Some Evidence on the Efficacy of the U.K. Inflation Targeting Regime: An Out-of-Sample Forecast Approach

i Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 565 September 1996 Some evidence on the efficacy of the UK inflation targeting regime: an out-of-sample forecast approach Chan Huh NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comments. References in publications to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors.

Abstract Inflationtargeting(IT)--apolicyframeworkthatdirectlytargetsanexplicitinflationgoal--hasgained widespreadattentionrecentlyas ithasbeenadoptedby severalOECDcountries. Thereisa growing bodyof literatureonthe ultimatelong-termbenefitsof pricestabilityandontheoreticalissuesrelated to inflationtargeting. Butthe shortdurationof thispracticehas limitedthe numberof worksthat empiricallyanalyzetheperformanceof ITregimes. Thispaperexaminesthe Britishinflationtargeting experiencesince1993by focusingontheout-of-sampleforecastperformanceof modelsfittedto the 1980s. The model over-predicts actual short-term and long-term interest rates, while it’sinflation forecast is on track for the recent period. This implies that it took less monetary tightening to obtain a favorable inflation outcome. Identical exercises were repeated for France and the US, countries that have not adopted IT but have experienced low inflation in the recent period. The results for these countries show that recent low inflation has not been unusual when compared to forecasts from the models designed to fit the second half of the 1980s. That is, given the level of inflation, the degree of actual monetary policy tightness (measured in terms of short-term interest rate) is about what the model expects. Findings of this paper could be explained by enhanced credibility of the UK monetary policy since the adoption of IT.

Someevidenceon the efficacyofthe UK inflationtargeting regime: an out-of-sample forecast approach Chan Huhl I. Introduction Inflation targeting (IT)--a policy framework that directly targets an explicit inflation goal--has gained widespread attention in recent periods.2 The ultimate long-term benefits of price stability to be realized through inflation targeting are expected to be quite large. Also, a body of papers that discuss related theoretical issues, such as optimal designs of targets, has grown over time. However, the fact that these regimes have only been in place for a relatively short period of time has limited the number of works that empirically analyze the performance of recent IT regimes.3’4 This paper empirically examines the question of whether there have been perceptible changes (or structural breaks) in how key macroeconomic variables interact since the introduction of the IT regime in the United Kingdom. Adopting a Vector Autoregression (VAR) modelling framework, I focus on the out-of-sample forecast performance of models fitted to the 1980sduring the recent IT 11receivedhelpfuldiscussionsandcommentsfromDavidBowman,DickFreeman,JoeGagnon,andDeb Lindner.I alsogratefullyacknowledgeprogrammingassistancebyRobertIngenitoandresearchassistanceby LeslieDavisandJohnHeitkemper.Anyremainingerrorsaremyown.Thispaperrepresentstheviewsofthe authorandshouldnotbeinterpretedasreflectingthoseoftheBoardofGovernorsoftheFederalReserve System,or theFederalReserveBankofSanFranciscoorothermembersoftheirstaff. Pleaseaddress correspondenceto:ChanHuh,EconomicResearchDepartment,FederalReserveBankof SanFrancisco,P. O. Box7702,S,anFrancisco,CA94120.E-nlail:chan.huh@sf.frb.erg,tel:415-974-2393. ‘NewZealand(1990),CanadaandIsrael(1991),theUnitedKingdom(1992),SwedenandFinland(1993), andSpain(1994). 3Fordiscussionsofthelong-termbenefits,seeFisher(1994),King(1994). Varioustheoreticalissuesare discussedinSvensson(1993,1996),HallandMankiw(1994),Woodford(1994),McCallum(1995). Ammer andFreeman(1995),FreemanandWillis(1995),thepapersinHaldane(1995),andinLeidermanandSvensson (1995)offerdetaileddescriptiveaccountsonsomeofthecountriesthatadoptedIT. 4TheUKhasbeenoneofthecountriesthatimplementedIT. However,thedepthandbreadthoffinancial marketssetthe{JKapart. Forkrther detailsofinila[iontargetingintheUK,seeKing(1994)andBowen (1995).

period. If there has been a noticeable structural shifi, the manner in which forecasts of the model - fitted totheearlier period mismatch thedata in the 1990sshould offer clues about the change. In particular, we focus on the model’sforecast errors during the IT period for inflation, the short-term interest rate, and the long-term interest rate to see if they show any unusual patterns. It should hold true that when monetary policy becomes more credible a less restrictive monetary policy would accompany low inflation, ceteris paribus. An alternative way of gauging a structural shifi is to assess changes in the terms of the tradeoff between output and inflation--i.e., the “sacrifice ratio”. However, such an approach requires making a structural interpretation of estimated reduced form models. The thrust of this exercise isnot to discern the nature of the dynamic relationships, but rather whether or not the structure of the dynamic relationship has remained intact throughout the sample period. Focusing on forecast performance consequently puts less demand on the estimated model Furthermore, the IT immediately followed the UK’smembership in the Exchange Rate Mechanism. During the ERM period, the British monetary policy stance was tight due to the need to support the pegged pound exchange rate prescribed by the ERM (Ammer and Freeman, 1995). Consequently, inflation was low when the inflation targeting monetary regime was first installed. This makes casting the UK’sexperience in terms of “sacrifice ratios” somewhat absurd.s Preliminary examinations are carried out using a VAR model of the UK economy consisting of six quarterly variables estimated using data up to 1990.6 The fit of the model to later periods ‘ThisisinstrongcontrasttoNewZealand’sexperienceduringtheperiodleadingupto IT.Aprotracted periodofmonetarytighteningwasnecessarytobringdowninflationwhichwasat 15percent,incuninga substantialouput loss. Hence,the“sacrificeratios”gainedcurrencyasthemeasureofeffectivenessoftheIT monetaryregime. SeeMayesandChapple(1995)foracriticalreviewofthisissue. %e variablesare:realGDPgrowth,unemploymentrate,inflationinretailpriceindex(RPIX),thetradeweightedpoundexchangerate,short-terminterestrate,andlong-terminterestrate. Thelasttwoeachrelateto monetarypolicystanceandinflationexpectationsplusriskpremium. -2-

deterioratesdrastically,suggestinginstabilityinthemodel. To furtherinvestigatethesepreliminary observations,a VARwithBayesianpriors(BVAR)isestimated. Thebaselinemodelthatrepresents macroeconomicdynamicsup to the late 1980s is estimated using data from the corresponding period.’ The results show a noticeable divergence in the model’sforecast performance since 1990with respect to inflation and the short-term interest rate and, to a lesser extent, the long-term interest rate. The model’sinflation forecast had large forecasting errors (over-prediction) during the ERM periods and the early part of the IT period. However, this over-prediction bias disappeared rapidly. The longterm interest rate forecast showed no clear bias until the beginning of the IT period. Then, the model consistently over-predicted the actual long-term rate. Most noticeably, the model consistently over-predicted short-term interest rates throughout the inflation targeting period afier showing a reasonable fit during the ERM period. In short, despite a monetary policy stance that has not been as tight as the model would suggest, inflation has remained close to the model’sforecast. The actual long-term rate also has been lower than the model’sprediction, suggesting lower-than-expected inflation expectations and inflation risk premium in later periods. One could attribute such constellations of forecast errors to an enhanced credibility of monetary policy. That is, monetary policy has become more effective in the sense that it takes less actual tightening to obtain a favorable inflation outcome as markets expect future monetary policy to beconducted along a path compatible with maintaining low inflation. Alternatively, favorable price shocks, such as a fall in import prices, could also explain low inflation. However, there have been several developments such as sterling depreciation during the period that have put upward pressure on 7TheBVARframework,whichwasdevelopedbyLittermanandSimsbasedonTheil’s(1972)mixing estimationmethodology,issuitableforthisexercise. Itentailsspecifyinga setofparametersthatrepresent priorknowledgeaboutthestructureoftheeconomy,whichisusedinconjunctionwithactualdataforthe estimationofthemodel. Wedeterminetheseparametersinanoptimizingfashionusingdatafrom1985to 1990. Theyinturnarefixedwhenforecastingtheout-of-sampleperiod(1990-1995).Inthissense,theyrepresenta salientdatastructureofthe1985-1990period. -3-

inflation. Hence the explanation of enhanced monetary policy effectiveness becomes plausible. A moderating inflation trend has not been unique to the UK but has been seen in many OECD countries in the 1990s. Presumably, the recession that visited major G-7 countries at the beginning of the 1990scould have caused this. However, the moderate trend has continued even when most of these economies moved well into recovery phases. This raises the possibility that the earlier finding of the mis-match between inflation and interest rates might not be unique to the British economy and particularly may have little to do with the institution of inflation targeting. To test this possibility, the identical exercise was repeated using data from the US and France, two countries that have not adopted explicit IT monetary policy regimes. The results for tliese countries show that recent low inflation has not been unusual when compared to forecasts from the models designed to fit the second half of the 1980s. That is, given the level of inflation, the degree of actual monetary policy tightness (measured in terms of short-term interest rate) is about what the model expects. Hence, it is unlikely that this paper’sresults on the UK are mainly due to an exogenous low inflation trend commonly seen in most OECD countries, but not captured by the model. A caveat to this paper’sfinding is in order. Despite the low and stable inflation in the 19!?0s, indicators of long-term inflation expectations have not shown noticeable change. C)neindicator is the yield spread between the British and the German long-term securities. Though the spread is narrower in the 1990scompared to the 1980son average, it has not narrowed noticeably during the recent period. Survey measures of inflation expectations also have declined at a glacial pace.8 They seem to reveal lingering doubts about whether current low inflation can be extended into the future. This in ‘However,anindicatorofinflationexpectationsderivedfromacomparisonofconventionalandinflationindexedbondyieldsshowsomechangeinpatternsinthe1990s. Currentinflationexpectationsmeasuresfor fiveandten-yearhorizonsrespectively150basispointslowerthanitspeak seeninearly1994(5-years)andin 1992-93(lO-years). However,boththecurrentlevelsofthebothmeasuresare2 to3percenthigherthanthe prevailinginflationrate(p.47, InflationReport(1996)). -4-

The remainder of the paper proceeds as follows. Section II reviews UK economic . developments in th ;late 1980sand early 1990s. Section 111describes the model and its estimation, respectively. Resu ts are examined in Section V, and Sectio~lVI concludes. II.EconomicDevelopmentsin the United Kingdom: 1985-1992 TheBritishmonetaryauthoritiesfocused,inturn,on broadaswellas narrowmoneta~ aggregatesas intermediatetargetsinthemid-1980s. Startinginthesecondhalfof the 1980s,the focus startedto shifi to exchange rates as the pound sterling steadily appreciated against the German mark. This began around the period of the 1985Plaza Accord and the Lourve Accord of 1987,at which time major industrialized countries agreed to lower the value of the dollar and to maintain stability in key exchange rates. A widening external imbalance brought about by rising imports in 1986also contributed to Britain’sshifi to managing exchange rates around that time. External balances, which first recorded a current account deficit of &l.5 billion in 1986after several years of surplus, deteriorated rapidly and reached about f29 billion in 1988fueled by a strong domestic demand for imports. GDP grew at 4 to 5 percent annually in real terms between 1985and 1988. In the meantime, the annual inflation in consumer prices, after ebbing to 4 percent in 1986, started to rise along with a surge in the domestic demand. This was followed by a substantial depreciation (about 10percent) of the pound exchange rate in 1989. However, this fall in the pound exchange rate was arrested as the short-term interest rates were raised by about 2 percentage points to 15percent at the end of 1989. Though short-rates gradually fell thereafter, it was not until March 1991that the rates fell to where they had been in January 1989. The exchange rate remained stable throughout 1990and 1991. At the same time, output growth that averaged about 4 percent in the preceding five years first sharply contracted in the third -5-

I quarter of 1990and then continued to be negative throughout 1991. On the other hand, inflation measured in terms of the year-over-year retail price index excluding mortgage interest payment reached a peak of 9-1/4 percent in the fourth quarter of 1990. This increase, from a 6 percent range a year earlier, was partly due to the run-up in oil prices associated with the Gulf-crisis. A substantial output contraction notwithstanding, both this high inflation as well as the need to support the pound exchange rate initially severely limited the options available to monetary authorities. In particular, due to high German interest rates associated with the financial burden of the unification, the UK rates had to be kept high to defend the pegged sterling exchange rate. On the fiscal policy side, the government’sbudget balance, which maintained a surplus for several years afier 1987,also started to deteriorate significantly starting in early 1991. By late 1991,the situation became more and more untenable. Finally sterling lefi the ERM in September 1992when it came under overwhelming pressure caused by a large-scale selling of sterling in the foreign exchange markets. This withdrawal subsequently lefi no nominal anchor to guide monetary policy. In October, the Chancellor of Exchequer announced the adoption of IT. 111.Estimation One way to examine the impact of this sequence of events on the relationship how variables interact is to rely on a general model. For such an investigation, a VAR model of six variables was fitted to the UK data as follows; (1) Xt=D +~ BiXt-i+et, i=l E[e(t)e(s)’] =Z if s =t, HereX = {y, un,n, ex, sr, /r }: y; real GDP growth, un;unemployment rate, z; inflation in retail -6-

=0 othewise. price index net of mortgage interest payments (RPIX), sr; short-term interestrateasthe keymeasure of monetary policy, ex; trade-weighted nominal pound exchangerate,andZr;the long-term interest rate.9 Figure1showsdataseriesfortheperiod1985Q1to 95Q3. First,the modelisestimatedusing datafrom 1972Q1to 90Q2. Next,whetherthisspecificationremainsstableisexaminedby inspecting residuals generated by fitting the model to data of the sample period not used for the estimation (i.e, 1990Q2-95Q3). Figure 2 shows two sets of residuals from the equations for inflation, short-term and long-term interestrates. Thetop panelshowsresidualsfromthemodelestimatedfrom 1972Q1-90Q2.andthe bottompanelshowsthosefromthemodelestimatedfrom 1972Q1-92Q3. Theseparticulardatesare chosensincethe firstdenotestheUK’sjoiningof ERM,andthesecondfortheUK’swithdrawalfrom theERMandthebeginningoftheIT regime. Twostandarderrorbandsforeachresidualarealso shown. Themis-matchbetweenthedataandthemodelforthepost-ERMperiodisevident. The model’sinflationforecastunder-predictsinflationandover-predictsinterestrates. Thistendency diminishesnoticeablyfor inflationandthe long-terminterestrate,butpersistsforthe short-term interestratewhenthemodelestimationperiodisextendedto includetheERMperiod(i.e., 1990Q3- 92Q3). Thestandarderrorbandforthe long-terminterestrateforecastalsonoticeablywidened. Theseobservationssuggestthattherehasbeenat leastoneperceptiblebreakinthe sample period. ThisbodeswellwiththeeventsintheBritisheconomy. In particular,the ERMandthe 9TherealGDPindexexcludingtheoilsectorisusedastheoutputvariable.Inflationismeasuredinterms oftheretailpriceindexexcludingmortgageinterestpayment. TheX-11filterwasappliedtothepriceindex, RPIX,toremoveseasonalitybeforecalculatinginflationrates. Atrade-weightendominalaverageexchangerate compiledbytheBankofEnglandisusedasthepoundexchangerate. Fortheshort-andlong-terminterest rates,theratesonthe3-monthinterbankloanandonthe3-1/2percentwarloan(consol)arerespectivelyused. GrowthratesofrealGDPandtheRPIX(ie.,inflationa)reused. Fortherestofthevariables,loggedseries wereused. Thelaglengthofsixwasdeterminedbytestingvariousalternativeussingthelog-likelihoordatio testofSims(1980).Usingtherealexchangerate,insteadofnominal,didnotmateriallyaffecttheresults. -7-

adoption of the IT regime each could offer distinct demarcation points, To push this further, I follow a variant of the VAR modeling approach, namely, VAR with Bayesian priors developed by Litterman (1984) and Sims (1982) based on the mixing estimation methodology of Theil (1972). The BVAR has mainly been used to improve long-term forecasting accuracy by estimating coefficients using both data and reasonable priors.’” This framework is useful since a modeler can choose specific values for the priors by means of optimizing criterion to a particular sub-sample period. Thus, one could tailor the model specification to incorporate the d~’namicstructure of the data, or economy, in the sense of a set of prior restrictions on coefficients. Consequently, by fixing the priors to the values determined at the earlier stage in subsequent estimations, one could preserve the dynamic structure of the baseline estimation period. This idea is implemented in the following way: First, an ordinary VAR was estimated using data from 1973Q2 to 84Q4. Second, a set of hyper-parameters representing ‘priors’are determined so as to minimize the one- to four-quarter ahead out-of-sample forecast of the VAR model for 1985Q1-90Q2. The end-product of step two is the BVAR version of (l). The prior distributions for the coefficients (b,s) are specified as follows: ii - N( 1,fla, ~, y]), for i =1 ati bi - N(O,f(a, ~, y)), for i>1. Here the subscript i denotes the lag length. This set of priors amounts to a random walk with a drifi. ]] ‘“Thepriorinformationisintroducedinthewayofhyper-parametersthatinfluence,ineachequation,the degreeofinteractionwithdependentvariable’sownlagsaswellasacrossdifferentvariablesingeneral,rather thanspecificindividualcoefficients. ‘*AlternativelyA, R(1)coefficientsestimatedusingtheinitialsampleperiod(1972-1984)wereusedasthe priorvalues. Resultswerenotsensitivetosuchchangesintheprior. -8-

i The variance of the prior distribution for a coefficient is given as f(.), which inversely reflects the degree to how certain the prior being imposed should be. That is, a small f(.) suggests that the chosen prior is very tightly distributed around the mean value. A large value for f(.) conversely suggests that the imposed prior has a large variance, hence a loose prior. The final estimation of coefficients is done by combining the prior and the actual data. Relatively speaking, the larger f(.), the stronger the influence of actual data on determining the coefficients.i2 To be specific, the values for the hyper-parameters a, ~, and y were chosen to optimize the model’sout-of-sample forecast performance of the VAR model estimated from the first stage for the sample period 1985Q1-90Q2. The optimization involves an objective function consisting of the sum of Root Mean-Squared-Errors over one- to four-quarter ahead forecasts. (2) F(. ) =~ w~~ [NSE [actual(m, t+k)-forec~t(m, t, k)]] , Zm k=l m.y where Wiand z~are respectively indexes for the forecast horizon, and the variables whose forecast errors are included in the objective function. The index k represents the forecast horizon. For example, the following expression stands for the difference between the output growth two periods 12Thevariancef(.)isdeterminedasafunctionofthreeparameters;a, ~,andy. Thethreeparameterseach representtheoveralltightnessoftheprior,howfasttheinfluenceoflaggedvaluesdecay,andthedegreeof cross-variabledynamics.Forexample,theelement~dictatestherateofdecreaseinthevalueoff(.)asthelag lengthincreases.Additionally,theparameteridictatestheinfluenceof(n,.z,n,.~,n,-x,.....yt2, yt.~,yt-q, ..., srt.2, sr,- 31 srt-4> ....,) on n,. Thus, a rapiddecaymeansatighterprioronthelaggedvaluesofthevariable. Equivalently,itreducesinfluencesfromlaggedvalues. Theparameterywoulddeterminehowmuchtheothervariables(eg.,y~-ly,~-z,yt.~,...,un~-,u,nt-z,un~couldinfluencen,. A largeryallowsmoreinfluencefromothervariablesinthe 3>..... Srt-l, srt.2, srt-3, ......) inflationequation. Forexample,acombinationofrapiddecayanda smallywouldreducetheroleoflagged valuesbeyondthefirstlag,andatthesametime,reducetheroleoftheothervariables. Hence,thiscombination pushesthemodeltowardsaunivariaterandomwalkspecification. Forfurtherdetaileddescriptions,seepp.8-17 -8-23 ofRATS4.2manual. -9-

hence andthe two-quarters-ahead output growth forecast the model made attimet. In the current estimation, w,= 1,z~= 1for i = 1- 4 and all ms. That is, the objective function includes all variables, and their one- to four-quarter ahead forecast errors are equally weighted. *3 A numerical search procedure was carried out over grids which define six different settings for three hyper-pararneters (eg. there are 63possible combinations) to minimize the objective function.14 The sample period begins in 1985. This is to allow for the fact that a new regime might hale been introduced in 1979with the beginning of the Thatcher aciministratic)n. Five additional >/earsare allowed as an adjustment period. The sample period ends with the UK’sparticipation in the ERM in 1990Q2. In addition to the RMSE, three types of accuracy measures are used in this exercise. They are: (1) mean errors (ME), (2) mean absolute errors (MAE), and (3) Theil’sU-statistics. (1) and (2) together convey information about the tendency of bias in the model’sforecast. Suppose that the ME is negative, and at the same time, the absolute size of the ME is close to the corresponding MAE for the same forecasting periods. This would indicate that the model consistently over-predicts over time (assuming forecast errors are measured as actual minus forecast). The Theil’sU-statistics are used as an indicator of the overa!l goodness of the forecast. This statistic is useful in particular because it offers a unit-free comparison of the model’s *31.e.t,hetwo-quarter-aheadforecasterror: actual(y, t+2) -forecast (y, t, 2). Onecouldsetsome~s to zero,orexcludetheRMSESofa subsetofvariableswhendesigningamodel. Similarly,onecouldchooseaparticularcombinationofforecasthorizon(s)bysettingsomewisequaltozero. 14Theprocesscanbedescribedasfollows:Pickapointonthe63gridofthehyper-parametervalues. Then, one-to four-quarteraheadforecastsaremadeforthesampleperiodfrom1985Q1-90Q2wherethecoefficients aresequentiallyupdatedovertimeafiereachforecast.Theforecasterrorsarecompiledforthewholeforecast period. Thisisrepeatedforallpossiblesettingsoftheh}’per-parametevraluesandforecasterrorsforeach settingofthehyper-parametersarescored. The optimal setting is chosenbyselectingtheonethatisassociated withtheminirnumRMSE. -10-

forecast against a random-walk model based forecast, or a no change forecast. It is calculated as the ratio between the RMSESof the model’sforecast and no-change forecast. Thus, a Theil statistic value greater than one indicates that the model’sforecast is less accurate than that of a random-walk model, and one could do better relying on the no-change forecast. Table 1shows the statistics for the current model. Examinations of different statistics suggest that the specification is reasonable. The mean-errors and mean-absolute-errors together suggest that the specification does not have a consistent over- or under-prediction bias. At the same time, Theil’s U-statistics show that the model forecast is superior to the no-change forecast, with the exception of the pound exchange rate. V. Results Oncethehyper-parameter values are chosen, we generate out-of-sample forecasts starting with 1985Q1. Then the forecast accuracy statistics for rolling ten-quarter intervals are compiled. For example, the first interval started 1985Q1and ended 87Q2. Similarly, the last interval is from 1993Q2 to 95Q3. Weapply this rolling method for both one- and four-quarter-ahead forecasts. For the fourquarter-ahead forecast, the model started forecasting 1984Q1, so the first four-quarter-ahead forecast is for 85Q1. For the purposes of exposition, a ten-quarter interval is treated as the unit interval for measuring forecast accuracy. This allows us several observations belonging exclusively to the IT period.15 The forecast accuracy statistics of these two periods are then compared with the rest of the sample periods, allowing one a ‘smallsample’feel about how close the model’sforecasts are. Furthermore,ifthereisa distinctpatternintheaccuracymeasures,we mightbeableto makean 15NotethattheUnitedKingdomwithdrewfromtheERMinSeptember1992,andinflationtargetingwas introducedthereafter. Accordingly,wehavetwoout-of-sampleobservationsthatareentirelymadeupofIT perioddata;onefor 1993Q1-95Q2,andtheotherfor 1993Q2-95Q3. - 11-

educated guess on its connection to the IT regime that has been in place since 1993. The figures show the forecast accuracy statistics for the rolling ten-quarter horizons. Figures with the suffix A (eg. 3.1.A) show MEs and MAEs on the same panel. For example, Figures 3.1.A and 3.4.A show the mean errors and mean absolute errors of the one-quarter-ahead and four-quarterahead inflation forecasts, respectively. Each point represents the ME and MAE for a ten-quarter interval ending at the date shown on the horizontal axis. Figures with the suffix B show Theil’sUstatistics. For example, Theil’sU-statistic dated 1995Q1represents the statistic calculated from the ten-quarter forecast horizon from the third quarter of 1992through the first quarter of 1995. Figures 3 through 5 respectively show the model’sinflation, short-term interest rate, and the long-term interest rate forecasts. The observation of these three variablesjointly is interesting in that each represents actual inflation outcome, the stance of monetary policy. and inflation expectations and risk premium, in turn. A. The ERM period: 1990Q3- 92Q4 Thereare someperceptibledifferencesacrossthesethreesetsof graphs. First,a marked deteriorationinthe forecastaccuracysetsinat differenttimesacrossvariables. Bothfor inflationand the long-terminterestrate,themodelfirststartsto over-predict,hence the forecast performance noticeably worsens starting sometime around the end of 1988,or the beginning of 1989. However, it was not until the end of 1990,or the beginning of 1991when the performance of the short-rate forecast started to deteriorate. Hence, there is at least a one year gap between the time the forecast performance started to get worse for the three variables. The ERM regime started in the third quarter of 1990. Thus, the behavior of inflation and inflation expectations/premium captured in the long-term interest rate during the period starting in late - 12-“

1988through late 1990could be attributed to an anticipation effect of the onset of the ERM regime.lG That is, once it became likely that the U.K. would participate in the ERM, markets anticipated a continuation of tight monetary policy to support a stronger pound exchange rate. Given that inflation was relatively high during these periods (RPI inflation of 4.4 and 5.7 in 1988and 1989),the nominal rate had to be pushed up to support the real short-term interest rate around 5 percent, which was the level seen in 1987. In fact, the yield curve remained inverted throughout this period as short-term rates were higher than long-term rates since the second quarter of 1988. This, in turn, implied lower future inflation as well as sluggish activity. Both actual inflation and the long-term interest rate thus reflected these, and adjusted even before the actual inauguration of the ERM regime. Since the model did not have this information, however, it persistently over-predicted actual inflation during the ERM period. In addition, there was a surge in RPI inflation in the second quarter of 1990caused by the Gulf crisis. This, in turn, generated a very large forecast error as the model’s forecast was far below the actual. The model took this to be a large unanticipated price shock and hence it introduced an upward bias in inflation forecast for subsequent periods. In terms of the short-term interest rate forecast, the model did not have the same information about the ERM. Hence, the persistent high short-term interest rate in late 1991,or the lack of lowering of rates, in the face of weak output came as a surprise. This explains the under-prediction of the short-rate around 1989-1992,as shown in figures 4.1.A and 4.4.A. B.Inflation Targeting Period: 1993Q1- 95Q3 IG’’StatemenbtsytheChancelloroftheExchequersoonafierthemeetingoftheGroupofSixFinancial MinistersinParis.....gavetheindicationthattheauthoritieswerepursuinganunannouncedexchangeratetarget.” (p.17,PaulTemperton,1990). TherewaswidelyknowndiscordbetweenChancellorLawson(pro-ERM)and PrimeMinisterThatcherwhichledtotheultimateremovalofLawsonfromhispositioninSeptember1989. However,perhapstheneedto findananchortoguidemonetarypolicyandtheimportanceoftheexternalsector totheeconomymighthavebeenperceivedtobemoreoverwhelming. -13-

Next, we turn to the post-ERM IT period. Let us first turn to the short-term interest rate forecast. An examination of Figures 4.1s and 4.4s show that the model consistently over-predicted the actual short-term interest rate. Though the ME and MAEappearto have reached a peak in 1994,the Theil statistics do not show a similarly improving trend. For example, the Theil statistic for the 10quarter interval 1993Q2-95Q3 is 1.484,which is the largest in the whole sample. The same statistic for the interval 1993Q1-95Q2 is the third largest in the whole sample. This becomes particularly obvious in the case of the four-quarter-ahead forecast. The lasttwo observations (for 1993Q1-95Q2 and 1993Q2-95Q3) are the two largest in the whole sample shown in figure 4.4.B. To the extent that this 3-month interest rate represents the stance of monetary policy, the model’sovenvhelming overprediction suggests that monetary conditions may hale been looser than the model would have anticipated, based on past experiences. In contrast, the model’sinflation forecast tnarkedly deteriorated in late 1992and/or early 1993, then improved rapidly. Such a pattern is clearly \’isible in all measures shown in Figure 3s. Particularly, with an exception of the four-quarter-ahead Theil, the inflation forecast noticeably improves in the last two forecast intervals that begin 1993Q1 and 93Q2. This pattern of improvement is in marked contrast to the case of the short-term interest rate. The long-term interest rate also exhibits a similar but much subdued pattern of improved forecast performance in the last two observations. In this case, the improvement is more visible in the four-quarter ahead forecast, though the one-quarter ahead forecast also shows gradual improvement. A comparison of MEs and MAEs in both inflation and the long rate cases indicates that the model’sforecast became less biased, unlike the ERM period that preceded the inflation targeting regime. These observations jointly suggest the following interpretation. To use the model’sshort-term interest rate forecast as a benchmark, monetary policy has not been overly restrictive. That is,the -14-

modelexpectedthe short-termratesto behigherthantheyactuallyturnedoutto be inthepost-1993 sampleperiod. However,despitetheselower-than-expectedconfigurationsofthe shortrates,actual inflationhadconvergedrapidlyto wherethemodelexpecteditto be. Totheextentthatthe long-term interestrateproxiesthe inflationexpectationandinflationriskpremium,themodelalsoover-predicted theseoverthisperiod. Thoughchangesinthecredibilityof monetarypolicyinthis inflationtargeting periodisthe likelyexplanation,otherpossibilitieswarrantourattention. A positivesupplyshockcouldexplainsuchan outcome. Thesterlinghasdepreciatedmoreor lesscontinuously since 1992. Though this is a favorable terms-of-trade shock, it has a definite inflationary implication. For example, the unit value of imports increased by 10and 3.4 percent, respectively in 1993and 1994. At the same time, there has been no evidence of extraordinarily favorable price shocks. In fact, the producer price index for input factors rose 4.7, 2.9, and 9.4 percent respectively for 1993, 1994,and 1995. Another possibility is that there was a favorable inflation environment in the form of low wage pressures during this period. Indeed, there have been few perceptible pressures on wages and unitlabor costs in the recent period, even with the robust activity seen in 1994,for example. The pace of growth in average earnings slowed to around 3-3/4 percent (from about 6 to 7 percent) in the lastthree years. This moderation in wage pressures could be attributed to cyclical as well as structural factors. The official claimant-count based unemployment rate has declined noticeably since the 1990-92 recession. However, the labor force participation rate has not increased proportionately, suggesting some residual slack in the labor markets. In addition, a large scale privatization of public corporations and a weakening labor union have been important changes British labor markets since the early 1980s. These developments affected patterns of wage settlements and hence should have influenced wage behavior in recent periods. However, a low inflation environment and increased credibility ofa low inflation monetary regime must have been factors contributing to such wage behavior. Workers would -15-

settle for a smaller rise in nominal wages if they expect slower erosion of the purchasing power of their nominal wages over the contract period, ceteris paribus. Hence, a lack of wage inflation can not be an independent explanation of the observed changes in the forecast performance pattern since 1990. Improvement in the effectiveness of monetary policy still remains a likely explanation. That is, despite a monetary policy stance that has not been as tight as the model would suggest, inflation has remained close to the model’sforecast. The model also expected a higher long-term interest rate (larger inflation expectations and risk premium). The UK monetary policy has become more effective in the sense that it has taken less tightening to obtain a favorable inflation outcome. This would not have been possible if markets fully discounted the credibility of the new IT regime. On the other hand, a moderating inflation trend has not been unique to the UK but has been seen in many OECD countries in the 1990s. Presumably, the recession that visited major G-7 countries at the beginning of the 1990scould partly explain this observation. However, the moderating trend has continued even when most of these economies moved well into recovery phases. This raises the possibility that the earlier finding of the mis-match between inflation and interest rates might not be unique to the British economy and particularly has little to do with the institution of inflation targeting. This possibility is examined in the next section. C. Cross country comparison: France and the US Thissectionexaminesresultsfromthe identicalexercisesrepeatedforthe keyG-7countries thathavenotadoptedan explicitIT regime,namely,FranceandtheUS.17Panelsin Figure 6 show *’Quarterlydatafrom1972-84Q4wereusedfortheinitialestimation,1985Q1-90Q2forthehyper-parameter estimation,and 1990Q3-95Q3fortheout-of-sampleforecast. Thesamesetofsixvariablesareused;growthin realGDP,inflationinconsumerpriceindexes,short-terminterestrates(one-monthParisinterbankmoneymarket rateforFrance,and3-monthT-billforUS),long-terminterestrates(long-termbellwetherbondyieldforFrance, and 10-yearrateforUS),trade-weightedexchangerates,andunemploymentrates. Parameterandweightsetups -16-

the inflation forecast errors for France and the US models respectively. The comparable figures from the British model are shown as dotted lines in all graphs to facilitate a direct comparison. In general, forecast errors for the two economies are smaller and less erratic. There isno discerniblebias tendencyin a one-quarter-forecasthorizon. Overa four-quarter-forecasthorizon,themodel’sforecast performanceforFrancesomewhattemporarilyworsenedintheearly 1990s,butotherwiseno clear trendcanbefound. Interestingly,accordingto theTheil’sstatisticsforthefour-quarter-aheadforecast, the USmodelhasdistinctlybeenover-predictingactualinflationsince1993intermsof theTheil statistics. ThiscorroborateswellwiththeperceptionthatinflationintheUS hasbecomeunusually well-behavedintherecentperiod. Panelsin Figure 7 show forecast errors for the short-term interest rate. For France, rising MAE and a falling ME pattern seen in a one-quarter-ahead forecast suggests that the model tended to over-predict short-term interest rates since 1993. However, this does not suggest a significant bias as patterns in both four-quarter-ahead MAE and ME and Theil statistics do not indicate such a tendency. On the other hand, both MAE and ME have been approaching the horizontal line from above and below in the recent period. This suggests that errors are small and evenly distributed between overand under-prediction. This generally improving trend is also reflected in Theil statistics. Interestingly, the pattern of Theil statistics for both France and the US show a markedly improving trend since 1994, in contrast to that seen in the UK. Panels in Figure 8 show forecast errors for the long-term interest rate. No particularly discernible patterns can be seen for France or the US. They are relatively more well-behaved in comparison to those for the UK. In general, forecast errors from models for France and the US tend to be more well behaved forhyper-parameterestimations--RMSEminimizationproceduredescribedin‘Estimation’ section--are identical tothosefortheUKcase. -17-

and smaller in absolute size since 1993. This suggests that the relative fit of the UK model is worse than those for the two other countries. Despite the similarity seen between most G-7 countries by way of low inflation, a more systematic comparison points to some perceptible differences between the inflation targeting UK and non-inflation targeting France and US. To summarize, the recent mild inflation seen in the latter two economies was not unusual in light of their experiences since the mid-1980s. However, in the case of the UK, it has been unusual. That is, the dynamic economic relationship of the 1980scaptured by the model can not explain the recent inflation behavior seen since the adoption of the IT regime.18 Conclusion Thispaperexaminedthe UK’sexperience with the IT monetary regime that started in 1993 and finds evidence that the regime has had some measurable effect on how monetary po icy and inflation interact. That is, despite a monetary policy stance that has not been as tight as the model would suggest, inflation remained close to the model’sforecast. It took less monetary tightening to obtain a favorable inflation outcome. This might be reflecting extant credibility effect of IT as markets expect future monetary policy to be conducted along a path compatible with maintaining lower inflation. Identical exercises were repeated for France and the US, countries that have not adopted IT but ha~’eexperienced low inflation in the recent period. Results show that, unlike the UK’s case, recent 18ThereexistsomeinterestingdifferenceswithinITcountries. TheUK’sITregimediffersfromthoseof othercountriesinthatitspecifiesonlythegoaltobeachieved. NewZealandandCanadanotordyspeci~ their explicitgoals,butalsospeci~ anexplicitpenaltyforfailure(NewZealand),andgrantagreatdealmore autonomytothecentralbanks(NewZealand,Canada). Thus,insomesense,theUK’sarrangementisless binding. Suchdifferencesnotwithstandingt,hispaper’sfindingindicatesanenhancedcredibilityofmonetary policy. Hence,perhapsthefactthatIToffersanobjectiveandexplicityardstickthatmonetaryauthorities’ performancecanbeheldtoisthekey. Intheeventoffailuretomeetthegoal,reactionsbypoliticiansand financialmarketscoulddealsevererepercussionstopolicymakers. Forexample,anincreaseinthe government’sfundingcostwouldbeoneconsequence. -18-

low inflation in these economies has not been unusual when compared to forecasts from models ~D•dˆesigned to fit the second half of the 1980s. That is, given the level of inflation, the degree of actual monetary policy tightness (measured in terms of short-term interest rate) is about what the model expects. Despitethefindingsofthisexercise,however,indicatorsof long-terminflationexpectations havenot been unanimous or unambiguous in pointing to low future inflationinthe 1990s. Survey measures of inflation expectations also have declined at a glacial pace. These suggest that establishing the monetary policy credibility over the long-term horizon is a highly costly commodity. -19-

References Ammer,J., andR. Freeman (1995), “Inflation Targets in the 1990s:The Experiences of new Zealand, Canada, and the United Kingdom,” ,Journalof Economics and Business 47 pp. 165-192, Bowen, A. (1995), “Inflation Targetry in the United Kingdom,” in Haldane, A. G. ed. Targeting Znjla/ionBank of England, London. Doan, T., R. Litterman, and C. A. Sims (1984), “Forecasting and Conditional Projection Using Realistic Prior Distributions,” Econometric Reviews vol. 3, pp. 1-100. Freeman, R., and J. Willis (1995), “Targeting Inflation in the 1990s:Recent Challenges,” FRB International Finance Discussion Paper No.525. Fischer, S. (1994), “Modern Central Banking,” paper presented at the Bank of England Tercentenary conference. Haldane, A. G. ed. (1995) Targe/ing l?lflationBank of England, London. Hall, R., and N. Gregory Mankiw (1994), “Nominal Income Targeting,” in N. Gregory Mankiw, cd., Mone/ary PoZic}~U, niversity of Chicago Press, Chicago. Inflation Reports (1996), The Bank of England. August. King, M. A.(1994), “Monetary Policy in the UK,” Fiscal Studies vol. 15,no.3, pp.109-128. Leiderman. L., and L. Svensson eds.(1995), Inflation Targets, CEPR, London. Litterman, R. (1981), “A Bayesian Procedure for Forecasting with Vector Autoregressions,” FRB of Minneapolis Working Paper. Mayes, D., and Bryan Chapple (1995), “The Costs and Benefits of Disinflation: a Critique of the Sacrifice Ratio,” ~eserve Bank Bulletin, Reserve Bank of New Zealand, pp. 9-21. McCallum, B. (1995), “Inflation Targeting in Canada, New Zealand, Sweden, the United Kingdom, and in General,” mimeo, Carnegie Mellon University. Sims, C. A. (1980), “Macroeconomics and Reality,” Econometric , vol. 48, pp. 1-49. Svensson, Lars (1996), “Inflation Forecast Targeting: Implementing and Monitoring Inflation Targets,” mimeo, Institute for International Economic Studies, Stockholm University. (1996), “Optimal Inflation Targets, ‘Conservative Central Banks, and Linear Inflation Contracts,” NBER Working Paper No. 5251. (1993), “The Simplest Test of Target Credibility,” NBER Working Paper No.4604. Temperton, P. (1991), UK A40ne/aryPoZicy,St. Martin’sPress. -20-

Theil, Henri (1972), Principles of Econometrics. New York: Wiley Woodford,M.(1994),“NonstandardIndicatorsforMonetaryPolicy:CanTheirUsefulnessbeJudged fromForecastingRegressions?”inN. GregoryMankiw,cd.,Monetary Policy, University of Chicago Press, Chicago. -21-

Table 1: One- to four-quarter ahead Forecast Error Statistics(1985Q1-90Q2) Variables Forecast Mean Errors Mean Root Mean Theil’sUhorizon Absolute Squared Statistics Errors Errors output l-quarter 0.878 2.310 2.851 0.829 growth ~) 4-quarter 0.729 2.107 2.499 0.766 Inflation in l-quarter -0.131 1.233 1.560 0.800 RPIX (n) 4-quarter -0.366 1.123 1.514 0.608 Unemploym- l-quarter -0.009 0.013 0.016 0.379 ent (un) 4-quarter -0.098 0.106 0.122 0.712 Exchange l-quarter -0.013 0.032 0.041 1.088 Rate (ex) 4-quarter -0.058 0.071 0.091 1.230 Short-term l-quarter 0.016 0.097 0.123 1.107 interest (sr) 4-quarter 0.022 0.162 0.182 0.891 Long-term l-quarter -0.003 0.037 0.046 0.882 interest (/r) 4-quarter -0.028 0.055 0.063 0.784 1.Thereare22 and 19observationsrespectivelyfor 1and4 quarteraheadforecasts. -22-

Figure1: DataPlot Output Growth Inflation Pound(weightedaverage) Unemploymentrate 12 10 8 6 4 1985 1990 1995 Short-termInterestrate Long-termInterestrate 1995 “1985 1990 1995 Note: The vertical linesdenote the beginningandthe end ofUKparticipationinthe ERM. The secondlinealsodenotes the beginning ofUKinflationtargeting. 23

i Figure 2: Residuals FromModelEstimated72Q -90Q2 FromModelEstimated72Q’-92Q3 Inflation Inflation 10 8 6 4 . 4 2 2 0 r... d 0 -2 -2 ....... 4 -4 -6 -fjL > ) 1995 1975 1980 1985 1990 Short-term Interest Rate Short-te~ InterestRate 0.4 0.4 . 4 ,... 0.2 0.2 ...... ----- -0.0 -0.0 -.. ..... -0.2 -0.2 -0.4 -0.4I -0.6 , 1 75 1980 1985 199~0 -0.6 1975 1980 1985 1990 1995 ..- Long-termInterestRate Long-termInterestRate 0.2, o.: 1 .................................... ...... 0.1 0.1 ! . . . . . . . . . . . . . . J -0.0 Ahn -0.( -0.1 -0.1 ------------------------------------ -0.2 -0.2 -0.3 , -0.3 -1 -0.4 1975 1980 1985 1990 1995 -0.4 75 1980 1985 1990 1995 24

m -0-l - *a -0 - m m -m -Nm -a m -0 -0-) a -m m 0 \ -1 ,\ 25

- N - (mn 7 -1 mm*woN~w.m. O“OO”O”OO” 000 26

r r , 1, > , - *m 51 m ‘1 i- 1 i- 1- I I 4- 1 * 1 , , 9 I t m 0-) m m m g m 0 b

i Figure6 Inflation: France(Dottedlineforthe UK) ME,MAEforlquarteraheadforecast THEILforIquarteraheadfore=st 2.0 4 7 --- - . . . .‘. # f . . . -.. 3 , , .. . o ‘, 2 1 . .# , . l MAE “ -•. . ‘. . .. 1.5 % , , , , * ,= # : ‘, J, ,---- .-., # # , . .-. ... 0 1.0 \ t ,0 . -1 . .. .. , . ...“- ‘. l., ---- ..... -2 ‘s, ME ,.-” “ 0.5 ‘..- , -3 -.. ,. # 0 ‘,, -4 987 ! 1988 I 1989 I 1990 1 1991 1 1992 I 1993 I 1994 1 199 0.0 987 I 1988 I 1989 t 1990 I 1991 1 1992 I 1993 1 1994 I 199 1 5 ME,MAEfor4-quarteraheadfore=st WEILfor4quarteraheadforeoast 6 --- -.. 4 2 .. , .. ‘. ‘. ‘. . ‘- -. 1.5 , , $ , . . t , , , , , l , . 0 . ------- ---- - - . 1.0 , * , * * , , , , , , .-. . , ,* -2 , ,* . -4 I . . l-, ,.-.. #..- ... I 0.5 I , , ., , , , , 00 ~ -6 I - -‘ . ~- ‘ ‘ ‘ “‘ s‘ ‘ 1- ‘ s‘ ‘ a‘ 1‘ s‘ ‘ ‘ ‘ ‘ ‘‘ ‘ 1 1987 1988 1989 1990 1991 1992 1993 1994 1995 1987 1988 1989 1990 1991 1992 1993 1994 1995 us ME,MAEforlquarter aheadforecast THEILforIquarteraheadforecast 4- ---- ---- . ,’.-.-.. 3 2 1 . . ~ -- 1 ------ ------ #. .. , , MA; .’. -. .. .- . . 1.5 . # , , . ,, .“* .-. .-. -’ , # #d # ,, , . ‘. 0 .“.-. - . 1.0 , -. -1 . . ‘. ..-------- ‘1 -2 , ‘.. ME .- -“’ 0.5 - - 4 3 1 . 987 1 1988 I 1989 1 1990 I 19 . 9 -. 1 [ - 1 - 9 - 9 . ‘ 2 . I .“ 1 . 9 # 93 1 1994 I 1995 0.0 1 I t 9“8‘7 ‘ 1‘9“88“ ‘ 1‘9 ‘ 8 a 9 ‘ 1 ‘ 9 ‘ 9 ‘ 0 ‘ 1 ‘ 9 ‘ 9 ‘ 1 ‘ 1 ‘ 9 ‘ 9 ‘ 2 ‘ 1 ‘ 9 ‘ 9 ‘ 3 ‘ 1 ‘ 9 “ 9 ‘ 4 ‘ 1 ‘ 9 ‘ 9 ‘ 5 ME,MAEfor4quarteraheadforecast THEILfor4quarteraheadforecast #, . # .,.-. 1 3.5r 1 3.0 ~O’iAE .’-*. “1~ 2 . ------- .-.--.- . .“ .“. ‘. . -. ----- 2 2 . . 5 0 0 -. . l l l. .- -.” -- 1.5 -2 . ME l 8“, .“ . 1.0 -4 , . l -, * ... .. # 0.5 ~1 ., -6 19 1 87 ’’’ 1 ’ 9 ” 8 ” 8 ’’ 1 ’ 9 ’ 8 ’’ 9 ’” 1 ’ 9 ” 90 19 ‘ 9 “ 1 ’”- 1 .” 9 .’ 9 ”’ 2 ’’ 1 ’’ 9 ’ 9 ’ 3 ’’’ 1 ” 994 1995 0.0 1 1 9“8‘7 ‘ 1‘9‘8‘8 ‘ 1‘9‘8“9 ‘ 1 ‘ 9 ‘ 9 “ 0 ‘ 1 “ 9 ‘ 9 “ 1 ‘ 1 ‘ 9 a 9 ‘ 2 ‘ 1 ‘ 9 ‘ 9 ‘ 3 ‘ 1 ‘ 9 ‘ 9 ‘ 4 ‘ 1 “ 9 ‘ 9 J 5 ME= MeanErrors MAE=MeanAbaoluteErrors THEIL=Theil”sU-Statistic 28

Figure7 Short-termInterestRate:France(Dottedlineforthe UK) ME,MAEfor1-quarteraheadforecast THEILforlquarteraheadforecast 0.2 2.0 1 [ -l 0.1.’‘--------------- 1.5 , --. . ----- , -- ., # 1.0 , , ,’ -0.1 ME ..... . .---- ,,-” 0.5 ‘. ..I -0.2 1987 I 1988 I 1989 I 1990 1 1991 1 1992 I 1993 I 1994 1 1995 0.0 1987 I 1988 I 1989 1 1990 I 1991I19921199311994I1995 ME,MAEfor4-quarteraheadforecast THEILfor4-quartearheadforecast 1.0- ,.1 0.5 . MAE . .,.. ---- ... . 2.0 . ,’ . 0.0 . ‘ ‘- ..- , . 1.5 . ,,. . ,, ? , , , r . , , ,, -“, ., , # # 1.0 - ““ ‘, -0.5 ME . . . ... .-. ” . ... 0.5 - . >.’ . . ‘.-.’ , . v . * , ,, , , --- -1.0 1987 I 1988 I 1989 I 1990 1 1991 t 1992 t 1993I1994I1995 1.1...1.,.,,.,,,,,,,,.,,.,,.,,,,1 0“01987 1988 1989 1990 1991 1992 1993 1994 1995 us ME,MAEfor1-quarteraheadforecast THEILforlqumer aheadforecast 0.2 2.0 , 1 0.1 ----------- -.----- ..-. . 1.5 . 0.0 ~---- # -. ... - . ---- 1.0 .— --- - ---- . . - . ,’ . ‘, . , ,’ ‘-., -0.1 0.5 ---- -0.2 387 1 1988 1 1989 1 1990 I 19911199211993I199411995 0.0-“’’’’’’,1,,.1.. . 1...1.,.1,1.,,, 1987 1988 1989 1990 1991 1992 1993 1994 1995 ME,MAEfor4-quarteraheadforecast THEILfor4-qutieraheadforecast 2.5 --- 1 ,.. 0.5 . 2.0 # . 0.0 - - - - - - - - - -. . .. /- 1.5 . . , , S 1 ,$ .‘‘t , / ,’ 1.0 >.’ ,, -0.5 . ME . . I --- ..” I 0.5 , ,1 - . .’ -1.0 1987 1 1988 1 1989 1 1990 t 1991 1 1992 I 1993 1 1994 1 1995 0.0 1 I 9 ‘ 8‘7 ‘ 1‘9‘8,8 J 1 , 9 , 89 I 1 , 9 . 9 . 0 I 1 L 9s9e1I1.9.9.2I1,9,9,3,1,9.9,4,1,9,9 I 5 ME. MeanErrors MAE. MeanAbsoluteErrors THEIL=Theil’sU-Statistic 29

Figure8 Long-termInterestRate:France(Dottedlineforthe UK) ME,MAEforl-quarteraheadforecast WEILforIquarteraheadfore~st 0.15 2.0 ,$ ,, 0.10[ MAE * , * * 0.05 .-- ..- --..- ‘- .-------- 1.5 .-” , , $ $ ,, ... 0.00 .. A ---- A. , . 1.0 1 .=- , # #, . . , , ----l.” .. -..------.--.,,J.-. ----- .- -0.05 ..’.-* -0.10I 0.5 . ME -o.I5 1 L 9 s 8 ~ 7 I 1 s 9 ~ 8 ~ 8 I 1 ~ 9 D 8 •, 9 ˆI 1 t 9 , 9 , 0 I 1 = 9 . 9 ~ 1 êIT 19 (cid:129)8. 9 •Ã. 2 (cid:129)I1,9.9.3I1.9.9,4I1,9.9j5 0.0 1987 I 1988 1 198911990I1991I1992I1993I199411W5 ME,MAEfor4-quarteraheadfore~st THEILfor4-qumer aheadforecast 1.0 2.5 1 ,, 0.5[ MAE 2.0 . , , , ,, 4#C .U 1 I ~ . t, , , , , , . ,#. . 1- ‘ ~oo / ,-- -“ I -. --- “ . .,“ w . -0.5 . ME J 0.5 ‘--- --- . . { -1.0 1 “ 9 “ 8 ’ 7 ”” 1 ’ 9 ’ 88 .. 19 ’ 8 .. 9 .1 1 . 990 19 . 91 .1 1 . 9 . 9 . 2 1 1 . 9 . 9 . 3 1. 1 .. 9 , 9 .. 4 < 1995 0.0 1 ~ 9“8 ‘ 7 ‘ 1‘9.8 n 8 ‘ 1 . 9 ~ 8 ~ 9 I 19 ~ 9 ~ 0I1.9.9.1 I 1 . 9 . 9 . 2!1.9.9.3, 1.9,9,4,1.9,9j5 us ME,MAEforl-quaneraheadfore~st THEILforl-quarteraheadforecast 0.15 2.0 r .. 0.10 MAE * 1 , , 0.05 --- - .. --- - .------ 1.5 . ..” , ,, , ,, ... -0.00[ 1 . ~ . * ... ’ . ------ ~ --- .. ---- I 1.0 ~~ , , , , ----- -0.05 ‘------ ‘.- -- , ..----- 1 ------ ‘.l .-. ----- --, ‘.., 0.5 -0.10 ME ‘0”’51987 1 1988 I 1989 I 1990 1 1991 1 1992 1 1993 I 1994 I 1995 0.0 1 ~ 1 9“8 s 7 ‘ 1‘9 n 88 . I 1 , 9 , 8 . 9 1. 19 . 90 . I 1 . 9 . 9 . 111t9.9.2I1,9,9,3I1.9,9,4,1,9,]9 { 5 ME,MAEfor4-quarteraheadforecast THEILfor4-quarteraheadforecast 1.0 I ------. 0.5 MAE 2.0t .’: 1I 7– = 1 I 0..::~:::- . . . ~ --- - - -: - -:------.---:- --- - - - ..-- - - . . 1.5 . ..-. , , , 0 c ! ,,11 1 1 ,’ ,’ ,“. , 1.0 ~ &#45;&#45;&#45; -0.5 t ME 0.5 . * ‘. ------ -. . , . I 1 -1.0 1 I 9 ‘ 8 ‘ 7 ‘ 1 “ 9 ‘ 8 a 8 ‘ 1 , 9 . 8 . 9 1 1 ~ 9 , 9 , 0 I 1 , 9 . 9 . 1 I 1 . 9 . 9 . 2 1 1 . 9 . 9 . 3 , 1 , 9 . 9 . 4 , 1 . 9 . 9 J 5 0.0 1 1 1 I t 1 I t 1987 1988 1989 1990 1991 1992 1993 1994 1995 ME= MeanErrors MAE. MeanAbsoluteErrors THEIL=Theil’sU-Statistic 30

International Finance Discussion Papers IFDP Number Titles Author(s) 1996 565 SomeEvidenceontheEfficacyoftheUKInflation Chan Huh TargetingRegime: An Out-of-SampleForecast Approach 564 TheUseoftheParallelMarketRateasa Guide Nita Ghei to SettingtheOfficialExchangeRate Steven B. Kamin 563 CountryFundDiscountsandtheMexicanCrisisof Jeffrey A. Frankel December1994: DidLocalResidentsTurn Sergio L. Schmukler Pessimistic Before International Investors? 562 EasternEuropeanExportPerformanceduring Nathan Sheets theTransition Simona Boata 561 Inflation-Adjusted Potential Output Jane T. Haltmaier 560 The Management of Financial Risks at German Allen B. Frankel Nonfinancial Firms: The Case of Metallgesellschafi David E. Palmer 559 Broad Money Demand and Financial Liberalization Neil R. Ericsson in Greece Sunil Sharrna 558 StockholdingBehaviorof U.S. Households:Evidence Carol C. Bertaut fromthe 1983-89Surveyof ConsumerFinances 557 Firm Size and the Impact of Profit-Margin Uncertainty Vivek Ghosal on Investment: Do Financing Constraints Play a Role? Prakash Loungani 556 Regulation and the Cost of Capital in Japan: A Case John Ammer Study Michael S. Gibson 555 The Sovereignty Option: The Quebec Referendum and Michael P. Leahy Market Views on the Canadian Dollar Charles P. Thomas 554 Real Exchange Rates and Inflation in Exchange-Rate Steven B. Kamin Based Stabilizations: An Empirical Examination 553 Macroeconomic State Variables as Determinants John Ammer of Asset Price Covariances Please addressrequestsforcopiesto InternationalFinanceDiscussionPapers,Division of InternationalFinance,Stop24,Boardof GovernorsoftheFederalReserveSystem, Washington,DC 20551, 31

International Finance Discussion Papers IFDP Number Titles Author(s) 1996 552 The Tequila Effect: Theory and Evidence from Martin Uribe Argentina 551 The Accumulation of Human Capital: Alternative Murat F. Iyigun Methods and Why They Matter Ann L. Owen 550 Alternatives in Human Capital Accumulation: Murat F. Iyigun Implications for Economic Growth Ann L. Owen 549 More Evidence on the Link between Bank Michael S. Gibson Health and Investment in Japan 548 The Syndrome of Exchange-Rate-Based Enrique G. Mendoza Stabilization and the Uncertain Duration of Martin Uribe Currency Pegs 547 German Unification: What Have We Learned Joseph E. Gagnon from Multi-Country Models? Paul R. Masson Warwick J. McKibbin 546 Returns to Scale in U.S. Production: Estimates Susanto Basu and Implications John G. Fernald 545 Mexico’s Balance-of-Payments Crisis: A Chronicle Guillermo A. Calvo of Death Foretold Enrique G. Mendoza 544 The Twin Crises: The Causes of Banking and Graciela L. Kaminsky Balance-of-Payments Problems Carmen M. Reinhart 543 High Real Interest Rates in the Afierrnath of Graciela L. Kaminsky Disinflation: Is it a Lack of Credibility? Leonardo Leiderman 542 Precautionary Portfolio Behavior from a Life-Cycle Carol C. Bertaut Perspective Michael Haliassos 541 Using Options Prices to Infer PDF’s for Asset Prices: William R. Melick An Application to Oil Prices During the Gulf Crisis Charles P. Thomas 540 Monetary Policy in the End-Game to Exchange-Rate Steven B. Kamin Based Stabilizations: The Case of Mexico John H. Rogers 539 Comparing the Welfare Costsand the Initial Dynamics Martin Uribe of Alternative Temporary Stabilization Policies 32

International Finance Discussion Papers IFDP Number Titles Author(s] 1996 538 LongMemoryinInflationExpectations:Evidence Joseph E. Gagnon fromInternationalFinancialMarkets 537 Using Measures of Expectations to Identi& the AlIan D. Brunner Effects of a Monetary Policy Shock 536 Regime Switching in the Dynamic Relationship Chan Huh between the Federal Funds Rate and Innovations in Nonborrowed Reserves 535 The Risksand Implications of External Financial Edwin M. Truman Shocks: Lessonsfrom Mexico 534 Currency Crashes in Emerging Markets: An Jeffrey A. Frankel Empirical Treatment Andrew K. Rose 533 Regional Patterns in the Law of One Price: The CharlesEngel Roles of Geography Vs. Currencies JohnH. Rogers 33