A New Test for the Success of Inflation Targeting

Transcription

1 ESCUELA DE NEGOCIOS Univeridad Torcuato Di Tella CIF Centro de Invetigación en Finanza Documento de Trabajo 03/2004 A New Tet for the Succe of Inflation Targeting Verónica Cohen Sabbán Banco Central de la República Argentina Martín Gonzalez Rozada Univeridad Torcuato Di Tella Andrew Powell Univeridad Torcuato Di Tella Miñone 2177, C1428ATG Bueno Aire Tel: interno 181 y Web ite:

2 A New Tet for the Succe of Inflation Targeting Verónica Cohen Sabbán Banco Central de la República Argentina Martín Gonzalez Rozada Univeridad Torcuato Di Tella Andrew Powell 1 Univeridad Torcuato Di Tella Revied April, 2003 Abtract We propoe a new tet, derived from a et of variance decompoition of a tructural VAR, for the ucce of inflation targeting. In contrat to tandard acrifice ratio thi tet conider change in the tructure of real and nominal hock; econd moment effect. We find trong upport for IT with 7 of the 9 countrie in our ample having negative acrifice and many countrie with benefit. However, we alo find very different performance acro IT countrie. We find that IT ucce depend on the ize of the real hock uffered but controlling for thi there are difference in country performance. JEL Categorie: E42, E52 Keyword: Inflation targeting, Structural Vector Auto-Regreion (VAR) 1 We are grateful to Ricardo Haumann, Leonardo Leiderman, Klau Schmidt Hebbel and Ted Truman for comment. Naturally all error remain are own. Pleae addre comment to Andrew Powell at apowell@utdt.edu

3 1. Introduction Inflation targeting ha become popular in both academic and policy-making circle. In thi paper we conider 9 countrie that have adopted IT regime 2, ranging from pioneering New Zealand to other indutrialied countrie, including Autralia, Canada, Sweden and the UK, to everal more recent emerging country convert including Mexico and Brazil, the pioneer in Latin America, Chile, and finally Irael 3. There i alo a growing academic literature on IT 4. Chapter 10 of Bernanke et al (1999) i devoted to a comparative analyi, conidering largely a ample of indutrialied countrie. Three quetion are poed; (1) doe IT make diinflation le cotly? (2) doe IT reduce inflation expectation?, and (3) doe IT change the behaviour of inflation? The preliminary concluion i that, countrie that have adopted IT have een inflation level and inflation expectation fall below what would be expected given extrapolation of pat behaviour. However, thi analyi find no upport for the view that diinflation i le cotly under IT 5. Still, the author hy away from concluding whether thi implie that IT fail to "create credibility" or whether 2 We alo comment in the text on the cae of Spain that adopted an IT regime before joining EMU. We do not include Spain in the empirical ection due to the very hort period of IT. 3 It i alo intereting that the two mot larget currency area by GDP, Euroland and the US, have not adopted IT regime - ee Bernanke et al 1999 for a dicuion. 4 Notable empirical contribution uing cro-country ample include Bernanke et al (1999), Schmidt Hebbel and Werner (2002), Corbo, Landerretche and Schmidt Hebbel (2001) and Debelle (1997). For a more theoretical review ee, for example, Svenon (1998). See Haldane (1999) for an attempt to draw out general leon from the UK experience and Mihkin (2000) for a review of the iue from the tandpoint of emerging countrie. 5 Thi reult tem from comparing acrifice ratio (normally defined a the output lo for each percentage point reduction in inflation) between IT and non-it countrie.

4 thi reflect a confirmation of previou reult that the tructure of monetary policy ha little impact on the output-inflation trade-off in the hort run. Schmidt Hebbel and Werner (2002) focu on the performance of emerging country inflation targeter in Latin America: namely Brazil, Chile and Mexico but alo preent comparative tatitic againt a wider control group of IT countrie. While the pot IT period ha been aociated with lower inflation, the acrifice ratio preented by thee author how a light deterioration for Brazil and Mexico and an improvement for Chile 6. They alo preent a variety of empirical exercie to attempt to undertand the effect of IT in thee three countrie. One analyi, that they claim i tandard, i to run a Vector Auto Regreion including a endogenou variable the real interet policy rate, the inflation target, core inflation, output growth, money growth and the real exchange rate. There are at leat two iue worth dicuing with thi methodology. One i that their ample period cover pre and pot-it period and hence they ue an expected inflation proxy for the inflation target in the pre IT period 7. Second, the inflation target i a tep variable and given that it i a target et by the Central Bank, it i not clear how to interpret the error term in the equation for 6 Sacrifice ratio are again the output lo (GDP and indutrial output) for each percentage point reduction in inflation. 7 The analyi i conducted recurively and the author dicu the difference in the impule repone in the pre and pot inflation-targeting period. While the reult are uggetive, if the IT target i a better predictor of expected inflation than the proxy employed in the pre IT period, then that could explain why the effect of target change become more ignificant after IT without that target having any real influence on actual inflation. It i alo not clear what the policy interet rate variable correpond to in the pre IT period.

5 that variable (treated a endogenou) nor what it tatitical propertie are. It i difficult to etimate how thee iue affect the reult preented 8. IT regime have in general developed a a way to move away from other nominal anchor and in particular from explicit or implicit exchange rate target or band toward greater exchange rate flexibility. One of the potential benefit of floating exchange rate i that thee rate can then repond to real hock and act a a type of hock aborber for the ret of the economy. At the ame time, floating exchange rate may imply a acrifice in that real exchange rate may become more volatile and PPP deviation more evere 9. On the other hand, if IT regime develop credibility and are ucceful then perhap we would expect that floating with IT would reult in only mall and nonperitent nominal hock on the real exchange rate and hence for real rate not to be unduly affected. The current 'tet' regarding the 'ucce' of IT regime a mentioned above tre the effect of IT on inflation expectation and outcome inflation level. Thee might be thought of a firt moment effect. However, a different way to conider the quetion of whether IT change the behaviour of inflation and other variable i to conider how the adoption of IT affect the tructure and the impact of nominal and real hock econd moment effect. A dicued we might, on the one hand, expect that the nominal exchange rate would become more volatile and be driven more by real hock a the exchange rate 8 The reult claim a trong and ignificant effect of IT target change on inflation in Chile and ignificant but quantitatively mall effect of IT target change in Mexico and Brazil. 9 A previou literature focued on the exce volatility of nominal rate with rational model of over-hooting and bubble a well a irrational panic and herd behavior potentially driving real rate away from equilibrium or fundamental value.

6 act a a hock aborber. Thi may be thought of a the benefit of floating with IT. However, on the other hand if the real exchange rate become more volatile and driven more by nominal hock, then thi would contitute a acrifice. Succeful IT regime hould then have only a mall (or no) acrifice and ignificant benefit. Thee argument ugget a different 'tet' regarding the 'ucce' of IT regime. Namely an analyi of the tructure of hock between the real and the nominal exchange rate and thi i indeed the approach taken in thi paper. We are helped in thi regard by Ender and Lee (1997) who conider the tructure of nominal and real exchange rate hock between G4 countrie in the pot Bretton Wood era. We follow their methodology here, which in turn employ the etimation of a tructural VAR a uggeted by Blanchard and Quah (1999). However our purpoe i different. We etimate tructural VARS for 9 countrie plitting the ample between 'prior IT' and 'pot IT' window 10. In ome cae, where data permit, and notably the UK and Chile, we alo plit the 'pot IT' ample into two to tet whether there i evidence of change a, perhap, the IT regime ha gained credibility over time. Our reult ugget that IT ha indeed allowed the nominal exchange rate to react more to real hock a nominal exchange rate regime have become cloer to pure float. In other word the benefit have been poitive. However, the reult on acrifice are urpriing to ay the leat. We find that for a et of IT regime, mainly in indutrialied countrie, there are no acrifice. 10 Note that thi methodology doe not employ the inflation target directly within the VAR. It i then a more indirect tet regarding change in the relation between nominal and real hock in the economy.

7 However, we alo find coniderable variation acro countrie a to how IT i performing. The paper i organied a follow. In ection 2, we briefly dicu the adoption of IT in the ample countrie. Section 3 develop the empirical methodology while ection 4 i devoted to a dicuion of the data and the reult. Section 5 conclude. 2. The Adoption of Inflation Targeting in a Sample of Indutrialied and Developing Countrie. In thi paper we conduct tet on 9 countrie that have adopted IT regime. To give ome background, in thi ection we briefly contrat the development in thee economie a they adopted IT. While each country ha it own particular context and monetary hitory, we attempt to delineate ome common theme acro the countrie in thi highly ummaried account. We alo dicu briefly the difference in the way in which IT ha been implemented acro the countrie in our ample 11. Adopting IT In Table 1 we detail the countrie in the ample and the date at which IT wa adopted 12,13. A can be een from the Table the firt country to adopt IT wa 11 Thi ection draw on Bernanke et al (1999), Schmidt Hebbel and Werrner (2000) and information from the webite of the relevant Central Bank. 12 Thee date are largely peaking the date when the relevant Central Bank claim to have adopted IT. 13 It i ometime debatable exactly when IT wa adopted a Central Bank do not alway give pecific date and other appear to 'jump the gun' in claiming IT adoption. See for

8 New Zealand in 1990 although other countrie followed rapidly. The latet IT country in our ample i Brazil and we ue 1999 a the date for it adoption. Countrie have adopted IT in quite different macroeconomic etting. Here we identify three main circumtance. Firt, everal countrie in our ample adopted IT after a period of poor economic performance, weak economic credibility and a deire to eek greater tranparency in monetary operation and a belief that with greater tranparency might come greater credibility and (hence) improved economic performance. Countrie fitting thi et of tylied fact include New Zealand, Canada, Autralia and arguably Mexico. Second, countrie have adopted IT a a way to obtain greater flexibility moving away from an exchange rate band after a ignificant deflation frequently obtained uing the exchange rate band a nominal anchor 14. Countrie in thi group include Irael, Chile and Spain 15. The third group are countrie that uffered an exchange rate crii a an exchange rate band wa forcibly abandoned and IT wa then adopted in it aftermath, in part, a a way to attempt to regain credibility. Countrie in thi category include Brazil, Sweden and the UK 16. example Benanke et al (1999) for a dicuion of the Autralian cae. The IT adoption date for Chile i in particular debatable with the Central Bank typically claiming 1991, but alo note Schmidt Hebbel and Werner (2002) uggeting that only, "In the Bank upgraded it monetary framework to full inflation targeting" pp36. Thi later date coincided with the dipoal of the exchange rate band and development of the Central Bank' modeling and forecating activitie. 14 Thi i not to ay that there were not alo ignificant deflation in other countrie uch a New Zealand but the countrie included here ued an exchange rate anchor during the deflation and the deflation wa immediately prior to the adoption of IT. 15 Spain adopted IT alo within the context of explicitly eeking EMU memberhip and within the context of following the rule of ERM II - including making the Central Bank independent. Spain ubequently joined EMU. We do not conider Spain in the empirical analyi due to the hort period in the pot IT regime prior to EMU entry. 16 Mexico alo uffered an exchange rate crii prior to adopting IT but the crii wa in early 1994 and the adoption of IT wa much later and hence we include Mexico in the firt group.

9 Table 1: Adopting IT Adoption of IT Before IT After IT Frequency New Zealand quarterly Canada monthly Chile monthly Irael monthly Autralia quarterly United Kingdom monthly Sweden monthly Brazil monthly Mexico monthly Prior to adopting IT, the countrie in our ample had a variety of monetary regime. Thee ranged from a non-tranparent (or "jut do it") tyle of monetary policy that characteried the prior IT regime of Autralia, Canada and New Zealand, and arguably Mexico, to the exchange rate targeting regime of Spain, Sweden and the UK to the more explicit exchange rate band of Brazil, Chile, Irael 17. Each of the countrie in the ample had ome meaure of nominal exchange rate flexibility in the prior IT regime. Perhap the hardet 'fix' wa the exchange rate band of the Brazilian real plan but even here the nominal exchange rate varied within band, the band crawled and the band wa alo adjuted a number of time. Characteritic of IT regime There i coniderable debate in the IT literature a to the appropriate way to conduct IT 18. Here we focu on ix particular characteritic to illutrate 17 Chile maintained the exchange rate band after adopting IT. However, Schmidt Hebbel and Werner (2002) claim that the inflation target alway had priority and the exchange rate band wa adjuted everal time. 18 Svennon (1988) ugget that IT, a practiced in New Zealand, Canada, the UK, Sweden and Autralia can be thought of a, (1) an explicit quantitative inflation target, (2) an

10 ome of the imilaritie and difference in IT regime acro the world. The characteritic are (i) if pecific IT legilation ha been enacted, (ii) if Central Bank are independent, (iii) how inflation i defined, (iv) the role of the exchange rate and whether an explicit and tranparent monetary condition index i ued (v) the type of target adopted and (vi) the degree of flexibility that exit around that target 19. Mot indutrialied countrie have enacted pecific legilation regarding IT - the exception being Canada. However, emerging countrie have tended to lag behind thi trend. Alo, mot IT countrie have given formal independence (objective or goal) to their Central Bank but Brazil and Mexico to date have not. There i coniderable debate in the IT literature a to what meaure of inflation hould be employed. Mot countrie ue headline CPI inflation however New Zealand and Autralia ue an underlying inflation index that take out certain volatile component. Inflation targeting appear to have become particularly popular in mall open economie a evidenced by our 9-country ample. The role of the exchange rate i then a crucial variable in all of thee IT regime. Indeed, ome countrie have defined an explicit Monetary Condition Index (MCI) or imilar - mot tranparently in Canada, New Zealand and Irael - for the operating procedure can be decribed a inflation forecat targeting. and (3) a high degree of tranparency and accountability. 19 Bernanke et al (1999) ugget ten operational and communication iue are paramount (which we have ummarized in the 6 lited above); (1) which meaure of inflation hould be ued?, (2) what numerical value hould the target have? (3) a price level or an inflation target? (4) what horizon? (5) a point or a range? (6) what information hould be ued in policy making? (7) when hould deviation from the target be allowed? (8) when i the bet time to implement IT? (9) what hould be communicated and in what forum? (10) to what degree hould central bank be held accountable?

11 purpoe of hort term monetary control. In the cae of Canada and New Zealand, for example, a change in the MCI i defined a the weighted um of change in 90 day commercial paper interet rate and the trade weighted exchange rate (with weight 3 to 1 for Canada and 2: 1 for New Zealand). In the cae of New Zealand a 2% rie in the exchange rate i then etimated to have the ame effect a a 1% point rie in interet rate. Thi ratio indicate particularly the importance of the exchange rate in the cae of NZ. Since July 1997, the reerve bank ha publihed a forecat of the MCI to indicate the path of monetary policy for the following three year if condition remain unchanged. Haldane (1999) contrat the automatic ue of an MCI index to govern monetary policy repone to exchange rate change veru what he refer to a a pot the hock approach where the monetary policy repone i conditioned on a view a to what caued the exchange rate to move. Schmidt- Hebbel and Werner (2002) alo devote a ection of their paper to the role of the exchange rate within the IT framework of Brazil, Chile and Mexico. Their focu i on the related iue of whether there i evidence of high pathrough from the exchange rate to interet rate behaviour and whether there ha been a 'fear of floating' in thee countrie On the latter they conclude that conidering the volatility of exchange rate and reerve, "fear of floating ha declined ubtantially with the adoption of the floating regime, and it appear to be levelling off toward level oberved in the more mature floater" (namely Autralia, Canada and New Zealand). Regarding "pa through", they ugget that, controlling for inflation, exchange rate movement do not lead to ignificant change in interet rate in the three countrie analyed.

12 There are alo difference a to how the target hould be pecified. The UK, Brazil, Chile Sweden and Autralia have tended to adopt point target wherea Canada, New Zealand and Irael have inflation range 21. Finally regarding flexibility, New Zealand alo ha the mot explicit rule-baed ytem. The UK ha a formal procedure if inflation i outide of the range involving a public letter from the Central Bank Governor to the Miniter explaining the ituation. Other countrie have fewer rule and hence, arguably, greater flexibility. Flexibility i often related to whether the Central Bank only ha the inflation target or other objective uch a output tabilization a well. Mot Central Bank in our ample have other goal a well a imply the inflation target although with arguably with varying weight 22. Thi brief decription of ome of the alient characteritic erve to highlight that while a variety of countrie claim to have adopted Inflation Targeting there i till coniderable debate a to what IT actually mean and difference a to how it i implemented acro the globe. We come back to thi a we dicu the reult. 21 See the dicuion in Haldane (1999) regarding the alleged benefit of a pecific target rather than a range. 22 See Svenon (1997) for a dicuion on trict veru flexible inflation targeting.

13 3. Econometric Methodology In thi ection, we decribe the methodology we ue to analye the performance of IT regime. Our approach i to conider how the adoption of IT by a country affect the behaviour of the nominal and real exchange rate. One poible alternative to tudy thi type of problem would be to conider a tructural model of exchange rate determination. However, the empirical evidence, a reviewed in Ender and Lee (1997) (EL hereafter), ugget that thee kind of model do not explain well real and nominal exchange rate movement. Intead of uing a tructural pecification, we follow EL and decompoe the exchange rate movement into change induced by real veru nominal factor 23. By tudying the tructure of real and nominal hock and their effect, before and after the adoption of IT, we can ae the performance of thi new monetary regime. The propoed decompoition i baed on the technique developed by Blanchard and Quah (1989) (BQ hereafter). The baic idea behind thi decompoition i the etimation of a reduced form vector autoregreive model (VAR) for both exchange rate, real and nominal, and then to decompoe the exchange rate erie into movement caued by real hock and thoe caued by nominal hock. 23 Moreover EL how that the decompoition i conitent with a number of theorie regarding exchange rate determination.

14 We aume that the real hock affect both real and nominal rate imilarly while the nominal hock affect both rate differently. In the latter cae, conitent with the idea of long-run money-neutrality, the nominal hock are aumed to have only a temporary, rather than permanent, effect on the real exchange rate. Thi i the key retriction that allow BQ to identify both type of hock. Before we can formulate the model, it i neceary to analye the tochatic propertie of both exchange rate erie. If both erie are tationary, an unretricted table VAR model in level can be pecified; if both erie are characteried by unit root procee and they are co-integrated, then the correct model to be pecified will be a retricted VAR for the variable in firt-difference incorporating the co-integrating vector to repreent the long run; finally, if both erie are non-tationary but they are not co-integrated, then the correponding model would be an unretricted table VAR for the variable in firt-difference 24. For eae of expoition, we conider the BQ decompoition technique for an unretricted VAR model for both rate expreed in firt-difference, 24 In a et of preliminary reult we find that in general the relevant erie are I(1) but are not co-integrated. The exception are Mexico and Irael in the pot-it period where we find the erie are I(1) and cointegrated. In thee cae we incorporate the cointegrating vector into the relevant empirical model.

15 ( ) ( ) ( ) ( ) t t S t S t t t t t S t S t t t r e r r e e ε γ β ε γ β + + = + + = = = = = (1) where e t and r t are the nominal and real exchange rate, repectively; the β, and γ are the parameter of the model; ε = [ε 1, ε 2 ] = [ e t, r t ] E([ e t, r t ] [ e t-, r t- ], 1), with var(ε) = Σ; and = 1-L, with L being the lag operator. In the empirical analyi below we ue the nominal dollar exchange rate for almot all countrie except for Sweden and United Kingdom for which we employ the German DM (and ubequently the Euro). We ue the WPI to calculate the relevant real exchange rate. Since the VAR model i table it ha a moving average repreentation given by, nt rt t nt rt t L u b L u b r L u b L u b e = = = = + = + = ) ( ) ( ) ( ) ( (2) where, u rt and u nt are zero-mean mutually uncorrelated real and nominal hock, repectively. BQ how that one can impoe retriction on any of the two model in order to identify real and nominal hock in the ytem. In particular, the retriction that the nominal hock have no long-run effect on the real exchange rate i repreented by the retriction that in the moving average repreentation of the ytem.

16 = 0 b22( ) = 0 That i the cumulative effect of u nt on r over time i zero. Similarly, becaue the um of b 12 from zero to infinity i the ame, a the effect of u n on r after an infinite number of period, the nominal hock ha only hort run effect on the real exchange rate 25. Thi retriction on the moving average repreentation can be equivalently expreed in term of the VAR model by the following expreion, β β 1 (0) + (0) = 0 0 1( ) 22 = t b = 0 2( t ) b 12 (4) The coefficient of the tructural model can then be repreented a follow: b b ( L) b12( L) ( ) ( ) L b22 L where = 0 () 0 b22 = (5) Thi mean that, a uggeted by the theory, in the long run real hock can caue permanent change in the real exchange rate, but nominal hock can caue only temporary movement in the real rate. In our applied work we would like to know the repone of one of the exchange rate to an exogenou impule in another variable. That i, we would like to trace out what happen to the etimated ytem during ubequent period after an exogenou hock ha occurred. Specifically, we would like to eparate the variation in an endogenou variable into the component hock to the VAR. Therefore we employ variance decompoition. The variance decompoition provide information about the relative importance of each random innovation in affecting the firt difference in the real and nominal 25 See Blanchard and Quah (1989) or Ender and Lee (1997).

17 exchange rate. Uing the variance decompoition we can then ee whether IT ha had any ytematic effect on the relative importance of the component hock driving the real and nominal exchange rate. To ae the tatitical ignificance of any change in the variance decompoition we compute 90% boottrapping confidence interval 26. Thee interval are computed in the following way. Firt, we etimate the correponding VAR ytem and obtain the etimated coefficient and reidual from both equation. Second, from the vector reidual we elect randomly a new vector of reidual that we call the boottrapping reidual vector. With thi new reidual vector and the etimated coefficient of tep one, we reproduce the value of both dependent variable. Third, once we have the boottrapping value of the dependent variable, we repeat the VAR etimation and produce the tructural variance decompoition. Next, we go to the econd tep and repeat thi procedure. We perform thi imulation 1000 time o that after the imulation proce i complete we have for each forecating period, 1000 replication of the variance decompoition. Then, for each forecating period we contruct a 90% empirical confidence interval for the variance decompoition.

18 To analye the ucce of IT regime, we propoe a new acrifice and a benefit ratio a a function of the variance decompoition. On the one hand, we expect the real exchange rate to be driven more by nominal hock. Hence we propoe a acrifice ratio expreed a: S = % of r explained by NS after IT - % of r explained by NS before IT Where r i the real exchange rate and NS i a nominal hock.. A poitive number for S repreent a acrifice in the ene that the real exchange rate i now affected more by nominal hock after IT than before 27. On the other hand, we may alo expect the nominal exchange rate to become more volatile and be driven more by the real hock. Thi in our view contitute the 'benefit' - the fact that nominal exchange rate can act more a a hock aborber for real hock. We then define a benefit ratio a: B = % of e rate explained by RS after IT - % of e explained by RS before IT Where e i the nominal exchange rate and RS i a real hock. 26 Naturally, thee confidence level alo erve to analyze whether the variance decompoition are tatitically ignificant in them-elve and not jut due to ay ampling error. 27 We note that the Blanchard and Quah (1989) identification procedure in the tructural VAR impoe the contraint that in the long run the nominal hock doe not affect the real exchange rate however thi doe not imply a retriction on the variance decompoition and doe not therefore invalidate the above ratio.

19 In thi cae, a poitive number for B repreent a benefit. Thi i the cae when an IT regime allow the nominal exchange rate to be affected more by the real hock and hence act more a a type of 'hock aborber'. In the cae of the UK and Chile we alo plit the pot-it ample to analye how the thee acrifice and benefit ratio change comparing the two pot-it period againt the period pre-it. Our idea i that if we find that the benefit ratio become more poitive and the acrifice ratio more negative, then there i evidence that the IT regime ha become more ucceful over time. The nominal exchange rate would then have become more affected by the real hock and the real exchange rate le affected by the nominal hock. Finally, we analye the reult acro countrie in the pot IT regime to conider how different IT regime in different countrie are functioning. In particular we eek to acertain whether there are ytemic difference in the tructure of the hock. We find trong and intereting country difference. 4. Empirical Reult In thi ection, we preent the reult of the empirical analyi in 9 countrie that have adopted IT. In Table 2 below we firt preent the reult for the tructural model identified uing the Blanchard and Quah technique for the 9 countrie. The table for each country give the long run coefficient for the tructural model and tandard error. A can be een many coefficient are ignificant at any tandard ignificance level and they are all of the expected ign.

20 Table 2: Pre - IT Pot - IT Chile ( ) ( ) ( ) ( ) ( ) ( ) Mexico ( ) ( ) ( ) ( ) ( ) Brazil ( ) ( ) ( ) ( ) ( ) ( ) United Kingdom ( ) ( ) ( ) ( ) ( ) ( ) New Zealand ( ) ( ) ( ) ( ) ( ) ( ) Autralia ( ) ( ) ( ) ( ) ( ) ( ) Sweden ( ) ( ) ( ) ( ) ( ) ( ) Irael ( ) ( ) ( ) ( ) ( ) ( ) Canada ( ) ( ) ( ) ( ) ( ) ( )

21 Given the coefficient of the tructural model, we then etimate the variance decompoition and confidence interval a decribed in the previou ection. In Table 3 below we detail the reult for each country for the pre and pot IT regime with 90% confidence interval for lag of 1, 5 and 10 period. A can be een the confidence interval are relatively tight indicating that the variance decompoition are in general very well defined. More pecifically, in only very few cae do we find that the confidence interval overlap in other word we find trong evidence for change in the tructure of hock in the pre and pot IT regime.

22 Table 3: Variance Decompoition and 90% Confidence Interval Nominal Exchange Rate Real Exchange Rate Nominal Shock Real Shock Nominal Shock Real Shock Pre IT Pot IT Pre IT Pot IT Pre IT Pot IT Pre IT Pot IT Chile Lag * 89.90* ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Mexico Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Brazil Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) United Kingdom Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) New Zealand Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Autralia Lag * 9.16* 91.13* 90.84* ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag * 90.48* ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Sweden Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Irael Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Canada Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Lag ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) * indicate that the 90% two ided confidence interval overlap. In all other cae we then find trong evidence of change between pre and pot IT regime

23 We are now in a poition to conider the benefit and acrifice ratio a defined above. The following table ummarie the reult obtained for both ratio and uing a 1-period, 5-period and 10-period horizon in the variance decompoition 28 : Table 4: Sacrifice and Benefit Ratio Lag 1 Lag 5 Lag 10 Sacrifice Ratio Benefit Ratio Sacrifice Ratio Benefit Ratio Sacrifice Ratio Benefit Ratio Chile Mexico Brazil New Zealand* Autralia* Canada Sweden United Kingdom Irael The reult regarding the acrifice and benefit ratio are urpriing. In particular it i notable that in 6 of the 9 countrie there appear to be negative acrifice ratio at the firt lag and in 7 of the 9 countrie at the tenth lag. Thi implie that with the adoption of inflation targeting, the real exchange rate i driven more by real hock and le by nominal hock. The two remaining countrie where acrifice are poitive at a lag length of ten are Chile, where the acrifice i relatively mall (an additional 3% of real exchange rate movement are explained by nominal hock pot IT) and Brazil where the acrifice i higher (ome 13% at the tenth lag). 28 For all countrie a period refer to one month except Autralia and New Zealand where we work with quarterly information. For thee countrie the lag are 1-period, 2-period and 4-period.

24 Moreover, the benefit ratio appear to be in general large and poitive implying that the nominal rate reflect much more the real hock pot inflation targeting. Thi i epecially the cae for New Zealand, Chile, Brazil, Autralia and the UK. The benefit ratio for Mexico and Canada are mall - although the acrifice ratio are negative and Irael i clearly a pecial cae where the benefit ratio and the acrifice ratio are both negative. Thi implie that in Irael after IT, the nominal exchange rate reflect more the nominal hock (negative benefit) and the real exchange rate reflect more the real hock (negative acrifice). It i a if the nominal and the real economie have become more independent from each other. For two countrie namely the Chile and the UK where the data allow, we conider whether there ha been any learning or gain in credibility over the period of pot IT. In both cae we plit the ample in For the cae of the UK thi correpond to when the Bank of England wa granted greater independence and pecifically intrument independence and in the cae of Chile when, arguably, the Central Bank wa applying a more pure inflationtargeting regime. The reult are preented in Table 5. Table 5: Lag 1 Lag 5 Lag 10 Sacrifice Ratio Benefit Ratio Sacrifice Ratio Benefit Ratio Sacrifice Ratio Benefit Ratio Chile United Kingdom

25 The reult indicate that comparing the two pot-it period with the pre-it period, in the cae of Chile the benefit ratio rie but o too doe the acrifice ratio. The nominal exchange rate wa then allowed greater flexibility to reflect more the real hock but at the cot that the real exchange rate reflected more the nominal hock. In the cae of the UK the benefit ratio in the econd period i greater and the acrifice ratio i more negative. Hence, the real exchange rate reflected more the real hock and the nominal exchange rate alo reflected more the nominal hock. In the cae of the UK, a cae can be made then that over time there wa an overall improvement in the functioning of the IT regime wherea in the cae of Chile there wa a tradeoff more benefit but alo a lightly higher acrifice. Table 4 give quite trong upport for inflation targeting, a in general there appear to be a ignificant benefit but with little or negative acrifice. However, there may be quite different reaon why we are obtaining thee reult in different countrie related to how IT i working acro different countrie. To invetigate thi further, in Table 6 we preent a matrix of the decompoition of real and nominal exchange rate movement into the real and nominal hock in the pot IT period for all the 9 countrie.

26 Table 6: Shock Exchange Rate Pre - IT Pot - IT Nominal Real Nominal Real Chile Mexico Brazil New Zealand* Autralia* Canada Sweden United Kingdom Irael Nominal Real Nominal Real Nominal Real Nominal Real Nominal Real Nominal Real Nominal Real Nominal Real Nominal Real * For New Zealand and Autralia we how lag 4, becaue they have quarterly data. Thi Table i revealing in a number of way. Firt it i extraordinary to note that in the cae of the UK, 95% of real exchange rate movement reflect real hock and ome 92% of nominal exchange rate movement alo reflect real hock! A imilar but not quite o extreme pattern i alo found in the cae of Sweden (95% and 83% repectively) and Autralia (90% and 97% repectively).

27 Canada, New Zealand and Chile repreent more mixed cae where in the cae of Canada the real exchange rate i driven 93% by real hock but the nominal exchange rate only reflect the real hock to the tune of 57% and for New Zealand the real exchange rate reflect 74% real hock and the nominal exchange rate 79% real hock. In the cae of Chile the real exchange rate reflect the real hock a much a 87% and the nominal exchange rate reflect the real hock ome 60%. In the cae of Mexico and Brazil, the real and the nominal exchange rate are driven more by the nominal hock. We might conclude that thee are cae where the full potential benefit of a ucceful IT regime have yet to be reaped. Irael, a before, i a pecial cae where 97% of the real exchange rate i driven by the real hock and 97% of the nominal exchange rate i driven by the nominal hock! It i a if the nominal and the real economy are virtually independent. The above reult can alo be ummarized in the form of a graph namely an x-y plot, where each point repreent a country in the pot IT regime and where the axe repreent the percentage of the real exchange rate explained by real hock and the percentage of the nominal exchange rate explained by the reall hock.

28 The graph how the different country group a identified above. In the top right corner we have a a firt group the UK, Sweden and Autralia illutrating that both the nominal and real exchange rate are very largely explained by real hock in thee countrie. A mixed econd group then comprie of Canada, New Zealand and Chile where either the nominal hock explain a higher percentage of the real exchange rate (New Zealand) or the nominal exchange rate (Canada and Chile). We then have a third group of Mexico and Brazil where both the real and nominal exchange rate are driven much more by the nominal hock and finally the pecial cae of Irael where the real exchange rate i driven almot entirely by the real hock and the nominal exchange rate by the nominal hock. Graph 1: Comparative IT Performance % Real Exchange Rate Explained by Nominal Shock 100% 95% 90% 85% 80% 75% 70% 65% 60% 55% Irael Mexico Brazil Canada Chile Sw eden UK Autralia New Zealand 50% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% % Nominal Exchange Rate Explained by Real Shock

29 Broadly peaking there are two theorie that might explain thee reult acro countrie. The firt i imply that the firt group of countrie ha extremely high credibility in their monetary regime and hence the influence of nominal hock i limited, while the third group ha yet to gain fullcredibility. The econd group of countrie i in an intermediary poition. An alternative view however i that credibility i in fact in itelf related to the ize of the real hock in that if a country uffer very high real volatility it i then extremely difficult to gain full credibility. Rather than credibility being an independent aet to do ay with intitution, thi view ugget that it i itelf endogenou to the ize of the real hock and countrie, where the effect of the nominal hock i large will alo be thoe countrie that uffer large real hock. Graph 2: Size of Real Shock Veru How Much Real Shock Explain Real Exchange Rate 100% 95% Irael U.K. Sweden 90% Canada Autralia 85% Chile 80% New Zealand 75% 70% 65% Mexico 60% 55% Brazil 50% 0.0% 1.0% 2.0% 3.0% 4.0% 5.0% 6.0% Standard Deviation of Real Shock

30 Our econometric methodology, by dientangling the real from the nominal hock, allow u to invetigate preciely thi iue. In Graph 2, we plot the ize of the real hock (or in other word the tandard error of the equation for the real exchange rate) againt the percentage of the movement of the real exchange rate explained by the real hock (from the variance decompoition of the tructural VAR). The Graph i not concluive of any particular relationhip. However, for thoe that claim a negative relationhip, it i clear that controlling for the ize of the negative hock, Autralia and New Zealand perform well relative to Brazil and Mexico. Autralia in particular ha uffered from very large real hock during it IT regime and yet the real exchange rate ha been affected very little by other factor apart from thoe real hock. A econd way to conider the importance of the nominal hock on the real exchange rate i not through the econd moment (the variation of the real exchange rate i explained), but rather by the firt moment or in the language of the tructural VAR by the impule repone. In thi cae we want to know how important i the abolute repone and hence we conider the tandardized impule repone multiplied by the tandard error of the nominal hock.. Thi then give u a comparative meaure of the importance of the immediate effect of the nominal hock on the real exchange rate. 29 In Graph 3 we then plot the ize of the tandard deviation of the real hock (tandard error of the equation for the real exchange rate) againt the firt period abolute impule repone of the nominal hock on the real exchange rate. 29 We find that the effect of the nominal hock a would be expected given the contraint on the tructural Var die out reaonably quickly.

31 There i ome evidence here of a poitive relationhip although, conitent with Graph 2, it i clear that controlling for the ize of the real hock Autralia doe well compared to Brazil (i.e.: there i a lower effect of nominal hock on the real exchange rate) and the effect of the nominal hock on the real exchange rate i about the ame for New Zealand a Mexico even though New Zealand uffered much larger real hock. Conitent with our previou reult, the UK ha urpriingly high real hock but a lower effect of the nominal hock on the real exchange rate compared to Chile and Sweden. Graph 3: Effect of Nominal Shock v Size of Real Shock Brazil Autralia Mexico New Zealand 2.0 Canada Chile 1.0 Irael Sweden UK % 1.0% 2.0% 3.0% 4.0% 5.0% 6.0% Standard Deviation of Real Shock

32 There i then mixed evidence with repect to the argument that the ize of the real hock matter in term of how important nominal hock are in explaining real exchange ate movement but that even taking into account a potential relation there remain difference acro countrie. In other word, to explain our reult, admittedly baed on the handful of country obervation that are available to date, there alo appear to be omething ele preent. We poit that thi omething might well be related to intitutional factor, reputation or credibility. 5. Concluion In thi paper we have propoed a new tet for the ucce of inflation targeting baed on a et of variance decompoition temming from a tructural VAR etimation in countrie that have adopted inflation targeting. We have preented reult on a ample of 9 countrie - 5 indutrialied and 4 emerging. Inflation targeting (IT) ha frequently been adopted to replace ome other kind of nominal anchor, typically an implicit or explicit exchange rate target or band and to allow for greater nominal exchange rate flexibility uch that the nominal exchange rate can act a a type of hock aborber for the real economy. Thi ugget that it might be expected that under IT, nominal exchange rate might reflect more the real hock but that the real exchange rate might reflect more nominal hock that drive the nominal exchange rate away from ome fundamental equilibrium value. The former we ue a the

33 bai of our propoed benefit ratio and the econd the bai of our propoed acrifice ratio. Our reult give quite trong upport for inflation targeting. Indeed conidering the 10 period horizon, 7 of the 9 countrie diplay negative acrifice ratio. And of the two that have a poitive acrifice it i relatively mall in the cae of Chile (4%) but a little higher for Brazil (13%). On the other hand, 6 of the 9 countrie have a benefit ratio higher than 10%. However, we alo find trong difference acro countie in term of how IT i working. In the cae of the UK, Sweden and Autralia we find that real exchange rate and nominal exchange rate change are almot excluively explained by real hock. More mixed cae are New Zealand, Canada and Chile. Brazil and Mexico have yet to fully reap the potential benefit of IT according to thee meaure and Irael appear a a pecial cae where the nominal exchange rate i almot entirely driven the nominal hock and the real exchange rate by real hock. Invetigating thee reult more cloely, we find that there i mixed evidence of a relationhip between the ize of the real hock and the importance of the nominal hock in explaining real exchange rate movement. However, even taking uch a relationhip into account there remain cro-country difference indicating for example, that given the ize of the real hock Autralia and New Zealand have performed extremely well and Brazil and Mexico not o

34 well. We conclude then that apart from thi relationhip there appear to be omething ele at work that may be related more to intitutional factor, reputation and credibility. While IT i till young in thee countrie it i notable that the mot important intitutional difference of the IT regime of thee two countrie and the more ucceful inflation targeter appear to be the degree of autonomy of the repective Central Bank. A more data become available for more countrie, it will be intereting to compare the reult of thi type of analyi againt the characteritic of the IT regime a dicued in the firt ection of thi paper. New Zealand i recognied a having the mot rule baed IT ytem and thi ha clearly led to ignificant benefit (and negative acrifice) comparing the pre and pot IT regime and comparing performance acro countrie given the ize of the real hock uffered. However, at the ame time the mot ucceful IT regime according to our meaure have greater in-built flexibility (namely the UK, Sweden) and there i alo evidence that the UK ha improved performance over time perhap a a reult of learning, greater Bank of England autonomy or greater pure credibility 30. Over time with more data to work with it might be poible to invetigate whether certain IT regime characteritic are ecret for ucce or whether there really i no one ize fit all and IT regime characteritic hould be modelled to particular country ituation. 30 Chile ha alo been relatively ucceful according to our meaure and increaed the benefit over time but at the cot of a lightly higher acrifice.