DEPARTAMENTO DE ECONOMIA PUC-RIO. TEXTO PARA DISCUSSÃO N o. 423 THE PASS-THROUGH FROM DEPRECIATION TO INFLATION: A PANEL STUDY

Transcription

1 DEPRTMENTO DE ECONOMI PUCRIO TEXTO PR DISCUSSÃO N o. 43 THE PSSTHROUGH FROM DEPRECITION TO INFLTION: PNEL STUDY ILN GOLDFJN SERGIO R. C. WERLNG BRIL 000

2 The Passthrough from Deprecaton to Inflaton: Panel Study 1 By Ilan Goldfajn and Sergo R.C. Werlang Pontfca Unversdade Catolca EPGE Departamento de Economa Fundação Getúlo Vargas Marques de Sao Vcente, 5 Deputy Governor Ro de Janero, Brazl Banco Central do Brasl Phone: ( Brasla, Brazl Fax: ( emal: goldfajn@econ.pucro.br bstract The paper studes the relatonshp between exchange rate deprecatons and nflaton usng a sample of 71 countres n the perod The man determnants of the extent of nflatonary passthrough of the deprecatons (apprecatons are the cyclcal component of output, the extent of ntal overvaluaton of the real exchange rate (RER, the ntal rate of nflaton, and the degree of openness of the economy. The paper fnds that the passthrough coeffcents ncrease the larger s the horzon measured, wth ts peak at 1months. It also fnds that RER msalgnment s the most mportant determnant of nflaton for emergng markets whle the ntal nflaton s the most mportant varable for developed countres. Usng the estmated model, the paper predcts somewhat hgher nflaton than actually observed n several well known large deprecaton cases, even f one takes nto account exstng measures of exchange rate expectatons. Ths suggests that polcy makers should use cauton when usng past models to predct future nflaton n the aftermath of large deprecatons. Keywords: Passthrough, Real Exchange Rate, Devaluatons 1 The authors would lke to thank Gabrel Srour for excellent Research ssstance. lso, to Morah urora Meyskens for edtoral help.

3 Table of Contents I. Introducton...3 II. Theoretcal Motvaton and Methodology...5 III. The effect of deprecaton on nflaton...8 IV. The determnants of the passthrough coeffcent...15 V. Forecasts...1 VI. Usng expectatons...7 VII. Robustness Checks...30 VIII. Conclusons...35 XI. References...37 X. ppendx...38

4 I. Introducton Currency crses and the resultng large deprecatons often brng about a fear that an nflatondeprecaton spral has just started. Usually ths fear s based on a prevous hstory of hgh nflaton that the exchange rate, as a useful nomnal anchor, has helped change. Sometmes, ths fear of the return of nflaton s based on the sheer sze of the deprecaton of the exchange rate, so large than even f t doesn t lead to an nflatondeprecaton spral t often overshoots ts long run value. However, the resultng observed nflaton has been lower than expected ntally. Fgure 1 below shows how small the passthrough coeffcent the extent that the large deprecaton has become nflaton n several crses durng the 1990 s, wth the notable excepton of Mexco n Therefore, the conventonal wsdom s ether n front of a new stylzed fact or unfamlar wth an old regularty. The fgure also shows that the passthrough ncreases the longer s the horzon. Is ths also a new stylzed fact? Once the ntal few months of the crss are over and the nflaton behaves better than expected, polcy makers often face the need to forecast nflaton for monetary polcy purposes n a new floatng exchange regme. In ths scenaro, the dffculty arses because new data on the behavor of the economy s stll not avalable and longterm data reflects partly the behavor of the economy under an old regme. In short, the past gves lttle help n forecastng the future. In ths context, the average experence of other countres could be of help, notwthstandng the specfctes of the partcular country nvolved. There s a vast theoretcal lterature on the passthrough of exchange from deprecaton to nflaton (Dornbusch 1987, Feenstra et al. 1994, Fsher 1989, Goldberg et al. 1997, and Klen There s also some emprcal work on the passthrough for specfc countres, regons ( mtrano et al 1997 and others. Few papers have nvestgated the passthrough n a large sample of countres over tme (an excepton s Borenszten and De Gregoro, 1999, that look at currency crses. 3

5 Passthrough: Effect of devaluaton on nflaton (% 80% 70% Inflaton over Devaluaton % 60% 50% 40% 30% 0% 1 months 4 months 10% 0% Italy Sweeden Unted Kngdom Mexco South Korea Currency Crses Cases Malaysa Indonesa Thaland Brazl Korea wth only months Ths paper ams at provdng addtonal emprcal evdence on the relatonshp between nflaton and deprecaton n a broad sample of country's (71 n the last 0 years. The results provded from ths paper could be of help for academcs and polcy makers. The paper fnds that the passthrough coeffcent ncrease the larger s the horzon measured, wth a maxmum value at 1month horzon, and that ts most robust determnants are the RER overvaluaton and the ntal nflaton. The GDP devaton and openness varables are sgnfcant determnants but are more senstve to the horzon and sample chosen. The paper also concludes that usng the coeffcents obtaned durng normal tmes to predct nflaton after currency crss generates an upward bas,.e., the model systematcally overpredcts future nflaton. Ths occurs even when one controls for dfferent exchange rate expectatons, usng survey data. Ths suggests that econometrcans should use cauton when usng past models to predct future nflaton n the aftermath of large deprecatons. The paper s organzed as follows. The next secton presents the theoretcal secton that motvates the emprcal exercse and sets up the methodology. Secton III estmates the passthrough coeffcent under dfferent tme horzons and samples. Secton IV nvestgates the determnants of the passthrough coeffcent. Secton V analyzes the ft of the estmated model forecasts n a few known currency crss cases. Secton VI uses survey data to test the effect of 4

6 ncludng exchange rate expectatons on the passthrough results. Secton VII performs several robustness tests on the results. Secton IX concludes and n the appendx we lst the countres under dfferent defntons. II. Theoretcal Motvaton and Methodology Ths paper estmate the passthrough from deprecaton to nflaton n a panel data framework. few seres for each country are constructed usng monthly data (from 1980 to 1998 of 71 countres amountng to approxmately 14,013 vald observatons. The perod examned s qute rch snce t contans the large swngs n the dollaryen relatonshp, as well as alternate perods of tranqulty and mportant exchange rate crses. The panel data analyss allows the paper to pool a larger amount of data that was used n prevous studes. The followng seres were dentfed as potental determnants of the passthrough from exchange rate deprecaton to nflaton. Frst, a proxy for the busness cycle (GDP devaton from an estmated trend was ncluded to capture the noton that wth ncreasng sales frms fnd t easer to passthrough ncreases n costs to fnal prces. The reverse s also true. Large deprecatons sometmes do not mply large prce ncreases because the economy s n recesson and frms do not adjust ther prces proportonally to ther ncrease n costs. Second, we dentfed the real exchange rate as potentally affectng the passthrough. Prevous studes have shown that the real exchange rate (RER overvaluaton s an mportant determnant of future deprecatons (Goldfajn and Valdes, These deprecatons need not call for hgher nflaton, f they smply restore the real exchange rate to ts steady state. In ths case, the overvaluaton would be corrected by a change n the relatve prce of tradables non tradables, and the deprecaton would not generate a generalzed ncrease n prces. On the other hand, large deprecatons that are not based on requred adjustments n relatve prces would ether nduce nflaton or reverse tself through a future nomnal apprecaton (the stylzed fact s that the correcton of excess nomnal deprecatons tend to occur through hgher nflaton (Goldfajn and Gupta, Ths effect was also dentfed by Borenszten and De Gregoro (

7 Thrd, the nflaton envronment may determne the wllngness of frms to ncrease prces n the presence of ncreasng costs. The pass through s determned by the perceved persstence of costs changes, that s nfluenced largely by the persstence of nflaton. Snce nflaton tends to be postvely correlated to the persstence of nflaton, t may also be postvely correlated to the passthrough, as argued recently by Taylor (1999. Inflatonary countres would tend to have a greater degree of pass through, whle stable countres wll tend to mantan current low nflaton, even n the presence of a large deprecaton. mtrano et al 1997 has shown that the latter was ndeed the case after the large deprecatons n Europe n 199. Fourth, the degree of openness of a country to the rest of the world should also affect the passthrough coeffcent. The lterature has concentrated on the drect effect of openness on nflaton, showng how openness puts a check on nflatonary fnance n a BarroGordon type model wthout a commtment technology (Romer, Ths effect mples that one should observe a negatve correlaton between nflaton and openness. The effect of openness on the passthrough coeffcent works n the opposte drecton. In a more open economy, wth larger presence of mports and exports, a gven deprecaton has a larger effect on prces. In the appendx we develop a smple model wth endogenously determned tradable and non tradable goods, where the degree of openness nduces more pass through from deprecaton to nflaton. Note that ths varable s closely lnked to foregn frm partcpaton n the domestc market, a fundamental mcroeconomc determnant n the passthrough lterature (see Dornbusch, In the paper, the nomnal exchange rate (E s expressed as domestc currency unts per unt of foregn currency, or n other words, the prce of foregn exchange n terms of domestc currency. Thus t s possble to have more than 100% devaluaton n local currency. The real exchange rate (ε s the rato of the domestc prce level over foregn prce level expressed n the same currency, ε = p/e p*, where p s the good s prce n domestc currency, p* s the prce n foregn currency. The passthrough coeffcent s defned as the relatonshp between accumulated nflaton n j perods, Π [t,tj], and the exchange rate deprecaton also accumulated n perods 6

8 ê [t1,tj1], but allowng at least one month lag of nflaton s response to a change n the exchange rate. passthrough coeffcent near 1 s equvalent to a total passthrough of exchange rate deprecaton to nflaton, whle a coeffcent near zero represents a total nelastcty of the economy s prces to a change n the nomnal exchange rate. The nflaton seres were bult for each country from monthly data of seasonally adjusted CPI (Consumer Prce Index from the Informaton Notce System (INS database of the IMF. ccumulated nflaton s the dfference between the CPI ndex at tme t1 and tme t. Deprecaton was calculated as changes n the effectve nomnal exchange rate ndex, defned as tradeweghted nomnal exchange rate of domestc currency over foregn currency. The source s also the INS database. ccumulated nflaton s the dfference between the CPI ndex at tme t1 and tme t. proxy for openness was created usng the sum of exports plus mports as a percentage of GDP, all from the IFS database of the IMF. monthly seres of multlateral real exchange rate (RER was also used n the regresson estmaton (upward movements n the ndex ndcate apprecaton. Msalgnment of the RER was constructed for each country usng the percentage dfference between the actual RER and a HodrckPrescott flter of the RER. To control for the economc cycle, a seres of monthly ndustral actvty s level was used, or GDP at constant prces. Data from Internatonal Fnancal Statstcs of the Internatonal Monetary Found was used. In the absence of the monthly GDP, the quarterly or the annual ndex was used. Usng the HodrckPrescott flter, the paper calculates the percentage of devaton of the actual GDP from an estmated trend. postve varaton ndcates that the country was growng faster than the trend, whle a negatve varaton represents the opposte. The same procedure was appled for the real effectve exchange rate. Therefore, the equlbrum exchange rate s proxed by the estmated trend, nstead of assumng a purchasng power party equlbrum exchange rate (see Goldfajn and Valdes, One mght consder that usng the ntal GDP Devaton to explan the accumulated nflaton n the next 1 months s not approprate, gven that the busness cycle stance may change throughout the perod. The alternatve s to use an average GDP devaton for the 1 month perod. However, ths may generate endogenety problems, n other words, the average GDP devaton may be affected by the nflaton accumulated durng the 1month perod. In 7

9 any case, for robustness purposes, the exercse wll be redone usng the average GDP n chapter VII. lso, the ntal nflaton varable s denomnated n the perod t1, n order to avod a spurous relatonshp wth accumulated nflaton n the perod t to t1. The trade openness varable s measured at the t moment. It s nterestng to observe some stylzed facts of the data. In Table 1, t s clear that there s a dfference between medans and averages reflectng the fact that the dstrbuton of nflaton and deprecaton across countres and tme has a few extreme observatons related to the hypernflaton and hgh nflaton epsodes. The average or medan GDP gap s small as well as the medan RER msalgnment by constructon of the msalgnment ndces. The average t s not exactly zero because we could not use n the overall regresson all the data ponts we used to construct the msalgnment seres. Table 1: Descrptve Statstcs Statstc Inflaton Exchange Rate Openness RER Msalgnment GDP Gap Deprecaton verage Medan Number of Observatons Number of Countres Note: ll data n percentage terms. Source: INS(IMF, IFS(IMF, GDP varous sources III. The effect of deprecaton on nflaton In ths secton, the paper estmates drectly the passthrough coeffcent n order to shed lght on the overall effect of the exchange rate deprecaton on nflaton and observe ts behavor over tme (to establsh the stylzed fact observed n Fgure 1 n the ntroducton, across regons and development status. In the next secton, we wll nclude cross terms on the deprecaton coeffcent to explan the specfc determnants of the passthrough coeffcent. The followng equaton, that only nclude the solated effects of the ndependent varables over accumulated nflaton, was adopted to estmate the passthrough coeffcent: 8

10 (1 Π,[t,tj] = β 0 β 1 ê,[t1,tj1] β RER,t(1 β 3 GDP, t(1 β 4 Π,t(1 β 5 OPE, t(1 u where ndcates the country and t the tme. The nflaton rate and nomnal exchange rate deprecaton are accumulated durng a perod of tme and the other control varables real exchange rate devaton, ntal nflaton, GDP Devaton and trade openness are ncluded at tme t1. The results of equaton (1 are summarzed n Table where the estmated regresson usng fxed effects and generalzed least squares results are presented. ll the varables ncluded are sgnfcant. s expected, deprecaton, ntal nflaton and GDP above trend are postvely related to nflaton, whle overvaluaton and openness (n some cases dampen nflaton. There are mportant changes as the tme perod expands. For a tme horzon of 1 month (lower panel n Table the coeffcent's values are stll relatvely low, a 10% deprecaton wll only cause a 1.4% nflaton n the followng month. It s nterestng to observe that a 10 % overvaluaton wll dampen nflaton by 1.1%. Therefore, n a hypothetcal case where a 10% nomnal devaluaton corrects an exchange rate overvaluaton of the same amount, one would not observe an ncrease n nflaton, ceters parbus. larger coeffcent n ths tme horzon s assocated only wth the ntal nflaton varable, where each 1% of monthly nflaton wll generate a 0.5% of nflaton next perod, evdence of nerta n the very short run. The coeffcents are larger as the tme horzon of the regresson s expanded. The pure passthrough coeffcent jumps from n the frst month to n 3 months and 0.46 after 6 months, fnally reachng 0.73 after 1 months. Fgure allows one to observe the passthrough coeffcent as a functon of the tme horzon. The evdence from the panel regresson seems to confrm the casual observaton shown n the ntroducton, an ncreasng passthrough coeffcent over tme. The fgure also shows that the passthrough coeffcent reaches a lmt after 1 months, and ts magntude s smaller than 1, evdence that the rsks of a spral nflaton deprecaton s not overwhelmng (n contrast to the average behavor of Latn merca shown n Fgure 3. 9

11 The magntudes of the coeffcents of the control varables are relevant. The real exchange rate overvaluaton s mportant (n 1 months, a 10% of overvaluaton decreases nflaton by 11.8%. Therefore, a nomnal devaluaton that does not overshoot ts requred adjustment would not have severe consequences n terms of nflaton. The effect of the GDP devaton s much smaller. Graphcally, n Fgure, we can track the effects of these varables on the passthrough coeffcent by omttng t n the regresson of equaton (1. Note that the passthrough coeffcent s curve that gnores the real exchange rate devaton s permanently below the pure passthrough coeffcent curve, snce an overvaluaton sgnfcantly reduces the exchange rate deprecaton s passthrough to nflaton. The opposte s true for the effect of a postve GDP devaton. However, the effect s much smaller. We can dvde the sample by dfferent crtera. One possble subdvson s by geographcal crtera, nto fve regons. nother s to classfy the countres accordng to ther socal economc condton, or n other words, nto emergng, developed, and developng countres. lternatvely, wth the same purpose, we dvde the countres between OECD members and nonmembers. The results of the regonal subdvson can be seen n Table 3 and Fgure 3. For the frst month after devaluaton, the passthrough coeffcent s relatvely smlar n all the regons. Startng n the frst quarter, one can see that the mercan and san regons possess a hgher degree of passthrough of the exchange rate to hgher prces, than other regons. Certanly the mercan coeffcent s strongly nfluenced by the South mercan country s nflatonary tendences. The European coeffcent s very small, whch n fact s the subject of a study by mtrano,. (1997. Oceana can be consdered as the regon wth the smallest passthrough coeffcent, where all four countres that make up ths subsample also ndvdually exhbt very small coeffcents. Overall, the ncrease of the passthrough coeffcent up to 1 months s evdent for all regons. World Bank Classfcaton. See the complete lst n the appendx. 10

12 Fgure : Passthrough passthrough gnorng GDP devaton passthrough passthrough gnorng Real E devaton months Note:Estmated wth fxed effects and cross secton weghts The subdvson of the sample by economc development produces the results shown n Table 4. In sx months, the passthrough coeffcent s much larger n developng (0.34 and emergng markets (0.394 than n developed countres (0.45. In 1 months, the emergng countres have an almost complete (0.91 passthrough coeffcent, whle the developed and other developng countres have a passthrough coeffcent equal to and respectvely. Nevertheless, the dampenng effect of the real exchange rate devaton s greater n emergng countres than n the developed countres, a 10% overvaluaton reduces nflaton by 15.34% versus 7.9% n developed economes. The ntal nflaton varable appears to be more mportant n developed countres due to ther greater nflatonary stablty durng the analyzed perod. The results for OECD members and nonmembers subdvson confrm the results above. The passthrough coeffcent s n fact much lower for OECD members than non OECD. For 6 months, for example, a 10% nomnal exchange rate devaluaton would lead to 1.13% nflaton for OECD members and 4.71% for nonmembers. 11

13 Table : Passthrough Panel Regressons Wthout Cross Terms Dependent Varable: ccumulated nflaton rate ( t to t1 Varable Coeffcent Expected Sgn tstatstc Pvalue GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Dependent Varable: ccumulated nflaton rate ( t to t6 Varable Coeffcent Expected Sgn tstatstc Pvalue GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Dependent Varable: ccumulated nflaton rate ( t to t3 Varable Coeffcent Expected Sgn tstatstc Pvalue GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Dependent Varable: ccumulated nflaton rate ( t to t1 Varable Coeffcent Expected Sgn tstatstc Pvalue GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Note: Estmated wth Fxed Effects and Cross Secton Weghts 1

14 Table 3: Passthrough coeffcent by Regons Months Total Europe frca merca Oceana sa 1 0,01 0,018 0,018 0,013 0,00 0, ,169 0,116 0,159 0,199 0,051 0, ,46 0,11 0,343 0,539 0,09 0, ,73 0,360 0,643 0,69 0,158 0, ,701 0,460 0,50 1,40 0,193 0,841 Fgure 3: Passthrough By Regon 1, 1 0,8 0,6 0,4 0, months Passthrough Passthrough Europe Passthrough frca Passthrough merca Passthrough Oceana Passthrough sa 13

15 Table 4: Passthrough Regressons by Type of Country 1 Months Dependent Varable ccumulated Inflaton Fxed Estmaton Developed Emergng Other Developng coeffcent tstatstc coeffcent tstatstc coeffcent tstatstc GDP devaton 0,033 4,313 0,00 0,09 0,03 1,766 Intal Inflaton 4,701 5,1 1,19,083 1,489 1,789 Openness 0,033 5,87 0,041 9,68 0,068,070 Real E devaton 0,797 5,855 1,534 7,0 1,045,638 ccumulated Deprecaton 0,605 5,916 0,91 7,56 0,506,0 6 Months Dependent Varable ccumulated Inflaton Fxed Estmaton Developed Emergng Other Developng coeffcent tstatstc coeffcent tstatstc coeffcent tstatstc GDP devaton 0,007,17 0,015 3,441 0,018 1,998 Intal Inflaton 3,151 6,513 1,170 3,980 1,11 4,061 Openness 0,016 5,90 0,01 10,178 0,016 4,48 Real E devaton 0,15 5, 0,41 6,045 0,470 5,994 ccumulated Deprecaton 0,45 4,539 0,394 5,15 0,340 4,393 Note1: GDP devaton and Openness are at tme t. Intal Inflaton, Real E devaton and ccumulated Deprecaton are at tme t1. WhteCorrected standard errors. Source: Lst of emergng countres n appendx 1 Months Dependent Varable ccumulated Inflaton Fxed Estmaton OECD nonoecd coeffcent tstatstc coeffcent tstatstc GDP devaton 0,09 3,36 0,014 3,886 Intal Inflaton 4,336 8,156,087 1,98 Openness 0,031 6,815 0,088 6,60 Real E devaton 0,58 10,975 1,314 3,739 ccumulated Deprecaton 0,188 10,51 0,754 3,841 6 Months Dependent Varable ccumulated Inflaton Fxed Estmaton OECD nonoecd coeffcent tstatstc coeffcent tstatstc GDP devaton 0,016 3,054 0,010 3,498 Intal Inflaton,447 9,810 1,553 5,975 Openness 0,0 9,34 0,07 1,380 Real E devaton 0,119 11,675 0,530 9,665 ccumulated Deprecaton 0,113 8,838 0,471 8,088 Note: GDP devaton and Openness are at tme t. Intal Inflaton, Real E devaton and ccumulated Deprecaton are at tme t1. WhteCorrected standard errors. Source: Lst of oecd countres n appendx 14

16 We can summarze the results obtaned n ths secton as follows. The passthrough coeffcent s ncreasng as the tme perod ncreases. For example, the 1months coeffcent s more than 4 tmes larger the 3month coeffcent. The coeffcents for the control varables have the correct sgn and are statstcally sgnfcant. The economcally relevant coeffcents are the degree of exchange rate overvaluaton and the ntal nflaton. For example, for the 1 months case, an overvaluaton of 10 percent reduces subsequent nflaton n 11.8 percent. The passthrough coeffcent n the mercan regon s the hghest (1.4 n 1 months reflectng the spral nflatondeprecaton n several Latn mercan countres. Europe, frca and Oceana have a substantal lower passthrough coeffcent than n sa and merca. The passthrough s substantally lower n OECD (or developed countres relatve to emergng market economes. In contrast, the dampenng effect of the ntal overvaluaton s larger n emergng and other developng countres. IV. The determnants of the passthrough coeffcent It s relevant to analyze the effect of several of the control varables drectly on the passthrough coeffcent. Does the passthrough coeffcent depend on the ntal overvaluaton, GDP devaton, ntal nflaton or degree of openness? drect approach would have been to calculate PT t = Π [t,t] / ê [t1,t1] and regress aganst the ndependent varables. However, some countres presented deprecaton rates close to zero (due for example to a fxed exchange rate wth postve nflaton rates leadng to very hgh passthrough numbers. 3 Gven the large volatlty of the passthrough observatons and the hgh standard errors t produces, and to avod arbtrary droppng of data, we have chosen nstead to analyze the model ncludng cross terms n equaton (1 to avod large standard devatons. The paper estmates a functonal form that ncludes both the nteracton terms as well as the drect effect of the proposed varables. In fact, t s expected that the real exchange rate devaton, for example, be as mportant n explanng drectly the nflaton rate, as n determnng the passthrough 3 New Zealand, for example, exhbts a passthrough coeffcent of 1,188 n January of 198. In January of 198, the country had nflaton of 1.3%, and n December of 1981 nomnal deprecaton was only 1,1x10 4 %. Even though the country had a reasonable passthrough coeffcent durng other perods, the average of the coeffcent was and the standard devaton was Evdently, the medan was only

17 coeffcent through ts ndrect effect. For smplcty, t s assumed that that passthrough coeffcent s a lnear functon of the other varables n queston: Π [t,tj] = α β 1 ê, [t1,tj1] β RER, t(1 β 3 GDP, t β 4 Π, t(1 β 5 OPE, t u where β 1 = β 6 β 7RER, t(1 β 8 GDP, t β 9 Π, t(1 β 10 OPE, t Or: (3 Π, [t,tj] = α β 6 ê, [t1,tj1] β 7 ê,[t1,tj1] * RER, t(1 β 8 ê, [t1,tj1] * GDP, t β 9 ê, [t1,tj1] * Π, t(1 β 10 ê, [t1,tj1] * OPE, t β RER, t(1 β 3 GDP, t β 4 Π, t(1 β 5 OPE, t u The estmates are presented n Table 5, whch has the results for 1, 3, 6, 1 and 18 months. For the frst month after the deprecaton, the nteractve terms (those that make up the passthrough coeffcent are not sgnfcant, wth the excepton of ntal nflaton that shows the opposte sgn to that expected. The coeffcent on the solated terms contnues to show the correct sgn and sgnfcance. For three months, the nteractve terms results mprove sgnfcantly. For each 10% of postve GDP devaton, the passthrough coeffcent ncreases 4.65% ndcatng the mportance of the busness cycle to determne the degree of the devaluaton s passthrough on nflaton. The ntal nflaton and the exchange rate devaton varables also nfluence the passthrough wth the expected sgn and were sgnfcantly dfferent from zero. For the sxth month perod, one observes that effect of the ntal nflaton or the busness cycle on the passthrough coeffcent reached ts lmt. t the same tme, the exchange rate overvaluaton varable contnues to be mportant n determnng the passthrough coeffcent and ncreases n value durng the months n queston. n exchange rate prevously 10% overvalued, decreases the passthrough coeffcent by 0.09%, 1.77%, 3.48%, 6.78% e 7.06% respectvely durng the perods of 1, 3, 6, 1 and 18 months after the exchange rate deprecaton. Fnally, wthn a year, the trade openness varable s effect on the passthrough produces the expected sgn and sgnfcance over the passthrough coeffcent. Nevertheless, the solated term s sgn s dfferent than expected. s n the prevous secton, exercses were done for the nteractve terms model n whch the sample was dvded n regons. Table 6 contans the results for 6 and 1 months for 16

18 each regon. For the frst semester, note the nteractve GDP devaton varable s mportance for merca and Oceana, where for each 10% of actvty level below the trend, the passthrough coeffcent decreases by 19.54% and 31.6% for the respectve months. The ntal nflaton s role as a component n determnng the passthrough coeffcent s more relevant n Europe and n sa, where each 10% of monthly ntal nflaton provokes a passthrough coeffcent 0% and 48% greater, respectvely. The cross effect of the real exchange rate devaton on the passthrough coeffcent s mportant and relatvely stable among the regons n queston. We could summarze the results obtaned n ths secton as follows. In general, the RER overvaluaton, ntal nflaton, trade openness and GDP devaton do affect the passthrough coeffcent but n dfferent degrees. The most robust determnants are the RER overvaluaton and the ntal nflaton. RER overvaluaton s partcularly mportant for the passthrough coeffcent n the mercan regon but has nfluence n other regons passthrough coeffcent too. Intal nflaton s a partcularly mportant determnant for European countres. The nfluence of RER ncreases as the horzon ncreases whle the nfluence of ntal nflaton s lmted to 6month horzon. The GDP devaton and openness varables are more senstve to the horzon and sample chosen. The GDP gap has an mportant effect wthn a 6month perod. However, the sgn reverses at a 1month horzon. The probable reason s that the level of actvty at t has lttle effect on the degree of passthrough close to t1. In essence, t s possble that devaluatons are countercyclcal makng the ntal GDP gap negatvely correlated to future passthrough of the nflaton. Openness s partcularly mportant for the passthrough coeffcent n frca and Oceana. 17

19 Table 5: Passthrough Panel Regressons Wth Cross Terms months coeffcent Pvalue coeffcent Pvalue coeffcent Pvalue coeffcent Pvalue GDP*Ê INF*Ê OPE*Ê RER*Ê GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Note: Estmated wth Fxed Effects and Cross Secton Weghts Expected Sgn

20 Table 6: Passthrough Regressons Usng CrossTerms fxed estmaton Dependent Varable: 19ccumulated Inflaton 1 Months Regresson merca frca Europe sa Oceana coeffcent Pvalue coeffcent Pvalue coeffcent Pvalue coeffcent Pvalue coeffcent Pvalue GDP*Ê 0,890 0,000 0,370 0,636 0,04 0,353 0,139 0,45 5,19 0,000 INF*Ê 0,04 0,001 0,048 0,04 0,550 0,001 3,895 0,000 1,119 0,309 OPE*Ê 1,090 0,000 0,493 0,000 0,006 0,874 0,065 0,00 0,9 0,000 RER*Ê 0,97 0,000 0,411 0,000 0,611 0,000 0,780 0,000 0,318 0,000 GDP devaton 0,673 0,004 0,060 0,374 0,030 0,014 0,005 0,77 0,35 0,000 Intal Inflaton 3,745 0,000 0,475 0,018 3,93 0,000 0,71 0,000,635 0,000 Openness 0,60 0,000 0,006 0,807 0,04 0,007 0,040 0,000 0,117 0,000 Real E devaton 0,780 0,000 0,867 0,000 0,555 0,000 0,733 0,000 0,15 0,000 ccumulated Deprecaton 1,316 0,000 0,373 0,000 0,411 0,000 0,40 0,000 0,058 0,049 6 Months Regresson merca frca Europe sa Oceana coeffcent Pvalue coeffcent Pvalue coeffcent Pvalue coeffcent Pvalue coeffcent Pvalue GDP*Ê 1,954 0,000 0,481 0,335 0,744 0,000 0,366 0,003 3,160 0,001 INF*Ê 0,838 0,000 0,777 0,000 1,999 0,000 4,806 0,000,885 0,014 OPE*Ê 0,58 0,000 0,36 0,000 0,045 0,07 0,015 0,315 0,144 0,003 RER*Ê 0,57 0,000 0,74 0,000 0,31 0,001 0,166 0,155 0,384 0,000 GDP devaton 0,147 0,004 0,076 0,0 0,008 0,144 0,005 0,401 0,196 0,000 Intal Inflaton 1,554 0,000 0,310 0,000, 0,000 0,557 0,000 1,788 0,000 Openness 0,07 0,003 0,007 0,456 0,014 0,000 0,017 0,000 0,08 0,114 Real E devaton 0,8 0,000 0,381 0,000 0,184 0,000 0,6 0,000 0,097 0,000 ccumulated Deprecaton 0,586 0,000 0,15 0,000 0, 0,000 0,185 0,000 0,016 0,548 Note1: GDP devaton and Openness are at tme t. Intal Inflaton, Real E devaton and ccumulated Deprecaton are at tme t1. 19

21 0 Table 7: Passthrough Regressons by Type of Country Includng CrossTerms 1 Months Dependent Varable ccumulated Inflaton Fxed Estmaton Developed Emergng Other Developng coeffcent Pvalue coeffcent Pvalue coeffcent Pvalue GDP*Ê INF*Ê OPE*Ê RER*Ê GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Expected Sgn 6 Months Dependent Varable ccumulated Inflaton Fxed Estmaton Developed Emergng Other Developng coeffcent Pvalue coeffcent Pvalue coeffcent Pvalue GDP*Ê INF*Ê OPE*Ê RER*Ê GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Expected Sgn Note: GDP devaton and Openness are at tme t. Intal Inflaton, Real E devaton and ccumulated Deprecaton are at tme t1. 0

22 V. Forecasts It s mportant to check the accuracy of the model s forecasts usng known cases of large devaluatons and comparng the results to ther actual nflaton. Most of the cases, wth the excepton of Brazl, are part of the sample, so the predctons below have to be nterpreted as mostly nsample predcton. Frst, the paper looks at the predcton for nflaton n emergng markets such as Indonesa, Brazl, Korea, Mexco, Thaland, Malaysa and Phllpnes. Next, the paper looks at some developed country s experence as n the U.K., Sweden, Italy, Span and Fnland. Note that all of the known cases are devaluaton cases durng currency crss events. Ths wll have a bear on the results. Table 8a fully presents the forecasts results and the man components that determne annual nflaton n the countres that encountered exchange rate crss begnnng n 1994, accordng to the two equatons presented n ths paper. 4 The forecasts are done usng the general coeffcents attaned by ncludng the entre avalable sample, but also, alternatvely, by usng the specfc coeffcents of emergng and developed countres. Usng the model s general coeffcents, t can be seen that forecasted nflaton s greater, n all cases, than observed nflaton. Clearly, the factor that prmarly contrbutes to nflaton s the actual exchange rate deprecaton tself and the man buffer s the exchange rate overvaluaton that these countres encountered before the devaluaton. Thaland, for example, has a general forecasted nflaton of 18% (n contrast to observed nflaton of 10%. If we consder each varable ndvdual effect, we fnd that the exchange rate devaluaton contrbuted to 34.1% to nflaton whle the overvaluaton reduced t by 19.%. Other countres present a smlar stuaton. Indonesa can be consdered an excepton. large local currency devaluaton of 85%, made t nevtable to predct a very hgh nflaton, although the actual nflaton was only 8%. Usng, nstead, the emergng country s coeffcents, the upward bas n the forecast of nflaton s magnfed, snce the passthrough coeffcent s larger for ths group of countres.

23 It s worth notng that part of deprecaton s strong nfluence s counterbalanced by (a also larger real exchange rate devaton coeffcent. One could argue, therefore, that these countres behaved lke developed countres durng exchange rate crss perods due to ther lower passthrough coeffcent. The subsample that ncludes the European crss countres n 1991, show low nflaton n spte of large deprecatons, even below the forecasted nflaton (Table 8b. For example, Sweden had less than 5% annual nflaton after a more than 30% declne n the nomnal value of t s currency. Smlar to the prevous sample, the deprecaton appears to strongly nfluence nflaton, although t s countered by the ntal devaton of the real exchange rate. The trade openness acts sgnfcantly, but wth less ntensty. One must stress that ncludng the constant decsvely affects the forecasted nflaton, ndcatng that even f there were no deprecatons, nflaton would be around 4% annually from nflaton nerta. When the forecast s calculated by usng coeffcents for the developed countres, there are no clear advantages, snce even though a smaller passthrough coeffcent exsts, whch would cause lower expected nflaton, the exchange rate overvaluaton coeffcent s also lower, wth less restrant on nflaton. Tables 8c and 8d use the nteractve term model to forecasted nflaton. In general, ths model succeeds n attanng a better forecast than the solated terms model. The results for emergng markets usng the general coeffcents gves an mportant role to nomnal deprecaton and real exchange rate overvaluaton. The latter effect may be subdvded nto two: the frst effect, through ts drect reducton of nflaton (solated coeffcent and the second, even more mportant, by reducng the passthrough coeffcent whch ndrectly dscourages nflaton. lso, the trade openness varable appears to strongly affect nflaton, through the passthrough coeffcent. The process n Indonesa serves as a good example to emphasze ths pont. In the forecast usng only solated terms, the model underforecasts nflaton due to the very large devaluaton. In the cross terms model, the real exchange rate devaton not only dampens nflaton drectly but also multples the nomnal deprecaton and ndrectly reduces nflaton 4 Ths emergng group ncludes: Brazl (1999, Indonesa (1997, Korea (1997, Mexco (1994, Thaland (1997, Malaysa (1997 and the Phlppnes (1997.

24 3 by decreasng the passthrough coeffcent. Therefore, the predcted nflaton n the cross term model s closer to the actual nflaton n Indonesa than the prevous model. The forecasts for the European countres tend to produce better results when usng developed economes passthrough coeffcents than the general coeffcents. Once agan exchange rate deprecaton and the ntal exchange rate devaton were the man determnants of nflaton. The ntal nflaton, although only margnally, was nfluental n consoldatng future prce ncreases, contrary to ts relatvely lesser contrbuton n determnng the nflaton n emergng countres. Table 9 compares the forecasts of the two models and the results of Borenszten and De Gregoro. On average, ths paper produces slghtly better forecasts than these authors forecasts. Notwthstandng, a common problem s the downward bas n the forecasts, ndcatng a possble structural break durng crss perods, thereby reducng the passthrough coeffcent. In sum, the estmatons presented n ths secton seem to predct better than prevous studes, n partcular, Borenszten and De Gregoro, J (1999 that concentrates on crses cases. The complete model, that ncludes the cross terms, s a better predctor than the smplfed model. The man lesson for emergng markets s that one must take nto account the RER overvaluaton n the country n order to better predct nflaton. Equvalently, the effect of the ntal nflaton s fundamental to access nflaton n Europe or n developed countres n general. Notwthstandng these results, we fnd that there s a systematc upward bas n the predcton of nflaton n these crses cases. Ths s prevalent across all models. One possblty s that n these cases the nomnal exchange rate overshot and the expectaton s that t wll revert partally to ts mean, nducng smaller prce adjustment than n normal tmes. We wll nvestgate ths possblty n the next secton. 3

25 4 Table 8a: 1Month Inflaton Predcton n Selected Crses Cases Dependent Varable Type of Regresson Contrbuton to Inflaton n: General Indonesa 97 Brazl99 Korea 97 Mexco94 Thaland 97 Malaysa 97 Phllpnes97 Constant GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Predcted Inflaton ctual Inflaton Dependent Varable Type of Regresson Contrbuton to Inflaton n: Emergng Indonesa 97 Brazl99 Korea 97 Mexco94 Thaland 97 Malaysa 97 Phllpnes97 Constant GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Predcted Inflaton ctual Inflaton Dependent Varable Type of Regresson Contrbuton to Inflaton n: Developed Indonesa 97 Brazl99 Korea 97 Mexco94 Thaland 97 Malaysa 97 Phllpnes97 GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Predcted Inflaton ctual Inflaton Table 8b: 1Month Inflaton Predcton n Selected European Crses Cases Dependent Varable Type of Regresson Contrbuton to Inflaton n: General England 9 Sweden 9 Italy 9 Span 9 Fnland 9 Constant GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Predcted Inflaton ctual Inflaton Dependent Varable Type of Regresson Contrbuton to Inflaton n: Developed England 9 Sweden 9 Italy 9 Span 9 Fnland 9 Constant GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Predcted Inflaton ctual Inflaton

26 5 Table 8c: 1Month Inflaton Predcton n Selected Crses Cases Usng Cross Terms Dependent Varable Type of Regresson Contrbuton to Inflaton n: General Indonesa 97 Brazl99 Korea 97 Mexco94 Thaland 97 Malaysa 97 Phllpnes97 Constant GDP*Ê INF*Ê OPE*Ê RER*Ê GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Predcted Inflaton ctual Inflaton Note: Brazl s an estmate for the frst 1Month Dependent Varable Type of Regresson Contrbuton to Inflaton n: Developed Indonesa 97 Brazl99 Korea 97 Mexco94 Thaland 97 Malaysa 97 Phllpnes97 Constant GDP*Ê INF*Ê OPE*Ê RER*Ê GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Predcted Inflaton ctual Inflaton Dependent Varable Type of Regresson Contrbuton to Inflaton n: Emergng Indonesa 97 Brazl99 Korea 97 Mexco94 Thaland 97 Malaysa 97 Phllpnes97 Constant GDP*Ê INF*Ê OPE*Ê RER*Ê GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Predcted Inflaton ctual Inflaton

27 6 Table 8d: 1Month Inflaton Predcton n Selected European Crses Cases Usng Cross Terms Dependent Varable Type of Regresson Contrbuton to Inflaton n: Developed England 9 Sweden 9 Italy 9 Span 9 Fnland 9 Constant GDP*Ê INF*Ê OPE*Ê RER*Ê GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Predcted Inflaton ctual Inflaton Dependent Varable Type of Regresson Contrbuton to Inflaton n: General England 9 Sweden 9 Italy 9 Span 9 Fnland 9 Constant GDP*Ê INF*Ê OPE*Ê RER*Ê GDP devaton Intal Inflaton Openness Real E devaton ccumulated Deprecaton Predcted Inflaton ctual Inflaton Table 9: Comparng 1Month Predctons of Dfferent Models ctual Predctons Borenszten & Wthout Cross Terms Wth Cross Terms Deprecaton Inflaton De Gregoro General Emergng Developed General Emergng Developed Indonesa Brazl na Korea Mexco Thaland Malaysa Phllpnes ctual Predctons Wthout Cross Terms Wth Cross Terms Deprecaton Inflaton General Developed General Developed England Sweden Italy Span Fnland

28 7 VI. Usng expectatons One possblty for the upward bas of the forecasts s that the specfcaton does not capture properly the role of expectatons. If expectatons are taken nto account one could argue that only changes n the nomnal exchange rate perceved to be permanent should affect nflaton. Therefore, controllng for expectatons, the drect passthrough effect should be smaller. The paper uses a surveybased measure of expectatons from the FT Currency Forecasters magazne. The magazne collects opnons/estmates of 140 specalsts for the future exchange rate for 7 countres. Ths s a much smaller sample than the one used n ths paper (171 countres. In order to compare the effect of expectatons on the results, we redd the exercse n the prevous sectons, usng only the restrcted sample. 5 The expectaton varable s calculated as the expected exchange rate varaton for the next 3 or 6 months. The estmaton s based n the followng lnear form: (4 Π, [t,tj] = β 0 β 1 ê, [t1,tj1] β RER, t(1 β 3 GDP, t β 4 Π, t(1 β 5 OPE, t β 6 E(ê, [t,tj] u t, It s mportant to note the tmng that the expectatons varable s ntroduced. Intally, the expectatons were ntroduced a one month after the begnnng of the nomnal exchange rate deprecaton (at tme t. In ths way, the expectatons are formed wth the knowledge of the ntal extent of deprecaton. If the agents expect a reversal n the deprecaton process n the next few months, as opposed to a permanent deprecaton, one would expect a smaller adjustment n domestc prces. The results are shown n Table 10. Usng expectatons for 3 and 6 months (longer horzons are not adequate snce one would expect some learnng durng the process and the expectatons at tme t would not be representatve of future adjustment of prces. The passthrough coeffcent ncreases when one ncludes the expectaton varable, but only margnally. 7

29 8 For 6 months, the devaluaton expectaton coeffcent was sgnfcantly postve, confrmng that the more the exchange rate s expected to deprecate the greater s the markup of prces. However, the coeffcent s relatvely small, suggestng that ether the expected permanent deprecaton s small or that the ntal exchange rate expectatons do not affect substantally nflaton n the next 3 or 6 months. 6 One could argue that t s mportant to measure expectatons before the agents know the extent of deprecaton (at tme t1. The ratonal s to check the effect of unexpected deprecaton on nflaton. However, n the bottom panel of Table 10, one can see that there are no sgnfcant changes n the results. The cross terms model results may shed lght on the effect of expected deprecaton on nflaton (Table 11. The effect of the control varables on the passthrough from expected deprecaton to nflaton s, n general, not sgnfcant (see the cross terms coeffcents at the bottom of the table. Ths s n contrast to the sgnfcant cross term coeffcents regardng the passthrough of actual deprecaton to nflaton. Interestngly, the pure expected devaluaton coeffcent s reasonably hgh compared to the pure devaluaton coeffcent, n accordance to the permanent hypothess. Therefore, the cross term regresson allows us to conclude that exchange rate expectatons work as expected drectly, but not through the same channels that actual deprecaton affects nflaton. Ths secton has nvestgated the role of expectatons on the passthrough coeffcents and ts determnants. We found that although statstcally sgnfcant, the effect of expectatons on nflaton s relatvely small. Ths result s robust to changng the tmng of the expectatons varable (t or t1. One probable reason for ths result s that the horzon analyzed s large enough such that agents expectatons at the begnnng of the perod may not represent ther belefs along the perod. Therefore, prce changes throughout the perod may not always be correlated to expectatons measured at the begnnng of the perod. Nevertheless, the effect of expectatons becomes relevant and domnates the drect mpact of deprecaton once the effect of our man determnants of passthrough (RER, openness, ntal nflaton, GDP devaton s factored n. In other words, t s only the drect mpact of the passthrough 5 In ths way, any type of bas problem when comparng the model wth and wthout expectatons s avoded. 6 The emphass here s on the fact that expectatons are measured at the begnnng of the perod. 8