Essays on Exchange Rates and Emerging Markets. Ezequiel Aguirre

Transcription

1 Essays on Exchange Raes and Emerging Markes Ezequiel Aguirre Submied in parial fulfillmen of he requiremens for he degree of Docor of Philosophy in he Graduae School of Ars and Sciences COLUMBIA UIVERSITY 2

2 2 Ezequiel Aguirre All righs reserved

3 ABSTRACT Essays on Exchange Raes and Emerging Markes Ezequiel Aguirre This disseraion consiss of hree essays on exchange raes and inernaional finance wih an emphasis on emerging economies. In Chaper, I provide empirical evidence ha suppors he hypohesis ha exchange rae based sabilizaion programs are expansionary during heir early phases. I derive a new se of sabilizaion episodes using exensive counry chronologies from Reinhar and Rogoff (24) and I find ha even afer conrolling for exernal condiions, he iniial expansion associaed wih he inroducion of an exchange rae based program, is caused by boh, he program iself and posiive exernal condiions. These expansionary effecs are robus o differen esimaion mehods and differen crieria for deecing sabilizaion episodes. In Chaper 2, I sudy he relaionship beween foreign ineres raes, counry spreads, erms of rade and macro fundamenals in emerging markes. I esimae a srucural VAR for 5 emerging economies. I find ha counry spreads explain 2% of oupu flucuaions, foreign ineres raes an addiional 7% and he erms of rade abou 5%. I also find ha counry spreads accoun for a quarer of real exchange rae variabiliy while he erms of rade accoun for jus %. To furher validae hese resuls, I develop a dynamic sochasic general equilibrium (DSGE) model for a small open economy. The model incorporaes several open economy fricions: i) bond-holding adjusmen coss, ii) invesmen adjusmen coss, iii) a working capial consrain, and iv) a counry spread componen ha depends upon macro fundamenals, which is aken from

4 he esimaed VAR. The model is able o replicae fairly good he propagaion effecs of foreign raes and counry spread shocks bu overesimaes he imporance of he erms of rades. In Chaper 3, I invesigae he relaion beween volailiy in he foreign exchange marke and excess reurns on carry rade porfolios for he G currencies. I develop and compare hree differen invesmen sraegies ha aim a avoiding losses when volailiy jumps, a common feaure of he carry rade. I find ha wo rading sraegies, one based on implied volailiy from F opions and he oher on exponenially-weighed moving averages, provide beer risk-adjused reurns han he sandard carry rade. A hird sraegy, based on Markov-swiching exchange rae forecass, provides excess reurns for some currencies bu fails for porfolios of currencies. I also show ha currency invesing provides superior Sharpe raios han a benchmark bond porfolio and a benchmark sock porfolio, even afer including he recen global financial crisis.

5 Conens Lis of Tables Lis of Figures Acknowledgmens iv vi viii Chaper Exchange rae based sabilizaion programs and exernal Condiions. Inroducion.2 Idenificaion of he Sabilizaion Programs Daa Descripion and Summary Saisics...4 Likelihood of Sabilizaion Plans Empirical Evidence Empirical Mehodology and Resuls 22.6 Robusness Checks Conclusions. 26 Chaper 2 Business Cycles in Emerging Markes and Implicaions for he Real Exchange Rae Inroducion Empirical Regulariies Daa Descripion Business Cycle Saisics. 42 i

6 2.2.3 Esimaion of he VAR Sysem Impulse Response Funcions Variance Decomposiions The Theoreical Model Households Timing of Evens Firms Compeiive Equilibrium Shock Processes and Funcional Forms The Real Exchange Rae Esimaion Calibraion and Bayesian Esimaion Priors and Poserios Esimaes Model Soluions Impulse Responses Conclusions 9 Chaper 3 The Carry Trade in he F Marke Inroducion Volailiy and Carry Trades Three Models for Currency Trading Markov Swiching Regime Model Exponenially Weighed Moving Average Model... 4 ii

7 3.3.3 Implied Volailiy Model Daa Sample and Sources. 3.5 Esimaion Resuls Cumulaive Reurns: Cumulaive Reurns: Sharpe Raios CAPM Regressions Conclusion. 3 Bibliography 33 Appendix A Appendix o Chaper 38 Appendix B Appendix o Chaper 2 4 B. Daa Appendix.. 4 iii

8 Lis of Tables. Reinhar & Rogoff classificaion scheme 8.2 Average saisics before and afer sabilizaions 6.3 Average saisics before and afer sabilizaion. 7.4 Probi model Panel regression: Real GDP growh Panel regression: Per-capia GDP Panel regression: Consumpion Panel regression: Invesmen Panel regression: Trade balance Panel regression: Alernaive ERBS idenificaion Panel regression; Alernaive ERBS idenificaion Business cycle saisics Business cycle saisics Business cycle saisics Parameer esimaes of VAR model Variance decomposiions Parameer values Summary saisics: F volailiy Reurn saisics and Sharpe raios, iv

9 3.3 Reurn saisics and Sharpe raios, CAPM regressions, CAPM regressions, A. Exchange rae based sabilizaion programs, v

10 Lis of Figures. Annual inflaion before and afer an ERBS was sared...2 Real GDP growh before and afer an ERBS was sared 2.3 Real GDP growh before and afer an ERBS plan sared relaive o world growh Oupu and ineres raes in emerging economies Oupu and exchange raes in emerging economies Real exchange raes and ineres raes in emerging economies Impulse response o an oupu innovaion Impulse response o a erms-of-rade innovaion Impulse response o a US rae innovaion Impulse response o a counry spread innovaion Priors disribuion Priors and poseriors Impulse response o a US rae shock Impulse response o a counry spread shock Impulse response o a TOT shock An example of he Markov-swiching esimaion model An example of EWMA volailiy for he AUD Volailiy in he G currencies 9 vi

11 3.4 Implied volailiy in G currencies Markov-swiching model: cumulaive reurns EWMA volailiy model: cumulaive reurns Currency porfolios: cumulaive reurns Implied volailiy model: cumulaive reurns Currency porfolios: cumulaive reurns. 2 vii

12 Acknowledgmens Several people helped me during my docorae sudies. I would like o firs acknowledge my advisors Richard Clarida and Marin Uribe, for heir invaluable suppor and guidance hroughou my disseraion wriing process. My undersanding of economic science have grealy benefied from simulaing conversaions wih hem. I would also like o express graiude o he oher members of my disseraion commiee: Guillermo Calvo, Sephanie Schmi-Grohe and Jaromir osal. I hank hem for heir helpful commens and suggesions. I am especially indebed o Juan Dubra, who encouraged me o pursue my graduae sudies in he firs place. His consan advice as an economis, simulaing discussions and uncondiional suppor as a friend were of immense value o me over all hese years. Finally, his disseraion would have no been possible wihou he suppor and love of my grandfaher Vicor, my parens Viviana and Fernando, my siser Vicoria and my uncle Ludovico. I dedicae his o all of hem. viii

13 A Vicoria y Vicor ix

14 Chaper Exchange rae based sabilizaion programs and exernal condiions. Inroducion High and persisen inflaion has long been a major economic issue for developing counries, being Lain America one of he mos affeced regions by chronic inflaion as defined by Pazos (972). Governmens across he region have underaken several sabilizaion programs in he pas in order o sop inflaion. Mos of he imes, hese programs used he exchange rae as a nominal anchor and no he money supply. Programs based on he money supply usually induced he expeced Phillips curve oucome of a recession a he beginning, followed by a laer economic boom, as shown by Calvo and Végh (994) and Fischer (988). On he oher hand, he empirical lieraure on exchange rae-based sabilizaion programs (henceforh ERBS) shows ha acually he opposie is rue. Specifically, hese plans are characerized by a differen se of sylized facs: iniial expansions in oupu and consumpion, appreciaion of real exchange raes, For an overview of he mos imporan sabilizaion episodes in Lain America, see Agenor and Moniel (999), chaper.

15 2 convergence of inflaion rae o he rae of devaluaion, curren accoun deficis and laer conracions of oupu (Kiguel and Liviaan 992, Végh 992, Calvo and Végh 999). This difference in he iming profile of he economic cycle (iniial recessions under money-based plans versus iniial booms under exchange rae based plans) has been named he recession now-recession laer hypohesis. Several heoreical models have been developed ha ry o accoun for he sylized facs associaed wih ERBS programs. They usually reduce o models wih inflaion ineria where inflaion expecaions are backward looking, like in Rodriguez (982) or models wih lack of credibiliy, where inflaion expecaions may be forward-looking bu governmen policies are no fully credible, as in Calvo and Végh (994). The empirical lieraure on ERBS programs suppors he noion ha his programs end o be associaed wih iniial expansions. I is no always easy and sraighforward o idenify sabilizaion aemps, no o menion he daing of he programs. There are wo differen approaches for he idenificaion of sabilizaion episodes. The episodic approach involves selecing only he bes-known sabilizaions episodes. This is he approach followed by for example Reinhar and Végh (994). The disadvanage of his approach is ha since i is subjecive, i may omi lesser-known episodes. A second alernaive, is o follow a mechanical approach, where sabilizaions are defined based on some objecive crieria, for example, inflaion behavior. This is he approach aken by Easerly (996). He defines as sabilizaion programs hose where a counry is going from 2 years or more of above 4% inflaion o 2 years or more of below 4% inflaion. He hen finds ha sabilizaions from chronic inflaion are ypically expansionary. Oupu

16 3 and per capia oupu increase in he firs years afer he launch of he sabilizaion program. 2 However, i should be noiced ha his approach idenifies programs based on heir oucome, so i will invariably bias he sample owards successful programs. Only hose counries ha succeeded in bringing down inflaion will be idenified. Therefore, we may be overesimaing he expansionary effecs of sabilizaion. Reinhar and Végh (994) use panel daa for seven Lain American counries from 964 o 993. They find ha ERBS programs are associaed wih an economic boom in he early phases of he programs, and wih a recession in he laer phases. However, one should be careful when concluding ha ERBS programs are indeed expansionary, ha is, ha hey cause an expansion. As was argued by Echenique and Foreza (2), he poliical decision of launching a sabilizaion program may be correlaed wih inernaional condiions. The empirical evidence in heir work cass doubs on he sylized fac ha he expansion now recession hen hypohesis is caused by he sabilizaion plan. They argue ha once variables represening exernal condiions are included in he regressions, he resuls are no longer robus. Calvo, Leiderman and Reinhar (993 and 996) also argue ha inernaional condiions may be criical in deermining capial inflows and hus limi he se of policies ha a counry can use in order o sabilize capial flows. This may sugges ha exchange rae programs are an endogenous response of governmens o exernal condiions. Calvo and Végh (999) run panel regressions on several dummies inended o capure he differen sages of he programs. They use a 2 Easerly (996) resuls are common o boh, ERBS and MBS programs. Furhermore, no evidence of laer recessions is found in his work.

17 4 smaller sample of ERBS programs and hey find ha even afer conrolling for exernal condiions, ERBS are expansionary a he beginning. In he presen paper, we ry o deermine if using he exchange rae as an anchor is unequivocally expansionary during he early sages. We ry o improve upon he previous work along wo dimensions. Firsly, we avoid using a pure episodic or a pure mechanical approach in deecing sabilizaion episodes, wih he associaed shorcomings of each mehod, by insead comparing inflaion raes wih acual exchange rae regimes for a large se of emerging economies. In order o do his, i is crucially imporan o know he acual, rue exchange rae regime for each counry a each poin of ime. We will use counry exchange rae arrangemens chronologies from Reinhar and Rogoff (24) for his purpose. Reinhar and Rogoff (24) develop a novel sysem of reclassifying hisorical exchange rae regimes. By using monhly daa on markedeermined parallel exchange raes, hey are able o pin down he acual, rue exchange rae policy ha were in place for almos any counry going back o 946. Secondly, we ry o improve he economeric mehodology by using Arellano-Bond esimaors. When working wih dynamic panel daa models, as in our case, addiional complicaion arise in esimaion of such a model. For example, in fixed and random effecs seings, he difficuly is ha he lagged dependen variable is correlaed wih he error erm. Arellano and Bond (99) provide a more convenien way for looking a efficien insrumens and using addiional informaion. The paper proceeds as follows. In secion.2, we discuss he mehodology used for sabilizaion episodes idenificaion. In secion.3, we describe he daa used and

18 5 provide some summary saisics. In secion.4, we esimae a probi model ha explains he probabiliy of launching a sabilizaion program. In secion.5, we analyze he empirical evidence of he effecs of sabilizaion programs on he business cycle. Secion.6 provides some robusness checks. Secion.7 concludes he paper..2 Idenificaion of he Sabilizaion Programs A criical mehodological issue when sudying sabilizaion programs involves idenifying he programs in he firs place. For his purpose, wo alernaive approaches have been used in he exising lieraure. Following Calvo and Végh (999) erminology, one approach may be named he episodic approach, which consis in working wih sabilizaion episodes ha have received a good deal of aenion. This is he approach followed by Reinhar and Végh (994), Calvo and Végh (999), and Echenique and Foreza (2) among ohers. As an example, hese hree papers use a subjecive lis of sabilizaion programs firs defined in Reinhar and Végh (994). 3 A poenial problem wih his approach is ha since i is subjecive i may omi smaller less-known episodes. The mechanism proposed in his paper is able o deec several addiional ERBS programs in Lain America, hus expanding he se of programs ha can be used o sudy empirical relaions beween sabilizaions and macroeconomic performance. 3 The original Reinhar and Végh (994) lis includes programs in 5 counries. Echenique and Foreza (2) add he Brazilian Plan Real o ha lis, whereas Calvo and Végh (999) work wih a subse of ha lis plus one program from Israel.

19 6 An alernaive approach migh be wha hey call he mechanical approach, were sabilizaions programs are defined according o an explici rule. A good example of his approach is Easerly (996), who defines a sabilizaion program as an episode characerized by a swich from a leas wo consecuive years wih inflaion raes above 4% followed by a period of a leas wo consecuive years wih inflaion raes below 4%. However, as was poined ou by Echenique and Foreza (2), his approach is problemaic since i defines a sabilizaion plan by is oucome, he inflaion rae, and hus possibly omiing programs ha failed and biasing he resuls owards successful programs. Given ha he previous wo approaches involve some disadvanages, we propose a new mehod here. The approach we will follow uses exensive counry chronologies from Reinhar and Rogoff (24), so i is worhwhile o review he main aspecs of heir work. The auhors develop a major reinerpreaion of modern hisory exchange rae arrangemens. Unil heir work, mos of he empirical lieraure dealing wih exchange raes used he official classificaion published in he IMF s Annual Repor on Exchange Rae Arrangemens and Exchange Resricions. The problem wih such classificaion is ha unil very recenly he IMF jus asked member saes o self-declare heir arrangemen ino one of four caegories ye many counries deviae from heir announced exchange rae regime. 4 To sor ou his problem, Reinhar and Rogoff (24) employ an exensive daabase on marke-deermined parallel exchange raes and hen use an 4 Calvo and Reinhar (22) and Alesina and Wagner (26) provide some explanaions why counries do no follow heir de jure exchange rae regimes.

20 7 algorihm o obain wha hey call a naural classificaion. The algorihm employed, relies mainly on he absolue percen change in he exchange rae on a monhly basis over wo and five year rolling windows, and hen soring hem across several exchange rae arrangemens. The differences from Reinhar & Rogoff s naural classificaions and ha of he IMF are no minor. As an example, hey find ha when he official classificaion is a form of a peg, almos half of he ime heir algorihm reveals he rue underlying moneary policy o be a varian of a floa. Conversely, when he official classificaion is a form of a floa, heir naural classificaion reveals ha he underlying policy was somehing closer o a de faco peg. They develop from his a deailed exchange rae chronology for 53 counries ha goes back o 946. Their algorihm allows for foureen differen caegories of exchange rae regimes, ranging from sric pegs and no separae legal ender o hyper-floas and freely falling. Table. liss all caegories in heir classificaion.

21 8 Table.: Reinhar & Rogoff Classificaion Scheme umber assigned o caegory in: aural classificaion bucke Fine grid Coarse grid o separae legal ender Preannounced peg or currency board arrangemen 2 Preannounced horizonal band narrower han or equal o ±2% 3 De faco peg 4 Preannounced crawling peg 5 2 Preannounced crawling band narrower han or equal o ±2% 6 2 De faco crawling peg 7 2 De faco crawling band ha is narrower han or equal o ±2% 8 2 Preannounced crawling band ha is wider han ±2% 9 2 De faco crawling band ha is narrower han or equal o ±5% 3 oncrawling band ha is narrower han or equal o ±2% 3 Managed floaing 2 3 Freely floaing 3 4 Freely falling (includes hyperfloa) 4 5 Source: Reinhar & Rogoff (24)

22 9 Using he counry classificaions in Reinhar and Rogoff (24), we proceed o idenify exchange rae-based sabilizaion episodes in he following manner. Firs, we divide he 4 exchange rae caegories ino wo groups: fixed (caegories -9) and floaing (caegories -4). The fixed caegory groups all R&R caegories ha involve some sor of exchange rae policy where he exchange rae is being used as an anchor. The floaing caegory groups all oher R&R caegories, namely, hose regimes where he exchange rae was largely le free. We hen look for all cases in which a counry moved from a floaing regime o a fixed regime, for all counries in he daase. Finally, we idenify ha change as an ERBS program, if he change in exchange rae policy was preceded by a leas wo years of high inflaion. We define as a high inflaion period, any year in which he inflaion rae was a leas 2 percen. 5 The reason for following he jus described procedure is ha i allows us o idenify no only well-known episodes, bu also smaller episodes ha may have no been publicly announced by he governmen. However, he change in exchange rae policy coupled wih he fac ha inflaion was above 2 percen for a leas he previous wo years, sugges ha policymakers were indeed using he exchange rae o sabilize inflaion. Using his procedure on emerging counries for he period we deec 34 ERBS programs. Mos of hese episodes ook place in Lain American counries. A 5 2 percen is our baseline cuoff for defining high inflaion periods. In secion 8, we perform some robusness checks by using wo differen cuoff levels, and 4 percen.

23 complee lis wih all exchange rae-based sabilizaion programs ha ook place over he period is presened in he appendix (Table A)..3 Daa Descripion and Summary Saisics The daase for exchange rae classificaions is from Reinhar and Rogoff (24). This daase classifies exchange raes arrangemens for 53 counries for he period The measure of inflaion we use is CPI percenage change from he Inernaional Financial Saisics (IFS) of he IMF. Daa on oupu, per capia oupu, consumpion, invesmen, rade balance, foreign direc invesmen and foreign reserves are from he IFS and World Developmen Indicaors (WDI) of he World Bank. Terms of Trade daa for Lain American counries is aken from he Unied aions Economic Commission for Lain America and he Caribbean (CEPAL). For he remaining counries, is aken from he IFS. We prefer he CEPAL daa for Lain American counries, since his daabase cover a larger period. Daa on several US ineres raes as well as on he SP5 sock marke index is from Global Insigh. Exchange rae-based sabilizaion programs are associaed wih oupu expansions and decreasing raes of inflaion, a leas iniially. Figures. and.2 make his poin clear. Figure.2 shows he median and mean inflaion raes in he sample before and afer an ERBS program began. The median inflaion rae declines from a peak of 74% wo years before a program sared, o a low of abou 9% by year 4. The mean inflaion

24 goes from almos,% o 29%. 6 However, by year 4, mean inflaion was again on is way up. Figure.: Annual inflaion before and afer an ERBS plan was sared. Mean Median Annual Inflaion.. =-5 =-4 =-3 =-2 =- = =2 =3 =4 =5 Year oe: Year is he year in which an ERBS program sared. Figure.2 shows he median and mean real GDP growh raes before and afer an ERBS program sars. GDP growh is around 2% during he earlier years before a program begins, and sharply increases o beween around 5% for he firs wo years of he ERBS program. By he fourh year, real GDP growh is again around is pre-program levels, suggesing ha he programs where only successful in he shor erm. 6 Mean raes are higher because he disribuion of inflaion has a well-known skewness o he righ.

25 2 Figure.2: Real GDP growh before and afer an ERBS plan was sared Mean Median 4. GDP growh =-5 =-4 =-3 =-2 =- = =2 =3 =4 =5 Year oe: Year is he year in which an ERBS program sared. The previous wo graphs seem o confirm he idea ha exchange rae-based sabilizaions are in fac expansionary in heir early sages. However, correlaion beween ERBS programs and early oupu expansions does no necessarily mean ha he sabilizaion programs caused he expansions. For example, oher variables may explain boh, he oupu performance and he ERBS programs. One simple way o check his is o see if oupu growh in counries under a sabilizaion plan was no only high, bu also high relaive o world oupu performance. Leing g i denoe real oupu growh

26 3 in counry i in year and y denoe world growh in ha year, hen he following measures oupu growh in each counry relaive o world oupu growh. g d i g y (.) i Figure.3 plos he average and median oupu growh relaive o world growh. Looking a oupu growh relaive o he world growh is relevan since he causaliy of sabilizaion programs on oupu would be less likely if a counry is booming a he same ime ha mos of he res of he world is. Growh is below world growh on each of he previous five years prior o an ERBS program. Once an ERBS sars, growh is above world growh and remains ha way for he firs four years afer he ERBS sared. 7 7 Median growh is above world growh during he firs four years afer he sar of an ERBS program. Mean growh is above world growh only he firs hree years.

27 4 Figure.3: Real GDP growh before and afer an ERBS plan sared relaive o world growh Mean Median GDP growh. -. =-5 =-4 =-3 =-2 =- = =2 =3 =4 = Year oe: Year is he year in which an ERBS program sared. In Tables.2 and.3 we es he significance of his before and afer effecs for a series of macroeconomic variables. We calculae sample averages for each of he four years preceding and ERBS and for he four years following he sar of an ERBS program. Mos of he expansionary effecs of sabilizaion occur in he firs wo years. If we look a GDP growh, per capia oupu growh and differenial growh (domesic growh above world growh), we observe always he same picure: very poor performance during he years preceding an ERBS sabilizaion program, followed by a

28 5 period of wo years of excepionally good performance and laer a reurn o more modes numbers. For example, oupu growh during hese wo years is 4.8 and 5.5 percen, respecively, above o wha i was in he pre-erbs years or o wha i will be in years 3 and 4. Per capia oupu growh presen a similar picure, wih growh raes above 3 percen in years and 2, well above o preceding years. In addiion, growh above inernaional growh is only significanly posiive in years and 2. I is also of ineres o see if he effecs of sabilizaion on consumpion and invesmen are similar o ha on oupu growh. In boh cases, he higher growh raes are observed in years and 2. In paricular, invesmen growh is excepionally high on hose ow years (4.9 and 3.2 percen respecively) as opposed o any of he previous four years of an ERBS, in which i was no significanly differen from zero. The rade balance and curren accoun balance, boh measured as a percenage of GDP, also behave as expeced. They go from small deficis o more pronounced deficis in years 2, 3 and 4 as absorpion increases a a faser rae han oupu.

29 6 Table.2: Average saisics before and afer sabilizaions GDP Per capia Growh Consumpion Invesmen growh growh differenial Sabilizaion x% x% x% x% x% years oes: x% are sample averages in percen, are heeroskedasiciy-robus -saisics. Growh differenial is counry GDP growh minus world oupu growh. Invesmen is measured as gross fixed capial formaion growh.

30 7 Table.3: Average saisics before and afer sabilizaions Trade balance Curren accoun Fiscal balance FDI Sabilizaion years x% x% x% x% oes: x% are sample averages in percen, are heeroskedasiciy-robus -saisics. Trade balance, curren accoun, fiscal balance and FDI are all percenages of GDP..4 Likelihood of Sabilizaion Plans The previous secion presened some basic saisics ha seem o suppor he hypohesis ha exchange rae-based sabilizaion plans are expansionary, a leas during he early sages. This simple evidence is, however, far from conclusive. One should ake ino accoun ha he poliical decision o sar a sabilizaion plan is no enirely exogenous. For example, Echenique and Foreza (2) argue ha governmens are more

31 8 likely o launch a plan when inernaional condiions are favorable. Saring a sabilizaion plan when he inernaional economic environmen is good migh reduce he coss associaed wih sabilizaion. Their work provides evidence ha favorable inernaional condiions, and no he sabilizaion plans, are he main reason of he iniial boom observed during sabilizaion. The purpose of his secion is o run a probi model similar o heirs. Echenique and Foreza (2) run a probi model using panel daa for five Lain American counries ha implemened ERBS programs during he period The dependen variable is a dummy variable ha akes he value one if and ERBS program sared in he counry in ha year, and zero oherwise. The independen variables hey use are differen domesic and inernaional variables ha may explain he iniiaion of an ERBS program. We run a similar regression wih wo modificaions. Firsly, we updae he daa se in order o include recen years. Thus, our sample period is This will allow us o include in our esimaion he crises ha several Souh-American counries suffered a he beginning of he presen decade. For example, real GDP declined by roughly percen in Argenina and Uruguay in 22. These recessions were preceded by abandonmen of he sabilizaion plans ha were iniiaed in boh counries in he early 9 s. In December 2, Argenina ended he Converibiliy Plan ha was inroduced in 99, and he exchange rae was depreciaed by almos 2 percen in he following monhs. In he case of Uruguay, he counry was under a sabilizaion plan based on crawling bands inroduced in December 99. On January 22, he band was widened and he pace of depreciaion was acceleraed. By June 22,

32 9 he sysem of bands was abandoned alogeher and he exchange rae was allowed o floa. Secondly, we updae he lis of sabilizaion programs using he Reinhar-Rogoff (24) algorihm, as was discussed in he previous secion. Afer his, we end up wih a bigger lis of ERBS programs across 23 differen emerging economies. 8 Table A2 in he Appendix, liss he sabilizaion programs along wih heir daes. In order o sudy he deerminans of he decision o sar a sabilizaion program, firs we creae a binary variable for ERBS plans: E equals when a ERBS plan is sared in a counry in year and zero oherwise. As in Echenique and Foreza (2), we assume ha he probabiliy of saring an inflaion sabilizaion plan is a funcion of domesic variables,, and exernal variables, Z, ha affec he decision of he policymakers. We hen esimae a probi regression of he likelihood of saring an exchange rae-based sabilizaion plan. The specificaions esimaed are of he form: P( E, Z) G( b dz) (.2) where ERBS is a binary variable ha akes a value equal o if counry i sared an exchange rae-based sabilizaion plan in year and oherwise. is an array of domesic explanaory variables ha influence he decision of saring a sabilizaion plan. This may conain conemporaneous as well as lagged variables. Z is an array of inernaional 8 The number of programs ha we deec depends upon he cuoff level of inflaion. Wih a 4% cuoff, we deec 25 sabilizaion episodes, wih a 2% cuoff we deec 34 episodes, and wih a % cuoff we deec 4 episodes.

33 2 variables ha influence he decision o sar a plan. I may also conain conemporaneous as well as lagged variables. The funcion G is he sandard normal cumulaive disribuion funcion (cdf) for he probi model. Table.4 shows he resuls using panel daa for he counries ha implemened ERBS. The dependen variable is a dummy variable ha akes he value one if and ERBS program sared in he counry in ha year, and zero oherwise. The domesic explanaory variables of he repored regression are he log of inflaion, he fiscal balance as share of oupu, he log of he inernaional reserves o GDP raio, he rade balance and he erms of rade, all lagged one year. The inernaional explanaory variables are he world growh rae of GDP, he percen increase in he S&P 5 reurn index (SP5) and he lagged Federal Funds rae. 9 In boh models, he coefficien for he lagged inflaion rae is saisically significan a one percen and exhibis he correc sign. Sabilizaion programs are more likely o be implemened when he inflaion rae in he previous year was higher. The coefficien for he fiscal balance is posiive and significan. The coefficien for he inernaional reserves o GDP raio is also significan and shows a posiive sign. Governmens prefer o launch sabilizaion plans when hey have more reserves and are running fiscal surpluses (or smaller deficis). The rade balance and erms of rade were no saisically significan in eiher model. As for he exernal variables, only he world oupu growh, wih a posiive esimae, proves o be saisically significan. The likelihood ha a counry will launch a sabilizaion plan is higher if he world economy is 9 Several inernaional ineres raes were used in he regressions yielding similar resuls.

34 2 growing a a faser rae, which ranslaes ino a favorable inernaional environmen. The remaining wo variables, inernaional ineres raes and he SP5 reurn, are no saisically significan. Our resuls are hus no conclusive as o wheher ERBS programs are sared when inernaional condiions are favorable. The domesic inflaion environmen seems o be he deerminan condiion for saring a sabilizaion program. Inernaional condiions play a role, a leas in our esimaed regressions, only o he exen ha high growh in he world economy creaes a favorable environmen in which he decision o sar a plan may involve less domesic coss..5 Empirical evidence on he real effecs of sabilizaion In he pas, i was hough ha sopping inflaion was cosly, in he sense ha i would bring sharp declines in oupu growh. Okun (978) esimae for he US economy he sacrifice raio of disinflaion, ha is, he percen oupu loss per percenage poin reducion in inflaion. Even for he small open economy, his view was hough o apply, as was argued by Fischer (986). Disinflaion is expeced o cause a recession, regardless of he nominal anchor used. This view is very much relaed o he Phillips-curve lieraure on he oupu-inflaion rade-off. However, he convenional wisdom was challenged by he sabilizaion episodes in he souhern-cone of Lain America of he sevenies. These programs used he

35 22 exchange rae as an anchor in order o sop inflaion, and conradicing he iniial view, hey were associaed wih srong iniial expansions in consumpion and oupu, despie he real appreciaion of he exchange rae. Several sudies have documened his empirical regulariy, wih Kiguel and Liviaan (992), Reinhar and Végh (994) and Calvo and Végh (999) among hem. For example, Reinhar and Végh (994) perform regression analysis using panel daa on seven counries ha implemened sabilizaion programs during In heir regressions, he growh rae of real GDP is he dependen variable, and several dummies designed o capure differen sages of he sabilizaion programs are he independen variables. They find ha he early sages of sabilizaion are associaed wih oupu expansions while he lae sages are associaed wih recessions. This is he ypical early expansion lae recession hypohesis ha has been ascribed wih exchange raebased sabilizaion..5. Empirical Mehodology and Resuls To es he effecs of exchange rae-based inflaion sabilizaion, we esimae panel regressions of he effecs of ERBS programs on oupu, per capia oupu, consumpion and invesmen growh as follows: y i, E 2E2 3E3 4E4 ' b Z' d y (.3) i, i i

36 23 where y i represens he dependen variable of ineres (oupu, per capia oupu consumpion, invesmen) for counry i in year. The variables E, E2, E3 and E4 are dummy variables mean o capure he differen sages of he sabilizaion period. E is a variable ha akes he value of if counry i is in year in he firs year of an ERBS program. E2 akes value of if he counry is in is second year of an ERBS program, and E3 and E4 are defined analogously. is a se of domesic conrol variables and Z a se of exern i are counry fixed effecs o accoun for any differences by counry. I is imporan o noice ha in equaion.3 we include as a righ hand side variable he lagged dependen variable, y i,-. Adding dynamics in his fashion is a major change in he model. In his case, he usual approach o esimaing a fixed-effecs model, he leas squares dummy variable esimaor (LSDV), generaes a biased esimae of he coefficiens, since now we have in he equaion, he enire hisory of he righ-hand side variables. To solve his problem, we use he Arellano-Bond dynamic panel esimaor, proposed by Arellano and Bond (99). This echnique relies on a GMM procedure for looking a efficien insrumenal variables for dynamic panel daa models. The resuls of he panel regressions are repored in Tables Columns -2 in hese ables are fixed and random effecs esimaors using only he dummy variables for he firs 4 years of sabilizaion. These are similar o he regressions repored on Reinhar and Végh (994) and he resuls yield similar conclusion. For oupu growh, per-capia oupu growh, consumpion and invesmen growh, he firs year of ERBS plans are

37 24 associaed wih expansions in hese variables. For he rade balance o oupu raio, he resuls are significan for he second year, wih a drop in he rade balance, as would be expeced from an economic expansion. Since our ineres is in invesigaing wheher his expansionary effec is a resul of sabilizaion or exernal condiions, we include a se of exernal conrol variables. The chosen variables are world GDP growh, he SP5 sock marke reurn, he yield on yr US Treasury bonds and he erms of rade for each counry. Columns 3 and 4 in Tables 5-9 show he resuls when inernaional variables are included. The coefficien on world growh is posiive and significan a he percen level in all specificaions wih he excepion of he equaion for invesmen growh, for which is significan only a he percen level. This is he expeced resul, implying ha increases in he domesic variables are more likely when he world oupu is also increasing. The coefficien on he US Treasury yield is negaive and significan a he percen level in all specificaions, wih he excepion for he rade-balance o oupu equaion. This resul is also as expeced, since lower inernaional ineres raes are favorable for emerging economies. The esimaes on he SP5 reurn index and he erms of rade are no saisically significan. The ineresing resul is ha, even afer including exernal conrol variables, he esimaes on he firs year dummy, E, are sill posiive and significan. Columns 5-7 of Tables.5-.9 summarize he resuls when he lagged dependen variable is included as a regressor, which should improve he efficiency of he esimaion. Columns 5 and 6 are he fixed and random effecs, respecively, and column 7 is he Several differen inernaional ineres raes were used yielding similar resuls.

38 25 Arellano-Bond esimaes. The inclusion of he lagged dependen variable in he fixed and random effecs models do no significanly aler he resuls. The coefficiens on he firs year dummy, E, remain posiive and saisically significan. As an example, oupu growh is 2.2 percen higher during he firs year of sabilizaion, consumpion growh is 2.7 percen higher, and invesmen growh is.9 percen higher. The Arellano-Bond esimaes also provide suppor ha sabilizaion cause increases in oupu, consumpion and invesmen, during he firs year of he programs..6 Robusness checks To check he robusness of our resuls, we esimae he panel regression wih alernaive definiions of ERBS programs. The resuls of he previous secions are derived from ERBS programs ha were defined using a 2 percen cu-off inflaion rae. We esimae he models using differen cu-offs of he inflaion rae. In paricular, we use a percen cuoff (which increases he number of idenified programs from 34 o 4) and a 4 percen cuoff (which reduces he number of idenified programs from 34 o 25). The resuls are summarized in Tables. and.. Table. shows he resuls for he 25 ERBS plans deeced using a 4 percen inflaion cu-off. The poin esimaes on he firs year dummies remain posiive and saisically significan a he one percen These are he resuls for he fixed effecs equaions. Fixed effecs esimaes are always consisen bu random effecs may be more efficien. However, he Hausman es rejecs he null hypohesis ha he coefficiens are he same under boh mehods for almos all specificaions, so random effecs are no consisen.

39 26 level for he equaions for oupu, consumpion and invesmen growh. In some cases, he posiive effec of sabilizaion is also seen in laer years. For example, using Arellano- Bond esimaors, consumpion is significanly higher during he firs four years of sabilizaion. Table. shows idenical resuls, when he crieria for deecing ERBS programs is a percen inflaion rae, which resuls in 4 programs being idenified. Again, he poin esimaes on he firs year dummies are posiive and saisically significan a he one percen level for all specificaions..7 Conclusions The empirical evidence presened in his paper, canno rejec he noion ha exchange rae based sabilizaion programs are expansionary during he early phases. Using a new lis of sabilizaion programs for emerging economies and daa from recen years, we conclude ha sabilizaion programs ha use he exchange rae as a nominal anchor, lead o iniial real GDP expansions, even when accoun is aken of inernaional condiions. The expansionary effecs are no exclusive of oupu. Real consumpion and invesmen also end o be higher during he early sages of sabilizaion, whereas he rade balance o oupu ends o be lower, as absorpion increases wih he economic boom. We derive a new lis of sabilizaion programs using he Reinhar-Rogoff s (22) exensive counry chronologies of exchange rae arrangemens. Unil very recenly, mos

40 27 of he empirical lieraure dealing wih exchange raes used he official classificaion published in he IMF s Annual Repor on Exchange Rae Arrangemens and Exchange Resricions. Ye, a closer look a he experience suggess ha hese official classificaions ofen fail o describe acual counry moneary policy. Afer using he Reinhar-Rogoff new counry chronologies, we deec new exchange rae based sabilizaion programs ha were omied on previous work. Our probi model, alhough no conclusive, sugges ha he saring of a sabilizaion program is more likely when he region faces favorable inernaional condiions. However, his does no imply ha he effecs of sabilizaion programs should be dismissed. In paricular, our panel regressions provide some evidence ha he iniial phases of an ERBS program, leads o an iniial boom in oupu, even afer conrolling for exernal variables. I is ineresing o noice ha one would also like o conrol for oher effecs, such as domesic reforms. Many of he sabilizaion programs sudied in his paper, have been accompanied wih rade and financial liberalizaion and srucural reforms. I remains an issue for fuure research o ake accoun of his, maybe by consrucing a liberalizaion index along he lines of work by De Melo, Denizer and Gelb (996) for ransiion economies.

41 28 Table.4: Probi Model. Dependen variable is E (dummy for firs year of ERBS) Explanaory Variables Model Model 2 Log Inflaion i,-.49***.782*** (4.88) (4.95) Reserves i,-.3*.62* (.68) (.74) Fiscal Balance i,-.7**.45* (2.2) (.89) Trade Balance i, (.37) (.62) Terms of Trade i, (-.55) (-.39) World Growh.276**.578** (2.32) (2.37) S&P5.9.2 (.97) (.5) Federal Funds (-.) (-.) Consan (-4.5) (-4.27) umber of obs McFadden R-sq.2.2 oe: Significance a he, 5 and percen level is indicaed by one, wo and hree sars, respecively. For Model an inflaion rae of 2% was used as a cu-off for idenifying he ERBS plans. For Model 2 an inflaion rae of 4% was used.

42 29 Esimaion Mehod Table.5: Panel regression. Dependen variable: Real GDP growh Fixed effecs Random effecs Fixed effecs Random effecs Fixed effecs Random effecs Arellano- Bond E.463*.392*.743**.586** 2.2***.979*** 3.43*** (.79) (.7) (2.3) (.95) (2.79) (2.54) (3.38) E * ** (.23) (.5) (.6) (.44) (.9) (.63) (2.4) E ** (.2) (.99) (.45) (.28) (.4) (.78) (2.2) E (.6) (.) (.63) (.47) (.38) (.6) (.34) World GDP.622***.68***.566***.538***.53*** (5.77) (5.83) (5.43) (5.33) (3.94) SP (-.6) (-.6) (-.92) (-.76) (-.7) YRBOD -.3*** -.29*** -.23*** -.2*** -.258** (-4.85) (-4.75) (-3.66) (-3.42) (-.9) TOT (.56) (.36) (.6) (.22) (.24) GPD -.286***.327***.262*** (8.78) (.3) (6.9) Consan 3.626*** 3.66*** 3.379*** 3.458*** 2.54*** 2.7**.6 (24.25) (3.84) (4.64) (4.78) (3.) (3.8) (.2) Observaions Adj R-sq Hausman Tes Chi (4) =.36 Chi (8) = 3.56 Chi (9) = P-value =.85 P-value =.89 P-value =. oes: -saisics in parenheses. One, wo and hree sars denoe significance a, 5 and percen respecively. Sabilizaion plans are idenified using a cuoff 2% inflaion rae. Sample period: 96 o 25.

43 3 Esimaion Mehod Table.6: Panel regression. Dependen variable: Per-capia GDP Fixed effecs Random effecs Fixed effecs Random effecs Fixed effecs Random effecs Arellano- Bond E.599**.587**.875**.84** 2.258*** 2.9*** 3.42*** (2.2) (2.2) (2.36) (2.3) (2.94) (2.88) (3.47) E *.445* ** (.4) (.37) (.78) (.67) (.27) (.9) (2.9) E ** (.23) (.23) (.63) (.53) (.3) (.) (2.36) E (.38) (.34) (.82) (.72) (.53) (.26) (.39) World GDP.5***.485***.482***.435***.542*** (4.88) (4.72) (4.77) (4.4) (4.6) SP (-.2) (-.5) (-.9) (-.75) (-.7) YRBOD -.32*** -.32*** -.25*** -.24*** -.28 (-5.2) (-5.3) (-4.6) (-4.) (-.43) TOT (-.4) (-.26) (.49) (-.27) (.28) PGDP -.264***.3***.244*** (8.3) (9.67) (5.69) Consan.568***.56*** 2.222*** 2.33***.434**.353**.28 (.83) (6.72) (3.4) (3.28) (2.7) (2.4) (.95) Observaions Adj R-sq Hausman Tes Chi (4) =.5 Chi (8) = 2.49 Chi (9) = 39.3 P-value =.97 P-value =. P-value =. oes: -saisics in parenheses. One, wo and hree sars denoe significance a, 5 and percen respecively. Sabilizaion plans are idenified using a cuoff 2% inflaion rae. Sample period: 96 o 25.

44 3 Esimaion Mehod Table.7: Panel regression. Dependen variable: Consumpion Fixed effecs Random effecs Fixed effecs Random effecs Fixed effecs Random effecs Arellano- Bond E 2.9**.93** 2.554** 2.23* 2.723** 2.93** 3.973*** (.88) (.79) (2.33) (.86) (2.52) (2.5) (3.9) E ** 2.386** 2.596** *** (.57) (.5) (2.42) (.99) (2.4) (.56) (3.38) E ** (.8) (.) (.78) (.38) (.66) (.5) (2.44) E *** (.35) (.28) (.47) (.9) (.5) (.7) (2.7) World GDP.463***.46***.498***.496***.35* (2.97) (3.2) (3.23) (3.32) (.86) SP (-.39) (-.36) (-.56) (-.52) (-.5) YRBOD -.27*** -.24*** -.25*** -.26** -.459** (-3.) (-2.84) (-2.79) (-2.47) (-2.3) TOT.**.6*.9**.4.47*** (2.37) (.68) (2.2) (.2) (3.57) CO -.3***.36***.7* (2.84) (3.82) (.6) Consan 3.734*** 3.79*** 2.562** 3.96*** 2.83** 2.478*** -.29 (7.93) (2.) (2.53) (3.6) (2.7) (2.56) (-.68) Observaions Adj R-sq Hausman Tes Chi (4) =.79 Chi (8) = 6.58 Chi (9) = P-value =.94 P-value =.3 P-value =. oes: -saisics in parenheses. One, wo and hree sars denoe significance a, 5 and percen respecively. Sabilizaion plans are idenified using a cuoff 2% inflaion rae. Sample period: 96 o 25.

45 32 Esimaion Mehod Table.8: Panel regression. Dependen variable: Invesmen Fixed effecs Random effecs Fixed effecs Random effecs Fixed effecs Random effecs Arellano- Bond E 9.467*** 8.279***.48*** 9.42***.953*** 9.93*** 3.778*** (2.92) (2.6) (3.33) (3.6) (3.52) (3.27) (3.67) E * * 5.86* ** (.66) (.38) (.94) (.73) (.62) (.36) (2.5) E (.67) (.37) (.87) (.66) (.49) (.27) (.48) E ** (.2) (.75) (.3) (.9) (.26) (.4) (2.8) World GDP.688*.736*.747*.88**.592 (.63) (.8) (.78) (2.) (.8) SP (.2) (.3) (.3) (.2) (-.44) YRBOD -.839*** -.79*** -.773*** -.72*** -.94 (-3.39) (-3.33) (-3.) (-3.) (-.5) TOT ** (.8) (.44) (.74) (.29) (2.3) IV -.62***.69***.2*** (4.49) (4.75) (.6) Consan 4.686*** 4.798*** 6.39** 6.829** 5.45* 5.548**.49 (7.97) (8.24) (2.29) (2.57) (.85) (2.9) (.) Observaions Adj R-sq Hausman Tes Chi (4) = 4. Chi (8) = 2.79 Chi (9) = 3.39 P-value =.4 P-value =.94 P-value =.94 oes: -saisics in parenheses. One, wo and hree sars denoe significance a, 5 and percen respecively. Sabilizaion plans are idenified using a cuoff 2% inflaion rae. Sample period: 96 o 25.

46 33 Esimaion Mehod Table.9: Panel regression. Dependen variable: Trade balance Fixed effecs Random effecs Fixed effecs Random effecs Fixed effecs Random effecs Arellano- Bond E (-.79) (-.72) (-.52) (-.47) (-.25) (-.85) (-.68) E2-3.*** -2.99*** -3.8*** -3.2*** -2.64*** -2.** -2.34*** (-2.79) (-2.74) (-2.86) (-2.8) (-3.9) (-2.45) (-2.69) E (-.59) (-.54) (-.44) (-.38) (.3) (.28) (-.33) E (-.53) (-.49) (-.35) (-.3) (-.6) (.4) (-.3) World GDP.63***.642***.333***.253**.45*** (4.66) (4.72) (3.28) (2.46) (3.95) SP (.5) (.5) (.64) (.2) (.5) YRBOD (.28) (.42) (.96) (.94) (.44) TOT *** (.9) (.3) (.6) (.7) (3.3) TBY -.665*** 858***.384*** (25.58) (45.3) (7.93) Consan -3.3*** -3.3*** -6.43*** -7.69*** -2.78*** -2.24***.29 (-8.8) (-3.35) (-7.9) (-5.9) (-4.2) (-3.25) (.3) Observaions Adj R-sq Hausman Tes Chi (4) = Chi (8) = 2.79 Chi (9) = 8.96 P-value =. P-value =.94 P-value =. oes: -saisics in parenheses. One, wo and hree sars denoe significance a, 5 and percen respecively. Sabilizaion plans are idenified using a cuoff 2% inflaion rae. Sample period: 96 o 25.

47 34 Dependen Var Esimaion Mehod Table.: Panel regression. Alernaive ERBS idenificaion GDP growh Consumpion Invesmen Fixed effecs Bond Fixed effecs Bond Fixed effecs Arellano- Arellano- Arellano- Bond E 2.669*** 4.4*** 3.869*** 5.34***.869*** 3.275*** (2.95) (3.77) (3.9) (3.69) (3.6) (3.7) E ** 2.473* 5.242*** (.93) (2.2) (.72) (3.9) (.35) (.58) E * ** (.26) (.85) (.39) (2.43) (.6) (.98) E *** ** (.6) (.57) (.54) (3.2) (.4) (2.2) World GDP.565***.59***.54***.342*.77*.62 (5.45) (3.92) (3.28) (.83) (.83) (.3) SP * (-.95) (-.22) (-.66) (-.64) (.4) (-.4) YRBOD -.227*** -.282** -.249*** -.487** -.76*** -.73 (-3.62) (-2.9) (-2.79) (-2.44) (-3.4) (-.28) TOT...9**.47***.9.8** Lagged Depvar (.) (.22) (2.3) (3.56) (.67) (2.8).286***.256***.6***.73*.62***.27*** (8.79) (5.95) (2.93) (.67) (4.49) (4.9) Consan 2.96***.3 2.8** *.3 (3.6) (.4) (2.7) (-.74) (.88) (.97) Observaions Adj R-sq oes: -saisics in parenheses. One, wo and hree sars denoe significance a, 5 and percen respecively. Sabilizaion plans are idenified using a cuoff 4% inflaion rae. Sample period: 96 o 25.

48 35 Dependen Var Esimaion Mehod Table.: Panel regression. Alernaive ERBS idenificaion GDP growh Consumpion Invesmen Fixed effecs Bond Fixed effecs Bond Fixed effecs Arellano- Arellano- Arellano- Bond E 2.*** 2.898*** 2.942*** 3.93*** 9.5*** 2.3*** (2.9) (3.38) (2.93) (3.37) (3.34) (3.57) E ** 2.782** 4.373*** 7.573**.57*** (.6) (2.33) (2.5) (3.36) (2.4) (2.72) E ** ** (.28) (2.6) (.9) (2.38) (.7) (.43) E * 2.762** 4.*** * (.32) (.66) (2.25) (2.75) (.35) (.66) World GDP.569***.57***.56***.36*.776*.625 (5.5) (3.97) (3.36) (.92) (.85) (.4) SP (-.94) (-.) (-.6) (-.48) (.9) (-.44) YRBOD -.239*** -.258* -.259*** -.449** -.8*** -.73 (-3.8) (-.9) (-2.9) (-2.25) (-3.2) (-.2) TOT..3.**.48***..89** Lagged Depvar (.28) (.35) (2.2) (3.63) (.8) (2.28).282***.26***.99***.69.59***.28*** (8.68) (6.6) (2.74) (.58) (4.4) (4.73) Consan 2.6***.6.952* *.54 (2.95) (.22) (.95) (-.63) (.79) (.4) Observaions Adj R-sq oes: -saisics in parenheses. One, wo and hree sars denoe significance a, 5 and percen respecively. Sabilizaion plans are idenified using a cuoff % inflaion rae. Sample period: 96 o 25.