Real long-term interest rates and monetary policy: a cross-country perspective



Similar documents
4. International Parity Conditions

The Greek financial crisis: growing imbalances and sovereign spreads. Heather D. Gibson, Stephan G. Hall and George S. Tavlas

Chapter 8: Regression with Lagged Explanatory Variables

BALANCE OF PAYMENTS. First quarter Balance of payments

Vector Autoregressions (VARs): Operational Perspectives

MACROECONOMIC FORECASTS AT THE MOF A LOOK INTO THE REAR VIEW MIRROR

11/6/2013. Chapter 14: Dynamic AD-AS. Introduction. Introduction. Keeping track of time. The model s elements

The Real Business Cycle paradigm. The RBC model emphasizes supply (technology) disturbances as the main source of

How To Calculate Price Elasiciy Per Capia Per Capi

Duration and Convexity ( ) 20 = Bond B has a maturity of 5 years and also has a required rate of return of 10%. Its price is $613.

Cointegration: The Engle and Granger approach

Usefulness of the Forward Curve in Forecasting Oil Prices

II.1. Debt reduction and fiscal multipliers. dbt da dpbal da dg. bal

Measuring macroeconomic volatility Applications to export revenue data,

Estimating Time-Varying Equity Risk Premium The Japanese Stock Market

PROFIT TEST MODELLING IN LIFE ASSURANCE USING SPREADSHEETS PART ONE

Morningstar Investor Return

Why Did the Demand for Cash Decrease Recently in Korea?

Supplementary Appendix for Depression Babies: Do Macroeconomic Experiences Affect Risk-Taking?

Relationships between Stock Prices and Accounting Information: A Review of the Residual Income and Ohlson Models. Scott Pirie* and Malcolm Smith**

A Note on Using the Svensson procedure to estimate the risk free rate in corporate valuation

CRISES AND THE FLEXIBLE PRICE MONETARY MODEL. Sarantis Kalyvitis

The Grantor Retained Annuity Trust (GRAT)

Chapter 8 Student Lecture Notes 8-1

Determinants of Capital Structure: Comparison of Empirical Evidence from the Use of Different Estimators

Chapter 6: Business Valuation (Income Approach)

Chapter 9 Bond Prices and Yield

SPEC model selection algorithm for ARCH models: an options pricing evaluation framework

Option Put-Call Parity Relations When the Underlying Security Pays Dividends

Risk Modelling of Collateralised Lending

Can Individual Investors Use Technical Trading Rules to Beat the Asian Markets?

Appendix D Flexibility Factor/Margin of Choice Desktop Research

The impact of Federal Reserve asset purchase programmes: another twist 1

DOES TRADING VOLUME INFLUENCE GARCH EFFECTS? SOME EVIDENCE FROM THE GREEK MARKET WITH SPECIAL REFERENCE TO BANKING SECTOR

Individual Health Insurance April 30, 2008 Pages

Principal components of stock market dynamics. Methodology and applications in brief (to be updated ) Andrei Bouzaev, bouzaev@ya.

Working paper No.3 Cyclically adjusting the public finances

Migration, Spillovers, and Trade Diversion: The Impact of Internationalization on Domestic Stock Market Activity

Do Credit Rating Agencies Add Value? Evidence from the Sovereign Rating Business Institutions

Impact of scripless trading on business practices of Sub-brokers.

Term Structure of Prices of Asian Options

Lecture Note on the Real Exchange Rate

Market Liquidity and the Impacts of the Computerized Trading System: Evidence from the Stock Exchange of Thailand

The Interest Rate Risk of Mortgage Loan Portfolio of Banks

BALANCE OF PAYMENTS AND FINANCIAL MA REPORT All officiell statistik finns på: Statistikservice: tfn

Why does the correlation between stock and bond returns vary over time?

Monetary Policy & Real Estate Investment Trusts *

Does Stock Price Synchronicity Represent Firm-Specific Information? The International Evidence

The yield curve, and spot and forward interest rates Moorad Choudhry

SURVEYING THE RELATIONSHIP BETWEEN STOCK MARKET MAKER AND LIQUIDITY IN TEHRAN STOCK EXCHANGE COMPANIES

ARCH Proceedings

Small and Large Trades Around Earnings Announcements: Does Trading Behavior Explain Post-Earnings-Announcement Drift?

Economics Honors Exam 2008 Solutions Question 5

Journal Of Business & Economics Research September 2005 Volume 3, Number 9

CURRENT ACCOUNTS IN THE EURO AREA: AN INTERTEMPORAL APPROACH. José Manuel Campa and Ángel Gavilán an. Documentos de Trabajo N.

Chapter 1.6 Financial Management

Explaining the NZ-Australian exchange rate occasional paper

Determinants of Bank Long-term Lending Behavior in the Central African Economic and Monetary Community (CEMAC)

Does Option Trading Have a Pervasive Impact on Underlying Stock Prices? *

USE OF EDUCATION TECHNOLOGY IN ENGLISH CLASSES

Hedging with Forwards and Futures

I. Basic Concepts (Ch. 1-4)

INTRODUCTION TO FORECASTING

CURRENT ACCOUNTS IN THE EURO AREA: AN INTERTEMPORAL APPROACH. José Manuel Campa Angel Gavilán

Real Return Bonds, Inflation Expectations, and the Break-Even Inflation Rate Ian Christensen, Frédéric Dion, and Christopher Reid

Does informed trading occur in the options market? Some revealing clues

The Identification of the Response of Interest Rates to Monetary Policy Actions Using Market-Based Measures of Monetary Policy Shocks

How does working capital management affect SMEs profitability? This paper analyzes the relation between working capital management and profitability

Swiss National Bank Working Papers

Consumer sentiment is arguably the

Bid-ask Spread and Order Size in the Foreign Exchange Market: An Empirical Investigation

Evidence from the Stock Market

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C.

Expecaion Heerogeneiy in Japanese Sock Index

When Is Growth Pro-Poor? Evidence from a Panel of Countries

Internal and External Factors for Credit Growth in Macao

Investor sentiment of lottery stock evidence from the Taiwan stock market

LECTURE: SOCIAL SECURITY HILARY HOYNES UC DAVIS EC230 OUTLINE OF LECTURE:

Present Value Methodology

INTERNATIONAL REAL ESTATE REVIEW 2003 Vol. 6 No. 1: pp Banking System, Real Estate Markets, and Nonperforming Loans

Debt Accumulation, Debt Reduction, and Debt Spillovers in Canada, *

WORKING PAPER SERIES STOCKS, BONDS, MONEY MARKETS AND EXCHANGE RATES MEASURING INTERNATIONAL FINANCIAL TRANSMISSION NO.

Florida State University Libraries

CHARGE AND DISCHARGE OF A CAPACITOR

LEASING VERSUSBUYING

Time Series Analysis Using SAS R Part I The Augmented Dickey-Fuller (ADF) Test

Estimating the Term Structure with Macro Dynamics in a Small Open Economy

Does Option Trading Have a Pervasive Impact on Underlying Stock Prices? *

New facts in finance. John H. Cochrane

MEDDELANDEN FRÅN SVENSKA HANDELSHÖGSKOLAN SWEDISH SCHOOL OF ECONOMICS AND BUSINESS ADMINISTRATION WORKING PAPERS

The Kinetics of the Stock Markets

Measuring the Effects of Exchange Rate Changes on Investment. in Australian Manufacturing Industry

Long-Run Stock Returns: Participating in the Real Economy

Table of contents Chapter 1 Interest rates and factors Chapter 2 Level annuities Chapter 3 Varying annuities

One dictionary: Native language - English/English - native language or English - English

The impact of the trading systems development on bid-ask spreads

Title: Who Influences Latin American Stock Market Returns? China versus USA

Market Analysis and Models of Investment. Product Development and Whole Life Cycle Costing

The effects of stock market movements on consumption and investment: does the shock matter?

CALCULATION OF OMX TALLINN

Transcription:

Real long-erm ineres raes and moneary policy: a cross-counry perspecive Chrisian Upper and Andreas Worms, 1 Deusche Bundesbank 1. Inroducion The real rae of ineres is a cenral concep in economics. I represens he price of he ineremporal allocaion of goods and hereby deermines saving, invesmen and, ulimaely, economic growh. Despie he imporance of he real rae for our undersanding of ineremporal choice, operaionalising he concep urns ou o be difficul. Firs of all, in a world of nominal conracs, real raes usually canno be observed direcly. This problem is paricularly imporan in he case of long-erm real raes in which long-erm inflaionary expecaions come ino play. Second, i is no clear o which exen moneary policy can affec real ineres raes. Alhough he cenral bank ses nominal ineres raes, by doing so i will also affec shor-erm real raes as long as rigidiies preven prices from adjusing immediaely. I is much less clear wheher moneary policy has much of an effec on medium- o long-erm real ineres raes. Third, even if we did have informaion on he level of real ineres raes and if he cenral bank could affec i, we sill need a benchmark in order o assess he curren level of he real rae, for example o deermine he sance of moneary policy. The ques for such benchmarks acually predaes he definiion of he real rae by Fisher (1930) and goes back o he work on he naural rae of ineres by Wicksell (1898). 2 More recen conribuions are Woodford (2002) and Neiss and Nelson (2001). This paper does no ackle he benchmark issue explicily, bu raher concenraes on he firs wo quesions, he answers o which can be seen as a prerequisie for an adequae analysis of he moneary policy/naural rae issue. Given hese wo problems, i comes as lile surprise ha policyoriened research on real ineres raes has mosly concenraed on he shor end of he erm srucure. An example is he lieraure on ineres rae feedback rules, mos prominenly associaed wih he work of John Taylor. In a Taylor rule, he (nominal) policy rae is se equal o a sor of naural real rae of ineres plus arge inflaion as well as wo erms proporional o he deviaions from arge of oupu and inflaion, respecively. Anoher example is he lieraure on inflaion forecass, where deviaions of he real rae of ineres from a benchmark are used o forecas fuure price developmens. Such real ineres gaps have been analysed, for example, in Neiss and Nelson (2001). This concenraion on he shor end of he erm srucure is a odds wih he fac ha boh saving and invesmen decisions end o be inherenly medium- o long-erm. I seems unlikely ha policy-induced flucuaions in shor-erm real ineres raes will have much impac on he real economy unless hey also affec long-erm real raes. I is his link beween shor-erm real raes as a proxy for moneary policy and long-erm real ineres raes ha is he focus of our paper. In paricular, we ask wheher such a relaion exiss, and - if so - wheher i has been sable over ime. The laer quesion is moivaed by he fac ha here may have been changes due o European moneary union (EMU) and/or financial globalisaion. Before proceeding o he esimaion par of he paper, we dwell on he measuremen problem relaed o he unobservabiliy of real ineres raes or inflaion expecaions. We argue ha his issue can be 1 2 We are graeful o Chrisian Dembiermon of Daa Bank Services a he BIS for racing oherwise unavailable ineres rae series. Par of he work on his paper was underaken when he firs auhor visied he Federal Reserve Bank of Kansas Ciy, whose hospialiy is graefully acknowledged. The opinions expressed are hose of he auhors and do no necessarily reflec he views of he Deusche Bundesbank. Wicksell s concep does no clearly disinguish beween nominal and real raes. His naural rae of ineres represens he reurn on fixed asses in an environmen wih no inflaion. Any divergence of he acual rae of ineres (which roughly corresponds o a nominal marke rae) from he naural rae will resul in an adjusmen of he price level. The realisaion ha here could be various naural ineres raes for differen growh pahs and employmen levels caused Keynes (1936) o develop he concep of he neural ineres rae, which presupposes full employmen. 234 BIS Papers No 19

addressed by using survey daa on inflaion expecaions. We consruc series of five- and 10-year ex ane real ineres raes for 10 indusrialised counries based on price expecaions published by Consensus Economics. Alhough survey daa have imporan drawbacks, he available evidence makes us believe ha he Consensus is no a bad measure for inflaion expecaions, especially when considering he alernaives. We hen esimae he effec of moneary and fiscal policy on long-erm real ineres raes. Moneary policy is proxied by he shor-erm real rae, and fiscal policy by governmen ne borrowing and public deb. The panel economeric echniques applied exploi boh he ime dimension and he cross-counry variaion of our daa se. Specifically, his sraegy allows us o perfecly conrol for world facors, which in his seing can be inerpreed as a pure ime effec. This, in urn, enables us o analyse wheher a change in a counry s moneary or fiscal policy relaive o he res of he world influences he deviaion of is long-erm real ineres rae from he world level. We find ha counry-specific moneary policy is an imporan facor deermining real raes, alhough he evidence suggess ha i has become less imporan since he lae 1990s. This is no purely relaed o EMU bu applies equally o he non-emu counries of our sample. Insead, i seems ha moneary policy has generally become more synchronised across counries, leaving less room for differences in naional long-erm real raes. The second imporan deerminan of long-erm real ineres raes is fiscal policy. In general, high deb or governmen borrowing end o be associaed wih high long-erm real ineres raes, alhough he recen Japanese experience of low real raes and very high deb provides a couner-example. In conras o he case of moneary policy, we do no find evidence ha counry-specific fiscal policy has become less effecive over ime. Our work exends he exising lieraure especially in wo dimensions: one is he use of inflaion forecass o consruc ex ane medium- and long-erm real raes. The oher is he use of panel daa mehods ha allow us o concenrae on cross-counry differences while perfecly conrolling for world facors. Neverheless, our mehodology is no wihou is own problems. The small size of our sample (a mos 10 counries and 25 semiannual observaions) makes i impossible o use sophisicaed economeric mehods ha permi us o beer idenify causaliy. As a consequence, reverse causaion canno be ruled ou, alhough we argue ha he problem is unlikely o be so large as o render our resuls useless. Moreover, he small size of our daa se prevens he esimaion of richer economeric models, such as dynamic panel models. As so ofen in empirical work, he alernaive is o live wih he shorcomings or no o use he daa a all. Our work should herefore raher be inerpreed as a complemen o he exising lieraure, no as a subsiue for i. The paper is srucured as follows. Secion 2 discusses he impac of moneary and fiscal policy on long-erm real ineres raes and reviews he relevan lieraure. Secion 3 addresses measuremen issues, followed by a descripive secion on developmens in long-erm ex ane real raes since he early 1990s. The esimaions ha form he core of he paper are presened in Secion 5. A final secion summarises our key resuls and concludes. 2. Macroeconomic policy and long-erm real ineres raes 2.1 Moneary policy Mos recen work in macroeconomics has been based on he undersanding ha prices are sicky in he shor run bu adjus over longer ime horizons. While here may be disagreemen on how long i akes for prices o adjus, i is probably safe o say ha hey end o be raher sicky for periods as shor as a quarer bu fairly flexible for horizons of several years. In such a world, a change in shorerm nominal ineres raes induced by moneary policy will have a large effec on shor-erm real raes. The impac on longer-erm real raes is likely o be much smaller. Take he simple example of an economy in which prices are se in advance for one period bu fully adjus aferwards. In equilibrium, he real rae of ineres is r, which is assumed o be consan. Now suppose ha in period he cenral bank wans o simulae he economy by cuing he one-period nominal rae by. Since prices canno adjus wihin he same period, he change in he nominal rae 1 brings abou a decline in he one-period real rae by an equal amoun o r = r (he superscrip BIS Papers No 19 235

indicaes ime o mauriy). In + 1, prices will adjus and compleely offse he moneary simulus. As a consequence, he expeced one-period real rae is Er 1 + 1 = r. According o he expecaions hypohesis of he erm srucure of ineres raes, long-erm raes should be equal o he expeced average shor raes over he same mauriy. This should hold for boh real and nominal ineres raes due o a noarbirage condiion. 3 N For example, he N-period real rae a, r, is relaed o he sequence of oneperiod raes by: N N 1 1 1 ( r ) = E( 1+ r )( 1+ r )...( 1 r ) 1+ + 1 + + N 1 Taking logarihms and rearranging yields: r N N 1 1 1 r + j 0 = E N j = 1 = N ( Nr ) = r N Thus he effec of moneary policy on longer-erm real raes declines wih he mauriy for which he raes apply. 4 In more realisic models, he decline may no be linear. In general erms: l s ( r r ) r = f, (1) The long-erm real rae r l is a funcion of he shor-erm real rae r s and some naural real rae r, which may vary over ime. Since r canno be observed, we rewrie equaion (1) as: l s ( r X ) r = f *, (1 ) where X is a vecor conaining he deerminans of r. We can isolae he effec of moneary policy by decomposing he shor real rae ino he naural rae and a erm indicaing he ighness of moneary gap s policy r = r r, which corresponds o he real ineres gap of Neiss and Nelson (2001). More formally: l gap r = * f * * r, X (1 ) gap Unforunaely, r canno be observed bu requires an esimae of r. Moreover, he oupu gap and he parameers of f** would have o be esimaed simulaneously in order o avoid inconsisencies. This can only be achieved wihin a well srucured model of real ineres raes. Economic heory does no provide much guidance in his respec, as he various heoreical models end o give conflicing resuls. 5 As a consequence, empirical models of real ineres raes have essenially been ad hoc. They end o include a number of possible facors ha shif he supply of and he demand for loanable funds and hence he equilibrium real ineres rae. We herefore prefer o esimae equaion (1 ) raher han (1 ). 2.2 Fiscal policy Public saving and dissaving ypically accoun for a subsanial proporion of oal saving in developed economies. I hus seems ha he scope for fiscal policy o affec long-erm real ineres raes can be much larger han ha for moneary policy. For example, he rise in real ineres raes in he early 1980s 3 4 5 Coss of arbirage would inroduce erm premia, which may vary over ime. This should no affec he validiy of he argumen unless he erm premium covaries wih he level of ineres raes. The resul ha he long raes move in he same direcion as he shor rae, only by a smaller amoun, does no necessarily exend o nominal raes. In ha case, long and shor raes may even move in differen direcions if he policy acion ha drives he shor rae affecs inflaion expecaions by a sufficienly large amoun. See Romer and Romer (2000) and Ellingsen and Södersröm (2001). For a review he reader is referred o Bliss (1999). See also he appendix in Deusche Bundesbank (2001). 236 BIS Papers No 19

has been aribued o he rise in governmen deficis in ha period. 6 This argumen, of course, hinges on he assumpion ha Ricardian equivalence does no hold. If, on he conrary, agens believe ha hey will have o pay for oday s high defici hrough higher axes in he fuure, hey may cu curren consumpion and hus offse any fiscal simulus. In ha case, fiscal policy would have no effec on real ineres raes. 2.3 Counry-specific and global variables A large number of papers, including Blanchard and Summers (1984), argue ha ineres raes are subsanially deermined worldwide, raher han domesically, because a large pool of capial flows owards naions wih high real raes ends o equalise raes around he world. They sress ha his seems o be rue no only for small open economies, bu also for large economies like he Unied Saes. This corresponds o he resuls presened in Barro and Sala-i-Marin (1990), who find for 10 OECD counries ha heir respecive expeced real ineres rae depends primarily on world facors, raher han on own counry facors. This is in line wih he more recen findings of Al Awad and Goodwin (1998) who - on he basis of coinegraion and Granger causaliy echniques - find a high degree of inegraion of inernaional asse markes, which implies a srong coinegraion among G10 ex ane real ineres raes. Neverheless, hey also find an imporan role for ransacion coss, which preven real ineres rae equalisaion across counries. Wu (1999) - by applying coinegraion echniques o German and Japanese real ineres rae and exchange rae daa - finds evidence in favour of a long-run relaionship beween real exchange rae and expeced real ineres differenials only if he curren accouns are explicily considered. The hypohesis of srong cross-counry linkages is also confirmed by Pain and Thomas (1997). Applying coinegraion echniques o daa from several indusrial counries, hey find evidence for a European shor-erm real ineres rae, wih Germany being he dominan player. Bu his resul does no seem o be robus wih respec o he inclusion of US raes, indicaing ha US raes deermine he rend in European raes. Ineresingly, hey also find evidence ha he degree of inegraion has increased over ime. This is in line wih he resuls presened in Founas and Wu (1999), who find evidence in favour of bilaeral real ineres rae convergence beween Germany and several oher counries for long-erm real ineres raes. They aach his o he growing degree of inegraion in he world financial markes. In a more recen paper hey apply a comparable coinegraion echnique o es for bilaeral real ineres rae convergence in he G7 agains he Unied Saes (Founas and Wu (2000)). They find srong evidence for bilaeral real shor-erm ineres rae convergence o a long-run relaionship beween US raes and raes in Canada, France, Germany and he Unied Kingdom. Moreover, hey find evidence in favour of a bilaeral real long-erm ineres rae convergence o a long-run relaionship beween US raes and raes in France and Germany. This means ha for France and Germany, long-erm real ineres rae changes are influenced by he US moneary policy sance. Given he srong resuls poining a a close inerrelaionship beween naional real raes, some papers esimae a world ineres rae raher han looking a naional ineres raes. Kraemer (1996) aggregaes naional daa of he G7 counries and esimaes a single-equaion error correcion model. He finds ha he resuling aggregae long-erm real ineres rae is mainly deermined by he real shor-erm ineres rae, capaciy uilisaion and srucural public borrowing. Orr e al (1995) pool daa from 17 counries o esimae an error correcion model. They find ha he low-frequency componen of he real ineres rae is mainly deermined by profiabiliy, a risk measure, he curren accoun, he governmen defici and a measure for surprise inflaion. The high-frequency componen is influenced principally by moneary and fiscal policy. They inerpre he low-frequency facors as he fundamenals ha influence savings and invesmen rends, whereas he high-frequency facors change he expecaions abou he fundamenal facors. On he basis of he idea of a world real ineres rae, Ford and Laxon (1999) analyse he role of global fiscal developmens. They find ha he increase in OECD-wide governmen deb since he lae 1970s was a major facor in he rise in real ineres raes. 6 Refer o Blanchard and Summers (1984) for a criique of his argumen and a discussion on he effecs of fiscal policy. They argue ha he aggregae inflaion-adjused srucural defici for he major naions of he OECD did no change significanly beween 1978 and 1984. Insead, hey claim, expeced profiabiliy has risen, as indicaed by an increase in sock prices. This drove up he demand for loanable funds and hereby he real rae of ineres. BIS Papers No 19 237

Conrasing hese sudies, Breedon e al (1999) find ha i is hard o argue ha naional real ineres raes converge o a single world rae, alhough inernaional facors are imporan. They also find ha he large and persisen differences in real ineres raes across counries canno be explained in erms of real exchange rae expecaions. Transacion coss and counry-specific facors, for example counry-specific risk and he porfolio home bias, sill seem o play a significan role. We, insead, are ineresed in hese world facors only in so far as we wan o assess which counryspecific facors deermine a counry s real ineres rae. In conras o he previous lieraure ha concenraed on ime series mehods, we use panel economeric echniques because hey allow us o conrol for all possible world facors. By inroducing a complee se of ime dummies in a fixed-effecs regression, all facors ha do no vary across counries (bu over ime) are filered ou as long as hey influence he counries real raes in a similar manner. Basically, his amouns o explaining differences in movemens of real raes across counries by cross-counry differences in movemens of he righhand variables. World facors, which by definiion do no differ across counries, are compleely capured by hese ime dummies. If he counry-specific facors urn ou o be insignifican, hen a counry s real rae is purely deermined by such world facors. If, on he conrary, counry-specific facors urn ou o be significan, hen domesic variables play a role for domesic real raes - which is imporan especially in he case of moneary and fiscal policy. I would hen be of ineres o find ou wheher he srengh of hese influences has changed over ime. 3. Measuring real ineres raes and daa Measuring real ineres raes is far from rivial. Ideally, one would like o derive hem from marke prices, for example from inflaion-indexed bonds or loan conracs. Such insrumens have been fairly common in environmens wih high inflaion and have recenly been inroduced in a number of lowinflaion counries. However, among he indusrialised counries only he Unied Kingdom provides series ha span more han a few years. 7 In he absence of inflaion-indexed securiies, real ineres raes can be compued by deflaing a nominal rae of ineres by a measure of expeced inflaion using he Fisher pariy: r e = i π (2) Here, r sands for he real ineres rae and i for he nominal ineres rae wih he same mauriy; π e represens he expeced inflaion rae for he period in quesion. 8 One has o bear in mind, hough, ha his simple Fisher pariy is based on a number of resricive assumpions. For insance, ax aspecs are omied alhough, in pracice, heir role is no negligible. In addiion, i assumes ha invesors are indifferen as o wheher heir invesmen is nominal or real, as long as he yield differenial is in line wih expeced inflaion. This is in conras wih recen work in finance which suggess ha an economically significan inflaion risk premium exiss and varies over ime. 9 Unforunaely, such risk premia are unobservable, and can herefore no be eliminaed when compuing real raes. As a consequence, any measure of real ineres raes compued from nominal raes and expeced inflaion will necessarily be pollued. Unforunaely, deflaing nominal ineres raes by expeced inflaion only ransfers he daa problem o anoher level as expecaions are inherenly unobservable. One possibiliy is o ask marke paricipans direcly. Such survey daa are available from Consensus Economics, who ask a number of professional forecasers based in a variey of counries abou heir expecaions of a wide range of economic variables. More specifically, we use heir long-erm forecass, which have been published biannually in April and Ocober since he auumn of 1989. They conain forecass for - iner alia - real 7 8 9 See Deacon and Derry (1998). Sricly speaking, he formula above only provides an approximaion. The exac form of he Fisher pariy is e (1 + i) = (1 + r ) (1 + π ). We use his correc formula for our compuaions. Buraschi and Jilsov (2002) find ha he inflaion risk premium conained in he yield of 10-year US reasuries is highly imevarying, flucuaing beween 0.2 and 1.6 %. 238 BIS Papers No 19

GDP growh and inflaion in he curren and each subsequen calendar year up o 10 years in he fuure. As is always he case wih surveys, one may doub (i) wheher he surveyed insiuions or individuals accuraely sae heir views, (ii) wheher hey pu enough effor ino heir answers o make hem meaningful, and (iii) wheher hese views reflec he inflaion expecaions of economic decisionmakers. Le us deal wih hese hree argumens in urn. (i) We can hink of wo disinc moives for respondens o deviae from heir rue beliefs. Firsly, hey may no wan o reveal heir informaion in order o secure rading profis. This moive seems paricularly relevan for predicions over he very shor erm on which rading profis can be made. I appears less perinen for he longer-erm forecass, which have a much bigger weigh in our esimaions. In addiion, he ime lag wih which he Consensus is published (abou one week) is relaively long by financial marke sandards, giving paneliss ample ime o rade on heir predicions. Anoher reason for missaing one s beliefs may be he generaion of publiciy. In he model of Laser e al (1999), for example, forecasers wages depend boh on he accuracy of he forecas and on he publiciy hey generae for heir employer. Since publiciy ends o be good publiciy or no publiciy (good performance is made public while negaive performance usually is no), forecasers have incenives o bias heir forecass in order o sand ou of he crowd if heir predicions urn ou o be rue. Such incenives are paricularly srong for independen forecasers serving occasional users of forecass, bu less so for hose working direcly for regular users such as banks or indusrial corporaions. This is borne ou by he daa. Laser e al find ha independen forecasers deviae from he Consensus by a far greaer exen han heir peers a banks or in indusry. Securiy firms come somewhere in beween. This sor of behaviour does no affec he mean forecas, alhough i drives up he variance. Missaemen cancels ou over ime (for each individual forecaser) or across firms. We hus conclude ha he Consensus Forecas does appear o give a fairly accurae picure of he average view of he respondens. (ii) We have argued ha i is unlikely ha he paneliss of he Consensus missae heir views in order o obain rading profis. Bu wha if here are no profis o be made from rading on heir forecass in he firs place, simply because hey are lile more han informed guesses? This poin seems paricularly perinen for long-erm predicions, which are less easily exploied by rading. Unforunaely, we canno rule ou his possibiliy. Neverheless, he fac ha he Consensus seems o be superior o mos oher forecass, including hose from inernaional public insiuions where forecasing over he longer erm is a main par of heir business, suggess ha i is no so bad afer all and gives reason for mued opimism. (iii) Le us urn o he hird quesion of wheher he Consensus correcly reflecs he view of marke paricipans. Unforunaely, here is lile we can say abou his save ha many of he analyss surveyed are employed by more or less presigious insiuions, and ha heir forecass receive a lo of press coverage. They should herefore be known o decisionmakers. Wheher or no hese agree wih heir conen is difficul o assess. Summing up, he available evidence makes us believe ha he Consensus is no a bad measure for inflaion expecaions, especially when considering he alernaives. Model-based esimaes can be criicised on he basis ha i is even less clear wheher hey have anyhing a all o do wih acual expecaions. Deusche Bundesbank (2001), for example, shows ha ARMA forecass for inflaion can differ considerably from he Consensus, especially during periods of srong inflaion or disinflaion. Furhermore, alhough ime series models may have a good forecasing power a few quarers ahead, he longer-erm predicions end o correspond o he uncondiional mean of he sample over which hey are esimaed. This may no be very plausible if he mean has been driven by excepional facors one does no expec o be repeaed. Examples for his are German reunificaion or oil price changes. Irrespecive of wheher inflaion expecaions are derived from surveys or wheher hey are esimaed, he mauriy and ype of he underlying nominal ineres rae as well as he choice of he price index are relevan issues. Analyses of real ineres raes are only useful if he ime horizon for such raes covers he whole invesmen period. In he case of capial invesmens his is usually several years, for savings decisions i migh even amoun o decades (eg for reiremen saving). Real ineres raes for shorer mauriies are no very informaive in his respec, unless assumpions are made abou price and ineres rae movemens in subsequen periods. For our purpose, he yields of nominal governmen bonds wih a residual mauriy of five and 10 years appear suiable. Such securiies have BIS Papers No 19 239

been available in mos developed economies since he 1970s or 1980s. Since governmen bonds end o be he leas risky asse, heir yields should roughly correspond o he opporuniy coss of invesmen, even if mos privae agens would have o pay higher raes if hey ook up deb. As well as he mauriy, he choice of he price index depends on he issue being analysed precisely. Moreover, daa availabiliy is crucial. We use he consumer price index (CPI), which is available for all he considered counries and provides a fair approximaion of he overall price level. Graph 1 Ex ane real ineres raes 10 Canada 10 France 10 Germany 8 8 8 6 6 6 4 4 4 2 2 2 0 0 0 1990 1992 1994 1996 1998 2000 1990 1992 1994 1996 1998 2000 1990 1992 1994 1996 1998 2000 10 Ialy 10 Japan 10 Norway 8 8 8 6 6 6 4 4 4 2 2 2 0 0 0 1990 1992 1994 1996 1998 2000 1990 1992 1994 1996 1998 2000 1990 1992 1994 1996 1998 2000 10 Spain 10 Sweden 10 Unied Kingdom 8 8 8 6 6 6 4 4 4 2 2 2 0 0 0 1990 1992 1994 1996 1998 2000 1990 1992 1994 1996 1998 2000 1990 1992 1994 1996 1998 2000 10 Unied Saes 8 6 4 2 0 1990 1992 1994 1996 1998 2000 10 year real rae 5 year real rae 240 BIS Papers No 19

4. Long-erm ex ane real ineres raes since he 1990s Graph 1 shows five-year and 10-year ex ane real raes for 10 indusrialised counries. The sample period begins in 1989, when Consensus Forecass for CPI inflaion are firs available for Canada, France, Germany, Ialy, Japan, he Unied Kingdom and he Unied Saes. In 1995, Spain and Sweden are added o he lis, and in 1998 Norway. A second daa consrain is he availabiliy of nominal ineres raes of he desired mauriy. This is paricularly relevan in he case of Japan, which did no inroduce five-year bonds before he lae 1990s. As can be seen from Graph 1, ex ane real ineres raes in indusrialised counries declined during mos of he 1990s and have remained in a relaively narrow corridor ranging from jus under 2% o jus over 4%. The mean five-year ex ane real rae halved from 6.5% in early 1990 o 3.2% in lae 1999, and he developmen of 10-year real raes has been similar. The decline in real ineres raes was briefly inerruped during 1993/94, bu hey soon revered o heir downward rend, which coninued unil 1998. They increased somewha during he following years, bu eased again in 2001. The only, alhough imporan, excepion o his overall picure is he Unied Saes. US real raes were much lower han hose in oher counries during he firs half of he 1990s, alhough hey oo were declining. Beween 1994 and 1997, however, US raes remained relaively consan beween 3 and 4%, and declined only in 1998 in line wih hose of he oher counries of our sample. The 1990s also saw a remarkable convergence of he real ineres raes in all counries bar Japan. The difference beween he minimum and he maximum rae a a single poin in ime declined from 4-5 percenage poins in he early 1990s o 1-2 percenage poins owards he end of he decade. The main facor behind his convergence was he sharp decline in he real ineres raes of Ialy, Spain and Sweden. During he early 1990s, Ialy had by far he highes raes in our sample. 10 In 1992, five-year ex ane real ineres raes were 4 percenage poins higher han hose in France and 5 percenage poins higher han he corresponding German raes. In 1999, he difference declined o a few basis poins. Ex ane raes for Spain and Sweden are available only as from 1995, so we canno say much abou he firs half of he 1990s. Bu we see real ineres rae convergence even during he sample ha is available. In early 1995, Spanish real ineres raes sood a 7.3%, compared o 3.7% in Germany. Four years laer in 1999, Spanish ex ane raes were half a percenage poin lower han German raes. Wha is ineresing is ha Swedish raes showed a similar decline, alhough Sweden is no, and a ha ime was no expeced o be, a member of EMU. This suggess ha alhough moneary union may have been a facor in explaining real rae convergence, i was no he only one. 5. Empirical resuls 5.1 Explanaory variables and esimaion mehodology As we have menioned in Secion 2, economic heory provides lile guidance as o wha a model for real ineres raes should look like. Our approach is herefore raher ad hoc. I relaes he level of fiveand 10-year ex ane real raes (r05cf and r10cf, respecively) o a moneary policy variable, fiscal variables and a se of conrols. A complee lis of he variables and heir sources is given in Table A1 of he Appendix. Our moneary policy variable is he hree-monh money marke rae, deflaed by inflaion over he previous 12 monhs (r3mre). Since inflaion is relaively persisen in he shor run, r3mre should be a good proxy for he ex ane real rae. Twelve-monh inflaion is preferable o hree-monh inflaion for several reasons. Firsly, year-on-year inflaion is he headline measure of inflaion and should herefore correspond more closely o he expecaions of agens han price increases over a shorer ime span. Secondly, inflaion over shor periods may be disored by emporary facors. This is 10 Ialian nominal and real raes declined sharply in he afermah of is ejecion from he EMS in Sepember 1992, bu sho up again abou a year laer. BIS Papers No 19 241

refleced by he fac ha longer-erm inflaion ends o have beer forecasing power for fuure price developmens han shor-erm price rises. Thirdly, using price increases for periods of under a year forces us o seasonally adjus he daa, hus adding o poenial mismeasuremen. We measure fiscal policy by gross governmen deb (ggdgs) and curren ne borrowing (nbg) or, alernaively, cyclically adjused ne borrowing (nbgca). Daa availabiliy is a big issue when i comes o conrolling for shifs in he supply of, or he demand for, funds ha are no relaed o eiher moneary or fiscal policy. 11 Le us begin wih he supply of funds, ie saving. Public saving is already capured by our fiscal variable. We proxy privae saving by boh curren and expeced GDP growh. According o he permanen income and life cycle hypoheses, household saving should depend on fuure expeced or life income, which we proxy by he Consensus Forecas for GDP growh over he five- and 10-year horizon, respecively (gdp_e5 and gdp_e10). In addiion, many sudies have found ha curren income does seem o have an imporan effec on consumpion and saving. 12 We herefore include curren GDP growh (gdp) as a furher explanaory variable for saving. On he demand side, he main driving force of invesmen is probably expeced profiabiliy. We proxy fuure profiabiliy by he price/earnings raio (peraio) of he relevan MSCI counry index and a compeiveness index based on relaive price levels (comix). The laer is closely relaed o he concep of he real exchange rae. 13 Boh expeced and curren GDP are also likely o be imporan driving forces for invesmen. We conrol for he possibiliy ha he economy is below is producion fronier by including he oupu gap (y_gap) and unemploymen (unre). As menioned before, our measure for he real ineres rae may be pollued by an inflaion risk premium. We address his issue by including he 12-monh inflaion rae (pi) as a proxy for inflaion risk. This is based on he assumpion ha a posiive relaionship exiss beween he level of he inflaion rae and is (expeced) variaion. Moreover, including he inflaion rae as such allows us o say somehing abou he effec of nominal shocks. In order o avoid any problems arising from having an unbalanced sample, we firs presen esimaes for he counries for which daa are available for he whole sample period. We hen corroborae he resuls using he full (unbalanced) sample of counries. The daa for he esimaions wih a balanced panel are biannual (April and Ocober) and end in he firs half of 2002. In he case of he five-year rae, hey sar in April 1990, in he case of he 10-year rae in he firs half of 1991, leaving us wih a maximum of 25 periods and 23 periods for a given counry, respecively. In he case of he five-year rae, his maximum number of observaions is reached by Canada, he Unied Saes, he Unied Kingdom, France, Ialy and Germany, in he case of he 10-year rae by he same group of counries and also by Japan. Unforunaely, he comparaively small cross-secional dimension of our daase resrics he range of possible panel esimaion mehods. This is especially rue for dynamic panel esimaions like GMM à la Arellano and Bond (1991), which need a cerain minimum number of cross-secional unis. Therefore, due o he low degree of freedom we resric ourselves o esimaing a simple fixed-effecs model of he following ype: r i, = αi + β X i, + d + ε (3) Here, r i, is he respecive real rae for counry i a ime, α i is a counry-specific consan (he fixed effec ), X i, is a marix of explanaory variables ha vary across ime and counry, d capures he pure ime effec and ε is an error erm which is assumed o be independen and idenically disribued. For he esimaion, he individual means are subraced from equaion (3) (wihin esimaor). This fixedeffecs regression amouns o esing wheher movemens in differences of he longer-erm real raes 11 12 13 For a lis of possible facors deermining he supply of and demand for funds, see, for example, Blanchard and Summers (1984) or Chadha and Dimsdale (1999). Curren income will play a direc role if consumers are credi consrained or if income shocks are permanen. If uncovered real ineres rae pariy holds, hen he deviaion of a counry s real ineres rae from world raes should be equal o he expeced change in he real exchange rae over he mauriy of he invesmen. The expeced change in he real exchange rae should be relaed o he curren level if one assumes ha PPP holds in he long run, ie over five or 10 years, respecively. 242 BIS Papers No 19

across counries can be aribued o movemens in differences of he explanaory variables across counries. We esimae equaion (3) in levels. Alhough he evidence on wheher or no real ineres raes are saionary is inconclusive, here is lile reason o believe ha he deviaion of naional real raes from he world rae has a uni roo. Unforunaely, he small number of observaions does no allow us o apply sandard uni roo ess. One shorcoming of saic regressions such as (3) is ha i is difficul o rule ou reverse causaliy from r i, o he variables conained in X i,. We address his problem by repeaing he esimaion using lagged values of r3mre. This reduces he problem of reverse causaliy bu does no eliminae i, since real ineres raes, boh shor-erm and long-erm, are fairly persisen. Unforunaely, due o he low number of observaions we canno insrumen variables. 5.2 Esimaion resuls The resuls of our basic esimaions wih a balanced panel are presened in Table 1. Le us firs urn o he esimaion for he five-year real rae ha includes all he righ-hand variables lised above (regression 1). The coefficien of he shor-erm real rae r3mre is posiive and highly significan. Of he fiscal variables, governmen deb (ggdgs) and cyclically adjused ne borrowing (nbgca) are significan a he 5% level. Subsiuing he laer by non-adjused ne borrowing (nbg) does no change he resuls qualiaively. Of he conrol variables, expeced GDP (gdp_e5), he unemploymen rae (unre), he oupu gap (y_gap), and he rae of inflaion (pi) are saisically significan. The posiive coefficien on expeced GDP growh suggess ha economic growh maers primarily hrough is impac on he demand for funds (invesmen) raher han he supply (saving). The negaive coefficiens on unemploymen and he oupu gap indicae ha he real ineres rae is lower if he economy operaes below poenial. Finally, he posiive relaionship beween real ineres raes and inflaion poins owards he exisence of an inflaion risk premium. In order o obain a more parsimonious model, we sequenially rejec he leas significan variable unil only significan variables remain. The resuls are given in he second column of Table 1 (regression 2). Only variables ha were significan in he general model remain so in he smaller regression. In he process of reducing he number of coefficiens, he oupu gap also drops ou. The oher previously significan variables remain significan and keep heir signs. Moreover, boh he coefficien esimaes and heir sandard errors are of comparable size. Le us now urn o he resuls for he 10-year rae given in columns 4 and 5 of Table 1. Noe ha he sample now also includes Japan, for which five-year raes were no available for he whole esimaion period. In order o ensure comparabiliy, we repea he regression excluding Japan (columns 7-9). We find ha he coefficien of he hree-monh rae (r3mre) is highly significan bu slighly smaller relaive o wha we found in he previous regressions. The resuls concerning he fiscal variables depend on wheher or no he sample includes Japan. If Japan is included (column 4), hen he fiscal variables urn ou o be insignifican if boh ne borrowing (nbg or nbgca, respecively) and governmen deb (ggdgs) are included. However, governmen deb becomes significan if we drop ne borrowing (see regression 5) and vice versa. If we exclude Japan (regression 7), governmen deb ggdgs is significan a he 5% level. Among he conrol variables, only he price/earnings raio (pera) and he rae of inflaion (pi) are significan a he 1% level. The remaining variables are always insignifican a he convenional confidence levels. BIS Papers No 19 243

244 BIS Papers No 19 Explanaory variables r3mre 0.5171** (0.0438) nbgca 0.0869* (0.0363) ggdgs 0.0262* (0.0120) gdp 0.0115 (0.0348) gdp_e5/10 0.6157** (0.2177) unre y_gap Table 1 Resuls from fixed-effecs esimaion of basic specificaion r05cf r10cf (including Japan) r10cf (excluding Japan) Full sample period Sample spli (98:01) Full sample period Sample spli (98:01) Full sample period Sample spli (98:01) 1 2 3 4 5 6 7 8 9 0.1915** (0.0675) 0.1377* (0.0659) comix 0.0075 (0.0066) pera 0.0163 (0.0191) pi 0.3932** (0.0562) 0.4883** (0.0422) 0.0746* (0.0316) 0.0343** (0.0105) 0.4724* (0.1872) 0.1092* (0.0496) 0.3569** (0.0511) 0.4652** (0.0445) 0.0646* (0.0313) 0.0636** (0.0119) 0.3487 (0.1969) 0.1857** (0.0549) 0.3095** (0.0525) 0.0012 (0.1298) 0.0433 (0.0860) 0.0688** (0.0148) 0.4979 (0.3417) 0.3758** (0.1025) 0.1304 (0.1279) 0.4467** (0.0405) 0.0274 (0.0347) 0.0121 (0.0092) 0.0246 (0.0343) 0.1156 (0.1955) 0.0842 (0.0700) 0.0468 (0.0687) 0.0086 (0.0062) 0.0362** (0.0108) 0.4354** (0.0658) 0.3960** (0.0341) 0.0213** (0.0050) 0.0320** (0.0097) 0.3702** (0.0514) 0.4038** (0.0343) 0.0251** (0.0069) 0.0332** (0.0102) 0.3445** (0.0525) 0.0814 (0.0928) 0.0137 (0.0077) 0.0023 (0.0220) 0.2619* (0.1061) 0.4735** (0.0459) 0.0241 (0.0417) 0.0283* (0.0127) 0.0557 (0.0358) 0.0731 (0.3047) 0.1379 (0.0784) 0.0293 (0.0716) 0.0003 (0.0072) 0.0547** (0.0203) 0.4429** (0.0695) 0.4183** (0.0367) 0.0393** (0.0089) 0.0527** (0.0194) 0.3453** (0.0561) 0.4557** (0.0412) 0.0474** (0.0104) 0.0583** (0.0210) 0.3371** (0.0578) 0.0284 (0.1061) 0.0422** (0.0129) 0.0216 (0.0396) 0.2518* (0.1220)

BIS Papers No 19 245 Explanaory variables Table 1 (con) Resuls from fixed-effecs esimaion of basic specificaion r05cf r10cf (including Japan) r10cf (excluding Japan) Full sample period Sample spli (98:01) Full sample period Sample spli (98:01) Full sample period Sample spli (98:01) 1 2 3 4 5 6 7 8 9 Sample period 90:01-02:01 90:01-02:01 90:01-02:01 91:01-02:01 91:01-02:01 91:01-02:01 91:01-02:01 91:01-02:01 91:01-02:01 No of periods 25 25 25 23 23 23 23 23 23 Counries 6: DE, FR, US 6: DE, FR, US 6: DE, FR, US 7: DE, FR, US, JP 7: DE, FR, US, JP 7: DE, FR, US, JP 6: DE, FR, US 6: DE, FR, US 6: DE, FR, US No of obs 150 150 150 161 161 161 138 138 138 R 2 (w); R 2 (b) 0.92; 0.74 0.92; 0.76 0.93; 0.55 0.87; 0.86 0.86; 0.62 0.88; 0.59 0.89; 0.96 0.88; 0.85 0.90; 0.82 F-es (p-val) 0.015 0.001 0.000 0.002 0.000 0.000 0.066 0.000 0.000 Time dummies yes yes yes yes yes yes yes yes yes Noe: Sandard deviaions in brackes; */** = significan a he 1% and 5% levels.

How sable are hese resuls over ime? In paricular, did EMU affec he deerminaion of long-erm real ineres raes? We ry o shed more ligh on hese quesions by spliing he sample in early 1998. 14 Alhough EMU did no ake place unil a year laer, by ha ime markes already anicipaed wih reasonable cerainy which counries would paricipae and which would remain ouside. The oupu of he spli-sample regressions for he five-year rae are given in column 3 of Table 1 and hose for he 10-year rae in columns 6 and 9. In all hree cases, he coefficien of he shor-erm real rae becomes insignifican in he second subperiod, indicaing ha he effec of moneary policy on longerm real raes los srengh afer he beginning of 1998. The resuls concerning fiscal policy are less clear. The coefficiens on deb are highly significan and posiive in boh subperiods as long as Japan is excluded from he sample. If we include Japan, i ceases o be significan afer 1998. To sum up, he shor-erm real rae is significan in all regressions in he firs period bu no in he second. In conras o he shor-erm ineres rae, here is no general evidence ha fiscal policy has los is imporance in he lae 1990s. Only in he case of 10-year real raes for he sample including Japan do we find ha governmen deb is significan before 1998, bu no aferwards (Table 1, regression 6). This suggess ha he high level of Japanese deb combined wih low real ineres raes since he lae 1990s drove he resuls for he full sample. Japan should herefore be viewed as an excepion. In all oher cases, we canno saisically rejec equaliy of coefficiens for he wo subsamples. 5.3 Robusness In order o check he robusness of he resuls wih respec o changes in he specificaion, we rerun he esimaions (a) wih an alernaive esimaion mehod, (b) wih dummy variables ha conrol for EMU counries, (c) wih a lagged shor-erm ineres rae and (d) for he whole (ie unbalanced) sample. (a): given ha we canno assume ha here is no cross-counry correlaion in he included variables and ha here is no auocorrelaion wihin he variables of a given counry, we reesimae he models wih Feasible GLS. This allows for heeroskedasic panels, cross-counry correlaion and counryspecific AR1 coefficiens. Given he small size of our sample, however, i is unlikely ha FGLS is efficien. The resuls given in Table A2 in he Appendix should herefore merely be used as a robusness check of he resuls of he basic fixed-effec regressions. As i urns ou, he resuls obained over he full sample period, namely ha he shor-erm real ineres rae and fiscal policy ha significanly deermine long-erm real raes, are confirmed. Bu we no longer find ha he effec of shor-erm real raes weakens over ime. (b): an obvious candidae explanaion for he srucural break found in regressions 3-9 of Table 1 is, of course, EMU, which consrains nominal raes in he member counries o be essenially idenical. 15 As a consequence, differences in real ineres raes are mainly due o differen inflaion expecaions. A counry wih an overheaed economy will herefore have lower real raes han a counry where he economy operaes below poenial - jus he opposie of wha one would normally expec. In order o es for possible EMU effecs, we define a dummy variable which is equal o one for EMU counries (DE, FR, IT) on and afer 1998:1, bu zero oherwise. We inerac his dummy wih he moneary and fiscal variables, as well as he conrols ha showed significan coefficiens in he previous regressions, and esimae his augmened model. The resuls of his exercise are reproduced in Table A3 (see Appendix). The EMU variables urn ou o be insignifican in all cases, suggesing ha he break we found in Table 1 canno merely be aribued o EMU bu is a more general phenomenon. (c): we ry o reduce he problem of reverse causaliy beween r3mre on he one hand, and r5cf and r10cf on he oher, by subsiuing he curren shor rae wih is value lagged by one period (Table A4 in he Appendix). We generally find ha moneary policy and fiscal policy are significan deerminans of long-erm real raes (regressions 2, 5 and 8), alhough for some reason cyclically adjused 14 15 The resuls are virually idenical if he spli dae is 1997:2 or 1998:2. Assuming ha premia for defaul and liquidiy risk are of minor imporance. 246 BIS Papers No 19

governmen borrowing seems o perform beer ha deb. In addiion, he resul ha moneary policy seems o have los srengh is confirmed (regressions 3, 6 and 9), whereas he evidence concerning fiscal policy is more mixed. In he case of he 10-year rae, esimaes for he sample excluding Japan show a highly significan coefficien for cyclically adjused ne borrowing in he firs subperiod, bu an insignifican coefficien for he second subperiod. In he regressions wihou Japan, governmen borrowing is eiher always insignifican (five-year raes) or always significan (10-year raes). 16 (d): we repea he esimaes using he unbalanced panel of 10 counries for which a leas some observaions are available. The resuls are colleced in Table A5 of he Appendix. Apar from some differences concerning he conrol variables, hey largely confirm previous esimaes as long as we only look a he enire sample period. If we spli he sample in 1998, imporan differences o previous resuls appear. In paricular, he evidence ha moneary policy has los is impac afer 1998 canno be corrobaed. Wih one excepion (regression 9), he coefficien on he shor-erm real rae remains significan, alhough he poin esimaes become somewha smaller. Fiscal policy appears o be equally effecive hroughou he 1990s and early 2000s, a leas ouside Japan. 5.4 Discussion Our esimaes indicae ha he shor-erm real ineres rae is one of he main facors driving long-erm real ineres raes. In all our regressions over he full sample ranging from he early 1990s o spring 2002, lower shor-erm real raes are ceeris paribus associaed wih lower long-erm real raes. This resul is robus o changes in he esimaion mehod or in he composiion of he sample. However, we find some evidence ha he effec of shor-erm real raes on long-erm raes has on average become less srong in recen years. In mos of our regressions, he coefficien for he shorerm real ineres rae urns ou o be insignifican in esimaions covering only he laer par of our sample. This is no simply he resul of EMU bu applies also o he counries ouside he euro area. In par, his may be due o he fac ha boh shor-erm and long-erm real ineres raes have become more synchronised across counries, leaving less variaion for our model o exploi. This is compaible wih he hypohesis ha he relaive imporance of world facors, or a leas inernaional facors, for naional long-erm real raes has increased over ime. This resul is in line wih he findings of he empirical lieraure on real raes, for example Pain and Thomas (1997) and Founas and Wu (1999, 2000). Since our esimaion mehodology is based on he differences beween counries, i is no surprising ha we do no obain significan coefficien esimaes. If i is rue ha moneary policy can conrol shor-erm real raes, hen our resuls - if aken a face value - sugges ha moneary policy provides a powerful ool o influence long-erm real ineres raes. However, can we safely ake hem a face value? One reason for scepicism is poenial reverse causaion from long-erm real raes o he sance of moneary policy. For example, if long-erm real ineres raes are high because he economy is expeced o grow rapidly over he nex few years, inflaionary fears may force he cenral bank o raise shor-erm ineres raes. Neverheless, he fac ha our resuls basically hold even if we use lagged shor real raes suggess ha reverse causaion is no he whole sory. Our resuls concerning fiscal variables are less clear han hose for he shor-erm real rae, alhough his seems o be driven mainly by he Japanese experience of he lae 1990s and aferwards. If one drops Japan, hen gross governmen deb urns ou o be an imporan driving force for real ineres raes, suggesing ha Ricardian equivalence does no hold. The resuls concerning cyclically adjused governmen borrowing are less robus. Our resuls are by and large in line wih he exising empirical lieraure, which has up o now concenraed on ime series economerics and no on panel mehods. Blanchard and Summers (1984) find ha he high real ineres raes of he lae 1970s and he early 1980s were probably due o he fiscal-moneary policy mix. Fiscal facors are also sressed by Barro and Sala-i-Marin (1990) and by Ford and Laxon (1999). Breedon e al (1999) also sress he imporance of domesic facors in deermining long-erm real ineres raes, among hem he governmen deb-o-gdp raio and real 16 These resuls hold if we use governmen deb insead of ne borrowing. BIS Papers No 19 247

shor-erm raes. The role of moneary policy is sressed by Allsopp and Glyn (1999), who conclude heir paper wih he saemen ha heir analysis leads hem... o conclude ha i would be wrong o hink of he real ineres rae as deermined... independen of moneary policy (p 15). As in he case of real ineres raes, our resuls concerning fiscal policy represen correlaions ha do no necessarily indicae causaliy. Reverse causaliy may be a problem in paricular for ne lending. In he long run, high real ineres raes add o deb servicing coss and herefore o he governmen s financing requiremens. Neverheless, ineres paymens on he public deb end o be relaively sicky due o he prevalence of long-erm nominal ineres raes in mos counries of our sample and persisen inflaion. This should reduce he problem of reverse causaliy, especially since we are esimaing a a half-yearly frequency. 6. Conclusions The aim of he paper was o shif he focus of he debae on real ineres raes more owards long-erm raes. We argue ha he analysis of real ineres raes is only useful if he ime horizon for such raes covers he whole invesmen period. In he case of capial invesmen, his can be several years, in he case of saving even decades. Given he imporance of long-erm real ineres raes, i is naural o ask o wha exen hey can be influenced by macroeconomic policy. In he case of moneary policy, his is he sage where shor-erm real ineres raes become imporan. In he presence of nominal rigidiies, shor-erm real raes are under he conrol of moneary policy. Alhough heir direc impac on real variables may be limied, hey may have a much greaer indirec effec hrough heir influence on longerm raes. This link beween moneary policy, proxied by he shor-erm real ineres rae, on he one hand, and long-erm real raes on he oher, is a he hear of he paper. We also consider he effec of fiscal policy, as proxied by governmen ne borrowing and governmen deb. In conras o moneary policy, fiscal policy affecs he long end of he erm srucure direcly by changing he demand for loanable funds. Empirical work on ex ane real ineres raes comes wih serious measuremen problems. One of he reasons why relaively lile work has been done in his field is scarciy of daa. In he absence of inflaion-indexed bonds, real raes canno be observed direcly. We believe ha survey evidence on inflaion expecaions can help us o overcome his problem. These price forecass can be used o deflae nominal ineres raes o yield esimaes of ex ane real raes. We discuss he problems associaed wih survey daa in deph, bu come o he conclusion ha hey seem o provide a fairly good approximaion o he rue beliefs of marke paricipans. Our esimaion resuls sugges ha boh moneary and fiscal policy generally play an imporan role in he deerminaion of long-erm real ineres raes. Neverheless, he imporance of moneary policy for long real raes appears o have diminished since he lae 1990s. An obvious candidae explanaion for his is EMU, in paricular since abou half of he counries in our sample belong o he euro area. This is no borne ou by he daa, however. We find ha moneary policy has also los is impac among he non-emu counries. This is consisen wih evidence ha moneary policy has become more synchronised across counries, leaving less room for naional real ineres raes o diverge. Unforunaely, we canno say wheher his implies ha moneary policy has indeed become less effecive or wheher he counries have simply no ried o seer long-erm real ineres raes away from he group of counries we consider. In conras o moneary policy, here is no evidence ouside Japan ha a counry-specific fiscal policy has become less imporan over ime. Only in Japan have low real ineres raes coincided wih high deb and governmen borrowing, hus producing insignifican esimaes for he coefficiens on he fiscal variables. 248 BIS Papers No 19

Daa appendix Table A1 Variables Descripion Source A. Moneary variables r3mre Three-monh real ineres rae (nominal rae minus annual inflaion over las 12 monhs) BIS B. Fiscal variables nbg Ne borrowing by general governmen, % of GDP (corresponds o ne lending in OECD daabase, muliplied by 1) OECD nbgca Ne lending by general governmen, cyclically adjused, % of GDP OECD ggdgs General governmen gross deb, % of GDP OECD C. Conrol variables gdp_e5/10 Real GDP growh expeced over he nex 5/10 years Consensus Forecas gdp Growh rae of curren GDP, 1995 prices OECD y_gap Oupu gap, % of GDP OECD unre Unemploymen rae OECD pi Inflaion over he previous 12 monhs OECD pera Price earnings raio based on analyss forecass for profis of firms conained in MSCI counry index in 12 monhs ime I/B/E/S comix Compeiiveness index, based on relaive consumer prices OECD BIS Papers No 19 249

250 BIS Papers No 19 Explanaory variables r3mre 0.4128** (0.0399) nbgca 0.0530 (0.0310) ggdgs 0.0360** (0.0110) gdp 0.0275 (0.0234) gdp_e5/10 0.5049** (0.1776) unre 0.1144 (0.0648) y_gap 0.0155 (0.0445) comix 0.0024 (0.0049) pera 0.0225 (0.0183) pi 0.2897** (0.0453) Table A2 Resuls from FGLS esimaion (heeroskedasic panels wih cross-counry correlaion, counry-specific AR1) r05cf r10cf (including Japan) r10cf (excluding Japan) Full sample period Sample spli (98:01) Full sample period Sample spli (98:01) Full sample period Sample spli (98:01) 1 2 3 4 5 6 7 8 9 0.4383** (0.0367) 0.0370** (0.0098) 0.5011** (0.1526) 0.3070** (0.0420) 0.4530** (0.0379) 0.0352** (0.0101) 0.5189** (0.1686) 0.3056** (0.0416) 0.3393** (0.0924) 0.0349** (0.0112) 0.3042 (0.3313) 0.2849** (0.0903) 0.3547** (0.0305) 0.0010 (0.0221) 0.0035 (0.0076) 0.0217 (0.0191) 0.1640 (0.1228) 0.0795 (0.0496) 0.0342 (0.0339) 0.0072* (0.0036) 0.0141 (0.0083) 0.3469** (0.0371) 0.3045** (0.0295) 0.0056 (0.0047) 0.2533** (0.0361) 0.3128** (0.0308) 0.0078 (0.0048) 0.2188** (0.0412) 0.2209** (0.0605) 0.0012 (0.0058) 0.1968** (0.0637) 0.3502** (0.0363) 0.0146 (0.0316) 0.0285* (0.0121) 0.0209 (0.0204) 0.0716 (0.1949) 0.1444* (0.0581) 0.0506 (0.0374) 0.0018 (0.0042) 0.0035 (0.0155) 0.3131** (0.0456) 0.2788** (0.0330) 0.0322** (0.0121) 0.2224** (0.0413) 0.3469** (0.0295) 0.0370** (0.0096) 0.2633** (0.0407) 0.0943 (0.0663) 0.0345** (0.0108) 0.1654** (0.0579)

BIS Papers No 19 251 Explanaory variables Table A2 (con) Resuls from FGLS esimaion (heeroskedasic panels wih cross-counry correlaion, counry-specific AR1) r05cf r10cf (including Japan) r10cf (excluding Japan) Full sample period Sample spli (98:01) Full sample period Sample spli (98:01) Full sample period Sample spli (98:01) 1 2 3 4 5 6 7 8 9 Sample period 90:01-02:01 90:01-02:01 90:01-02:01 91:01-02:01 91:01-02:01 91:01-02:01 91:01-02:01 91:01-02:01 91:01-02:01 No of periods 25 25 25 23 23 23 23 23 23 Counries 6: DE, FR, US 6: DE, FR, US 6: DE, FR, US 7: DE, FR, US, JP 7: DE, FR, US, JP 7: DE, FR, US, JP 6: DE, FR, US 6: DE, FR, US 6: DE, FR, US No of obs 150 150 150 161 161 161 138 138 138 Cross-counry correlaion Counry-specific AR1 yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes Time dummies yes yes yes yes yes yes yes yes yes Noe: Sandard deviaions in brackes; */** = significan a he 1% and 5% levels.

Table A3 Resuls from fixed-effecs esimaion: esing for he effecs of EMU Explanaory variables r05cf Including Japan r10cf Excluding Japan r3mre 0.4863** (0.0484) nbgca 0.0791* (0.0342) ggdgs 0.0434** (0.0136) gdp_e5/10 0.4267* (0.1915) unre 0.0681 (0.0523) 1 2 3 EMU EMU EMU 0.1828 (0.2176) 0.1224 (0.1214) 0.0043 (0.0120) 0.1288 (0.4990) 0.0943 (0.1383) 0.3836** (0.0409) 0.0160** (0.0056) pera 0.0231* (0.0104) pi 0.35113** (0.0564) 0.0753 (0.2838) 0.3421** (0.0584) 0.0581 (0.1940) 0.0120 (0.0068) 0.0555 (0.0290) 0.0045 (0.1973) 0.4102** (0.0470) 0.0355** (0.0137) 0.0467* (0.0209) 0.3372** (0.0649) Sample period 90:01-02:01 91:01-02:01 91:01-02:01 No of periods 25 23 23 0.1259 (0.2073) 0.0037 (0.0086) 0.0351 (0.0326) 0.1150 (0.2311) Counries 6: DE, FR, US 7: DE, FR, US, JP 6: DE, FR, US No of obs 150 161 138 R 2 (w); R 2 (b) 0.92; 0.74 0.87; 0.80 0.88; 0.86 F-es (p-val) 0.004 0.000 0.025 Time dummies yes yes yes Noe: Sandard deviaions in brackes; */** = significan a he 1% and 5% levels. 252 BIS Papers No 19

BIS Papers No 19 253 Explanaory variables r3mre ( 1) 0.2483** Table A4 Resuls from fixed-effecs esimaion: including lagged shor-erm ineres rae r05cf r10cf (including Japan) r10cf (excluding Japan) Full sample period Sample spli (98:01) Full sample period Sample spli (98:01) Full sample period Sample spli (98:01) 1 2 3 4 5 6 7 8 9 (0.0544) nbgca 0.1726** (0.0520) ggdgs 0.0158 (0.0173) gdp 0.0377 (0.0476) gdp_e5/10 0.3028 (0.3008) unre 0.3565** (0.0912) y_gap 0.0034 (0.0889) comix 0.0051 (0.0092) pera 0.0117 (0.0267) pi 0.1143 (0.0693) 0.2199** (0.0476) 0.1958** (0.0358) 0.3574** (0.0568) 0.1820** (0.0525) 0.1494** (0.0406) 0.4025** (0.0663) 0.1102 (0.1531) 0.2446* (0.0989) 0.6155** (0.1002) 0.2258** (0.0450) 0.0466 (0.0446) 0.0227 (0.0117) 0.0153 (0.0438) 0.0726 (0.2515) 0.0653 (0.0881) 0.0562 (0.0873) 0.0062 (0.0080) 0.0312* (0.0139) 0.2088** (0.0803) 0.2398** (0.0422) 0.0644** (0.0231) 0.2359** (0.0646) 0.2608** (0.0446) 0.0368 (0.0321) 0.2498** (0.0727) 0.1097 (0.1214) 0.0338 (0.0682) 0.0535 (0.1563) 0.2196** (0.0544) 0.0987 (0.0588) 0.0214 (0.0191) 0.0052 (0.0475) 0.2161 (0.4084) 0.0409 (0.0974) 0.0661 (0.0943) 0.0005 (0.0096) 0.0293 (0.0269) 0.1956* (0.0878) 0.2282** (0.0455) 0.1375** (0.0378) 0.2321** (0.0739) 0.2527** (0.0493) 0.1080** (0.0407) 0.2532** (0.0820) 0.0804 (0.1459) 0.0129 (0.0824) 0.0429 (0.1839)