ASYMMETRY AND INFLATION DYNAMICS IN DIFFERENT SPECIFICATIONS OF THE PHILLIPS CURVE FOR THE EURO AREA

Transcription

1 ASYMMETRY AND INFLATION DYNAMICS IN DIFFERENT SPECIFICATIONS OF THE PHILLIPS CURVE FOR THE EURO AREA Maria Paloviia and David Mayes Bank of Finland Absrac This paper sudies he differences in inflaion dynamics in he euro area and eleven EMU economies by comparing he empirical performance of wo oupu gap based Phillips relaions: he expecaions augmened Phillips curve and he New Keynesian Phillips curve. Direc measures of inflaion expecaions are used from OECD forecass which offers some advanages over simply imposing raional expecaions. No only do we find ha inflaion expecaions are cenral o he inflaion process in all eleven EMU counries bu ha here is very considerable heerogeneiy of inflaion dynamics. Boh models correspond quie well o he daa, aided perhaps by he persisence in expecaions. However, in mos counries i appears ha he New Keynesian approach may offer he beer explanaion. The difference beween he wo specificaions has imporan implicaions for moneary policy and for he inflaion dynamics policy generaes, wih a greaer immediae effec in he New Keynesian case. Given he subsanial differences in responsiveness among he member saes and differences in oupu gaps, his has considerable implicaions for he seing of he single moneary policy. Inflaion dynamics have been heavily debaed over he years. Three main facors have recenly reignied ineres. The firs is he major changes in he conjuncure of low inflaion and real economic growh in many indusrialised counries in he 1990s (i.e. behaviour seems o have changed). Secondly, he policy problem has changed and new moneary insiuions have emerged. Economic and Moneary Union in Europe has led o he esablishmen of a single moneary policy ha has o cope wih srikingly differen processes of price developmen in he welve member saes. Lasly, recen heoreical advances have produced alernaive views of he dynamic process, which have crucially differen implicaions for he design of opimal moneary policy. In paricular we can conras he implicaions of he expecaions augmened Phillips curve in he radiion of Phelps (1967), Friedman (1968) and Lucas (1976) wih he New Keynesian Phillips curve of Taylor (1980), Calvo (1983) and Fuhrer and Moore (1995). Our ineres in his paper is hree-fold. In he firs place we wish o esablish empirically wha hese wo main alernaive specificaions of he Phillips relaion imply for he dynamics of inflaion in he euro area. Secondly, we wan o compare he wo approaches in a common framework, which involves making assumpions abou he deerminaion of he oupu gap. Finally, in he ligh of he variaion across he welve member counries and across he wo models we seek o explore he implicaions for he single moneary policy ha has o be decided for he euro area by he Governing Council of he European Cenral Bank. 1 1 Greece had o be omied from esimaion due o lack of daa. 1

2 To pre-judge our resuls, boh versions of he Phillips relaion do a reasonably good job in accouning for inflaion dynamics in he euro area. Saisical ess appear o favour he New Keynesian approach as an overall explanaion in he euro area and mos individual counries bu he magniude and significance of some of he coefficiens are more problemaic. The difference in slopes of he Phillips curve under eiher specificaion is subsanial across he member saes, implying ha he same oupu gap would have srikingly differen implicaions for inflaion across he euro area. The persisence of boh inflaion and he oupu gap, while also showing considerable variaion across counries, is no so divergen. However, combining he wo facors o show he dynamics of inflaion exacerbaes he differences boh among he member saes and beween he wo models. Wha is paricularly noable is he difference beween he funcions esimaed a he euro area level and hose for he individual counries. In he expecaions augmened model he euro area Phillips relaion is much more responsive han i is for he wo larges economies, Germany and France. Whereas, in he New Keynesian case euro area inflaion is so persisen ha he influence of he Phillips relaionship is small and insignifican if he GDP deflaor is used. Thus if euro area policy were o be based on hese relaionships he wo specificaions would lead o differen conclusions and ones ha could be considerably differen from ha implied by behaviour in individual counries. Expecaions play a cenral role in comparing he wo heories. Since we are ineresed in pracical policy implicaions we require a measure of expecaions ha is available a he ime of decision-making. Insead of imposing model-based raional expecaions, we use a direc measure of inflaion expecaions as an empirical proxy. We ake he published OECD inflaion forecass o represen prevailing inflaion expecaions in wage and price formaion. 2 They need no necessarily be raional alhough hey may be. The Phillips curves are fied o daa for he period by a single equaion generalised mehod of momens. Conrary o many oher empirical sudies, he oupu gap appears o be an adequae empirical measure of cyclical inflaionary pressure in he Phillips relaion. The paper proceeds as follows. Secion 1 derives he wo Phillips relaions ha we esimae and places hem in he conex of exising empirical sudies of he new Keynesian specificaion. Secion 2 presens he empirical framework, esing sraegy and he daa. Secion 3 considers he empirical resuls for he euro area and eleven EMU counries and explores heir implicaions for inflaion dynamics and he moneary policy of he euro area. Secion 4 concludes. 1 Two Phillips Curve Specificaions This secion presens he wo alernaive Phillips curve specificaions, which are applied in his sudy, aking he expecaions-augmened Phillips curve is described firs. The approach o modelling he Phillips curve has been heavily influenced by hisory. Reducing inflaion has been seen o be cosly. Problems in credibiliy and hus expecaions formaion have helped lead o subsanial employmen and oupu losses in he process of disinflaion. While research on inflaion dynamics has been largely based on he Phillips curve since he lae 1950s, is economic modelling has changed considerably over he years. Originally Phillips (1958) and 2 Similar sudies wih survey based expecaions have been done by Robers (1997, 1998) for he US economy. 2

3 Samuelson and Solow (1960) hypohesised a sable negaive relaionship beween unemploymen and inflaion wihou paying special aenion o he role of expecaions. This sabiliy seemed o disappear and abou en years laer Phelps (1967) and Friedman (1968) developed he expecaionsaugmened Phillips curve wih a cenral role of expecaions in he inflaion process hrough wage bargaining and price seing o reinsae i. In urn in he 1970s, Lucas (1976) offered a 'raional expecaions' approach, o overcome he sysemaic deviaion of inflaion expecaions from acual inflaion ha had been occurring in many formulaions as inflaion rose. In laer empirical work on he expecaions-augmened Phillips curve raional expecaions are ypically assumed and real economic aciviy is measured by acual oupu relaive o poenial, he oupu gap. Recen advances in he heoreical modelling of inflaion dynamics have led o a specificaion labelled he New Keynesian Phillips curve, which is based on work by Taylor (1980) and Calvo (1983). This heory has been buil explicily from microfoundaions. I suggess ha prices are sicky and inflaion depends enirely on curren and expeced fuure economic condiions. Currenly expeced fuure inflaion deermines inflaion in his heory, conrary o he expecaions-augmened Phillips relaion, where inflaion is dependen on previously expeced curren inflaion. In he New Keynesian heory, excess demand eners hrough real marginal coss, he empirical measure of which in he lieraure is ypically he oupu gap or real labour coss (labour income share). 1.1 The Expecaions-augmened Phillips Curve The expecaions-augmened Phillips curve can usefully be presened using Fischer s (1977) model wih one-period wage conracs. In his model expeced inflaion plays a cenral role in wage bargaining, as nominal wages are predeermined a he beginning of every period. However, oupu and he price level can hen adjus during each period as a reacion o random disurbances affecing he economy. If he goal of wage seing is assumed o be a given real wage, he nominal wage conrac for period depends on he expeced price level in he following way: 1W 1 p (1) where 1 W is he logarihm of period nominal wage, whose conrac has been made a he end of period 1. The erm 1 p is he expeced price level for period, according o he expecaion formed a he end of period 1. For simpliciy, he scale facor,, is henceforh assumed o be zero. In his framework he aggregae supply relaion beween he price level, he expeced price level and excess demand can be rewrien in erms of inflaion, expeced inflaion and excess demand. Hence, wih one-period wage conracs he expecaions-augmened Phillips curve can be expressed as 1 ŷ (2) where denoes he period inflaion rae, defined as he rae of change of prices from period 1 o period. The erm 1 refers o he corresponding expeced rae of inflaion i.e. he expeced curren inflaion in period. The erm ŷ denoes he period excess demand. In he expecaionsaugmened Phillips relaion inflaion expecaions perain o he curren inflaion. In principle, he heory allows for differen expecaions formaion mechanisms. Saic, adapive and raional 3

4 expecaions have been posulaed in empirical sudies. The sandard formulaion wih raional expecaions is E 1 ŷ (3) where E 1 is he expecaions operaor condiional on informaion available in period 1. The mean value of forecas errors is zero under raional expecaions. The oupu gap, which is here defined as he difference in logarihms of acual and poenial oupu, is ofen used o indicae he excess demand in he economy. Oher measures, like he capaciy uilisaion rae and he unemploymen gap, which denoes he difference beween he acual rae of unemploymen, and he non-acceleraing inflaion rae of unemploymen, he NAIRU, have also been used. Alernaive specificaions may include addiional lags of excess demand. The expecaions-augmened Phillips curve approach has suppor from many empirical sudies, see for example King and Wason (1994) and Lown and Rich (1997). In applied work, oil price and oher addiional variables have ofen been incorporaed o he model in order o capure he supply shocks of he 1990s. 1.2 The New Keynesian Phillips Curve The new Keynesian approach has incorporaed recen mehodological advances in dynamic general equilibrium heory. In his approach, inflaion is wholly explained by he curren and expeced fuure economic condiions, and non-neural effecs of moneary policy are based on emporary nominal price rigidiies. Furhermore, he heory uses explici microfoundaions, assuming firms are monopolisically compeiive and nominal price seing is saggered. Price-seing decisions of individual firms are derived from an explici opimisaion problem, as every firm maximises profis subjec o consrains on he frequency of price adjusmens or alernaively subjec o menu coss relaed o changing prices (Roemberg, 1982). Generally, aggregaing he decision rules of firms is raher complicaed when nominal price seing is saggered, bu i can be simplified following he approach in Calvo (1983), where each firm has a fixed probabiliy (1 ) of changing is price in any period. Correspondingly, is a probabiliy of keeping he price fixed. The probabiliy of adjusing he price is independen of he lengh of ime since he las price change. The expression 1/(1 ) describes he average ime a price is fixed. Thus, nominal rigidiy increases, as rises. Aggregaing across he price seing decisions of he individual firms yields he following linearised relaionship beween curren inflaion, expeced fuure inflaion and real marginal coss: 1 mc E (4) where mc is he period log deviaion of he firms real marginal coss from he seady sae value and expecaions are condiional on informaion available in period. Thus, in his specificaion, inflaion is dependen on currenly expeced fuure inflaion. The parameer is he subjecive discoun facor and he coefficien on real marginal coss, he erm, is a funcion of wo srucural parameers of he model in he following way: = (1 )(1 )/, where is he degree of price 4

5 rigidiy. Because is decreasing in, he longer prices are fixed on average, he less sensiive inflaion is o curren variaion in real marginal coss. In his framework forward-looking firms have o se prices for possible muliple periods and base heir pricing decisions on he expeced fuure developmens of marginal coss. Ieraing equaion (4) forward yields: k E mc k. (5) k 0 As inflaion is equal o he discouned sream of fuure marginal coss, i is enirely forward looking wih no persisence in i. The credibiliy issues of moneary policy are relevan in his framework. Wih credible moneary policy inflaion can be reduced wihou oupu losses via expecaions of fuure moneary policy. In empirical sudies, he oupu gap is a commonly used proxy for he real marginal coss, bu labour coss have also been used. These variables are assumed o capure changes in real marginal coss associaed wih variaion in excess demand in he economy. Under cerain assumpions abou echnology, preferences and he srucure of labour markes we can link he oupu o real marginal coss wihin a local neighbourhood of he seady sae of log real marginal coss according o: mc ˆ y (6) Subsiuing he above relaion ino (4), we obain he ypical specificaion for he New Keynesian Phillips relaion: where =. yˆ, E 1 (7) 1.3 Evidence from Earlier Empirical Sudies Alhough he oupu gap based New Keynesian Phillips curve under raional expecaions has ofen been used in moneary policy analysis, he empirical validiy of he model has no been firmly esablished. Inflaion seems o be poorly capured and he oupu gap is ofen insignifican or has a wrong, negaive, sign. To represen he daa beer, he model has ypically been modified, using he lagged inflaion rae and alernaive measures of real economic aciviy. The real marginal coss implied by he New Keynesian heory are difficul o measure empirically. This has lead o a debae in applied work over wheher he oupu gap or labour coss is he appropriae measure of cyclical inflaionary pressure. The oupu gap has been criicised for having problems in measuremen and i may no move proporionally wih real marginal coss due o he failure o accoun for labour marke fricions. Moreover, he relaion beween unobservable firm-level marginal coss and observable aggregae marginal coss appears o be problemaic and he esimaion resuls seem o be highly sensiive o he specificaion of labour coss. Empirical sudies have raised quesions abou he adequacy of he underlying heory. The heoreical model under raional expecaions does no imply any lags of inflaion. Such lags have 5

6 been inerpreed as represening agens who only look backwards seing prices. The poor empirical fi may hus be associaed wih he possible inaccuracy of he raional expecaions hypohesis assumpion. Fuhrer and Moore (1995) have shown wih US daa ha he New Keynesian model under raional expecaions hypohesis wihou lags canno capure inflaion dynamics. In he recen sudy by Gagnon and Khan (2001) inflaion responds beer o anicipaed movemens in labour coss han o he oupu gap. When he oupu gap is used in ha sudy, he esimaed slope of he New Keynesian Phillips curve is negaive for he euro area and he US bu posiive for Canada. When lagged inflaion is added o he oupu gap based model, he slope of he curve is posiive for he euro area and Canada, bu negaive (wrong) for he US. In mos of he oupu gap based esimaions by Gagnon and Khan, he esimaed coefficien on oupu gap iself is insignifican. Gali and Gerler (1999) conclude ha he New Keynesian Phillips curve provides a reasonable approximaion o inflaion dynamics in he US, when he labour income share is used o measure real marginal coss. Alhough backward looking price seing is saisically significan in heir sudy, i is no quaniaively imporan. Gali, Gerler and Lopez-Salido (2001) favour he New Keynesian Phillips curve condiional on real uni labour coss insead of he oupu gap boh for he euro area and he US. While here seems o be some backward looking in inflaion dynamics, hey find ha he New Keynesian Phillips curve fis he euro area daa. Sbordobe (2002) argues also ha he New Keynesian Phillips curve performs beer when real labour coss are used. By conras, Rudd and Whelan (2002) argue ha he empirical fi of he New Keynesian model is no improved when using real labour cos insead of he oupu gap. Benigno and Lopez-Salido (2002) have compared inflaion dynamics in five major EMU counries. When sudying France, Germany, Ialy, he Neherlands and Spain hey provide evidence on heerogeneiy in price changes across he counries. This sudy suggess ha inflaion has a dominan forward looking componen only in Germany and here is a significan backward looking componen in inflaion process in he oher four counries. Robers (1997, 1998) has analysed inflaion dynamics in he US wih he New Keynesian specificaion by using survey esimaes of inflaion expecaions insead of raional expecaions assumpion. He finds evidence ha inflaion expecaions are no raional, which appears o be conneced o he poor empirical fi of he New Keynesian heory. However, he resuls are highly dependen on inflaion surveys, which have been criicised as being unreliable esimaes of inflaion expecaions. 2 Empirical Evidence 2.1 The Empirical Framework In order o make he wo Phillips curve specificaions as comparable as possible for he euro area daa, we have applied he same mehod of operaionalising expecaions and he same measure of excess demand o boh cases. The role of expecaions is cenral when comparing he wo heories. The expecaionsaugmened Phillips curve and he New Keynesian Phillips curve differ in a way expecaions are linked o inflaion process. In he expecaions-augmened specificaion curren inflaion is relaed 6

7 o he previously expeced curren inflaion, as shown in (3). We have modified his slighly in applying i o he daa, o allow us o es for non-neuraliy in inflaion process: E 1 ŷ. (3') Under neuraliy he parameer = 1. I also makes he model direcly comparable o he new Keynesian specificaion. In he New Keynesian specificaion (7), curren inflaion is dependen on he currenly expeced fuure inflaion. In his case he parameer is he discoun facor, which is less han bu very close o uniy. Inflaion varies posiively wih he oupu gap in boh specificaions and he parameers and measure he sensiiviy of inflaion o variaion of excess demand. In he New Keynesian model he fuure oupu gap changes are assumed o capure he fuure changes in real marginal coss exacly. We used wo inflaion measures in esimaion: he annual changes of he GDP deflaor and he privae consumpion deflaor, because boh measures are widely used in he exising lieraure. The corresponding OECD forecass are used o measure inflaion expecaions. By using direc measures of inflaion expecaions, we can avoid he problem faced by many previous sudies of inflaion dynamics, of having o es dual hypoheses, abou he specificaion of he Phillips curve and he formaion of expecaions, a he same ime. Thus, in our sudy we can allow for he possibiliy ha he expecaions hemselves may adjus slowly or move for spurious reasons. The OECD forecass are likely o be more reliable proxies for inflaion expecaions han some survey esimaes ha have been used, as hey are based on sysemaic monioring of economic developmens and economeric models. While using hese proxies for inflaion expecaions, we can also es wheher he lagged inflaion rae is needed o improve he empirical fi of he Phillips relaion. We use he encompassing es of Davidson and MacKinnon (1993) o analyse wheher he expeced fuure inflaion or he expeced curren inflaion dominaes inflaion process: 1 (1 )E ŷ E 1. (8) Equaion (8) includes boh expecaions variables (from (3') and (7)) and hence encompasses boh models under consideraion as special cases. If he hypohesis (1 ) = 0 is no rejeced and he hypohesis = 0 is rejeced, he inflaion process is dominaed by he currenly expeced fuure inflaion. Accordingly, in he converse case, i.e. he hypohesis = 0 is no rejeced and he hypohesis (1 ) = 0 is rejeced, he model reduces o he expecaions-augmened Phillips curve wih he previously expeced curren inflaion. If he sum of he esimaed coefficiens is resriced o one, i is possible o analyse relaive weighs of he alernaive componens in he inflaion process. The wo Phillips relaions can also be compared using a non-nesed es (Davidson and MacKinnon, 1993). This es analyses wheher one Phillips curve specificaion has explanaory power over he oher, which would indicae ha he second is misspecified. However, compared o he ess based on parameer resricions wih cerain assumpions abou disribuions, non-nesed ess are subsanially weaker. A bes hey only indicae which model fis daa beer. In applying he non-nesed es o he wo Phillips relaions, we express he wo models of he alernaive Phillips curve specificaions in a manner ha allows us o es hem boh agains a more general specificaion: 7

8 E ˆ ˆ ( 1 ) 1 y (9) and E ˆ ( 1) y ~ 1 (10) where he erms ˆ and ~ denoe he fied values of he single equaion esimaes of he expecaions-augmened Phillips curve and he New Keynesian Phillips curve, respecively. The validiy of he alernaive specificaions can be analysed by combining he es resuls. As in all such non-nesed ess, hree classes of oucome are possible: one model is beer suppored by he daa han he oher; neiher is saisfacorily suppored and lasly boh are suppored. For example, if he hypohesis = 0 is no rejeced in (9), he expecaions-augmened specificaion has no explanaory power over he New Keynesian one. Accordingly, if he hypohesis = 0 is rejeced in (10), he New Keynesian specificaion has explanaory power over he oher one. So, when considering he resuls of boh equaions we can conclude ha he New Keynesian specificaion suppors evidence agains he expecaions-augmened one in his example. I he hird case where boh hypoheses are suppored ha provides he greaes problem. I could simply mean ha here is nohing much o choose beween he expecaions-augmened and he New Keynesian Phillips relaions. However, i may be ha he paricular daa sample is inadequae for esing inflaion dynamics beween he wo Phillips curves. Since he implicaions of he wo relaions for he conduc of moneary policy are noiceably differen such an oucome would be unhelpful. 2.2 The Daa The annual daa cover he period For all counries, he inflaion raes and he oupu gap series are consruced using he OECD Economic Oulook daa se. Two alernaive inflaion measures are used, he annual changes of he GDP deflaor and he privae consumpion deflaor. The corresponding OECD inflaion forecass for every counry have been colleced from he OECD Economic Oulook publicaions. The OECD makes forecass wice a year. Here, he December esimaes for he nex year are considered. Thus, in he expecaions-augmened Phillips curve specificaion, in (3'), he previous year s December forecass for he curren year are used. Similarly, he curren year s December forecass for he nex year measure inflaion expecaions in he New Keynesian specificaion (7). The oupu gap is defined as he difference beween he log real GDP and he HP-filered log real GDP. When aggregaing inflaion raes, inflaion forecass and he oupu gap for he euro area, he ECB counry weighs for he eleven EMU counries have been used. 4 The four bigges economies, Germany, France, Ialy and Spain dominae he euro area wih a weigh of over 80 per cen. 3 The daa for Luxembourg are available only for he period The inflaion raes and he inflaion forecass were aggregaed from individual counry series by using he ECB counry weighs for 11 EMU counries. The percenage weighs from 1999 are: Germany 32.4, France 22, Ialy 18, Spain 9.1, The Neherlands 6, Belgium 3.8, Ausria 3.2, Finland 2, Porugal 1.7, Ireland 1.4 and Luxembourg 0.3. The oupu gap: he log real aggregaed GDP minus he HP-filered log real aggregaed GDP. Real GDP was aggregaed 8

9 Figures 1 and 2 show ha he inflaion raes and inflaion forecass for he GDP and privae consumpion deflaors for he euro area, while having a similar paern, differ in imporan respecs. In he beginning of he 1980s inflaion was abou 10 per cen, bu i decreased o approximaely 3 per cen during he nex few years. Since a small peak of abou 4.5 per cen in he beginning of he 1990s inflaion has remained subdued in he euro area. When looking a he inflaion forecass for he curren year, i can be seen ha he decreasing inflaion was overprediced in he firs half of he 1980s. In addiion, he acceleraing inflaion in he lae 1980s was no direcly capured by he forecass. Wih low and sable inflaion, recen inflaion forecass are close o acual inflaion. However, he privae consumpion deflaor has been he more difficul o predic, including he upurn in Euro area real oupu and poenial oupu show a simpler paern (Figure 3). Before he end of he 1980s real growh was quie modes and he oupu gap was negaive. In he lae 1980s and in he beginning of he 1990s oupu was above poenial and economic growh was fairly brisk. Afer ha he oupu gap was negaive again unil he year 1999 because of slow growh in many euro area economies. The figures for he individual counry variables (shown in Appendix Figure A1) are much more complex. We have divided he counries ino wo groups, boh o make each counry disinguishable and o highligh he ouliers. The inflaion hisory varies clearly across he EMU counries, especially in he 1980s, Porugal being he main oulier. In he 1980s and 1990s he oupu gap pah was quie similar in he seven bigges euro area economies wih he excepion Germany, where we can see he effec of unificaion. Compared o hese counries, he four smalles economies have experienced quie differen and more diverse developmens in he oupu gap, wih Finland being he obvious oulier wih is crisis a he beginning of he 1990s. Figure 1. Euro area inflaion and inflaion forecass, using he GDP deflaor The inflaion rae The previous year's inflaion forecas for he curren year The curren year's inflaion forecas for he nex year using ECB counry weighs for 11 EMU counries and aking ino accoun he unificaion in Germany by spliing he HP-filering for Germany in wo. Real GDP up o 1990 was exended by hree years by excluding he Eas German share of 6.5 per cen from he combined real GDP series. Similarly, real GDP since 1991 (for he unified Germany) was exended backwards by hree years by adding he share of Eas Germany o he original series. An HP-filer was hen applied o each series separaely before combining hem, using he firs series up unil 1990 and he second herafer. For he years 1981 and 1982, he missing Luxembourg daa were assumed o have he same characerisics as Belgium. 9

10 Figure 2. Euro area inflaion and inflaion forecass, using he privae consumpion deflaor The inflaion rae The previous year's inflaion forecas for he curren year The curren year's inflaion forecas for he nex year Figure 3. Euro area real oupu and poenial oupu Resuls Real oupu Poenial oupu 3.1 Single Equaion Esimaion The parameers for boh Phillips relaions were esimaed separaely using he generalised mehod of momens (Hansen, 1982). In boh cases he insrumens used are chosen o represen variables which are predeermined a ime. For comparabiliy, he insrumen se for boh specificaions consiss of he same hree variables: he lagged inflaion forecas, E 1 { }, and wo lags of he oupu gap, y ˆ 1 and y ˆ 2. Thus lagged inflaion expecaions are reaed as exogenous (predeermined) and he oupu gap is he only endogenous righ hand side variable in he expecaions-augmened Phillips curve. The sandard errors of he esimaed parameers are 10

11 modified using he whie-covariance correcion. The J-saisic is used o es he overidenifying resricions ie validiy of he insrumens. Overall, he esimaion resuls using he GDP deflaor (shown in Table 1) are fairly plausible for he euro area and for individual economies. The esimaed coefficiens on he expeced inflaion are higher in he New Keynesian specificaion wih only wo excepions. In he expecaions-augmened specificaion none of he poin esimaes is significanly over uniy, bu some raher low values (clearly less han 0.9) can be found. By conras, in he New Keynesian specificaion he poin esimaes for Ialy, Porugal and Spain exceed uniy and he hypohesis of = 1 is rejeced a he 5 per cen level. In he New Keynesian specificaion, only he poin esimae for he Neherlands is below For he euro area i is plausible a Table 1. Single equaion esimaion resuls using he GDP deflaor as he measure of inflaion Expecaions-augmened specificaion π π βe 1 λyˆ Parameers Tes J EU (0.035) * (19.078) * Ausria (0.077) * (26.107) * Belgium (0.059) * (19.956) * Finland (0.062) * (13.197) France (0.043) * (9.231) Germany (0.062) * (10.575) Ireland (0.132) * (27.599) Ialy (0.051) * (37.412) * Luxembourg (0.163) * (45.517) * Neherlands (0.062) * (21.799) * Porugal (0.071) * (32.983) Spain (0.043) * (15.885) New Keynesian specificaion π κyˆ π βe 1 Parameers Tes J EU (0.021) * (10.132) Ausria (0.057) * (17.822) Belgium (0.064) * (20.041) Finland (0.019) * (6.010) France (0.031) * (8.384) Germany (0.050) * (8.159) Ireland (0.082) * (23.068) Ialy (0.013) * (9.984) * Luxembourg (0.174) * (32.036) * Neherlands (0.036) * (11.898) Porugal (0.047) * (21.732) Spain (0.043) * (8.683) Noes: Sample period wih he excepion of Luxembourg: Numbers in parenheses are sandard errors, * indicaes significance a 5 per cen level. J-saisic corresponds o he Hansen es of he overidenifying resricions. The J-saisic imes he number of observaions is asympoically 2 wih one degrees of freedom, he criical value of which is Insrumens: he lagged inflaion expecaion, E 1{ } and he wo lags of he oupu gap, ŷ 1 and ŷ 2. 11

12 The esimaed oupu gap eners always wih a posiive sign when he expecaions-augmened model is considered. In he New Keynesian specificaion he coefficien on he oupu gap is posiive for he euro area and for eigh ou of eleven counries. The esimaed coefficiens on he oupu gap are clearly higher in he expecaions-augmened specificaion compared o he oher case. Hence, inflaion appears o be more sensiive o changes in curren excess demand, when expecaions are measured by he inflaion forecas for he curren year insead of he inflaion forecas for he nex year. This is in par because he excess demand effec is incorporaed in expeced fuure inflaion. A more direc comparison of he inflaion dynamics accouning for difference in lag srucures in he models follows in Secion 3.3. The saisical reliabiliy of he oupu gap coefficiens is greaer in he expecaions-augmened specificaion for he euro area as well as for almos all counries. The resuls are slighly differen, when using he privae consumpion deflaor as he measure of inflaion, as shown in Table 2. The esimaes of he expecaions variable coninues o be higher wih he New Keynesian specificaion. Alhough he esimaed coefficien on he expeced inflaion for he euro area is oo high in he New Keynesian specificaion, 1.022, i is no saisically differen from uniy. The hypohesis of = 1 is rejeced a he 5 per cen level in he New Keynesian specificaion for he same hree counries: Ialy, Porugal and Spain, as for he GDP deflaor. Wih he privae consumpion deflaor, all he esimaed oupu gap coefficiens are posiive in he expecaions-augmened specificaion and here is only one esimae wih he wrong sign in he New Keynsian case. The expecaions-augmened specificaion gives noably higher and saisically more reliable coefficiens for he oupu gap han for he New Keynesian specificaion. The overidenifying resricions are rejeced, according o he J-saisic, in only a few cases in Tables 1 and 2. The poenial weakness of he insrumens was also checked by regressing he regressors of he Phillips curves on he hree insrumens; see Appendix Table A1. In he case of expeced fuure inflaion, in over half of he cases R-squared from he regressions was above 0.80 and only in one case below For he oupu gap only 8 ou of 24 of he R 2 s were below Thus, overall, he chosen insrumens seem o be relevan o he variables we seek o model. Table 2. Single equaion esimaion resuls using he privae consumpion deflaor as he measure of inflaion Expecaions-augmened specificaion π βe 1 π λyˆ 12 Parameers Tes J EU (0.043) * (18.566) * Ausria (0.088) * (33.707) Belgium (0.092) * (31.093) Finland (0.058) * (6.479) * France (0.055) * (11.416) * Germany (0.076) * (10.137) * Ireland (0.105) * (26.085) Ialy (0.042) * (24.650) * Luxembourg (0.077) * (28.803) Neherlands (0.053) * (16.117) * Porugal (0.077) * (43.146) Spain (0.029) * (12.593) * 0.041

13 New Keynesian specificaion π κyˆ π βe 1 Parameers Tes J EU (0.023) * (10.840) Ausria (0.064) * (23.794) Belgium (0.084) * (27.230) Finland (0.038) * (4.971) * France (0.044) * (10.630) Germany (0.064) * (6.118) * Ireland (0.086) * (16.323) Ialy (0.025) * (19.182) * Luxembourg (0.065) * (18.613) Neherlands (0.030) * (10.845) * Porugal (0.051) * (33.272) * Spain (0.034) * (9.161) Noes: See Table 1. Boh Phillips curve specificaions wih boh inflaion measures show reasonable inflaion dynamics for his daa se, where he oupu gap is a proxy for excess demand and inflaion expecaions are measured direcly. Qualiaively similar esimaion resuls are obained from single equaion esimaion wih ordinary leas squares (see Appendix 4 of Paloviia (2002)) and pooled leas squares mehods, as repored in Appendix Table A2. The esimaes of he impac of expeced inflaion are greaer when he New Keynesian specificaion is applied. Whereas, while mos coefficiens on he oupu gap are posiive, hey are higher and more reliable wih he expecaionsaugmened specificaion. The resuls are comparable wih he earlier sudies of inflaion dynamics in he euro area. The oupu gap based Keynesian Phillips curve under raional expecaions in Gali, Gerler and Lopez- Salido (2001) and Gagnon and Khan (2001) show he same esimaed coefficien on he expeced inflaion, 0.991, as we obain, for he case of he GDP deflaor. However, he slope of he Phillips curve is negaive in boh of he sudies. We hus find more supporive evidence for he empirical validiy of he oupu gap. The slope of he Phillips curve is posiive in 44 of he 48 cases for he euro area and he individual counries ha we show in Tables 1 and Comparison of he Two Hypoheses On he basis of he resuls repored above, a clear saisical preference canno ye be claimed for eiher of he Phillips relaions. Boh of hem seem o capure inflaion dynamics fairly well. We herefore urn o he wo ess we described in Secion 2.1. The encompassing es (8), shown in Appendix Tables A3 and A4, indicaes ha he currenly expeced fuure inflaion seems o dominae he inflaion process compared wih he previously expeced curren inflaion for he euro area wih boh inflaion measures. 5 In oher words he weigh on is clearly larger han ha on 1 -. (The weighs are consrained o sum o uniy in he regression.) Similar resuls are found for mos of he individual counries. However, he opposie 5 The insrumens are he same as before, ie he lagged inflaion forecas and wo lags of he oupu gap. 13

14 resul occurs for he same hree counries ha exhibied a poor empirical fi for he New Keynesian Phillips curve. The encompassing es indicaes ha he larger weigh falls on he previously expeced curren inflaion for Ialy and Spain under boh inflaion measures and for Porugal wih he GDP deflaor. Some of he esimaed oupu gap coefficiens are negaive, for example, ha for he euro area wih he GDP deflaor. Overall, he J-saisics indicae ha he insrumen se for encompassing es is accepable is almos all cases. The encompassing es was also applied o he euro area and individual counries wihou imposing uni sum consrain bu he resuls (no repored here) were qualiaively similar. Thus, aken ogeher, he encompassing ess appear o be more favourable o he New Keynesian specificaion. However, because of he number of implausible coefficiens and he variaion in resuls beween he inflaion measure, his es may no be a very reliable means of evaluaion of he relaive meris of he wo Phillips relaions. However, he same conclusions appear from he non-nesed es derived from (9) and (10) as shown in Appendix Tables A5 and A6. 6 Secion A of he boh ables ess he expecaionsaugmened model agains he general model while secion B reverses he procedure and ess he New Keynesian model agains he general model. Secion C summarises he oucomes. The New Keynesian Phillips curve is preferred in he euro area under boh inflaion measures, as he expecaions-augmened specificaion has no explanaory power over he New Keynesian bu he new Keynesian specificaion has explanaory power over he expecaions-augmened. Similar resuls can be seen for mos individual counries, wih 9 (8) ou of 11 counries for he GDP (privae consumpion) deflaor showing a significan resul for he second hypohesis. By conras, his resul is obained for he firs hypohesis in only wo of he counries wih boh inflaion measures. So if we combine he wo ses of resuls as in panel C of he Tables, boh panels A and B come ou in favour of he New Keynesian version in seven ou of eleven cases for he GDP deflaor (six ou of eleven for he consumpion deflaor). For all oher cases he resuls are mixed in four cases boh hypoheses are preferred and in five, neiher. The expecaions-augmened specificaion is no favoured for any counry wih eiher deflaor. 3.3 A Direc Comparison of he Price Dynamics As we noed above i is no possible o compare he slopes of he Phillips curves in he New Keynesian case jus by comparing he esimaes of from (7) because he expeced fuure inflaion erm is no predeermined bu also depends upon curren economic aciviy. In he same way we canno compare he impac curren economic aciviy on inflaion from he Phillips curve alone. We need o include some sor of deerminaion process for he oupu gap over ime. We can, for example, describe he ime series properies of he oupu gap by a simple auoregressive process: y y 1. (11) By successive forward subsiuion in (7) we obain j 0 j E y. (12) j Then using he auoregressive process in (11) for he expeced values of he oupu gap, under he assumpion ha is a shock whose expeced value is zero, (12) becomes simply 6 In boh specificaions he lagged US inflaion rae was used in addiion o he insrumens used in he previous es. 14

15 1 y * Defining / 1 /. (13) we can now compare * across each of he esimaed New Keynesian Phillips curves as a measure of he impac of a change in he oupu gap on inflaion, as we have now allowed for differen persisence in he oupu gap. In he same way we can compare he * impac in he expecaions-augmened specificaion (3'),, wih in order o compare he impac in he wo differen specificaions. Figures 4 and 5 make hese comparisons for he GDP and privae consumpion deflaors derived in Tables 1 and 2. The values of are shown in he Tables below he respecive figures. There are main messages we can derive from his comparison. The firs is ha he broad ranking of counries under he wo models is similar. There are some noable differences for Ireland, Porugal, Finland and Ausria bu he larger counries, represening some 80% of he GDP of he euro area remain in he same order and relaive posiion. The second obvious resul is ha he persisence in he oupu gap no merely varies subsanially across counries bu ha i increases he spread of he New Keynesian curves, paricularly for he more exreme cases. This increase in spread is o be expeced unless were uniy. As a generaliy is a decreasing funcion of he slope of Phillips curve so he changes are less han proporionae. The hird noiceable feaure is ha he resuls under he privae consumpion deflaor vary less han under he GDP deflaor. The exen is imporan when i comes o comparing policy based on he EU11 aggregae relaionship compared o he posiion for he individual saes. Wih he privae consumpion deflaor case he euro area resuls are similar under he wo models. While of he major counries Spain and o a lesser exen France lie below he euro area slope and Ialy clearly lies above i, he discrepancies are no so sriking for he privae consumpion deflaor. Wih he GDP deflaor he Phillips curve is virually horizonal in he New Keynesian case. This would again be a good represenaion of he counry resul for France and Germany and o a lesser exen Spain. Ialy remains an oulier. The expecaions augmened curve, however, has far more in common wih he consumpion deflaor values and is very differen from he esimae one would ge by weighing he individual counry esimaes by heir economic size. Differences among he member saes of clear policy imporance here and he applicaion of an aggregae model would have very differen implicaions from a disaggregae one. Even if we ignore he more exreme resuls and concenrae on he consumpion deflaor as his is he arge variable for moneary policy i sill remains he case ha over his sample he range of counry experience is sill over a facor from 2 o 3. This order of difference is very similar o hose repored in he survey of sudies of he euro area in Mayes and Virén (2000). The final feaure of inflaion dynamics worh considering from his simple analysis is he degree of persisence. As expeced, i is clear from Figure 6, ha wih he excepion of Belgium and Luxembourg, here is very considerable persisence over he daa period, wih inflaion persisence following a similar paern o oupu gap persisence. 15

16 Figure 4. Slopes of he Phillips curves when using he GDP deflaor as he measure of inflaion 3.0 The slopes of he expecaions-augmened Phillips curve and he New Keynesian Phillips curve wih he correcion facor 3.0 The slopes of he expecaions-augmened Phillips curve and he New Keynesian Phillips curve wihou he correcion facor EU11 NK EU11 EA EA NK EU11 NK EU11 EA EA NK France Finland Germany Ireland Porugal Spain Belgium Neherlands Ausria Luxembourg Ialy France Finland Germany Ireland Porugal Spain Belgium Neherlands Ausria Luxembourg Ialy 3.0 The slopes of he New Keynesian Phillips curve wih and wihou he correcion facor EU11 NK wih EU11 NK -0.5 France Finland Germany Ireland Porugal Spain Belgium Neherlands Ausria Luxembourg Ialy The slopes of he Phillips curves when using he GDP deflaor as he measure of inflaion NK NK EA wih corr.facor wihou corr.facor Auocorrelaion Correcion facor EU France Finland Germany Ireland Porugal Spain Belgium Neherlands Ausria Luxembourg Ialy

17 Figure 5. Slopes of he Phillips curves when using he privae consumion deflaor as he measure of inflaion 3.0 The slopes of he expecaions-augmened Phillips curve and he New Keynesian Philips curve wih he correcion facor 3.0 The slopes of he expecaions-augmened Phillips curve and he New Keynesian Philips curve wihou he correcion facor EU11 NK EU11 EA EA NK EU11 NK EU11 EA EA NK Ireland Spain Finland France Germany Luxembourg Ausria Belgium Neherlands Porugal Ialy -0.5 Ireland Spain Finland France Germany Luxembourg Ausria Belgium Neherlands Porugal Ialy 3.0 The slopes of he New Keynesian Phillips curve wih and wihou he correcion facor EU11 NK wih EU11 NK -0.5 Ireland Spain Finland France Germany Luxembourg Ausria Belgium Neherlands Porugal Ialy The slopes of he Phillips curves when using he privae consumpion deflaor as he measure of inflaion NK NK EA wih corr.facor wihou corr.facor Auocorrelaion Correcion facor EU Ireland Spain Finland France Germany Luxembourg Ausria Belgium Neherlands Porugal Ialy

18 EU 11 EU 11 The persisence (1 s-order auocorrelaion) of he oupu gap and he inflaion rae when using he GDP deflaor as he measure of inflaion Luxembourg Belgium Ireland Oupu gap Ialy Ausria Neherlands Germany Porugal France The inflaion rae / gdp-defl. Finland The persisence (1s-order auocorrelaion) of he oupu gap and he inflaion rae when using he privae consumpion deflaor as he measure of inflaion Luxembourg Belgium Ireland Oupu gap Ialy Ausria Neherlands Germany Porugal France The inflaion rae / pcp Finland Spain Spain Oupu gap The inflaion rae The inflaion rae gdp-defl. pcp EU Luxembourg Belgium Ireland Ialy Ausria Neherlands Germany Porugal France Finland Spain Figure 6 Persisence in Inflaion and he Oupu Gap in he euro area

19 4 Conclusions In recen years here has been an increasing ineres in he issue of inflaion dynamics because of he puzzling price developmens and he new euro area. Also he developmen in he underlying heoreical frameworks has renewed he sudy of he radeoff beween inflaion and oupu, which is cenral in moneary policy analysis. In empirical sudies of inflaion dynamics only limied suppor has been found for he oupu gap based New Keynesian Phillips curve under raional expecaions. In his paper inflaion dynamics in he euro area and eleven EMU economies have been sudied by comparing he empirical performance of he wo alernaive oupu gap based Phillips relaions: he expecaions-augmened Phillips curve and he New Keynesian Phillips curve. In applied work he Phillips relaion and he hypohesis of expecaions formaion have been ypically esed joinly. By conras, in his sudy he raional expecaions hypohesis has no been mainained, as direc measures of marke expecaions have been used. In he expecaions-augmened specificaion he OECD inflaion forecas for he curren year has been used o measure inflaion expecaions. Correspondingly, he OECD inflaion forecas for he nex year has measured inflaion expecaions in he New Keynesian specificaion. In addiion o he single equaion esimaion wih alernaive inflaion measures, he empirical superioriy of he alernaive Phillips relaions has been compared using encompassing and non-nesed ess. Comparison of he impac of he oupu gap on inflaion across models and counries requires aking accoun of he deerminaion process of he oupu gap, as i is no predeermined in he New Keynesian model. The empirical evidence suggess ha he oupu gap based expecaions-augmened Phillips curve provides a reasonable approximaion o he inflaion dynamics in he euro area, when expecaions i.e. he previously expeced curren inflaion is measured direcly using he OECD inflaion forecass. Also, he oupu gap based New Keynesian Phillips curve wih he currenly expeced fuure inflaion racks inflaion variaion plausibly, when he OECD forecass are used. A conclusive comparison beween he alernaive models canno be made on he basis of he single equaion esimaion resuls, bu he empirical es resuls show evidence more in favour of he new Keynesian specificaion boh for he euro area and for mos EMU counries. Resuls are qualiaively fairly similar regardless of he inflaion measure, he GDP deflaor or he privae consumpion deflaor. However, here appears o be subsanial heerogeneiy in inflaion processes across he euro area counries, wih he slope of he Phillips curve varying by a facor of wo o hree across he member saes, afer omiing he ouliers. A policy based on a common relaion would herefore have srikingly differen oucomes across he area. If addiionally here was subsanial asymmery in he shocks (size of he oupu gap) across he area, he appropriaeness of a single policy based on aggregae esimaes could amplify he discrepancies even furher if he paricular disribuion of shocks happened o be correlaed wih he disribuion of slopes. Mayes and Virén (2000, 2002) show ha hese discrepancies are raher greaer if he Phillips 'curve' is acually allowed o be more like a curve and no he linear relaionship ha is esimaed in he wo cases here. They also use OECD forecass as he measure of expeced inflaion in he more recen of heir aricles. I is, however, imporan o noe ha he sample of daa used here covers he enire period from 1980 o As Figure 1 showed here has been a subsanial change in he behaviour of inflaion since he beginning of he 1990s. Furhermore, wih he inroducion of he euro area and he 19

20 Sabiliy and Growh Pac one migh also expec some changes in he relaionships, quie possibly in he direcion of convergence. Conrary o many oher sudies under raional expecaions, inflaion dynamics is adequaely capured in his sudy wihou need o use lagged inflaion as an explanaory variable. The resuls also provide evidence supporing he use of he oupu gap as a measure of cyclical variaion in he Phillips relaion. The plausible empirical fi of he models may be associaed wih he assumpion of expecaions formaion. Boh Phillips curves seem o capure price changes, as inflaion expecaions may adjus slowly and agens need no necessarily be raional. The presen resuls are in line wih he sudies by Robers (1997, 1998), which show evidence on he New Keynesian Phillips relaion using survey based expecaions from he US. Using he OECD forecass as direc measures of inflaion expecaions may herefore be worh sudying furher. The role of expecaions in inflaion dynamics is crucial when designing moneary policy. Under he expecaions-augmened Phillips curve, moneary policy change does no change inflaion direcly, even if economic agens modify heir expecaions as a reacion o a policy change. The only effec comes hrough he excess demand channel. Insead, if inflaion dynamics follows he New Keynesian Phillips curve, a ransiion o a new policy regime affecs inflaion immediaely, as curren inflaion is dependen on currenly expeced fuure inflaion. The beer he policy change is undersood by he privae secor, he more curren inflaion is affeced according o he New Keynesian Phillips relaion hrough he change in expecaions. The debae over he appropriae moneary policy becomes more complex as i is no jus he seing of policy ha maers in some mechanical way bu is impac on expecaions as well ha maers. There may herefore be some benefi o undersanding inflaion dynamics from easing combinaion of he Phillips relaion wih raional expecaions as he mainained hypohesis. Insead, he use of direc measures of inflaion expecaions may improve our undersanding of inflaion processes. By using direc esimaes of inflaion expecaions, in his paper we have been able o sugges wo main resuls for he euro area. Firs of all in he simple specificaions used he New Keynesian Phillips curve fis he euro area daa slighly beer han he expecaions-augmened Phillips curve. Secondly, whaever he specificaion here has been very considerable heerogeneiy in he inflaion process across he member saes of he euro area in he las 20 years. The role of expecaions in inflaion dynamics is imporan in conducing moneary policy and sill open o debae. Research on expecaions formaion would be an imporan complemen o he presen resuls. Given he exen of he asymmeries in behaviour i appears ha sudying he process disaggregaed a leas o he member sae level is necessary in he design of he single moneary policy. 20