ECONOMETRIC TESTING OF UNCOVERED INTEREST RATE PARITY IN SERBIA

Transcription

1 ECONOMIC ANNALS, Volume LXIII, No. 216 / January March 2018 UDC: 3.33 ISSN: hps://doi.org/ /eka m Zorica Mladenović* Jelena Rašković** ECONOMETRIC TESTING OF UNCOVERED INTEREST RATE PARITY IN SERBIA ABSTRACT: This paper provides economeric evidence of he ineres pariy puzzle in Serbia over he period Economeric findings are derived from he following echniques: long-run parameer esimaion based on he auoregressive disribued lag model, impulse response funcion compued from he bivariae vecor auoregressive model, and esimaion of he wo-regime Markov swiching parameer model. Our resuls indicae ha a posiive ineres differenial correced for counry risk leads o significan dinar appreciaion agains he euro. The inensiy of his impac is differen across sub-periods of low exchange rae variabiliy and high variabiliy. Exchange rae movemens are found o appreciae more srongly during lower variabiliy episodes. Preliminary economeric invesigaion of four oher European emerging economies documens similar findings only for Romania. Our resuls sugges ha here is a huge incenive for shorerm carry rades in Serbia, regardless of subsanial risks. KEY WORDS: Uncovered ineres rae pariy, Counry risk, Ineres pariy puzzle, ARDL bounds es, VAR model, Regime swiching. JEL CLASSIFICATION: F31; F41; E43; C32 * Full Professor, Faculy of Economics, Universiy of Belgrade, zvujos@ekof.bg.ac.rs ** PhD Suden, Teaching Assisan, Faculy of Economics, Universiy of Belgrade, raskovicj@ekof.bg.ac.rs 35

2 Economic Annals, Volume LXIII, No. 216 / January March INTRODUCTION In recen years, financial inegraion has become more visible and global capial flows have become more imporan. On one hand a growing body of evidence suggess ha financial openness in developing counries enhances growh, while on he oher hand his increases a counry s vulnerabiliy during financial crisis and he possibiliy of speculaive aacks. Emerging economies ineres raes are usually high and are poenially aracive o invesors looking o make a profi. Remarkably, mos of hese economies are characerized by underdeveloped insiuions, weak macroeconomic performance, poliical insabiliy, and volaile economic condiions. These cause higher risk premiums for boh currency risk and counry risk. Despie he high risk, such markes are aracive o foreign invesors for shor-run ransacions. Therefore, hese counries use governmen resricions on capial accoun ransacions exensively, which are associaed wih complex policies regarding capial flows across naional borders. The main objecive of his paper is o economerically idenify he opporuniy o gain arbirage profi in Serbia by modelling uncovered ineres rae pariy (UIP). Because Serbia has had very high ineres differenials, even when correced for counry risk (6.23% on average in he pas 10 years), i could be a poenial arge for speculaive capial. We consider changes in he average monhly exchange rae beween he Serbian dinar and he euro (FX changes) and he average money marke monhly ineres rae differenial for he period Sepember 2005 o December The sample period is chosen by daa availabiliy. Because expeced exchange raes are no available ime series, we es he ex pos UIP condiion ha assumes raional expecaions. Some auhors (Rojas-Suarez and Soelo 2007; Ferreira 2009) emphasize he significance of counry defaul risk (or counry risk) in emerging economies and find a srong relaionship beween counry defaul risk and ineres raes. Hence, we incorporae counry risk a priori in he model using he Emerging Marke Bond Index (EMBI). The key economeric resuls are derived from bounds-esing mehodology wihin he auoregressive disribued lag (ARDL) model advanced by Pesaran e al. (2001). This provides a framework for examining he exisence of a long-run relaionship. This approach is suiable for boh saionary and non-saionary daa, and for a mixure of saionary and uni roo ime series. Our empirical findings suppor he exisence of he ineres pariy puzzle in Serbia. We also 36

3 ECONOMETRIC TESTING OF UIRP IN SERBIA employ he vecor auoregressive (VAR) model as an addiional ool o reassess he relaionship beween FX and ineres rae differenials. Based on he compuaion of he impulse response funcion, we explore how sensiive variables are o unexpeced random shocks in he sysem hroughou ime. Finally, o accoun for he possibiliy ha UIP model parameers may be ime varying we use he Markov-swiching (MS) model (Hamilon 1989), assuming ha hey change randomly across differen regimes. Such a framework may link swiches in parameers o sub-periods of currency depreciaion and appreciaion. This sudy conribues o economeric lieraure on he Serbian economy in wo ways. Firs, previously a priori counry risk has no been considered par of he UIP model. The explanaory power of he model is increased by he inclusion of counry risk, which makes saisical inference more reliable. Second, he differen economeric ime series echniques implemened improve he undersanding of he naure of his relaionship. The res of he paper is organized as follows. Secion 2 presens a lieraure review. Secion 3 gives an overview of he heory of ineres rae pariy. Secion 4 describes he ime series of our sample and briefly discusses he mehodological framework. Economeric resuls for Serbia are given in Secion 5. Preliminary resuls for some oher emerging economies (Czech Republic, Hungary, Poland, and Romania) obained by similar mehodology are discussed and compared wih Serbia in Secion 6. Concluding remarks are offered in Secion LITERATURE REVIEW There are many papers in he lieraure ha invesigae he condiions for UIP, mos of hem referring o developed counries. UIP has become a favourie heoreical absracion ha is rejeced by he daa. In he lieraure i is recognized as an ineres pariy puzzle (or forward premium puzzle). The ineres pariy puzzle finds ha over he shor-erm horizon (one week o a quarer) he posiive ineres differenial is associaed wih an appreciaing currency, raher han a depreciaing currency as UIP predics. Fama (1994) and Froo and Thaler (1990) provide early evidence of his rejecion. One of he possible explanaions of he puzzle is relaed o he ime-varying risk premium (Engel 1996). Engel (2015) finds ha he relaively higher real ineres rae srengh of a currency is even greaer han prediced by he heory of 37

4 Economic Annals, Volume LXIII, No. 216 / January March 2018 raional expecaions of fuure shor-erm nominal ineres. Based on Fama regression and he vecor equilibrium correcion model, Engel produces evidence of risk reversal over ime (higher risk in he shor run, and lower in he long run) based on monhly panel daa for G7 counries relaive o he US dollar over he period June 1979 Ocober Risk reversal is explained by nonpecuniary liquidiy reurn on asses. Ismailov and Rossi (2017) inroduce a new exchange rae uncerainy index ino he model and show ha UIP fails in five indusrialized counries (Canada, Japan, Europe, Swizerland, UK) relaive o he US dollar during periods of high uncerainy, bu i holds a imes when uncerainy is low. They es he ex ane UIP relaionship based on monhly daa spanning November 1993 o January 2015, considering regression where he consan and slope parameer migh be ime-varying. The second explanaion is based on exchange rae expecaions. The vas majoriy of he papers assume raional expecaions. Juselius and Assenmacher (2017) emphasize ha mos of he puzzle vanishes if an uncerainy premium (proxied by he persisen PPP gap) in he foreign exchange marke is inroduced, which suppors imperfec knowledge-based expecaions insead of raional expecaions. This evidence is he resul of applying he coinegraed VAR model o Swiss-US pariy for monhly daa. Ter Ellen e al. (2013) es heerogeneous expecaions using a daase of survey expecaions for four exchange raes ($/, $/, /, /$) on foreign exchange markes for he period January 2003 February 2008, and show ha in he shor run (1 and 3 monhs), posiive ineres differenials lead o a carry rades sraegy, while in he long run (12 monhs) he expecaions are in line wih UIP heory. By applying he srucural VAR model in hree small open economies (Ausralia, Canada, he UK) and using money marke quarerly daa ( ), Felcser and Vonnak (2014) show ha an unexpeced shock in he ineres rae affecs exchange rae appreciaion and carry-rade movemens. Brunnermeier e al. (2008) arrive a he same resuls. Also, a poenial reason for he puzzle could be he presence of a differen group of invesors in he marke and heir ineracion (Froo and Frankel 1990; Bacchea and van Wincoop 2010). McCallum (1994) considers cenral bank inervenion regarding he UIP condiion for hree exchange raes ($/DM, $/, and $/ ). I is argued ha empirical applicaions of UIP (OLS esimaes) give a negaive slope esimae due o moneary auhoriies reacing o he movemen in he exchange rae by seing a policy ineres rae. In his way, all iems (ineres differenial and expeced depreciaion) in he UIP condiion are deermined by he cenral bank. 38

5 ECONOMETRIC TESTING OF UIRP IN SERBIA Addiionally, Chinn and Meredih (2004) expand McCallum s model by including inervenion oupu and inflaion for G7 counries in he funcion of he cenral bank. They conclude ha deviaions in UIP are primarily he resul of he cenral bank reacion o shocks in exchange raes in he shor run (1 monh). Similarly, Lohian and Wu (2012), using French francs and US dollars o he pound serling, show ha in periods of excepionally high inflaion (lae 1970s 1980s) he regression slope esimaes become negaive. Lohian (2016) assesses he UIP relaionship over an exremely long ime horizon (wo cenuries) using panel esimaion for 16 developed counries (Ausralia, Belgium, Canada, Denmark, Finland, France, Germany, Ialy, Japan, Neherlands, New Zealand, Norway, Spain, Sweden, Swizerland, Unied Kingdom) and based on annual long-erm bond yields. The resuls confirm he UIP condiion ha i holds for a longer ime horizon, in line wih previous lieraure (Alexius 2001; Chin 2006). The empirical evidence for emerging economies is less frequen. Since he UIP condiion holds under he assumpion of full capial mobiliy, low ransacion coss, and risk neuraliy (idenical asses in erms of defaul risk), i is expeced ha all hese characerisics conribue o larger UIP deviaions in emerging economies han in developed economies. However, empirical findings do no suppor his. Some auhors find ha UIP raher holds for counries wih higher inflaion raes and higher volailiy because high inflaion raes are followed by large exchange rae depreciaion (pass-hrough effec) and high ineres differenials (Bansal and Dahlquis 2000; Bacchea and Wincoop 2006; Flood and Rose 2002). Ferreira (2004) shows ha moneary auhoriies inervenions deermine boh expeced exchange raes and he ineres differenial, which induce a simulaneiy bias on he ineres differenial and expeced exchange rae for observed emerging markes. The model proposed by McCallum (1994) is used o es his for five emerging markes (Argenina, Brasil, Chile, Mexico and Turkey) from Cavoli and Rajan (2006) analyse his phenomenon for Eas Asian counries from January 1990 o May 1997 and show ha large capial inflows in hose counries have a negligible effec on UIP deviaion due o he parial capial mobiliy ha leads moneary auhoriies o serilize subsanial pars of ha inflow. Also, Io and Chinn (2007), based on panel esimaes for 21 indusrial counries ( ), find ha capial accoun openness and greaer financial developmen decrease he wide range of deviaions, while greaer inflaion volailiy and higher per capia income increase he size of UIP deviaions. 39

6 Economic Annals, Volume LXIII, No. 216 / January March 2018 Only a few papers invesigae he UIP condiion for Cenral and Eas European (CEE) counries. In some of hem hese counries are included as par of a wider group of emerging markes, so ha he overall resuls are very general. Sarmidi and Salleh (2011) pu Lain America counries, Asian emerging markes, and CEE counries ino he panel and repor ha he UIP condiion holds beer for a longer ime horizon (12 monhs) han for shorer periods (1 and 3 monhs). However, for all ime horizons hey find exchange rae depreciaion, raher han appreciaion as appears in he puzzle. The dynamic naure of he UIP condiion has also been examined. By comparing he monhly daa of 18 indusrialized economies and 25 emerging markes (including CEE counries) and following he linear regression approach, Burnside (2014) confirms more UIP failure for indusrialized counries han for emerging economies. Alhough carry rade profiabiliy is demonsraed for boh groups, he ime varying risk premium only explains he reurns in indusrialized counries. In emerging markes he reurns mainly address he high ineres rae differenials. Vasilyev e al. (2017) show ha UIP holds beer for Russia han for oher emerging economies for monhly daa, which is explained by accumulaion of foreign exchange reserves. Addiionally, Burnside s (2014) resuls are confirmed, based on panel daa for 10 advanced and 15 emerging economies. Triandufal and Richer (2012) rejec he UIP condiion for five CEE counries (Bulgaria, Czech Republic, Hungary, Poland, and Romania) using he GARCH model based on money marke daa over he period. Filipozzi and Saehr (2012) apply rolling regressions o five CEE counries (Bulgaria, Czech Republic, Hungary, Poland, and Romania) using hree-monh daa and ge negaive esimaes of slope parameers for all counries excep Romania. Jiang e al. (2013) and Cuesas e al. (2015) explore wheher UIP deviaion is a saionary process for each counry using money marke monhly daa for ineres raes and ex pos exchange raes. Based on he hreshold auoregressive model, Jiang e al. (2013) conclude for seven CEE counries (Bulgaria, Croaia, Czech Republic, Hungary, Poland, Romania, Russia) ha UIP deviaion is a non-linear saionary process, which confirms he heory; however, he uni roo was found for Belarus, Lavia, and Macedonia. Cuesas e al. (2015) apply several uni roo ess (wih and wihou srucural break analysis) for a similar sample and confirm saionariy wih consan erms in mos economies. 40

7 ECONOMETRIC TESTING OF UIRP IN SERBIA 3. UNCOVERED INTEREST RATE PARITY (UIP): OVERVIEW The relaionship beween ineres raes and he forward exchange rae is defined as covered ineres rae pariy (Levi 2005, Ch. 8). In a world wihou resricions on capial flows or ransacion coss and wih profi-oriened marke players, in equilibrium Covered Ineres rae Pariy (CIP) would hold: 1 i 1 i * F S (1) where i and i * denoe he ineres rae beween ime and +1 on domesic and foreign asses respecively, S is he spo exchange rae a ime (domesic currency price of foreign currency), and F is forward exchange rae (domesic currency price of foreign exchange delivered a +1). This would hold due o invesors (maximizing profi agens), who always monior his relaionship, rying o exploi mispricing beween hese four key variables and gaining riskless profi. I applies o counries a he same level of developmen. The CIP does no hold for foreign exchange raes beween an emerging marke economy and a developed economy due o he counry defaul risk premium (counry risk). For example, he ineres differenial beween he dinar ineres rae in Serbia and he euro ineres rae in Germany does no equal he forward exchange rae. I mus be correced for counry risk. Acually he appropriae forward exchange rae in Serbia is deermined by he ineres differenial beween he ineres rae on RSD bonds and he ineres rae on EUR bonds, boh issued by he Serbian governmen. Assuming ha he forward exchange rae equals he expeced fuure exchange rae (F =S e +1), we end up wih he Uncovered Ineres rae Pariy (UIP) condiion ha mus hold in equilibrium: e 1 i S 1 1 i * S (2) e where S 1 is expeced exchange rae a ime +1. Conrary o CIP, UIP does no consider invesors who lock in fuure exchange raes on he forward exchange marke, bu invesors who buy foreign asses and expose hemselves o 41

8 Economic Annals, Volume LXIII, No. 216 / January March 2018 he risk ha he money earned could be squeezed ou by unexpeced movemens in he exchange raes. Taking he logarihm of (2), we ge: s s i i e * 1 (3) where he small case leers denoe logs. There is anoher way o express he ineres rae differenial, which is imporan for emerging marke economies (f denoes log value of F ): i i * * e e ( i i ) ( f s ) f s s s 1 1 (4) If we rearrange he relaionship in (4), we ge a modified UIP equaion: e * * e s i i i i ) ( f s f s s 1 ( ) 1 (5) Counry risk Currency risk We may observe ha in (5) he ineres rae differenial is correced by wo erms, denoed respecively as counry risk and currency (exchange rae) risk. As menioned above, if we consider wo developed counries, hen CIP should always hold. However, in emerging markes here is a resounding risk associaed wih he issuer of he asse. This risk is known as counry risk. The risk associaed wih variabiliy of he currency iself in erms of anoher currency is recognized as currency risk. Fama (1984) highlighs ha he forward exchange rae conrac mus conain a premium (F =S e +1+RP, where RP is a risk premium). Throughou he lieraure i is ofen idenified as he main cause of UIP rejecion, primarily for developed counries. Given ha he ime series of expeced exchange raes are unavailable in emerging economies, he ex pos UIP condiion is esed: * * s s i i i i ) ( f s f s 1 ( ) 1 (6) 42

9 ECONOMETRIC TESTING OF UIRP IN SERBIA The subsiuion of ex ane exchange raes wih ex pos exchange raes is due o he assumpion of raional expecaions. Finally, he empirical model considered in his paper akes he following form: s 1 s * i i CR (7) where and are parameers, is an error erm, and CR is he counry risk. Currency risk is omied from he model because ime series of he expeced FX movemens and forward raes are unavailable. However, counry risk is included and subraced from he ineres differenials. If i is posiive, hen he domesic counry is higher risk han he foreign counry, and vice versa. Mos previous sudies have no included he counry risk. The following hypohesis is commonly esed in he empirical lieraure: H 0 : α = 0, β = 1. As already menioned, parameer β is ofen esimaed o be negaive, which is defined in he lieraure as he ineres pariy puzzle or forward premium puzzle. 4. DATA DESCRIPTION AND OVERVIEW OF THE METHODOLOGY This sudy is based on average money marke ineres raes and average monhly changes in exchange raes, since financial insiuions underake much of he arbirage raher han individual invesors. The ineres raes of Serbian governmen bonds could also be used, bu due o he shor ime series and low liquidiy he resuls migh be spurious. As a measure of ineres raes we ake he money marke rae BELIBOR in Serbia, and EURIBOR for EMU. The exchange rae is represened by he average monhly price of he euro in unis of dinars. This daa is aken from he Naional Bank of Serbia websie and he Eurosa daabase. We op for he Emerging Marke Bond Index (EMBI) as a proxy for counry risk. EMBI is spread beween inernaional governmen bonds issued by he Serbian governmen in dollars and US Treasury Bonds. This is no he mos appropriae measure, given ha deb in he index is more han one year o mauriy, while we consider he UIP condiion over a one-monh horizon. However, his incorporaes he relaively high volailiy of he Serbian economy in he modelling. The daa is obained from Bloomberg. The sample period is 43

10 Economic Annals, Volume LXIII, No. 216 / January March 2018 defined by availabiliy of EMBI index daa and covers he period from Sepember 2005 o December Exchange rae movemens, he ineres rae differenial, he EMBI index, and he ineres rae differenial correced by he EMBI are shown in Figures 1 3. Figure 1. Monhly Exchange Rae Movemens (RSD/EUR) FX movemen 2005M09=100 Log of FX changes Figure 1 shows a disinc upward rend of he nominal exchange rae, especially afer he financial crisis in 2008 when he Serbian dinar experienced sharp depreciaion (abou 25% in he six monhs from July 2008 o February 2009). Figure 2. Ineres Rae Differenial (BELIBOR 1M EURIBOR 1M) The highes ineres rae (Fig. 2) is recorded a he beginning of 2006 (over 20%). This is he resul of he Serbian moneary auhoriy s endency o curb inflaion hrough changing he policy rae (he reference rae), direcly affecing money marke ineres, since hese raes appear in inerbank markes. The inflaion rae in 2005 was 16.1%. 44

11 ECONOMETRIC TESTING OF UIRP IN SERBIA Figure 3. EMBI Index and Ineres Rae Differenial (IRD) correced by EMBI EMBI IRD-EMBI The EMBI index (Fig. 3) also reached a peak of above 12% during he financial crisis. Basic descripive saisics are presened in Table 1. We can observe ha Serbia had very aracive ineres raes during he sample period, even correced for counry risk, and could have been a poenial arge for carry rade invesors. The average ineres rae differenial correced for counry risk was 6.23%, ranging beween 0.27% and as much as 18.68%. Table 1. Descripive saisics of considered ime series RSD/EUR i i* EMBI (CR) i i* CR Mean % 3.77% 6.23% Minimum % 1.45% 0.27% Maximum % 12.24% 18.68% Sd. Deviaion % 1.78% 3.98% To es he validiy of he UIP model he following economeric approach is employed. Firs, we implemen several uni roo ess o find ou if he ime series are saionary. Specificaion (7) is a valid regression equaion only if boh he variables on eiher side of he equaion are saionary or non-saionary, bu coinegraed. In he second sep we apply bounds esing mehodology based on he ARDL model suggesed by Pesaran and Shin (1999) and Pesaran e al. (2001). The approach is flexible so ha he long-run relaionship (level relaionship) among ime series of a differen order of inegraion can be considered. Also, differen shor-erm dynamics across variables are allowed. If 45

12 Economic Annals, Volume LXIII, No. 216 / January March 2018 a long-run relaionship exiss, hen he ARDL model has an adequae equilibrium correcion form (ECM) ha provides informaion abou long- and shor-run adjusmens. The baseline ARDL(k+1,q+1) model can be reformulaed as an unresriced condiional ECM specificaion: (8) where y denoes he FX changes and x represens he ineres rae differenial correced for counry risk. Δ is he firs difference operaor and is an error erm. Afer verifying he assumpions of his model (sabiliy and approximaion of an error erm by he Gaussian whie noise process), he bounds esing procedure is applied. Firs, as in he convenional ECM model, he F-es is used o es for he absence of a level relaionship beween he variables (H 0 : θ 0 =θ 1 =0). A null hypohesis rejecion implies he exisence of a level relaionship beween FX changes and ineres rae differenials correced for counry risk. As he disribuion of he es saisic is non-sandard and criical values are no available for a mixure of I(0) and I(1) ime series, Pesaran e al. (2001) provide lower and upper asympoic criical value bounds for he disribuion of he F- saisic for a differen number of variables. The consan erm is resriced o be a par of he level relaionship. Corresponding criical values a he 5% significance level are given in Table 2. Table 2. Asympoic criical value bounds for F-saisics (resriced inercep and no rend, Pesaran e al. 2001) Bound F-es I(0) 3.62 I(1) 4.16 If he compued F-saisic is below he lower bound criical value, i implies ha all variables are I(0) and hence here is no level relaionship. If he F-saisic is above he upper bound criical value a level relaionship exiss. If he F-saisic is in beween criical values he es is inconclusive. Assuming ha he bounds 46

13 ECONOMETRIC TESTING OF UIRP IN SERBIA es implies long-run exisence, i can be exraced from equaion (8) given ha β = - (θ 1 /θ 0 ) holds for he slope parameer. The hird sep esimaes he sandard VAR model. I consiss of our wo key ime series, bu one purely exogenous variable is also included (moneary inervenion). This model esimaes he impulse response funcion so ha he impac of unanicipaed random shocks on FX changes and ineres rae differenials can be raced over ime. Finally, he MS approach is implemened o allow for changes in he parameers of he UIP model, assuming ha inercep, slope, and error-erm variabiliy differ beween wo regimes. The specificaion akes he following form (similar versions are used, for example, in Engel and Hamilon 1990; Bekaer and Hodrick 1993; and Ichiue and Koyama 2011): y α 0 α1 P 0 1 P x h0 h1 P (9) P is he unobserved random variable ha follows a firs-order Markov chain defined by ransiion probabiliies beween wo saes. The full marix of ransiion probabiliies for wo saes reads as follows: Sae a +1 Condiion a P =0 P =1 P +1 =0 q=p 0/0 f=p 0/1 P +1 =1 p 1/0 p 1/1 Noe: Probabiliies in each column sum o 1. Shifs of he economy from sae 0 o sae 1 are governed by he inroduced random variable P. Under his specificaion we have wo differen regimes: regime 0 (i.e., P =0) and regime 1 (i.e., P =1). Index 0 is associaed wih he parameers of regime 0. The parameers α 1, β 1, h 1 capure he changes in he inercep, slope parameer, and random erm variabiliy respecively during regime 1 relaive o regime 0. Such a specificaion enables a more deailed analysis of UIP relaionship (7) because i may deec a differen reacion of FX changes ha could also depend on he level and variabiliy of FX changes. 47

14 Economic Annals, Volume LXIII, No. 216 / January March EMPIRICAL RESULTS To find ou if a uni roo is presen in he ime series we have applied he following uni roo ess: Augmened Dickey-Fuller (ADF), Kwiakowski- Phillips-Schmid-Shin (KPSS), Phillips-Perron (PP), and Ellio-Rohenberg- Sock (ERS). Resuls are shown in Table 3. We are able o clearly rejec he uni roo null hypohesis for s +1 -s, bu no for i i * - CR. Namely, while he ADF and he ERS ess indicae ha he ineres differenial is a saionary series, he applicaion of he KPSS and he PP ess implies a uni roo presence. Table 3. Resuls of uni roo esing Level Variable ADF KPSS PP ERS s +1 -s i i * - CR Noes: The number of lags included o ake care of he auocorrelaion is 0 in FX changes and 3 in ineres rae differenial correced for counry risk. I is deermined by he SC crierion. Consan and rend are used as deerminisic componens. The 5% criical values are: 3.44, and 3.00 for ADF (PP), KPSS, and ERS ess respecively. Nex, we proceed wih he selecion of he appropriae ARDL model. Using he crierion of he lowes value of he Akaike informaion crierion (AIC) saring wih a maximum of 12 lags, he opimal lags in he ARDL model for FX changes and he ineres differenial are chosen o be 5 and 1 respecively (ARDL(5,1)). From here, we formulae he condiional ECM model, bu wihou imposing any resricion on heir coefficiens. The resuls of he bounds ess clearly indicae he presence of a level relaionship (Table 4), wih esimaes given in Table 5. 48

15 ECONOMETRIC TESTING OF UIRP IN SERBIA Table 4. Bounds Tesing Resuls Compued value of F-saisic The 5% criical value I(0) I(1) Table 5. Coefficien esimaes of level equaion Coefficien Esimae Sandard error α 0.007** β 1.354** Noe: The form of level equaion is presened in equaion (7). The symbol ** denoes significance a he 0.05 level. Esimaes of boh α and β are significan a he 5% significance level. The consan α is close o zero (0.007) and he β coefficien is negaive ( 1.354). Finally, we end up wih he usual ECM in Table 6: Table 6. Equilibrium correcion form of he ARDL(5,1) UIP equaion Regressor Coefficien Sandard error p-value z NA Δy Δy Δy Δy Δx Box-Ljung Q(8)=6.48(0.59), Q(12)=10.15(0.60), Jarque-Bera JB =0.25(0.88), Whie WH=10.12(0.34) Noes: This regression is based on he condiional ECM model given in equaion (8) using he ARDL(5,1) specificaion where he dependen variable y is defined as s+1 - s and x as i - i*-cr. z-1 is an equilibrium correcion erm lagged one period ha is based on esimaes in Table 5. Addiionally, wo dummy variables are included: impulse dummy o ake only non-zero value 1 for Ocober 2008, and ransiory dummy wih non-zero values 1 and 1 for December 2008 and January 2009 respecively. The model performs well saisically, as confirmed by several misspecificaion ess. 49

16 Economic Annals, Volume LXIII, No. 216 / January March 2018 The same approach has been employed for he sysem of variables ha do no include informaion abou counry risk. However, he resuls indicae an insignifican level influence of ineres rae differenials on FX changes, while he esimaed uncondiional ECM model has lower explanaory power han he corresponding model in Table 6. The obained resuls correspond o oher findings in he lieraure on he ineres pariy puzzle (forward premium puzzle) over a one-monh horizon. A posiive ineres rae differenial correced for counry risk affecs dinar appreciaion agains he euro, no depreciaion as UIP predics. In fac, a high ineres rae differenial correced for counry risk leaves space for excessive reurns in dinars compared o he euro. In spie of individual invesors no being able o reach hese reurns because capial conrols were imposed, foreign financial insiuions were permied o perform carry rades. This affeced he dinar appreciaion. Our resul is in line wih he findings in he lieraure (Burnside 2014; Ter Ellen e al. 2013; Felcser and Vonnak 2014; Brunnermeier e al. 2008) ha in he shor-run, posiive ineres differenials lead o a carry rade sraegy. Alhough hese findings are mosly based on daa from developed counries, our evidence poins o he huge incenive for shor-run carry rades in Serbia, regardless of he resounding risks. Burnside (2014) emphasizes ha a ime-varying risk premium only explains he reurns in indusrialized counries, whereas in emerging markes he reurns are mainly he resul of he high ineres rae differenials. In he nex sage of our analysis he dynamic relaionship is revealed using he VAR model of FX changes and ineres rae differenials ha includes exchange rae inervenion as an exogenous variable (daa is from he Naional Bank of Serbia websie). Due o a lack of reliable daa on exchange rae inervenion, he analysis is based on a sample ha sars in January A VAR model of order 4 is chosen following he general-o-specific approach. The model performs well saisically, as confirmed in Table 7. The model is sable because all roos of he companion marix are less han one in he modulus (Table 8). The resuls of he Granger causaliy es show here is only one-way significan causaliy, running from ineres differenials correced for counry risk oward 50

17 FX changes, which is in line wih he UIP equaion. Namely, he null hypohesis ha he ineres rae differenial correced for counry risk does no Grangercause FX differenial is srongly rejeced (24.89 wih p-value = 0.0). Granger causaliy in he opposie direcion is no deeced (6.90 wih p-value = 0.23). Table 7. Mulivariae es saisics Tes for Value D. of p-value freedom Auocorrelaion of order Auocorrelaion of order Normaliy Heeroskedasiciy Noes: Auocorrelaion is esed by he mulivariae version of he Box-Ljung es. Normaliy is assessed by he mulivariae version of he Doornik-Hansen es. Heeroskedasiciy is esed by he mulivariae ype of he Whie es. The esimaed VAR model conains hree dummy variables ha are included o ake care of several ouliers. Two of hem are already defined in he Noe o Table 6. The hird one is an impulse dummy variable designed o ake only non-zero value 1 for 2008M3. Table 8. Roos of he companion marix, in modulus ECONOMETRIC TESTING OF UIRP IN SERBIA Roo 1 Roo 2 Roo 3 Roo 4 Roo 5 Roo 6 Roo 7 Roo The findings of he Granger-causaliy es indicae how ordering for he Cholesky decomposiion of he residual covariance marix should be chosen. Therefore, he impulse response funcions are compued based on he ordering: FX changes ineres rae differenial. The esimaes from ordinary impulse response funcions are depiced in Figure 4, whereas accumulaed esimaes are ploed in Figure 5. In his way we race he reacion of he given variable from he sysem o one sandard deviaion shock in he oher variable. These esimaes are dynamic such ha he variable reacion can be followed hrough differen monhs. Unexpeced shocks o ineres rae differenials correced for counry risk (in unis of one sandard deviaion) cause dinar appreciaion wih one lag, which coninues in he following four monhs. The esimaes appear o be precise, 51

18 Economic Annals, Volume LXIII, No. 216 / January March 2018 given he narrow confidence inerval bands. On he oher hand, unanicipaed shocks in FX changes seem o be negligible for ineres rae differenial responses. As confidence bounds are relaively wide, hese esimaes are insufficienly reliable. The Serbian economy experienced relaively high inflaion raes and inflaion volailiy over he firs half of he period considered (10% on average), which forced moneary auhoriies o increase policy raes o moderae demand. The policy raes referred o reference raes, which direcly affec money marke ineres raes. Hence, high inflaion raes were one of he main sources of he unexpeced shocks in he ineres differenial. Moreover, he very aracive policy raes influenced financial insiuions (mainly commercial banks) o keep money in he cenral bank, enabling he cenral bank o wihdraw liquidiy from he markes. Consequenly, he reduced amoun of Serbian dinars in he marke sysem induced dinar appreciaion. Figure 4. Esimaion of Ordinary Impulse Response Funcion based on Cholesky One-sandard Deviaion Shock Response of FX changes o shocks in ineres rae differenial Response of ineres rae differenial o shocks in FX changes Figure 5. Esimaion of Accumulaed Impulse Response Funcion from Figure 4 Accumulaed response of FX changes o shocks in ineres rae differenial Accumulaed response of ineres rae differenial o shocks in FX changes Noe o Figures 4 5: Doed lines represen he 95% confidence inerval bands calculaed by Mone-Carlo simulaions wih 1000 replicaions

19 ECONOMETRIC TESTING OF UIRP IN SERBIA In he final sage of our economeric research an MS model is esimaed (Table 9). Two differen regimes are deeced as being significan. Regime 0 is characerized by he lower inercep and variabiliy of FX changes, whereas he impac of he ineres rae differenial is esimaed as Regime 1 is found o be associaed wih higher inercep and variabiliy of FX changes. The slope parameer ha measures he reacion of FX changes o ineres rae differenial in regime 1 is esimaed as Therefore, we may argue ha regime swiches in he relaionship beween FX changes and ineres rae differenials significanly deermine FX changes. A visual inspecion of he regimes in Figure 6 shows ha regime 0 is closely relaed o sub-periods of currency appreciaion and/or relaively sable FX level. Regime 1 is deeced for sub-periods of currency depreciaion. The probabiliy q of remaining in he currency appreciaion regime while being in ha regime is The probabiliy f of swiching from he currency depreciaion regime o he currency appreciaion regime is small and equal o 0.15, indicaing ha he probabiliy of saying in he currency depreciaion regime is relaively high, The economy remains in he currency depreciaion regime 64% of he ime, while he remaining 36% is associaed wih he currency appreciaion/sabiliy regime. The average duraion of he currency depreciaion regime is 8.6 monhs, whereas he average duraion of he opposie regime is shorer, 4.8 monhs. The parameers of he model appear o be muually dependen o some exen. Namely, he slope parameer is bigger during currency depreciaion ( 0.86) when relaively higher variabiliy in FX changes is observed. The slope parameer drops significanly, by abou hree imes, wih he appreciaion rend when lower variabiliy in FX changes is idenified. Therefore, over sub-periods of he dinar s lower variabiliy, sronger dinar appreciaion occurred as a resul of higher ineres rae differenials. However, higher-variabiliy sub-periods are characerized by much lower appreciaion. There is a clear incenive for carry-rade aciviy during periods of less uncerainy (relaively sable dinar) wih high ineres rae differenials. However, during periods of higher variabiliy and persisen depreciaion hese aciviies are subsanially reduced. 53

20 Economic Annals, Volume LXIII, No. 216 / January March 2018 This resul srongly suppors a carry rade presence and overlaps wih some of he previous lieraure. For example, Ichiue and Koyama (2011), also using he regime-swiching approach, find ha low volailiy influences low-ineres-rae currencies o depreciae, which is he cause of he puzzling relaionship. This evidence is derived for developed counries (Japan, Grea Briain, Swizerland, and Germany) and we would expec even greaer appreciaion for emerging markes, due o higher ineres raes. This is wha we have idenified for Serbia. Table 9. Esimaed MS version of UIP model Parameer α 0 (α 0 + α 1 ) β 0 (β 0 + β 1 ) h 0 (h 0 + h 1 ) q f Esimae raio p-value Lineariy es LR = 22.73(0.00) Box-Ljung Q(8)=12.89(0.12), Q(12)=18.21(0.11), Jarque-Bera JB =0.86(0.65), ARCH(12) = 1.38(0.19) Noes: Esimaion is based on he BFGS algorihm. Robus sandard errors are used o calculae - raios. The model addiionally conains wo impulse dummy variables defined similarly o dummies previously inroduced. The firs one has only wo non-zero values of 1 for he following monhs: 2008 M10 and M11. The second one akes 1 for 2009M1 and 0 oherwise. The model performs saisically well, as shown by several misspecificaion ess compued for one-sep predicion error. The value of he lineariy es srongly confirms he presence of non-lineariy capured by MS specificaion. 54

21 Figure 6. Esimaed regimes from MS version of UIP model ECONOMETRIC TESTING OF UIRP IN SERBIA 0.05 FX changes Regime 0 Fied P[Regime 0] smoohed P[Regime 1] smoohed PRELIMINARY ECONOMETRIC RESULTS FOR FOUR EUROPEAN EMERGING ECONOMIES The bounds esing procedure is applied for four CEE counries wih mainly floaing exchange raes, he Czech Republic, Hungary, Poland, and Romania. Prior o his we ake ino accoun descripive saisics of he ineres rae differenial correced for counry risk; he resuls are shown in Table 10. All counries have posiive ineres differenials over he whole period considered, excep ha in he Czech Republic, unypically for an emerging economy, he average ineres differenial is negaive or close o 0. The resul for he oher hree counries is consisen wih he general rend of posiive ineres rae differenials in emerging economies. Similar o Serbia, Romania has he highes ineres differenials correced for counry risk, peaking in 2003 and 2004 (16.42%), which correspond o he high inflaion raes during ha ime (15.3% and 11.9% respecively). Resuls of he economeric invesigaion are summarized in Table

22 Economic Annals, Volume LXIII, No. 216 / January March 2018 Table 10. Descripive saisics of ineres rae differenials correced for counry risk for CEE counries Czech Republic Hungary Poland Romania Mean 0.55% 2.68% 1.40% 3.53% Minimum 2.34% 0.98% 0.77% 0.78% Maximum 0.45% 10.32% 4.14% 16.42% Sd. Dev. 0.55% 2.79% 1.09% 4.37% Table 11. Uni roo es, Bounds es and Level equaion for CEE counries Counry Period Czech 2004M1 2016M12 Hungary 2003M1 2016M12 Poland 2003M1 2016M12 Romania 2004M1 2016M12 ADF for level Level equaion ARDL F sa. s +1 -s i i* - CR α β (2,4) (9,0) (2,0) (1,0) ** Noes: The Augmened Dickey Fuller es is applied o examine saionariy. The number of lags included o ake care of he auocorrelaion is 0 in FX changes for all counries and in ineres differenial correced for counry risk is 4 for he Czech Republic, 1 for Poland, and 0 for Hungary and Romania. I is deermined by he SC crierion. Consan and rend are used as deerminisic componens, excep ha for Poland only consan is included. The 5% criical values are 3.44 for he model wih consan and rend and 2.88 for he model wih consan only. The appropriae ARDL model is seleced based on AIC. Several dummy variables are included in he modelling, defined in he following way. For he Czech Republic hree dummies are used (one ransiory o ake 1 and 1 for 2008M7 and 2008M8 respecively and 0 oherwise, and wo impulse dummies o ake only non-zero value 1 for 2009M1 and 2013M11). For Poland hree impulse dummies are employed having he non-zero value of 1 only for 2008M10, 2008M12, and 2009M2. The model for Hungary has wo impulse dummies: one akes only he non-zero value 1 for 2008M10 and he oher one wo non-zero values of 1 for 2009 M1 and M2. Romania has one impulse dummy designed o have only a non-zero value of 1 for 2009M1. The symbol ** denoes significance a he 0.05 level. The Emerging Marke Bond Index (EMBI) is used as a proxy for counry risk for Poland and Hungary and credi defaul swap (CDS) for he Czech Republic and Romania. 56

23 ECONOMETRIC TESTING OF UIRP IN SERBIA I is eviden ha FX changes are a saionary ime series, whereas he ineres rae differenial correced for counry risk has one uni roo in all counries excep he Czech Republic, where i is saionary. Hence, he necessiy of applying he ARDL procedure arises. As all calculaed F-saisics are above he upper bound a he 5% significance level (Table 2), he level relaionship is confirmed for all counries. The las columns of Table 11 give esimaed coefficiens. In hree economies he slope esimae β is negaive, and i is only significan in he case of Romania. For he Czech Republic he slope is esimaed o be posiive, bu appears o be insignifican. The consan erm is insignifican in all counries. Recall ha resuls for Serbia are quie differen from hose for mos of he oher considered CEE counries. The Serbian resuls are in line wih he findings repored for Romania, alhough he slope coefficien is esimaed o be wice as low as in Romania ( 1.35 in Serbia and 0.77 in Romania). These are he only wo counries ha have significanly negaive coefficiens of he ineres rae differenial correced for counry risk. This is no surprising, given he remendously high ineres rae differenials in boh counries even when correced for counry risk. Romania is also characerized by weak macroeconomic performance and a volaile economy ha resuls in higher ineres raes, which migh be a poenial arge for carry-rades in he shor run. On he oher hand, Poland and he Czech Republic are he mos developed counries in he CEE region, wih sable economic condiions and relaively good performance compared wih he res of he region. This explains why he ineres differenial is lower in hese counries. We confirm he exisence of he forward premium puzzle in mos of he considered counries, wih significan slopes in only Serbia and Romania. These resuls correspond o oher evidence in he lieraure showing ha he ineres rae puzzle appears when esing he UIP condiion in a shor-erm horizon (one week o a quarer). 7. CONCLUSION This paper examines an ex pos uncovered ineres rae pariy condiion in Serbia over a one-monh horizon. The money marke ineres raes BELIBOR and EURIBOR and EUR/RSD exchange raes are used for he period Sepember 2005 o December Counry risk is incorporaed a priori in he model by aking ino accoun he Emerging Marke Bond Index (EMBI). Firs, 57

24 Economic Annals, Volume LXIII, No. 216 / January March 2018 economeric resuls are derived from long-run parameer esimaion based on he auoregressive disribued lag model. In Serbia a posiive ineres differenial correced for counry risk is found o cause he dinar o appreciae, insead of depreciaing as UIP predics. Hence, he ineres pariy puzzle is confirmed, as previously deeced by he OLS approach (Božović and Talijan 2015). The resuls are in line wih he growing body of lieraure, mainly derived for developed markes. Unexpeced shocks o ineres rae differenials correced for counry risk lead o dinar appreciaion wih one lag, which coninues in he following four monhs. On he oher hand, he effec of unanicipaed shocks in FX changes on ineres rae differenial responses seems o be negligible. This resul is derived from he impulse response funcion compued from he bivariae vecor auoregressive model. The wo-regime Markov swiching parameer model is also esimaed. I suggess ha regime swiches in he relaionship beween FX changes and ineres rae differenials are significan in deermining FX changes. Sronger dinar appreciaion occurred as a resul of higher ineres raes during lowervariabiliy sub-periods of he dinar. Higher-variabiliy sub-periods are characerized by much lower appreciaion. Our resuls underline ha here was a huge incenive for shor-erm carry rades in Serbia, regardless of he subsanial risk. This finding migh be useful when evaluaing exising capial conrols and ailoring new ones, so ha policymakers pay paricular aenion o shor-erm lenghs. Finally, preliminary economeric analysis conduced for oher four European emerging economies only documens similar resuls for Romania. 58

25 ECONOMETRIC TESTING OF UIRP IN SERBIA REFERENCES Alexius, A. (2001). Uncovered Ineres Pariy Revisied. Review of Inernaional Economics, 9 (3), pp Bacchea, P. & van Wincoop, E. (2006). Incomplee Informaion Processing: A Soluion o he Forward Discoun Puzzle. Working Paper WP/06/35, Federal Reserve Bank of San Francisco. Bacchea, P. & van Wincoop, E. (2010). Infrequen Porfolio Decisions: A Soluion o he Forward Discoun Puzzle. American Economic Review, 100 (3), Bansal, R. & Dahlquis, M. (2000). The forward premium puzzle: differen ales from developed and emerging economies. Journal of Inernaional Economics, 51 (1), pp Bekaer, G. & Hodrick, B. (1993). On Biases in he Measuremen of Foreign Exchange Risk Premiums. Journal of Inernaional Money and Finance, 12 (2), pp Božović, M. & Talijan, M. (2015). The anomalous forward premium of EUR/RSD exchange rae. Indusrija, 43 (4), pp Brunnermeier, M.K., Nagel S. & Pedersen, L.H. (2008). Carry rades and currency crashes. NBER Working Paper WP Naional Bureau of Economic Research. Burnside, C. (2014). The carry rade in indusrialized and emerging markes. Journal Economia Chilena, 17, pp Cavoli, T. & Rajan, R. S. (2006). Capial Inflows Problem in Seleced Asian Counries in he 1990s Revisied: The Role of Moneary Serilizaion. Asian Economic Journal, 20 (4), pp Chinn, M. & Meredih, G. (2004). Moneary Policy and Long Horizon Uncovered Ineres Pariy. IMF Saff Papers, 51 (3), pp Cuesas, J.C., Filipozzi, F. & Saehr, K. (2015). Uncovered Ineres Pariy in Cenral and Easern Europe: Sample, Expecaions and Srucural Break. Research Paper , The Universiy of Sheffield, Deparmen of Economics. Enders, W. & Lee, J. (2012). A Uni Roo Tes Using a Fourier Series o Approximae Smooh Breaks. Oxford Bullein of Economics and Saisics, 74 (4), pp Engel, C. (1996). The Forward Discoun Anomaly and he Risk Premium: A Survey of Recen Evidence. Journal of Empirical Finance, 3 (2), pp Engel, C. (2016). Exchange Raes, Ineres Raes, and he Risk Premium. American Economic Review, 106 (2), pp

26 Economic Annals, Volume LXIII, No. 216 / January March 2018 Engel, C. & Hamilon, J.D. (1990). Long Swings in he Dollar: Are They in he Daa and Do Markes Know I? American Economic Review, 80 (4), pp Fama, E. (1984). Forward And Spo Exchange Raes. Journal Of Moneary Economics, 14 (3), pp Felcser, D. & Vonnak, B. (2014). Carry Trade, Uncovered Ineres Pariy and Moneary Policy. MNB Working Paper WP/14/3. Magyar Naional Bank. Ferreira, A. L. (2004). Leaning Agains Pariy. Working Paper WP/04/13, Universiy of Ken. Ferreira, A.L. (2009). Is i Risk? An Auomaed Approach o Explaining he ex ane UIP Deviaions of Brazil. Cuadernos de Economía, 46, pp Filipozzi, F. & Saehr, K. (2012). Uncovered Ineres Pariy in Cenral and Easern Europe: Convergence and he Global Financial Crisis. Discussions on Esonian Economic Policy: Theory and Pracice of Economic Policy 20 (1). Flood, R.B. & Rose, A.K. (2002). Uncovered Ineres Pariy in Crisis. Inernaional Moneary Fund Saff Papers, 49 (2), pp Frankel, J. A. & K.A. Froo (1990). Chariss, Fundamenaliss, and Trading in he Foreign Exchange Marke. American Economic Review, 80 (2), pp Froo, K.A. & Thaler, R.H. (1990). Anomalies: Foreign Exchange. Journal of Economic Perspecives, 4 (3), pp Hamilon, J.D. (1989). A new approach o he economic analysis of nonsaionary ime series and he business cycle. Economerica, 57 (2), pp Ichiue, H. & Koyama. K. (2011). Regime Swiches in Exchange Rae Volailiy and Uncovered Ineres Pariy. Journal of Inernaional Money and Finance, 30 (7), pp Ismailov, A. & Rossi, B. (2017). Uncerainy and deviaions from uncovered ineres rae pariy. Journal of Inernaional Money and Finance, aricle in press. Io, H. & Chinn, M. (2007). Price-based Measuremen of Financial Globalizaion: A Cross-counry Sudy of Ineres Rae Pariy. Pacific Economic Review, 12 (4), pp Jiang, C., Li, X., Chang, H. & Su, C. (2013). Uncovered ineres pariy and risk premium convergence in Cenral and Easern European Counries. Economic Modelling, 33, pp Juselius, K. & Assenmacher, K. (2017). Real exchange rae persisence and he excess reurn puzzle: The case of Swizerland versus he US. Journal of Applied Economerics, 32 (6), pp Levi, M.D. (2005). Inernaional Finance, Rouledge. 60