Establishing Credibility: Evolving Perceptions of the European Central Bank

Transcription

1 Esablishing Credibiliy: Evolving Percepions of he European Cenral Bank Linda S. Goldberg! Federal Reserve Bank of New York and NBER Michael W. Klein Flecher School, Tufs Universiy and N.B.E.R. March 4, 2005 Absrac: The credibiliy of a cenral bank s ani-inflaion sance, a key deerminan of is success, may reflec insiuional srucure or, more dynamically, he hisory of policy decisions. The firs years of he European Cenral Bank (ECB) provide a naural experimen for considering wheher, and how, cenral bank credibiliy evolves. n his paper, we firs presen a model demonsraing how he highfrequency response of asse prices o news reflecs marke percepions of he aniinflaion sance of a cenral bank. Empirical ess of his model, regressing boh he change in he slope of he German yield curve and he change in he euro/dollar exchange rae on he surprise componen of Unied Saes price news, sugges significan breaks in he marke s percepion of he policy sance of he ECB during is firs five years of operaion. The daes of hese breaks are linked o he policies underaken by he ECB. Similar ess on he response of he slope of he Unied Saes yield curve o price news during his period fail o find comparable breaks in marke percepions of he conduc of moneary policy by he Federal Reserve. Keywords: Cenral Banking, European Cenral Bank, Federal Reserve, inflaion, exchange rae, credibiliy, yield curve JEL Classificaion: F3, E5, E6! The views expressed in his paper are hose of he individual auhors and do no necessarily reflec he posiion of he Federal Reserve Bank of New York or he Federal Reserve Sysem. Address correspondences o Linda S. Goldberg, Federal Reserve Bank of NY, Research Deparmen, 33 Libery S, New York, N.Y Linda.Goldberg@ny.frb.org, or Michael W. Klein, Flecher School, Tufs Universiy, Medford MA michael.klein@ufs.edu. The auhors hank Dia Ganguly and Eleanor Dillon for excellen research assisance

2 Esablishing Credibiliy, Goldberg and Klein. nroducion The credibiliy of he ani-inflaion sance of a cenral bank plays a key role in deermining wheher he goal of low inflaion is aained. This poin is, by now, a sandard heoreical resul. 1 is also received wisdom among praciioners. n a survey of he heads of 84 cenral banks, as well as 52 prominen academic moneary economiss, Blinder (2000) finds ha credibiliy is considered vially imporan and helps keep inflaion low. This consensus on he imporance of credibiliy naurally leads o he quesion of how i is achieved, and wheher and how i evolves over ime. One view is ha esablishing an appropriae insiuional srucure is he key elemen in aaining credibiliy. A second, more dynamic, view focuses on he role ha acual policy conduc plays in building he repuaion of a cenral bank. These wo differen views have disinc implicaions for he relaive imporance of he srucure of a cenral bank as compared o is conduc for aaining and mainaining is credibiliy. A majoriy of respondens o Blinder s survey believe ha cenral bank credibiliy is based more is hisory of acions han on he consrucion of insiuional srucures ha insulae a cenral bank from poliical concerns and afford i independence. Noneheless, here is also a consensus among respondens ha srucure maers. This laer view is consisen wih empirical research ha has found, in cross secions of counries, ha insiuional srucure is associaed wih economic performance, perhaps because i indicaes he abiliy of an insiuion o ie is hands and commi o a policy ha may cause shorerm pain in he pursui of longer-run gain. 2 There is less evidence, however, on wheher and how he credibiliy of a paricular cenral bank evolves over ime in response o he conduc of policy. The quesion of he achievemen and he mainenance of credibiliy is especially relevan for a new cenral bank. An analysis of he experience of he European Cenral Band 1 Seminal conribuions on he role of credibiliy includes Kydland and Presco (1977), Calvo (1978) and Barro and Gordon (1983). 2 For example, Cukierman (1992) analyzes he charers of cenral banks and shows, in a cross-counry panel, ha average inflaion is lower in counries in which laws afford cenral banks greaer independence. Alesina and Summers (1993) also find cross-counry evidence ha he level of inflaion, as well as is variabiliy, is negaively associaed wih indicaors of cenral bank independence, bu here is no associaion beween cenral bank independence and real variables. Quesions have been raised, however, abou wheher he de jure srucure is closely linked o he de faco behavior of insiuions (Forder 1999). 1

3 Esablishing Credibiliy, Goldberg and Klein (ECB) during is early years of operaion provides a naural experimen for considering his quesion. The archiecs of he insiuional srucure of he ECB were mindful of lessons from economic heory concerning he imporance of independence from poliical consideraions. 3 The role of conduc was also clearly apparen. As indicaed by he survey resuls in Blinder (2000), he direcors of cenral banks are vially aware ha heir policies are closely scruinized for indicaions of general endencies. This may be especially rue wih a new cenral bank where each policy choice can lead o a larger updaing of marke priors han would be he case for a long-esablished cenral bank. n Secion, we presen a framework for a novel es of he evoluion of marke percepions of cenral bank policy. This es uses high frequency asse price daa and he surprise componens of economic news announcemens o esimae wheher he marke percepion of he ani-inflaion credibiliy of a cenral bank changes over ime. 4 The key insigh from his model is ha a given surprise increase in inflaionary pressures will resul in a greaer increase in a long ineres rae relaive o a shor ineres rae, and a larger exchange rae depreciaion, when a cenral bank is perceived as having a weaker ani-inflaionary policy sance. f unvarying insiuional srucure is he dominan deerminan of a new cenral bank s credibiliy, hen one would no expec o find a change in he relaionship beween news and asse prices over ime. Bu if credibiliy for a new cenral bank is earned hrough he conduc of is policy, one would find a significan break in he relaionship beween news and asse prices as credibiliy evolves over ime. 5 3 Despie hese lessons, some poliicians coninued o ry o influence policy direcion. For example, Oscar Lafonaine, appoined Finance Miniser of Germany in he Auumn of 1998, called for he new ECB o lower ineres raes from he ime of his appoinmen unil his resignaion in March n response, Wim Duisenberg, he firs presiden of he ECB, saed in November 1998 ha i was a normal phenomenon for poliicians o offer heir views on he conduc of moneary policy, bu i would be very abnormal if hose suggesions were o be lisened o. See Wim Duisenberg, Banker o a New Europe, The Economis, November 26, Forward marke informaion has been used in oher ess of policy regime credibiliy. For example, Svensson (1991) shows ha forward exchange raes were no wihin he arge zone band of he European Moneary Sysem (EMS) in he 1980s, a resul he inerpres as indicaing ha he EMS generally did no offer credible bands on is members currencies. Svensson (1993) presens a similar se of ess o deermine wheher he inflaion arges of Canada, New Zealand and Sweden were consisen wih marke yields. These ess, while informaive, require he presence of an explici arge, like an exchange rae band or an inflaion arge, o judge credibiliy. Oher relaed empirical analyses on he policy credibiliy of an exchange rae arge zone use inervenion daa o esimae perceived arge zone bands (Klein and Lewis 1993 and Lewis 1995). 5 Klein, Mizrach and Murphy (1991) develop a similar ype of analysis concerning differences in he responsiveness of asse prices o news as policy evolves in heir sudy of he changing responsiveness of dollar exchange raes o news abou he Unied Saes curren accoun. They find he 1985 Plaza Accord alered percepions of he degree o which American policy was concerned wih he U.S. curren accoun defici. 2

4 Esablishing Credibiliy, Goldberg and Klein n Secion we apply his es o sudy he evoluion of he credibiliy of he European Cenral Bank from he ime i began is operaions in January 1999 hrough mid We find evidence ha he marke s percepion of he ani-inflaion credibiliy of he ECB evolved over ime and responded o is policy acions. As a benchmark for our analysis, and also o idenify wheher he resuls we found for he ECB could be aribued o changes in he economic environmen raher han in specific views of is credibiliy, we also es for changes in he marke s percepion of he ani-inflaion sance of he Federal Reserve over he same sample period. n conras o our resuls for he ECB, we find no evidence of changing percepions of he policy sance of he Federal Reserve, a resul ha is no surprising given he Fed s long-sanding commimen o price sabiliy under he chairmanship of boh Alan Greenspan and Paul Volker.. Cenral Bank Policy and Marke Responses o News n his secion we presen a model ha shows how changes in percepions abou a policy sance can aler he response of asse prices o news. This is based on a modificaion of he sandard framework for undersanding he response of an asse price o news in which q q +, he change in an asse price beween ime -, jus before an announcemen, and ime +, jus afer ha announcemen, is a linear funcion of news, ha is, he difference beween he announced value of a variable, x +, and he expeced value of ha variable a ime -, E x +. The sandard empirical specificaion is q + = α + β + (1) ( + + ) + q where ε + is a whie-noise error erm. This parsimonious specificaion is mos appropriae when he ime horizon is beween + and - is shor, for example, when i is measured in minues raher han days, and when news abou he variable x does no become available a he same ime (ha is, wihin he span - o + ) as announcemens abou some oher relevan variable. The acual se of variables ha consiue x depends upon he asse sudied bu, in general, any variable ha markes consrue as revealing informaion abou curren and fuure economic aciviy may be appropriae for sudy. A more general version of equaion (1) akes ino accoun expecaions abou he policy response o news. x E x ε 3

5 Esablishing Credibiliy, Goldberg and Klein q q + = α + β + φ + (2) ( + + ) ( + + ) + n his formulaion, he erm ( M E M + ) x + E x M φ represens he manner in which he perceived difference beween acual policy pah afer he announcemen, E M ε M +, and is expeced pah before he announcemen occurs, E + affecs he change in he price of he asse. The M parameer φ may be posiive or negaive, depending upon he policy and he asse. The inclusion of ( E M + ) M reflecs he presence of a perceived policy reacion funcion in + which M depends upon x, such as (3) M = λ x x ) i ( i where λ i is he response of he perceived policy pah o news, x i is a baseline value of he news variable, and he subscrip i is included o allow for he possibiliy of changing policy pah over ime. For example, over a sample period, he perceived responsiveness of cenral bank policy o price shocks may change from Period A o Period B, and, herefore, he parameers in he policy reacion funcion may change such ha value of he news variable may change such ha x x. A B λ λ or he baseline f (2) is he rue model, an esimae of he effec of observed news announcemens on asse prices using equaion (1) will yield q q + = α + β φλi + (4) ( )( + + ) + and he esimaed coefficien on news is β φλ ). n he case in which he perceived policy x ( i reacion funcion does no change over he sample period sudied, here will be a consan esimaed response of asse prices o news. For example, his would occur if an unvarying insiuional srucure deermines he reacion funcion. Bu, in he case where here is a change in he perceived policy reacion funcion, perhaps as a consequence of he pah of acual policy, here will be a corresponding change in he esimaed response of asse prices o news. Therefore, we can check for changes in he marke s percepions of policy sances over a sample period by considering wheher here are significan differences over ime in he response of asse prices o news. E x ε A B 4

6 Esablishing Credibiliy, Goldberg and Klein We apply his general framework o he quesion of wheher he marke percepion of he ani-inflaionary sance of a cenral bank evolves over ime or remains consan. n our applicaion, he dependen variable, q q +, represens he change in he slope of he yield curve over alernaive asse mauriies or he change in he exchange rae, reflecing he inflaion oulook. For example, for 10 year and 2 year bonds, q ( 10 2 ) ( 10 2 r r π π ) = =, so ha beween he ime before a news announcemen, -, and he ime afer ha announcemen, +, ( r + r + ) E ( r + r + ) = ( r + E r + ) ( r + E r + ) = ( π + E π + ) ( π + E π 2 + ) An increase in he slope of he yield curve in response o news implies ha news has a bigger impac on he change in he expeced value of inflaion a he 10-year horizon han on he expeced value of inflaion a he 2 year horizon. 6 n conras, a flaening of he yield curve in response o economic news suggess ha he effec of ha news has a smaller expeced effec on inflaion a he longer horizon han i does on inflaion a he shorer horizon. This discussion shows how o inerpre changes in he coefficiens of an equaion like (4) when q ( r r ) and q ( r r ). Consider wo differen marke views of + = + + = policy, a view ha policy has a sronger ani-inflaion il (Policy S) and a view ha policy has a weaker ani-inflaion endency (Policy W). These policies are disinguished by he condiion ha λ > λ when x in (1) represens a posiive price shock. Therefore, if he S W percepion shifed owards a view ha a cenral bank was more commied o a srong aniinflaion policy as ime wen on, we would expec o find a larger value for he esimaed coefficien ha represens he response of he slope of ha counry s yield curve o inflaionary news in he earlier period of he sample han in he laer period since β φλ ) > ( β φλ ). n conras, if percepions of policy were unvarying, perhaps because ( W S hese percepions were driven by he iniial and unvarying insiuional srucure, we would 6 News effecs on equilibrium real ineres raes are common o reurns a all horizons along he yield curve. As a consequence, when we difference across he reurns of long and shor-daed bonds we absrac from he effec of news on equilibrium real reurns. We also are absracing from he effec of news on erm premia or liquidiy premia, which seems reasonable given he shor ime horizon beween imes - and +. n a similar vein, Fleming and Remolona (1999) argue ha he effecs of news on asse prices of differen mauriies reveals informaion abou marke paricipan beliefs abou cenral bank reacion funcions. 5

7 Esablishing Credibiliy, Goldberg and Klein no expec o find a significanly differen esimaed coefficien on he response of he slope of a cenral bank s yield curve o price news across he sample period sudied. The robusness of hese resuls can be sudied by running a similar regression using, as he dependen variable, he change in he exchange rae, i.e. q s + and q s where + = s is he number of unis of domesic currency per uni of foreign currency. Through purchasing power pariy, a nominal exchange rae evolves in accordance wih he rajecories of home versus foreign inflaion. The sign of he coefficien of news on s depends upon wheher he news is expeced o have a bigger effec on expeced inflaion in he foreign counry or in he home counry. This, in urn, depends upon he expeced responses of he respecive cenral banks o hese inflaionary pressures. =. Evolving Percepions of European Cenral Bank Policy Based on he discussion in he previous secion we es for changes in he marke s percepion of he policy of he ECB during is firs five years by esing for a significan change in he coefficien on he surprise componen of he news variable in a regression on he slope of he German yield curve or in a regression on he euro/dollar exchange rae. is also useful, as a benchmark, o es for a significan change in responsiveness of he slope of he Unied Saes yield curve o price news. f one finds no significan change in he regressions on he Unied Saes yield curve, bu a significan change in he regressions on he German yield curve and on he euro/dollar exchange rae, hen his is consisen wih a shif in marke percepions of he policy sance of he ECB, wih no comparable shif in he marke percepions of he policy sance of he Federal Reserve. Furhermore, if he esimaed coefficien on he price news variable is greaer in an earlier period han in a laer period, his suggess ha he marke is ascribing o he ECB a more ani-inflaionary sance over ime. The iming of breaks in slopes across periods can be compared o he evoluion of acual policy o examine wheher credibiliy evolved in response o conduc. Alernaively, if here is no evidence of a significan break in he responsiveness of news over his period, he credibiliy of he ECB may have been aained hrough he saues ha se up is insiuional srucure. 6

8 Esablishing Credibiliy, Goldberg and Klein 3.1 Empirical Mehods. We implemen he model presened in Secion by running regressions ha ake he form (5) q + q = α + β ( x + E x + ) + β ( x + E x + ) + ε + where ( x E x + ) + is he surprise componen of news across announcemen evens in he firs par of he sample period and ( x E x + ) + is he surprise componen of news across announcemen evens in he second par of he sample period. A series of regressions are run, each wih a successively laer break poin ha divides he sample ino is firs and second pars. For a sample wih n observaions, we run n 20 regressions; one in which he break poin is he 10 h observaion and, herefore, he firs period runs from observaion 1 o observaion 10 and he second period runs from observaion 11 o observaion n; a second regression in which he break poin is he 11 h observaion so he firs period runs from observaion 1 o observaion 11 while he second period runs from observaion 12 o observaion n; and so on, unil he final regression, in which he break poin is he n 11 h observaion and he firs period runs from observaion 1 o observaion n 11 while he second period runs from observaion n 10 o observaion n. For each of hese regressions, we plo boh β and β, and heir respecive confidence inervals. Also, in a separae graph, we plo he -saisic associaed wih he es β β = 0 agains he daes associaed wih he break poins. A able presens he regressions associaed wih break poin for he firs ime ha he es β β = 0 is significanly differen from zero a he 95 percen level of confidence, and also regressions associaed wih he break poin ha yields he maximum - value for he es β β = The daa. We use a high frequency daa base on hourly changes in sovereign deb yields for he Unied Saes and Germany, spanning he period from wih from January 1999 o Sepember The daa are as repored by Reuers, for he on-he-run U.S. and German wo- and en-year noes. 7 Using his daa, we consruc differences beween he one-hour change in 10 year bond ineres raes and he one-hour change in 2 year bond ineres raes 7 ndicaions are capured hourly beween 0:00 and 23:00 easern sandard ime for hese markes. 7

9 Esablishing Credibiliy, Goldberg and Klein (ha is, he change in he slope of he yield curve) for boh German and U.S. bonds in he hour surrounding economic news announcemens, ha is, 10, German 10, German 2, German 2, German ( r+ r ) ( r+ r ) 10, US 10, US 2, US 2, US ( r+ r ) ( r+ r ) where he superscrips refer o 10 year German bonds (10,German), 2 year German bonds (2,German), 10 year U.S. bonds (10,US) and 2 year U.S. bonds (2,US). We also es for changes in he effecs of economic news on he euro/ dollar exchange rae, ( Euro / $) ( Euro / $) + The independen variable in all regressions is he surprise componen of U.S. price announcemens. Price news is he mos direc indicaor of changes in inflaion. While our modeling mehods also can be applied o oher ypes of economic news variables, news abou real macroeconomic variables may have more ambiguous relaionships o changes in percepion abou he underlying inflaionary environmen. For example, news ha real oupu is higher han expeced may be associaed wih an increase in expeced inflaion, if he higher oupu is due o greaer demand, or a decrease in expeced inflaion, if larger han expeced produciviy gains are he source of he increase in oupu. Furhermore, an invesigaion of changing policy responses associaed wih news on real macroeconomic variables can conflae perceived policy shifs of boh he moneary and he fiscal auhoriies. We focus on news abou Unied Saes prices, raher han news abou German or European prices, because of possible problems wih he news conen of German and European price announcemens. Goldberg and Leonard (2003) and Ehrmann and Frascher (2004) presen evidence of he effec of Unied Saes macroeconomic news on American, European, and German asse prices, and he weakness or absence of an effec of German or European macroeconomic news on asse prices in boh Europe or America. 8 These resuls, consisen wih he empirical regulariies found by Chinn and Frankel (2003), are srongly supporive of news effecs originaing in he Unied Saes and being ransmied o he euro 8 Gadzinski and Orlandi (2004) argue ha inflaion persisence is comparable across he Unied Saes and he euro area. 8

10 Esablishing Credibiliy, Goldberg and Klein area, wihou much reverse causaion. The four differen indicaors of U.S. prices are he CP, he PP, he Core CP, and he Employmen Cos ndex (EC). As in Andersen e al. (2003), he news componen of an economic daa release represens he difference beween he acual release and he markes expecaion of he conens of he release. Also following heir approach, we obain comparabiliy across he four price series we use by dividing news for each of he series by he respecive sandard deviaion of news for ha series over he sample period. The expecaions daa ha we rely on are median responses from weekly surveys of marke paricipans conduced by Money Marke Services, a division of Sandard & Poor s, and more recenly from Acion Economics. 9 Table 1 presens mean values of news and he associaed sandard errors, for each of he four separae announcemens as well as for he se of he wo CP announcemens and for he se of all four announcemens. This able also includes he resuls of regressions of he value of he announcemen on he expeced value of he announcemen. These regressions, which ake he form x γ 1E = + + x + + u + 0 γ should resul in γ 1 if he expecaions of he announcemens are unbiased forecass.10 1 = The regression resuls show ha we fail o rejec he hypohesis ha γ 1 a he 95 percen level of confidence for he Core CP announcemens and for se of announcemens including he Core CP and he CP. While hese resuls sugges ha he se of he wo CP announcemens is he preferred for implemening equaion (5), we repor, in he main body of he paper, resuls using all four announcemen series since his provides us wih more observaions. Bu, mindful of he resuls presened in Table 1, in secion 3.4 we presen comparable resuls obained wih he smaller se of observaions from he wo CP series only, which show he robusness of he resuls using all four series. 1 = 9 Money Marke Services were he source of hese daa hrough December Haver Analyics provided coninuous expecaions and announcemen daa hrough 2004 using daa from Acion Economics. 10 Anoher condiion is ha γ 0 0 =, which is me for he Core CP announcemens and he se of Core CP and CP announcemens. 9

11 Esablishing Credibiliy, Goldberg and Klein Table 1: Saisics on News, Announcemens and Expecaions x Resuls of x + γ + 1E x + + u + News = ( E x + ) + = 0 γ Obs. Mean s.d. γ s.e. 0 γ 1 s.e. R 2 Core C.P C.P P.P EC All Series All CP Series Bold = γ 0 = 0 or γ 1 = 1, respecively, a 95 percen level of confidence. 3.3 Marke Percepions of Policy. Resuls on he possible changes in he percepions of policy, as embodied in a comparison of he value of he coefficiens β and β in regressions based on specificaion (5), are presened in hree differen ways. Figures 1, 2 and 3 presen he coefficiens β and β, and heir respecive confidence inervals, for a se of rolling regressions in which he daes of he break poins change. Table 2 presens esimaes for regressions wih paricular break poins, as well as full sample regressions wih no break poins. A hird se of resuls, presened in Figure 4, plos he -saisics for he equaliy of β and β. n each of of hese hree cases, he regressor is he sandardized surprise componen of Unied Saes price news using all four announcemen series. Each of he hree ses of resuls include esimaes of he effec of his news variable on he change in he slope of he he German yield curve, he change in he slope of he Unied Saes yield curve, and he change in he value of he euro / dollar exchange rae. The overall sample period runs from February 5, 1999 o Augus 17, 2004, while break poins are considered from observaions on April 29, 1999 (corresponding o he 10 h observaion in he daa) hrough June 15, 2004 (corresponding o he 208 h observaions of he daa). Figures 1, 2 and 3 show he evoluion of β and β for each of he hree regressands. n each of hese graphs, he value of β and β for a paricular dae depiced on he horizonal axis represens esimaed values of he respecive coefficien for a regression in which he early period includes all observaions up o and including ha dae, and he 10

12 Esablishing Credibiliy, Goldberg and Klein laer period consiss of all observaions afer his dae. An unvarying view of he cenral bank s policy sance would resul in an overlap of he confidence inervals for β and β over he full sample period while an evolving view owards a greaer ani-inflaion sance is refleced in, a leas iniially, a significanly greaer value of β as compared o β (ha is, confidence inervals ha firs are disinc and laer overlap). Figure 1 shows ha he esimaed effec of news on he change in he slope of he German yield curve is significanly greaer for he early par of he sample han for he laer par of he sample when he daes defining he early par of he sample are beween mid-1999 and early While β > β for he enire sample, he confidence inervals for β and β overlap when he break poin is eiher very early in he sample or afer he beginning of n conras, as shown in Figure 2, β and β are saisically indisinguishable for any break poins when he regressand is he change in he slope of he Unied Saes yield curve. Figure 3, in which he regressand is he change in he euro / dollar exchange rae, presens a resul more similar o Figure 1 han o Figure 2, since, in his case, β > β and β β saisically disinguishable when he break poin is beween April 2000 and June and are Figure 1: Effec of U.S. Price Announcemens on German Yield Curve Beas for Early and Lae Periods Early & Lae Period Beas, (and confidence inervals) /29/1999 7/29/ /29/1999 1/29/2000 4/29/2000 7/29/ /29/2000 1/29/2001 4/29/2001 7/29/ /29/2001 1/29/2002 4/29/2002 7/29/ /29/2002 1/29/2003 4/29/2003 7/29/ /29/2003 1/29/2004 4/29/2004 Dae Confidence nerval, Early Period Coefficien for Early Period Confidence nerval, Laer Period Coefficien for Laer Period 11

13 Esablishing Credibiliy, Goldberg and Klein Figure 2: Effec of U.S. Price Announcemens on U.S. Yield Curve Beas for Early and Lae Periods 0.05 Early & Lae Period Beas, (and confidence inervals) Confidence nerval, Early Period Coefficien for Early Period Dae Confidence nerval, Laer Period Coefficien for Laer Period 4/29/1999 7/29/ /29/1999 1/29/2000 4/29/2000 7/29/ /29/2000 1/29/2001 4/29/2001 7/29/ /29/2001 1/29/2002 4/29/2002 7/29/ /29/2002 1/29/2003 4/29/2003 7/29/ /29/2003 1/29/2004 Figure 3: Effec of U.S. Price Announcemens on Euro/$ Beas for Early and Lae Periods Early & Lae Period Beas, (and confidence inervals) /29/1999 7/29/ /29/1999 1/29/2000 4/29/2000 7/29/ /29/2000 1/29/2001 4/29/2001 7/29/ /29/2001 1/29/2002 4/29/2002 7/29/ /29/2002 1/29/2003 4/29/2003 7/29/ /29/2003 1/29/2004 Dae Confidence nerval, Early Period Coefficien for Early Period Confidence nerval, Laer Period Coefficien for Laer Period 12

14 Esablishing Credibiliy, Goldberg and Klein The resuls presened in hese figures shows ha he effec of he surprise componen of US price announcemens on he change of he slope of he German yield curve varied over he sample period, while here is no evidence of a change in he responsiveness of he slope of he US yield curve o price news announcemens during his period. 11 Furhermore, he fac ha he coefficiens in he earlier period are significanly larger han he respecive coefficiens in he laer period for he regressions on change in he slope of he German yield curve is consisen wih an increasing amoun of credibiliy over ime in he ani-inflaion sance of he ECB. These resuls suppor he view ha he ECB gained credibiliy over ime hrough is acions, no ha is insiuional srucure garnered an unvarying level of credibiliy from is incepion. The fac ha here is no significan break in he responsiveness of he slope of he Unied Saes yield curve o Unied Saes price news over his period bolsers his resul since, if here were a change in he overall economic environmen, one would expec o find some consisency in he paern of break poins across counries. This inerpreaion is consisen as well wih he paern of significance of he regressions of surprise componen of Unied Saes price news on he change in he euro/dollar exchange rae. n his case, he inerpreaion of he resul depends upon wheher he impac of an unexpecedly large increase in US prices has a bigger effec on expeced inflaion in he Unied Saes or in Europe, and wheher he relaive magniude of hese effecs varies over ime. The fac ha β > β for all break poins, and β β and are saisically disinguishable when he break poins are in he early par of he sample, combined wih he differences in he resuls beween he Unied Saes and German yield curve regressions, is consisen wih he inerpreaion above of a varying percepion of he ani-inflaion sance of he ECB. Figure 4 presens he -saisics associaed wih he es β β = 0 for each of he 198 possible break poins in he sample, for each of he hree regressands. The hree lines in he figure represen ess ha employ, as he regressand, he change in he slope of he Unied Saes yield curve (he line consising of longer dashes), he change in he slope of he 11 While numerous sudies have shown he implicaions of such news for he U.S. yield curve, and for wo or en year bond yields, he slope of he yield curve is iself unaffeced by his economic news over he period examined. 13

15 Esablishing Credibiliy, Goldberg and Klein German yield curve, (he solid line), and he change in he euro/dollar exchange rae (he line consising of shorer dashes). Figure 4: High-Frequency Responses o U.S. Price News -saisic for Equaliy of Coefficiens in 2 Periods 6 max Ger. 1/14/00 4/14/ /15/99 7 max Euro: 3/16/01 10/25/01 -Saisic /18/1999 5/18/1999 6/16/99 8/18/ /18/1999 2/18/2000 5/18/2000 4/14/00 8/18/ /18/2000 2/18/2001 5/18/2001 8/18/ /18/2001 2/18/2002 Dae 1/30/03 6/13/0 5/18/2002 8/18/ /18/2002 2/18/2003 5/18/2003 8/18/ /18/2003 2/18/2004 5/18/2004 Germany: Yield Curve Slope Euro/Dollar Exchange Rae U.S.: Yield Curve Slope Figure 4, like Figures 1 and 3, show ha he economeric ess suppor a range of possible break poins associaed wih he effec of he surprise componen of price news announcemens on he change in he slope of he German yield curve and on he change in he euro/dollar exchange rae. Again, hese resuls are consisen wih an evolving view of he policy sance of he ECB, raher han a view of a consisen policy sance ha would arise if unvarying insiuional srucures dominaed he percepion of he ani-inflaion sance of he ECB. This inerpreaion is bolsered by he absence of a significan break poin for regressions of news on he change in he slope of he Unied Saes yield curve. The absence of a significan break poin in his case suggess ha, in conras o he ECB, he long and relaively sable hisory of Federal Reserve policy led o a consisen view of is policy sance over his sample period. An examinaion of he evoluion of he -saisics wih respec o daes reveals how policy acions are linked o credibiliy. The significance of he break poin for he regression of he German yield curve on he surprise componen of U.S. price news coninues o rise over he firs par of he sample, wih he -saisic reaching global maxima on January 14, 14

16 Esablishing Credibiliy, Goldberg and Klein 2000 and on April 14, On his laer dae, here is also a large jump in he significance of a break poin for he regression of he euro/dollar exchange rae on he surprise componen of US price news from 0.96 on April 13, 2000 o 2.96 on April 14, There is no evidence of a significan break poin afer mid is ineresing o consider hese daes in ligh of he acions of he ECB around ha ime. November 4, 1999 marked he firs ime ha he ECB raised is key ineres rae since i began operaions on January 1, A ha ime, his ineres rae, he rae for main refinancing operaions, was raised from 2.5 percen o 3.0 percen. This was followed by anoher 25 basis poin increase on February 3, 2000, addiional 25 basis poin increases on March 16 and April 27, and a 50 poin basis poin increase o 4.25 percen on June 9. Wihin our model, he spike in he significance of an esimaed break poin for he slope of he German yield curve in January 2000, as well as he large increase in he significance of an esimaed break poin for he euro/dollar exchange rae in mid-april 2000, sugges ha hese acions by he ECB served o aler he marke s percepion of is policy. n paricular, he fac ha he -saisic on he es β β = 0 is posiive indicaes an evoluion owards he view ha he ECB was becoming more srongly ani-inflaionary in he wake of his round of ineres rae increases in The European Cenral Bank swiched owards a more accommodaive moneary policy on May 11, 2001 wih a decrease in he minimum bid rae for he main refinancing operaions by 25 basis poins, o 4.50 percen. 13 As shown in Figure 4, he esimaed - saisic for he es of β β = 0 in regressions of he euro/dollar exchange rae on he surprise componen of he U.S. price news reaches is peak jus before his policy shif, in mid-march Afer his, here is a more-or-less seady decrease in he -saisic for an esimaed break poin hrough he end of he sample, during which ime he ECB coninued o lower ineres raes. The decline in he -saisic for he German yield curve regressions begins on November 9, 2001, abou six monhs afer he firs indicaion of his policy shif, 12 The key ineres rae on fixed rae enders was a 3.00 percen from January 1, 1999 hrough April 9, when i dropped by 50 basis poins o 2.50 percen. On November 4, 2000 a period of moneary ighening sared. For main refinancing operaions, changes in he rae are effecive from he firs operaion following he dae when changes were indicaed. 13 On June he ECB announced ha, saring June 28, 2000, he main refinancing operaions of he Eurosysem would swich from fixed rae enders o variable rae enders. Thereafer he key ineres rae se by he ECB was he minimum bid rae of he variable rae enders for he main refinancing operaions. See 15

17 Esablishing Credibiliy, Goldberg and Klein and afer hree addiional ineres rae cus by he ECB on Augus 30, 2001 (when he ineres rae was cu by 25 basis poins), on Sepember 17, 2001 (when he ineres rae was cu by 50 basis poins), and on November 8, 2001 (when he ineres rae was cu by 50 basis poins). The -saisic for a break poin in he se of German yield curve regressions decreases from 4.63 for a break poin on Ocober 25, 2001, o 3.89 for a break poin on he nex dae of a price announcemen, November 9, 2001, he day afer a 50 basis poin cu in he key ineres rae by he ECB. 14 Figure 4 shows a more-or-less seady decline in he -saisics for he esimaed break poin for boh he euro/dollar exchange rae and for he German yield curve during he remainder of he sample period afer November During his ime, here were hree more rae cus, on December 5, 2002 (a 50 basis poin cu), March 6, 2003 (a 25 basis poin cu) and June 6, 2003 (a 50 basis poin cu). The main refinancing ineres rae remained a 2.00 percen from he ime of his las rae cu unil he end of he sample period in Sepember Early in 2003 he ECB refined heir wo pillar approach, wih he imporance of M3 apparenly reduced and he arge for inflaion a or slighly below wo percen. This communicaion was a significan even. There is no evidence of a significan break poin afer mid An inerpreaion could be ha, by ha ime, he marke had esablished is views of he ECB and, given he experience o ha poin as well as he lack of more recen policy surprises, hese views did no change. is noeworhy ha his sabiliy persised even hrough he ime of he change of he presidency of he ECB from Wim Duisenberg o Jean-Claude Triche on November 1, Table 2 isolaes resuls for regressions wih paricular break poins and also presens regression resuls for he full sample under he assumpion of no break poin. This able includes regressions represening he earlies break poin for which here is a saisically significan difference (a he 95 percen confidence inerval) beween β and β, and regressions represening in which he break poin yields he larges saisical difference beween β and β. Regressions wih he separae coefficiens β β and in which he 14 One may be concerned ha, because his period includes Sepember 11, 2001, he effec of he erroris aacks on ha day may be responsible for he shif in he -saisics. However, in his figure we do no observe comparable changes in he -saisics from he regressions on he U.S. yield curve in he afermah of he 9/11 aacks (he large change in he -saisic for hese regressions ha is apparen in Figure 1 occurs in lae-july 2001). As will be seen below, here is some evidence of a break poin in he regressions on he slope of he US yield curve around he ime of Sepember 11, 2001 when we use only he wo CP price series announcemens. 16

18 Esablishing Credibiliy, Goldberg and Klein regressand is he change in he Unied Saes yield curve are no presened, however, since here is break poin ha generaes a significan difference in β and β in his case. The firs day when he -saisic for he es of he equaliy of he wo coefficiens exceeds 2.00 is June 16, 1999 for he regression on he slope of he German yield curve, and April 14, 2000 for he regression on he exchange rae. The esimaes repored in Table 2 for he wo regressions wih hese respecive break poins show ha he coefficiens on he price news for he earlier par of he subperiods are abou hree imes larger han he coefficiens for he laer par of he sample in boh cases. Also, he coefficiens on β and β in each of he wo regressions are significan a beer han he 95 percen level of confidence. We also noe ha he adjused goodness of fi improves by abou 29 percen in he regression on he slope of he German yield curve and by abou 24 percen in he regression on he exchange rae when comparing hese regression wih he respecive ones wih no break poin. The final se of resuls repored in Table 2 use, as he break poin, he day ha generaes he maximum value of he -saisic for he es of he equaliy of he wo slope coefficiens. This is January 14, 2000 for he regression on he slope of he German yield curve, and March 16, 2001 for he regression on he exchange rae. The adjused R 2 saisics for hese regressions are abou 2.4 imes larger han he adjused R 2 saisics in he respecive regressions ha do no allow for differen coefficiens during he sample period. n he regression on he slope of he German yield curve, he esimaed value of β is almos eigh imes larger han he esimaed coefficien for β, and he -saisic for he former coefficien is 6.82 while he -saisic for he laer is n he case of he euro/dollar exchange rae regression, he esimaed coefficien in he earlier period, which has a -saisic of 5.96, is fifeen imes he size of he esimaed coefficien in he laer period, which has a -saisic of

19 Esablishing Credibiliy, Goldberg and Klein No Break Poins Firs Significan Break Poin (6/16/1999; 4/14/2000) Break Poin wih Max. Significance (1/14/2000; 3/16/2001) Table 2 Effecs of Surprise Componen of 4 US Price News Series on Yield Curve Slopes and Exchange Rae Slope of German Yield Curve Euro/Dollar Exchange Rae Slope of US Yield Curve Coef. s.e. Coef. s.e. Coef. s.e. α β Adj. R Α Β Β Adj. R sa: β = β Α Β Β Adj. R sa: β = β α = inercep, β = coefficien on surprise componen of US price news for full sample, β, β coefficiens on surprise componen of US price news for 1 s or 2 nd par of sample, respecively. Daes lised for Firs Significan Break Poin and Maximum Significan Break Poin are for (German Yield Curve; Euro/Dollar Exchange Rae). Bold = significan a beer han 95 percen level of confidence. See ex for definiions of regressands and surprise componen of US price news. 3.4 Robusness The resuls presened in Table 1 indicae ha he expeced values of he wo CP price announcemens, he CP iself and he Core CP, are good predicors of he acual announcemens over he sample period in ha we canno rejec he hypohesis ha he esimaed coefficien on he expeced value of he announcemen is equal o 1 a he 95 percen level of confidence in a regression of he acual announcemen on is expeced value. As was seen in Table 1, his resul does no exend o he oher wo price series, he PP and he Employmen Compensaion ndex. Therefore, as a check of he robusness of he resuls 18

20 Esablishing Credibiliy, Goldberg and Klein presened in Figure 1, we also calculaed he -saisics associaed wih a break in he responsiveness of he slope of he German yield curve, he slope of he U.S. yield curve, and he euro/dollar exchange rae o news abou he CP series only. The resuls from his exercise, which includes 130 observaions and, herefore, -saisics for 110 possible break poins from June 16, 1999 hrough March 17, 2004, are ploed in Appendix Figure 1. The resuls from his exercise are generally consisen wih hose presened in regressions using he full se of four price series in consrucing news. The firs significan break poins occur on he same daes for boh he German yield curve and exchange rae regressions using eiher all four price series, as in Figure 1, or only he wo CP series. 15 The dae on which he -saisic associaed wih he break poin reaches is maximum value is he same across he ses of esimaes for he German yield curve, and is abou six monhs earlier for he exchange rae when using only he wo CP series han when using all four series. The las dae on which he -saisic associaed wih a break poin is greaer han 2.00 is virually he same in he wo figures for he exchange rae series, bu i is abou six monhs laer for he German yield curve esimaes when only he wo CP series are used raher han all four price series. The one sriking difference beween he resuls is ha here is a significan break in he slope of he US yield curve series when only he wo CP announcemen series are used. The period over which a break is significan is relaively shor, and occurs beween Augus 16, 2001 and November 16, 2001, a period ha spans he erroris aack on he World Trade Cener. The larges -saisic is associaed wih a break poin on Ocober 19, The regression esimae wih his break poin is ( ) ( r ) = ( x ) ( x 10, US 2, US 10, US 2, US r + r + r + E x + + E x + (0.0017) (0.0026) (0.0024) wih an R 2 of and a -saisic of 2.37 for he es β = β (sandard errors are presened in parenheses below he coefficien esimaes). According o his resul, he period around he 9/11 aack may have marked a significan break poin for he markes view of he behavior of U.S. inflaion. Bu, given he weakness of he US economy in he pos 9/11 period, i can be argued ha break poin could ) 15 The dae for he break poin for he German yield curve, June 16, 1999, is he firs possible break poin in Figure 2 because of he smaller se of observaions available wih only he CP announcemens. 19

21 Esablishing Credibiliy, Goldberg and Klein represen a change in any of he parameers in he erm ( β φλ) of equaion 4, no jus a change in λ as was posied in ha model. is also noeworhy, wih respec o he perceived policy of he ECB, ha here was no a upward spike in he -saisics associaed wih a break poin for eiher he German yield curve or he exchange rae regressions in he period around he erroris aack. Thus, i is likely ha he change ha may have occurred in β or φ wih respec o he U.S. economy a ha ime was no mached by a comparable change in β or φ for he German economy, or for ha of Europe as a whole. Therefore, he fac ha here is a shor period of a significan esimaed break poin for he responsiveness of he slope of he US yield curve o a subse of he price news announcemen daa se does no aler our view ha he significan breaks in he regressions for he slope of he German yield curve and he exchange rae could reflec changing views abou he policy sance of he European Cenral Bank. V. Conclusions The imporance of he repuaion of a cenral bank for he success of is operaions is sressed in heory and is eviden from pracical experience. An imporan quesion is wheher a cenral bank gains credibiliy hrough is insiuional srucure or hrough he conduc of policy. This quesion is especially relevan for a newly esablished cenral bank ha faces he challenge of esablishing is repuaion, someimes in he face of poliical conroversy over he appropriae conduc of moneary policy. The evoluion of he markes percepions of he policy sance of he European Cenral Bank since i began operaions in January 1999 is ineresing for a number of reasons. One of hese reasons is he inheren ineres of he economic experience of he eurozone. A second reason is ha he esablishmen of he European Cenral bank provides a naural experimen for considering how he repuaion of a cenral bank evolves over ime. This episode is a paricularly rich vein o mine because of he conroversy surrounding he conduc of moneary policy in Europe as he ECB began is operaions. n his paper, we propose a novel es for he sudy of he evoluion of marke percepions abou he policy sance of a cenral bank. When we apply his es o daa on he slope of he German yield curve and he euro / dollar exchange rae, we find, in fac, evidence of a seady shif owards he view ha he ECB policy is geared owards keeping 20

22 Esablishing Credibiliy, Goldberg and Klein inflaion low. There is no a similar shif in he marke s percepion of he policy sance of he Federal Reserve, a period marked by he sabiliy in is leadership, he consisency of is saed goals, and he lack of poliical opposiion o he manner in which i conducs policy. 21

23 Esablishing Credibiliy, Goldberg and Klein References Alesina, Albero and Lawrence H. Summers, 1993, Cenral Bank ndependence and Macroeconomic Performance: Some Comparaive Evidence, Journal of Money, Credi and Banking, vol. 25, no. 2 (May), pp Andersen, Torben G., Tim Bollerslev, Francis X. Diebold and Clara Vega, 2003, Micro Effecs of Macro Announcemens: Real-Time Discovery in Foreign Exchange, American Economic Review, vol. 93, no. 1 (March), pp Ball, Laurence and Niamh Sheridan, Does nflaion Targeing Maer? in Ben Bernanke and Michael Woodford, ediors The nflaion-targeing Debae, Universiy of Chicago Press, Chicago, llinois, 2005, pp Barro, Rober J. and David B. Gordon, 1983, A Posiive Theory of Moneary Policy in a Naural Rae Model, The Journal of Poliical Economy, vol. 91, no. 4 (Augus), pp Bernanke, Ben and Michael Woodford, ediors, The nflaion-targeing Debae, Universiy of Chicago Press, Chicago, llinois, Blinder, Alan, Cenral Bank Credibiliy: Why Do We Care? How Do We Build?, 2000, American Economic Review, vol. 90, no. 5 (December), pp Calvo, Guillermo, On he Time Consisency of Opimal Policy in a Moneary Economy, 1978, Economerica, vol. 46, no. 6 (November 1978), pp Chinn, Menzie and Jeffrey Frankel, 2004, The Euro Area and World neres Raes manuscrip, Ocober. Cukierman, Alex, Cenral Bank Sraegy, Credibiliy and ndependence: Theory and Evidence, The MT Press, Cambridge, Massachuses, Ehrmann, Michael and Marcel Frazscher, 2004, Equal Size, Equal Role: neres Rae nerdependence beween he Euro Area and he Unied Saes European Cenral Bank working paper series no. 342, April. Fleming, Michael and Eli Remolona, 1999, Price Formaion and Liquidiy in he U.S. Treasury Marke: The Response o Public nformaion Journal of Finance vol. LV, no. 5 (Ocober). Forder, James, Cenral Bank ndependence and Credibiliy: s here a shred of evidence?: Review, 1999, nernaional Finance, vol. 3, no. 1 (April), pp Gadzinski, Gregory and Fabrice Orlandi nflaion persisence in he European Union, he euro area, and he Unied Saes. European Cenral Bank working paper series no. 414, November. 22

24 Esablishing Credibiliy, Goldberg and Klein Gerler, Mark Commen on Does nflaion Targeing Maer? in Ben Bernanke and Michael Woodford, ediors The nflaion-targeing Debae, Universiy of Chicago Press, Chicago, llinois, 2005, pp Goldberg, Linda and Deborah Leonard, 2003, Wha Moves Sovereign Bond Markes? The Effecs of Economic News on U.S. and German Yields Federal Reserve Bank of New York, Curren ssues in Economics and Finance volume 9, no. 9 (Sepember). Klein, Michael W., and Karen K. Lewis, 1993, Learning Abou nervenion Targe Zones, Journal of nernaional Economics, vol. 35, no. 3/4, pp Klein, Michael W., Bruce Mizrach and Rober G. Murphy, 1991 Managing he Dollar: Has he Plaza Maered? Journal of Money, Credi and Banking, vol. 23, no. 4 (November), pp Kydland, Finn and Edward Presco, 1977, Rules Raher han Discreion: The nconsisence of Opimal Plans, The Journal of Poliical Economy, vol. 85, no. 3 (June), pp Levin, Andrew ''mperfec Credibiliy and nflaion Persisence'' (wih Chrisopher J. Erceg), Journal of Moneary Economics (forhcoming). Lewis, Karen K., Occasional nervenions o Targe Raes, 1995, American Economic Review, vol. 85, no. 4 (Sepember), pp Svensson, Lars E.O., The Simples Tes of Targe Zone Credibiliy, 1991, MF Saff Papers, vol. 38, pp , The Simples Tes of nflaion Targe credibiliy, 1993, NBER Working Paper No. 4604, December