UNCOVERED INTEREST PARITY IN CENTRAL AND EASTERN EUROPE: CONVERGENCE AND THE GLOBAL FINANCIAL CRISIS 1

Transcription

1 UNCOVERED INTEREST PARITY IN CENTRAL AND EASTERN EUROPE: CONVERGENCE AND THE GLOBAL FINANCIAL CRISIS 1 Abstract Fabio Filipozzi 2, Karsten Staehr Tallinn University of Technology, Bank of Estonia This paper presents tests of uncovered interest parity in Croatia, the Czech Republic, Hungary, Poland and Roania; all countries in Central and Eastern Europe with floating exchange rates. Data are onthly and the trading horizon is three onths. The estiations show that the UIP hypothesis is rejected for the full saple fro 1999 to 211 for all five countries. A nuber of reasons for the rejection were investigated. Rolling regressions show that standard versions of the UIP essentially lose all explanatory power in 28-1, which was a period in which the global financial crisis led to instability in currency and interest arkets in Central and Eastern Europe. Two indicators of global risk aversion were also found to enter significantly in the any UIP estiations. Finally, the size of the interest rates spread also sees to be of iportance, at least for Poland and Roania. Keywords: UIP, financial integration, global financial crisis, Central and Eastern Europe JEL Classification: E43, F36, G1, G15 1. Introduction Uncovered interest rate parity reains a key assuption in international econoics despite the assive body of epirical evidence against the hypothesis. A. Alexius (21, p. 55) This paper presents the results of econoetric analyses testing the uncovered interest parity (UIP) hypothesis on data fro Poland, the Czech Republic, Hungary, Roania and Croatia. The data saple starts in 1999 or shortly afterwards and ends in Septeber 211, and as such spans a period in which the countries experienced both rapid econoic and financial integration and also the fallout fro the global financial crisis. The UIP hypothesis is tested for a trading horizon of three onths using onthly data. The five countries in the saple are the ain countries in 1 The authors would like to thank Juan Carlos Cuestas, Kalev Jõgi, Jaan Masso and Kärt Tooel for valuable coents to an earlier version of the paper. All reaining errors are the responsibility of the authors. The views expressed are those of the authors and not necessarily those of Eesti Pank. 2 Corresponding author: Eesti Pank, Estonia pst. 13, 1595 Tallinn, Estonia. Tel.: E-ail: fabio.filipozzi@eestipank.ee. 58

2 Central and Eastern Europe having floating or essentially floating exchange rate regies during the saple period. 3 Poland, the Czech Republic and Hungary joined the European Union in May 24 and Roania in January 27, while Croatia was in the final stages of ebership negotiations at the tie of writing in August 211. The hypothesis of uncovered interest parity rests on the idea that arbitrage leads to equalisation of the return on assets or liabilities in the doestic currency and the expected return on coparable assets or liabilities in a foreign currency. Testing the UIP hypothesis ay thus provide inforation as to whether the exchange and interest arkets under consideration function so that all the gains fro trade are exploited, i.e. whether the arkets are efficient. In practice, however, divergence between doestic and expected foreign returns ay also be due to issues such as transaction costs, different risk profiles and non-syetric tax treatents. This paper presents tests of the UIP hypothesis for Croatia, the Czech Republic, Hungary, Poland and Roania. Section 2 provides a survey of epirical studies of the UIP hypothesis with a particular focus on studies dealing with countries in Central and Eastern Europe (CEE). There are only a very liited nuber of studies that exaine the UIP hypothesis for Central and East European countries, particularly studies which use data covering the EU accession and the global financial crisis. The CEE countries liberalised their capital arkets and reoved their reaining exchange rate restrictions before joining the EU (European Coission 21a). Many of the countries experienced substantial capital inflows in the years iediately before and after accession to the EU, just to see a reversal of the flows in 28-9 following the global financial crisis (Jevcak et al. 211). It is a largely un-researched question whether these abrupt changes in capital flows have affected the relationship between exchange rates and interest rates in the CEE countries. Testing the UIP hypothesis for the CEE countries is also iportant because households and firs in any countries in the region have borrowed extensively in foreign currencies, ostly the euro and the Swiss franc (Rosenberg & Tirpak 28). In essence borrowers expect that borrowing in a foreign currency is cheaper than doestic currency borrowing, eaning they have bet that the UIP will not hold within the horizon of the loan contract. Speculators without an underlying otive of borrowing or saving have also taken positions, carry trade, in the currencies of the CEE countries. Rosenberg & Tirpak (28) and Brzoza-Brzezina et al. (21) find that the interest differential between doestic and foreign rates is an iportant deterinant of borrowing and saving in foreign currencies in the CEE countries. 4 3 The study excludes countries with fixed exchanges and countries that adopted the euro during the saple period. 4 Batini & Dowling (211) use a UIP fraework to decopose exchange rate oveents between ajor currencies and the US dollar into shocks steing fro US onetary policy and other sources. The sharp depreciation of ost of the saple currencies against the US dollar during the global financial crisis cannot be attributed to changes in the interest rate spread, but rather to changes in the risk preia. The subsequent appreciation of any of the 59

3 This paper seeks to contribute to the epirical literature on the UIP by investigating its epirical validity in the ain CEE countries that have a floating exchange rate. The paper tests the UIP hypothesis using individual regressions for each of the five CEE countries. As typically found in the literature, the UIP holds better for soe countries than for others and better in soe periods than in others. The paper investigates factors that ay explain the variation across countries and across tie, linking the findings to the different stages of convergence attained in the countries and to the global financial crisis that unfolded in The rest of the paper is organised as follows: Section 2 discusses the theoretical foundation of the UIP hypothesis. Section 3 surveys a nuber of epirical studies with a particular ephasis on the CEE countries. Section 4 docuents the data and shows the results of unit root tests. Section 5 presents the baseline estiations using the full saple available. Section 6 contains the estiations when structural change is identified using rolling windows. Section 7 considers whether there are non-linear effects. Section 8 shows the results when different proxies of external deterinants of the risk preiu are included. Finally, Section 9 suarises the results. 2. The theory of uncovered interest parity The theory underlying the Uncovered Interest Parity is fairly siple as it builds on the assuption of arbitrage equalising expected returns in different arkets (Levi 25, Ch. 8). Consider the investent decision of an investor who at tie t seeks to invest a su for a period of tie units. Assuing that the interest rate is constant and equal to i, for the entire investent horizon, the gross return fro investing doestically t is 1 per tie unit leading to i t, ( 1 t, i ) copounded during the periods of the investent. The su can alternatively be exchanged at the spot exchange rate S t and invested abroad at the interest rate * t, * t, i. The foreign denoinated gross return after periods is (1 i ) / S and this su can be exchanged into doestic currency at the exchange rate S. t t In practice the exchange rate periods ahead is unknown, so the investor will have to for expectations for this exchange rate. The variable e t S denotes the expectation in period t for the exchange rate in period t. A risk-neutral investor would be indifferent as to whether to invest in the doestically denoinated asset or in the foreign denoinated asset if the expected returns are identical, i.e. if uncovered interest parity holds: currencies ay partly reflect the carry trade exploiting low US interest rates and higher interest rates in other countries. None of the CEE countries are included in the saple. 6

4 e t * S (1 it, ) (1 it, ) (1) S t This condition is usually log-linearised. We adopt the notation logs e t t e St log = logs, which is approxiately the relative change in the exchange rate over the -period horizon of the investent. The variable e St log is positive if the investor expects that the doestic currency will depreciate fro period t to period t + and negative if the investor expects that the doestic currency will appreciate. Using this notation eq. (1) becoes: e log St log( 1 i ) log( 1 * t, it, ) (2) * * t, t, Using the approxiations it, log( 1 it, ) and i (1 i ) and lowercase s t to denote the logarith of the exchange rate, i.e. s t log( S t ) and s log( e t St ), the version of the UIP in eq. (2) can be rewritten as: e st * it, it, (3) The left-hand side is the annualised average expected capital gain fro the foreign currency investent. The right hand side is the spread between the doestic and foreign interest rates. The upshot is that a positive spread is consistent with the UIP hypothesis only if the spot rate is expected to depreciate in the way given in eq. (3), i.e. investent in the foreign denoinated asset will only take place if the positive interest spread is copensated for by a corresponding capital gain. 5 Eq. (3) can be tested epirically if a easure of the expected spot exchange rate periods ahead is available, for instance fro surveys or arket data. A ore coon ethodology, however, is based on the assuption of rational expectations, i.e. s e t / = st / t, where E t [ t ], i.e. the atheatical expectation of is zero, conditional on inforation in period t. t This epirical version of the UIP is: st it i *, t, t (4) e 5 The doestic interest rate that is consistent with UIP follows directly fro Eq. (3), i.e. * e i t, t, t / i s. 61

5 A siple epirical ethodology for a test of the UIP hypothesis entails estiation of the following standard UIP regression odel: s t ( i * t, it, ) t (5) Eq. (5) is the odel used in ost estiations in the paper. The UIP corresponds to the joint null hypothesis that the constant α =, the slope coefficient β = 1 and E t [ t ] ; the UIP hypothesis cannot be rejected if none of these conditions can be rejected. 6 Three coents are appropriate: First, the assuption that E t [ t ] iplies that the residuals are serially uncorrelated if the investent horizon coincides with the sapling frequency. If, however, the investent horizon exceeds the investent frequency (as would be the case with, for instance, onthly data and a quarterly investent horizon), overlapping data eerge and the residual will be subject to serial correlation of order 1 even if E t [ t ] is satisfied for the investent horizon (Baillie & Bollerslev 2). Second, the test iplies essentially a joint test of several hypotheses, including the hypothesis that arbitrage equalises the expected currency gain and the interest rate differential and the hypothesis that investors have rational expectations (Alper et al. 29). If α = and β = 1 cannot be rejected (in a odel with non-serially correlated residuals), it is reasonable to assue that both hypotheses are satisfied. Rejection iplies that the UIP does not hold, but the underlying reason (such as absence of arbitrage trades or non-rational expectations) cannot be identified right away. Third, the test entails the estiation of one coefficient of the interest spread * t, it, i, not separate coefficients for each of the interest rates. The iplicit assuption is that the investors react only to the interest rate spread, i.e. in siilarly sized but opposing ways to each of the two interest rates (Mehl & Cappiello 27). In practice, the assuption is convenient as it typically iplies that the interest spread * t, it, i is stationary, but this ay not be the case for each interest rate considered individually. The theoretical odel in eq. (3) and the epirical odel in eq. (5) are based on the assuption that the investors are risk-neutral and do not require a risk preiu to hold one currency or the other. This assuption is unrealistic in practice insofar as investors are risk averse. A constant risk preiu can be included by allowing the 6 Faa (1984) suggests a narrower test of the UIP hypothesis, essentially testing whether the forward rate is an unbiased estiator of the future exchange rate. The Faa regression entails that the forward preiu is regressed on the future exchange rate change and a slope coefficient of one is interpreted as confiration of the efficient arket hypothesis. 62

6 constant α to differ fro zero. 7 This assuption ight be too restrictive if the risk preiu is non-constant, but it would then be necessary to odel the risk preiu. The presence of a risk preiu and in particular a non-constant risk-preiu does not contradict the UIP hypothesis per se, but it coplicates the epirical testing as it requires that the risk preiu can be identified epirically. Beyond the presence of a risk preiu, it is possible to point out a nuber of factors which would entail that eq. (3) would not hold (Levi 25, Ch. 8): Financial arkets ay not be fully integrated because of regulation, institutional barriers or undeveloped trading possibilities (lack of instruents). In this case, the trades needed to arbitrage different expected returns ay not be available. Illiquidity or thin arkets ay lead to arket inefficiency as prices ay not reflect available inforation. Illiquidity creates ore risks and coplicates arbitrage trades, but this ay not play a ajor role in currency arkets with large turnovers. Transaction costs ay ake it unprofitable to execute trades that exploit sall deviations fro the UIP. Inforation costs ay be high, in part because inforation is needed for expectations about exchange rate oveents to be fored. Investors in exchange and interest arkets ay not have fully rational expectations. Investors ay use echanical or oentu-based trading strategies, essentially disregarding the available inforation. Liquidity preference ay favour investent in doestic currency assets, as investent in foreign currency assets ay be ore difficult to wind down if there is a sudden need for liquidity in the doestic currency. The asyetric tax treatent of interest returns and returns fro capital gains (here steing fro exchange rate changes) ay ean that the strict UIP hypothesis which does not take account of taxation would not hold. 3. Epirical studies The uncovered interest parity hypothesis has been tested epirically for a long tie, but better financial data have continuously expanded the possibilities for testing. We will briefly discuss the results of studies using datasets covering developed econoies, eerging arket econoies and countries in Central and Eastern Europe. Meese & Rogoff (1983) is an influential early study showing that the interest rate spread has essentially no predictive power for the future exchange rate oveents of the US dollar when evaluated on data fro the 197s. 7 If the exchange rate is expected to reain constant ( / ) and α >, the doestic interest rate st e * i t, ust exceed the foreign currency interest i t, in order for UIP to hold. 63

7 A range of epirical studies have subsequently exained the UIP hypothesis using different currency and tie saples and different econoetric ethods. Froot & Thaler (199) survey 75 published estiates and conclude that the strict version of the UIP hypothesis is rejected in alost all cases. Siilar conclusions have been reached in other subsequent survey papers (e.g. Engel 1996, Alexius 21). The consistent finding that the estiated slope coefficient is far below one and often negative has been labelled the forward preiu anoaly (Froot & Thaler 199, Booth & Longworth 1986, Olo & Pilbea 211). Most studies are based on data with investent horizons of one onth, three onths or six onths as such data are readily available. Studies suggest, however, that the UIP ay hold better at longer investent horizons. Chinn & Meredith (24) study the epirical validity of the UIP hypothesis for the currencies of the G7 countries using a saple fro 1983 to 2. For short investent horizons, the UIP is rejected in all cases, but when the UIP regression is estiated using 5 or 1 year horizons, the slope coefficient is always positive and in any cases not statistically different fro one. 8 Qualitatively siilar results are obtained by Alexius (21) and Mehl & Cappiello (27) although the UIP hypothesis is still rejected for soe countries. The tie saple also sees to be of iportance, which is unsurprising given that financial arkets and regulatory schees change over tie. Lothiana & Wu (211) use a saple of 2 years and consider the UIP hypothesis between the dollar and sterling and between the franc and sterling. They find that the slope estiate β typically is positive although far fro one until 198, but then turns negative for ost periods after that. It is argued that the liited support for the UIP hypothesis is the result of expectations that ex-post are wrong for extended periods of tie. Flood & Rose (22) reach different conclusions using data fro the 199s and a broad saple of high-incoe and eerging econoies. Estiation of standard UIP regressions leads to the conclusion that the hypothesis received ore support fro their data fro the 199s than fro earlier data, although the overall conclusion is still negative as spelled out in the title: Uncovered interest parity in crisis. Baillie & Bollerslev (2) suggest that the forward preiu anoaly can, at least partly, be explained by the different tie series properties of the variables in the standard UIP regression. The relative exchange rate change ( ) is close to s t / a rando walk (at least at relatively high frequencies), while the interest rate spread * t, t, ( i i ) typically exhibits substantial persistence (but not a unit root). Baillie & Bollerslev (2) siulate data based on these characteristics and show that the resulting slope, although centred around one, exhibits a very high variance. The upshot is that estiations with relatively few observations are likely to produce 8 The finding that the UIP hypothesis generally holds better for long investent horizons than for short horizons can be related to the peso proble (Froot & Thaler 199). In this context, the peso proble iplies that adjustents of the exchange rate to the UIP ay occur in discrete and infrequent steps of substantial agnitude. 64

8 coefficient estiates that are sensitive to saple changes and that ay differ significantly fro one even if the UIP is in fact satisfied. It is typically found that the UIP holds better for cases where the interest rate spread is substantial and less well for cases where the interest rate spread is sall. Mehl & Cappiello (27) find that UIP relations estiated for soe high-incoe and eerging arket econoies exhibit non-linearities. They estiate a sooth transition regression iplying different arginal effects of the interest rate spread when the interest rate spread is sall and when it is large. The upshot is that the standard linear odel ixes the effects of different regies. Using data for selected European currencies, Lothiana & Wu (211) find ore support for the UIP hypothesis in periods in which the interest rate spread is large. This result sees intuitively reasonable as factors such as risks and transaction costs ay not warrant arbitrage trading if the returns fro such trades are liited (Froot & Thaler 199). Alper et al. (29) survey the literature on UIP testing in eerging arket econoies. On the one hand, the high trend inflation observed in any eerging arkets facilitates the forecasting of exchange rate developents and therefore akes it ore likely that the UIP hypothesis does hold. On the other hand, structural breaks and uncertainties are likely to be ore pronounced in eerging arkets, which would suggest that the UIP does not hold. Epirical studies confir that UIP estiations frequently exhibit different properties for eerging arkets and for high-incoe econoies. Alper et al. (29, p. 123) conclude that identifying and odelling structural breaks provide roo for iproveent for further research on the UIP condition for [eerging arkets]. Bansal & Dahlquist (2) provide an explicit coparison of results for high-incoe and eerging arket econoies and conclude that the UIP is ore likely to hold for eerging arkets than for highincoe econoies. Different per capita GNP, average inflation and inflation volatility are factors that ay explain the different results. Only a sall nuber of studies have exained the epirical validity of the UIP hypothesis for countries in Central and Eastern Europe. Brasili & Sitzia (23) estiate panel odels based on CEE data in which future exchange rate changes are explained by the interest rate spread and a range of other factors that ay be considered proxies of the risk preiu. The spread is not statistically significant in a specification in which it enters linearly, but a non-linear transforation of the spread attains statistical significance, suggesting that non-linearities play an iportant role. Ho & Ariff (29) also use a panel explaining the future exchange rate change with any variables along with the interest rate spread. A range of specifications all produce positive and statistically significant coefficients to the interest rate spread for the saple of Eastern European countries, but the coefficients vary substantially across different specifications. The use of panel data in these two studies precludes the estiation of country-specific coefficients of the interest rate spread. Mansori (23) copares results for the Czech Republic, Hungary and Poland fro 1994 to 22 with results for a nuber of West European countries. There is ore 65

9 support for the UIP hypothesis for the three East European countries, especially the Czech Republic and Hungary, than for the West European countries. The results for the CEE countries are however very sensitive to changes in the tie saple, possibly as a result of the convergence processes underway during the period analysed. Horobet et al. (29, 21) estiate standard UIP regressions for eight countries, including four fro Central and Eastern Europe using onthly data fro 26 to 29. The estiated slope coefficients are positive in all cases, but neither econoically nor statistically different fro zero. This result sees to hold whether or not exchange arket volatility is taken into account. 4. Data and unit root tests This section provides an overview of the dataset and the ain features of the series for the five saple countries, Croatia, the Czech Republic, Hungary, Poland and Roania. The saples vary across the five countries but generally span a bit ore than a decade, starting in 1999 and ending in Septeber 211. The five countries all had floating exchange rates during this period, although Poland forally used anaged devaluations until April 2 and Hungary used different corridors until The analyses are undertaken for positions with a 3-onth horizon, iplying that the returns fro the currency exposure and the interest rate differential are both calculated for a 3-onth holding period. As discussed in the literature survey in Section 3, the results ay vary with the investent horizon, but the 3-onth horizon has been chosen because the 3-onth oney arket is one of the ost liquid segents of the arket. The five countries saw increased integration with Western Europe, and in particular with the euro area, during the saple period. The reference area is therefore taken to be the euro area: the exchange rates are in units of local currency per euro and the interest rate spreads of the local interest rate are against the Euribor rate. It is noticeable that the countries considered here were at different stages of their processes of convergence with Western Europe during the saple period. 1 9 The Hungarian bands changed frequently before they were finally reoved in February 28. Until May 21, the anaged devaluation was based on a daily rate of devaluation against, in 1999, a basket (3 percent USD, 7 percent EUR) and, thereafter, the euro. The band around the central rate of the devaluation path was +/ 2.25 percent. Fro May to October 21 the band around the central rate was increased to +/ 15 percent. Fro October 21 the central parity was fixed at HUF/EUR and in June 23 to HUF/EUR, while the band reained at +/ 15 percent. 1 For an overview of the stages of convergence, see the European Coission (21a, 21b). Different indicators can be used to assess the degree of convergence of the CEE countries with Western Europe. European Coission (21a, 21b) asserts that the convergence process in Roania and Croatia has been slower than that in the other three CEE countries in our saple. 66

10 Most of the estiations are based on only two variables, cf. eq. (5). 11 The variable FX_CHG is the percentage change of the spot exchange rate over a 3-onth period, where the exchange rate denotes units of local currency per euro at the end of onth. A positive value of FX_CHG indicates a depreciation of the local currency against the euro over the 3-onth period; a negative value indicates an appreciation. The variable INT_SP is the annualised interest spread between a 3-onth doestic currency deposit and the 3-onth Euribor. The available saple of data varies across the countries. For Croatia, the series on the noinal exchange rate starts in Noveber 1999, iplying that the 3-onth FX_CHG variable starts in February 2. For Poland, the local 3-onth interest rate is available fro the beginning of 21. Table 1 reports suary statistics of the exchange rate changes and the interest rate spreads for the five saple countries. Table 1. Descriptive statistics for 3-onth exchange rate change and 3-onth interest rate spread FX_CHG Mean Median Max. Min. Std. Dev. Obs. Croatia Czech Republic Hungary Poland Roania INT_SP Mean Median Max. Min. Std. Dev. Obs. Croatia Czech Republic Hungary Poland Roania Figure 1 depicts the noinal exchange rate of each Eastern European country against the euro fro the beginning of 1999 and until Deceber 211. The first thing to notice is that the exchange rate dynaics vary considerably across the five saple countries. The currencies of Croatia and the Czech Republic have tended to appreciate against the euro, while the currency of Roania has tended to depreciate. The currencies of Hungary and Poland have been relatively stable with exchange rates fluctuating around a relatively constant level. 11 The variables are calculated based on Ecowin source data. 67

11 7.8 Croatia 4 Czech Republic Hungary 5. Poland Roania Figure 1. Noinal exchange rate of local currency against euro. The different exchange rate developent across the saple countries is the result of any factors. The process of integration into EU structures, and the associated confidence effects, has affected the exchange rate dynaics in the Central and Eastern European countries. The speed of and coitent to integration has differed across the countries. 12 The ain essage for our analyses is that there is no Central and Eastern European block with closely co-oving exchange rates; the exchange rate developents are fundaentally different across the five saple countries. Figure 2 depicts the 3-onth annualised change of the exchange rate against the euro. The series are very volatile, which suggests that, for the UIP to hold, the interest rate differential between the country and the euro area would also have to be volatile. 12 The Roanian case is noticeable because the period fro 23 to 25 represents a political and econoic regie switch. During this period Roania joined the Council of Europe and the WTO, and becae an associated eber of the European Union. These steps were part of the process of stabilising the political and econoic situation in the country, and helped to increase the confidence of financial arkets in the Roanian econoy (European Coission 21a). 68

12 Croatia Hungary Czech Republic Poland Roania Figure 2. Annualised changes of local currency versus euro over 3-onth period, %. Figure 3 reports the spread between the local 3-onth interbank interest rate and the 3-onth Euribor. The volatility of the interest rates spread is uch saller than the volatility of the foreign exchange rate changes on the sae horizon. 69

13 Croatia Hungary Czech Republic Poland Roania Figure 3. Annualised interest rate spreads on 3-onth deposits, %. The tie series properties of the exchange rate changes and the interest rate spreads have been exained by eans of Augented Dickey-Fuller tests. Given that the variables are either changes in percentage ters (for currency pairs) or spreads (interest rates), the test is perfored at the level of the variables and an intercept, but no tie trend, is included in the estiations. The nuber of lags used is chosen by eans of the Schwartz selection criterion. The results are reported in Table 2. The hypothesis of a unit root can be rejected in all cases; the series are I() for all five saple countries. 7

14 Table 2. Augented Dickey-Fuller unit root tests FX_CHG 1% C.V. 5% C.V. 1% C.V. Statistic Prob. Process Croatia I() Czech Republic I() Hungary I() Poland I() Roania I() INT_SP 1% C.V. 5% C.V. 1% C.V. Statistic Prob. Process Croatia I() Czech Republic I() Hungary I() Poland I() Roania I() Note: C.V. denotes critical value. 5. Uncovered interest parity We start by rewriting eq. (5) using our epirical notation in which a bracket after the variable nae is used to indicate a tie shift (in onth) of the variable: FX_CHG(3) = α + β INT_SP + ε(3) (6) Eq. (6) is estiated for each country individually using OLS. The results are reported in Table 3. The choice of a 3-onth investent horizon but onthly data leads to first- and second order-autocorrelation of the residuals. We therefore report Newey-West robust standard errors. The strict version of the UIP holds if α = and β = 1 and the residuals do not exhibit serial correlation of the third or a higher order. The table reports the F-statistics for the Wald test of the joint hypothesis α = and β = 1. Exaination of the residuals reveals the existence of autocorrelation of first and soeties second order, but never of higher orders. The estiation results reveal that the coefficients of deterination, R 2, of all the regressions are extreely low. This is not surprising in light of Figures 2 and 3 and is found in all tests of the UIP hypothesis (Flood 1996). The foreign exchange return is uch ore volatile than the interest rate spread, which liits the ability of the interest rate spread to explain the foreign exchange change. 71

15 Table 3. UIP estiation results (OLS) Croatia Czech Republic Hungary Poland Roania ˆ ˆ F-stat R 2 Saple Obs (.888) (1.718) * (5.76) (6.479) 2.23 (3.29) ** (.21) (.972) * (.711) (1.319).38 *** (.87) [.] [.] 1.12 [.].642 [.528] [.] :2-1999:1-1999:1-21:1-1999:1- Note: Newey-West standard errors are shown in round brackets. Superscripts ***, **, * denote that the coefficient estiate is statistically different fro at the 1, 5 and 1% level of significance respectively. The null hypothesis of the F-test is that α = and β = 1; the p-value is shown in square brackets. The estiated slope coefficients in Table 3 are different fro 1 at the 1% level of significance for all five saple countries. For all countries except Roania, the coefficients are also negative, which is in accordance with the forward preiu anoaly found in any other studies (cf. Section 3). For Roania, the estiated coefficient is positive and significantly different fro zero (but also significantly different fro one). This would be consistent with the finding that the UIP hypothesis is ore likely to hold when the interest rates spread is large (Froot & Thaler 199, Mehl & Cappiello 27, Lothiana & Wu 211). It follows fro Figure 3 that the spread between the Roanian 3-onths interest rate and the 3-onths Euribor rate was in the double digits until 25 and also afterwards reained uch higher than for the other saple countries. The large interest spread reflects that Roania has experienced a ore prolonged convergence process the other saple countries. The estiated constant ters are, with the exception of the Czech Republic, positive, but statistically significantly different fro only for one country. As already noted, this coefficient should indicate the presence of either a risk preiu or barriers to entry. While it is probable that barriers to entry or other parts of the regulatory landscape do not change very often, previous research and anecdotal evidence (again, fro the recent financial crisis) indicates that the risk preiu varies across tie and econoic cycles, and therefore to odel the as a constant would be to ipose a tight constraint on the odel The residuals generally exhibit soe heteroskedasticity. To assess the ipact, we estiated eq. (6) using a GARCH specification. Although the GARCH coefficients are statistically significant in any cases, the effects on the estiated α and β and the explanatory power of the regressions are odest. 72

16 The F-statistics reported in Table 3 shows that Poland is the only country for which the null hypothesis cannot be rejected. The Polish case is predicated by the fact that the standard errors of the two coefficient estiates are very high for this country. For all other countries in the saple, the joint hypothesis that α and β take values in accordance with the UIP is rejected. 6. Uncovered interest parity across tie The test of the UIP in Section 5 is undertaken on the entire available tie saple fro the turn of the century to Septeber 211. The recent global financial crisis has, however, provoked very sharp reactions in inter alia foreign exchange and interest arkets. Eastern European countries largely escaped the first part of the crisis (the sub-prie phase fro suer 27), but the default of Lehan Brothers in Septeber 28 affected the region greatly. This is also shown by Figures 1 and 3, in which sudden depreciations of the currencies against the euro and a jup in the spreads between local interest rates and the Euribor are evident. In order to shed further light on the ipact on the UIP of the global financial crisis, and ore generally to shed light on the tie diension, we undertake rolling windows estiations with saples of onthly observations for five years. The estiations are based on eq. (6), i.e. the siple linear version of the UIP. Figure 4 shows the coefficient of deterination, while Figures 5 and 6 show the estiated constants and slope coefficients for the five countries. For all three figures, the date reported on the horizontal axis indicates the end of the saple Roania Poland Hungary Czech Republic Croatia Figure 4. Coefficient of deterination, 5-year rolling windows. Figure 4 reveals that the explanatory power of the regressions is always very low for Poland and Croatia, but relatively high before the crisis for the three other countries. 73

17 This could be an indication that Poland and Croatia ay have been ore closed or insulated fro external influences than the other three countries in the saple (Jevcak et al. 211). Moreover, when the windows consist largely of the period around the global financial crisis, the siple UIP specification (without crisis indicators and with fixed coefficients) basically has no explanatory power for the five saple countries. Further insights into developents before and after the global financial crisis hit the region can be gained fro Figures 5 and 6. The coefficient estiate and +/ 2 ties the Newey-West standard errors are depicted in each figure. The estiated constants and slopes for all the saple countries display extree variation. This could be due to the relatively short span of the saple (five years for each rolling regression), or to an inherent instability in the relation between interest rate spreads and currency returns (Baillie & Bollerslev 2). 8 Croatia 5 Czech Republic Hungary Poland Roania Figure 5. Estiated constants, 5-year rolling windows. 74

18 2 Croatia 1 Czech Republic Hungary Poland Roania Figure 6. Estiated slope coefficients, 5-year rolling windows. The UIP specifications exhibit soe explanatory power for the Czech Republic, Hungary and Roania in the pre-crisis period. For the Czech Republic the constant was close to zero and the slope was negative. The absolute value of the slope estiate is extreely large when the period 2-21 is included in the saple; this was a period in which the Czech koruna appreciated rapidly. For Hungary the slope estiate is also negative (below -1), while the constant is positive. For Roania the slope is positive and the constant is negative. Moreover, the slope is close to one for all of the period before 27 but turned negative later. This suggests that the UIP was satisfied in the transition period when the interest spread was very high, but not in later periods when the spread was reduced. The conclusion fro the estiations in Sections 5 and 6 is that the UIP has liited epirical validity in the saple of CEE countries. Still, there are noticeable differences across the saple countries and across different tie saples. The rest of the paper exaines a nuber of possible reasons for these findings. Transaction costs ay liit arbitrage when the interest rate spread is sall (Section 7) and the risk preiu ay be tie-varying (Section 8). 75

19 7. Non-linearities The size of the interest rate spread ay affect whether or not the UIP hypothesis is supported. Transaction and inforation costs are likely to keep investors fro exploiting deviations fro the UIP when the interest rate spread is sall, but not when the spread is high (Froot & Thaler 199). The conjecture has soe epirical support (Mehl & Cappiella 27, Lothiana & Wu 211). The extree volatility of the FX_CHG variable has ade us pursue a siple and robust way to odel the presence of different regies for different levels of interest rate spreads. We separate the interest spread into two series. Taking the average spread over the saple for each country, two series of interest rate spreads are coputed: the variable INT_SP_LO equals the spread when the spread is lower than the average, and zero otherwise; the variable INT_SP_HI equals the spread when the spread is higher than the average, and zero otherwise. Both spread variables are included in the UIP specification: FX_CHG(3) = α + β LO INT_SP_LO + β HI INT_SP_HI + ε(3) (7) The results of the regressions are reported in Table 4. The results are as expected for Poland and Roania; the slope coefficients for high interest rate spreads are in both cases positive and statistically different fro zero, while the coefficients for low spreads are statistically insignificant. The results are inconclusive for the other three countries; the slope coefficients of the high interest rate spreads are negative and the coefficients are generally estiated iprecisely. Overall, Table 4 provides soe support to the hypothesis that the UIP should hold better when the interest rate spread is large than when it is low, at least for Poland and Roania. Table 4. UIP estiation results, high and low interest rate spread variables ˆ LO ˆ HI ˆ F-stat R 2 Saple Obs. Croatia Czech Republic Hungary Poland Roania (1.355) (1.68) (9.84) (5.111) 5.79 (4.523) (.729).17 (3.328) (1.894).221 (.497) (.462) ** (.225) * (.955) (.934).464 (.221).266 *** (.89) [.] [.] [.] [.2] [.] :2-1999:1-1999:1-21:1-1999:1- Notes: OLS estiation. Newey-West standard errors are shown in round brackets. Superscripts ***, **, * denote that the coefficient estiate is statistically different fro at the 1, 5 and 1% level of significance respectively. The null hypothesis of the F-test is that α =, β L = 1and β H = 1; the p-value is shown in square brackets

20 We have also ipleented two other specifications of the non-linear relation fro the interest spread to the foreign exchange rate change (results not shown). One approach was the sooth transition odel of Granger & Teräsvirta (1993), but we generally had probles estiating the non-linear relation. Another approach was to use a Taylor order approxiation up to the third order of the Granger & Teräsvirta odel and then to estiate coefficients to all the included powers. In any cases the estiated coefficients attained iplausible sign and size and the R 2 of the regressions did not change fro the base case (results not shown). In conclusion, non-linearities see to play only a inor role for the UIP estiations, i.e. transaction and inforation costs are unlikely to be behind the weak support of the UIP for the CEE countries. 8. Risk aversion and financial instability A possible explanation for the low explanatory power of the UIP estiations is that the risk preiu is in fact not constant. We include different proxies of the risk preiu. We start by including the VIX index as a proxy of the risk preiu. The VIX index is an iplied volatility index calculated fro option prices on the S&P5 equity index and is often seen as a ain indicator of risk aversion in global financial arkets. A higher value of the VIX index is tantaount to larger financial uncertainty. We include VIX as an additional explanatory factor in the epirical UIP specification: FX_CHG(3) = α + β INT_SP + γ VIX + ε(3) (8) The results are reported in Table 5. While the R 2 of the estiations do not iprove arkedly, the coefficient of VIX is positive for all the countries and also statistically significant for Croatia and Roania. More financial instability in global financial arkets puts ceteris paribus depreciation pressure on the local currency. The slope coefficients stay largely unchanged, while the constants change sign for three countries, becoing (with the exception of Hungary) negative, but ostly not significant. This suggests that when global risk aversion is taken into account, the tie-invariant reaining part captured by the constant loses its explanatory power. 77

21 Table 5. UIP estiation results, including VIX Croatia ˆ ˆ ˆ F-stat R 2 Saple Obs (1.912) Czech Republic ** (6.133) Hungary Poland Roania (8.785) (9.412) * (6.385) -.65 *** (.23) * (1.232) * (.757) (1.488).271 *** (.82).185 ** (.94).488 (.32).373 (.43).748 (.614).639 ** (.294) [.] [.26] 5.58 [.5] [.318] [.] :2-1999:1-1999:1-21:1-1999:1- Notes: OLS estiation. Newey-West standard errors are shown in round brackets. Superscripts ***, **, * denote that the coefficient estiate is statistically different fro at the 1, 5 and 1% level of significance respectively. The null hypothesis of the F-test is that α = and β = 1; the p-value is shown in square brackets. An alternative easure of risk aversion, less global and ore linked to European foreign exchange arkets, ay be based on other currency pairs in the region. As a rough easure of the external risk aversion affecting currency arkets in Europe, we use the 3-onth return of the Swedish krona against the euro. Sweden had a floating exchange rate throughout the saple period and the exchange rate is likely be affected by currency arket pressures. The estiated equation is the following, where SWE_FX_CHG denotes the annualised 3-onth depreciation of the Swedish krona against the euro: FX_CHG(3) = α + β INT_SP + δ SWE_FX_CHG(3) + ε(3) (9) The results are reported in Table 6. The R 2 are higher and the coefficients of the Swedish krona return are always statistically significant (with the exception of the results for Croatia) and have positive signs. It sees that including the currency pressure on the Swedish krona gives the sae overall result as was given when the VIX variable were included, but in an arguably stronger way. Unlike in the equation with VIX, the constants becoe insignificant, with the exception of the one for the Czech Republic, where the constant is still significant and negative

22 Table 6. UIP estiation results, including change in Swedish krona foreign exchange rate Croatia Czech Republic Hungary Poland Roania ˆ ˆ ˆ F-stat R 2 Saple Obs (.915) -3.5 ** (1.449) (5.655) (5.75) (2.758) * (.221).147 (.846) (.754) -.34 (1.168).324 *** (.68).94 (.71).484 *** (.173).61 *** (.211) *** (.312).87 *** (.129) 3.88 [.] 5.42 [.8] [.1].736 [.481] [.] :2-1999:1-1999:1-21:1-1999:1- Notes: OLS estiation. Newey-West standard errors are shown in round brackets. Superscripts ***, **, * denote that the coefficient estiate is statistically different fro at the 1, 5 and 1% level of significance respectively. The null hypothesis of the F-test is that α = and β = 1; the p-value is shown in square brackets. Concluding this section, the two indicators of risk aversion in international financial arkets see to exhibit substantial explanatory power. The estiated coefficients attain the expected sign and are statistically significant in any cases. The addition of these risk aversion easures, however, does not change the conclusions about the estiated slope coefficient, but has, as expected, an ipact on the constant ter, which becoes statistically insignificant Suary This paper presented the results of epirical tests of uncovered interest parity in Croatia, the Czech Republic, Hungary, Poland and Roania during the first decade of the 21 st century. The objective was to exaine whether the UIP would obtain epirical support in this particular saple, and to ascertain to which extent the convergence process and the global financial crisis have affected the UIP relation. We proceeded fro siple estiations of the link between the return on 3-onth exposure to local currencies against the euro and the spread between local interest rates and Euribor. The stability of the estiated paraeters was analysed using rolling windows. The analysis exained the iportance of a nuber of issues that ay affect the results. Estiations took into account the possibility of different regies depending on the size of the interest rate spread. Various indicators of risk and risk aversion were included, chiefly to capture the effect of the global financial crisis. The ain results are suarised below For the Czech Republic, Hungary and Poland we tried to use the Exchange Market Pressure (EMP) index in Filipozzi & Harkann (21). The coefficients of the EMP index were not statistically significant (not reported). 79

23 The basic odel used to test the UIP in the CEE countries gave a result in line with ost of the previous literature, naely that the UIP relation cannot be supported in general. The forward preiu anoaly is confired in the present saple of Central and Eastern European countries; the estiated slope coefficient is negative in all cases except Roania. Rolling window regressions showed that the coefficient estiates generally are unstable and depend on the choice of saple. The rolling regressions also cast soe light on the effect of global financial crisis on the UIP relations in the five CEE countries. At least for the Czech Republic, Hungary and Roania, there is a clear change after the crisis as the explanatory power of the UIP regressions drops draatically after 27. Transaction and inforation costs do not see to affect the UIP estiations in ways which can be clearly discerned through the inclusion of non-linearities in the UIP relation. It is clear, however, that the iportance of the interest rate spread varies between low and high interest rate spread regies, but the picture is not unifor across the saple countries. For Poland and Roania, the slope coefficient is positive when the interest rate spread is large, although the estiate is still statistically different fro one. There is substantial evidence suggesting that the risk preiu is not constant. Both the global volatility index VIX and the oveents in the Swedish exchange rate see to exhibit substantial explanatory power although not syetrically across all five countries. This suggests that global risk factors have considerable ipact on the liquidity of financial arkets and the arbitrage processes underlying the UIP in the five countries fro Central and Eastern Europe. References 1. Alexius, A. (21): Uncovered interest parity revisited, Review of International Econoics, Vol. 9, No. 3, pp Alper, C. E., O. P. Ardic & S. Fendoglu (29): The econoics of uncovered interest parity condition for eerging arkets: a survey, Journal of Econoic Surveys, Vol. 23, No. 1, pp Baillie, R. T. & T. Bollerslev (2): The forward preiu anoaly is not as bad as you think, Journal of International Money and Finance, Vol. 19, No. 4, pp Bansal, R. & M. Dahlquist (2): The forward preiu puzzle: different tales fro developed and eerging econoies, Journal of International Econoics, Vol. 51, No. 1, pp Benassi-Quere, A., S. Larribeau & R. MacDonald (23): Models of exchange rate expectations: how uch heterogeneity?, Journal of International Financial Markets, Institutions & Money, Vol. 13, No. 2, pp Booth, P. & Longworth, D. (1986): Foreign exchange arket efficiency tests: Iplications of recent epirical findings, Journal of International Money Finance, Vol. 5, No. 2, pp