Carry Trade, Uncovered Interest Parity and Monetary Policy *

Transcription

1 Carry Trade, Uncovered Ineres Pariy and Moneary Policy * Daniel Felcser Balazs Vonnak November 21, 213 Absrac I is well documened in he lieraure ha idenified vecor auoregression (VAR) models ofen produce puzzling resuls when he effec of unexpeced moneary policy movemens is esimaed. Many auhors find ha raising ineres rae generaes proraced appreciaion of he exchange rae (he so-called delayed overshooing puzzle) which is in conradicion wih radiional heory of exchange rae dynamics based on uncovered ineres pariy. We esimae a VAR for a panel of small open economies ha are considered as arges for carry rade sraegies. We idenify srucural shocks by allowing he ineres rae and exchange rae o reac simulaneously o moneary policy and changes in expeced risk premium. Our resuls show ha he delayed overshooing is no a robus finding. Exchange rae appreciaion and carry rade movemens ake place almos on impac afer an unexpeced ineres rae hike. Roughly half of he variaion in carry rade posiions can be explained by domesic ineres rae changes and risk premium shocks. * The auhors would like o hank Zolán Reppa and Zolán Schepp for useful commens and suggesions. Any remaining errors and omissions are ours. Direcorae Moneary Policy and Financial Marke Analysis, Magyar Nemzei Bank (he cenral bank of Hungary), felcserd@mnb.hu Direcorae Financial Sysem Analysis, Magyar Nemzei Bank (he cenral bank of Hungary), vonnakb@mnb.hu 1

2 1 Inroducion There is an unpleasan disconnec beween he bes pracice of moneary policy and empirical ess of exchange rae heories. While cenral banks forecass and decision-making rely on models assuming some sor of uncovered ineres pariy (UIP), here seems o be a wide consensus among economericians ha UIP can be rejeced wih high cerainy. Neverheless, he dynamic relaionship beween exchange rae and ineres rae is of special ineres for cenral banks in small open economies where he exchange rae channel of moneary ransmission mechanism is imporan. In our paper we focus only on one of he empirical puzzles, he so-called delayed overshooing (DOS). According o Dornbusch (1976) s model in which UIP holds, afer an (unexpeced) moneary ighening he nominal exchange rae appreciaes insananeously and hen gradually depreciaes o is new level consisen wih purchasing power pariy. However, srucural VAR esimaes, like Eichenbaum and Evans (1995) or more recenly Scholl and Uhlig (28), ofen show a proraced appreciaion lasing even for years. There are, however, some auhors who challenge he idenificaion sraegy of he sudies reporing DOS. Already McCallum (1994) emphasized ha he empirical failure of UIP may be caused by shocks o he exchange rae o which he moneary policy reacs wihin one period. Since in small open economies exchange rae movemens can have a large impac on inflaion and oupu, a quick response of cenral banks o hose shocks can be jusified. Cushman and Zha (1997), Kim and Roubini (2), Faus and Rogers (23), Bjørnland (29), Jarocinski (21) and Vonnák (21) relax he assumpion ha moneary policy reacs o exchange rae shocks only wih delay, which is ofen implicily made by Cholesky decomposiion of he variancecovariance marix. Allowing simulaneiy beween moneary policy and he exchange rae yields impulse responses resembling o Dornbusch s overshooing model. I should be noed, however, ha Scholl and Uhlig (28) find DOS wihou assuming recursive srucure among he shocks and he variables. Anoher issue which recenly has received much aenion and is presumably relaed o delayed overshooing is carry rade aciviy. Carry raders borrow in low-ineres-rae currency and lend in high-ineres-rae currency. As long as UIP holds, he profi of his sraegy is zero on average, since he ineres rae premium is perfecly offse by he exchange rae depreciaion. If we augmen he UIP by a (ime-varying) risk premium erm, he reurn of a carry rade posiion correlaes wih his risk premium which can be regarded as he compensaion for aking he exchange rae risk. Sill, as long as exogenous changes in he policy rae do no affec he risk premium, UIP holds condiionally, 1 and afer he infiniesimally shor period during which he exchange rae jumps according o Dornbusch s model, here is no incenive for carry raders o change heir exposure. Thus, afer a moneary shock we would expec only a very emporary change in speculaive posiions. On he conrary, he delayed reacion of exchange rae o moneary policy provides excess reurn for several periods. Afer an ineres rae hike a carry rader could make profi from higher reurn on domesic asses as well as from he appreciaion of he currency. Since he exchange rae appreciaes gradually, DOS would imply proraced carry rade inflow. In he seminal paper of Brunnermeier, Nagel, and Pedersen (29) slowly moving carry raders and DOS are shown o be he wo sides of he same coin. The auhors esimae a VAR and show ha he reacion of boh he exchange rae and carry raders o an ineres rae shock is proraced. In he model of Planin and Shin (211) carry raders endogenously amplify he effec of 1 By UIP holding condiionally we mean ha afer a shock he realized reurn on a carry posiion does no change which implies ha he effec of he shock on he ineres rae differenial is equal o is effec on he (expeced) change in exchange rae. 2

3 moneary policy on he exchange rae. They assume ha carry raders while going long in local currency, increase he credi supply and herefore generae an overheaing in domesic demand. In response o his, he cenral bank increases furher he ineres rae which aracs more capial from abroad. 2 The resul is a moneary policy generaed bubble ha ends up in a currency crash. In his model he role of carry raders is desabilizing, as opposed o he convenional UIP framework where i is he carry rader who helps he pariy condiion o be fulfilled quickly. In our paper we deploy a robus economeric mehodology o invesigae he relaionship beween moneary policy, exchange rae and carry rade. Similarly o Brunnermeier, Nagel, and Pedersen (29) we esimae he effec of he ineres rae change on carry rade aciviy and he exchange rae wihin he same model. By doing his, we have he chance o uncover no only how moneary policy affecs he exchange rae and carry rade, bu also how carry rade ransmis moneary policy shocks. We esimae our srucural VAR on a panel of hree small open economies (Ausralia, Canada, U.K.) having currencies considered o have been arge for carry rade. However, unlike he above-menioned paper, we disinguish beween endogenous and exogenous ineres rae movemens by idenifying moneary policy and oher srucural shocks. Following he previously menioned sudies ha challenged he exisence of DOS, we allow he moneary policy o reac simulaneously o exchange rae or risk premium shocks by imposing sign insead of zero resricions. Our second conribuion o he lieraure is ha we ry o find he main driving forces behind carry rade. To his end, we idenify four domesic and four foreign shocks. The variance decomposiion of carry rade daa may inform us abou wheher he exchange rae is a shock absorber or a source of idiosyncraic shocks, and wheher raders on he FX-marke help he exchange rae reac quickly o changes in fundamenals or generae undesired volailiy. Our approach is similar o ha of Anzuini and Fornari (212). Alhough hey focus more on deerminans of carry rade and less on DOS, heir approach is common wih ours in recognizing he imporance of he idenificaion of economically meaningful shocks. However, here are essenial differences in he model specificaion. Probably he mos imporan is ha while hey esimae a VAR on relaive variables (domesic minus foreign), we use he original ime series. This may have crucial consequences, since domesic variables are more likely o rack he foreign ones han vice versa. Imposing idenifying resricions on he relaive variables may cause subsanial bias when here is high asymmery in how foreign and domesic variables reac o each oher. The mos obvious example is moneary policy, as we expec he cenral bank of a small open counry o follow some exen he moneary sance in he big economy, bu no he oher way around. Therefore, we expec a beer idenificaion of he relevan srucural shocks in our model. Our resuls show ha delayed overshooing is no a robus finding. Our exchange rae impulse response funcions resemble raher Dornbusch (1976) s overshooing model, consisenly wih UIP. Comparison wih he Cholesky idenificaion scheme confirms previous findings ha improper idenifying resricions embedded implicily in he recursive approach can be responsible o some exen for he puzzle found in some of he referred sudies. Anoher imporan finding is ha carry raders reac o moneary policy according o he UIP: he exogenous shif in moneary policy sance induces a conemporaneous change in speculaive currency posiions which sar revering already in he nex period. These resuls sugges ha he exchange rae channel of moneary ransmission mechanism works as in he Dornbusch model and carry raders play an imporan role in i. Our findings are in line wih hose of Kisgergely (212), who could rejec he hypohesis ha ineres sensiive capial flows can reverse he effec 2 This mechanism is also known as Tosovsky-dilemma, named afer an earlier governor of he Czech cenral bank and appears ofen in cenral bank publicaions and financial marke expers analyses. 3

4 of moneary policy. Variance decomposiion shows ha roughly half of he carry rade movemens can be aribued o surprise movemens in domesic moneary policy sance and changes in risk premium of he domesic currency. While he inerpreaion of he former is sraighforward, he laer is no. On one hand, he dynamics of he exchange rae and carry rade afer a moneary policy shock sugges ha speculaive posiion-aking help he UIP o resore quickly. On he oher hand, he risk premium of a currency can change for wo reasons: eiher because he fundamenals have changed and carry raders adjus heir demand for compensaion for aking risk, or because here is an idiosyncraic shock o carry rade aciviy. In he firs case he role of currency speculaion can be considered as greasing, as he new informaion abou he curren or fuure sae of he economy is channelled ino he exchange rae by carry raders. In he second case, however, currency speculaion is a source of shocks ha can lead o welfare losses. Unforunaely wihin our modelling framework i is no possible o decompose risk premium shocks o changes in he risk profile of he economy and changes in risk appeie, herefore we canno draw firm conclusions o wha exen carry rade aciviy is welfare-improving. The remainder of he paper is srucured as follows. Secion 2 presens our economeric model and he resricions used in he idenificaion of he shocks. Secion 3 describes our daase. Secion 4 presens he resuls. Secion 5 shows resuls from alernaive specificaions as a robusness check. Finally, Secion 6 concludes. 2 Modelling sraegy During he empirical analysis we build on he mehodology presened in Uhlig (25). By using a srucural vecor auoregression (SVAR) model we can idenify srucural, economically meaningful exogenous shocks and causal relaionships beween hem and he endogenous variables. In paricular, a VAR is esimaed in he form ha is given by y = A 1 y 1 + A 2 y A p y p + Cz + ϵ (1) E(ϵ ϵ ) ϵ (2) where = 1...T, y is he vecor of endogenous variables included in he VAR: he log of real gross domesic produc, log of consumer prices, log of 3-monh money marke ineres rae, log of he nominal exchange rae and a proxy for carry rade posiions. A is he coefficien marix and is he variance-covariance marix for he one-sep ahead predicion error. z is he vecor of exogenous variables. Inrinsically, we are ineresed in he parameers of he srucural VAR B y = B 1 y 1 + B 2 y B p y p + Dz + u (3) E(u u ) u = I n (4) where u is he vecor of muually uncorrelaed srucural shocks, I n is an n-dimensional ideniy marix and ϵ = B 1 u (5) Equaion (2) imposes n(n+1)/2 resricions on B. To fully disenangle he srucural shocks from he reduced-form innovaions, we need n(n 1)/ 2 addiional resricions. One can find several approaches in he lieraure o carry ha ou. One is assuming a recursive srucure among 4

5 shocks and heir conemporaneous effec on he endogenous variables, he so-called Cholesky ordering. Kim and Roubini (2) s idenificaion scheme builds on non-recursive zero resricions. Scholl and Uhlig (28) use sign resricions on impulse responses for a prolonged period (one year) afer he shock. Bjørnland (29) employs long-run neuraliy resricions. Bjørnland and Halvorsen (21) combine sign and shor-run resricions. Since in a small open economy boh moneary policy shocks and sudden swings in carry rade (exchange rae or risk premium shocks) may affec he ineres rae and he exchange rae simulaneously, recursive ordering is no appropriae for our purposes. Therefore, we idenify he srucural shocks using mainly sign resricions. Zero resricions are used only for separae financial shocks from hose originaing in real economy. Sign resricions have he advanage of robusness a he price of wider confidence bands of impulse responses han in jus-idenified VARs. The endogenous par of our VAR consiss of GDP, CPI, shor-erm ineres rae, exchange rae and carry rade. Following he slighly modified noaion of Kilian (211), B 1 can be wrien as ϵ Prod ϵ Prices + + u Moneary ϵ Ineres + u Risk ϵ Exchange = u Demand (6) + u Supply ϵ Carry + u 5 where + and denoes he sign of he resriced impac response, indicaes zero resricion and denoes no resricion. According o he resricions, an unanicipaed moneary ighening causes he domesic ineres rae o increase and he exchange rae o appreciae on impac. Carry raders ake long posiion in local currency due o higher ineres rae. 3 An unexpeced increase in he risk premium leads o higher ineres rae and weakening of he currency, accompanied by a fall in carry rade. We do no impose any resricions on prices, while, in boh cases, he conemporaneous effec of he shock on producion is zero, ha is GDP responds o hese shocks wih delay. The laer assumpion may receive some criicism as in small open economy producion can be sensiive o exchange rae movemens wihin he same quarer. In order o check o wha exen our conclusions depend on hese resricions, we esimae a model on monhly daa as well as wih a pure sign resricion approach. The resuls repored in Secion 5 confirm he main resuls of he benchmark model. We use he sandard sign resricions o idenify domesic demand and supply shocks. An unanicipaed posiive supply shock causes producion o increase and prices o fall, while a demand shock causes boh producion and prices o increase on impac. Demand shocks are associaed wih an increase in he ineres rae as moneary policy ries o counerac inflaion. Finally, we leave he fifh domesic shock unidenified. Besides domesic facors, foreign shocks may be imporan drivers of carry rade aciviy. 3 A firs glance i may seem conradicing o idenify he effec of moneary policy shocks on exchange rae and carry rade by imposing resricions on exchange rae and carry rade hemselves. Indeed, imposing sign resricion on he impac response and being compleely indifferen in he second period response may cause a bias agains hump-shaped response funcions. Sill, we hink ha his bias is no ha big as o influence significanly our resuls. Firsly, among our impulse responses here are several examples when a conemporaneous sign resricion is imposed, bu he resul is hump-shaped. Secondly, we esimaed he same model by imposing he sign resricions for 4 quarers and we go he same qualiaive resuls. We also esimaed i wihou imposing any resricions on carry rade. Again, he resuls are very similar. 5

6 Thus, we idenify foreign shocks as well. The corresponding resricions are similar o he domesic ones, and are described in deails in Secion Daa Due o he relaively shor ime series we prefer he panel approach o he counry-by-counry esimaions, similarly o Brunnermeier, Nagel, and Pedersen (29). Our panel consiss of hree developed counries (Ausralia, Canada and he Unied Kingdom) ha can be considered as arges of carry rade aciviy on our sample. 4 Our choice of his paricular group of counries was deermined primarily by he availabiliy of carry rade saisics. We have quarerly daa for he macroeconomic variables from 1992Q2 o 27Q4 aken from he Inernaional Financial Saisics (IFS) daabase. 5 In his way we leave ou he recen financial crisis from he sample, as we are ineresed in moneary ransmission and exchange rae dynamics in normal imes. 6 The saring period was chosen based on carry rade daa availabiliy. Anoher opion would be o include all he counries having long enough carry rade daa, like Japan and he Unied Saes. The reason for invesigaing only hese hree counries is ha pooling hem ogeher wih big, closed economies would quesion our seup as we assume ha he main dynamic properies of he vecor of variables are approximaely he same across counries. All GDP and CPI daa are seasonally adjused in he IFS daabase. However, Unied Kingdom CPI daa seemed o have some remained seasonaliy, herefore we correced for ha. 7 The end-of-period nominal exchange raes vis-à-vis he U.S. dollar are defined as he local currency price of one uni of foreign currency, hus an increase in he exchange rae means depreciaion. The ineres rae daa is he quarerly average of shor-erm money marke rae. To conrol for foreign shocks, we use U.S. GDP, CPI, ineres rae and exchange rae daa as exogenous in he VAR. U.S. dollar exchange rae vis-à-vis he euro is aken from Eurosa. An increase in he exchange rae means depreciaion of he dollar. Following Brunnermeier, Nagel, and Pedersen (29), we use he fuures posiion daa from he Commodiy Fuures Trading Commission (CFTC) as a proxy for carry rade aciviy. I is a widely used measure of speculaive posiions. We use he laes available daa for each quarer o consruc he ne fuures posiion of non-commercial raders in Ausralian Dollar (AUD), Canadian Dollar (CAD) and Briish Pound Serling (GBP), expressed as a fracion of oal open ineres. 8 According o Brunnermeier, Nagel, and Pedersen (29), despie is shorcomings, i is he bes publicly available daa for carry rade aciviy. 9 4 See he resuls of Brunnermeier, Nagel, and Pedersen (29). 5 See Table 1 in Appendix for deails. 6 The sample ends before he recen global financial crisis, due o he possibiliy of nonlinear effecs caused by he severe shocks ha may pose a bias o he esimaion of he (linear) VAR model. For a robusness check, we exended he esimaion of he baseline model o he 1992Q2-212Q2 period and found ha he main resuls qualiaively sill hold. 7 In 2Q3 he Ausralian Governmen inroduced a Goods and Services Tax, which resuls in a level shif in Ausralian CPI daa. Conrolling for his wih a dummy variable does no aler our resuls, herefore we use he original daa. 8 Classified by he CFTC, non-commercial raders use fuures for speculaive purposes and no for hedging agains currency risk. 9 One of he main deficiencies is ha i does no cover all speculaive exchange rae posiions as, for insance, hedge funds reporedly rade more in forward markes han in fuures markes. Oher proxies for carry rade aciviy also exis, bu none of hem seem o be more suiable enough o jusify a deviaion from he approach of Brunnermeier, Nagel, and Pedersen (29). Reurns of Exchange Traded Funds (ETFs) and Exchange Traded Noes (ETNs) are linked o carry-rade sraegies making hem appealing candidaes. Bu hey have he same 6

7 Of course, he sum of speculaive posiions repored o CFTC is only a fracion of oal open ineres. Hence, behaviour of CFTC carry rade does no necessarily apply o all ineres sensiive posiion-aking. Sill, if we find ha CFTC carry raders eliminae excess reurn quickly, we can safely assume ha here are no incenives o ake posiions by oher marke paricipans. A Bayesian VAR wih 4 lags is esimaed on quarerly frequency using he previously inroduced panel daa se. 1 Conemporaneous and one period lagged U.S. daa appear as exogenous variables. We use counry-specific inerceps in he VAR. Following Uhlig (25), we use fla prior for he VAR. The coefficiens are drawn from he poserior disribuion, which is a normal-inverse- Wishar disribuion parameerized by he OLS esimaes of coefficien and variance-covariance marices. Calculaion of poserior disribuions is made following Reppa (29). 2 draws saisfying he sign resricions have been generaed. In order o measure he failure of UIP, we calculae excess reurn impulse responses. We define (expeced) excess reurn as he sum of he ineres rae and he (expeced) appreciaion expressed in annual erms 11 : z = i 4(e +1 e ) (7) where (i ) is he (log) nominal ineres rae and (e ) is he (log) nominal exchange rae in period. If UIP holds condiionally afer a shock, a posiive ineres rae differenial is offse by he depreciaion of he domesic currency resuling in no excess reurn. In oher words, he condiional expecaion E z +p mus be zero for all p as long as UIP holds. The effec of he srucural shocks on excess reurn can be calculaed from he impulse responses of he domesic and U.S. ineres raes, and he exchange rae. 4 Resuls In his secion we discuss he empirical resuls obained from our preferred idenificaion scheme, and hen briefly compare our resuls wih he Cholesky decomposiion. I is followed by an analysis of he effec of foreign shocks. Finally, we presen variance decomposiion wih focus on he deerminans of carry rade aciviy. 4.1 The effec of moneary policy and risk premium shocks We are ineresed firs of all in he effec of moneary policy shocks. As menioned earlier, separaion of hem from risk premium shocks can be crucial. Therefore we focus here on hese shocks. Responses o all he idenified domesic shocks can be found in he Appendix (Figure 13). Figure 1 shows he esimaed impulse responses o a domesic conracionary moneary policy shock and an unfavourable risk premium shock, respecively, up o 5 years afer he shock. We repor he median, he 2.5h, 16h, 84h and 97.5h perceniles of he poserior disribuion. weakness as CFTC daa as ETFs and ETNs are mosly used by reail invesors and are unlikely o represen a large par of overall carry rade aciviy; and heir ime series sar only in mid-2s resuling in relaively shor sample periods. Anoher poenial proxy is he BIS inernaional banking saisics ha measure he amoun of cross-border lending, including a currency breakdown of banks inernaional asses and liabiliies. Unforunaely, banks repor only heir on-balance shee posiions, wihou explicily disinguishing beween carry rade posiions and oher aciviies, same problem as in case of CFTC fuures posiions daa. For more deails, see Curcuru, Vega, and Hoek (211). 1 Sandard selecion crieria sugges 1-2 lags for he VAR; however, we include 4 lags o be able o rejec serial correlaion in he residuals. 11 Since U.S. ineres rae is assumed o be exogenous and no affeced by domesic shocks, when calculaing excess reurn we can ignore i. 7

8 Figure 1: Responses o a moneary policy and a risk premium shock GDP o MONETARY.2.4 CPI o MONETARY.35 IR o MONETARY.5 ER o MONETARY CARRY o MONETARY GDP o RISK.5 CPI o RISK.35 IR o RISK.3 ER o RISK.15 CARRY o RISK Noe: The solid line is he poinwise median of all successful draws. Dashed and doed lines encompass he middle 68 and 95 per cen of he poserior disribuion, respecively. The impulse responses are inuiive, albei no always significanly differen from zero. A ypical moneary policy shock can be characerized by a 15 basis poins ineres rae increase, which hen sars decreasing, finally dropping slighly below is iniial level. The gradual wihdrawal of he iniial ighening sance reflecs some smoohing in he conduc of moneary policy, which is a well-known finding in he lieraure. The exchange rae appreciaes by 1.5 per cen on impac, which is followed by a gradual depreciaion owards is iniial level. Hence, he adjusmen of he exchange rae is insananeous wihou any delayed overshooing paern, in line wih he predicion of he Dornbusch (1976) model. We do no observe any significan price puzzle eiher; he price index sars declining in he second year afer he shock, bu he effec of he moneary conracion is no significan. Oupu declines quickly and o a saisically significan 12 exen, which is in line wih wha we expec afer a conracionary moneary policy shock. The fas GDP and slow CPI responses resembles he moneary ransmission in New Keynesian sicky price models. Shocks o risk premium increase shor-erm ineres rae and depreciaes he currency on impac, according o our idenifying assumpions. The effec of higher ineres rae and weaker currency on GDP is no significanly differen from zero. They affec domesic producion in differen ways: while he increase in ineres rae reduces domesic demand, he depreciaion makes 12 Since we adop a Bayesian approach, significan means ha large par of he poserior disribuion is below or above a cerain value. In his paricular case he lower 84 (i.e. he enire middle 68) percen of he impulse response poserior is below zero. 8

9 expor more compeiive. In he CPI response he exchange rae channel seems o dominae: domesic prices increase, presumably due o he weaker currency. The iniial drop in carry rade posiion is followed by a gradual recovery of risk appeie. In he second year afer he shock long speculaive posiions are significanly higher han originally. This can be explained by he higher ineres rae and he sill appreciaing exchange rae. The mos imporan resul is ha exchange rae and carry rade seem o reac quickly o moneary policy, and here is no sign of delayed overshooing or prolonged carry rade inflow. Our impulse responses are in favour of Dornbusch (1976) and conradic o Brunnermeier, Nagel, and Pedersen (29) and Planin and Shin (211). Since drawing conclusions abou he shape of impulse responses based on poinwise median can be misleading (Sims and Zha, 1999), we repor he poserior disribuion of he horizon when exchange rae and carry rade have heir maximum response. We calculae wo measures o describe he peak response. We call urning poin he earlies quarer when appreciaion urns o depreciaion. We call minimum he quarer where he exchange rae response has is minimum value over he 2 quarers horizon. These definiions apply o carry rade wih similar logic. The hisograms confirm our previous finding ha carry raders respond o moneary policy wihin he same quarer which resuls in a promp adjusmen of he exchange rae (Figure 2). According o he lef panel, he peaks of he impulse responses are in he firs period in mos of he cases. The firs peak mosly also coincides wih he exreme value of he impulse response, as shown in he righ panel. A more direc way o assess he role and incenives of carry rade is o quanify he realized reurn afer a shock. If he exchange rae appreciaes fas enough o an unexpeced rae hike by he cenral bank, he subsequen depreciaion can eliminae he excess reurn, which is he Figure 2: Poserior disribuion of he locaion of peak response o a moneary policy shock (as a percen of oal draws) 1 HISTOGRAM of EXCHANGE RATE TURNING POINT 1 HISTOGRAM of EXCHANGE RATE MINIMUM HISTOGRAM of CARRY TRADE TURNING POINT 1 HISTOGRAM of CARRY TRADE MAXIMUM

10 Figure 3: Excess reurn o moneary policy and risk premium shocks.5 EXCESS RETURN o MONETARY.5 EXCESS RETURN o RISK Noe: The solid line is he poinwise median of all successful draws. Dashed and doed lines encompass he middle 68 and 95 per cen of he poserior disribuion, respecively. logic of he uncovered ineres pariy heorem. The impulse response of (predicable) excess reurn suggess ha he reacion of he exchange rae is even sronger a bi han wha he UIP would imply (Figure 3). Righ afer he shock he excess reurn becomes slighly negaive, suggesing ha he rae a which he exchange rae depreciaes afer he quick appreciaion is a bi faser han he corresponding ineres rae differenial. In he case of risk premium shock he response of excess reurn is posiive for several quarers. This means ha he exchange rae depreciaes immediaely, and during he subsequen periods, ogeher wih he higher ineres rae, is gradual recovery offers an excess reurn o compensae for he lower risk appeie or he higher perceived risk. 4.2 Comparison wih he Cholesky decomposiion We briefly discuss he resuls wih he Cholesky decomposiion of he covariance marix wih a recursive ordering. This is a sandard saring poin in he lieraure sudying he effec of moneary policy shocks (see Bjørnland, 29; Uhlig, 25, among ohers). Besides, i allows us o highligh he main heoreical difference beween he recursive and he sign resricions approach. In he former case as long as he ineres rae is ordered before he exchange rae, which is usually he case i is (implicily) assumed ha an exchange rae or risk premium shock has no immediae effec on he ineres rae. However, cenral banks end o incorporae informaion abou he exchange rae ino heir decisions as well as any oher daa ha may influence he evoluion of he key variables like consumer prices or oupu gap. Therefore, we need o ake his channel ino accoun o properly idenify moneary policy shocks. Using he same VAR model, we calculae he impulse responses assuming a recursive srucure of shocks, oo. Our ordering is he following: GDP, CPI, ineres rae, exchange rae and carry rade. Here we idenify moneary policy shocks as an unexpeced increase in he ineres rae ha affecs GDP and CPI only wih delay. Noe again ha he recursive scheme implies ha moneary policy does no reac o he las wo shocks (exchange rae and carry rade) on impac. The resuls are displayed in Figure 4. Conrary o he findings in he benchmark model, he dynamic response of he exchange rae exhibis delayed overshooing, reaching is peak response nearly 2 years afer he shock. The sluggishness of he exchange rae response is comparable o wha Scholl and Uhlig (28) have found using sign resricions and somewha longer han in Bouakez and Normandin (21). I is also similar o he Cholesky decomposiion resuls of 1

11 Figure 4: Responses o a moneary policy (ineres rae) shock, Cholesky decomposiion.5 GDP o IR.2 CPI o IR.4 IR o IR.6 ER o IR.5 CARRY o IR Noe: The solid line is he poinwise median of all successful draws. Dashed and doed lines encompass he middle 68 and 95 per cen of he poserior disribuion, respecively. Bjørnland (29). Consisenly wih he delayed appreciaion, significan carry rade aciviy can be deeced even one year afer he shock. Our median impulse responses are quie similar o wha Brunnermeier, Nagel, and Pedersen (29) have found, which is line wih expecaions since hey also applied Cholesky decomposiion. The poserior disribuion of he peak exchange rae response as well as he excess reurn confirm ha he recursive idenificaion scheme does favour for prolonged UIP failure more han he sign resricion approach (Figure 5 and 6). Our conclusion is ha idenificaion based on Cholesky decomposiion may indeed generae delayed overshooing Idenificaion of foreign shocks Since foreign shocks may be imporan drivers of carry aciviy, idenificaion of hem is essenial for a horough variance decomposiion analysis. In he lieraure cied above, some aemps have been already made o disinguish domesic and foreign srucural shocks. Kim and Roubini (2), for insance, include he Federal Funds rae o conrol for foreign moneary policy, bu since hey do no idenify U.S. moneary policy shocks, is movemens may reflec oher srucural shocks as well. Anzuini and Fornari (212) use heir variables in erms of differences o he corresponding U.S. variables, so hey canno separaely idenify he effec of foreign shocks. We ake i for imporan o disinguish beween domesic and foreign shocks because even if hey may have similar shor-run effec on he differences, due o he asymmeric behaviour beween a small and a big counry, he medium and long-run effecs may differ a lo. A rivial example is a moneary policy shock. While in he small counry we expec he moneary policy o reac o he change in he foreign ineres rae, he same is no expeced from he cenral bank of he big counry. Thus, he implicaion on exchange rae and carry rade response may differ subsanially. In order o idenify foreign shocks, we esimae a srucural VAR separaely for he U.S. variables wih he same mehodology as in he domesic case. The VAR includes he same four U.S. variables used in he panel VAR as exogenous, wih 4 lags 14 on he same sample. The 13 I is noeworhy ha delayed overshooing is no a robus finding even wih Cholesky idenificaion. Using 2 lags in he VAR, he mode of peak responses wih recursive ordering akes place much earlier. This is in line wih Isrefi and Vonnak (212) who find ha Cholesky decomposiion does no always yield delayed overshooing. 14 The number of lags was seleced by looking a he usual informaion crieria and making sure ha he residuals are free of auocorrelaion. 11

12 Figure 5: Locaion of exchange rae peak response under various idenificaion schemes o a moneary policy (or ineres rae) shock (as a percen of oal draws) 1 Per cen Per cen Cholesky Sign resricion quarers Noe: The locaion where exchange rae impulse response has is minimum value are shown, see Figure 2. Figure 6: Response of excess reurn o a moneary policy shock using sign resricions (lef) and Cholesky decomposiion (righ).3 EXCESS RETURN o MONETARY.25 EXCESS RETURN o IR Noe: The solid line is he poinwise median of all successful draws. Dashed and doed lines encompass he middle 68 and 95 per cen of he poserior disribuion, respecively. 12

13 only difference is ha we did no include carry rade daa and exogenous variables. We idenify demand, supply, moneary policy and risk premium shocks using he same resricions as in he panel model presened before, obviously wihou he resricions on carry rade. In order o calculae he effec of foreign shocks o domesic variables, we randomly draw from he poserior of U.S. impulse responses for each draw from he panel VAR, and feed he former ino he laer hrough he exogenous U.S. variables. This way he shocks from he U.S. economy are channelled ino he small open economy VAR. Figure 14 in he Appendix depics he esimaed impulse responses of he U.S. VAR. Regarding he response of domesic variables o U.S. shocks, domesic ineres rae reacs posiively and he exchange rae depreciaes afer a conracionary U.S. moneary policy shock (Figure 15 in he Appendix). GDP and CPI do no show saisically significan responses, neiher he main variable of ineres, he carry posiion, alhough is immediae response is inuiive. Carry rade jumps o an U.S. risk premium shock significanly, bu he magniude is again much smaller han in he case of domesic shock. This suggess ha U.S. shocks have a minor role in carry rade aciviy. 4.4 Variance decomposiion Figure 7 shows he decomposiion of he variance of k-sep ahead forecas error of he carry rade. According o he median esimaes, domesic moneary policy and risk premium shocks explain more han 2-2 per cen of carry rade variabiliy over almos he whole 5-year horizon. Median esimaes are surrounded by large poserior uncerainy. The oher shocks seem o play only a minor role in carry rade. This is consisen wih he variance decomposiion of forecas error of he exchange rae (Figure 16 in he Appendix), where he explanaory power of domesic moneary policy and risk premium shocks is similarly high. 15 The median of unexplained variance remained less han 1 per cen a each horizon. I is worh menioning ha he role of U.S. shocks, including moneary policy shocks is of second order in explaining carry rade variaion. The main reason for i can be ha domesic moneary policy reacs o foreign shocks so ha ineres rae differenial does no change oo much, which discourages carry rade and hereby miigaes he exchange rae response. This inerpreaion is confirmed by he resuls as he poserior disribuion of he ineres rae differenial afer a U.S. moneary policy shock 16 is quie symmeric around zero a each horizon. On he oher hand, domesic moneary policy shocks are imporan for carry rade, because hey lead o persisen changes in ineres rae differenial. Noe again, ha in order o ge hese resuls boh domesic and foreign shocks have o be idenified. 5 Robusness analysis In his secion we es he robusness of he resuls o he idenificaion assumpions and he daa frequency. Boh ess are moivaed by he zero resricions we imposed in our benchmark model. As menioned earlier, he assumpion ha GDP and prices respond o moneary policy and risk premium shocks wih several monhs delay can be criicised in case of small open economies where he exchange rae channel is srong. 15 Anoher issue in recen lieraure is he connecion beween moneary policy shocks and exchange rae variaion. Our resuls show ha moneary policy shocks explain 2 per cen of exchange rae flucuaions a shorer horizon, while he conribuion is 5 per cen a longer horizon. This is broadly in line wih Scholl and Uhlig (28) bu smaller han wha Bouakez and Normandin (21) have repored. Kim and Roubini (2) have found much higher conribuion, around 6 per cen a shor horizon. 16 No shown in he paper, bu available upon reques. 13

14 Figure 7: Variance decomposiion of carry rade 1 Moneary 1 Risk 1 Demand 1 Supply 1 Unexplained US Moneary 1 US Risk 1 US Demand 1 US Supply Noe: The solid line is he poinwise median of all successful draws. Dashed and doed lines encompass he middle 68 and 95 per cen of he poserior disribuion, respecively. In he firs experimen we relax he zero resricions. frequency daa o make zero resricions more credible. In he second case we use higher 5.1 Pure sign resricion approach Firs, we consider a pure sign resricion approach as an alernaive idenificaion sraegy. More specifically, our resricions are he following in his case: ϵ Prod ϵ Prices + + u Moneary ϵ Ineres + + u Risk ϵ Exchange = u Demand (8) + + u Supply + ϵ Carry where he noaion is he same as in he benchmark model. Compared o he baseline specificaion, some addiional resricions are necessary o disenangle he shocks of ineres. Paricularly, we use sign resricions for he responses of GDP and prices, wih he excepion of he unresriced response of GDP o a risk premium shock (see he upper lef 2 2 marix). Furhermore, we assume ha he exchange rae appreciaes afer an unexpeced demand shock, and depreciaes following a supply shock, which is broadly in line wih sandard macroeconomic models. Impulse responses ha are in he cenre of our ineres do no aler significanly compared o he benchmark model (Figure 8). The price puzzle is now avoided by consrucion, as in Uhlig u 5 14

15 Figure 8: Responses o a moneary policy and a risk premium shock wih pure sign resricions.1 GDP o MONETARY.1 CPI o MONETARY.3 IR o MONETARY.1 ER o MONETARY CARRY o MONETARY GDP o RISK.5 CPI o RISK.3 IR o RISK.3 ER o RISK.15 CARRY o RISK Noe: The solid line is he poinwise median of all successful draws. Dashed and doed lines encompass he middle 68 and 95 per cen of he poserior disribuion, respecively. (25). No delayed overshooing of he exchange rae can be observed eiher in his case, and he responses of carry aciviy (which were unresriced) move in he direcion presened previously (Figure 9). Turning o he variance decomposiion, Figure 1 shows ha moneary policy and risk premium shocks explain less variance in carry aciviy, and larger explanaory power is aribued o oher domesic shocks compared o he benchmark resul. This can be a consequence of resricing he sign of he exchange rae response o oher shocks as well. 5.2 Monhly frequency To make he zero resricions more credible, we esimae he same model on monhly frequency. The assumpion ha he response of GDP and CPI o moneary and risk premium shocks is lagged by one monh is more defendable han he one quarer delay. We use monhly daa from 1992M4 o 27M12 and he VAR includes 3 lags of he endogenous variables. In he U.S. VAR we use 7 lags. 17 As GDP daa is no available on monhly frequency, we op for indusrial producion insead. The auhoriies of Ausralia do no publish monhly daa on consumer prices and economic aciviy, herefore we have o resric our panel sample o Canada and he Unied Kingdom. 17 The choice of lag numbers was moivaed by he rejecion of serial correlaion in he residuals. We esimaed an alernaive version wih 9 lags for boh panel and U.S. models as in his case no serial correlaion was deeced eiher. Resuls do no change significanly compared o he case described above. 15

16 Figure 9: Poserior disribuion of he locaion of peak response wih pure sign resricions o a moneary policy shock (as a percen of oal draws) 1 HISTOGRAM of EXCHANGE RATE TURNING POINT 1 HISTOGRAM of EXCHANGE RATE MINIMUM HISTOGRAM of CARRY TRADE TURNING POINT HISTOGRAM of CARRY TRADE MAXIMUM Figure 1: Variance decomposiion of carry rade wih pure sign resricions 1 Moneary 1 Risk 1 Demand 1 Supply 1 Unexplained US Moneary 1 US Risk 1 US Demand 1 US Supply Noe: The solid line is he poinwise median of all successful draws. Dashed and doed lines encompass he middle 68 and 95 per cen of he poserior disribuion, respecively. 16

17 Figure 11: Responses o a moneary policy and a risk premium shock wih monhly daa IPROD o MONETARY.4.2 CPI o MONETARY.5 IR o MONETARY.5 ER o MONETARY CARRY o MONETARY IPROD o RISK.25 CPI o RISK.4 IR o RISK.2 ER o RISK.5 CARRY o RISK Noe: The solid line is he poinwise median of all successful draws. Dashed and doed lines encompass he middle 68 and 95 per cen of he poserior disribuion, respecively. The resuls depiced in Figure 11 confirm ha he exchange rae and carry aciviy reac wihin a quarer (a mos 1-2 monhs) o a moneary policy shock, generally. Unforunaely, he responses of indusrial producion and consumer prices are no significanly differen from zero. Variance decomposiions lead o he same conclusion, wih moneary policy and risk premium remaining dominan in explaining he oal variance of carry aciviy. Our robusness checks confirm he main resuls of he benchmark model. However, heir impulse responses are less convincing in general han he original specificaion. Moreover, relaxing zero resricions requires addiional resricions, and changing he frequency decreases he sample. Taking all hese rade-offs ogeher, we find i reasonable o sick o he benchmark model. 6 Conclusions In our paper we invesigaed he effec of moneary policy on exchange rae and he role carry rade plays in he exchange rae channel of moneary ransmission wihin he same model. We esimaed a VAR for a panel of hree small open economies regarded as arge of carry rade sraegies. We idenified domesic and foreign srucural shocks by using sign resricions. We found ha allowing for simulaneous ineres rae and exchange rae reacions o boh moneary policy and risk premium shocks, he delayed overshooing found by oher auhors can 17

18 be rejeced by high probabiliy. The exchange rae behaves as prediced by uncovered ineres pariy. Our resul suggess ha speculaive posiion-aking plays an imporan role in i. Afer an unexpeced moneary policy shock carry raders reac promply helping he exchange rae jump and eliminae excess reurn, which may be an imporan conribuion o he lieraure. Variance decomposiion shows ha he main drivers of carry rade are domesic moneary policy and risk premium shocks. While in he firs case we aribue a beneficial role o currency speculaion in ransmiing moneary policy, in he second case he idiosyncraic exchange rae shocks generaed by carry rade aciviy may incur welfare losses. We esed he robusness of our resuls o he choice of resricions used o idenify he VAR, and o he daa frequency. Our main findings proved o be fairly robus, wih he excepion of variance decomposiion, which proved o be sensiive o idenifying resricions. References Anzuini, Alessio and Fabio Fornari (Aug. 212). Macroeconomic Deerminans of Carry Trade Aciviy. In: Review of Inernaional Economics 2.3, pp url: hp://ideas.re pec.org/a/bla/reviec/v2y212i3p hml. Bjørnland, Hilde C. (29). Moneary policy and exchange rae overshooing: Dornbusch was righ afer all. In: Journal of Inernaional Economics 79.1, pp url: hp://ideas. repec.org/a/eee/inecon/v79y29i1p64-77.hml. Bjørnland, Hilde C. and Jørn I. Halvorsen (Nov. 21). How does moneary policy respond o exchange rae movemens? New inernaional evidence. Working Papers 1. Cenre for Applied Macro- and Peroleum economics (CAMP), BI Norwegian Business School. url: hp://ideas.repec.org/p/bny/wpaper/1.hml. Bouakez, Hafedh and Michel Normandin (May 21). Flucuaions in he foreign exchange marke: How imporan are moneary policy shocks? In: Journal of Inernaional Economics 81.1, pp url: hp:// ideas. repec.org/ a/ eee/ inecon/v81y21i1p hml. Brunnermeier, Markus K., Sefan Nagel, and Lasse H. Pedersen (29). Carry Trades and Currency Crashes. In: NBER Macroeconomics Annual 28, Volume 23. NBER Chapers. Naional Bureau of Economic Research, Inc, pp url: hp://ideas.repec.org/ h/nbr/nberch/7286.hml. Curcuru, Sephanie, Clara Vega, and Jasper Hoek (211). Measuring carry rade aciviy. In: Proceedings of he IFC Conference on Iniiaives o address daa gaps revealed by he financial crisis, Basel, Augus 21. Ed. by Bank for Inernaional Selemens. Vol. 34. IFC Bulleins chapers. Bank for Inernaional Selemens, pp url: hp://ideas.re pec.org/h/bis/bisifc/34-31.hml. Cushman, David O. and Tao Zha (1997). Idenifying moneary policy in a small open economy under flexible exchange raes. In: Journal of Moneary Economics 39.3, pp url: hp://ideas.repec.org/a/eee/moneco/v39y1997i3p hml. Dornbusch, Rudiger (1976). Expecaions and exchange raes dynamics. In: Journal of Poliical Economy 84, pp Eichenbaum, Marin and Charles L. Evans (1995). Some Empirical Evidence on he Effecs of Shocks o Moneary Policy on Exchange Raes. In: The Quarerly Journal of Economics 11.4, pp Faus, Jon and John H. Rogers (23). Moneary policy s role in exchange rae behavior. In: Journal of Moneary Economics 5.7, pp url: hp://ideas.repec.org/a/ee e/moneco/v5y23i7p hml. 18

19 Isrefi, Klodiana and Balazs Vonnak (212). Delayed Overshooing Puzzle in Srucural Vecor Auoregression Models. mimeo. Goehe Universiä, Frankfur. Jarocinski, Marek (21). Responses o moneary policy shocks in he eas and he wes of Europe: a comparison. In: Journal of Applied Economerics 25.5, pp url: hp: //ideas.repec.org/a/jae/japme/v25y21i5p hml. Kilian, Luz (Aug. 211). Srucural Vecor Auoregressions. CEPR Discussion Papers C.E.P.R. Discussion Papers. url: hp://ideas.repec.org/p/cpr/ceprdp/8515.hml. Kim, Soyoung and Nouriel Roubini (2). Exchange rae anomalies in he indusrial counries: A soluion wih a srucural VAR approach. In: Journal of Moneary Economics 45.3, pp url: hp://ideas.repec.org/a/eee/moneco/v45y2i3p hml. Kisgergely, Kornél (212). Is here a carry rade channel of moneary policy in emerging counries? MNB Working Papers 212/3. Magyar Nemzei Bank (he cenral bank of Hungary). url: hp://ideas.repec.org/p/mnb/wpaper/212-3.hml. McCallum, Benne T. (1994). A reconsideraion of he uncovered ineres pariy relaionship. In: Journal of Moneary Economics 33.1, pp url: hp://ideas.repec.org/a/e ee/moneco/v33y1994i1p hml. Planin, Guillaume and Hyun Song Shin (Feb. 211). Carry Trades, Moneary Policy and Speculaive Dynamics. CEPR Discussion Papers C.E.P.R. Discussion Papers. url: hp: //ideas.repec.org/p/cpr/ceprdp/8224.hml. Reppa, Zolán (29). A join macroeconomic-yield curve model for Hungary. MNB Working Papers 29/1. Magyar Nemzei Bank (he cenral bank of Hungary). url: hp://ideas. repec.org/p/mnb/wpaper/29-1.hml. Scholl, Almuh and Harald Uhlig (28). New evidence on he puzzles: Resuls from agnosic idenificaion on moneary policy and exchange raes. In: Journal of Inernaional Economics 76.1, pp url: hp://ideas.repec.org/a/eee/inecon/v76y28i1p1-13.hml. Sims, Chrisopher A. and Tao Zha (1999). Error Bands for Impulse Responses. In: Economerica 67.5, pp url: hp://ideas.repec.org/a/ecm/emerp/v67y1999i5p hml. Uhlig, Harald (25). Wha are he effecs of moneary policy on oupu? Resuls from an agnosic idenificaion procedure. In: Journal of Moneary Economics 52.2, pp url: hp://ideas.repec.org/a/eee/moneco/v52y25i2p hml. Vonnák, Balázs (21). Risk premium shocks, moneary policy and exchange rae pass-hrough in he Czech Republic, Hungary and Poland. MNB Working Papers 21/1. Magyar Nemzei Bank (he cenral bank of Hungary). url: hp://ideas.repec.org/p/mnb/wpaper/21-1.hml. 19

20 Appendix Table 1: IFS daa Counry Variable IFS Code Ausralia NEER from INS 193..NECZF... Ausralia GDP vol. (25=1) 19399BVRZF... Ausralia CPI: all groups, six capials ZF... Ausralia Average rae on money marke 1936B..ZF... Canada NEER from INS 156..NECZF... Canada GDP vol. (25=1) 15699BVRZF... Canada CPI: all ciies pop. over 3, ZF... Canada Overnigh money marke rae 1566B..ZF... Unied Kingdom NEER from INS 112..NECZF... Unied Kingdom GDP vol. (25=1) 11299BVRZF... Unied Kingdom CPI: all iems ZF... Unied Kingdom Overnigh inerbank min. 1126B..ZF... 2