Working Paper No. 81. Exchange Rate Pass-through and Partial Dollarization: Is there a Link?

Transcription

1 Workng Paper No. 81 Exchange Rate Pass-through and Partal Dollarzaton: Is there a Lnk? by Jose Antono Gonzalez Anaya 1 December 2000 Stanford Unversty John A. and Cyntha Fry Gunn Buldng 366 Galvez Street Stanford, CA Senor Research Scholar, Center for Research on Economc Development and Polcy Reform

2 Exchange Rate Pass-through and Partal Dollarzaton: Is there a Lnk? Prelmnary and Incomplete Comments are Welcome November 5, 2000 José Antono González Anaya 1 Center for Research on Economc Development and Polcy Reform Stanford Unversty , joseg@stanford.edu Abstract Ths work examnes the way dollarzaton affects the transmsson channel of nomnal exchange rate fluctuatons nto domestc nflaton for 13 countres n Latn Amerca between Contrary to conventonal wsdom I fnd that dollarzaton across countres s not postvely assocated wth a hgher degree of exchange rate pass-through. Moreover, wthn a country, ncreases n dollarzaton are not assocated wth an ncrease n pass-through. The polcy mplcatons are that partally dollarzed countres can stll adjust ther real exchange rate through nomnal devaluatons and that ncreases n dollarzaton do not hnder ther ablty to do so. Two alternatve methodologes are followed: () long run pass-through s estmated usng an Error Correcton Model. When usng the 20-year tme perod, pass-through estmates for most countres are close to one supportng a long-term stable relatonshp close to PPP. When breakng the tme perod nto two, pass-through estmates are more heterogeneous. Long term pass-through coeffcents and dollarzaton are not sgnfcantly correlated across countres. There s a consstent but also nsgnfcant relatonshp between a hgher speed of adjustment and the degree of dollarzaton. () Yearly pass-through coeffcents were computed and regressed n a panel framework usng dollarzaton and nflaton as explanatory varables. The crosscountry dollarzaton coeffents are ambguous supportng the long run pass-through results. Wthn country estmates are nsgnfcant and splt n sgn. Inflaton s negatvely but not sgnfcantly correlated wth the degree of yearly pass-through both wthn and across countres. 1 Ths s a revsed verson of a paper prepared for the World Bank Conference on Dollarzaton n Latn Amerca held at the Unversdad Torcuato d Tella n Buenos Ares, Argentna on June The author thanks partcpants for many useful comments. Remanng errors are mne

3 1. Introducton Perhaps as a result of the nternatonal captal markets turmol, academc and polcy makers have revved the old debate of fxed versus flexble rates for developng countres wth a new varaton: the desrablty of relnqushng one s own domestc currency altogether and adoptng a hard currency nstead, namely the dollar. Ths paper takes the vew that n practce, the choce s not a dscrete one and that there s a contnuum of possbltes where from one extreme to the other a country can let t currency float freely, ntervene n the exchange market wth or wthout pre-determned rules, pursue a fxed exchange rate, mplement a currency board, or adopt the dollar as legal tender. Just n Latn Amerca, on one extreme, there s Panama where the dollar s legal tender and t s the only currency avalable, Ecuador s n the process of dollarzng, and Argentna has a currency board. On the other extreme, there are countres lke Brazl and Venezuela where dollar bank accounts are not allowed. In the mddle, there have emerged a whole seres of ad hoc regmes, whch vary n ther degree of dollarzaton: 2 Bolva, Peru, Ncaragua, and Uruguay mantan ther own currency and most transactons are denomnated n ther own currency but a hgh proporton of bank savngs are n dollars. The purpose of ths paper s not to advocate or even evaluate the welfare mplcatons a hgher or lower degree of dollarzaton but rather to nvestgate macroeconomc polcy makng under these partally dollarzed regmes. In partcular, whether dollarzaton affects the degree and speed of transmsson of nomnal exchange rate movements nto domestc nflaton;.e. the degree and speed of exchange rate pass-through of the nomnal exchange rate nto domestc nflaton. 3 The study of exchange rate pass-through n Latn Amerca s mportant to polcy makers for two reasons: () Because monetary polcy affects only the nomnal exchange rate drectly. The effects on the real exchange rate depend on the degree to whch movements n the nomnal exchange rate translate nto nflaton. A frm tenet of the monetary approach s that n the long run, a nomnal devaluaton translates one for one nto an ncrease n the domestc prce level elmnatng any changes n the real exchange rate. In practce, a nomnal devaluaton wll translate nto a real deprecaton f wages and prces are less than fully flexble and there s not an accommodatng monetary polcy. () When faced wth an external balance polcy makers n partally dollarzed economes have been reluctant to devalue the nomnal exchange rate (or ncrease the crawl dependng on the regme) on the arguments that because of a hgh degree of dollarzaton, nomnal exchange rate movements translate entrely and quckly nto nflaton;.e a hgher degree of dollarzaton mples as hgher and faster exchange rate pass-through. 2 The defnton of dollarzaton used n ths work s the proporton of bankng labltes denomnated n dollars. 3 The defnton of pass-through n ths work s gven by P / P φ = and s nterpreted as the degree to whch ( EP*)/ EP * changes n the nomnal exchange rate translate nto domestc prces. The clarfcaton s mportant because durng the late 1980s revval of the study of pass-through n the Unted States the nterest was on the effects of exchange rate movements on mport and export prces (See Baldwn and Krugman a,b,c, for the orgnal arguments)

4 The effects and presence of dollar ndexaton are well known: an economy that s strongly ndexed- and n partcular, wth exchange rate nfluences on ndexaton an attempt at creatng employment va easy money would be frustrated and exchange rate deprecaton precptates offsettng wage and prce nflaton. 4 In an economy that s perfectly dollar ndexed, a nomnal devaluaton translates nstantaneously nto a rse n all prces and wages barrng any real effects. In a hgh nflaton regme where prces are revsed very often, the nomnal exchange rate s often used as a benchmark and studes of hgh nflaton show close cumulatve movements of nternal prces and the exchange rate;.e. dollar ndexaton. 5 The argument n partally dollarzed economes arses because most of them experenced hgh nflaton and, n many cases, hypernflatons n the 1980s, durng whch the nomnal prce ndex tracked nomnal prce ndex qute closely becomng heavly dollar ndexed. In fact many prces were quoted n dollars on a day-to-day bass n countres lke Argentna, Bolva, and Peru. As nflaton abated, dollarzaton ratos ncreased, and many prces contnued to be quoted n dollars, conventonal wsdom smply assumed that hgh dollarzaton ratos mpled that the level of dollar ndexaton remaned hgh. 6 The ssue has seldom been contested to the pont even IMF country documents justfy the polcy almost as a matter of fact: the real effects of a nomnal devaluaton n country. are lmted due to the hgh degree of dollarzaton. 7 Usng prce and dollarzaton data for 13 countres n Latn Amerca, I found no sgnfcant cross-country or wthn-country relatonshp between dollarzaton and pass-through contradctng, the conventonal wsdom that predcts a postve relatonshp between the level of exchange rate pass-through and the degree of dollarzaton. The polcy mplcatons are that () countres wth a hgher degree of dollarzaton do not have a smaller ablty to adjust the real exchange rate through nomnal exchange rate fluctuatons and () that as a country s degree of dollarzaton ncreases, t does not loose ts channel to adjust the real exchange rate through nomnal devaluatons. I dd fnd a postve (but not sgnfcant) cross-country relatonshp between the speed of adjustment and dollarzaton and the level of nflaton. In addton, n a very long tme perod of 20 years, pass-through estmates for an expanded sample of Latn Amercan of countres are close to one (although only 3 out of 16 are statstcally nsgnfcant dfferent from one) makng Purchasng Power Party (PPP) a surprsngly reasonable benchmark. The results of ths paper suggest that the degree of dollarzaton and the degree of dollar ndexaton are not necessarly the same or even correlated. The prces of non-tradables are set by real determnants nsde the country just as theory would suggest. An example may clarfy the ssue. Prces n the hgh-end real state market n Mexco are quoted n dollars. Yet when a devaluaton occurs and economc actvty n the country falls, property prces n dollars are adjusted downwards to clear demand. Thus, even a perfectly dollarzed market wll not adjust fully and nstantaneously. 4 In a paper descrbng the Mexcan stablzaton experence Cordoba (1991) hghlghts the lack of ndexaton to past nflaton or the nomnal exchange rate n Mexco and warns of the effects n countres wth ndexaton. 5 Dornbusch (1988) although the same studes were quck to pont relatve prces showed large devatons 6 Where the author s part of ths conventonal wsdom. 7 The quote above s one of many examples from mssons to partally dollarzed countres but these IMF reports are confdental precludng makng specfc references

5 The rest of the paper s organzed as follows: Secton 2 dscusses some basc theory and the transmsson mechansms of nomnal exchange rate movements nto the domestc prce level. Secton three and the estmaton process. Secton 3 dscusses the data, the estmaton process and the results. And Secton 4 presents some conclusons. 2. Basc Defntons and Transmsson Mechansms The theory for the determnaton of exchange rate pass-through s not new. Ths sectons merely ctes the most relevant work. The pont of departure for the study of pass-through s the law of one prce and the Purchasng Power Party (PPP) lterature. Dornbusch (1988) n the New Palgrave presents an excellent defnton and a revew of the lterature. From the begnnng t became clear that there were varous reasons why the law of one prce and the PPP would not hold. Most of the tme, PPP does not hold n any nterestng sense. Certanly the noton that the equlbrum exchange rate s such that a dollar should buy the same basket of goods n the Unted States and Japan would only hold n an extraordnary world. What s really nterestng about PPP are the systematc drectons n whch the lterature has documented dvergences from PPP. 8 The purpose of ths secton s to brefly (re-)state the transmsson mechansms and the theoretcal reasons for devatons from PPP. The strong verson of PPP states that ther prce levels determne the exchange rate between two countres n any perod of tme. Therefore, f PPP holds, exchange rate fluctuatons translate nto proportonal movements n the domestc prce level;.e. pass-through s equal to one. PPP requres two restrctve assumptons: () That there s nstantaneous costless, and frctonless arbtrage. () That the same goods enter the basket of goods wth the same weght n every country. Surely nether of the above can hold all the tme leadng to the weak or relatve verson of PPP where the law of one prce holds up to a constant. It has also come to be known as the nflaton theory of exchange rates suggestng changes n the exchange rate between two countres are determned by the dfference of ther nflaton levels. (.e. eˆ = Pˆ Pˆ * ). The weak verson elmnates the requrement that arbtrage s costless but t does requre that t does occur at a constant cost. Ths wll clearly not be the case f there are quanttatve restrctons n place or f there are modfcatons n trade polcy. More mportantly, the determnaton of domestc nflaton may use dfferent shares of goods n ther respectve baskets and certanly non-traded groups are not the same and cannot be arbtraged. The lterature has dentfed dfferent types of structural and transtory devatons from PPP although pnpontng the source of the devatons has proved to be dffcult. The most mportant structural devaton from strong and weak verson of PPP arses from dfferences n productvtes or dfferences n productvty changes respectvely. The phenomenon was frst noted by Rcardo who noted that real prces of home goods are hgh n countres where manufactures floursh. 9 The mechansm, now called the Balassa-Samuelson effect, assumes the law of one prce apples to tradables. An ncrease n productvty n the traded sector puts upward pressure on the nomnal wage. Wthout a commensurate ncrease productvty n the 8 Dornbusch, R., Exchange Rates and Prces Introducton to Part III. 9 As quoted n Dornbusch

6 home goods sector, non-tradable prces ncrease. 10 Thus a country who s catchng because t has greater ncreases n productvty wll have observe an apprecaton of ts domestc prce level when measured n a common currency;.e. ts real exchange rate wll apprecate. The phenomenon has been documented n country cross-sectons and long-term tme seres. Other structural devatons from PPP can arse because of supply shocks, permanent Terms of Trade (TOT) shocks, changes tastes between traded and non traded goods, or changes n commercal polcy. 11 Countres n Latn Amerca were subject to most, f not all, of the above. Transtory devatons from PPP beyond those caused by transportaton and nformaton costs whch make arbtrage dffcult on a contnuous bass arse because of stcky prces and wages compared to exchange rates. The lterature has theoretcally justfed slow adjustng domestc prces and wages for many reasons. The mplcatons have been explored extensvely. Indeed slow adjustng prces are mplct n the standard Mundell-Flemng model of nternatonal macroeconomcs. The queston that the emprcal secton addresses s the sze and duraton of these temporary devatons. There s one notable type of shock whch does create a devaton from PPP even f the domestc prce ndexes have dfferent shares and goods n ther baskets: namely, a monetary shock when the condtons for homogenety postulate of monetary theory exst. In ths case, a change n the money supply wll lead to a proportonate change n all prces ncludng the exchange rate. Ths type of shock s mportant because of the hgh nflaton levels n Latn Amerca. Durng hgh nflaton, ncreases the spral ncreases n the money supply and the prce level can dwarf real shocks and offer support for PPP. There s a long hstory of emprcal evdence on PPP both n support and aganst t. Dornbusch 1988 presents a revew of the emprcal evdence untl that date. Most developments snce then have concentrated on expandng the data set usng more narrowly defned goods and mprovng the econometrc technques. The evdence from both lookng at nflaton dfferentals and at narrowly defned manufactured goods leaves lttle doubt that they have been large and persstent. If the evdence for PPP has been so tme dependent and nconclusve, why s ths exercse relevant for Latn Amercan countres at ths pont. The answer has to do wth nvestgatng the way the transmsson mechansms of monetary polcy have changed wth dollarzaton. Monetary polcy can cave real effects by changng relatve prces of tradable and non-tradable and by changng the cost of money n the asset markets. Moreover, the approach s to estmate the degree of exchange rate pass-through wthout attemptng to pass-judgment on the theory of PPP. In partcular, the emprcal estmates test whether ths transmsson mechansms has been affected by the dollarzaton of fnancal assets. Lookng at pass-through rather than at a verson of PPP or even the real exchange rate s the rght approach of lookng at the transmsson mechansms of nomnal devaluatons nto the prce level because one needs not be concerned about the factors that can upset PPP. 10 The result was re-stated by Harrod n the 1930s, Balassa n the 1960s, Samuelson who formalzed n the 1960s, and more recently n the Dornbusch-Fscher-Samuelson (1977) framework. 11 A good reference for a formal exposton of the mechansms s Dornbusch Macroeconomcs (1980)

7 Transmsson Mechansms The transmsson mechansms of fluctuatons of the nomnal exchange rate nto the domestc prce level are standard ones and wll not be formally developed: () Fluctuatons n the nomnal exchange rate change the prce of mports whch drectly affect the overall prce ndex. For a small country wth perfect competton at home and abroad, the change n the prce of mports should be one for one. However, Dornbusch (1987) showed that f there s olgopolstc competton, and/or mport and domestc goods are mperfect substtutes, exchange rate passthrough can be less than one as frms strategcally modfy ther prcng behavor and consumers change ther pattern of consumpton to ncrease or decrease mported goods. 12 Although the market structure models are n partal equlbrum, they provde nsght nto the way segmented markets of traded goods can systematcally devate from PPP. Usng a Cournot settng, he fnds that mport share and lower concentraton ncrease the degree of pass-through. In two types of models of mperfect substtutes he shows that pass-through falls as product dfferentaton ncreases. A body of lterature arose n the 1980s to explan the prcng to market behavor durng the pronounced dollar apprecaton and deprecatons. The arguments centered on mperfect competton n the form of sunk costs to entry (Baldwn and Krugman a,b, and c), and market share (Froot and Klemperer 1989, see Knetter 1992 for a good revew of ths lterature). Although Latn Amercan Countres are becomng more mportant to the Unted States, I assume these affects are small and temporary. () Besdes the drect effect that fluctuatons of mport prces has on the domestc consumpton basket, mports have an ndrect effect on the domestc prce level when they are use as nputs. The hgher the use of mported goods as nputs or ntermedate goods, the hgher the effect on the domestc prce level. () A thrd way n whch nomnal exchange rate fluctuatons can affect the prce level s through dollar ndexaton as dscussed before. The hgher the proporton of contracts denomnated n dollars, the hgher the degree of pass-through nto the domestc prce level. Specfcally, ths work wll estmate f dollarzaton has affected the degree of dollar ndexaton n partally dollarzed economes. The man factor affectng the speed of adjustment of the domestc prce ndex s the level of nflaton. The hgher the level of nflaton, the more often economc agents wll change ther prces. Therefore, a nomnal exchange rate shock wll be transmtted faster through the channels outlned above. The extreme case s n a hypernflaton when prces are adjusted daly or even more often. 3. Estmaton Process and Results Data I collected data for 16 countres n Latn Amerca. The data set for each country conssts of monthly seres from of the Consumer Prce Index as a measure of the domestc prce level; the nomnal dollar exchange rate; and the Unted States PPI and the G7 PPI as 12 See Dornbusch (1987) for a formal presentaton

8 measures of nternatonal prces. The exchange rate data came from the IFS. The consumer prce ndexes were obtaned ether from the IFS or drectly from natonal sources. Monthly data provded the approprate tme frame to look at prce fluctuatons and elmnatng the hgh varance that typcally accompanes more frequent observatons of the exchange rate. The domestc prce seres for most countres durng ths perod reflect a hgh degree of nstablty to the pont many countres resorted to elmnatng zeroes from ther currences to make transactons easer. In general, at the begnnng of the 1980s, the nflaton was moderate but rsng for most countres n the sample. Inflaton typcally rose out of control at some pont durng the 1980s. Fnally, some type of stablzaton plan was mplemented as early as the early 1980s as n Chle and Bolva and as late as 1994 n Brazl. The nternatonal prce seres exhbted a relatvely steady upward trend. Thus, most of the varance of the explanatory varables comes from fluctuatons n the nomnal exchange rate. Although conceptually one would want to use the nomnal effectve exchange rate nstead of the dollar exchange rate, usng the nomnal exchange rate makes the process consderably more cumbersome, the data s less relable, and t s not clear that much s ganed. For Central Amercan countres and Mexco the dollar nomnal exchange rate and the nomnal effectve exchange rate are very close. The ssue s relevant only for South Amercan countres where the Unted States s not the man tradng partner. The process s cumbersome because the weghts of the nomnal effectve exchange rate change wth trade patterns. More mportantly, nstead of usng the U.S. or the G7 PPI as an ndex of nternatonal prces, one would have to construct an ndex of nternatonal export prces usng the prce levels of tradng partners. 13 For example, f Peru and Bolva devalue ther currency concurrently wth respect to the dollar, then the nomnal effectve exchange rate wll not be affected for ths blateral trade relatonshp. The nternatonal prce level faced by ether of these countres wll change as her export prces change due to the blateral devaluaton. The problem s that most countres do not have export prces forcng the use of an mperfect measure such as the WPI. Two further reasons justfy usng the nomnal dollar exchange rate and the U.S. or G7 PPI: () It gave all the countres a common world prce whch s what one would expect from traded commodtes. Gven that all of these countres are small, ther nfluence on the prce of tradables should me small. () Followng McKnnon s (a, b, and c) argument for Asa, a case could be made, that countres n Latn Amerca use the dollar as a standard and that nternatonal trade n ths regon s heavly dollar ndexed. () In the very long run, for most countres n the sample dollar exchange rate devaluatons translated nto almost full pass-through of the domestc prce level ndcatng that n the end the prce of tradables n dollars can be assumed to be gven. The measure of dollarzaton used n ths study s the amount of bankng labltes whch are n dollars as a share of total bankng assets or M4, between 1990 and 2000;.e. the percentage of savngs n dollars. Gven the large efforts to prevent exchange rate exposure n banks, usng bankng assets or labltes as a percentage of a broad measure of money does not make a dfference. For the purposes of ths study, the approprate measure s the level of dollar ndexaton n the form of the share of contracts that are conducted n dollars. However, that data s not avalable. Moreover, ths s the conventonal measure of dollarzaton and snce a prmary 13 Obvously the weghts to construct the nternatonal prce ndex for country X have to be the same as those used for the nomnal effectve exchange rate

9 purpose of ths work s to evaluate whether dollarzaton (as defned by polcy makers) ndeed mples that a nomnal exchange rate fluctuaton translates nto a faster and closer to proportonal ncrease of the domestc prces. The reason for choosng the tme-perod s that t was mportant to have a long enough tme horzon. Many countres n ths Latn Amercan experenced pronounced apprecatons and reversals that lasted a few years. For example, Mexco s apprecatons and reversal cycles last sx years, Chle experenced a contnuous deprecaton from the outset of the debt cycle n 1982 untl roughly the mddle of 1988, before a steady apprecaton snce then. Ignorng the 1980s mght provde lead to the wrong conclusons. Estmaton Process The pont of departure for the estmaton process s the weak verson of PPP. In logs ths translates nto: * p = α + φ ( e p ) (2) where country,, t, t t p, t s the domestc prce level for country at tme t. α s the mark up constant for e, t s the nomnal exchange rate of country n pesos per dollar, nternatonal prce level, and φ s the estmated coeffcent of pass-through for country. p * t s the The frst estmaton problem encountered s that, not surprsngly, all of the seres were non-statonary and contaned a unt root makng conventonal estmaton methods napproprate. 14 The smplest alternatve s to dfference the data and test for statonarty. The problem s that the regresson estmates loose long term memory and the long-run equlbrum propertes embedded n the orgnal seres. In ths case, the long run equlbrum propertes of seres are exactly the law of one prce and PPP. The earler dscusson suggests that temporary devatons are expected for varous reasons but that there should be a stable underlyng long run relatonshp between the domestc and the nternatonal prce level when measured n a common currency. Therefore, the theory behnd the jont behavor of the seres naturally suggests a contegraton framework to estmate equaton 2. The estmaton method does not mpose PPP or the law of one prce. Rather t allows temporary devatons whle t estmates a long run relatonshp between the varables (f t exsts), whch may or may not be consstent wth PPP and the law of one prce. (See the dscusson below). For all the countres, the seres of domestc prces and nternatonal prces measured n domestc currency faled to reject a contegraton vector at the 5% sgnfcance level usng a Johansen Test both for the and for the perod. (The results of the tests are avalable upon request). The contegraton relaton allowed for trends n the seres and a constant because nomnal varables were used but t dd not allow a trend n the contegraton relatonshp. 14 Nelson and Plosser (1982) frst put forth the argument. See Gonzalez (1999) for a bref non-techncal descrpton of the ssues. See Hamlton for a formal treatment of the subject

10 From an economc standpont, the falure to reject ths smple form of the contegraton merely states that there s n fact a long-term stable relatonshp between the domestc and the nternatonal prce level when measured n a common currency. The result s conceptually consstent but does not mpose or by tself mply PPP. However, the stable long run relatonshp does mply that temporary devatons do cancel out 15 and that as such, only the permanent structural devatons from PPP dscussed n Secton 2 need to be consdered;.e. permanent or secular TOT shocks, trade lberalzaton, and most mportantly, productvty dfferentals. Econometrcally, the exstence of contegraton mples that the errors from the contegratng regresson are statonary, that there wll not be any spurous regresson results and that one can use an Error Correcton Model of the form p, t = µ + φ f + ( 1 δ )( p, t 1 f, t 1) + laggeddff + ε, t *, t e, t pt (3) where f =, φ s the long run pass-through coeffcent for country, and ( 1 δ ) s the speed of adjustment wll have statonary resduals. The speed of adjustment term s the coeffcent of the error correcton term and can be nterpreted as the dsequlbrum response to shocks. The ECM descrbed above s the approprate estmaton tool for varous reasons. Frst, t provdes an out of equlbrum response of the seres that reveals how quckly the seres return to equlbrum. Snce, all of these countres are small and wll not affect the world prce (measured as the U.S. or G7 PPI) then the speed of adjustment really does measure how fast the domestc prce level of each country returns to ts long term stable relatonshp wth the nternatonal prce level. And second, t estmates the long term stable relatonshp between the domestc and the nternatonal prce levels wthout mposng PPP or any other relatonshp on the data. 16 The ECM wll allow the long-term stable relatonshp to be one to one corroboratng PPP or dfferent from one because of any of the structural theoretcal arguments put forth n Secton 2. The estmaton process allows long-term apprecatng or deprecatng trends due to permanent changes n the determnants of the real exchange rate: productvty dfferentals workng through the Balassa-Samuelson effect, secular TOT shocks, permanent trade lberalzatons. In fact, I argue that n general en ECM s a better way of obtanng relevant nformaton from the PPP relatonshp. The lterature has concentrated on testng whether PPP holds n the strctest sense. Yet, there are many theoretcally sound and practcal realtes that cause ths relatonshp to fal hold n predcable ways. Ths estmaton process allows for these real shocks to occur but at the same tme preserves the fact that there s an underlyng reason for the domestc and the nternatonal prce seres to return to ther long run relatonshp. One can say PPP holds only when the pass-through coeffcent φ s equal to one. The result mples that, at least n the long run, nomnal exchange rate movements 17 are reflected n 15 Ths s not to say that each type of temporary shock cancels tself out, but that the set of temporary set cancel each other out n the aggregate. 16 The exstence of a stable long-term relatonshp was already establshed through the contegraton test. 17 Strctly speakng, movements n the nternatonal prce level. However, as dscussed before most of the varance comes from fluctuatons n the nomnal exchange rate

11 proportonal ncreases n the domestc prces level. Gven the restrctve theoretcal assumptons dscussed before the emergence of PPP s qute remarkable. It mples that aggregaton ssues do not play a systematc role. Moreover, t mples that the structural reasons for devatons from PPP n the aggregate cancelled each other out. For example, t s possble for a country to have a postve productvty dfferental (to be catchng-up ) that was offset by adverse TOT shocks and trade lberalzaton. In our estmaton process we cannot tell ether of the underlyng ssues at work and the estmate s smply consstent wth PPP. A pass-through coeffcent φ lower than 1, mples movements n the nomnal exchange rate (tmes the nternatonal prce level) translated on average nto less than proportonal ncreases n the domestc prce level;.e. on average the real exchange rate deprecated durng the perod. Permanent equlbrum real deprecatons n the real exchange rate arse because of a permanent an adverse TOT shock, trade lberalzaton, and most mportantly because productvty n the home country ncreased at a slower pace than the reference country. As dscussed before, snce these effects are not beng studed separately one can only conclude that n the aggregate the result was the domestc prce ndex deprecated even though t s qute possble that the one or more of the determnants of the real exchange rate had an apprecatng pressure. An exercse snce 1930 for England vs. the U.S. would exhbt ths result. Conversely, a pass-through coeffcent φ larger than one mples that country experenced a real apprecaton of ts currency on average. The reasonng s exactly as above and t would arse n a world where convergence domnated the other effects. A post war exercse for Japan would show ths result strongly. The out of equlbrum coeffcent δ allows us to calculate the speed at whch the domestc prce level of each country returns to ts stable long term relatonshp wth the world prce level. A lower speed of adjustment n country mples that the domestc prce level n that country takes a longer tme to return to ts long term relatonshp wth the prce level and that the relatonshp between the domestc and the nternatonal prce level s not as tght. A slow speed of adjustment also makes the estmated long-term relatonshp a poor predctor of the nomnal exchange rate n the short term. Long perods of adjustment can also result n changes to the real economy f there are non-convextes lke the ones explored n the 1980s U.S. passthrough lterature. Exchange Rate Pass-through Results The frst step was to estmate pass-through n the very long run usng equaton (3). A lkelhood rato test was performed to determne the optmal lag length. Although for some of the cases two lags were enough to render the resduals whte nose, for some of the countres the dynamcs were not completely captured by a two lag Vector Error Correcton (VEC). For unformty, all of the VEC were estmated wth three lags. Table 1 shows the long run pass-through coeffcents from The results show that n the very long run, 20 years, PPP s not a bad benchmark to predct exchange rates n Latn Amerca n the very long term. On one hand, out of 16 countres, only Chle, Colomba, the Domncan Republc and Paraguay, have coeffcents that are more than 20% off the PPP

12 benchmark of one. On the other hand, n strct terms, only Brazl, El Salvador and Ncaragua follow PPP wth estmates that are not statstcally sgnfcantly dfferent from one. Part of the reason s that estmates are very precse wth an average standard error of 3 percent of the value of the coeffcent. Table 1: Long Run Pass-Through Pass-through Coeffcent Standard Error Bolva Brazl Chle Colomba Costa Rca Domncan Republc Ecuador El Salvador Guatemala Honduras Mexco Ncaragua Paraguay Peru Uruguay Venezuela Regonal Average The plots of the real exchange rate shown n the Appendx ndcate that despte ths apparent convergence towards PPP, the real exchange rates fluctuate ostensbly n the regon. The frst pont to note s that durng the early 1980s most Latn Amercan countres observed large real deprecatons of ther currences. Most countres have not seen that level of apprecaton of the exchange rate snce. The second gross generalzaton s that real exchange rates tend to apprecate gradually and wth abrupt reversons. The tme perods between correctons vary by country wth some lastng a few years. Thus t s mportant to nsure that the estmaton perod ncludes at least a complete cycle. For most of the countres n the sample, pass-through was less than one ndcatng that domestc basket of goods lost value compared to the nternatonal basket when measured n a common currency. Only the Domncan Republc, El Salvador and Venezuela observed sustaned apprecatons of ther currency durng ths tme perod. Real exchange rate plots n the appendx corroborate the regresson results. The overall deprecaton can be explaned because of the large unsustanable collectve apprecaton leadng up to the debt crss, or to any of the structural arguments put forth above. However, as the plots show and as the regresson results argue below post stablzaton performance has been domnated by a catch up reflected n sustaned albet gradual real apprecatons

13 Table 2 shows pass-through coeffcents usng the same methodology splttng the tme perod nto two, from and from Ideally, one would lke to choose the tme perods ndvdually for each country to nsure that a complete apprecaton and reversal cycle was ncluded. For unformty, a decade was deemed long enough. Estmates are more dspersed than n the regressons and there s heterogenety n the pont estmates both wthn and across countres. The average standard error of the 1980s and the 1990s s fve and ten tmes larger than n the 20-year regresson respectvely. Econometrcally, the cuttng the tme perod n half had a strong detrmental effect on the precson of the estmates. Economcally, the larger standard error n the 1990s mples that the varance of the real exchange rates ncreased. On one hand, the result s counterntutve gven the number of countres that mplemented prce stablzaton plans. On the other t mples that pure monetary dsturbances, whch are the only ones for whch PPP hold wth less restrctons, played a smaller role n the 1990s. Table 2: Decade Run Pass-Through Coeffcent Standard Error Coeffcent Standard Error Bolva Brazl Chle Colomba Costa Rca Domncan Republc Ecuador El Salvador Guatemala Honduras Mexco Ncaragua Paraguay Peru Uruguay Venezuela Average* *Not ncludng El Salvador Ncaragua snce 1992 Surprsngly, even for as long a tme perod as a decade, pass-through coeffcents are qute dfferent from one. Lookng at the evdence across countres, only Colomba n the frst 18 It s reassurng to know that the estmates of pass-through obtaned here concde wth estmates elsewhere n the lterature. Hausman, Panzza, and Sten (2000) fnd estmates of pass-through for Colomba, Mexco, and Peru that are remarkably close and estmates for Paraguay and Guatemala whch are wthn 10 percent. More over, the Mexcan estmate s consstent wth an exercse performed by Garce-Daz when the same tme perod was used

14 decade s consstent wth PPP. 19 Durng the 1990s, seven countres are consstent wth PPP: Bolva, Chle, Colomba, Costa Rca, Guatemala, Ncaragua, and Uruguay whle Brazl and Mexco have pont estmates that are close to one. However, n half of the sample, the pont estmates dffer from one by more 20%. Had there not been convergence to one n the 20-year sample, the evdence would be nterpreted as smply another example that PPP does not hold. However, the gven the 20-year result, the dsperson decade results combned wth the speed of adjustment dscusson below ndcate that temporary devatons from the long run relatonshp between the domestc and the nternatonal prce can take a long tme. For wthn country estmates, four countres n the sample have statstcally dfferent pass-through estmates between the 1980s and the 1990s: Chle, Venezuela, Guatemala, El Salvador, and Honduras. For the frst three the pattern follows a deep deprecaton n the aftermath of the debt crss and a sustaned recovery n the 1990s. The rest of the countres have estmates that are not statstcally dfferent from each other. The result could mean stablty n the exchange real exchange rate but as the plots n the Appendx show, most of the tme t smply means that the reversal took place wthn the estmaton perod, as n Mexco. The results for Mexco and Brazl present an nterestng contrast. Mexco exhbted a constant coeffcent n both decades that s close to one. Brazl s pass-through estmate ncreased from 0.82 to 0.92 reachng a level smlar to Mexco s despte drastcally dfferent measures of openness. Part of the explanaton s that although both countres had mportant departures from PPP, the reversals occurred wthn a decade makng the pass-through estmate converge. Furthermore, pass-through estmates remaned constant despte the fact that the nflaton average was substantally dfferent between countres and decades ndcatng that nflaton does not affect the degree of pass-through nto domestc prces although as argued before and documented below, t may affect the speed of adjustment. The results for the Central Amercan Countres are a puzzle. Pass-through estmates devate wldly from PPP. On one hand, ths s not what one would expect from small countres wth most of ther trade wth the Unted States. On the other, most of these countres are prmary commodty producng countres subject to large TOT shocks and restrcted access to captal markets to smooth out the shocks. In addton, ther tradable sector s small makng the transmsson channel nto domestc prces small. Table 3 shows half-lfe duraton of shocks..e. how long does t take for a the domestc prce level to absorb half of the shock. The adjustment process slowed down n the 1990s compared to the 1980s. The average half-lfe ncreased from 21 months to 35 months. As argued n secton 2, hgher rates of nflaton are assocated wth more frequent prce changes and the average nflaton rate n the regon durng the 1980s was hgher. 19 Although Costa Rca, Ecuador and Peru 95 percent confdence ntervals are wthn 0.01 of ncludng the PPP benchmark value of

15 Table 3: Pass-through half lves n months 1980s 1990s Bolva 9 34 Brazl Chle Colomba Costa Rca Domncan Republc Ecuador El Salvador Guatemala Honduras Mexco Ncaragua Paraguay Peru 4 5 Uruguay Venezuela Regon Average Table 4 alternate presentaton shows the tremendous heterogenety n the estmates of the speed of adjustment. On one hand, n Peru half of the shock s transmtted n to the domestc prce level n four or fve months whle n Honduras t takes eght years! Although not entrely convncng, the results are consstent wth other results n the lterature. 21 Nevertheless, an avenue for further research s to refne the estmaton process and nvestgate f the speed of adjustment results s robust because the cross-country results are not ntutve. For example, Peru s quck translaton of exchange rate movements nto the domestc prce level s contrasted wth Bolva s 2-3 years half-lfe despte the fact that both countres have a smlar degree of openness, level of nflaton n the 1990s and, as wll be dscussed below, smlar dollarzaton ratons. The man dfference could only be explaned by the fact the end of the hypernflaton n Peru came n 1992 n contrast wth Bolva, whch came n 1986, and the dollar ndexaton memory n Peru has mantaned a quck transmsson mechansm. If that were the case one would expect the speed of adjustment n Peru to fall n the future. 20 The average excludes El Salvador and Honduras. 21 Housman et al (2000) estmated half-lves that were consstent wth those estmated here for the countres that overlapped. For example, they found that n Germany, the half-lfe was 130 months. Over 10 years long

16 Table 4: Dstrbuton of Duraton of Exchange Rate Shocks Pass-through half-lfe 1980s 1990s Less than 6 moths Peru Peru 6 months-year Bolva, Costa Rca Ncaragua 1-2 years Brazl, Chle, Brazl, Domncan Guatemala, Republc, Guatemala, Mexco, Paraguay, Mexco, Ncaragua, Venezuela 2-3 years Colomba, Ecuador, Ncaragua, Uruguay More than 3 years Domncan Republc, El Salvador, Honduras Paraguay, Uruguay Bolva Chle, Colomba, Costa Rca, Ecuador, El Salvador, Honduras, Venezuela Once agan the Central Amercan Countres present a puzzle. On one hand Guatemala, and Ncaragua have half-lves between 1-2 years. On the other, Costa Rca, El Salvador and Honduras have half-lves of over three years. A possble explanaton s that most dsturbances to the prce level are not monetary but due to TOT shocks. Exchange Rate Pass-through and Dollarzaton Wthn and Across Countres Fgures 1 a) and b) present plots for the degree of dollarzaton n 11countres n Latn Amerca. For expostonal purposes only, the sample was dvded nto hgh and low dollarzaton countres. In one end, Bolva and Uruguay exhbt dollarzaton ratos of over 80% by the end of the decade. Peru and Ncaragua ncreased ther dollarzaton raton from about around 30% n 1990 to about 70% by the end of the decade. And on the other end, Mexco and other Central Amercan countres had low farly constant dollarzaton ratos whle Brazl and Venezuela have a zero degree of dollarzaton by ths defnton. Modelng the drvng forces behnd dollarzaton s left as a future venue of research but there are two casual observatons: () The degree of dollarzaton presents a hgh degree of statonarty, and apparent rreversblty. Only Mexco and Chle exhbt sustaned reductons n the dollarzaton rato, albet small declnes at low levels. Thus, once a country s bankng or monetary aggregates become dollarzed, t appears to be dffcult to reverse the dollarzaton process. It may be that authortes have not actvely tred to reverse the stuaton but f they have, efforts have been unsuccessful. () Dollarzaton appears to ncrease as a result of extreme prce stablty and a stablzaton program that lberalzed the fnancal sector enough to permt savngs n dollars. Unfortunately, the data for Bolva was not avalable snce 1986 when dollar bank accounts were frst allowed but the fact remans that by 1990 the dollarzaton rato was 80%. Moreover, the pattern n the early 1990s for Peru suggests that dollarzaton ncreases after extreme prce stablty but after lberalzaton n the fnancal sector. Ecuador s meteorc rse snce the mddle 1990s also arses from the nstablty n the prce level at frst and then because of the delberate move towards dollarzaton. It s nterestng that the as of 1999 there were stll a

17 sgnfcant proporton of savngs n the domestc currency. The rse n dollarzaton rato n Paraguay concdes frst wth the openng of the captal account and the fnancal sector, and second wth the loss n prce stablty and bankng falures. 0.4 Fgure 1a: Low Dollarzaton Latn Amercan Economes Proporton of Bankng Assets n Dollars crdolr hodolr chdolr esdolr mxdolr Fgure1b: Hghly Dollarzed Latn Amercan Economes Proporton of Bankng Assets n Dollars bodolr ndolr pedolr urdolr ecdolr3 pydolr Fgure 2 shows long run exchange rate pass-through from vs the dollarzaton rato n 1990, the mddle of the perod. As dscussed above, long run pass-through tends to center around one and from the plot the degree of pass-through appears to be ndependent of the degree of dollarzaton. There appears to be more pass-through varance n countres wth lower dollarzaton ratos but that may be due to the fact the country sample has few hghly dollarzed economes at the begnnng of the 1990s

18 Fgure 2: Exchange Rate Pass-through vs Dollarzaton Rato Dollarzaton Rato (FCD/M3) Uruguay Bolva Peru Ncaragua Costa Rca Chle Ecuador Paraguay Mexco El Salvador Honduras Brazl Venezuela Pass-through Fgures 3a,b present estmates of pass-through n the 1980s and n the 1990s vs the dollarzaton rato at the begnnng and at the end of the decade respectvely. In both tme perods the relatonshp between dollarzaton and pass-through s not sgnfcant. Durng the 1980s, there s no obvous pattern. Durng the 1990s, f anythng the plot suggests a negatve relatonshp between pass-through and dollarzaton because hghly dollarzed economes tended to have lower pass-through. Moreover, the dollarzaton rato explans very lttle of the varance of pass-through evdence by low R-squared. The plots suggest that the degree of dollarzaton as measured here does not play a sgnfcant role n the transmsson mechansm of nomnal exchange rate fluctuatons nto the prce level;.e. dollarzaton as conventonally measured s not the same as the dollar ndexaton phenomenon whch occurred n many of these countres durng hgh nflaton. Fgure 3a: Pass-through vs Dollarzaton Rato (1990) Dollarzaton Rato (FCD/M3) PT = DR (t=-1.64) R2=0.05 Venezuela Chle Ncaragua Paraguay Brazl Bolva Mexco Uruguay Peru Costa Rca Ecuador El Salvador Honduras Pass-through

19 Fgure 3b: Pass-through vs. Dollarzaton Rato 2000 Dollarzaton Rato (FCD/M3) PT = DR (t=-1.64) R 2 = 0.20 Peru Ncaragua Paraguay Bolva Uruguay Mexco Brazl Honduras Ecuador Costa Rca El Salvador Chle Venezuela Pass-through In short, n countres wth a hgher degree of dollarzaton a nomnal devaluaton does not necessarly translate more fully nto an ncrease n the prce level. Therefore, a more dollarzed economy does not have a lesser ablty to adjust ts exchange rate through nomnal devaluatons than a less dollarzed economy. The mportant queston s whether as a country becomes ncreasngly dollarzed, does exchange rate pass-through ncrease and the authortes loose ther ablty to adjust the real exchange rate through nomnal exchange rate devaluatons? The answer cannot be derved from the plots but t wll be dealt wth n the regresson analyss. Fgures 4a,b shows a plot of the speed of adjustment vs the dollarzaton rato. The relatonshp s postve but, not sgnfcant. 22 Agan, the dollarzaton rato explans a small fracton of the varance of the speed of adjustment. The b-polar dstrbuton of the speed of adjustment s odd and as dscussed before the slow speed of adjustment for Central Amercan Countres s a puzzle. Brazl and Mexco are notable outlers wth quck responses and low dollarzaton ratos. The explanaton, partcularly for Mexco, could be that the degree of dollarzaton measured here understates the prevalence of dollar ndexaton, whch arses because of strong lnks wth, and the fact that anecdotal evdence suggests that many ctzens hold accounts n the Unted States. The faster speed of adjustment means that nomnal dsturbances are shorter lved, but not that they are dfferent n magntude. 22 The negatve slopes ndcate that a hgher dollarzaton rato s (weakly) correlated wth a faster speed of adjustment

20 Fgure 4a: Speed of Adjustment and thedollarzaton Rato Bolva Uruguay HL= DR (t=-0.57) R 2 = 0.06 Dollarzaton Rato Peru Costa Rca Ncaragua Chle Ecuador Mexco Paraguay Venezuela El Salvador Brazl Honduras Half Lfe of pass-through adjustment n months Fgure 4b: Speed of Adjustment and thedollarzaton Rato Dollarzaton Rato Peru Ncaragua Brazl Uruguay Paraguay Mexco Bolva Chle HL= DR (t=1.49) R 2 = 0.18 Costa Rca Ecuador Honduras El Salvador Venezuela Half Lfe of pass-through adjustment n months A Smple Panel Regresson Analyss I constructed yearly pass-through coeffcents wth a geometrc dstrbuted lag/vector Auto Regresson wth one lag. Ths technque was chosen over an error correcton vector because the tme perod s too short to assume contegraton. Usng a VAR permts gnorng the other factors that affect the domestc prce level wthout creatng an omtted varable bas. The prce settng process for each country s gven by: p, t j= 0 = λ + φ δ f ε (4), t t +, t

21 where 1 δ s the factor by whch the mpact of an exchange rate (nternatonal prce) shock dmnshes every month. I follow the standard procedure to estmate these types of equatons to obtan: 23 pt = α + φ f, t + δ p, t 1 + υ t (5) where α = λ ( 1 δ ) and υ t = ε t δε t I estmated equaton 5 for each country between to obtan 10 yearly passthrough coeffcents for each of the 11 countres. The pass-through coeffcents from equaton 5 can be nterpreted as the effects of a nomnal exchange rate fluctuaton on the domestc prce level n any gven year. The coeffcent s not the long-run relatonshp of the domestc and the nternatonal prce levels but the yearly contemporary effect. Therefore, to some degree the yearly pass-through coeffcent embodes both the level and the speed of adjustment;.e. a country wth hgh pass-through but a slow speed of adjustment wll be estmated as havng a low coeffcent. The conceptual dscusson and the graphcal dscusson above suggested that exchange rate pass-through could depend on the degree of dollar ndexaton (measure by the dollarzaton rato), and the level of nflaton. Therefore, the next step was to use the estmated exchange rate pass-through coeffcents from equaton 5 as dependent varables to construct a balanced panel of 11 countres for 10 years from resultng n a total of 109 observatons. The panel regresson took the form where φ Dollr +,, t µ + β, t, t + θπ, t ζ t = (6) Dollr, t and π, t are the yearly dollarzaton ratos and the nflaton rates respectvely. Table 5 presents the pooled and wthn estmators of the panel regresson. The pooled, cross-country results corroborate the plots n Fgures 3 and 4 a) and b). There s no sgnfcant relatonshp between dollarzaton and exchange rate pass-through across countres. Wth and wthout yearly nflaton the coeffcent s postve but nsgnfcant. Thus, countres wth a hgher degree of dollarzaton do not necessarly have a hgher pass-through and they can stll nfluence ther real exchange rate through nomnal devaluatons. The dscusson suggested that nflaton should ncrease the speed of adjustment and thus the yearly pass-through coeffcents. Wth more frequent prce revsons, hgher nflaton would be expected to be assocated wth a faster speed of adjustment of the domestc prce level to nomnal exchange rate shocks. However, the pont estmate was negatve although nsgnfcant. As a whole the both of the explanatory varables explan only a ffth of the varance of the passthrough coeffcents ndcatng that there are other more mportant varables. 23 One multples an expresson for, t 1 equaton I gnored the second-order term. p tmes δ and substracts t from the t p, expresson. In the resultng