Working Paper Series. The role of central bank. for guiding. forecasts. No 1146 / January 2010

Transcription

1 Working Paper Series No 1146 / The role of cenral bank ransparency for guiding privae secor forecass by Michael Ehrmann, Sylveser Eijffinger and Marcel Frazscher

2 WORKING PAPER SERIES NO 1146 / JANUARY 2010 THE ROLE OF CENTRAL BANK TRANSPARENCY FOR GUIDING PRIVATE SECTOR FORECASTS 1 by Michael Ehrmann 2, Sylveser Eijffinger 3 and Marcel Frazscher 4 In 2010 all publicaions feaure a moif aken from he 500 banknoe. This paper can be downloaded wihou charge from hp:// or from he Social Science Research Nework elecronic library a hp://ssrn.com/absrac_id= We would like o hank Nergiz Dincer and Barry Eichengreen for sharing heir cenral bank ransparency daa, David-Jan Jansen, Erneso Villanueva and Hermann Hansen for help in he consrucion of he cenral bank communicaion daa, Qian Shen for excellen research assisance and seminar paricipans a he for helpful commens and suggesions. Eijffinger hanks he for is hospialiy during his say in Spring 2009 as Wim Duisenberg Research Fellow. This paper presens he auhors personal opinions and does no necessarily reflec he views of he European Cenral Bank or he Eurosysem. 2 European Cenral Bank, Kaisersrasse 29, Frankfur am Main, Germany; michael.ehrmann@ecb.europa.eu 3 European Banking Cener and CenER, Tilburg Universiy and CEPR; s.c.w.eijffinger@uv.nl 4 European Cenral Bank, Kaisersrasse 29, Frankfur am Main, Germany; marcel.frazscher@ecb.europa.eu

3 European Cenral Bank, 2010 Address Kaisersrasse Frankfur am Main, Germany Posal address Posfach Frankfur am Main, Germany Telephone Websie hp:// Fax All righs reserved. Any reproducion, publicaion and reprin in he form of a differen publicaion, wheher prined or produced elecronically, in whole or in par, is permied only wih he explici wrien auhorisaion of he or he auhor(s). The views expressed in his paper do no necessarily reflec hose of he European Cenral Bank. Informaion on all of he working papers published in he s Working Paper Series can be found on he s websie, hp:// c/ wps/dae/hml/index.en.hml ISSN (online)

4 CONTENTS Absrac 4 Non-echnical summary 5 1 Inroducion 6 2 The daa and he economeric model Privae secor forecas daa Measures of cenral bank ransparency Conrol variables 13 3 The effecs of cenral bank ransparency on privae secor forecass Overall effecs The response o shocks Probing he limis 17 4 Conclusions 18 References 20 Tables 22 Appendix 37 3

5 Absrac There is a broad consensus in he lieraure ha coss of informaion processing and acquisiion may generae cosly disagreemens in expecaions among economic agens, and ha cenral banks may play a cenral role in reducing such dispersion in expecaions. This paper analyses empirically wheher enhanced cenral bank ransparency lowers dispersion among professional forecasers of key economic variables, using a large se of proxies for cenral bank ransparency in 12 advanced economies. I finds evidence for a significan and sizeable effec of cenral bank ransparency on forecas dispersion, be i by means of announcing a quanified inflaion objecive, oher forms of communicaion, or by publishing cenral banks inflaion and oupu forecass. However, here also appear o be limis o cenral bank ransparency, wih decreasing marginal reurns o enhancing (economic) ransparency, and given our findings ha disagreemen among inflaion expecaions in he general public is no affeced by he various cenral bank ransparency measures analyzed in his paper. JEL classificaion: E37, E52, C53. Keywords: cenral banking, ransparency, disagreemen, survey expecaions, moneary policy, inflaion argeing, cenral bank communicaion, forecasing. 4

6 Non-echnical summary Deviaing from he assumpion of raional expecaions, e.g. by assuming coss of informaion processing, leads o macroeconomic models ha can generae disagreemen among economic agens. In such models, here is a role for cenral bank ransparency and communicaion o reduce he cos of informaion processing, eiher by increasing he precision of signals received by economic agens, or by making he informaion acquisiion less cosly. Over he las decades, cenral banks have underaken grea effors in ha direcion, leading o major improvemens in he predicabiliy of heir decisions and/or in he anchoring of inflaion expecaions. This paper sudies o wha exen greaer cenral bank ransparency and communicaion have conribued o a convergence in expecaions of economic agens, by esing for heir effecs on he disagreemen in forecass of key macroeconomic variables. Compared o earlier conribuions o his lieraure, our ess and proxies for cenral bank ransparency and communicaion are subsanially broader and more sysemaic. The paper finds empirical evidence ha several of hese measures (in paricular he announcemen of a quanified inflaion objecive and enhanced ransparency abou economic dimensions of he conduc of moneary policy, such as he release of he cenral bank s inernal forecass of inflaion and oupu) are effecive in ha regard, and ha he effecs are parially addiive. The reducion in forecaser disagreemen is no only saisically significan, bu also economically imporan. The findings of he paper indicae ha his reducion has been achieved because forecasers in counries wih more ransparen and open cenral banks manage o updae heir forecass in response o news in a smooher fashion, generaing less disagreemen. A he same ime, he findings of he paper also poin a possible limis o he effecs of cenral bank ransparency. Increasing economic ransparency seems o be paricularly powerful a low levels of ransparency. Ye a higher levels, he addiional effecs ha can be reaped by furher enhancemens of economic ransparency appear o be much smaller. Furhermore, he presen paper does no find ha he dispersion in expecaions among he general public (in conras o hose of professional forecasers) is affeced by any of he various measures of cenral bank ransparency ha i analyzes. 5

7 1. Inroducion Many modern macroeconomic models assume ha all agens form full-informaion raional expecaions. Under his assumpion, all economic agens share a common informaion se and form expecaions condiional on ha informaion, indicaing ha everyone has he same expecaions, and here are no disagreemens. However, his assumpion is easily rejeced when looking a forecas survey daa, where disagreemens, e.g. in inflaion forecasing, appear o be paricularly subsanial. A variey of explanaions have been offered o explain hese findings. For example, Mankiw and Reis (2002) propose a sicky-informaion model, in which economic agens updae heir expecaions only periodically because of coss of collecing and processing informaion. Using his model, Mankiw, Reis and Wolfers (2003) show ha disagreemen exhibis subsanial variaion hrough ime, moving wih inflaion, he absolue value of he change in inflaion, and relaive price variabiliy. Woodford (2001) argues for an imperfec-informaion model, in which agens observe a noisy signal abou he curren sae. Sims (2003) argues ha individual agen has limied capaciy for processing informaion, and we mus hus add informaion-processing consrains o model behavior in macroeconomic models. In all hese cases, economic oucomes could be improved if i was possible o ensure ha agens updae heir informaion ses more frequenly, ha he signals were observed wih less noise, or if he cos of informaion processing was reduced. Accordingly, here migh be a role for cenral bank ransparency and communicaion in shaping beer economic oucomes. Of course, neiher is relevan when assuming ha he public has full-informaion raional expecaions. If he public undersood moneary policy perfecly, any sysemaic paern in he way ha policy is conduced should be correcly inferred from he cenral bank's observed behavior (Woodford, 2005). Thus, when predicing fuure ineres raes, i suffices o inerpre (forecass of) economic daa in view of he cenral bank's policy rule. On he conrary, he assumpion of imperfec informaion is crucial for cenral bank communicaion and ransparency. Imperfec informaion generaes disagreemens among forecasers, making he economy volaile. Via communicaion and ransparency, cenral banks may anchor marke expecaions, reduce volailiy in he economy, and hus achieve a beer economic (inflaion) performance. Cenral banks have over he las decades sared o play his role, by becoming remarkably more ransparen, and by acively using communicaion as an insrumen o seer agens expecaions. A large academic lieraure has accompanied hese processes, and has idenified numerous ways in which hey have urned ou o be beneficial o he conduc of moneary policy. As poined ou, e.g., in he survey of he lieraure on cenral bank communicaion by Blinder e al. (2008), he anchoring of inflaion expecaions has been one poin in case, and an improved predicabiliy of moneary policy decisions anoher. In his paper we invesigae wheher and o wha exen cenral bank communicaion and increased ransparency have affeced he disagreemen among privae agens forecass of inflaion, ineres raes, and oher macroeconomic variables. Our hypohesis is ha we should see a reducion in forecas dispersion if i) informaion-relaed deviaions from he raional expecaion assumpions oulined above hold, and if ii) cenral banks manage o increase he signal-o-noise raio of heir public communicaion, or alernaively o reduce he cos of informaion processing by making relevan pieces of informaion available more readily. For ha purpose, we use wo kinds of daases in our empirical analysis. The firs one is he Consensus Economics forecas daa from 1990 o 2008, which covers professional analyss forecass of various macroeconomic variables for 12 advanced economies, over differen forecasing horizons The second is he European Commission household survey daa (Eurobaromeer) from 1985 o 2008, which measures inflaion expecaions of he general public. We find ha cenral bank ransparency and communicaion are indeed powerful ools o reduce he dispersion of professional forecasers views. These effecs can be generaed by announcing a quanified inflaion objecive, by being ransparen abou he economic daa and he policy models, as well as by publishing inernal forecass, and are herefore based on a 6

8 broad se of cenral bank policies. They are also economically large, wih reducions in dispersion reaching up o 20%. How is his reducion in heerogeneiy brough abou? The paper shows ha he updaing process of forecass in response o macroeconomic news generaes less dispersion in he case of more ransparen cenral banks, or if he news are accompanied by explici cenral bank communicaion. This suggess ha he privae secor can gain a clearer undersanding of he implicaions of news on he fuure evoluion of he economy if aided by cenral banks. A he same ime, he paper also idenifies possible limis o he effeciveness of hese cenral bank measures. In paricular, here is some evidence for diminishing marginal reurns in (economic) ransparency, and he paper finds ha neiher of he cenral bank measures affecs he disagreemen among he general publi while he levels of inflaion expecaions of he public are responsive o he announcemen of a quanified inflaion objecive or increasing economic ransparency. This paper relaes o an emerging lieraure ha sudies he deerminans of disagreemen among economic agens, and he role of cenral banks (for a deailed overview of he respecive papers, see Table A1 in he Appendix). 1 As o he deerminans of disagreemen, Coibion and Gorodnichenko (2009) use hree differen surveys of economic forecass o assess boh he suppor for and he properies of informaional rigidiies faced by agens. Specifically, hey rack he impulse responses of mean forecas errors and disagreemen among agens afer exogenous srucural shocks. They conclude ha afer srucural shocks, agens fail o adjus heir forecass by a sufficien amoun, inducing a non-zero response of forecas errors. As ime goes by, forecas errors converge monoonically o he full informaion oucome. They inerpre hese resuls as providing a robus empirical basis for models of informaional rigidiies ha has previously been sorely lacking. In a similar fashion, Andrade and Le Bihan (2009) find ha raional inaenion can explain some of he disagreemen in he s Survey of Professional Forecasers, alhough oher facors mus also be a work, given ha hese models canno generae sufficien sickiness in expecaions. Capisran and Timmermann (2009) go beyond he models ha base differen views of forecasers on differences in informaion ses and assume ha agens weigh he consequences of over- and underpredicions differenly and, as a resul, calculae heir forecass under asymmeric loss wih a shape of he loss funcion ha differs across agens. The auhors find empirical evidence of such asymmeries in forecasers loss funcions. Finally, Dovern, Frische and Slacalek (2009) invesigae deerminans of disagreemen abou six key economic indicaors using he Consensus Economics daase from G7 counries. Their esimaes documen a dichoomy beween disagreemen abou real variables (GDP, consumpion, invesmen and unemploymen), which is more srongly affeced by real facors, and disagreemen abou nominal variables (inflaion and ineres rae), which reacs o he insiuional seing of moneary policy (in paricular cenral bank independence). Disagreemen abou real variables inensifies srongly during recessions. Disagreemen abou nominal variables is considerably lower under independen cenral banks. Cross-secional dispersion for boh groups increases wih uncerainy abou he underlying indicaors. Their findings sugges ha more credible moneary policy can subsanially conribue o he anchoring of expecaions abou nominal variables; however is effecs on disagreemen abou real variables are moderae. Oher papers on he role of cenral banks for forecaser disagreemen can be divided by he cenral bank measures ha are analyzed. Swanson (2006) focuses on cenral bank ransparency in general, and finds ha wih increased ransparency of he US Federal Reserve, privae secor forecass of US ineres raes have become more precise, boh by improving he average qualiy of forecass as well as by reducing heir dispersion across forecasers. A relaed finding is presened in Bauer e al. (2006), who show ha since 1994 (when he FOMC began o release saemens accompanying changes in he policy rae) forecass for key macroeconomic variables by marke paricipans have become subsanially more synchronized. Fujiwara (2005) examines how a cenral bank s economic forecass affec forecass by 1 For a sudy on he effec of cenral bank communicaion on he raionaliy of privae secor forecass, see Chorareas e al. (2009). 7

9 professional forecasers and vice versa, looking a he case of he Bank of Japan. Empirical resuls show ha while cenral bank economic forecass are no significanly influenced by professional forecass of inflaion, hey have a noable effec on professional forecass. Furhermore, ess on he second momen sugges ha he Bank of Japan s forecas reduces professional forecasers uncerainy abou he fuure. A number of sudies relae o he announcemen of a quanified inflaion objecive. Beechey, Johannsen and Levin (2007) compare he recen evoluion of long-run inflaion expecaions in he euro area and he Unied Saes as provided in he respecive Surveys of Professional Forecasers. They reveal subsanially greaer dispersion across forecasers long-horizon projecions of US han of euro area inflaion, and relae his o he fac ha he has announced a definiion of price sabiliy, whereas he Federal Reserve has no. Crowe (2006) es wheher inflaion argeing enhances ransparency, using inflaion forecas daa obained from he Consensus Economics daase for 11 inflaion argeers. The paper oulines a simple signal-exracion model and derives a esable proposiion: if inflaion argeing enhances ransparency in he manner assumed in he model, hen is inroducion should promoe convergence o lower forecas errors. The auhor finds ha convergence occurs in all counries due o mean-reversion, bu ha he adopion of inflaion argeing leads o greaer convergence, as prediced by he model, which serves as srong evidence ha inflaion argeing does indeed enhance ransparency. In a similar vein, Crowe and Meade (2007, 2008) find ha enhanced ransparency pracices are associaed wih he privae secor making greaer use of informaion provided by he cenral bank, which suppors Crowe's (2006) finding ha he inroducion of inflaion argeing (hereby increasing in heir ransparency score) is associaed wih a convergence in forecas errors among he privae secor. A he same ime, Cecchei and Hakkio (2009) find no or only small effecs of he adopion of inflaion arges on forecas dispersion, and Capisran and Ramos-Francia (2009) idenify such effecs only for developing counries. All he sudies lised so far relae o he disagreemen among professional forecasers. Much less work has been done wih regard o he views of he general publi mos likely given he lack of daa availabiliy in his conex. 2 Two excepions are Maag and Lamla (2009) as well as Badarinza and Buchmann (2009), boh of which find evidence ha media coverage affecs inflaion forecas disagreemen of households. This paper conribues o hese srands of he lieraure by expanding he relevan ess. Whereas mos papers focus on one measure of cenral bank ransparency or communicaion, we broaden he analysis in his respec, allowing for a comparison of he relaive srenghs of he effecs. We will do so for a large se of counries, and furhermore comparing he responsiveness of professional forecasers o he one of he general public. The paper is organized in he following way. Secion 2 discusses he daa used in he empirical analysis. Secion 3 repors on he effecs of cenral bank ransparency and communicaion on privae secor forecass. Secion 4 summarizes he findings and draws some policy implicaions. 2. The daa and he economeric model To es for he effecs of cenral bank ransparency on privae secor forecass, we employ various daa ses. This secion provides a deailed descripion of he privae secor forecas daa, our measures for cenral bank ransparency and communicaion, and oher conrol variables. 2 In his respec i should be noiced ha he effecs of cenral bank ransparency and communicaion on professional forecasers are likely o be more direc and significan han hose regarding he general publi given ha he laer receive heir informaion indirecly via he media or via professionals (Carroll 2003). Van der Cruijsen and Eijffinger (2007) show, using household survey daa, ha acual and perceived ransparency in he general public may deviae, making he effecs of ransparency on he public less obvious. 8

10 2.1 Privae secor forecas daa As o he privae secor forecass, we focus on he daa provided by Consensus Economics. These comprise professional forecass for a large range of variables, for differen horizons, and are available for a reasonably long hisory for a se of indusrialized counries. They have also been used in he relaed sudies by Crowe (2006) and Dovern e al. (2009). In paricular, we have available he micro daa for 7 counries of he European Union (France, Germany, Ialy, he Neherlands, Spain, Sweden and he Unied Kingdom) and 5 oher counries (Canada, Japan, Norway, Swizerland and he Unied Saes). The daa are monhly, saring from January 1990 (January 1995 for he Neherlands, Spain and Sweden; June 1998 for Norway and Swizerland) and running hrough November For he 5 euro area counries in our sample (France, Germany, Ialy, he Neherlands and Spain), we do herefore cover he ime prior o moneary union as well as membership in he euro area. Noe, however, ha he consensus forecas daa are counry-specific (e.g., French forecasers provide heir esimaes of French inflaion, ec.). Accordingly, he dependen variables in our regressions remain counry-specific even afer While some of he explanaory variables (namely he ransparency indices and he dummy variable for cenral banks wih a quanified inflaion objecive) relae o he, and are hus idenical across he euro area counries afer 1999, he cenral bank communicaion variable remains counry-specific. In he economeric esimaes, i will herefore be imporan o calculae appropriae sandard errors. To cover a broad range of economic indicaors, we will analyze forecass for consumer price inflaion (% change p.a.), real GDP growh (% change p.a.), unemploymen (% of labor force), 3-monh ineres raes and 10-year governmen bond yields. Imporanly, he firs hree forecass are made for he end of he curren year and for he end of he nex year, whereas ineres rae forecass are made for a fixed horizon of 3 and 12 monhs. In paricular he year-end forecass herefore require careful modeling: over he course of a given year, he forecas horizon decreases; whereas a year-end forecas in January spans nearly an enire year, he forecasing problem in November is much simpler, as much of he year s daa are already realized and released. A model of dispersion or forecas errors will herefore have o conrol for he forecas horizon, as we would expec boh o decline over he course of a year. As we will describe below, all economeric models will herefore conain a full se of monh-fixed effecs. We have deliberaely oped for a broad coverage of economic indicaors o probe he exen o which forecass are affeced. Forecass of ineres raes are obviously very closely linked o expecaions abou he fuure course of moneary policy, and hese in urn should be more aligned o he exen ha he cenral bank s reacion funcion and assessmen of he curren economic siuaion are beer undersood. Wih regard o forecass for inflaion, GDP growh and unemploymen, a more homogeneous undersanding of he cenral bank s reacion funcion will only lead o a parial alignmen, as long as agens do no have similar views abou he ransmission mechanism of moneary policy. However, some of our proxies for cenral bank ransparency (discussed in more deail below) relae o wheher cenral banks publish heir macroeconomic policy models which, in urn, could align agens views abou he ransmission process. Also, we will es wheher he publicaion of he cenral bank s inflaion and oupu growh forecass reduces dispersion among he privae secor forecasers, such ha we are paricularly ineresed in he effecs on precisely hose variables. On average, he daase comprises 18 forecasers per counry and monh, alhough here is some variaion. For insance, survey paricipaion is relaively smaller in he Neherlands and Norway, wih 10 forecasers on average, whereas he number of forecasers in he Unied Kingdom is relaively large, wih 29 on average. Table 1 provides an overview of he forecaser coverage. TABLE 1 HERE Noe ha analyss need no give heir forecas for every single variable and every forecasing horizon each ime. Accordingly, he number of acual forecass differs slighly across variables. Our daase 9

11 conains around 41,000 o 42,000 observaions for inflaion and GDP forecass, whereas he number of forecass for unemploymen and ineres raes is somewha lower, a around 31,000 o 38,000. For he purposes of his paper, we are ineresed in measures of cross-secional dispersion on he one hand, and of forecas accuracy on he oher. As explained in he inroducion, our focus is on he effecs of cenral bank ransparency and communicaion on cross-secional dispersion, as we are ineresed in he exen o which cenral banks can make he informaion acquisiion of he privae secor more effecive. A he same ime, however, i is also imporan o es how forecas accuracy is affeced, as we need o ensure ha ransparen and communicaive cenral banks do no align forecass a lower levels of accuracy. While we would expec ha more ransparency and communicaion improves accuracy, his need no necessarily show up in our economeric esimaes, for he following reason: Forecass wih more ransparen and beer communicaing cenral banks migh be qualiaively more accurae han oherwise ex ane. Wha we observe, however, is only forecas accuracy ex pos, i.e. afer he realizaion of shocks. The ex ane forecas accuracy is no observable. If i were he case ha cenral bank ransparency and communicaion is correlaed wih he incidence of large, unexpeced shocks (for insance, because cenral banks in small open economies which are more subjec o shocks and herefore more volaile are more inclined o sabilize agens expecaions by being more ransparen), ex pos forecas accuracy migh urn ou o be lower under more ransparen cenral banks. 3 We will consequenly ake he analysis of forecas accuracy as a robusness es raher han as he focal poin of his paper. Our preferred measure of cross-secional dispersion is, in line wih he lieraure (Mankiw e al. 2003, Dovern e al. 2009), he iner-quarile range of forecass in a given counry and monh. The advanage of his measure over he simple sandard deviaion is ha i is insensiive o ouliers, which migh be imporan in he analysis of survey daa. We will, in any case, check for robusness of our resuls by using he sandard deviaion. 4 For a measure of forecas accuracy, we will calculae he average absolue forecas error per counry and monh. Table 2 provides some summary saisics for he resuling variables. The daase comprises 2342 counry-monh observaions for mos forecass (albei for unemploymen, only 1589 observaions are available). The iner-quarile range is on average around 0.2 o 0.5 percenage poins, alhough i ranges from forecass where he 25 h as well as he 75 h percenile of forecasers expressed he same view o hose where he differences amouned o around 2 percenage poins. Mean absolue forecas errors are comparably scaled, ranging on average from around 0.4 o 1.2 percenage poins, and spanning for he individual variables a broader range from 0 o 6.7 percenage poins. TABLE 2 HERE In order o conras he resuls for he professional forecass o hose for he general publi we will also make use of a second daase, which is based on he European Commission s Consumer Survey. This survey is also conduced monhly, covering around 1000 respondens in 20 European Union counries. The daa are available o us saring from January 1985 a he earlies, hrough November For his and oher reasons, our economeric models will include counry fixed effecs. Sill, he same argumen migh apply if he change in cenral bank ransparency is correlaed wih he change in an economy s exposure o shocks. 4 I is imporan o noe ha his measure is no necessarily a good proxy for uncerainy abou he variable o be forecased. As shown by D Amico and Orphanides (2008), disagreemen across forecasers is no necessarily equivalen o he inflaion uncerainy expressed by forecasers in he form of probabilisic responses, as in he Survey of Professional Forecasers. While his is imporan o keep in mind when inerpreing our resuls, we are primarily ineresed in he deerminans of disagreemen raher han uncerainy. 5 More precisely, he daa cover Belgium, Ireland, Greece, France, Ialy, he Neherlands and he Unied Kingdom since 1985m1; Spain (1986m6); Lavia (1993m1); Hungary (1993m2); Esonia (1993m4); Czech Republic (1995m1); Ausria and Sweden (1995m10); Finland (1995m11); Slovenia (1996m3); Cyprus, Lihuania, Poland and Romania (2001m5). The remaining counries (amongs which, for insance, Germany) conduc he survey, bu do no make he breakdown of he response shares per answer caegory available on he Eurosa websie. 10

12 Differen o he Consensus Economics daase, only inflaion forecass are conained in he consumer survey and furhermore, hese are no quaniaive, bu raher only qualiaive. Respondens are asked o indicae wheher, in heir view, prices will, over he nex 12 monhs, i) increase more rapidly, ii) increase a he same rae, iii) increase a a slower rae, iv) say abou he same, or v) fall. While he micro daa are no direcly available, he share of responses in he various caegories is mosly subsanial. This allows us o consruc a measure of dispersion, which needs o reflec he qualiaive naure of he responses. We herefore oped for he measure proposed by Lacy (2006), and also applied in Badarinza and Buchmann (2009): 2 4 d F j (1 F j ) (1) j1 where F j is he cumulaive relaive frequency of he j h caegory. Noe ha he 5 h caegory can be excluded, given ha is cumulaive relaive frequency is equal o one, and accordingly does no provide any relevan informaion abou he disribuion of he variable. 6 Unforunaely, he qualiaive naure of he quesion does no allow consrucing a measure of forecas accuracy. The European Commission releases a balance saisic ha is supposed o provide an aggregaed measure of he responses, based on he formula B ( PP 0.5P) (0.5M MM ) (2) where PP denoes he share of responses expecing prices o increase more rapidly, P he share of responses ha prices will increase a he same rae, M equivalenly for expecaions of prices o say abou he same, and MM referring o expecaions ha prices will fall. We will analyze his balance saisic o es wheher cenral bank ransparency and communicaion can affec he level of inflaion expecaions. Given ha quaniaive surveys on inflaion expecaions have ypically concluded ha consumers end o overesimae inflaion by non-negligible amouns (Bryan and Venkau 2001), we ake his o sugges ha a reducion in he level of inflaion expecaions is equivalen o an improvemen in forecas accuracy. For boh measures (d 2 and he balance saisic), our daase comprises 3967 counrymonh observaions. I is of course apparen ha he wo daases differ along various dimensions. Our main ineres are he differen arge groups across he wo daa sources, wih professional forecasers in he Consensus Economics survey, and he general public in he EC s Consumer Survey. A he same ime, however, he laer survey asks a qualiaive quesion, whereas he former is ineresed in a quanified forecas, and he coverage of counries and across ime differs. This of course limis he comparabiliy of resuls. A he same ime, we are no aware of any oher consumer survey ha is done in a comparable fashion across counries, asks quaniaive quesions, and spans a sufficienly long ime sample. 2.2 Measures of cenral bank ransparency Being ineresed in how cenral bank ransparency and communicaion affec privae secor forecass, we need o define measures for he former. To be as broad as possible, we have consruced four such measures. The firs one relaes o wheher or no a cenral bank has announced a quanified inflaion objecive. This measure is consruced as a dummy variable, aking he value of one as of he monh 6 To consruc his measure, we have excluded he don know response shares, and calculaed he cumulaive relaive frequencies jus aking ino accoun he response shares of he remaining answer caegories. This measure has he advanage ha i does no require he quanificaion of he disances beween he various answer caegories. The measure will be equal o zero if all responses fall ino a single caegory, and i will be equal o one, is maximum value, if 50% of respondens expec prices o rise more rapidly and 50% expec prices o fall. 11

13 when he quanified inflaion objecive was adoped (as, e.g., in Ball and Sheridan 2005), according o he cenral bank websies. Alernaives o he adopion daes migh be he announcemen dae (as, e.g., in Bernanke e al. 1999), or alernaively a laer dae o allow for a build-up of credibiliy (see, e.g., Goldberg and Klein 2005 for he ). By oping for he adopion dae, we place ourselves in he middle of hese alernaives. For he Consensus Economics daase, 7 ou of he 12 counries adop a quanified inflaion objecive over he course of our sample period, wo counries (Spain and Sweden) are considered o have a quanified inflaion objecive hroughou (noe ha heir daa only sar in 1995), and he remaining hree counries (Swizerland, Japan and he Unied Saes) do no have a quanified inflaion objecive over he enire sample. A second se of variables relaes o more general measures of cenral bank ransparency. Eijffinger and Geraas (2006) have developed a cenral bank ransparency index, which is an aggregae of subindices relaing o (1) poliical, (2) economi (3) procedural, (4) policy and (5) operaional ransparency. This index has recenly been exended and updaed for a large se of counries by Dincer and Eichengreen (2009). The daa is annual and covers he period from 1998 o In ligh of he large number of subindices, we will concenrae on he overall index, as well as he subindex relaed o economic ransparency. 7 A deailed explanaion of he consrucion of hese daa is provided in Eijffinger and Geraas (2006); for he purposes of his paper, i shall suffice o sae ha cenral banks are given ranks beween 0 and 3 for economic ransparency, depending on (a) wheher hey provide imely informaion on key economic daa (money supply, inflaion, GDP, unemploymen rae and capaciy uilizaion), (b) wheher or no cenral banks announce wha policy models are employed inernally, and finally (and probably mos imporan for our purposes) (c) wheher or no cenral banks release numerical inernal forecass for inflaion and/or oupu. The overall index is a simple aggregae of he 5 subindices, each of which can vary from 0 o 3, such ha he overall index is defined over a range from 0 o 15. For he Consensus Economics daase, he economic ransparency index ranges from 1 o 3, whereas he overall index akes values in beween 6 and 15. One imporan issue relaes o he relaively shor ime sample for which hese daa are available. To mainain a maximum number of observaions, we have decided o use he 1998 daa for all earlier years, and o assume ha in he years 2007 and 2008, he 2006 indices are sill applicable. Of course, we have esed for he robusness of our resuls using only he acual ransparency daa. A las measure for cenral bank ransparency is closely relaed o one aspec of he economic ransparency index, namely wheher cenral banks publish heir inernal forecass for inflaion and oupu. We have consruced a variable (which we will label cenral bank communicaion) ha is equal o one in hose monhs where such forecass are published, and zero oherwise. Pracices across cenral banks differ widely wih regard o he specific naure of he forecass; for insance, some cenral banks release forecass ha are owned by he decision-making body, whereas ohers provide saff forecass. Also, pracices differ wih regard o he underlying assumpions abou he fuure pah of ineres raes, which could be assumed o be consan, based on marke ineres raes, or reflec he cenral bank s own forecas. Finally, he forecasing horizons are also differen across cenral banks. Unforunaely, he daase does no conain sufficien variaion in order o es for he effecs of differen pracices separaely. Insead, we pool all hese forecas releases in one dummy variable, assuming ha regardless of heir precise naure, hey provide useful signals o he public which in urn migh align privae secor forecass. For he counries of he euro area, we consider boh he relevan publicaions of he as well as of he Naional Cenral Banks. Whereas he former provides saff projecions for he euro area as a whole, we prefer o rea his as relevan informaion also for he naional forecass, given ha his poenially allows forecasers o beer assess he fuure course of moneary policy, which in urn migh allow for more accurae (or more aligned) naional forecass. A robusness es for he inclusion of saff projecions for he euro area will be conduced. A final noe on he consrucion of he cenral bank communicaion variable relaes o he precise iming. Given ha he Consensus Economics survey akes place a he beginning of a monh, we have made sure ha he cenral bank publicaion can affec he 7 I should be noed ha we have esed also he subindices for poliical, procedural, policy and operaional ransparency, which didn prove o be significan in reducing he dispersion among professional forecasers. 12

14 survey. If i akes place afer he collecion of he Consensus Economics daa, we ener i in he subsequen monh. One migh expec ha hese four variables measure relaed conceps, and are herefore posiively correlaed. Table 3 displays he correlaion coefficiens, some of which are indeed relaively large. I is apparen ha cenral banks wih a quanified inflaion objecive are on average more ransparen. A he same ime, he correlaion coefficien wih economic ransparency is no ha high, presumably because also many oher cenral banks fare relaively high on he economic ransparency ranking. Cenral bank communicaion is leas correlaed wih he oher measures, which is explained by he fac ha i conains much more ime variaion: whereas all oher measures are bes described by sep funcions, our communicaion measure swiches beween he values zero and one repeaedly (in many cases, he relevan publicaions ake place quarerly, or semi-annually, leading o non-zero enries for 4 or 2 monhs a year, respecively). These correlaion coefficiens sugges ha in our empirical analysis i will no always be possible o clearly idenify which measure is riggering an effec. A he same ime, we are comfored by he relaively low correlaion measures beween quanified inflaion objecives, economic ransparency, and cenral bank communicaion. Noe ha we will subsume all of hese variables under he heading of cenral bank ransparency, even hough we will use he erm also for he narrower definiion of he cenral bank ransparency index. TABLE 3 HERE 2.3 Conrol variables When explaining cross-secional forecaser dispersion, a number of oher facors need o be considered. Mos sraighforwardly, i is imporan o include monh fixed effecs, as well as counry fixed effecs in any model. The former are necessary due o he changing forecasing horizon in he Consensus Economics survey, as discussed above. The laer can ake accoun of a myriad of issues ha migh have a bearing on he dispersion ha we observe on average in a given counry, such as he qualiy of he forecaser pool, he difficuly in forecasing a given economy (e.g. because smaller economies are more prone o shocks, and as such migh ceeris paribus be relaively more volaile), or possibly he availabiliy of a well-esablished forecasing insiuion ha migh affec he views expressed in his survey. Following Capisran and Timmermann (2009), we also include he condiional volailiy of he variable ha needs o be forecased. Such a conrol variable is useful for wo reasons. Firs, higher volailiy can imply a more difficul forecasing ask which, in urn, migh increase cross-secional dispersion. Second, our esimaion sample is in large par coinciding wih he so-called grea moderaion, i.e. a sable macroeconomic environmen. A he same ime, cenral banks have become more ransparen over he sample. In order no o confuse he wo ime variaions, i is imporan o conrol for he impac of he grea moderaion. We do his by including he condiional volailiy of he variable ha needs o be forecased. 8 In concordance wih he approach of Capisran and Timmermann (2009), we esimaed GARCH(1,1) models (where for our purposes 2 lags proved o be sufficien o reduce remaining serial correlaion), and exraced he esimaes of he condiional volailiy. Adding he level of he variable o be forecased (e.g. because inflaion migh be more difficul o forecas when i is high) does no affec our resuls, and generaes issues of correlaed regressors, given ha he condiional volailiy and he level of he variables are highly correlaed. We have, herefore, decided o only include he condiional volailiy measure. The acual daa on he variables o be forecased was sourced from he OECD s Main Economic Indicaors daabase. 8 An alernaive would be o include ime fixed effecs in he model, which would conrol for all ime variaion ha is common across counries, i.e. boh for possible common variaions due o he grea moderaion and due o changes in cenral bank ransparency. However, effecs semming from a common ime variaion in cenral bank ransparency would hen no longer be idenifiable. Furhermore, he condiional volailiy of he variable o be forecased is more direc, as i allows for a counry-specific conrol for he grea moderaion. 13

15 Furhermore, we include he absolue change in oil prices (sourced from he Bank for Inernaional Selemens) ha was observed prior o he survey, o incorporae possible effecs his variable migh have on he evoluion of forecass, and he disagreemen among forecasers in paricular. Imporanly, he inclusion of his variable does no change our resuls. A final se of variables ha we include in he exension of he empirical models relaes o he news componen of macroeconomic announcemens. Wih new macroeconomic daa arriving, forecasers need o updae heir informaion se and adjus heir forecass. This in iself migh affec forecaser disagreemen; an ineresing hypohesis in ha respec is wheher he adjusmen in forecass can ake place in a smooher fashion under ransparen cenral banks, and as such would lead o a relaively smaller increase in cross-secional dispersion. 9 To es for his possibiliy, we have consruced a series of macroeconomic announcemens and he surprise componen conained herein. Following he sandard in he announcemen lieraure (see, e.g., Andersen e al. 2003), we deduc he expecaion of he announcemen from he acual announcemen value of he variable o ge a measure of he surprise componen, and sandardize hese surprises by heir own, naional, sandard deviaion. Due o his sandardizaion, he regression coefficiens for each series can be inerpreed as a response per one naional sandard deviaion surprise. As is sandard in his lieraure, we obained daa on expecaions of he macroeconomic releases from a survey among financial marke paricipans conduced by Bloomberg Financial Services, and use he median response as our measure of expecaions. I is imporan o noe, however, ha hese daa are no available for he enire sample period. To maximize he sample, we focus on he releases of inflaion and unemploymen, for which he Bloomberg surveys have he longes hisory. As wih he cenral bank communicaion variable, we ensure ha he daa release is appropriaely assigned o he relevan Consensus Economics forecas round The effecs of cenral bank ransparency on privae secor forecass Wha effecs do cenral banks exer on privae secor forecass by means of heir ransparency and communicaion? This secion presens he empirical resuls. We will firs focus on a benchmark regression ha analyzes he deerminans of he iner-quarile range in he Consensus Economics forecass, before expanding he analysis and esing is robusness in several ways. Following his, we will urn o an analysis of he responsiveness o news shocks, and hen probe he limis of he effecs of cenral bank ransparency. 3.1 Overall effecs Our benchmark economeric model is given as i, c m x 1i, 1 2i, 3 oil 1 i, (3) where i, denoes our dependen variable, i.e. he cross-secional forecas dispersion for forecas i in counry c a ime, or alernaively he absolue mean forecas error, and c and m sand for he 9 This es is relaed o he influenial work by Gürkaynak, Sack and Swanson (2005), who show ha, in response o macroeconomic news shocks, long-erm inflaion expecaions are beer anchored in inflaion-argeing counries. 10 An alernaive migh be o use he magniude of he change in he announced values for he macroeconomic releases, assuming ha larger changes require more subsanial forecas revisions, and as such could raise disagreemen. Our preferred measure is based on he shocks assuming ha if large pars of a macroeconomic release have no been foreseen, his should rigger more subsanive revisions, whereas a large change in he announcemen could have been foreseen, and herefore already facored ino he earlier forecas rounds. 14

16 counry and monh fixed effecs. Our measures for cenral bank ransparency and communicaion are given by x c, ; i, marks he condiional volailiy of he variable o be forecased, and oil 1 denoes he absolue change in oil prices in he preceding monh. As oulined in deail in secion 2, our hypoheses are ha disagreemen among forecasers should depend negaively on cenral bank ransparency and communicaion ( 0 ), ha disagreemen is likely o be persisen ( 1 0 ), and ha higher volailiy of he variable increases disagreemen ( 2 0 ), as do larger oil price changes ( 3 0 ). We esimae differen varians of his model, including differen measures of (iner-quarile range and mean absolue error), for differen forecass i (varying he variable o be forecased as well as he horizon), and differen cenral bank ransparency and communicaion measures x each ime (alhough we will also esimae a join model for hree cenral bank measures for each forecas). The models will be esimaed by simple OLS. Since our observaions are clusered by counry, he sandard errors are esimaed aking he cluser srucure ino accoun. 11 Table 4 provides he resuls parameer esimaes obained in he benchmark model. Column (1) conains he esimaes esing for he imporance of quanified inflaion objecives, column (2) for overall ransparency, column (3) for economic ransparency, and column (4) includes our cenral bank communicaion variable. Column (5) combines quanified inflaion objecives, economic ransparency, and communicaion (he hree cenral bank measures wih he lowes correlaion coefficiens) in a join model. TABLE 4 HERE All models conain he various conrol variables discussed in he preceding secion. To save space, he regression coefficiens for he monh and counry fixed effecs are no provided. I is imporan o noe, however, ha especially he former are exremely relevan for he year-end forecass. No only are hey highly saisically significan a he 1% level and large in magniude (he iner-quarile range in December is up o 80% lower han in January for he curren-year forecass, and up o 20% for he nexyear forecass), bu heir imporance also increases monoonically over he course of he year. Counry effecs are a imes significan, bu no as consisenly so as he monh effecs. The resuls for he oher conrol variables show ha disagreemen is indeed persisen. The own lag has saisically significan and sizable coefficiens. As hese are significanly smaller han one, disagreemen is, however, no described by a uni roo process. Also he condiional volailiy of a variable is apparenly an imporan deerminan of disagreemen, wih large and (for all forecass) saisically significan effecs. Oil prices, in conras, seem o affec only a number of forecass, such as ineres rae forecass a he 3-monh horizon and inflaion forecass for he subsequen year. In hese cases, larger oil price changes do, as o be expeced, increase disagreemen. Turning o our variables of ineres, he cenral bank ransparency and communicaion measures, i is apparen ha here is a large majoriy of cases where more ransparency reduces disagreemen. 12 In many cases he effec is saisically significan, and he coefficiens are furhermore always negaive. The larges effecs are exered by he adopion of quanified inflaion objecives and by increasing he level of economic ransparency: wih a quanified inflaion objecive, he iner-quarile range of forecas 11 This assumes ha errors are correlaed wihin, bu uncorrelaed across counries. For he euro area counries, i migh be possible ha errors are also correlaed across counries, given ha he cenral bank measures are parially idenical. We have herefore esed wheher allowing for correlaed errors wihin he non-euro area counries and wihin he euro area affecs our resuls, and noe ha his is no he case. If anyhing, he sandard errors when clusering by counry are slighly more conservaive, and are herefore repored in his paper. 12 I migh be surprising o find ha he disagreemen in CPI forecass is no affeced by all cenral bank ransparency measures. A he same ime, however, i is imporan o noe ha moneary policy affecs inflaion only wih raher long lags, which are ofen esimaed in he range of 1.5 o 2 years. The forecas horizons in his daase migh herefore be oo shor o reflec possible effecs of enhanced cenral bank ransparency. 15

17 dispersion is reduced by 7 o 19% (depending on he variable we look a), and an improvemen in he economic ransparency index by one lowers he iner-quarile range by 6 o 20%. This compares o a reducion of up o 5% if he overall ransparency index increases by one, and of up o 10% in he cenral bank communicaion monhs. In addiion, hese effecs are parially addiive, especially when i comes o he ineres rae forecass. The models in column (5) show ha in many cases, here are join effecs of quanified inflaion objecives, economic ransparency and/or communicaion, which add o larger reducions han esimaed in he separae models. Anoher imporan resul is ha he dispersion-reducing effec, which is apparen for he overall ransparency index is nearly enirely generaed by he conribuion of economic ransparency. Consrucing an overall ransparency index excluding economic ransparency, and reesimaing he economeric models shows ha he oher dimensions of ransparency manage o reduce dispersion only in he case of long-erm ineres rae forecass. These resuls are robus o a number of model variaions, such as using he sandard deviaion insead of he iner-quarile range, or allowing for correlaed errors wihin he non-euro area counries and wihin he euro area (i.e. dropping he assumpion of uncorrelaed errors across he euro area counries). Similary, a reclassificaion of he Deusche Bundesbank (prior o 1999) as a cenral bank wih a quanified inflaion objecive, given ha i had released he so-called inflaion norm, does no affec resuls. Wha is imporan, hough, is he reamen of he. Excluding he saff projecions from he communicaion variable, or dropping he from he se of cenral banks wih quanified inflaion objecives (such ha all remaining cenral banks in he group are defined as formal inflaion argeers) implies much reduced effecs for he communicaion variable and he quanified inflaion objecive on dispersion, respecively. Agains he background ha 5 of our 12 counries in he sample are members of he euro area, his change in resuls is no all oo surprising, and suggess he following. Firs, he saff projecions, alhough for he euro area, are useful informaion in predicing he main macroeconomic variables in he euro area naional economies. Second, for he purpose of aligning agens expecaions of hese variables, wha maers is no ha a cenral bank follows a formal inflaion argeing policy, bu raher ha i announces a quanified inflaion objecive. For breviy, hese robusness ess are no repored in he paper. Anoher robusness es is conained in Table 5. As menioned above, we have expanded he sample for he ransparency indices beyond hose originally available, by exrapolaing using he firs observaion forward, i.e. for he years prior o 1998, and he las observaion backward, for 2007 and Using only he original daa covering he years 1998 o 2006 reduces sample size by around half. Accordingly, here are fewer saisically significan effecs. However, for a number of forecass, saisical significance remains, and wih only one excepion, hese cases poin o a reducion in forecaser disagreemen when ransparency is enhanced. This is paricularly he case for he ineres rae forecass. TABLES 5 AND 6 HERE The final robusness es is provided in Table 6. Here, we aemp o explain he mean absolue error, i.e. a measure of forecas accuracy, raher han he iner-quarile range, our measure of dispersion, in oherwise idenical models. As we had conjecured in he preceding secion, here is some, bu weaker evidence ha cenral bank ransparency and communicaion improves (ex pos) forecas accuracy. Taken ogeher, hese resuls sugges ha here are various ways hrough which cenral banks can reduce forecas dispersion, wih he announcemen of a quanified inflaion objecive and enhanced economic ransparency being he mos poen channels. Addiional effecs can be achieved in he monhs when cenral banks communicae heir inflaion and oupu forecass. 3.2 The response o shocks Wha are he underlying mechanisms by which enhanced cenral bank ransparency and communicaion leads o a convergence in he views of he privae secor? We ry o shed ligh on his quesion by 16