Do Real Exchange Rate Appreciations Matter for Growth?* by Matthieu Bussière, Claude Lopez and Cédric Tille

Transcription

1 Do eal Exchange ae Appreciaions Maer for Growh?* by Mahieu Bussière Claude Lopez and Cédric ille o. 9 Mai 04

2 Kiel Insiue for he World Economy Kiellinie Kiel Germany Kiel Working Paper 9 Mai 04 Do eal Exchange ae Appreciaions Maer for Growh?* Mahieu Bussière Claude Lopez and Cédric ille Absrac: While he impac of exchange rae changes on economic growh has long been an issue of key imporance in inernaional macroeconomics i has received renewed aenion in recen years owing o weaker growh raes and he debae on currency wars. However in spie of is prevalence in he policy debae he connecion beween real exchange raes and growh remains an unseled quesion in he academic lieraure. We fill his gap by providing an empirical assessmen based on a broad sample of emerging and advanced economies. We assess he impac of appreciaions produciviy booms and capial inflows surges using a propensiy-score maching approach o address causaliy issues. We show ha appreciaions associaed wih higher produciviy have a larger impac on growh han appreciaions associaed wih capial inflows. Furhermore he appreciaion per se ends o have a negaive impac on growh. We provide a simple heoreical model ha delivers he conrased growh-appreciaion paern depending on he underlying shock. he model also implies adverse effecs of shocks o inernaional capial flows so concerns abou an appreciaion are no inconsisen wih concerns abou a depreciaion. he presence of an exernaliy hrough firms desrucion leads o inefficien allocaions. oneheless addressing hem does no require a dampening of exchange rae movemens. Keywords: exchange rae currency crises endaka inernaional rade inernaional capial flows lending booms small open economy macroeconomics JEL classificaion: F0 F30 F4 Mahieu Bussière Banque de France 3 rue Croix des Peis Champs 7500 Paris; mahieu.bussiere@banque-france.fr Claude Lopez Banque de France 3 rue Croix des Peis Champs 7500 Paris; claude.lopez@banque-france.fr Cédric ille Graduae Insiue of Inernaional and Developmen Sudies P.O. Box 36 Genève Swizerland; cedric.ille@graduaeinsiue.ch * Paper prepared for presenaion a he 59h Panel Meeing of Economic Policy. he views presened in his paper are hose of he auhors and do no necessarily reflec hose of he Banque de France or he Eurosysem. We are graeful o Philippe Bacchea Chrisiane Baumeiser Bruno Cabrillac Menzie Chinn James Hamilon Luz Kilian Céline Misrea ober Vigfusson seminar audiences a he DIW insiue and he European Cenral Bank efe Gürkaynak and wo anonymous referees for helpful suggesions. We also would like o hank Caherine Marin for ousanding research assisance.

3 . Inroducion Policy makers hroughou he world have long been wary of real exchange rae appreciaions owing o heir effec on growh hrough reduced price compeiiveness in inernaional markes. Japanese policy makers for insance have repeaedly expressed heir worries having experienced several episodes of large real exchange rae appreciaions in he early 970s in he wake of he Smihsonian Agreemen from 985 o 995 afer he Plaza Accord and in he curren crisis. he concern is however no universally shared. Paul Krugman 994 has dubbed i a dangerous obsession and argues ha concerns abou compeiiveness are as an empirical maer almos compleely unfounded. his long-sanding quesion has received renewed ineres in recen years wih he debae on currency wars iniiaed by Brazil s Finance Miniser Guido Manega in Sepember 00. Indeed he Grea Financial crisis of has lowered growh raes hroughou he world making policy makers more concerned of he sources of economic growh and simulaneously led cenral banks o underake unprecedened acions whose effec on exchange raes was no always negligible promping concerns over poenial beggar-hy-neighbour effecs. Several counries especially among emerging marke economies expressed concerns regarding surges in capial inflows and he associaed real exchange rae appreciaions hey recorded in he wake of moneary policy decisions in foreign counries. his acive debae raises he quesion wheher appreciaions have indeed an adverse effec on growh a poin ha has no been firmly esablished. his paper akes a sep owards filling he gap by providing empirical evidence and inerpreing i hrough a heoreical framework of he join deerminaion of he real exchange rae and growh. Boh he empirical and heoreical elemens sress how he relaion beween growh and he exchange rae depends on he specific underlying shock. Our empirical assessmen relies on a broad annual sample of 68 counries 30 advanced economies and 38 emerging markes from 960 o 0. We focus on he relaion beween exchange rae appreciaions and oupu conrasing he paern beween appreciaions associaed here is even a special word in Japanese o refer o a period of srong appreciaion endaka. wo quoes illusrae hese differen views. On he one hand Mishkin 007 clearly expresses concern abou he compeiiveness channel An appreciaion of he dollar in urn resrains expors because he price of U.S. goods rises when measured in foreign currencies and simulaes impors because impors become cheaper in dollar erms. he resuling decrease in ne expors implies a reducion in aggregae demand. On he oher hand oyer 007 brings a more balanced view: I is clear ha he price-compeiiveness of French indusries has deerioraed significanly in recen years. Has he euro s appreciaion played a role in his? On he one hand i undoubedly penalises expor secors whose compeiors are locaed in oher moneary areas. Bu on he oher hand i benefis hose secors which are large consumers of impored commodiies. A his sage he overall effec on France s growh and exernal balance is no clearly apparen.

4 wih unusual developmens in he financial secor from hose associaed wih a srong increase in domesic produciviy. A challenge in evaluaing he impac of appreciaion on growh is o disenangle he direcion of causaliy. We handle his issue by relying on he propensiy-score maching mehod. While i is commonly used o assess he impac of a public policy in labor economics is applicaions in inernaional economics are only few and recen. Our analysis makes hree poins. Firs only advanced economies suffer from an appreciaion and his only in he mos general case of increase in real exchange rae. Second boh emerging and advanced economies reac posiively o he combinaion of appreciaion and produciviy increase bu he conribuion of he appreciaion on he enhanced growh rae differs depending on he group of counries considered. Specifically growh is srenghened in advanced economies compared o he case of solely a produciviy shock while i is weakened for he emerging economies. hird appreciaions ha are associaed wih a surge in capial inflows are characerized by weaker growh compared o episodes wih produciviy shock. Furhermore emerging economies are sensiive o an appreciaion in presence of a surge in capial inflows. Ineresingly focusing on large appreciaion episodes does no change hese conclusions. We develop a simple small open economy model in which he paern of growh and appreciaion depends on he underlying shocks consisen wih he empirical resuls. he model is furhermore characerized by asymmeric effecs for some shocks wih a recession under any exchange rae movemen and an exernaliy. Policy makers are hus righ o be concerned abou exchange rae movemens and worrying abou an appreciaion is no necessarily inconsisen wih worrying abou a depreciaion. he key feaure of he model is ha firms face idiosyncraic fixed coss so ha in he iniial seady sae he gross profis of he marginal firm in each secor jus covers is fixed coss. Shocks affec gross profis and we assume ha in he shor run he firms ha canno cover heir fixed cos shu down. In addiion we assume ha no firm can be creaed in he shor run. 3 While sark his assumpion capures he fac ha i akes more ime and resources o esablish a new firm han o shu down an exising one. A shock ha lowers gross profis in he raded secor reduces he number of raded firms wihou any new non-raded firms being creaed. As individual firms use a echnology wih decreasing reurns o scale he lower number of firms leads o lower oupu as in Bacchea and van Wincoop 03. he impac on 3 For breviy we assume ha afer one period he number of firms recovers o is iniial value. he model could be enriched by allowing for an endogenous number for firms in he long run a he cos of addiional complexiy. 3

5 he number of firms leads o an asymmery. For insance an increase in he world ineres rae leads o a real depreciaion lowering profis in he non-raded secor. his lowers he number of firms in ha secor as well as overall oupu. he recession following an ineres rae increase does no imply ha a decrease leads o a boom o he conrary. A lower ineres rae reduces profis and he number of firms in he raded secor. Shocks o ineres raes hus lower oupu regardless of heir direcion. he model also exhibis an exernaliy as under he decenralized allocaion agens do no inernalize he impac of heir decisions on firms profiabiliy. We assume ha he fixed coss faced by firms merely consis of a ransfer o he household and hus do no represen a real cos. A planner is hus no bou by hese coss and chooses o offse hem hrough lump sum subsidies. he planner recognizes he cos of lowering he number of firms and hus keeps i a is iniial level in boh secors hereby prevening inefficien recessions. An ineresing feaure of he model is ha he planner s allocaion does no dampen he movemens in he real exchange rae which are acually larger han under he decenralized allocaion. his indicaes ha he firs bes policy canno simply be implemened by leaning agains he wind. he res of he paper is organised as follows. Secion reviews he relevan lieraure. Secion 3 inroduces he definiion of exchange rae appreciaions and oher benchmarks he propensiy maching mehod and is resuls. Secion 4 presens he heoreical model. Secion 5 concludes and discusses possible policy implicaions.. eview of he lieraure Mos of he exising sudies on he link beween growh and he exchange rae predominanly focus on episodes of weakening currencies more specifically sharp depreciaions or currency crises. his is undersandable given ha currency crises generally have powerful adverse effecs on growh as documened by Cerra and Saxena 008 and Bussière Saxena and ovar 0 among ohers. One excepion is Kappler e al. 0 who look a he impac of an appreciaion on he curren accoun balance and on real oupu. hey build heir empirical resuls of a formal definiion of large exchange rae appreciaion and find ha large appreciaions lead o a deerioraion of he curren accoun hrough lower savings and lower expors he effecs being larger in emerging and developing economies. hey however find lile impac on overall GDP as domesic demand 4

6 and ne expors move in opposie direcions. We build on heir work in several ways. Firs we do no limi our analysis o only large appreciaion bu consider such a possibiliy as a special case of our empirical approach. Second our mehod of impac measuremen conrols for counry specificiies or bias selecion. hird we consider he underlying reasons for he appreciaion. While nonrivial such a disincion is imporan as a given appreciaion can be associaed wih very differen movemens in growh depending on he driving shock. In paricular we disinguish episodes driven by movemens in inernaional capial flows from episodes driven by domesic produciviy shocks. Our emphasis on movemens in capial flows which o our knowledge has no been aken previously fis wih a growing emphasis in inernaional economics on capial flow bonanzas einhar and einhar 008 and lending booms Gourinchas e al. 00. Movemens in inernaional capial flows can in principle reflec he fundamenals of he paricular economy or global fundamenals. A growing body of lieraure sresses he prominen role of he laer. Forbes and Warnock 0 argue ha episodes of large movemens in capial inflows and ouflows are associaed wih changes in global risk especially for flows in deb insrumens while local fundamenals do no have a robus effec. Ghosh e al. 0 also documen he role of global facors for episodes of large ne flows wih local facors playing a secondary albei relevan role. ey 03 sresses he relevance of global financial cycles in driving economic condiions regardless of he exchange rae regime. 4 A prominen sudy of episodes of large appreciaions is Goldfajn and Valdes 999 who however focus on he persisence and unwinding of episodes where he appreciaion is ou of line wih fundamenals which is a differen focus han ours. Our heoreical conribuion is relaed o he lieraure ha assesses he drivers of real exchange raes. Saring from he exbook Balassa-Samuelson effec where higher produciviy in he raded secor leads o a currency appreciaion he lieraure has considered he impac of shocks o financial flows and he presence of exernaliies. 5 Benigno and omei 0 consider financial shocks in he form of a ighening of borrowing consrains. hey absrac from he disincion beween raded and non-raded goods which plays a key role in our model. Benigno and Fornaro 04 develop a model where a real appreciaion of he exchange rae leads o a cosly exernaliy. heir model considers endogenous growh in he raded secor. A real 4 Ghosh e al. 0 however find ha surges of capial flows are less frequen and smaller in counries wih flexible exchange raes. 5 Benigno and hoenissen 003 show ha when raded goods are differeniaed a produciviy increase in he raded secor can lead o a real depreciaion because of offseing impacs on he erms-of-rade and he relaive price of non-raded goods. 5

7 depreciaion shifs demand away from raded goods and hus leads o a reducion of produciviy growh ha is no inernalized by agens. While we also consider ha exchange rae movemens have a cosly exernaliy hrough a desrucion of firms our framework ensures ha his is he case for appreciaions and depreciaions. he impac of exchange rae movemens on he number of firms follows from Bacchea and van Wincoop 03 who however focus on he issue of muliple equilibria under sicky prices. While he focus on he consequence of real appreciaions is relaively recen an exensive lieraure has considered he impac of real depreciaion. A cenral mechanism in ha line of research is he presence of balance shee consrains wih a depreciaion raising he local currency value of foreign deb and he local cos of impored inpus as in Cespedes Chang and Velasco 004. While our approach differs from heir specific modeling we also consider coss of exchange rae movemens hrough firms profiabiliy. 3. Growh appreciaions capial surges and produciviy boom: an empirical assessmen his secion presens our economeric analysis of he effecs of an appreciaion on growh where we allow for he paern o vary depending on wheher he economy experiences a produciviy boom and/or shocks o capial flows. We sar by defining our measures of appreciaions produciviy and capial flow surges. We hen presen he mehod of propensiy score maching used in he analysis before reporing our resuls. 3. Definiions Our analysis relies on annual daa from 960 o 0 for a broad sample of 68 counries 30 advanced and 38 emerging. 6 An appreciaion episode is defined as a year-o-year appreciaion of he real effecive exchange rae which is no preceded by a depreciaion over he previous year in order o absrac from episodes of cach-up afer a depreciaion. 7 he average real appreciaion in hese episodes amouns o 5.4 percen compared o he previous year. As poined above he empirical lieraure sresses he role of swings in inernaional financial flows as a major driver of economic performance in emerging economies as well as advanced economies as he curren crisis highlighs. We consider gross capial inflows measured as a percenage of GDP and define a capial surge as an annual increase in inflows ha is one sandard deviaion 6 We decide o use yearly daa as quarerly daa especially for capial flows end o be noisy and have a limied ime span. 7 We consider he effecive raher han he bilaeral exchange rae as we are ineresed in he aggregae macroeconomic oucome. 6

8 above he average increase over he five previous years following Forbes and Warnock 0. 8 In our sample surge episodes are characerized by increases of gross capial inflows by a leas 7 percenage poins of GDP over he previous year. We follow a similar approach in defining a produciviy increase episode. Specifically we consider siuaion where he annual increase in produciviy is one sandard deviaion above he average increase over he five previous years. 9 hese definiions enable us o consider hree ypes of appreciaion episodes: a sole appreciaion an appreciaion combined wih a produciviy increase and an appreciaion combined wih a capial surge. able repors heir incidence. he figures on he diagonal show he oal number of episodes here are for insance 35 cases of produciviy increases and he figures in he lower riangle show he numbers of episode where wo ou of he hree characerisics are observed. here is a reasonable amoun of informaion regarding an appreciaion in presence of a produciviy shock 77 observaions or a capial surge 96 observaions. Ineresingly here are only a few insances where we observe an appreciaion a capial surge and a produciviy increase 6 observaions. We herefore ignore he overlap of all hree characerisics in our analysis. 3.. Propensiy score maching Our assessmen of he linkages beween growh and appreciaions canno merely consis of an examinaion of he co-movemens beween hese variables. his is because we need o address he main challenge in our exercise namely endogeneiy and selecion bias. he selecion bias occurs when a real appreciaion he reamen is no randomly allocaed across counries bu is insead correlaed wih oher variables. A difference in growh beween counries faced wih an appreciaion he so-called reaed group and he oher counries he so-called conrol group could hen be aribuable o sysemaic differences in some variables beween he reaed and conrol group raher han he effec of he reamen iself. A sandard approach is o rely on an insrumenal variable ha affecs he appreciaion bu does no direcly affec growh. Conrolling for he differences across counries hrough an effecive insrumen is however quie difficul especially in presence of limied amoun of daa. 8 Gross capial inflows are he sum of inflows of direc invesmen porfolio inflows and oher inflows. 9 he measure of produciviy is he oupu per person provided by Oxford Economics. We do no repor he benchmark as he oupu is measured in domesic currency: an average across currencies would no be useful. 7

9 One alernaive commonly used in labor economics and medical research is he Propensiy Score Maching PSM approach as developed in osenbaum and ubin 983. his mehodology has so far only been used in a small number of sudies in inernaional macroeconomics including Glick Guo and Huchinson 006 who look a currency crisis Das and Bergsrom 0 who assess capial accoun liberalizaion Forbes Frazscher and Sraub 03 who consider capial conrol and prudenial measures and Forbes and Klein 03 who analyze he policy responses o crisis. o our knowledge he mehod has no been used o assess he linkage beween exchange rae movemens and growh. he idea behind he PSM approach is o deermine wheher a reamen in our case he hree differen ypes of appreciaions wih or wihou produciviy increases and capial surges leads o differen oucomes han he absence of reamen by maching reaed observaions wih conrol observaions ha share similar characerisics oher han he presence of he reamen. In oher words i consrucs a counerfacual for he reamen based on a se of observable characerisics. o illusrae he PSM mehodology we denoe he indicaor reamen for counry i by D i which is equal o when here is an appreciaion i.e. he reamen and 0 oherwise. he counry i growh rae is 8

10 independence assumpion. his assumpion saes ha condiional on a vecor of observable characerisics X he variable of ineres he growh rae for us is independen of he reamen saus. Condiional on his vecor X he expeced growh rae in absence of an appreciaion would hen be he same for paired counries ha is E[Y i0 D X] E[Y i0 D 0X] and he bias would disappear. Under his assumpion hen A effec is wrien as: A E[Y i D X] - E[Y i0 D 0X] where E[Y i D X] conrols for he relevan se of characerisics X. his se should include variables ha are co-deerminans of boh appreciaion and growh and condiioning on all relevan variables may be a challenge. osenbaum and ubin 983 and Imbens 004 show ha if he hypohesis of condiional independence hold hen all biases due o observable componens can be removed by condiioning on he propensiy score. Hence A becomes: A E[Y i D px] - E[Y i0 D 0pX] where E[Y i D px] denoes he fac ha we conrol for he probabiliy of observing he reamen condiional on he se X of variables. o obain A we firs need o esimae he propensiy score o conrol for he likelihood of receiving he reamen and hen use a maching algorihm o pair he observaions based on observable characerisics. We consider four differen maching algorihms all wih advanages and drawbacks. he firs is he neares-neighbor ha pairs each observaion in he reaed group wih he closes observaion in erm of propensiy score from he conrol group. his limis he incidence of bad maches a he cos of excluding a lo of poenially useful informaion. We use his algorihm wih replacemen meaning ha he conrol observaions can be used as a mach more han once. his choice decreases he bias increases he average qualiy of maching and he resuls do no depend on he order in which observaions ge mached. he rade-off is ha i increases he esimaor variance if we compare wih he alernaive of no replacemen. he second maching algorihm is he five neares neighbors ha akes five counries insead of one from he conrol group. his maching rades reduced variance as more informaion is used o consruc he counerfacual a he cos of some bias as 9

11 he average mach is poorer han in he previous case. he las wo maching algorihms he kernel and local-linear mehod are non-parameric maching esimaors ha compare he oucome of each reaed observaion o a weighed average of he oucomes of all conrol observaions wih he highes weigh being placed on he conrol observaions wih he closes propensiy scores o he reaed observaion. hese algorihms have a lower variance as hey use more informaion a he cos of being more exposed o bad maches. Applying hese maching mehods requires ha wo hypoheses mus be saisfied. he firs is he condiional independence assumpion described above. he second is he common suppor condiion which ensures ha here is sufficien overlap in he characerisics of he reaed and unreaed o find adequae maches Esimaion of he impac of appreciaions he explanaory variables considered in he propensiy score maching are seleced o capure counry specificiies as well as global condiions. We conrol for global condiions hrough he VXO as a measure of global uncerainy he log of he commodiy price index and he US ineres rae. Counry characerisics are capured by changes in real GDP he raio of privae credi o GDP and produciviy. We also include inflaion he differenial beween he US and he counry ineres rae he level of reserves scaled by GDP he curren accoun scaled by GDP he indicaor capial accoun openness by Chinn and Io 008 updaed in 03 and a dummy equal o one if he exchange rae is pegged. 0 All variables used in he logi regression are lagged in order o ensure ha hey are no affeced by he reamen. he propensiy score px specificaion ha we reain reflecs a compromise beween he poenial influence of a variable on he oucome and is abiliy o improve he maching. able repors he logi esimaions used o produce he final propensiy score specificaion for he hree cases of appreciaion wih and wihou produciviy increase and surge. As expeced no all he variables are saisically significan bu overall he variables help capure he specificiies while esimaing he PSM help he maching or boh. 0 he IMF and he World Bank are he source of he daa wih he excepion of he daa on produciviy from Oxford Economics he Chinn and Io index from Chinn and Io he Goldman and Sachs commodiy index and he VXO from Daasream and he peg dummy is defined as in Shambaugh 004. he resuls are robus o several alernaives based on he lis of variables provided as long as we do no reduce oo much he number of reaed observaions. 0

12 As poined above our approach requires ha he condiional independence condiion hold ha is ha he value of he various conrol variables do no significanly differ beween he reamen and conrol groups once he maching is compued. able 3 assesses his poin for he four maching algorihms considered. 3 he p-values repored correspond o he es of mean equaliy for a variable beween he reaed and he conrol group before he maching unmached and afer. We observe ha no significan difference remains in he daa afer any of he four maching procedures. All four maching algorihms are hus clearly suiable for our analysis and perform quie well. Finally he common suppor condiion of sufficien overlap in he characerisics of he reaed and unreaed observaions also needs o be validaed. o do so he remaining of he analysis focuses on reaed observaions ha have a propensiy score beween he maximum and he minimum propensiy score of he conrol group. Having esablished ha all condiions required for he use of our mehod hold we now compare he impac of he differen appreciaions on growh by performing he maches and esimaing he A effecs. he resuls are repored in able 4 focusing on he year of he appreciaion. Each panel in he able corresponds o one of he four maching algorihms. In each panel we underake he analysis for all counries and hen focus on emerging economies EME and advanced economies AE separaely. For each we consider he hree cases of appreciaion sole wih produciviy boom wih capial surge as well as he cases of produciviy booms or capial surge ha are no accompanied by an appreciaion. his allows us o idenify he overall impac of he differen cases of appreciaion as well as isolae he impac of he appreciaion iself when combined wih anoher change. While he size of he A esimaes vary somewha depending on he algorihms heir direcion and saisical srengh are quie robus. Four main oucomes can be highlighed. Firs an appreciaion ha is no accompanied by a produciviy boom or a capial surge does no have any significan growh impac overall or for emerging economies bu has a negaive impac for advanced economies subracing beween 0.8 and percenage poin of growh compared o he conrol group depending on he maching algorihm. Second growh is boosed in counries experiencing an appreciaion accompanied by a produciviy boom. his is observed across all counries and is especially pronounced for emerging economies where i adds beween.8 and We repor he case wih only an appreciaion bu he oucomes are similar for he oher cases. 3 We use he Saa module PSMACH developed by Leuven and Sianesi 003.

13 3.03 percenage poins o growh. here is also a posiive effec in advanced economies albei of a smaller magniude. hird an appreciaion associaed wih a capial flows surge does no yield a significan effec on growh excep for emerging economies where i adds up o.77 percenage poins. he significance of he effec is however sensiive o he maching algorihm considered. Finally a produciviy increase no accompanied by an appreciaion clearly booss growh in all economies while a capial surge has some posiive impac in emerging economies. Having esablished he impac of an appreciaion combined wih a produciviy boom or a capial surge we now urn o he quesion of he conribuion of he appreciaion per se. he esimaes of he impac of higher produciviy and capial inflows in able 4 end o be larger in he absence of appreciaion. We hus specifically es wheher he growh variaions observed in he presence of an appreciaion combined wih anoher change is saisically differen from he one observed wihou an appreciaion. he resuls are repored in able 5 where negaive values indicae ha he appreciaion per se reduces growh. While he significance of esimaes vary depending on he maching algorihm considered we observe ha in he presence of a produciviy boom an appreciaion ends o lower growh by up o 0.65 percenage poins especially in emerging economies. By conras he appreciaion ends o boos growh in advanced economies by up o 0.5 percenage poins. urning o he impac of an appreciaion during a capial flow surge able 5 shows ha he appreciaion has an adverse effec on growh for boh emerging and advanced economies wih some sensiiviy o he maching algorihm considered. he final sep of our analysis is o assess wheher he magniude of appreciaion maers. Inuiively he adverse impac of an appreciaion can come hrough a reduced compeiiveness of he raded secor. While exporers can handle a moderae appreciaion by lowering heir profi margin hey may have o cu down on producion if he appreciaion becomes large. We herefore consider wheher our resuls sill hold if we were o consider only episodes of srong appreciaions. Allegaions of currency war and hrea of enhanced capial conrol measures during he pos financial crisis period are a naural moivaion for his quesion. Our definiion of a srong appreciaion parallels he one of produciviy booms and capial flow surges. Specifically a srong appreciaion occurs when he annual increase if he real effecive exchange rae is a leas one sandard deviaion above he average increase over he five previous years. We also consider a reference period of hree years as a robusness check. Furhermore a period of srong appreciaion canno be preceded by a depreciaion of equal size. o rule ou cach

14 up episodes. Quaniaively large appreciaions correspond o an average increase in he real effecive exchange rae by 7.33 percen when he reference period is of five years and by 5.58 percen for he hree years reference period. 4 A drawback of considering large appreciaions is ha he maching process dramaically reduces he number of observaions usable in his analysis which limis our abiliy o draw conclusions beyond he observed sample. he resuling A esimaes are repored in able 6 and 7 which correspond o able 4 for he five and hree year reference periods respecively. Overall he paern is similar o able 4 wih a weakening of some saisical evidence. he ables make hree poins. Firs a srong appreciaion has a negaive growh impac in advanced economies varying from no significan effec using he hree years reference period o up o -.39 percen using he five year period. Second emerging economies remain sensiive o an appreciaion combined wih higher produciviy which adds beween.37 and 3.53 percenage poins o growh. By conras here is no clear evidence of an impac for advanced economies. Finally a srong appreciaion combined wih a capial surge seems o have no growh impac across any of he hree groups of counries considered. his robusness analysis shows ha focusing on srong appreciaion episodes does no change he overall message of our resuls. Moreover he enhanced growh rae observed in emerging economies ends up being sronger in he presence of a srong appreciaion han in he presence of an appreciaion. o sum up our empirical analysis shows ha he link beween growh and he real exchange rae depends on he underlying shock. An appreciaion accompanied by higher produciviy is associaed o higher growh. For emerging economies he posiive growh primarily reflecs he produciviy gains as he appreciaion per se ends o lower i. By conras an appreciaion accompanied by a capial flow surge ends no o be associaed wih higher growh and he effec of he appreciaion iself is negaive. he nex secion presens a simple model where he link beween appreciaion and growh depends on he underlying shock wih he possibiliy of inefficien allocaions ha a planner would wan o correc. 4 hese averages are across he counries and he ime period as a resul i is only useful in comparison wih he average appreciaion repored in he previous secion

15 4. A simple model of he real exchange rae In his secion we presen a small open economy model ha conrass he impac of produciviy and ineres rae shocks on he real exchange rae oupu and he curren accoun. he model makes hree poins. Firs appreciaions semming from capial inflows surge are associaed wih weak growh while appreciaions associaed due produciviy shocks are accompanied by high growh in line wih our empirical evidence. Second he weak growh observed during a capial flow driven appreciaion does no imply ha a mirroring depreciaion is associaed wih high growh. Insead he fricions in he model imply ha exchange rae movemens driven by capial flows shocks are associaed wih low growh. hird he model includes an exernaliy ha leads o inefficienly low growh following capial flows shocks hereby jusifying policy makers concerns abou boh capial flows and sudden sops. For breviy we focus on he main feaures and resuls and leave more deails o he appendix. 5 We firs presen he building blocks and he soluion mehod. We hen derive he soluion and discuss he resuls hrough a numerical illusraion. 4. Building blocks We consider a small open economy where a represenaive agen consumes a baske C of raded and non-raded goods: C C C where denoes ime C and C are he consumpions of he homogeneous raded and nonraded goods respecively and is he share of raded goods in he consumpion baske. he price of he raded good P is exogenously se in he world marke and he consumer price index is P P - where P / P is he relaive price of he non-raded good. he allocaion of consumpion across raded and non-raded goods akes he sandard form. he real exchange rae is relaed o wih an increase in he relaive price of he non-raded good leading o a real appreciaion. here are n firms producing he raded good and n firms producing he non-raded good a ime. Producion uses labor wih decreasing reurns o scale. he oupu of individual firms denoed by i in he raded and non-raded secor are Y i A L i - and Y i A 5 he fully deailed seps of he model soluion are available on reques. 4

16 5 L i - respecively and A k is an exogenous produciviy erm in secor k. he parameer reflecs he degree of reurns o scale. he case of corresponds o an endowmen economy while he case of 0 corresponds o consan reurns o scale. he oal labor inpu is se exogenously o. he assumpion of decreasing reurns o scale implies ha for a given oal labor inpu in a secor L k spread evenly across firms oupu is an increasing funcion of he number of firms in he secor: 0 0 / k k k n k k k n k k k L A n di n L A di i L A Y k k Firms in all secors pay a wage equal o he marginal produc of labor. Combining he ensuing labor demand wih he echnology we wrie he profis of individual firms in he raded and nonraded good secors as: / / / A w Π / / / A w Π where w is he real wage and profis are measured in unis of he consumpion baske. he oupu levels in he wo secors and he aggregae oupu in erms of he consumpion baske: / / / / / / / / Y Y Y A w n Y A w n Y he household can freely lend and borrow from world markes in a bond denominaed in he raded good wih an exogenous ineres rae r. he income of he agen is he wage received on he oal labor supply and he profis of all firms. he budge consrain for period is hen: Π Π B r n n w B C 3 he clearing of he non-raded good marke is given by: / / / A w n C 4 he represenaive Home household maximizes he following ineremporal uiliy of consumpion: 0 ln s s s C U β 5 he household maximizes 5 subjec o he budge consrain 3 aking he value of profis - he real wage and he numbers of firms as given. he ensuing Euler condiion is:

17 C 6 βc r / he las building block of he model is he deerminaion of he number of firms in each secor. We consider ha in he iniial allocaion firms face an idiosyncraic fixed cos. Specifically he n s firm in he raded secor and he n s firm in he non-raded secor face he following fixed coss denominaed in unis of he consumpion baske: σ σ Z n ; Z n 7 We assume ha hese coss do no enail a loss of real resources bu are ransferred by firms o he household and hus do no ener he budge consrain 3. Firms in each secor are hus ranked according o heir fixed cos. σ capures he sensiiviy of he cos o he rank of he marginal firm. he number of firms in he iniial allocaion is such ha he marginal firm in each secor makes zero profis including he fixed cos: Z Z σ / / / n w A 8 σ / / / n w A 9 4. Soluion mehod We solve he model in erms of linear log approximaions around an iniial seady sae where he counry does no hold inernaional bonds and he world ineres rae offses he discoun facor. σ We se produciviy o A in boh secor and Z A in 7 and σ Z A in 8. his parameerizaion ensures ha: n L C Y A w A n L Π A We consider ha in period he shor run he economy is hi by a one-period shock in he ineres rae or by permanen produciviy shocks. From period onwards he economy reaches a new seady sae he long run. A cenral feaure of he model is he deerminaion of he number of firms. We follow he model by Bacchea and van Wincoop 03 where firms making insufficien profis shu down. Specifically shocks lead o shor run movemens in wages demand and he relaive price of he non-raded good which in urn affec profis. We assume ha firms canno be creaed in he shor run. While his assumpion is resricive i is moivaed by he fac ha firm creaion is a more cosly process han firm desrucion. If all firms in a secor make posiive profis in he shor run 6 Π Y A Y A

18 he number of firms remains a is iniial level. If he shocks lead o lower shor run profis he marginal firms are no able o cover heir fixed coss and we assume ha hey shu down in he shor run. he number of firms in he secor is hen given by 8 or 9. In he long run we assume ha he number of firms in each secor reurns o he iniial values of and - so any shor run shudown is emporary. his assumpion is done for racabiliy. Anoher alernaive would be o allow for firm creaion and hus le he long-run number of firms be deermined by 8 and 9. his alernaive however raises he complexiy of he model. A hird alernaive would be o assume persisence in he number of firms so ha firms shuing down in he shor run never reappear. Such an alernaive would however be quesionable in he long run. We sar by solving for he long run allocaion condiional on he bond holdings acquired in he shor run. Denoing log deviaions by haed values we wrie: where: C L A β B A β β B β A agg Π A w Y A agg β B A β A agg dif and agg β diff B A β β Π B A β agg 0 A A. A produciviy increase raises consumpion wages real oupu evaluaed a he iniial relaive prices and profis and affecs he real exchange rae hrough he sandard Balassa-Samuelson effec. Higher wealh B > 0 raises consumpion and leads o a real appreciaion ha shifs labor owards he non-raded secor and raises profis in ha secor relaive o he raded secor. We now urn o he shor run allocaion. We firs solve for he allocaion condiional on he shor run number of firms in each secor. Using he long-run soluion 0 he Euler condiion 6 he budge consrain 3 he clearing of he non-raded good marke 4 and he fac ha he marginal produc of labor is equalized across secors we wrie: 7

19 C B β r n L β r n A A agg β r n dif n n agg dif w A agg n agg where: agg n n n and dif n n n. he shor run profis are given by: Π Π A A agg agg n n agg agg n n dif dif β r n β r n shows ha he profis reflec he shocks and he number of firms in each secor. If he shocks are such ha he profis in are posiive for an unchanged number of firms hen he number of firms in ha secor remains a he iniial value. Oherwise he number of firms adjuss so ha he marginal firm makes zero profis: Π σ n Π σn In general he soluion depends on wheher he zero marginal profi consrains are binding or no. For breviy we illusrae he resuls by aking hree paricular cases. he welfare impac of shocks can be assessed by aking a linear approximaion of he uiliy 5 and using 0 and : β β agg agg U C C β A n he welfare hus only reflecs he produciviy shocks and any changes in he number of firms. In paricular a reducion in he number of firms in any secor is welfare reducing. A social planner would hus choose o deliver a lump sum subsidy o firms in he secor facing low profis paid for by a lump sum ax on he household in order o preven he desrucion of firms. As he fixed 8

20 coss faced by firms are only a ransfer and do no enail a use of real resources hey are no a consrain for he planner hree specific cases We firs consider he case of a permanen produciviy increase in he raded secor A > 0. shows ha profis in boh secors are posiive when holding he number of firms o heir iniial values. here is hen no desrucion of firms in he shor run and he shock permanenly booss consumpion wages and profis and appreciaes he currency: C C w w Π Π Π Π A A B L 0 3 Profis equally rise in he wo secors bu for differen reasons. Firms in he non-raded secor benefi from he higher relaive price of heir oupu while firms in he raded secor benefi from heir higher produciviy which more han offses heir lowered compeiiveness relaive o nonraded firms. We nex urn o a emporary increase in he world ineres rae r > 0 a shock ha can be inerpreed as a sudden-sop of inernaional capial flows. shows ha profis are reduced in he non-raded secor if we hold he number of firms unchanged in boh secors: β β Π 0 n n r Π n n 0 r 4 I hus canno be he case ha he number of non-raded firms remains consan. Insead i is given by he zero-profi condiion Π σ. We show in he appendix and in he numerical n example below ha in equilibrium he number of firms in he non-raded secor he relaive price of he non-raded good consumpion and wages all decrease. Labor is reallocaed owards he raded secor bu he ensuing increased in raded oupu is no enough o offse he conracion in non-raded oupu due o labor reallocaion and he reducion of he number of 6 We can explicily solve for he planner s allocaion around he seady sae and show ha he planner never lowers he number of firms from he seady sae values. he planner s allocaion is hus he one wih consan numbers of firms. 9

21 firms and overall oupu falls. he conracion of overall oupu solely reflecs he reducion of he number of firms in he non-raded secor. We finally consider a emporary decrease in he world ineres rae r < 0 a shock ha can be inerpreed as a capial inflows surge. 4 shows ha profis are reduced in he raded secor if we hold he number of firms unchanged in boh secors. We show below ha in equilibrium he number of firms in he raded secor consumpion and wages all decrease. he relaive price of he non-raded good increases leading o labor reallocaion owards he non-raded secor. he ensuing increase in non-raded oupu is no enough o offse he conracion in raded oupu and overall oupu falls. he conracion of overall oupu solely reflecs he reducion of he number of firms in he raded secor. 4.4 umerical illusraion We illusrae our resuls wih a numerical example. We assume ha raded goods accoun for 30 percen of he consumpion baske 0.3 ha here are decreasing reurns o scale 0.3 and se he discoun facor β o We se he sensiiviy of he fixed cos o he number of firms σ o uniy. Figure shows he shor run impac on Home counry variables in response o he hree shocks discussed above. he response o a permanen produciviy increase in he raded secor is given by he sriped bar he black bars shows he impac of a reducion in he world ineres rae and he doed bar shows he impac of an ineres rae increase. he shocks are parameerized o lead o a uni response in he relaive price of he non-raded good. he op panel of Figure presens he main variables namely he relaive price overall consumpion he curren accoun and overall oupu. he boom panel shows a broader range of variables. he main message of Figure is ha ineres rae shocks are conracionnary regardless of heir direcion while a produciviy increase raises oupu. he key reason behind he reducion in overall oupu in response o ineres rae shocks is ha hese shocks lower profis in a secor raded for an ineres rae decrease and non-raded for an increase and lead o he desrucion of some firms in he secors. Because of he decreasing reurns o scale in producion shifing he labor inpu used by he desroyed firms o he surviving ones does no generae enough oupu o replace he foregone one and overall oupu falls. 0

22 Ineresingly here is a gap beween he impacs of ineres rae decreases and increases. For insance oupu and firm desrucions are more acue following a decrease. Even for variables ha move in opposie direcions across he wo shocks such as he labor inpu in he non-raded secor we observe ha he effecs are of a larger magniude following an ineres rae decrease. his is because n eners and no only in a symmeric manner as affecing he real ineres rae expressed in erms of he consumpion baske he n does bu also by r n erm. A fall in he number of raded firms direcly reduces he relaive price of he non-raded good 7 which in urn lowers he real ineres rae expressed in erms of he consumpion baske equal o r and hus magnifies he iniial shock. A reducion in he number of nonraded firms does no generae such an effec. As poined above a cenral planner would no allow for he number of firms o decrease as his lowers welfare and is no sricly speaking necessary because he fixed coss faced by firms only represen a ransfer and hus do no enail a real use of resources. he planner s allocaion corresponds o n n 0. We hus assess how he planner s allocaion differs from he decenralized one for he hree shocks we consider. here is no discrepancy following an increase in produciviy as his does no lead o any desrucion of firms under he decenralized allocaion. Figure conrass he allocaions following a decrease in he ineres rae wih he shock parameerized o lead o a uni movemen in he relaive price of he non-raded good under he decenralized allocaion. he planner s allocaion prevens he reducion in he number of firms in he raded secor. his subsanially reduces he conracion in raded oupu and avoids he decrease in overall oupu. he boom in consumpion and associaed curren accoun defici are reduced under he planner s allocaion. Ineresingly he increase in he relaive price of he non-raded good real appreciaion is higher under he planner s allocaion han under he decenralized oucome. Our model hus shows ha while relaive price movemens have derimenal effecs under he decenralized allocaion prevening hese effecs does no imply ha price movemens should be dampened. A similar paern is observed in Figure 3 which conrass he allocaions following an increase in he ineres rae. We again see ha he planner prevens he reducion in he number 7 In a seady sae where he number of firms can change he relaive price is posiively affeced by n n.

23 of firms in he non-raded secor hereby limiing he conracion of oupu in ha secor and prevening an overall recession. he decrease in consumpion is dampened bu his is no accompanied by a dampening of he curren accoun surplus. In addiion he real depreciaion of he currency is larger under he planner s allocaion. 4.5 Insighs from he heory Our model delivers an asymmery beween he impac of appreciaions and depreciaions semming from shocks o capial flows as well as an exernaliy ha makes he decenralized allocaion inefficien in some cases. he asymmery reflecs he fac ha shocks o capial flows affec profis in he raded and non-raded secors differenly. hey always lead o lower profis and firms shudown in one secor. his canno be offse by firms creaion in he oher secor as we assume ha creaion is no possible in he shor run and we herefore always ge a lower number of firms. Combined wih decreasing reurns o scale his leads o a conracion in oupu. Our model can hus delivers adverse effecs boh of real depreciaions linking wih he lieraure on sudden sops and appreciaions linking wih he recen concerns of policy makers. An exernaliy is also presen in our seing as he fixed coss faced by firms which drive he movemens in he number of firms do no represen a real cos and hus are ignored by a planner. A planner would always preven he reducion in he number of firms o suppor oupu and welfare. Ineresingly he movemen in he real exchange rae is larger under he planner s allocaion despie he fac ha his movemen is wha leads o profis falling in a secor. A limiaion of our model is ha i does no deliver any hreshold effecs where he impacs of small and large shocks differ. For insance he impac of a percenage poin move in he ineres rae is simply wice he impac of a 50 basis poins move. One could exend he model o allow for such hreshold effec for insance by allowing for he marginal firm in each secor o make posiive profis in he seady sae. A small reducion in is profis would hen no endanger i while i would have o shu down following a large reducion of profis. Allowing for such hreshold effecs would however make he model subsanially more complex. he model we consider is deliberaely kep simple for breviy and can be exended in many direcions. One could include differeniaed raded goods leading o movemens in he erms-ofrade. Anoher exension would be o allow for an endogenous labor supply. Overall oupu hen

24 would no jus reflec produciviy and he number of firms hus dampening he conracion following shocks o he ineres rae. 8 Finally we assume ha he planner can implemen lump sum axes and subsidies o preven he numbers of firms from falling. If such subsidies are no feasible he model could be used o assess wheher axes or subsidies on he consumpion of raded or non-raded goods or on inernaional borrowing could also keep he numbers firms sable. 6. Conclusion his paper invesigaes he connecion beween srong appreciaions and growh. Our main message is ha one canno draw a simple link beween he wo variables and insead needs o consider he specific underlying shock. We firs esablish hree main sylized facs from a broad daase of emerging and advanced economies. Firs exchange rae appreciaions are associaed wih weaker growh only for advanced economies when compared o normal imes. Second boh emerging and advanced economies observe enhanced growh when appreciaions are accompanied by a produciviy increase. oneheless he growh primarily reflecs he produciviy gain. he appreciaion per se reduces growh in emerging economies bu no in advanced economies. hird appreciaions ha are associaed wih a surge in capial inflows are characerized by weaker growh compared o episodes wih a produciviy increase. In addiion he impac of he appreciaion per se in a capial surge episode is clearly negaive. A robusness analysis shows ha hese resuls sill hold when considering only cases of srong appreciaion. We develop a simple model ha generaes a paern of real exchange rae and growh consisen wih our empirical findings. In addiion he model also jusifies a concern from policy makers abou exchange rae movemens driven by capial flows surges and sudden sops. Fricions in he secoral re-allocaion of oupu imply ha surges and sudden sops are boh associaed wih a weak oupu performance. Furhermore his performance is inefficien and hus jusifies a policy response albei one ha would no dampen he movemens in he real exchange rae. By 8 Boh feaures were considered in an early version of he paper where we assumed consan number of firms. Boh feaures enriched he model bu did no radically aler is predicions. Given he higher complexiy of he soluion under differeniaed raded goods and endogenous labor supply we oped o absrac from hem for breviy. 3