monetary policy.) Our model is only illustrative and we can increase the effects by using larger shocks, altering their timing to affect when the regime switches or adding the asymmetric version of the reaction function. If instead of using the panel data model we were to allow different parameter values for each individual country, then we would observe a much bigger variety of timing and size of regime shifts even under a single monetary policy reaction function. Our analysis does not offer much scope for a discussion of the causes of asymmetry. In their tests of causes of asymmetry in the Phillips curve Dupasquier and Ricketts (1998) are able to isolate some evidence for the hypotheses of costly adjustment, capacity constraints and misperception (of aggregate and relative price shocks). The nominal wage resistance hypothesis was not obviously sustained, a result consistent with Yates (1998). Although to some extent these causes should be separable the results from their joint inclusion were not well determined. Eliasson’s (1999) finding that the Phillips curve, using unemployment not an output gap as the determining variable, shows different sources of nonlinearity in Sweden and Australia is also helpful. In the Swedish case it is the rate of change of inflation expectations that is important, while for Australia it is the rate of change of unemployment.5 The former

5

Buxton and Mayes (1986) also made this finding for the importance of the rate of change of unemployment in the case of the UK.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

186 Asymmetry and Aggregation in the EU

case will have particularly important implications for the conduct of monetary policy. Moreover the fact that the sources of nonlinearity differ for these two countries and are not found in the case of the US in contrast to Laxton et al. (1999) emphasises the potential problem of aggregation that we have outlined for the euro area. In the next section we go to expand the scope of the analysis to include both stock prices and house prices within the ambit of monetary policy. We can do this in a straightforward manner by adding both of these variables to the Taylor rule, not of course implying that John Taylor would advocate this. This simply creates a more extensive reaction function where not simply inflation and output are deemed important for stabilisation but also asset prices. However, we have to take into account the role that asset prices play in the economy as a whole, as clearly, while asset prices can be taken into account in formulating monetary policy this would be largely without value if these asset prices did not themselves have some measurable impact on the stability of the nominal and real economies. We have already explored how asset prices help in the explanation of aggregate demand and consumers’ expenditure, in particular in Chapter 5. Our interest is in the asymmetry. Asset prices, as is very well illustrated in the present downturn, have a very asymmetric relationship with economic behaviour. For example, in a downturn people have to sell assets to survive, as a result driving asset prices very low. This has a particularly adverse effect in the banking sector, where both collateral and capital values are reduced. Banks whose capital values have fallen need to reduce their lending in order to return to adequate capitalisation. Reducing lending sharply has a highly adverse effect on economic activity causing further losses to banks when firms cannot continue if their lending is not rolled over. This discussion is thus of particular significance at present.

8.2 The asymmetric role of asset prices in the European economy The asymmetric behaviour of asset prices over the cycle interacts with monetary policy. Prices evolve in the light of actual and expected policy but policy itself responds to process. An asymmetric approach by monetary policy to stock prices over the cycle has been set out in Blinder and Reis (2005) for the Greenspan years in the US and confirmed by Greenspan himself (Greenspan, 2007). The line of argument runs as follows. As it is difficult to decide whether stock market bubbles exist and to judge

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Monetary Policy 187

188 Asymmetry and Aggregation in the EU

their extent, the central bank is better employed in warning people and pointing out the difficulties they may face than in trying to decide when and how to prick such a putative bubble.

The famous remarks on ‘irrational exuberance’ (Greenspan, 1996) illustrate this in practice – as Greenspan (2007) admits with the benefit of hindsight, he may have been a little premature. In pricking a ‘bubble’ the central bank may on the one hand slow real growth in the economy unnecessarily or on the other provoke a precipitate decline. The Greenspan approach instead would see monetary policy continuing to tighten slowly as inflation risks increased but moving much more rapidly when the downturn sets in to avoid the rapid fall in stock prices and associated financial concern turning into an outright recession with the danger of debt deflation. It is already clear from the monetary policy decisions of the Bernanke period that asymmetry in policy remains with much more rapid cuts, followed by a raft quantitative and credit easing measures as the zero bound was reached in the face of financial difficulties and an economic downturn than the steady and predictable rises as the economy grew and inflation started to rise. This approach is now subject to intense scrutiny in the light of the severity of the present crisis and measures to dampen the openness of the economy to fluctuations can be expected. Nevertheless, as Milne (2009) suggests, this will come largely through changes in structure and the regulation of financial institutions rather than through macroeconomic policy per se. The asymmetry in house prices is somewhat different from that associated with stock prices. Traditionally, for people who own their own homes, when prices peak, many sellers are reluctant to sell at a loss, especially if this means that they would realise negative equity. Hence the market tends to dry up and prices tend to fall rather more slowly than they increased, without the sudden and rapid declines that can characterise stock prices. However, their contribution to inflation tends to be rather more important. This can perhaps best be characterised as a liquidity channel as people cannot sell and collateral values fall. Changes in

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Experience with past asset price boom episodes tells us that we should be very careful in calling a boom, which is observable, a bubble … Not all boom or bubble episodes threaten financial stability. Policy-makers should not fall into the trap of attempting to eliminate all risk from the financial system. They would either be unsuccessful (due to moral hazard) or they would likely hamper the appropriate functioning of a market economy where risk-taking is of the essence. Jean-Claude Trichet (2005)

Monetary Policy 189

house prices can thus act as an indicator of the proportion of liquidity constrained households. However, housing is increasingly becoming an investment as incomes and wealth rise, hence the constraints and cyclical pressures may well be changing. Altissimo et al. (2005) suggest that for stock prices, in addition to the obvious wealth effect on consumption, there might be an effect on investment through Tobin’s Q , a balance sheet effect and a confidence effect, although the empirical evidence for the latter three is quite weak. Monetary policy

The next step in our analysis is to see how much the asymmetry in behaviour may be due to asymmetry in policy responses. The obvious policy to look at is monetary policy, partly for practical reasons, as much of fiscal policy is set on an annual basis. However, our work on fiscal policy set out in the next chapter (see also Mayes and Virén, 2007) suggests that not only does policy in the euro area countries show a clear asymmetry in the sense that governments tend to ease up on consolidation during the up phase of the cycle but that there has been a clear shift in behaviour since 1996, first with the run up to qualification for Stage 3 of EMU and then with its operation.6 There are, however, some problems, as it is difficult to describe the monetary policy of all the countries and over the whole period as being in the same regime. Many countries were shadowing the deutschemark and effectively following an exchange rate target in the period up to the formation of the euro area, while others were inflation targeting. Inside the euro area interest rates are even more tightly linked. Nevertheless, as shown in section 8.1, a Taylor rule seems to provide quite a reasonable representation of a wide range of policies. Our estimated interest rate equation here therefore is a basic Taylor rule (with interest rate smoothing), as set out in equation (8.4), which is augmented with house and stock prices. It takes the form: rit = β0 + β1∆yit + β2 pit + β3HPit + β4SPit + β5rit–1 + εit

(8.5)

where r is the (nominal) short-term rate, p the rate of inflation and HP and SP rates of change of nominal house and stock prices, respectively. ε is the error term. Estimating (8.5) allows us to see whether there has

6 The asymmetry is found in taxation rather than expenditure, which tends to be fairly symmetric over the cycle, reflecting automatic stabilisation. However, taxes tend to be cut when the economy is in the up phase, thereby only partially offsetting the extended deficits that occur in downturns. Not all countries follow this pattern and Finland, for example, has shown much more symmetry and as a result its debt ratio has fallen more consistently than in some of its partner countries.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

8.2.1

190 Asymmetry and Aggregation in the EU Impact of house and stock prices on interest rates 1 ∆y

0.086 (4.45)

2 0.115 (6.73)

3

4

5

6

0.091 (6.33)

gap

0.181 (6.36)

0.208 (4.61)

0.250 (7.89)

p

0.087 (3.56)

0.087 (4.29)

0.022 (1.28)

0.070 (3.04)

–0.016 (0.93)

0.072 (2.15)

HP

0.000 (0.06)

0.0001 (0.17)

–0.003 (0.71)

0.005 (0.10)

–0.003 (0.62)

–0.002 (0.30)

SP

–0.002 (1.63)

–0.001 (0.58)

0.003 (1.87)

0.001 (0.98)

0.004 (2.66)

0.002 (1.28)

r–1

0.938 (50.00)

0.944 (78.93)

0.881 (18.08)

0.931 (48.77)

0.819 (13.93)

0.884 (44.14)

R2

0.953

0.953

0.900

0.954

0.911

SEE

0.833

0.877

0.375

0.868

0.353

DW

1.853

1.750

1.709

1.897

1.778

Estimator Panel Sample N

1.164 ..

LS

CLS

LS

LS

LS

GMM

CFE

CFE

CFE

CFE

CFE

Dif

1979–07

1979–07

1999–07

1979–07

1999–07

1979–07

1076

1076

460

1076

460

1061

Notation as above, except where indicated. The dependent variable is the short-term interest rate, denoted by r. p denotes the rate of inflation while hp and sp denote (here) the growth rates of nominal house and stock prices, respectively. Equations 3 and 5 are estimated from the sample of the EMU period 1999Q1–2007Q3. Otherwise, the estimation period is 1979Q1–2007Q3.

been any role for ‘activist’ monetary policy in which also asset price inflation is accounted for. We set the estimation up in a matching form to the IS curve, shown in Tables 3.7 to 3.8. In this case (Table 8.7) monetary policy seems little affected by house prices. There are however important differences between looking at the period as a whole and when we confine ourselves simply to the years when the euro area has been in existence. In the period as whole a Taylor rule works quite well. Both inflation and output, whether in growth rate or output gap format, have a clear influence. Yet in the euro area period, inflation seems of little importance. Indeed with the output gap it has a perverse sign. This seems more difficult to explain. As

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 8.7

we noted earlier, in part it is simply a reflection of the success of policy. Inflation in Europe has not in general been much outside the target range. However, it would perhaps be more appropriate to replace both the output gap and inflation by their forecast values as monetary policy is forward-looking. To do this it would be necessary to incorporate the forecasts used by the policy-makers. While this was possible for New Zealand (Huang et al., 2001) it is not possible for Europe as a whole, although some of the central banks have been publishing forecasts in recent years – driven initially by the adoption of inflation targeting, but more recently by a general realisation that greater transparency will make policy more understandable and hence help to focus inflation expectations on the target. We cannot get round this by using leading values of the variables as these are policy inclusive. In any case an appeal to rational expectations here would be inappropriate as we are concerned with the forecasts of the decision-makers not the economy as a whole; see Mayes and Tarkka (2002) and Paloviita and Mayes (2005). Stock prices do appear to have a slight influence in an intuitive manner – in an output gap framework, stock prices and interest rates tend to work in opposite directions in their influence on inflation. High growth rates on the other hand can occur in the period immediately after a downturn and hence could have a positive link. To some extent these results reflect the form of the equation and we get some different results from alternative specifications. So we also estimate interest rate equations which represent a standard term structure equation augmented with our additional regressors. The basic structure of these equations is: ∆rLit = γ0 + γ1∆rit + γ2(rLit–1 – rit-1) + γ3 ∆y it + γ4 pit + γ5HPit + γ6SPit + υit (8.6) where rL (r) is the (nominal) long (short) rate and υit is the error term. If we set γ1 equal to zero the equation is a standard term-structure equation while if γ1 is nonzero (possibly one) it comes close to the equations used in e.g. the NiGEM model.7 Here the results are somewhat different (Table 8.8). The influence of stock prices is still weak but that of house prices is now apparent when longer-term interest rates are used. These are not the monetary policy instrument but reflect the change in monetary conditions and hence the bite of monetary policy. There are other respects in which monetary policy is asymmetric, which will affect our results. Monetary policy appears to react much more 7

http://nimodel.niesr.ac.uk/.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Monetary Policy 191

192 Asymmetry and Aggregation in the EU

Dependent variable constant

Estimates of alternative interest rate equations

∆rL

∆rL

∆r

∆r

∆rL

∆rL

–0.080 (2.39)

–0.109 (3.06)

–0.310 (3.55)

–0.352 (3.50)

–0.011 (0.30)

–0.030 (0.80)

0.225 (5.39)

0.223 (5.39)

∆r (rL – r)–1

0.015 (1.24)

0.007 (0.54)

0.168 (3.46)

0.163 (3.41)

–0.023 (1.54)

–0.030 (2.05)

∇y

0.073 (5.35)

0.058 (4.09)

0.167 (6.22)

0.164 (5.89)

0.038 (2.59)

0.022 (1.48)

p

–0.002 (0.27)

–0.011 (1.22)

0.031 (2.06)

0.025 (1.71)

–0.009 (1.14)

–0.017 (1.96)

HP

0.010 (3.24)

0.006 (1.24)

0.008 (3.13)

SP

–0.000 (0.58)

0.002 (1.53)

–0.001 (1.56)

R2

0.046

0.061

0.144

0.150

0.194

0.206

SEE

0.489

0.486

0.858

0.856

0.451

0.448

DW

1.314

1.325

1.781

1.802

1.390

1.391

LS

LS

LS

LS

LS

LS

CFE

CFE

CFE

CFE

CFE

CFE

Estimator Panel

rL is the long-term interest rate (government bond yield). Otherwise notation is the same as in Table 8.7. The sample period is 1979Q1–2007Q3. Number of observations is 962.

strongly when there are serious threats of inflation than when the threats are fairly minor (Mayes and Virén, 2005). This asymmetry does not have a clear match with the phases of the cycle or rising or falling asset prices. The thresholds for this asymmetry are more complex and will tend to occur near the peaks and the troughs of the cycle. Clearly there are several ways we could try to incorporate this. Instead of looking at asset price inflation we could look at the acceleration in these prices, as sharp rises or falls may be far more likely to provoke reactions in monetary policy. Unfortunately we do not have enough data to explore these hypotheses properly. It is of some interest to carry out some sort of contra-factual simulation with the conventional Taylor rule (which does include asset prices) for the EMU period to see how interest rates have deviated from

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 8.8

Monetary Policy 193 Figure 8.1

Interest rate forecast for the euro period

uk sw sp pt nl lx it ie gr ge fr fi

be at –4

–3

–2

–1

0

1

2

those predicted by a model that is estimated with the pre-1999 data. Figure 8.1 gives some idea of the result: if the pre-EMU regime had continued after 1998 and interest rates had been much higher in all countries except Germany (and Portugal for which we have a very short pre-1999 data set). The result can be interpreted in many ways. One may say that that the EMU has succeeded in gaining the same credibility as Germany used to have in earlier days. Alternatively, one may argue. EMU has pursued ‘too loose’ monetary policy. This interpretation comes close to Ahrend’s (2008) findings. His interpretation is that particularly the 2002–2005 period was characterised by loose monetary policy. This conclusion is re-enforced by computation of what is called a Financial Condition Index (FCI) as discussed in Chapter 3. In the FCI, the stance of monetary policy is measured not only by the real interest rate but also by the real exchange rate and change rates of real asset prices. It is an extension of the idea of a Monetary Conditions Index (see Chapter 3 and Mayes and Virén, 2000a; Ericsson et al., 1997). The idea is that more financial variables than simply real interest rates have an impact on the economy, even though the only instrument used by

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

dk

194 Asymmetry and Aggregation in the EU Figure 8.2

An FCI for the euro area

12

10

8

6

4

2

–2 88

90

92

94

96

Median RR

98

00

02

04

06

08

Median FCI

the central bank in setting policy may be a short rate in the money market as is the case for the Eurosystem. Each item in the index needs to be assigned a weight so they can be cumulated into a single value. Computing such an index (Figure 8.2) quite clearly shows that most of the EMU period can be characterised with relatively easy monetary policy.8 Appreciation of the US dollar after 2000 and the recent slowdown of stock and house prices represent some sort of exceptions to this rule. This in turn suggests that if asset prices developments had been properly accounted for monetary policy would indeed have been less accommodative. It is, however, important to recognise that while the nominal interest rates and exchange rates may be the same for all euro area countries, stock prices and house prices can be quite different at the country level, as indeed can the consumer prices used to deflate the interest 8 The FCI is computed using the following weights: rr 1.0, rt 0.3, hp 0.1 and sp 0.05. For details of constructing the FCI, see e.g. Mayes and Virén (2002b).

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

0

Monetary Policy 195 Figure 8.3

An FCI for Germany (with panel data parameters)

9

35

8

30

7

25

6

20

5

15

4

10

3

5

2

0

1

–5 –10 88

90

92

94

96

98

RR_GE

Figure 8.4

00

02

04

06

08

FCI_GE

An FCI for the UK (with panel parameters)

9

120

8

100

7

80

6

60

5

40

4

20

3

0

2

–20

1

–40

0

–60 88

90

92

94

96 RR_UK

98

00

02

04

06

08

FCI_UK

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

0

196 Asymmetry and Aggregation in the EU Figure 8.5 An FCI for the euro area with an increased weight on the real exchange rate 10

8

6

4

2

–2 88

90

92

94

96

Median RR

98

00

02

04

06

08

Median fci

and exchange rates. We illustrate this for both Germany and the UK in Figures 8.3 and 8.4. In the German case the FCI normally adds to the impact of financial variables implied by the real interest rate alone. Thus if one were to tend to the view that Eurosystem monetary policy has been largely run with Germany in mind, the policy would be considerably tighter than that shown in Figure 8.2. We also include the UK, which is not a euro area for comparison. Here of course interest rates and the exchange rate are also different from those applied in the euro area. Here there is no indication that the general run of policy was loose compared with the settings implied by an FCI. This may reflect the fact that the Bank of England was explicitly following an inflation target over the period and that the Monetary Policy Committee did take all relevant information into account. However, when we get to the country level it is less clear that common parameters should be applied. The UK with its greater openness, greater use of market rather than bank finance for corporates and more extensive home ownership might be expected to be rather more sensitive to noninterest rate factors, particularly through the consumption function. That is well-known to policy-makers.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

0

Monetary Policy 197

Interestingly enough the FCI does not seem to be heavily affected by changes in the weights. Increasing the weight of the real exchange rate from 0.3 to 0.5 (compared to 1 for the real interest rate has only a limited effect, see Figure 8.5. This is perhaps a little surprising as the exchange rate placed a heavily dampening effect on inflation over the period from 2002 onwards (Figure 8.6). The big difference at the country level is the importance of housing. If we were to permit each country to have its own coefficients rather than impose the panel estimates for the euro area as a whole, we could get very different results (Figure 8.7). Germany and to a lesser extent Portugal exhibit strikingly higher effects but in the former case this may simply reflect that house prices have remained rather static in Germany over the period.

Concluding remarks

There is a continuing debate over whether asset prices should be included in central banks’ targets of price stability, irrespective of whether they have explicit inflation targets (Cecchetti et al., 2000; Goodhart, 2001). Until now, the general view has been that they should not be explicitly in the target or if they are it should be with a low weight (Bernanke and Gertler, 1999). However, the present crisis has heightened the view that Figure 8.6 area

The real interest rate, real exchange rate and asset prices in the euro

12

50

10

40

8

30

6

20

4

10

2

0

0

–10

–2

–20 real interest rate real ex rate asset prices

–4

–30

–6

–40 88

90

92

94

96

98

00

02

04

06

08

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

8.2

198 Asymmetry and Aggregation in the EU Figure 8.7

The impact house prices on a country by country basis

Coef. of real house prices BE DK FI FR GE GR IE IT

NO PT SP SW UK 00

0,2

0,4

0,6

0,8

central banks should react to asset prices more explicitly when they appear to be rising implausibly far or fast. Housing is clearly a consumption item so the cost of housing services should be included, although it presents measurement problems. Mortgage interest rates are typically excluded from such costs as their inclusion would leave central banks chasing their own tail since those costs reflect the setting of the monetary policy instrument. One of the simple issues is volatility. Asset prices are highly volatile and Woodford (2003) argues that central banks should put their policy emphasis on the stickiest prices. Hence prices should be inversely weighted in the target by their volatility. Wynne (2008) shows that for the United States, such an index of consumer prices, weighted both by expenditure shares and by inverse frequency of price changes, is highly correlated with headline inflation. The picture for the euro area

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

NL

is somewhat different. Simply weighting by inverse frequency of changes is much more closely related to core measures of inflation, while the double-weighted series has somewhat weaker correlations. There is no doubt that asset prices are an information variable and hence should be taken into account. Whether they should be included in some wider measure of inflation as suggested by Reis and Watson (2007), included in indicators9 – such as EuroCOIN or the Chicago Fed National Activity Index (CFNAI)10 – or treated less formally in the decision-making discussion, they should clearly be used in the modelling of economic behaviour, even though the lack of forecastability of stock prices in particular makes them useful in scenario and risk analysis rather than in forecasts beyond the short run. Our research shows that both stock prices and especially house prices have a clear role in business cycles and in the inflationary process (see Table 8.9 for a summary).11 Whether or not central banks use asset prices extensively in setting monetary policy, asset prices have a clear correlation with both short-run interest rates and the slope of the yield curve in European countries over the last 30 years, with some slight differences in the period of the euro area’s existence. We have argued that a simple way to consider these pressures is to construct a Financial Conditions Index, which adds the weighted contribution of stock prices and house prices to those of real interest rates and the real exchange rate in affecting inflation.12 This can be seen

9

Forni et al. (2001). Stock and Watson (2000) suggest that asset prices make an important but rather unstable contribution to a composite indicator. Bryan et al. (2001) offer an alternative method. 11 The part of the table is drawn from Table 3.9. We introduce a further robustness test in this table by asking whether house prices and stock prices simply act as indicators of the state of the economic cycle and are hence proxying other more fundamental variables. We therefore insert a confidence indicator, ci, published by the EU, in the consumption function (see data Appendix in Chapter 3) column 8. It is immediately apparent that while it is indeed a helpful explanatory variable in its own right its introduction has virtually no impact on the effect of either house prices or stock prices, both of which remain significant. This is a little surprising as a simple glance at the data suggests that confidence and house prices seem to move together (Figure 8.8). 12 Wynne (2008) makes the interesting observation that there is an apparent correlation between inflation targeting and house price volatility among the OECD countries. It is of course debatable which is the cause and which the effect and other third factors may be at work as in Germany (Mayes, 2008) but it does raise the suspicion that maybe a wider view of price volatility in the economy has a role to play in ensuring monetary and financial stability. 10

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Monetary Policy 199

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

–0.027 (2.99) 0.036 (7.49)

rr/100

hp

0.023 (7.14)

–0.011 (1.29)

0.078 (0.61)

0.417 (10.38)

gap

0.018 (3.13)

–0.031 (2.86)

0.205 (8.09)

∆4cq

∆4cq

0.017 (2.77)

–0.039 (3.08)

0.169 (5.92)

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

0.888 (4.13)

0.281 (6.80)

∆y

The effect of house and stock prices: An update

re/100

∆4pc

gapw

∆4yw

gap

∆ 4y

Dep var →

Table 8.9

–0.000 (0.07)

0.116 (3.09)

0.189 (6.92)

rs

–0.000 (0.05)

0.142 (3.59)

0.085 (4.05)

rs

200

0.0100 2.12

SEE DW

2.23

0.0069

0.716

1.925

0.0095

0.808

(26.76)

1.928

0.0091

0.826

(25.45)

1.843

0.841

0.954

(36.37)

0.911

0.001 (0.95)

rs

1.781

0.862

0.952

(36.27)

0.914

–0.002 (1.62)

rs

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

cq denotes the (annual) change rate of private consumption and ci denotes the (annual) change rate of the EU confidence indicator. hp and sp denote growth rates of real house and stock prices. rr is the short-term (three month) real interest rate. In Taylor rule equations hp and sp are, however, nominal change rates. The data cover the period 1979q1–2008q3. In the third equation, the hypothesis that the two multiplicative terms are identically equal to zero can be rejected by F test (F(2,1089) = 3.96).

0.803

R2

(23.71)

(25.49)

dep.var

0.663

0.696

0.669

0.040 (2.98)

∆4cq

0.645

0.041 (3.92)

∆4cq

lagged

0.001 (0.79)

gap

0.007 (2.29)

0.009 (5.77)

∆y

The effect of house and stock prices: An update – continued

ci

sp

Dep var →

Table 8.9

201

202 Asymmetry and Aggregation in the EU Figure 8.8

Confidence and house prices

120

110

100

90

Confidence indicator Forecast with house prices 70 88

90

92

94

96

98

00

02

04

06

from Figure 8.2 which also shows the usefulness of a close and systematic scrutiny of asset price developments when assessing the stance of monetary policy. One major feature of the present analysis is that it confirms the suggestion that asset prices have an asymmetric effect on the economy and on policy. When the economy is expanding more rapidly or the output gap is positive then both house prices and real interest rates have about twice as great an effect on the economy as they do when there is a negative gap. In contrast, the effect of interest rates is much stronger when house prices or stock prices are falling than when they are rising. It has been argued that for the United States at any rate there has been a stronger asymmetric relationship between monetary policy and stock prices in the form of the ‘Greenspan standard’ (Blinder and Reis, 2005; Greenspan, 2007). As prices rise beyond levels that seem to make sense from the point of view of fundamentals, policy will only tighten cautiously as the rise may be justified and there will not be strong pressure to prick any supposed bubble. On the way down however policy will react much more swiftly to head off any dangers of a damaging debt-deflation

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

80

Monetary Policy 203

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

spiral (King, 1994). A similar approach seems to have emerged under the present Chairman of the Federal Reserve Board, Ben Bernanke, although house prices have also been playing an important role in the downturn. Our results only relate to Europe. Interest rates have been much more stable than in the US, particularly in the euro area. There, any influence from stock prices is quite weak but the influence of house prices is clear. There is some asymmetry in the responsiveness of policy depending on the strength of inflationary or deflationary pressures but it is not clear whether this relates to asset prices. Our threshold approach is well suited in this regard as it enables a direct test of whether there is some sort of tolerance limit beyond which more vigorous action will ensue. In general, we see less smoothing of asset price fluctuations by monetary policy than in the US but this is aided by stickiness of house prices in Germany, which is the largest economy in the euro area.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

9

We noted in Chapter 7 that fiscal policy was faced by a number of challenges in the EU. In the first place there are longer-run pressures from ageing and from the competition by countries such as China, with low wage rates and managed exchange rates. These are both requiring politically difficult readjustments to ensure that the fiscal position is sustainable. Second, the nature of the economic cycle tends to mean that downturns are more effective in shaking out labour than upturns of the same size are in (re)employing it. This therefore tends to add further pressures in the same direction. The longer-term pressures have been addressed through a number of routes, particularly the Lisbon Strategy to increase the sustainable rate of growth by 1 per cent a year and the Broad Economic Policy Guidelines that seek to coordinate the macroeconomic responses. Ameliorating the consequences from the economic cycle are treated in part by monetary policy but also by the Stability and Growth Pact (SGP) which seeks on the one hand to encourage the longer-term improvement in fiscal positions by trying to ensure that budgets are normally in balance or in surplus and on the other by preventing excessive deficits (deficits exceeding 3 per cent of GDP except under extreme pressures). The cyclical pressures on fiscal policy are greatest for euro area countries that are out of phase with the bulk of the euro area and hence with the monetary policy set for the area as a whole. The same of course applies to countries that have a fixed exchange rate with the euro, such as those with currency boards. However, several member states have made the problems worse by failing to ensure a sustainable longer-term fiscal stance, notably Greece. Such failures are clearly asymmetric. There are no examples of countries that have persistently made errors that lead to them running excess surpluses and finding that they are facing problems 204

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Fiscal Responses

Fiscal Responses 205

from their accumulation of assets rather than debts. (Norway of course is building up assets but this is as the result of a deliberate policy to replace oil by financial assets as it steadily gets used up not errors in fiscal forecasting.) In this chapter we explore the extent of this asymmetry and show that contrary to some expectations the problems appear to be more in unsustainable tax reductions than unsustainable expenditure increases. However, we also show that even though problems may have increased in recent years they are a lot smaller than they were before the build up to monetary union. Whether or not the SGP is responsible, fiscal responsibility is clearly much greater now than it was 20 years ago. We do not extend our analysis to sufficient non-euro area countries to use a difference in differences approach to establish whether such improvements are simply general rather than being a feature only of the euro area.

In this section we turn directly an aspect of fiscal policy that is subject to constraint under the SGP, namely whether the current rules impose excessive constraints on the running of deficits. If fluctuations round a prudent longer-term policy would exceed the 3 per cent deficit limit without themselves being destabilising then prima facie the constraint is too tight. Avoiding the deficits without altering the overall setting of the fiscal system would involve tightening in the most difficult years just when it is most harmful to economic stability to do so (and presumably some loosening in better years, which might be difficult to organise without contributing to unwanted inflation). Clearly this would defeat the point of stabilising policy. However, to permit such fluctuations in difficult years without adjusting the structure of the tax and benefit system a country might have to move quite strongly into surplus in normal years, such that it would be effectively repaying its debt as a proportion of GDP. For a country with a low debt ratio this would be a strange strategy if it inhibited growth enhancing (or revenue enhancing) investment. For the euro area as a whole of course reducing the debt to GDP ratio is precisely what is required at present. Most countries are not starting from what is thought to be a sustainable position and need to consolidate. Indeed for some countries, Finland for example, there has been no contradiction in needing to run a surplus in normal times as the government wanted to run the debt ratio down substantially, both to leave room to act in the event of another serious shock like the banking crisis at the beginning of the 1990s and to cover unplanned difficulties with the

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

9.1 How sensitive is the budget balance to cyclical fluctuations in the EU?

ageing of the population for the funding of pensions or provision of services. Indeed Finland went further in building up buffer funds so that it could absorb some of the shock to unemployment and pensions from a downturn without the need to borrow, increase taxation or reduce benefits. This strategy has proved its worth in the present crisis, where Finland has experienced a greater downturn than most euro area members but without a catastrophic impact on its debt. However, Finland is not typical. Other countries have behaved differently. The UK was prepared to run considerable deficits in the growth phase of the cycle even before the present crisis put such pressure on it.1 Indeed one of the reasons why the UK’s attempts to offset the financial shock have been so modest is that it had very little leeway. However, at some point this fortunate co-incidence between the need to consolidate and the constraints of the excessive deficit procedure may not exist. Views vary as to whether output shocks have substantial effects on the fiscal balance. If ‘automatic’ stabilisers are important then the balance will move in a strongly counter-cyclical manner (Buti et al., 1998). The effects may be particularly strong if buffer funds are used, as exist in Finland (Mayes and Suvanto, 2002) and Sweden.2 However, given discretionary behaviour by governments, the effects may be attenuated (Melitz, 1997). For example, when revenues rise governments may be tempted to be somewhat more lax in their fight against rising expenditures or may take the opportunity to cut taxes. However, the process may not be symmetric, as cutting expenditures or raising taxes in downturns tend not to be attractive electorally.3 1 We do not pursue here the debate about the appropriateness of alternative simple rules for maintaining prudence, as practised inter alia by the UK. A rule that only permits borrowing for investment by the public sector is not necessarily stable since the return on many public investments are not purely financial and may not necessarily pay for themselves. Direct required rates of return may not reflect the appropriate valuation of the social benefits from the investment. 2 Johansson et al. (2002) sets out a good case for expanded buffer funds to enable a strong fiscal reaction against adverse shocks yet remain within the terms of the SGP. 3 This is the essence of the EDP in the SGP, which is intended to deter countries from getting into the position where they have imposed a fiscal tightening in the middle of a downturn. It is expected to be a vote loser and hence a very strong incentive for governments to avoid getting into that position. Experience thus far is decidedly mixed as to whether that incentive has worked. Many of the countries that have had to act have miscalculated how well their system would stand up in a downturn or simply wrongly forecast the rate of growth and ended up with lower revenues/higher expenditures than expected.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

206 Asymmetry and Aggregation in the EU

Fiscal Responses 207

There is considerable debate over how to measure the appropriate balances and Virén (2000a) computes the results for a wide range of definitions as well as for the expenditure and revenue components separately. Here we deal with just three definitions using the common specification d/y– = b0 + b1 (d/y– )t–1 + b2 t + b3 ∆y– + b4 ∆y+ + b5 r + b6 D/ y– + u

(9.1)

where d refers to the measure of the deficit, D refers to debt, y to GDP, y– indicates the trend value of y, t a time trend, r the nominal rate of Selected country-specific estimates of equation (9.1)

Austria Belgium Denmark Finland France Germany Greece Ireland Italy Netherlands Portugal Spain Sweden UK data

∆y < 0

∆y ≥ 0

∆y < 0

∆y ≥ 0

∆y < 0

∆y ≥ 0

2.115 (1.04) 1.115 (2.34) 2.084 (2.01) 1.158 (6.01) 1.092 (2.17) –

0.140 (1.21) 0.212 (1.78) 0.381 (2.51) 0.168 (1.55) 0.368 (3.62) –

0.021 (0.09) –8.362 (0.144) 0.718 (1.44) 0.134 (0.15) 0.155 (0.43) 1.757 92.67) 3.112 (5.36) –

0.306 (2.51) 0.048 (0.54) 0.149 (0.54) 0.241 (1.54) 0.298 (2.39) 0.182 (2.67) 0.128 (0.49) –

1.166 (0.60) 0.816 (1.79) 2.006 (1.78) 0.897 (5.66) 1.329 (3.07) 1.344 (1.86) 0.168 (0.79) –7.130 (1.26) 0.861 (1.26) 0.404 (0.48) 0.510 (1.59) 1.217 (1.94) 2.852 (4.74) –0.424 (0.93)

0.279 (3.10) 0.090 (0.98) 0.494 (2.92) 0.177 (2.33) 0.246 (2.97) 0.106 (1.05) 0.145 (1.90) 0.041 (0.49) –0.051 (0.49) 0.187 (1.38) 0.210 (2.12) 0.206 (3.12) 0.059 (0.22) 0.309 (2.10)

0.864 (0.40 –0.238 (0.47) 1.726 (1.79) 0.554 (3.17) 0.628 (1.33) 1.168 (1.52) –0.338 (1.47) –7.086 (0.96) 0.258 (0.96) –0.293 (0.32) –0.143 (0.41) 1.013 (1.45) 2.314 (3.84) –0.615 (1.44)

–0.032 (0.33) –0.105 (1.01) –0.229 (1.56) –0.359 (4.31) –0.060 (0.62) –0.321 (3.02) 0.061 (0.75) –0.155 (1.33) –0.179 (1.33) –0.301 (2.05) 0.079 (0.75) –0.216 (2.88) –0.634 (2.29) –0.269 (1.96)

def 1972–99

def 1972–99

defp 1961–99

defp 1961–99

defa 1961–99

defa 1961–99

def denotes net lending, defp net lending excluding interest expenses, defa the structural deficit. All of these are related to trend GDP. SUR estimates.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 9.1

interest and u an error term and ∆ denotes a growth rate. This a straightforward example of our threshold model, where, in this case, the threshold is applied to the growth rate. Thus –/+ denotes whether the growth rate is below or above the threshold (normally zero), ∆y+ includes only the above threshold values and ∆y–only the on and below threshold values. Using data for the period 1960–99 from the EUROSTAT databank for the 15 EU countries excluding Luxembourg, the country specific estimates with respect to ∆y–/+ are shown in Table 9.1. The three deficit measures, shown in the Notes to the table, are net lending, net lending less interest payments and the cyclically adjusted deficit according to the Commission of the EU. Between them these three cover the range of concepts one might want to address. The cyclically adjusted deficit gives an idea of the overall stance of fiscal policy, although the appropriate cyclical adjustment is difficult to achieve. It can be computed after the event but the policy stance is a forward-looking concept that depends on the forecast of what the trend is likely to be over the medium term – something that can often be seriously erroneous. We use a well-established definition rather than entering the debate, especially since it is this definition that is used in the official EU discussions about the stance of policy. Similarly, while interest payments are a function of the overall stance, they too vary over the course of the cycle with the fluctuations in interest rates and outstanding debt. The main implications of the results in the table are: (1) Fiscal policy seems to respond to business cycles quite considerably. Thus, the deficit elasticities with respect to output growth appear to be around 0.2–0.3 for a one-year horizon (clearly more than obtained by Melitz (1997)). (2) There appears to be strong evidence of asymmetric cyclical behaviour in government deficits. The output effects on deficits seem to differ depending on the business cycle regime: they appear to be much strong in depressions (output falling) than in booms. The hypothesis of equal coefficients for these regimes can be rejected quite clearly.4 (3) Asymmetries mainly relate to the structural deficit. Thus, the cyclical component of the government deficit seems to behave more or less symmetrically in terms of output fluctuations. This means that when output decreases structural deficits increase but when output increases

4 The threshold estimated by the maximum likelihood procedure we describe was close to zero so the results using it are not reported.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

208 Asymmetry and Aggregation in the EU

structural deficits also tend to increase (surpluses decrease). The problem thus lies with discretionary behaviour rather than with automatic stabilisation. In good times discretionary policy appears to have been perverse. (4) The different cyclical effects show up in both revenues and expenditures. Revenues seem to be more sensitive to output growth in depressions than in booms. Thus, when output grows, the revenue/trend output ratio remains more or less constant, while in depressions it decreases quite markedly. Expenditures seem to increase in depressions and decrease in booms. This probably reflects changes in government transfers (e.g. unemployment benefits). (5) The direct effect of interest rates on deficits can be clearly discerned. The effect is particularly strong with net lending but it also shows in primary deficits. Thus, an increase in interest rates leads to some loosening of fiscal policies, and vice versa. The net lending effect obviously reflects the direct expenditure effect on interest expenses but the primary deficit effect is a bit hard to be interpreted. (6) More interestingly, the effect of government debt also turns out to be both significant and of ‘correct’ sign and magnitude. Larger debt leads to some correction in the form of lower deficits. We do however have to be rather cautious in interpreting these results, as the reverse impact of the fiscal balance on output has not been taken into account in estimation on the grounds that it occurs with a lag (while the effect of growth on the deficit is contemporaneous). Virén (2000a) therefore extends the analysis, first of all by estimating a simple VAR using ∆y, rr, d/y and ∆yOECD, where rr denotes the real rate of interest and ∆yOECD the rate of growth in OECD as a whole. He then uses the NiGEM model to simulate how a 1 per cent increase in GDP affects the deficit/GDP ratio as a comparison. Figure 9.1 illustrates the EU average of the impulse responses of d/y to output shocks. The impact builds up quite quickly over the first two years before dropping away to zero after ten years. The peak of 0.4 is similar to values derived from Table 7.1 for the individual countries in the sample. The fact that deficits in EU countries appear to be quite sensitive to cyclical fluctuations is good news in the sense that it may help to solve problems caused by country-specific output shocks. If the elasticity of surplus/output ratio to GDP growth is of the magnitude 0.2 to 0.3, the lack of a federal budget may not be such a serious problem as it would be otherwise. Interestingly, the output growth effects on deficits seem to be more important in depressions than in ‘normal times’. This

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Fiscal Responses 209

210 Asymmetry and Aggregation in the EU Figure 9.1

An EU average of the impulse responses of d/y* to growth shocks

0,5

0,4

0,3

0,2

0 1

2

3

4

5

6

7

8

9

10

seems to be because policies appear to be quite different in these two regimes. Examination of the cyclically adjusted deficits reveals that policy seems to be counter-cyclical in bad times but that the opposite holds in good times. Thus, output growth leads to smaller surplus/GDP ratios. This could be explained by tax cuts or discretionary increases in expenditures in boom periods. Given the data, it is rather difficult to say which of these mechanisms dominates for EU countries. The explanation appears to lie on revenue side in the sense that the cyclically adjusted revenues (in relation to trend GDP) seem to decrease when output increases. On the expenditure side, the coefficient of ∆y|∆y > 0 points in the same direction (i.e. to a procyclical output growth effect). In the case of recessions, cyclically adjusted expenditures seem to behave counter-cyclically, while the revenue side is quite passive. Thus in bad times fiscal policy operates mainly via increases in expenditure. And, as mentioned above, in good times discretionary action mainly affects taxes in the form of tax cuts. From the viewpoint of counter-cyclical fiscal policy, the main problem appears to be behaviour in ‘good times’. Although automatic stabilisers seem to operate in this case as well, discretionary action does

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

0,1

Fiscal Responses 211

not seem to help to smooth the output growth path. Expenditures are not cut but instead taxes are lowered rather than increased. Thus in many respects the SGP is quite well adjusted to the natural inclinations of the member states in setting fiscal policy. It needs to push them towards using discretionary policy in a more symmetric and sustainable manner. Thus the current emphasis on trying to operate in surplus or near balance would provide both countervailing pressure and not inhibit the operation of normal automatic stabilisers.

Changes over the period of monetary union

In the previous section we have concentrated on the period before the formation of the euro area, although the data do cover the period of qualification for Stage 3 once the objectives were agreed in the Maastricht Treaty. Since the requirements were to be able to keep the public sector deficit below 3 per cent of GDP and the debt to GDP ratio below 60 per cent (or show adequate progress in bringing the debt ratio down to that level) we could expect a change in behaviour as countries tried to qualify. The assessment was made in the first part of 1998 and hence member states needed to qualify from around 1997 onwards, which implies action in the years before then if the general structure of the fiscal stance needed adjustment.5 In the same way, within Stage 3 the members are bound by the terms of the SGP which also attempts to get countries to bring their debt to GDP ratios down and to keep their deficits less than 3 per cent of GDP. Indeed, as discussed earlier, the objective has been to try to attain a surplus or at least balance in normal times. However, once membership was achieved, the sanctions became different. Countries could be subject to an Excessive Deficit Procedure (EDP) under the terms of the SGP. Thus far no sanctions have been applied and the SGP itself was revised in 2005, making the occurrence of an excessive deficit less likely, nevertheless, the chances are that countries would become increasingly concerned

5 We can probably neglect the first possible qualification date in 1996 as at that point only Luxembourg qualified on all criteria and it was clear early on that there would be insufficient countries to make starting the third stage of monetary union politically feasible. Hence countries are unlikely to have made special efforts during that period. It is a matter of some irony that New Zealand also met all the Maastricht criteria in 1996, although it was in danger of appreciating its exchange rate too far. Thus qualification is something that could occur simply through running a prudent macroeconomic policy, irrespective of whether membership of monetary union was on offer.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

9.2

212 Asymmetry and Aggregation in the EU

Figure 9.2

Median of fiscal variables before and after the EMU

2

.05

1

.04

0

.03

–1

.02

–2

.01

–3

.00

–4

–.01

–5

–.02

–6

–.03

–7 1975

1980

1985

1990

Net lending/GDP

1995

2000

–.04 2005

100 90

0

80

–1

70

–2

60

–3

50

–4

40

–5

30

–6

20 10 1985

1990

Net lending/GDP

8 4 0 –4 –8 1975

1995

2000

Debt/GDP

1980

1985

1990

Net lending/GDP

1

1980

12

Output gap

2

–7 1975

16

2005

1995

2000

2005

Inflation

60 55 50 45 40 35 30 1975

1980

1985

1990

Expenditures/GDP

1995

2000

2005

Revenues/GDP

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

to control deficits as they rose as a proportion of GDP. Although the present recession has meant that much larger deficits are tolerated until growth returns. The implications behind this are that we would see two or possibly three regimes in the data. The period of qualification would entail a very specific effort as failure would lead to not being in monetary union at the start. Hence the sanction is clearly harsher than that through the EDP. A look at the data, Figure 9.2, confirms this. Deficits did indeed decrease markedly after 1995. Similarly the debt ratio began to fall. It is clear that the main adjustment came through expenditures rather than revenues. After monetary union there is little further change and the improvement in debt ratios tails off. Of course if we were to add the most recent years we would see a drastic worsening. However, these remarks ignore the state of the economic cycle. As is clear much of the adjustment is simply on the back of the upturn between 1996 and 2000 and the worsening thereafter reflects the downturn. Nevertheless the fact that deficits in the 2001–2 downturn do not fall as far as their values in the previous

two peaks in the 1975 to 1990 period shows there has been a major change in behaviour. To some extent this may reflect the fact that by this time inflation has also been brought firmly under control. On the one hand, inflation erodes the real value of debt and makes it easier to have negative real interest but on the other low inflation restricts the automatic creep in revenues through a progressive tax system. We do not have enough data to determine all these possible break points in behaviour econometrically but we can explore whether there is a change in behaviour in 1995, when convergence began in earnest, as well as whether there was one in 1999 when the euro area started (Table 9.2). From the first eight rows of Table 9.2 it appears that the disciplining effect of debt on deficits is if anything a little lower after the start of Stage 3. This is surprising as not only is there the traditional constraint from the increased cost of servicing but the Maastricht convergence criteria, which continued into an ongoing commitment, also tried to keep debt ratios below 60 per cent of GDP and encourage steady improvement in fiscal prudence, thus doubling up the incentives. However, the clearest change in behaviour is in the period 1995–2001, when the member states needed to qualify and then before the performance of the euro area began to weaken. The equation used in the estimation is a simplified version of (9.1), where the interest rate has been replaced by the rate of inflation. We should also allow for the fact that, as demonstrated in earlier chapters, other aspects of economic behaviour also show a structural break over this period. Hence where possible we have used the Arellano-Bond panel GMM estimator or GLS. Estimates of the disciplinary effect of debt vary a lot depending on the specification estimated and on the time period. The EMU period appears to be somewhat different from earlier periods e.g. in terms of cyclical sensitivity and the role of inflation but it appears that the disciplinary role of debt is not very significant. In fact, it is the late 1990s which appears to be somewhat different in this respect. The difference can be seen quite clearly by computing a time-varying coefficient for the lagged debt/GDP ratio (Figure 9.3). On the basis of the figure one might say that it is 1995 or 1996 when fiscal behaviour changed towards more disciplinary direction but already in 2000 some deterioration took place. (It is also clear from the figure that each of the oil crises, 1975, 1981 and 1995 caused a step up in impact of debt, only first of which was reversed.) The nature of the change may be better understood by scrutinising the behaviour of expenditures and revenues (see the subsequent four

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Fiscal Responses 213

214 Asymmetry and Aggregation in the EU Evidence of changing fiscal behaviour

Sample Dep.var

g

Lagged def/y–

DW J-stat(df)

debt–1

∆p

R2/SEE

–0.024 (3.33)

0.055 (1.33)

0.403 2.881

0.621

GLS

2.00

GLS

Estimator

1971– 2006 def/y–

0.461 (6.78)

1971– 2006 def/y–

0.385 (8.44)

0.830 (16.07)

0.017 (3.20)

0.023 (0.88)

0.853 0.015

1971– 1998 def/y–

0.243 (5.39)

0.652 (16.74)

0.064 (5.77)

0.120 (3.11)

.. 0.021

.. 38.8 (35)

GMM/AB

1999– 2006 def/y–

0.556 (5.52)

0.610 (5.45)

0.021 (1.06)

–0.104 (0.97)

.. 0.017

.. 13.1 (16)

GMM/AB

1995– 2001 def/y–

0.402 (3.66)

0.673 (10.84)

0.046 (2.09)

–0.264 (2.87)

.. 0.017

.. 18.1 (15)

GMM/AB

1970– 1998 exp/y–

–0.897 (7.45)

0.095 (4.68)

–0.332 (3.44)

0.722 0.037

0.332

LS

1999– 2006 exp/y–

–0.221 (2.68)

0.059 (2.91)

0.102 (0.70)

0.970 0.012

0.897

LS

1970– 1998 tax/y–

–0.431 (4.88)

0.069 (5.81)

–0.167 (2.70)

0.834 0.026

0.375

LS

1999– 2006 tax/y–

0.324 (3.58)

0.044 (2.33)

–0.079 (0.44)

0.971 0.012

0.904

LS

g|gap>0

debt–1

∆log(P)

R2/SEE

1970– 1998 def/y–

0.349 (2.59)

0.552 (4.97)

–0.024 (1.78)

0.154 (2.19)

0.492 2.844

0.518

LS

1995– 2001 def/y–

–0.058 (0.22)

0.542 (2.74)

–0.024 (0.53)

–0.531 (1.90)

0.632 2.024

1.157

LS

1999– 2006 def/y–

0.157 (0.81)

0.519 (4.47)

–0.027 (1.29)

–0.377 (1.25)

0.746 1.450

1.302

LS

g|gap<0

DW

Def denotes net lending (thus positive values represent surpluses), y denotes GDP and y– trend GDP (constructed by the Hodrick–Prescott filter). exp denotes government expenditures and tax government revenues. Debt denotes general government debt in relation to GDP and p the GDP deflator. g denotes the growth rate of GDP. LS denotes panel least squares (with fixed cross-section effects) estimator, GLS the corresponding generalised least squares estimator and GMM/AB the Arellano-Bond GMM estimator with first differences. If one tests the hypothesis that the coefficients of g|gap<0 and g|gap>0 are equal with the last three equations, the F statistics and marginal significance values are: 2.76 (0.098), 16.67 (0.000) and 7.96 (0.006), respectively.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Table 9.2

Fiscal Responses 215 Figure 9.3

Change in the responsiveness to the debt ratio Coefficient of the lagged debt/GDP variable

0.03 0.02 0.01 0.00 –0.01 –0.02 –0.03 –0.04

–0.06 1970

1975

1980

1985

1990

1995

2000

2005

rows in Table 9.2). As these rows show, the effect is not symmetric on expenditures and revenues. Expenditures fell quite strongly compared to GDP when growth rates rose before Stage 3 but the effect was clearly more limited thereafter. Before Stage 3 tax revenues were if anything pro-cyclical. We can see the extent of the asymmetry if we allow the coefficient on the growth rate to be different in down and up phases of the cycle (the last two rows of the table). In the period before Stage 3 there was indeed asymmetry with the response being less when output gaps were negative. In Stage 3 this effect has become stronger (the hypothesis of symmetry can be rejected more decisively). This does imply that the expected effect has occurred and there has been a stronger attempt to contain deficits in downturns.

9.3

Concluding remarks

The sources of asymmetry within the euro economy and the asymmetry of monetary policy addressed in previous chapters set some clear challenges for fiscal policy. Policy needs to be asymmetric itself in order

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

–0.05

to counteract them. Downward pressures on the economy create greater problems for unemployment and participation rates than subsequent upturns of the same size unwind. The limited impact of negative output gaps on inflation, while the positive gaps can have substantial effects, encourages the monetary authority to make sure that inflation does not take off, thereby imposing a limiting factor on the upside. Downside threats however permit and indeed require much stronger policy reactions and here the asymmetry in the behaviour of the monetary authorities suggests that their actions will be very much in tune with the fiscal authorities in that phase of the cycle. However, as we have seen in the present crisis, monetary policy can effectively reach a lower bound where it has only a limited effect, despite the unusual actions in the UK of quantitative easing and credit easing. It is however here that the SGP should cut in as the permitted extent of deficits is limited. This does not appear to be problem for automatic stabilisation but with discretionary actions. Even with extensive buffers, normal fluctuations round a sustainable trend do not seem to generate excessive deficits. Abnormal shocks like the global financial crisis in any case generate exceptions to the excessive deficit procedure because of the decline in GDP (even before the 2005 changes). The problem with discretionary actions is that in good times taxes appear to be cut more than sustainable but are not raised again when the deficit promises to become too large. Correspondingly governments do not cut back on expenditure in good times well enough to balance out the tax cuts and are rather too ready to raise expenditure in the downturn compared to their reluctance to raise taxes. There is therefore a deficit bias across the cycle, a feature the SGP seems designed to help counter. The emphasis of the SGP and wider EU level macroeconomic policy on reducing the general level of debt also seems appropriate as the member states appear to have reached the point where the share of public spending is sufficiently great that it may impair the overall growth rate of the economy. There may therefore be tension between policies designed to offset the impact of downturns and those aimed at faster growth. Matching up the two would require a different balance to the pattern of tax cutting and expenditure increases over the course of the cycle. The SGP pushes in that direction in the downphase but some other pressure is needed to increase the pressure/incentives in the up phase. The present crisis has now made the problem much worse. Some countries that had reached a sustainable position now face politically difficult budgetary consolidation to return to that path. In some cases this can be achieved over the course of the cycle but in others it returns them to the difficulties

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

216 Asymmetry and Aggregation in the EU

that prevailed before the lure of being able to join monetary union led them to change markedly. There is no matching lure now and the pressure will come simply from the difficulties themselves. For some countries, such as Greece, this appears to be insufficient. This leads naturally to one issue that remains – the appropriateness of the ‘penalty’. Imposing financial penalties on those in difficulty makes their short-run position even worse, whether or not the penalty has to be levied. The chances are that the excessive deficits will only be triggered when a country is in a downswing. Thus avoiding the excessive deficit would involve a fiscal tightening exactly when the inclination would if anything be to do the exact opposite. Thus the economy would be pushed into more of a difficulty than it would otherwise. This problem is a good incentive structure for the time consistency problem. If a member state organises itself prudently under normal times then the chance of it being faced by unfortunate pressure to tighten in a downturn will be small. It is thus well motivated not to get into that sort of position. The problem then comes if a country has deliberately or through bad luck got to the point where it will have to apply unfortunate policy or face the fine. The temptation then must be to defy the rules. The better social outcome is probably to carry on with the mistake and then put it right later on when the economy is doing better (even though our results suggest that this tends not to occur). Downturns tend not to be very prolonged so the opportunity to correct the underlying balance (and pay the fine) would not necessarily be delayed for more than a few years. Thus if anything the problem is that the SGP does not threaten effective enough sanctions, especially if the actual behaviour is going to be that the Council of Ministers will shy away from harsh implementation of the Pact once important member states get into difficulty. The softening of the Pact in 2005 would be credible if member states had shown more willingness in the past to adjust without the sanctions. In the longer term, however, when there is no particular call for consolidation one might very well want to move a system that had a rather more sophisticated way of judging whether policy was prudent. However, even now it is necessary to address the issue of how to handle member states that are already well within the debt criterion. There is a second issue of stability here. If all the member states were simultaneously to switch to a much more expansionary stance this will have a much bigger effect on overall policy and the interaction with monetary policy, We have already seen (Chapter 8) that monetary policy reacts much more vigorously to substantial threats to price

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Fiscal Responses 217

218 Asymmetry and Aggregation in the EU

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

stability than to small ones. In part this will be in reaction to the observed behaviour of the fiscal authorities. If the individual member states are out of phase or there are asymmetric shocks then, even if individual governments make big swings in the fiscal policy stance this will have little impact on the overall fiscal balance of the euro area or on monetary policy. It is correlated actions that cause the difficulty. Clearly the SGP would have to become much more complex if its rules for each individual country were to be contingent on the general position of the EU. Since all countries could be trying to improve their own position compared to the others this would result in a very complex game to determine the overall outcome. It would be very understandable if the EU were to stick with rules that apply to each individual country and were contingent purely on that country’s actions and prospects. The more opaque or complex the rule and the more it is open to discussion before it is applied then the more contentious will be the political debate on each occasion. Simple, hard and fast (but fair) rules seem a more likely prospect.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Adam, K. and Padula, M. (2003) ‘Inflation Dynamics and Subjective Expectations in the United States’, European Central Bank Working Paper no.222. Agell, J., Lindh, T. and Ohlsson, H. (1997) ‘Growth and Public Sector: A Critical Review Essay’, European Journal of Political Economy, vol.13, pp.33–52. Ahrend, R. (2008) ‘Monetary Ease – A Factor behind Financial Crises? Some Evidence from OECD Countries’, OECD Discussion Paper 44 (available at http://www.economics-ejournal.org/economics/discussionpapers/2008-44). Aigner, D., Lovell, C.A.K. and Schmidt, P. (1977) ‘Formulation and Estimation of Stochastic Frontier Production Functions’, Journal of Econometrics, vol.6, pp.21–37. Alesina, A., Blanchard, O., Galí, J., Giavazzi, F. and Uhlig, H. (2001) Defining a Macroeconomic Framework for the Euro Area: Monitoring the European Central Bank 3, London: CEPR. Alesina, A., Broeck, M., Prati, A. and Tabellini, G. (1992) ‘Default Risk on Government Debt in OECD Countries’, Economic Policy, vol.15, pp.428–51. Altissimo, F., Georgiou, E., Sastre, T., Valderrama, M.T., Sterne, G., Stocker, M., Weth, M., Whelan, K. and Willman, A. (2005) ‘Wealth and Asset Price Effects on Economic Activity’, ECB Occasional Paper 29. Anderson, G. and Hussey, P. (1999) Health and Population Aging: Multinational Comparison, New York: The Commonwealth Fund. Archibald, G.C. (1969) ‘The Phillips Curve and the Distribution of Unemployment’, American Economic Review, vol.59(2), pp.124–34. Arellano, M. and Bond, S. (1991) ‘Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations’, The Review of Econometric Studies, vol.58, pp.277–97. Artis, M. and Buti, M. (2000) ‘“Close to Balance or in Surplus”: A Policy Maker’s Guide to the Implementation of the Stability and Growth Pact’, Mimeo European University Institute. Artis, M., Krolzig, H.-M. and Toro, J. (1999) ‘The European Business Cycle’, European University Institute Working Paper 99/24. Arulampalam, W. (2001) ‘Is Unemployment Really Scarring? Effects of Unemployment Experiences on Wages’, Economic Journal, vol.111, pp.F585–606. Atkinson, A.B. (1995) ‘The Welfare State and Economic Performance’, National Tax Journal, vol.48(2). Atkinson, A.B. (1999) The Economic Consequences of Rolling Back the Welfare State, Cambridge MA: MIT Press. Attfield, C. and Silverstone, B. (1998) ‘Okun’s Law, Cointegration and Gap Variables’, Journal of Macroeconomics, vol.20, pp.125–37. Baldwin, R. and Krugman, P. (1989) ‘Persistent Trade Effects of Large Exchange Rate Shocks’, Quarterly Journal of Economics, vol.104(4), pp.635–45. Ball, L. (1993) ‘What Determines the Sacrifice Ratio’, NBER Working Paper 4306. Ball, L. and Mankiw, N.G. (1995) ‘Relative Price Changes as Aggregate Supply Shocks’, Quarterly Journal of Economics, vol.110, pp.161–94. 219

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

References

Banque de France (1996) ‘Les Indicateurs des Conditions Monetaire’, Bulletin de la Banque de France, vol.30, Paris. Barro, R.J. (1990) ‘Government Spending in a Simple Model of Economic Growth’, Journal of Political Economy, vol.98, pp.5103–25. Baxter, M. (1994) ‘Real Exchange Rates and Real Interest Differentials: Have We Missed the Business Cycle Relationship?’, Journal of Monetary Economics, vol.33, pp.5–37. Begg, I., Berghman, J., Chassard, Y., Kosonen, P., Madsen, P.K., Matsaganis, M., Mayes, D., Muffels, R., Salais, R. and Tsakloglou, P. (2001) Social Exclusion and Social Protection in the European Union: Policy Issues and Proposals for the Future Role of the EU, London: London South Bank University. Benigno, P. and Lopez-Salido, D. (2002) ‘Inflation Persistence and Optimal Policy in the Euro Area’, International Research Forum on Monetary Policy, European Central Bank. Bernanke, B. and Gertler, M. (1995) ‘Inside the Black Box: The Credit Channel of Monetary Transmission’, Journal of Economic Perspectives, vol.9(4), pp.27–48. Bernanke, B. and Gertler, M. (1999) Monetary Policy and Asset Price Volatility in New Challenges for Monetary Policy, Federal Reserve Bank of Kansas City, pp.77–128. Bienkowski, W., Brada, J.C. and Radlo, M. (2008) Growth versus Security: Old and New EU Members Quest for a New Economic and Social Model, Basingstoke: Macmillan. Blanchard, O. (1998) ‘Revisiting European Unemployment: Unemployment, Capital Accumulation, and Factor Prices’, NBER Working Papers no.6566. Blanchard, O. (2000) ‘What Do We Know about Macroeconomics that Fisher and Wicksell Did Not?’, Quarterly Journal of Economics, vol.115(4), pp.1375–409. Blanchard, O. and Perotti, R. (1999) ‘An Empirical Characterisation of the Dynamic Effects of Changes in Government Spending and Taxes on Output’, NBER Working Paper no.79. Blanchard, O. and Wolfers, J. (2000) ‘The Role of Shocks and Institutions in the Rise of European Unemployment: The Aggregate Evidence’, Economic Journal, vol.110, pp.C1–33. Blinder, A. and Reis, R. (2005) ‘Understanding the Greenspan Standard’, Proceedings of the Jackson Hole Conference, pp.11–96, Kansas City: Federal Reserve Bank of Kansas City. Blinder, A. and Solow, R. (1973) ‘Does Fiscal Policy Matter?’, Journal of Public Economics, vol.2, pp.319–37. Brainard, W. (1967) ‘Uncertainty and Effectiveness of Policy’, American Economic Review, vol.57, pp.411–25. Brandner, P., Diebalek, L. and Schuberth, H. (1998) ‘Structural Budget Deficits and Sustainability of Fiscal Positions in the European Union’, Österreichische Nationalbank, Working Paper no.26. Brechling, F. (1973) ‘Wage Inflation and the Structure of Regional Unemployment’, Journal of Money, Credit and Banking, vol.5(1), pp.355–79. Bryan, M.F., Cecchetti, S.G. and O’Sullivan, R. (2001) Asset Prices in the Measurement of Inflation, De Nederlandsche Bank Staff Report no.62. Buti, M., Franco, D. and Ongena, H. (1998) ‘Fiscal Discipline and Flexibility in EMU: The Implementation of the Stability and Growth Pact’, Oxford Review of Economic Policy, vol.14, pp.81–97.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

220 References

Buxton, A. and Mayes, D.G. (1986) ‘The Persistence of Unemployment’, Quarterly Bulletin, no.88 (December), pp.29–37. Caballero, R.J. and Hammour, M.L. (1996) ‘On the Timing and Efficiency of Creative Destruction’, Quarterly Journal of Economics, vol.111(3), pp.805–52. Campbell, J.Y. and Cocco, J.F. (2007) ‘How Do House Prices Affect Consumption? Evidence from Micro Data’, Journal of Monetary Economics, vol.54 (April), pp.591–621. Carroll, C., Otsuka, M. and Slacalek, J. (2006) ‘What is the Wealth Effect on Consumption: A New Approach’, mimeo, John Hopkins University. Case, K.E., Quigley, J.M. and Shiller, R.J. (2005) ‘Comparing Wealth Effects: The Stock Market Versus the Housing Market’, Advances in Macroeconomics, vol.5, pp.1–32. Catte, P., Girouard, N., Price, R. and André, C. (2004) ‘Housing Market Wealth and the Business Cycle’, Working Paper 17, OECD. Cayen, J.-P. and van Norden, S. (2002) ‘La fiabilité des estimations du l’écart de production au Canada’, Bank of Canada Working Paper 2002–10. Cecchetti, S.G., Genburg, H., Lipsky, J. and Wadhwani, S. (2000) Asset Prices and Central Bank Policy. Geneva Report on the World Economy, no.2, International Center for Monetary and Banking Studies and CEPR. Chan, K.S. (1993) ‘Consistency and Limiting Distribution of the Least Squares Estimator of a Threshold Autoregressive Model’, Annals of Statistics, vol.21, pp.520–33. Chassard, Y. (2001) ‘European Integration and Social Protection: From the Spaak Report to the Open Method of Coordination’, pp.277–305, in Mayes et al. (2001). Chetty, V. and Heckman, J. (1986) ‘A Dynamic Model of Aggregate Supply, Factor Demand and Entry and Exit for a Competitive Industry with Heterogeneous Plants’, Journal of Econometrics, vol.64, pp.237–62. Cho, S. and Moreno, A. (2006) ‘A Small-Sample Study of the New-Keynesian Macro Model’, Journal of Money, Credit and Banking, vol.38(6), pp.1461–81. Christoffel, K., Coenen, G. and Warne, A. (2008) ‘The New Area-Wide Model of the Euro Area: A Micro Founded Open-Economy Model for Forecasting and Policy Analysis’, ECB Working Paper 944. Clark, P.B. and Laxton, D. (1997) ‘Phillips Curves, Phillips Lines and the Unemployment Costs of Overheating’, IMF (August). Clark, P.B., Laxton, D. and Rose, D. (1996) ‘Asymmetry in the US Output-Inflation Nexus: Issues and Evidence’, IMF Staff Papers, vol.43, no.1. Coats, W., Laxton, D. and Rose, D. (2003) The Czech National Bank’s Forecasting and Policy Analysis System, Prague: CNB. Cohen, D. and Folette, G. (1999) ‘The Automatic Stabilisers: Quietly Doing Their Thing’, Federal Reserve Board. Collins, S. and Siklos, P. (2002) ‘Optimal Reaction Functions, Taylor’s Rule and Inflation Targets: The Experience of Dollar Bloc Countries’, mimeo, Wilfrid Laurier University. Corrado, L. and Holly, S. (2003) ‘Non Linear Phillips Curve, Mixing Feedback Rules and the Distribution of Inflation and Output’, CEIS Tor Vergata Research Paper 37, October. Cuaresma, J.C. (2003) ‘Okun’s Law Revisited’, Oxford Bulletin of Economics and Statistics, vol.65(4), pp.439–51.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

References 221

Dalsgaard, T. and De Serres, A. (1999) ‘Estimating Prudent Budgetary Margins for 11 EU Countries: A Simulated SVAR Model Approach’, OECD Economics Department Working Papers no.216. Daveri, F. and Tabellini, G. (2000) ‘Unemployment, Growth and Taxation in Industrial Countries’, Economic Policy, vol.30, pp.47–104. Davis, S.J., Haltiwanger, J.C. and Schuh, S. (1996) Job Creation and Destruction, Cambridge MA: MIT Press. Debelle, G. and Laxton, D. (1996) ‘Is the Phillips Curve Really a Curve? Some Evidence for Canada, the United Kingdom and the United States’, IMF Working Paper WP/96/111 (October). Diebold, F.X. and Rudebusch, G.D. (1990) ‘A Nonparametric Investigation of Duration Dependence in the American Business Cycle’, Journal of Political Economy, vol.98(3), pp.596–616. Diebold, F.X. and Rudebusch, G. (1999) Business Cycles: Durations, Dynamics and Forecasting, Princeton: Princeton University Press. Disney, R., Gathergood, J. and Henley, A. (2007) ‘House Price Shocks, Negative Equity and Household Consumption in the United Kingdom’, mimeo, University of Nottingham. Dornbusch, R., Favero, C. and Giavazzi, F. (1998) ‘Immediate Challenges for the European Central Bank’, Economic Policy, no.26, pp.17–52. Duesenberry, J. (1949) Income, Saving, and the Theory of Consumer Behavior, Cambridge, MA: Harvard University Press. Duguay, P. (1994) ‘Empirical Evidence on the Strength of the Monetary Transmission Mechanism in Canada’, Journal of Monetary Economics, vol.33, pp.39–61. Dupasquier, C., Lalonde, R. and St Amant, P. (1997) ‘Optimum Currency Areas as Applied to Canada and the United States’, in Bank of Canada, Exchange Rates and Monetary Policy, Ottawa. Dupasquier, C. and Ricketts, N. (1998) ‘Nonlinearities in the Output-Inflation Relationship: Some Empirical Results from Canada’, Bank of Canada Working Paper 98–14 also in Bank of Canada, Exchange Rates and Monetary Policy, Ottawa. Eichengreen, B. and Wyplosz, C. (1998) ‘The Stability Pact: More than a Minor Nuisance?’, Economic Policy, no.26, pp.67–104. Eika, K.H., Ericsson, N.R. and Nymoen, R. (1997) ‘Hazards in Implementing a Monetary Conditions Index’, Oxford Bulletin of Economics and Statistics, vol.58(4), pp.765–90. Eliasson, A.-C. (1999) ‘Is the Short-run Phillips Curve Nonlinear? Empirical Evidence from Australia, Sweden and the United States’, mimeo, Department of Economic Statistics, Stockholm School of Economics. Enders, W. and Siklos, P. (1999) ‘Cointegration and Threshold Adjustment’, Working Paper, University of California, San Diego. Enders, W. and Siklos, P. (2001) ‘Cointegration and Threshold Adjustment’, Journal of Business and Economic Statistics, vol.19(2), pp.166–76. Ericsson, N., Jansen, E., Kerbishian, N. and Nymoen, R. (1997) ‘Understanding a Monetary Conditions Index’, Federal Reserve Board, mimeo. Esping-Adersen, G. (1990) The Three Worlds of Welfare Capitalism, Oxford: Polity Press. European Commission (2001) The 2001 Broad Economic Policy Guidelines, Brussels.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

222 References

European Commission (2002) ‘Co-ordination of Economic Policies in the EU: A Presentation of Key Features of the Main Procedures’, Euro Paper no.45, Directorate General for Economic and Financial Affairs, Brussels. Evans, G.W. and Honkapohja, S. (2002) ‘Adaptive Learning and Monetary Policy Design’, Bank of Finland Discussion Paper, 29/2002. Feldstein, M. (1997) ‘The Costs and Benefits of Going from Low Inflation to Price Stability’, in NBER Reducing Inflation: Motivation and Strategy, pp.123–66, New York: NBER. Folk, D., Van Dijk, D. and Franses, P. (2004) ‘Forecasting Aggregates Using Panels of Nonlinear Time Series’, Econometric Institute Report 2004–44. Forni, M., Hallin, M., Lippi, M. and Reichlin, L. (2001) ‘Coincident and Leading Indicators for the Euro Area’, Economic Journal, vol.111, pp.62–85. Friedman, M. (1964) ‘Monetary Studies of the National Bureau’, The National Bureau Enters Its 45th Year, 44th Annual Report, 7–25; reprinted in Friedman, M., The Optimum Quantity of Money and Other Essays, Ch.12, pp.261–84, Chicago: Aldine, 1969. Friedman, M. (1968) ‘The Role of Monetary Policy’, American Economic Review, vol.58, pp.1–17. Friedman, M. (1993) ‘The “Plucking Model” of Business Fluctuations Revisited’, Economic Enquiry (April), pp.171–7. Fuhrer, J.C. and Moore, G.R. (1995) ‘Inflation Persistence’, Quarterly Journal of Economics, vol.114, pp.127–59. Fujita, M., Krugman, P. and Venables, A.J. (1999) The Spatial Economy, Cambridge MA: MIT Press. Gagnon, E. and Kahn, H. (2001) ‘New Phillips Curve with Alternative Marginal Cost Measures for Canada, US and Euro Area’, Bank of Canada Working Paper 2001-25. Gaiotti, E. and Generale, A. (2001) ‘Does Monetary Policy Have Asymmetric Effects? A Look at the Investment Decisions of Italian Firms’, ECB Working Paper 110. Galí, J. and Gertler, M. (1999) ‘Inflation Dynamics: A Structural Econometric Analysis’, Journal of Monetary Economics, vol.44, pp.115–57. Galí, J., Gertler, M. and López-Salido, D. (2001) ‘European Inflation Dynamics’, European Economic Review, vol.45, pp.1237–70. Goodfriend, M. and King, R.G. (2001) ‘The Case for Price Stability’, NBER Working Paper 8423. Goodhart, C.A.E. (2001) ‘What Weight Should Be Given to Asset Prices in the Measurement of Inflation?’, Economic Journal, vol.111 (June), F336–57. Goodhart, C.A.E. and Hofmann, B. (2000) ‘Asset Prices and the Conduct of Monetary Policy’, Sveriges Riksbank and Stockholm School of Economics Conference on Asset Markets and Monetary Policy, Stockholm, June. Goodhart, C. and Hofmann, B. (2005) ‘The IS Curve and the Transmission of Monetary Policy: Is There a Puzzle?’, Applied Economics, vol.37(1), pp.29–36. Gordon, R. (1997) ‘The Time Varying NAIRU and Its Implications for Economic Policy’, Journal of Economic Perspectives, vol.11 (winter), pp.11–32. Gordon, R. (2000) ‘Does the New Economy Match Up to the Great Inventions of the Past?’, Journal of Economic Perspectives, vol.14(4), pp.49–74. Grande, G. (1997) ‘Properties of the Monetary Conditions Index’, Banca d’Italia Temi di Discussione no.324.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

References 223

Granger, C. (1987) ‘Implications of Aggregation with Common Factors’, Econometric Theory, vol.3, pp.208–22. Granger, C. (1988) ‘Aggregation of Time Series Variables: A Survey’, Federal Reserve Bank of Minneapolis Institute for Empirical Macroeconomics Discussion Paper 1. Granger, C. and Lee, T.-H. (1993) ‘The Effect of Aggregation on Nonlinearity’, in Mariano, R. (ed.) Advances in Statistical Analysis and Statistics Computing iii, Greenwich Coon: JAI Press. Granger, C. and Lee, T.-H. (1999) ‘The Effect of Aggregation on Nonlinearity’, Econometric Reviews, vol.18(3), pp.259–69. Granger, C.W.J. and Teräsvirta, T. (1993) Modelling Nonlinear Economic Relationships, Oxford: Oxford University Press. Greenspan, A. (1996) ‘The Challenge of Central Banking in a Democratic Society’, Francis Boyer Lecture of the American Enterprise Institute for Public Policy Research, Washington, D.C., December 5. Greenspan, A. (2007) The Age of Turbulence: Adventures in a New World, New York: Penguin Books. Gruen, D., Robinson, T. and Stone, A. (2002) ‘Output Gaps in Real Time: Are They Reliable Enough to Use for Monetary Policy?’, Reserve Bank of Australia, Research Discussion Paper 2002-06. Haldane, A. and Quah, D. (1999) ‘UK Phillips Curves and Monetary Policy’, Bank of England, October. Haltiwanger, J.C. and Schuh, S. (1999) ‘Gross Job Flows between Plants and Industries’, New England Economic Review (March/April), pp.41–64. Hamilton, J. (2008) ‘Regime-Switching Models’, in Durlauf, S. and Blume, L. (eds) New Palgrave Dictionary of Economics, 2nd edition, Basingstoke: Palgrave Macmillan. Hamermesh, D. and Soss, N. (1975) ‘An Economic Theory of Suicide’, Journal of Political Economy, vol.83, pp.83–98. Hansen, B. (1999a) ‘Testing for Linearity’, Journal of Economic Surveys, vol.13, pp.551–76. Hansen, B. (1999b) ‘Threshold Effects in Non-Dynamic Panels: Estimation, Testing and Inference’, Journal of Econometrics, vol.93, pp.345–68. Hansen, B. (2000) ‘Testing for Structural Change in Conditional Models’, Journal of Econometrics, vol.97, pp.93–115. Hansen, L. (1982) ‘Large Sample Properties of Generalized Method of Moments Estimators’, Econometrica, vol.50, pp.1029–54. Haque, N., Pesaran, M. and Sharma, S. (1999) ‘Neglected Heterogeneity and Dynamics in Cross Country Savings Regressions’, IMF Working Paper 99/128. Harris, R. and Silverstone, B. (1999a) ‘Asymmetric Adjustment of Unemployment and Output in New Zealand: Rediscovering Okun’s Law’, University of Waikato Working Paper 99/1. Harris, R. and Silverstone, B. (1999b) ‘Asymmetric Adjustment of Unemployment and Output: Cross-country Estimates of Okun’s Law’, University of Waikato Discussion Paper 99/2. Hendry, D.F. (1995) Dynamic Econometrics, Oxford: Oxford University Press. Hodson, D. and Maher, I. (2001) ‘The Open Method as a New Mode of Governance: The Case of Soft Economic Policy Co-ordination’, Journal of Common Market Studies, vol.39(4), pp.719–46.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

224 References

Holmes, M. and Maghrebi, N. (2005) ‘Are International Interest Rate Linkages Characterized by Asymmetric Adjustment?’, Journal of International Financial Markets, Institutions and Money, vol.16(4), pp.384–96. Holmes, M. and Silverstone, B. (2005) ‘Okun’s Law, Asymmetries and Jobless Recoveries in the United States: A Markov Switching Model’, University of Waikato Working Paper in Economics 6/05. Honkapohja, S. and Mitra, K. (2005) ‘Performance of Monetary Policy with Internal Central Banks Forecasting’, Journal of Economic Dynamics and Control, vol.29(4), pp.627–58. Hosomatsu, Y. (1970) ‘Asymmetric Loss Function and Optimal Policy Under Uncertainty’, Decision Sciences, vol.11, pp.577–85. Huang, A., Margaritis, D. and Mayes, D. (2001) ‘Monetary Policy Rules in Practice: Evidence from New Zealand’, Multinational Finance Journal, vol.5(3), pp.175–200. Iacoviello, M. (2000) ‘House Prices and the Macroeconomy in Europe: Results from a Structural VAR Analysis’, European Central Bank Working Paper no.18, April. Issing, O. (2002) ‘On Macro-economic Policy Co-ordination in EMU’, Journal of Common Market Studies, vol.40(2), pp.345–58. Issing, O., Gaspar, V., Angeloni, A. and Tristani, O. (2001) Monetary Policy in the Euro Area, Cambridge: Cambridge University Press. Jacquet, P. and Pisani-Ferry, J. (2001) Economic Policy Co-ordination in the EuroZone: What has been Achieved? What Should be Done?, London: Centre for European Reform. Johansson, F. et al. (2002) ‘The Committee on Stabilisation Policy for Full Employment if Sweden Joins the Monetary Union’, Stockholm. Jondeau, E. and Le Bihan, H. (2003) ‘ML vs. GMM Estimates of Hybrid Macroeconomic Models: With an Application to the “New Phillips Curve”’, Banque de France. Jones, D., Peat, M. and Stevenson, M. (1996) ‘Does the Process of Spatial Aggregation of the U.K. Unemployment Rate Series to Induce or Remove Evidence of Asymmetry in the Business Cycle?’, School of Finance and Economics in University of Technology Sydney WP 67. Karras, G. (1996) ‘The Optimal Government Size: Further International Evidence on the Productivity of Government Services’, Economic Inquiry, vol.34, pp.193–203. Kaufman, S. (1988) Likelihood Based Inference in Cointegrated Vector Autoregressive Models, Oxford: Oxford University Press. Kavkler, A., Mikek, P., Böhm, B. and Borsic, D. (2008) ‘Nonlinear Econometric Models: The Smooth Transition Regression Approach’, available at http://www. cerge.cuni.cz/ pdf/gdn/RRCVI_36_paper_01.pdf Keynes, J.M. (1936) The General Theory of Employment, Interest and Money, London: Macmillan. Kiander, J. and Virén, M. (2001) ‘Measuring Labour Market Flexibility in the OECD Countries’, Empirica, vol.2, pp.187–202. Kilponen, J., Mayes, D.G. and Vilmunen, J. (2000) ‘Labour Market Flexibility in the Euro Area’, European Business Journal, vol.12(2), pp.100–10. Kim, C.-J. and Nelson, C.R. (1999) ‘Friedman’s Plucking Model of Business Fluctuations: Tests and Estimates of Permanent and Transitory Components’, Journal of Money Credit and Banking, vol.31(3), pp.317–34.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

References 225

King, M. (1994) ‘Debt Deflation: Theory and Evidence’, European Economic Review, vol.38, pp.419–35. Koenders, K. and Rogerson, R. (2005) ‘Organizational Dynamics over the Business Cycle: A View on Jobless Recoveries’, Federal Reserve Bank of St. Louis Review, July, pp.555–80. Kortelainen, M. (2000) ‘Actual and Perceived Monetary Policy Rules in a Dynamic General Equilibrium Model of the Euro Area’, mimeo, Bank of Finland (July). Kortelainen, M. (2001) ‘Heterogenous Expectations of Monetary Policy: Some Numerical Experiments with a Calibrated Euro-area Dynamic General Equilibrium Model’, mimeo, Bank of Finland. Kortelainen, M. and Mayes, D.G. (2001) ‘Using EDGE – A Dynamic General Equilibrium Model of the Euro Area’, pp.41–78, in Hall, S.G., Heilemann, U. and Pauly, P. (eds) Macroeconomic Models and European Monetary Union, Berlin: Schriftenreihe, Duncker and Humbolt. Koskela, E. and Virén, M. (1990) ‘Monetary Policy Reaction Functions and the Saving Investment Correlations: Some Cross-Country Evidence’, Bank of Finland Discussion Paper 11/90. Koskela, E. and Virén, M. (2000) ‘Is There a Laffer Curve Between Aggregate Output and Public Sector Employment?’, ETLA (the Research Institute for the Finnish Economy) Discussion Paper 737, October. Labhard, V., Sterne, G. and Young, C. (2005) ‘The Wealth Effects on Consumption in Industrialised Countries’, Bank of England Working Paper 275. Land, K., Cantor, D. and Russell, S. (1995) ‘Unemployment and Crime Rate Fluctuations in the Post-World War II United States: Statistical Time-Series Properties and Alternative Models’, in Hagan, J. and Peterson, R. (eds) Crime and Inequality, pp.55–79, Stanford: Stanford University Press. Laxton, D., Meredith, G. and Rose, D. (1995) ‘Asymmetric Effects of Economic Activity on Inflation’, IMF Staff Papers, vol.42(2). Laxton, D., Rose, D. and Tambakis, D. (1999) ‘The US Phillips Curve: The Case for Asymmetry’, Journal of Economic Dynamics & Control, vol.23, pp.1459–85. Laxton, D., Rose, D. and Tetlow, R. (1993) ‘Problems in Identifying Nonlinear Phillips Curves: Some Further Consequences of Mismeasuring Potential Output’, Bank of Canada Working Paper 93-6. Lilien, D. (1982) ‘Sectoral Shifts and Cyclical Unemployment’, Journal of Political Economy, vol.90(4), pp.777–93. Lindé, J. (2002) ‘Estimating New Keynesian Phillips Curves: A Full Information Maximum Likelihood Approach’, Sveriges Riksbank Working Paper no.129, April. Linzert, T. (2005) ‘The Unemployment Inflation Trade-Off in the Euro Area’, IZA Discussion Papers 1699, Institute for the Study of Labor (IZA). Lipsey, R.G. (1960) ‘The Relationship Between Unemployment and the Rate of Change of Money Wages in the United Kingdom 1862–1957: A Further Analysis’, Economica, vol.27, pp.1–31. Loupias, C., Savignac, F. and Sevestre, P. (2001) ‘Monetary Policy and Bank Lending in France: Are There Asymmetries?’, ECB Working Paper 101. Lucas, R.E. (1972) ‘Expectations and the Neutrality of Money’, Journal of Economic Theory, vol.4, pp.103–24. Lucas, R.E. (1973) ‘Some International Evidence on Output-inflation Tradeoffs’, American Economic Review, vol.63, pp.326–34. Lucas, R.E. (1976) ‘Econometric Policy Evaluations: A Critique’, in Brunner, K. and Meltzer, A. (eds) The Phillips Curve and Labour Markets, Carnegie-Rochester

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

226 References

Conference Series on Public Policy, vol.1, pp.19–46. New York: American Elsevier. Ludwig, A. and Sløk, T. (2002) ‘The Impact of Changes in Stock Prices and House Prices on Consumption in OECD Countries’, IMF Working Paper. Luukkonen, R., Saikkonen, P. and Teräsvirta, T. (1988) ‘Testing Linearity Against Smooth Transition Autoregressive Models’, Biometrika, vol.75(3), pp.491–9. Macleannan, D., Muellbauer, J. and Stephens, M. (1999) ‘Asymmetries in Housing and Financial Market Institutions and EMU’, CEPR Discussion Paper 2062. Mayes, D.G. (1979) The Property Boom, Oxford: Martin Robertson. Mayes, D.G. (1986) Change, Occasional Paper NZ Institute of Economic Research. Mayes, D.G. (1996) Sources of Productivity Growth, Cambridge: Cambridge University Press. Mayes, D.G. (2007) ‘Monetary Policy Strategy and Practice in Australia, Canada and New Zealand: Some Current Issues’, BoFOnline, 2007, no.12. Mayes, D.G. (2008) ‘How Should Central Banks Define Price Stability: Comments on Mark A. Wynne’, in Mayes, D.G. and Wood, G.E. (eds) Designing Central Banks, London: Routledge. Mayes, D.G., Berghman, J. and Salais, R. (2001) Social Exclusion and European Policy, Cheltenham: Edward Elgar. Mayes, D.G. and Chapple, B.R. (1995) ‘The Costs and Benefits of Disinflation: A Critique of the Sacrifice Ratio’, Reserve Bank of New Zealand Bulletin, vol.589(1), pp.9–21. Mayes, D.G., Harris, C. and Lansbury, M. (1994) Inefficiency in Industry, Hemel Hempstead: Harvester Wheatsheaf. Mayes, D.G. and Silverstone, B. (1998) ‘Institutional Reform in a Labour Market: The Case of New Zealand’, Tinbergen Institute, May. Mayes, D.G. and Suvanto, A. (2002) ‘Beyond the Fringe: Finland and the Choice of Currency’, Journal of Public Policy, vol.22, pp.43–64. Mayes, D.G. and Tarkka, J. (2002) ‘Transparency in European Central Banks: The Role of Forecasts’, Contemporary Politics and Economics of Europe, vol.11(3), pp.245–70. Mayes, D.G. and Virén, M. (1998) ‘The Exchange Rate and Monetary Conditions in the Euro Area’, Bank of Finland Discussion Paper 27/98. Mayes, D.G. and Virén, M. (2000a) ‘Exchange Rate Considerations in a Small Open Economy: A Critical Look at the MCI as a Possible Solution’, pp.585–628 in Mahadeva, L. and Sterne, G. (eds) Monetary Policy Regimes in a Global Context, London: Routledge. Mayes, D.G. and Virén, M. (2000b) ‘The Exchange Rate and Monetary Conditions in the Euro Area’, Weltwirtschaftliches Archiv, vol.136(2), pp.199–231. Mayes, D.G. and Virén, M. (2001) ‘Macroeconomic Factors, Policies and the Development of Social Exclusion’, Centre for Research on Europe Discussion Paper, University of Canterbury New Zealand, May. Mayes, D.G. and Virén, M. (2002a) ‘Asymmetry and the Problem of Aggregation in the Euro Area’, Empirica, vol.29(1), pp.47–73. Mayes, D.G. and Virén, M. (2002b) ‘Financial Conditions Indexes’, Economia Internazionale, vol.55(4), pp.522–50. Mayes, D.G. and Virén, M. (2002c) ‘Macroeconomic Factors, Policies and the Development of Social Exclusion’, in Muffels, R.J.A., Tsakloglou, P. and Mayes, D.G. (eds) Social Exclusion in European Welfare States, pp.21–50. Cheltenham: Edward Elgar.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

References 227

Mayes, D.G. and Virén, M. (2003) ‘Problems of Asymmetry in European Monetary Policy’, 8th EUSA Conference, Nashville, March. Mayes, D.G. and Virén, M. (2004) ‘Pressures on the Stability and Growth Pact from Asymmetries in Policy’, Journal of European Public Policy, vol.11, pp.781–97. Mayes, D.G. and Virén, M. (2005) ‘Monetary Policy Problems for Currency Unions’, Economic Modelling, vol.22, pp.219–51. Mayes, D.G. and Virén, M. (2007) ‘The SGP and the ECB: An Exercise in Asymmetry’, Journal of Financial Transformation, vol.19, pp.159–75. McDonald, I.M. (1997) ‘The Australian Inflation Target and the Shape of the Short-run Phillips Curve’, University of Melbourne, mimeo. Melitz, J. (1997) ‘Some Cross-Country Evidence about Debt, Deficits and the Behaviour of Monetary and Fiscal Authorities’, CEPR Discussion Paper 1653. Milne, A.K.L. (2009) ‘Macro-prudential Policy: Is It Really So New?’, mimeo, Cass Business School. Mishkin, F. (1995) ‘The Channels of Monetary Transmission: Lessons for Monetary Policy’, NBER Working Paper 5464. Moosa, I.A. (1997) ‘A Cross Country Comparison of Okun’s Coefficient’, Journal of Comparative Economics, vol.24, pp.335–56. Morduch, J. (1995) ‘Income Smoothing and Consumption Smoothing’, Journal of Economic Perspectives, vol.9(3), pp.103–14. Muellbauer, J. (1992) ‘Anglo-German Differences in Housing Market Dynamics: The Role of Institutions and Macroeconomic Policy’, European Economic Review, vol.36, pp.539–48. Muffels, R., Tsakloglou, P. and Mayes, D. (2002) Social Exclusion and European Welfare States, Cheltenham: Edward Elgar. Nadal De Simone, F. and Clarke, S. (2007) ‘Asymmetry in Business Fluctuations: International Evidence on Friedman’s Plucking Model’, Journal of International Money and Finance, vol.26(1), pp.64–85. Neiss, K.S. and Nelson, E. (2002) ‘Inflation Dynamics, Marginal Cost and the Output Gap: Evidence from Three Countries’, Federal Reserve Bank of San Francisco available at http://www.frbsf.org/economics/conferences/0203/nkpenn.pdf. Nelson, E. and Nikolov, K. (2001) ‘UK Inflation in the 1970s and 1980s: The Role of Output Gap Mismeasurement’, Bank of England Working Paper 148. Nickell, S. (1998) ‘Unemployment: Questions and Some Answers’, Economic Journal, vol.108, pp.802–16. Nickell, S. and Bell, B. (1995) ‘The Collapse in Demand for Unskilled and Unemployment across the OECD’, Oxford Review of Economic Policy, vol.11(1), pp.40–62. Nickell, S., Jones, P. and Quintini, G. (2002) ‘A Picture of Job Insecurity Facing British Men’, Economic Journal, vol.112, pp.1–27. NiGEM (1999) The World Model Manual. National Institute of Economic and Social Research. London. Novo, A. and Pinheiro, M. (2003) ‘Uncertainty and Risk Analysis of Macroeconomic Forecasts: Fan Charts Revisited’, Banco de Portugal Working Paper 19-03. Oates, W.E. (1999) ‘An Essay on Fiscal Federalism’, Journal of Economic Literature, vol.37(3), pp.1120–49. Obstfeld, M. and Peri, G. (1998) ‘Asymmetric Shocks’, Economic Policy, no.26, pp.207–47.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

228 References

Okun, A. (1962) ‘Potential GDP: Its Measurement and Significance’, Proceedings of the Business and Economics Statistics Section of the American Statistical Association, pp.98–104. Oliner, S.D. and Sichel, D.E. (2000) ‘The Resurgence of Growth in the Late 1990s: Is Information Technology the Story?’, mimeo, Board of Governors of the Federal Reserve System (February). Orphanides, A. (2001) ‘Monetary Policy Rules Based on Real Time Data’, American Economic Review, vol.91(4), pp.964–85. Orphanides, A. and van Norden, S. (2002) ‘The Reliability of Output Gap Estimates in Real Time’, Review of Economics and Statistics, vol.84(4), pp.569–83. Orphanides, A. and Wilcox, D. (1996) ‘The Opportunistic Approach to Disinflation’, Board of Governors of the Federal Reserve System Discussion Paper 96-24. Oswald, A. (1997) ‘The Missing Piece of the Unemployment Puzzle’, Inaugural Lecture, University of Warwick. Pagan, A. (1984) ‘Econometric Issues in the Analysis of Regressions with Generated Regressors’, International Economic Review, vol.25(1), pp.221–47. Palley, T. (1993) ‘Okun’s Law and the Asymmetric and Changing Cyclical Behaviour of the US Economy’, International Review of Applied Economics, vol.7, pp.144–62. Paloviita, M. and Mayes, D.G. (2005) ‘The Use Real Time Information in Phillips Curves for the Euro Area’, North American Journal of Economics and Finance, vol.16(3), pp.415–34. Paloviita, M. and Mayes, D.G. (2006) ‘Asymmetry and Inflation Dynamics in Different Specifications of the Phillips Curve for the Euro Area’, in Crowley, P., EU Economic Policy at the Dawn of the Century, pp.107–37, New York: Nova Science and in Current Politics and Economics of Europe, vol.19(3–4), pp.203–32, 2008. Paloviita, M. and Virén, M. (2005) ‘The Role of Expectations in the Inflation Process in the Euro Area’, Bank of Finland Discussion Paper 2005/6. Peeters, M. (1998) ‘Monetary Conditions in Europe: A Methodological Analysis’, De Nederlandsche Bank. Peltzman, S. (2000) ‘Prices Rise Faster than They Fall’, Journal of Political Economy, vol.108(3), pp.466–502. Phelps, E. (1967) ‘Phillips Curves, Expectations of Inflation and Optimal Inflation Over Time’, Economica NS, vol.35, pp.254–81. Phelps, E. and Zoega, G. (2001) ‘Structural Booms’, Economic Policy, no.32, pp.83–126. Phillips, A.W. (1958) ‘The Relationship between Unemployment and the Rate of Change of Money Wage Rates in the United Kingdom, 1861–1957’, Economica NS, vol.25, pp.283–99. Prachowny, M. (1993) ‘Okun’s Law: Theoretical Foundations and Revised Estimates’, Review of Economics and Statistics, vol.75, pp.331–6. Priesley, M. (1988) Non-linear and Non-stationary Time Series Analysis, London: Academic Press. Pyyhtiä, I. (1999) ‘The Nonlinearity of the Phillips Curve and European Monetary Policy’, Bank of Finland Discussion Paper 17/99. Quandt, R. (1988) The Econometrics of Disequilibrium, New York: Basil Blackwell.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

References 229

Razzak, W. (1997) ‘The Inflation-Output Trade-off: Is the Phillips Curve Symmetric? A Policy Lesson from New Zealand’, Discussion Paper G97/2, Reserve Bank of New Zealand. Reis, R. and Watson, M. (2007) ‘Measuring Changes in the Value of the Numeraire’, mimeo, Princeton University, May. Roberts, J.M. (1997) ‘Is Inflation Sticky?’, Journal of Monetary Economics, vol.39, pp.173–96. Roberts, J.M. (1998) ‘Inflation Expectations and the Transmission of Monetary Policy’, Board of Governors of the Federal Reserve System. Robinson, T., Stone, A. and van Zyl, M. (2003) ‘The Real-time Forecasting Performance of Phillips Curves’, Reserve Bank of Australia Research Discussion Paper 2003-12. Rogoff, K. (1985) ‘The Optimum Degree of Commitment to a Monetary Target’, Quarterly Journal of Economics, vol.100, pp.1169–90. Rudd, J. and Whelan, K. (2001) ‘New Tests of the New-Keynesian Phillips Curve’, Federal Reserve Board, Division of Research and Statistics. Rudd, J. and Whelan, K. (2002) ‘Should Monetary Policy Target Labor’s Share of Income?’, Federal Reserve Board, Division of Research and Statistics. Rünstler, G. (2002) ‘The Information Content of Real-Time Output Gap Estimates: An Application to the Euro Area’, European Central Bank Working Paper 182. Sapir, A. (2006) ‘Globalization and the Reform of European Social Models’, Journal of Common Market Studies, vol.44(2), pp.369–90. Sargent, T. (2001) The Conquest of American Inflation, Princeton: Princeton University Press. Sbordone, A.M. (2002) ‘Prices and Unit Labor Costs: A New Test of Price Stickiness’, Journal of Monetary Economics, vol.49, pp.265–92. Schaling, E. (1999) ‘The Non-linear Phillips Curve and Inflation Forecast Targeting’, Bank of England Working Papers 98, Bank of England. Scherer, F.M. (1999) New Perspectives on Economic Growth and Technological Innovation, Washington DC: Brookings Institution. Scherer, F.M. (2001) ‘An Early Application of the Average Total Cost Concept’, Journal of Economic Literature, vol.39(3), pp.897–901. Shiller, R. (2000) Irrational Exuberance, Cambridge MA: MIT Press. Silvapulle, P., Moosa, I. and Sivapulle, M. (2004) ‘Asymmetry in Okun’s Law’, Canadian Journal of Economics, vol.37, pp.353–74. Silverstone, B. and Harris, R. (2001) ‘Testing for Asymmetry in Okun’s Law: A Cross-country Comparison’, Economics Bulletin, vol.5(2), pp.1–13. Sims, C.A. and Zha, T. (2006) ‘Were There Regime Switches in US Monetary Policy?’, American Economic Review, vol.96(1), pp.54–81. Slacalek, J. (2006) ‘What Drives Personal Consumption? The Role of Housing and Financial Wealth’, DIW Discussion paper 647. Søndergaard, L. (2003) ‘Using Instrumental Variables to Estimate the Share of Backward-Looking Firms’, Working Paper 2003-09 Georgetown University. Stenius, M. and Virén, M. (1984) ‘Some Further Results on Rosen and Quandt’s Labor Model’, European Economic Review, vol.26, pp.369–77. Stiglitz, J. (1984) ‘Price Rigidities and Market Structure’, American Economic Review, vol.74, pp.350–5. Stiglitz, J. (1997) ‘Reflections on the Natural Rate Hypothesis’, Journal of Economic Perspectives, vol.11 (winter), pp.3–10.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

230 References

Stock, J. and Watson, M. (2000) ‘Forecasting Output and Inflation: The Role of Asset Prices’, Sveriges Riksbank and Stockholm School of Economics conference on Asset Markets and Monetary Policy, Stockholm, June. Stock, J.H. and Watson, M.W. (1999) ‘Forecasting Inflation’, Journal of Monetary Economics, vol.44, pp.293–335. Tambakis, D.N. (1999) ‘Monetary Policy with a Nonlinear Phillips Curve and Asymmetric Loss’, Studies in Nonlinear Dynamics & Econometrics, Berkeley Electronic Press, vol.3(4). Taylor, J.B. (1993) ‘Discretion versus Policy Rules in Practice’, Carnegie-Rochester Conference Series on Public Policy, vol.39, pp.195–214. Taylor, J.B. (2009) Getting Off Track: How Government Actions Caused, Prolonged and Worsened the Financial Crisis, Stanford: Hoover Institution Press. Teräsvirta, T. and Eliasson, A. (1998) ‘Nonlinear Error-correction and the UK Demand for Broad Money, 1878–1993’, Stockholm School of Economics, Working Paper no. 265. Teräsvirta, T. and Granger, C.W.J. (1993) Modelling Nonlinear Economic Relationships, Oxford: Oxford University Press. Tong, H. (1983) Threshold Models in Nonlinear Time Series Analysis (Lecture Notes in Statistics), New York: Springer Verlag. Trichet, J.-C. (2005) ‘Asset Price Bubbles and Monetary Policy’, Bank of Spain, June 8. Verbrugge, R. (1998) ‘A Cross-country Investigation of Macroeconomic Asymmetries’, mimeo, Department of Economics, Virginia Polytechnic Institute and State University. Virén, M. (1993) ‘Interest Rates and Policy Reactions: Some International Evidence’, in Stokking, E.J. and Zambruno, G. (eds) Recent Research in Financial Modelling, pp. 75–90, Heildelberg: Physica Verlag. Virén, M. (2000a) ‘Fiscal Policy, Automatic Stabilisers and Policy Coordination in EMU’, in Brunila, A., Buti, M. and Franco, M. (eds) The Stability and Growth Pact, Basingstoke: Palgrave. Virén, M. (2000b) ‘How Sensitive is the Budget Balance to Cyclical Fluctuations in the EU?’, Helsinki: VATT (Government Institute for Economic Research). Walsh, C. (1995) ‘Optimal Contracts for Central Bankers’, American Economic Review, vol.85, pp.150–67. Walsh, C. (2001) ‘The Output Gap and Optimal Monetary Policy’, mimeo, University of California Santa Cruz. Waud, R.N. (1970) ‘Public Interpretation of Federal Reserve Discount Rate Changes: Evidence on the “Announcement Effect”’, Econometrica, vol.38, March, pp.231–50. Weber, C. (1995) ‘Cyclical Output, Cyclical Unemployment and Okun’s Coefficient: A New Approach’, Journal of Applied Econometrics, vol.10, pp.433–45. Woodford, M. (2001) ‘Optimal Monetary Policy Inertia’, NBER Working Paper 7261. Woodford, M. (2003) Interest and Prices: Foundations of a Theory of Monetary Policy, Princeton: Princeton University Press. Wynne, M. (2008) ‘How Should Central Banks Define Price Stability?’, in Mayes, D.G. and Wood, G.E. (eds) Designing Central Banks, London: Routledge. Yates, A. (1998) ‘Downward Rigidity and Monetary Policy’, Bank of England, Working Paper 114.

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

References 231

Index

backward-looking 44 balance sheet effect 188 bank credit 51 finance 196 Bank of Canada 45 England 196 banks 159 Barro rule 164 benchmark 17 BEPG see Broad Economic Policy Guidelines bias 112 borrowing constraints 72 Brainard-uncertainty 117 Broad Economic Policy Guidelines 165, 172, 204

bubbles 16, 187, 188, 202 budget balance 205–11 buffer funds xiii, 157, 206 business cycle 7, 79 Canada 63, 70, 82, 106, 135, 143, 166 capacity 64, 81 constraints 65, 94, 145, 186 capital values 187 capitalisation 187 Cardiff process 165 central bank 16, 31, 51 independence 153 Chicago Fed National Activity Index 199 China 135, 204 Cholesky decomposition 129 cointegrating relationship 137 collateral 187 prices 64 values 188 Cologne process 165 commodity price 55 common monetary policy 31 company finance 75 confidence 202 indicator 199 consumer prices 175, 198 coefficient of variation 128 consensus economics 88 forecasts 71, 85, 86, 87 consolidation 55, 189, 216, 217 construction 52, 64, 123, 126 consumer expenditure 57, 71–5 prices 175 consumption 1 deflator 15 function 68, 196 smoothing 74 continuity of future supply 65 convergence 8 convexity 127 232

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

active labour market policies 6, 159 activist monetary policy 189 adaptive expectations 119 model 132 aggregate supply 42–76 aggregation 29–41 aging 206 agriculture 52, 123, 126 Alan Greenspan see Greenspan ARCH/GARCH 19 area-wide model xii, 31 ARIMA model 33 asset management companies 56 price bubble 5, 176 prices 15, 16, 56–7, 58, 59, 61, 65, 67, 68, 69, 72, 75, 175, 187–8, 197, 198, 199, 202, 203 asymptotic periodic form 20 Australia 94, 106, 135, 143, 145, 161, 186 Austria 92, 108 automatic stabilisers 167, 177, 189, 206, 209, 210

coordinated fiscal expansion 168 coordination 158 corporate taxation 159 corridor 15, 22, 182, 185 model 181 specification 20 Council of Ministers 217 counter-cyclical fiscal policy 155 CPI 86, 87 credit channel 56, 79 easing 216 currency boards 204 cyclical fluctuations 205–11 cyclically adjusted deficit 208, 210 data generating process 33 mechanism 36 debt 209, 213, 216 deflation 4, 188, 202 spiral 56, 176 levels 135 ratio 165, 189, 205, 211, 212, 213 deficit 205, 208, 212, 213, 215 bias 216 procedure 166 deflation 174 demand 42–76 pressures 2, 86, 87 Denmark 180 depth 3 Deutsche Mark 50, 189 disequilibrium model 24, 26 disparities 186 dispersion 115, 128–30 disposable income 72 DNB-nor 159 dot.com boom 5 downturns 158 downward rigidity 95 duration 3 Dynamic Information Matrix test 33 Stochastic General Equilibrium 42 earning capacity ECB 31, 185 ECOFIN 172

Economic and Monetary Union see EMU economic cycles 3 economies of scale 151 EDP see Excessive Deficit Procedure EDGE 156 employability 151 employment 4, 118 EMU 6, 8, 55, 89, 108, 155, 165, 180, 189, 192, 193 endogenous variables 129 end point problem 78, 100, 106–7, 113, 114 equilibrium 64 equity 160 ERM 45, 50, 178, 184 crisis 182 error-correction 14, 20 mechanism 137 model 19 term 141 errors in variables 101 EU 138 Euler equation 11, 44 euro area xii, 2, 30, 31, 44, 45, 46, 50–6, 51, 59, 70, 72, 81, 82, 86, 92, 95, 106, 108, 145, 147, 160, 175, 180, 182, 189, 190, 196, 197, 198, 203, 204, 213, 218 EuroCOIN 199 European Central Bank 5 Monetary System 45, 50 Eurosystem 176, 185, 193, 194 Excessive Deficit 158, 204 Procedure 166, 206, 211–12 exchange rate xii, 46, 88 mechanism see ERM targeting 178 exclusion 153 exit and entry 63 expansions 29 expectations 13, 64, 65, 70, 97, 113 channel 91 formation 91, 107, 109

151 failing firms 159 FCI 45, 194, 195, 196

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Index 233

234 Index

GDP deflator 86, 87, 103, 105, 108, 110, 119 Generalised Method of Moments see GMM Least Squares 24 Germany 68, 82, 93, 143, 148, 166, 192, 193, 194, 195, 196, 197, 199, 203 global financial crisis xii, 1 GMM 12–13, 24, 44, 59, 60, 80, 101, 111, 132, 213 government bond yield 192 consumption 161

great moderation xii, 51, 83, 180 Greece 141, 156, 204, 217 Greenspan 5, 188 effect 66 standard 4, 176, 202 Gross Domestic Product 5 growth 56 rates 69 recession 55 Helsinki 131 heterogeneity 32 Hodrick–Prescott 10, 42 filter 17, 76, 182 home ownership 196 house prices 56, 57, 58, 61, 65, 67, 68, 69, 71, 72, 74, 75, 175, 187, 188, 189, 190, 191, 198, 199, 200, 202, 203 volatility 199 housing cycles 64 finance 57 wealth 68 HP filter 100, 103, 106 human capital 155 Hybrid Euler equation 43 New Keynesian model 132 model 79 hysteresis 127, 141, 154 identification 145 problem 180, 186 impaired assets 56 import prices 87, 88, 124 impulse response 44, 46, 130, 210 incentive structures 168, 217 infrastructure investment 160 inflation 4, 6, 54, 55, 56, 58, 59, 70, 79, 80, 81, 83, 86, 87, 88, 89, 91, 93, 94, 96, 97, 103, 109, 111, 116, 119, 124, 127, 128, 159, 174, 180, 184, 187, 190, 191, 199, 205, 212, 213, 216 bias 177 dynamics 82, 98, 113 expectations 78, 81, 89, 91, 98, 129

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Federal Reserve 5 Board 203 financial conditions index 12, 44, 193, 199 crisis 131 innovations 51 market liberalisation 51 markets 64, 74 penalties 217 services 126 stability 175, 199 Finland xii–xiii, 8, 50, 51, 52, 56, 62, 72, 92, 108, 130–3, 141, 157, 171, 172, 178, 189, 205, 206 firms 2 fiscal coordination 172 deficits 6 federalism 157, 166 policy 55, 158, 167, 168, 204 problems 41 responsibility 205 sector 16 stance 204, 211 fixed effects 75, 119 forecasts 199 error 44 forward-looking model 70–1 variables 43 France 161 FRB/US macroeconomic model 169 frictions 32 full capacity 3

Index 235

Japan 135, 141, 143, 174 job destruction 3 losses 4 reallocation 144 search 159 security 155 Kernel densities 36 knowledge-based skills

155

labour market 62, 115, 116, 134–59, 185 institutions 116, 155 productivity 134, 158 share of income 79 shortages 88 Laffer curve 153 latent states 26 layoffs 148 learning 98 curve 2 Lilian index 118

liquidity channel 188 constrained households 73, 188 constraints 74 Lisbon agenda 135 Strategy 152, 155, 165, 204 long-term solvency constraint 170 loss functions 177 lower bound 186 problem 95 Luxembourg 148, 211 process 165 Maastricht convergence criteria 213 criteria 6, 8, 55, 211 Treaty 211 macroeconomic policy 2 manufacturing 126 Markov switching 7 process 26 measurement errors 20, 44 mismatch 118, 186 mobility 152 model consistent expectations 97 momentum threshold autoregressive models 18 monetary conditions 44–50, 70, 191 index 45, 193 discipline 128 policy 5, 44, 51, 56–7, 88, 91, 93, 94, 95, 112, 116, 117, 124, 144, 145, 146, 153, 172, 174–203, 215, 217, 218 committee 196 targets 175 reaction function 15 union 54, 147, 205, 211–15 money market 193 targeting 178 Monte Carlo simulations 34 mortgage interest rates 198 Muelbauer-Hajivassiliou condition 25

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

rates 69 target 93, 175 targeting 172, 178, 189, 191, 199 inflationary bias 93, 95 pressure 88, 93 innovation 151 instruments 113 Instrumental Variables 12 interest payments 208 rates 172, 174, 190, 191, 209 forecast 193 smoothing 15, 178, 189 interpolation 99 intertemporal substitution 44 investment 188 goods 52 Ireland 8, 51, 52, 148, 166 irrational exuberance 5, 188 IS curve 10–12, 24, 42, 43, 144, 174, 176, 189 Italy 92, 93, 106, 141, 143, 148, 161, 166

NAIRU 121, 153, 156 National Action Plans 156 natural wastage 4 negative equity 73–4, 188 neural network 39 new economic geography 151 economy 146, 154 New Keynesian 83 hybrid 81 model 78, 82, 88 Phillips Curve 12, 79, 83, 89, 91 Zealand 113, 135, 141, 143, 180, 190, 211 NiGEM 168, 172 nominal rigidities 186 wage resistance hypothesis 94 Nordea 159 Nordbanken 159 Nordic crisis 150, 159 Norway 159, 161, 205 OECD forecasts 83, 88, 90, 98, 99, 129 Okun curve 4, 14, 16, 24, 70, 116, 134–59, 176, 182, 185, 186 omitted variables 79 one-sided filters 100, 107 open economy 42–76 method of coordination 173 openness 51 opportunistic approach 177 optimal government size 164 outliers 92 output 134–59, 187 gap 2, 7, 18, 60, 63, 70, 73, 76, 77, 78, 80, 81, 82, 86, 87, 88, 89, 93, 94, 95, 96, 97, 100, 102, 103, 104, 105, 106, 109, 111, 112, 113, 114, 119, 121, 122, 123, 127, 137, 141, 148, 174, 178, 180, 182, 184, 185, 190, 202, 212, 216 growth 54, 58, 60, 66, 148 overshooting 3

panel unit root test 76 participation 116 rate 152, 216 pensions xiii, 206 policies 147 persistence 73, 79, 83, 112 Phillips curve 2, 4, 12–14, 20, 24, 29, 30, 51, 65, 70, 77–114, 115, 116, 118, 119, 120, 122, 124, 126, 127, 131, 132, 133, 143, 144, 145, 174, 176, 177, 182, 185, 186 plucking model 3, 27, 29, 64 policy coordination 165–73 reaction function 178 rules 20, 116, 117 pooling 108 Portugal 148, 192, 197 positive shocks 62 potential output 3, 137 price expectations 77 level target 93 rigidity 97 stability 15, 117, 180, 197 stickiness 29 primary deficits 209 production function 100, 106 approach 104 method 107 productivity 2, 63, 154, 160 product markets 185 progressive tax system 213 public consumption 168, 169, 170, 171 employment 159, 163 expenditure 5, 160 rate 69 sector 158–73 consumption 160 services 126, 164 quantitative easing

216

rational expectations 11, 43, 77, 81, 83, 88, 97, 113, 191 hypothesis 12, 44, 82, 129 reaction function 185, 186, 187

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

236 Index

real exchange rate 44, 45, 47, 48, 50, 51, 52, 54, 57, 58, 65, 68, 193, 196, 197 interest rate 47, 48, 50, 51, 52, 54, 55, 57, 59, 65, 66, 72, 197, 202 marginal costs 78–9, 81, 82 rigidities 95 time data 78 estimation 106–14 information 97–104 unit labour costs 82 recession 1, 212 redistribution 166 rigidities 153 regime change 1, 26, 60 switch 63, 64 switching models 7, 26 regimes 20 regional dispersion 117–21 REH see rational expectations hypothesis relocation 151 replacement rate 152, 157 required rates of return 206 Reserve Bank of New Zealand 45 RESET test 33 residential property 56 reskilling 153 retraining 151–2, 159 revaluation effect 73 revisions 102 rigidities 116 robustness 15, 75, 80, 86 rolling regressions 182 rules 20 sacrifice ratio 116, 127 sanction 212, 217 Seemingly Unrelated Regression 24 sectoral dispersion 121–7 services 123 SGP see Stability and Growth Pact shake out 158 share prices 175 short-side rule 24 single monetary policy 46, 155

skew 55 skills 151 slack demand 119 smoothing 99, 184, 203 smooth transition model 68, 141 regression 13, 17, 65 model 22, 80 snake 178 social deprivation 161 exclusion 155, 161, 166 inclusion 152 Spain 92, 141, 178 specialisation 151 speed limit 178 speeds of adjustment 141 stabilisation 156, 165–73 Stability 217 and Growth Pact xiii, 6, 155–6, 158, 165, 166, 172, 204, 205, 211, 216, 217, 218 start up costs 159 state -space models 26 subsidies 156 states 26 stationarity 104 steepness 3 sterling 50 stock market prices 57 prices 58, 59, 61, 65, 67, 68, 69, 71, 72, 73, 187, 188, 189, 190, 191, 199, 200, 202, 203 STR see smooth transition regression structural breaks 46 deficit 208 funds 166 policies 128 sub-prime crisis 180 market 58 surplus 6, 205 survey-based expectations 98 surveys 113 sustainable fiscal system 167 SVAR 169

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Index 237

238 Index

target band 175 tax and benefit system 153 cuts 210 taxation 156, 189 Taylor principle 182 rule 15, 20, 24, 178, 179, 180, 181, 182, 183, 187, 189, 190, 192 term structure 51 threshold autoregressive 18 model 17, 20, 21, 23, 34, 36, 38, 39, 65, 119 variable 18 time consistency problem 217 Tobin’s Q 188 tolerance band 175 total factor 134 trade 126 union 156 density 153 transition variable 17, 18, 26 transitory shocks 62 transmission mechanism 51, 57, 68, 75, 79, 94, 97 transparency 191 Tsay test 33 UK

52, 57, 72, 88, 92, 106, 115, 118, 135, 141, 143, 147, 160, 174, 178, 194, 195, 196, 206

unbiasedness 105 uncertainty 91 unemployment xiii, 4, 54, 77, 78, 80, 83, 86, 87, 88, 95, 115, 116, 117, 118, 121, 123, 132, 133, 134–59, 159, 161, 186, 206, 216 benefits 209 dispersion 182 gap 85, 88, 122, 161 United States 5, 55–6, 82, 106, 113, 115, 134–5, 137–8, 143, 145, 166, 174–6, 178, 180, 198, 202 US dollar 50 vacancies 159 variance of unemployment 119, 150 VAR 168, 209 model 128, 129, 130 VAT 124 volatility 198

118,

wage inflation 115 resistance 145 stickiness 14, 29 Wald test 23, 35 wealth 72, 73, 75 effect 71 welfare capitalism 148 regimes 159 wellbeing 153 West Germany 135 zero bound problems

5, 188 174

10.1057/9780230304642 - Asymmetry and Aggregation in the EU, David G. Mayes and Matti Virén

Copyright material from www.palgraveconnect.com - licensed to ETH Zuerich - PalgraveConnect - 2011-04-21

Sweden 51, 94, 145, 159, 178, 186, 206 Swedish krona 50 switching regression model 21 Switzerland 141