Carry Trades and Sovereign CDS Spreads: Evidence from Asia-Pacific Markets

Transcription

1 Carry Trades and Sovereign CDS Spreads: Evidence from Asia-Pacific Markets Ivelina Pavlova* Assistant Professor of Finance School of Business 700 Bay Area Blvd., Box 70 University of Houston Clear Lake, Houston, TX, Tel: (81) Maria E. de Boyrie Associate Professor Department of Finance, MSC 3FIN College of Business, P.O. BOX New Mexico State University, Las Cruces, NM Tel: (575) ; Fax: (575) JEL Classification: F31 * Corresponding author

2 Carry Trades and Sovereign CDS Spreads: Evidence from Asia-Pacific Markets Abstract This paper extends the current literature on currency carry trades by investigating the first and second moment interactions between carry trade returns and changes in sovereign credit default swap spreads. Using a VAR- EGARCH model and a sample of nine Asia-Pacific currencies, we examine the relation between sovereign spreads and carry trade returns with and without the inclusion of the 008 global financial crisis period. Our results show that carry trade returns and sovereign spread changes are negatively correlated, with Granger causality running in both directions. We also find significant volatility spillover effects. High conditional correlation between the currency component of carry trades and sovereign spreads is documented, which is amplified if the crisis period is considered. Global risk affects carry trade volatility more during turbulent periods. JEL Classification: F31

3 Carry Trades and Sovereign CDS Spreads: Evidence from Asia-Pacific Markets 1. Introduction Carry trades are speculative investment strategies in the foreign exchange market, where investors borrow low-yielding (funding) currencies and invest in high-yielding (investment) currencies. While uncovered interest parity hypothesis states that the interest rate differential should be offset by depreciation in the investment currency, empirical studies have shown that the reverse holds, i.e., investment currencies tend to appreciate ( forward premium puzzle, Fama, 1984). Carry trade volume is typically boosted by lower exchange-rate volatility, higher interest-rate differentials and lower investors risk aversion. Brunnermeier et al. (008) find that higher levels of global risk and risk aversion measured by the volatility index (VIX) correspond to unwinding carry trade positions and carry trade losses while emphasizing that crash risk may be the reason why speculators do not take positions that are sufficient to enforce uncovered interest parity. Currency crash risk has been extensively studied in the finance literature (Eichengreen et al., 1995; Frankel and Rose, 1996; Kumar et al., 003, among others) and is largely related to macroeconomic indicators. The strength of a currency is also related to the political and economic stability of a country, as increased political-economic instability could lead to higher sovereign credit risk, currency depreciation and higher currency volatility (Hui and Chung, 011). The authors study the information transmission between sovereign credit default swap (CDS) spreads (reflecting sovereign credit risk) and show a relation between sovereign CDS spreads and euro crash risk during the sovereign debt crisis of A sovereign CDS is an over-the-counter contract that makes a contingent payment in the event of a predefined credit event, therefore protecting investors against losses on sovereign debt. 3

4 The CDS indices are a basket of single country CDSs. In recent years this standardized security has been found to have higher liquidity and lower transaction costs than single name CDS. Theoretically, changes in sovereign CDS spreads should reflect mostly the creditworthiness of the sovereign borrower. In addition to country-specific risk, Longstaff et al. (007) argue that US stock and bond market returns, as well as global volatility, can explain a large portion of the variation in sovereign credit spreads. Sovereign credit risk is found to be the dominant macroeconomic fundamental risk in evaluating carry trade returns by Huang and MacDonand (013). They link carry trade returns to sovereign CDS premia along two dimensions: by including a sovereign risk premia in the term structure of interest rates, and through the valuation channel of global imbalances. The exchange rates of countries with net external debt include risk premia to compensate foreign creditors who are financing balance of payments current-account deficits. With regard to global imbalances, Abhyankar et al. (011) find that the time-varying risk premium in carry trade returns is captured by C-CAPM using Net Foreign Assets (NFA) as a conditioning variable. They show that changes in the trade balance and the net foreign-asset position forecast future portfolio returns or future net exports, where the channels of adjustment to a net export imbalance are either through trade or asset valuation channels. Motivated by earlier studies on the relation between crash risk, carry trades, and sovereign credit risk, we focus on the relation between carry trades and sovereign CDS spreads. Using data from September 008 till August 013 and employing bivariate and multivariate VAR-EGARCH models, we examine the first and second moment interactions between daily returns of carry trade strategies selecting from nine Asian currencies and changes in the itraxx SovX Asia Pacific CDS index. Our paper contributes to the literature on market interactions by focusing on the information 4

5 transmission mechanism and testing whether innovations in CDS markets and currency carry trades can have an impact on volatility asymmetrically. Since carry trades experienced significant position unwinding during the 008 global financial crisis turmoil, we also check the robustness of our results by excluding the financial crisis period. To the best of our knowledge, no previous research investigates Granger causality and volatility spillovers between carry trade returns and sovereign spread changes. The only other study that links carry trades and CDS swaps is by Huang and MacDonald (013), who use an option pricing model to measure carry trade positionunwinding risk and show that sovereign credit risk and position-unwinding risk can explain a large proportion of the variation of cross-sectional carry trade returns. Our findings can be summarized along four dimensions. First, we find bidirectional causality between carry trade returns and sovereign spread index changes. Second, when we focus on the second moment interactions, significant bidirectional volatility spillover effects are documented between sovereign spreads and carry trade returns. Negative asymmetry is shown in CT3 (carry trade strategy using the three highest and three lowest interest rate currencies) return volatility, i.e., negative returns lead to higher volatility of spreads than positive returns. Third, when carry trade returns are broken down into currency return and interest rate components, we find higher conditional correlation between the currency component and the sovereign spread changes than between the interest rate component and spreads. Lastly, we find significant spillover from both the global and idiosyncratic risk components of the CDS spread to carry trades, which is diminished when the crisis period is not included in the sample. The rest of the paper is organized as follows. Section offers a brief survey of the relevant literature, and sections 3 and 4 present the data and methodology. The empirical results are discussed in section 5, while section 6 offers concluding remarks. 5

6 . Literature Review Carry Trade Returns Determinants A number of studies investigate the determinants of carry trade returns. Among the recent ones, Menkhoff et al. (01) examine the cross-section of carry trade returns and find that global foreign exchange risk is a significant determinant. They show that high-yielding currencies have a negative relation with changes in global currency volatility and, as a result, carry trades yield low/negative returns in turbulent times. Lusting et al. (011) also study factors explaining the cross-section of carry trade returns and identify a slope factor in exchange rates, closely related to global equity volatility. Adding to the determinants of carry trade returns, Brunnermeier et al. (008) find a significant relation between carry trades and crash risk. They show that currency crashes are associated with higher levels of the VIX and the TED spread, which indicate funding constraints. A different explanation for carry trade returns is proposed by Burnside et al. (011), who argue that carry trade returns reflect peso problems. Using a smooth transition regression approach, Christiansen et al. (011) investigate currency carry trade returns of the G10 currencies and document a time-varying systematic risk factor. They find that carry trade risk is regime dependent and increases in times of turbulence. To model regimes, they use market risk proxies (FX volatility and VIX) and liquidity proxies (bidask spread and TED spread) as transition variables. The relation between carry returns and the stock market has also been investigated by Tse and Zhao (01). Using a bivariate EGARCH model and G10 currencies data, they document significant volatility spillover from the US stock market to G10 carry trade returns, with higher correlation during turbulent times. Granger causality is not detected in either direction. Tse and 6

7 Zhao (01) show that in both carry trades and stock markets, negative innovations have greater impact on volatility than do positive innovations. Currency Returns and Sovereign CDS The relationship between currency options volatility, risk premiums in sovereign CDS markets and the VIX is modeled by Pan and Singleton (008). Using data on Mexico, Turkey and Korea in their analysis, they find that the VIX is significantly related to sovereign CDS risk. Another study on currency options and sovereign CDS is by Carr and Wu (007). They use data for Mexico and Brazil and show that sovereign spreads have a strong relation with option implied volatility. Pu and Zhang (01) take a different perspective and investigate the relation between dualcurrency sovereign spreads and exchange rate returns. They find that the CDS spread differential contains information for predicting exchange rates and plays an important role in price discovery. Another study related to sovereign CDS spreads is by Hui and Chung (011), which focuses on crash risk and sovereign CDSs during the sovereign debt crisis of They study the five-year sovereign CDS spreads of 11 countries in the euro area and dollar-euro option implied volatility and find that the creditworthiness of the countries in the study is an important determinant of euro crash risk during the period studied. Hui and Chung (011) also detect information transmission from the CDS to the currency options market.. The only recent study that directly links carry trade returns and sovereign risk premia is by Huang and MacDonald (013). They build a model to measure carry trade position-unwinding risk and show that sovereign credit risk and position-unwinding risk can explain a large proportion of the variation of cross-sectional carry trade returns. Using data for 6 countries spanning and an option-pricing model to measure position-unwinding risk, Huang and McDonald 7

8 show that sovereign credit risk is the dominant factor in cross-sectional carry trade returns, where high-yielding currencies load positively on sovereign risk and low-yielding currencies could be used as a hedge against this risk. They also propose a carry trade strategy immunized from currency crash risk. Our paper differs from Huang and MacDonald in three important aspects: 1) we focus on market linkages and the direction of influence between CDS and currency markets; ) we study the information transmission mechanism within these markets by examining the volatility spillover effects and testing for asymmetric response to positive and negative innovations; and 3) we use data for the Asia-Pacific region, which includes some of the most actively used currencies in carry trades. 3. Data Exchange Rates and Interest Rates In our study we use the currencies of nine Asian countries that are included in the itraxx SovX Asia Pacific index. We gather daily spot- and three-month forward closing rates from WM/Reuters (through Datastream) on the Australian dollar (AUD), Japanese yen (JPY), Thai baht (THB), Malaysian ringgit (MYR), New Zealand dollar (NZD), Indonesian rupiah (IDR), South Korean won (KRW), Philippine peso (PHP) and Vietnamese dong (VND) versus the US dollar. We also collect three-month interbank rates from Datastream for those nine countries and the US. Following Brunnermeier et al. (008) and Christiansen et al. (011), we calculate the exchange rate return in excess of the lagged interest rate differential of a given country and the US: rr kk tt = ss kk tt ss kk kk UUUU tt 1 + ii tt 1 ii tt 1 (1) 8

9 kk kk where ss tt is the log of the nominal exchange rate of currency k versus the USD, ii tt 1 is the one- UUUU day lagged log interest rate of country k and, ii tt 1 is the one-day lagged log US interest rate. In this way we obtain the return in excess of the return predicted by uncovered interest parity. Carry trade portfolio returns are obtained in a way similar to the strategy followed by Deutsche Bank s G10 Currency Harvest Index and used by Christiansen et al. (011) and Tse and Zhao (01). The strategy involves taking a short position in the three currencies with the lowest interest rates and a long position in the three currencies with the highest interest rates, equally weighted. The currencies are ranked every three months by their interest rates and the portfolio is rebalanced accordingly. We denote the return on the portfolio with a short position in the three currencies with the lowest interest rates and a long position in the three currencies with the highest interest rates by CT3 and we also calculate CT1, which used only the lowest and highest yielding currency (following Tse and Zhao, 01). Typically, the low-yielding currencies from our data sample are the Japanese yen and Thai baht, and the high-yielding currencies are the Indonesian rupiah and the Australian dollar. The third currency varies in both the long- and short legs of the carry trade portfolio. CDS Index Data The Markit itraxx SovX Asia Pacific index daily data for the five-year CDS maturity are obtained from Datastream for September, 008 August 19, 013. The limited availability of the SovX data limits our sample to 1,8 observations for almost five years of daily data. The countries included in the SovX Asia Pacific index are Australia, Japan, Thailand, Malaysia, New Zealand, China, Indonesia, Korea, Philippines and Vietnam. 1 The key benefit of using index data is the higher liquidity compared to using single-name, sovereign CDS data for the countries in our 9

10 sample. The index is tradable, equally weighted and rolls every six months based on the liquidity of the constituents. The volume of trading in credit indices has increased significantly in recent years, due to lower transaction costs, greater transparency and operational efficiency. Coupon payments of CDS indices are made from the protection buyer to the protection seller and are paid quarterly. The SovX Asia Pacific index trades on spread. The buyer of the SovX Asia Pacific index takes on the credit exposure (i.e., it is similar to buying a portfolio of bonds). The buyer of the index sells protection to the seller and receives the coupon payment. Selling the index is equivalent to transferring credit exposure to another party. 4. Methodology To model the dynamic relation among carry trades returns, sovereign CDS spread changes and volatility we use a multivariate VAR-EGARCH model following Koutmos and Booth (1995) and Koutmos (1996): nn YY ii,tt = ii,oo + jj=1 iiii YY jj,tt 1 + εε ii,tt ffffff ii, jj = 1,, nn () = eeeeee ii,0 + jj=1 ii,jj ff jj zz jj,tt 1 + ii ln (σσ ii,tt 1 ) ffffff ii, jj = 1,, nn (3) σσ ii,tt nn ff jj zz jj,tt 1 = ( zz jj,tt 1 EE zz jj,tt 1 + jj zz jj,tt 1 ) ffffff ii, jj = 1,, nn (4) σσ ii,jj,tt = ρρ ii,jj σσ ii,tt σσ jj,tt ffffff ii, jj = 1,, nn aaaaaa ii jj (5) Equation () models the dynamic relation between variables as a vector autoregression (VAR). YY ii,tt denotes a time series vector of returns/changes at time t for variable i,. For example, ii = 1, denotes 1 = the returns of the carry trade strategy (CT1 or CT3), and = changes in the spread ( Spread). The conditional mean of each variable is a function of its own lagged values, as well 10

11 as the lagged values of the other variables. The iiii coefficients represent the lead-lag relationships in the model (for ii jj). We use a bivariate, as well as a multivariate, VAR in our analysis. To test the causality from carry trades to sovereign spreads and vice versa we impose restrictions on the coefficients as follows: HH 0,1 : iiii = 0 ffffff ii, jj = 1,, nn aaaaaa ii jj (6) Equation (3) is used to model the conditional variance and examine the volatility spillover effects. We employ a multivariate EGARCH model, in which the conditional variance is an exponential function of its own lagged and other market lagged standardized innovations. The conditional variance and covariance are denoted by σσ ii,tt and σσ ii,jj,tt, respectively, εε ii,tt is the innovation at time t, and zz ii,tt = εε ii,tt /σσ ii,tt is the standardized innovation. In this equation, both the lagged value of the conditional variance and cross-market standardized innovations are allowed to impact the volatility of a given market. For volatility to spill over from one market to another, ii,jj, for i j, must be significantly different from zero. Equation (4) presents the functional form of ff jj zz jj,tt 1, which allows the standardized innovation and the other market innovations to affect the conditional variance asymmetrically. In particular, zz jj,tt 1 EE zz jj,tt 1 measures the magnitude (size) effect and jj zz jj,tt 1 the sign effect and jj measures the asymmetric impact on the volatility of the market under study. This impact is measured as: ff jj zz jj,tt zz jj,tt = 1 + jj, ffffff zz jj > 0 ff jj zz jj,tt zz jj,tt = 1 + jj, ffffff zz jj < 0 (7) A negative jj, combined with a positive and significant ii,jj, suggests that negative return shocks in one market increase volatility more than positive returns do in the other market; while a 11

12 negative zz tt accompanied by a negative jj increases the magnitude effect. Volatility spillovers across the markets are measured by ii,jj for ii, jj = 1,, nn and ii jj. The persistence of volatility is measured by ii in equation (3), where a ii equal to one can be interpreted as the nonexistence of an unconditional varianceand a conditional variance that follows an I(1) process The conditional covariance specification is presented in equation (5), where ρρ ii,jj represents the cross-market correlation coefficients between volatility of returns and σσ ii,tt and σσ jj,tt represent the conditional covariance between the markets, and suggests that the correlations between markets are constant. The log likelihood for the multivariate VAR-EGARCH model, assuming normality, is presented as follows: TT LL(ΘΘ) = 0.5(NNNN) llll(ππ) 0.5( tt=1 llll SS tt + tt=1 εε tt SS 1 tt εε tt ) (8) TT where N is the number of equations, T is the number of observations, Θ is the parameter vector of the model, εε tt is the vector of innovations at time t, and SS tt is the time-varying conditional variancecovariance matrix. In the estimation we follow Koutmos (1996), who uses numerical maximization techniques to estimate the log likelihood function and applies the BHHH (Berndt et al., 1974) algorithm. The log likelihood is estimated via Quasi maximum likelihood estimation (QMLE). Bollerslev and Wooldridge (199) show that when the first two conditional moments are correctly specified, the robust standard errors computed via QMLE are valid under nonnormality. Two widely cited explanations of asymmetric volatility, or why volatility increases when the past return innovation is negative, are the leverage- and the volatility-feedback effects. The leverage effect (Black, 1976; Christie, 198) suggests that a negative stock return implies a drop in the value of equity, which increases firm leverage, and this leads to higher stock return volatility. Figlewski and Wang (000) among others, argue that the leverage effect is not due to firm leverage 1

13 and should be called a down market effect. The volatility feedback effect (Pindyck, 1984; French et al., 1987) implies that an increase in volatility raises the expected future stock returns, which in turn causes stock prices to fall and amplifies the initial negative return. An alternative explanation of asymmetric volatility in stock markets related to trading activity is proposed by Avramov et al. (006). While the evidence in equity markets overwhelmingly supports the notion of asymmetric volatility, the evidence in currency markets is mixed. One of the reasons is the two-sided nature of exchange rates (Wang and Yang, 009); a positive return shock to one currency is a negative shock for the other. The only recent paper studying volatility transmission in carry trades is by Tse and Zhao (01). Using a bivariate EGARCH model, they document asymmetric volatility in the case of carry trades and the stock market, with volatility spillovers from the US stock market to carry trade returns. 5. Empirical Findings Descriptive Statistics In Table 1 we report summary statistics for the variables under study. The average return of the carry trade portfolios (CT1 and CT3) is consistently negative when studying the full period but turns positive when the crisis period (September 3, 008 June 30, 009) is excluded. The mean changes in the sovereign spread index are negative both in the full sample, as well as when the crisis period is not included. Invariably the mean return and standard deviation of the carry trade strategy including only the highest and lowest interest rate currencies (CT1) are greater than that of the strategy including the three highest and three lowest interest rate currencies (CT3) and of the change in spread. Similar results are found while examining the measures of skewness and kurtosis in that the return series are negatively skewed (a sign of the carry trade s crash risk) and highly leptokurtic with respect to the normal distribution in the full period, but mostly positive but 13

14 still leptokurtic otherwise. The Kolmogorov-Smirnov test rejects normality for each of the series at the 1% level of significance. Table 1 also reports statistics on the currency return and interest rate return components of the carry trade portfolios. For the CT1 portfolio we denote the interest rate component as CT1Int and the currency component as CT1FX. The mean of the interest rate component appears positive for both CT portfolios, even when the crisis is excluded. The mean of the currency component is consistently negative, with a much higher standard deviation. Lastly, we present descriptive statistics on our measure of global and regional risk. We use changes in the VIX as a proxy for global risk and the residual from a regression of the CDS spread on VIX as a proxy for Asia-Pacific regional risk, not correlated with global risk factors. The standard deviation of changes in the VIX is much higher than the standard deviation of the regional risk factor and changes in CDS spread. Consistently, correlation between the variables show that CT1 and CT3 are significantly negatively correlated with the change in spread much like the change in VIX is negatively correlated to our regional risk variable when the crisis period is excluded. CT1FX and CT3FX are also negatively correlated with change in VIX and the regional risk variable. The highest correlations are found when we study the full period. Causality Results Granger (1969) causality tests are used to establish the causal relations between variables. The results presented in Table point to a feedback relationship between carry trade returns and changes in spread in all instances. Given that the purpose of the test is to determine if lagged values of one variable can be used to forecast another variable, the results obtained emphasize the idea that CDS spreads have predictive power over future carry trade positions and vice versa. Our findings on causality running from sovereign CDS spreads to carry trade returns can be related to 14

15 the recent study by Huang and MacDonald (013) who show that sovereign credit risk is one of the main determinants of cross-sectional carry trade returns and high-yielding currencies load positively on sovereign default risk. They also show that sovereign credit risk Granger-causes country-specific volatility and liquidity risk. On the other hand, our results indicating that causality runs from carry trades to sovereign spreads could be explained by the information about investors attitude towards global risk contained in carry trade. Attinasi et al. (009) show that 56% of the daily changes in sovereign bond spreads can be explained by international risk aversion. In bad times, when banks become highly risk averse, speculators involved in carry trades may face a liquidity squeeze, which can result in unwinding of carry trades. Furthermore, the volatilities of carry trades and sovereign spreads could be affected by a similar set of factors including liquidity shocks, funding constraints and risk premium differences between markets. Longstaff (010), studying the mechanism of spillover between markets, shows that, during the crisis period, market activity and funding-liquidity effects were a major part of the volatility transmission mechanism between markets. Volatility Spillover Between Carry Trades and Sovereign Spreads Koutmos and Booth s (1995) multivariate VAR-EGARCH model for volatility spillover is used in order to detect linkages between the return of two carry trade strategies and changes in the sovereign risk premia while analyzing two periods: the full period that includes the financial crisis of 008, and a shorter period that excludes the crisis. The analysis was performed using these two periods in order to determine if the crisis period influenced our results. The maximum likelihood estimators are reported in Tables 3 and 4, respectively. In terms of first order interdependence (conditional means, β) reported in Panel A of Tables 3 and 4, there exist significant lead-lag effects. When the entire period is considered in Table 3, 15

16 both CT1 and CT3 are influenced by past levels of the CDS spread, with feedback from carry trades to spread in both cases. However, the results differ when the crisis period is excluded in Table 4 in that only lagged values of carry trade influence spread. An important point to note is that in most instances the feedback effects are negative. Focusing on the second moment interdependencies in Tables 3 and 4 (volatility spillover, α), we find that during both periods under study the conditional variance of carry trade returns (CT1 and CT3) and changes in spread are all affected by their own innovations as well as past innovations generated by the other variable, except in the case of CT3, when the crisis period is excluded, where there is no volatility spillover from spread to carry trades. Significant spillover is documented from carry trades to the sovereign CDS index spread in all models. The coefficient measuring asymmetry (δ) is negative and significant in the case of CT1 and CT3 under both periods. The coefficient measuring asymmetry is positive and significant in the case of the sovereign spread in the whole sample, in the one excluding the crisis, and in both models including CT1 and CT3. In all instances the degree of volatility persistence (γ) is high and significant. Given that coefficients are close to unity, the hypothesis of unit root in each series cannot be rejected and the possibility of omitted variables exists. The low correlations between the portfolios and changes in spread, as explained by Ebrahim (000), indicate the possibility that the spillovers are due to contagion effects or that investors process the information generated by the sovereign risk premia. Based on the Ljung-Box statistics, neither linear nor nonlinear dependence in the standardized residuals can be found. This would imply that the bivariate VAR-EGARCH model employed reasonably explains the relations between carry trade and the sovereign risk premia. In addition, the cross product of the standardized residual test further demonstrates the existence of no serial correlation. 16

17 Panels C and D of Tables 3 and 4, respectively, show the degree of volatility persistence and the degree of the asymmetric impact of negative and positive innovations. The volatility shocks, based on the half-life of a shock measured as ln(0.5) /ln ( ii ), lasted longer in the period including the global financial crisis for both currency carry trade returns and CDS spreads compared to the period excluding the crisis. The degree of asymmetric impact, 1 + ii (1 + ii ), a measure of the number of times that a negative innovation increases volatility more than a positive innovation, shows that negative movements in CT1 and CT3 have a greater effect on conditional volatility than positive changes. The degree of asymmetry is weaker for sovereign spread changes. The diagnostics on the standardized residuals reported in Tables 3 and 4 suggest that the model explains well the interactions between the markets. The Ljung-Box statistics up to 1 lags are insignificant, showing no evidence of serial correlations in the residuals. The assumption of constant conditional correlation appears to be a valid specification for the variance-covariance structure of the carry trade returns and sovereign spreads. Panel E of Tables 3 and 4 reports the total impact of the effect of negative or positive shocks within one market on volatility in the other market. The impact of negative innovations from the carry trade returns (CT1 and CT3) on CDS spread volatility is larger than the impact of positive innovations. For instance, 1% positive innovation (i.e., ii,jj (1 + jj )) in carry trade returns (CT1) increases spread volatility by %, while 1% negative innovation (i.e., ii,jj 1 + jj ) in carry trade returns increases spread volatility by %. This is true whether we examine the whole period under study or just the crisis period. The results for CT3 point in the same direction. Negative innovations in spread, however, tend to increase CT1 volatility less than positive innovations do. As such, informational asymmetry is shown to exist between the carry 17

18 trade returns and spread. Note that, considering the entire period in Panel E of Table 3, the impact of +1% change in the sovereign CDS spread brings about % change in CT1, while the percentage change in CT3 is only %. Similar results are obtained in Table 4 where the crisis is excluded. Currency and Interest Rate Return Decomposition To investigate whether the volatility spillover between sovereign CDS index spreads and carry trades returns comes from the interest rate differential or the currency return, we decompose carry trade returns into foreign exchange (CTFX) and interest rate (CTInt) components. 3 The estimates of the multivariate VAR-EGARCH model for the whole period, as well as for the subsample excluding the financial crisis, are presented in Tables 5 and 6, respectively. Focusing first on the results for the trading strategy involving the highest- and lowest-yielding currencies (CT1) we document a significant bidirectional volatility spillover between sovereign spread changes and both the interest rate and the currency component of carry trade returns. This results holds both when the entire sample period is considered (Panel A in Table 5), and also when the crisis is excluded (Panel A in Table 6). When returns on the carry trades using the three highest- and lowest-yielding currencies are considered, we do not show spillover from sovereign spreads to the interest rate differential but we show a significant volatility effects between the FX component and spreads. This result is confirmed by causality results, which show significant bidirectional causality between spreads and the currency component of carry trade returns, but not between the interest rate component and spreads. The causality results are not reported for brevity and are available from the authors upon request. Panel B of Tables 5 and 6 shows the conditional correlation matrix. In Table 5, when the entire sample is considered, the correlation between the CT1FX component and sovereign spread 18

19 is -9.75% and the correlation between the CT3FX component and spread is -3.64%. The correlation between CDS spread and the interest rate differential is much lower at -1.78% for CT1Int and -.48% for CT3Int. Similar results are found in Table 6 where the crisis period is excluded. We also examine the impact of a shock in one series on the conditional volatility of the other series in Panel C of Tables 5 and 6. Positive innovations in spread bring about larger changes in interest rate differential volatility than negative innovations do when the entire period is considered. Global Versus Idiosyncratic Effects of CDS Spreads Studies on carry trades and stock market returns show positive correlation between the two and cite the relation of both to investors attitude toward risk as an explanation of the positive relation (Tse and Zhao, 01). In our study, we focus on sovereign CDSs s, and a good question in this case is how much of the relation between carry trades and sovereign spreads is due to exposure to global volatility, including stock market volatility. 4 A recent study by Longstaff et al. (007) on sovereign CDS spreads argues that sovereign spreads are more related to global risk, US stock and high-yield bond markets than to regional- and country-specific forces. To disentangle the global and idiosyncratic effects in the SovX Asia Pacific index we run a regression of changes in the SovX Asia Pacific index spread on changes in the CBOE volatility index VIX. We save the residual of this regression and use it as a proxy of the idiosyncratic risk component of sovereign CDS spreads, which is not due to global risk factors proxied by the VIX. We name the new variable regional risk. Tables 7 and 8 report results from the multivariate VAR-EGARCH model for the entire sample period and for the subsample excluding the crisis, respectively. Looking at the volatility feedback for CT1 in Table 7, we document a significant spillover from both the VIX and the 19

20 regional risk component of spreads. Regional risk volatility is also affected by changes in VIX and CT1 returns, while VIX shows no feedback with it. When the crisis period is excluded in Table 8, CT1 volatility appears influenced by regional, but not global, risk. The results for CT3 also show significant spillover from both VIX and regional risk to CT3 when the entire period is considered (Table 7), but no such effect is shown in Table 8. The conditional correlation points to significant negative correlation between both CT1, VIX and regional risk in CDS spreads in Table 7. The result is similar for CT3. The correlation between carry trade returns and global and idiosyncratic risk is much lower after July 009 (Table 8, panel B). We also show an asymmetric impact of VIX and regional risk factors on carry trade returns in Panel C of Tables 7 and 8. Positive changes in the VIX and in regional risk lead to higher volatility of carry trades than negative changes do. Another interesting finding is that when the carry trade uses only the currencies with the highest- and lowest-interest rate (CT1), the impact of a positive innovation in the VIX is lower than the impact of a positive innovation in the regional risk factor (e.g., % versus % in Panel C of Tables 7 and 8). The opposite is true when we study CT3 returns. When more currencies are involved in the carry trade strategy, 1% change in global risk leads to % change in CT3 volatility, as opposed to % impact of regional risk on CT3 volatility (Table 7, Panel C). Single Currency Carry Trade Analysis In our next analysis we focus on the individual currency carry trade returns versus the US dollar for the currencies used in the CT1 and CT3 portfolios. The currencies included are those of Japan, Malaysia, Thailand, Indonesia, New Zealand, Australia and Korea. We calculate single currency carry trade return versus the USD. Japan, Malaysia, Thailand and Indonesia s currencies are mostly in the funding currencies, while New Zealand, Australia and Korea are more often in the 0

21 high-interest rate group. Table 9 reports selected volatility spillover coefficients ( iiii ) from the model presented in equations () through (5) above. However, in this instance, i,j=1,,8 and 1 denotes single currency carry trade return versus the USD for Japan, Malaysia, 3 Thailand, 4 Indonesia, 5 New Zealand, 6 Australia, 7 Korea and 8 changes in the SovX Asia Pacific index spread in equation (5). We present results both for the full sample and for when the crisis period is excluded. We focus on spillover between single currencies and sovereign spread changes. Using data for the full period, there is significant volatility spillover going in both directions from each single currency carry trade return to spread and vice versa. When the crisis is excluded, then significant spillover is only documented from the sovereign spread changes to Thailand and Indonesia. Thailand and Indonesia s currencies are two of the currencies typically included in the low-interest rate group of the CT3 portfolio and also two of the countries with lower GDP compared to the rest (Japan, New Zealand, Australia, Korea). In terms of spillover to sovereign spread, there is no clear pattern when the crisis is excluded. The spread's volatility is mostly affected by Malaysia, Thailand, Korea and Australia. The results on spillover between single currencies and sovereign spread changes during the crisis may be linked to the literature on funding liquidity risk. Along the CDS dimension, Fontana and Scheicher (010) show that sovereign CDS premia increased when the funding liquidity decreased during the global financial crisis. They studied the CDS spreads of 10 European countries from 006 till 010. Along the currency carry trade dimension, Ferreira Filipe and Suominen (014) find that the measures of funding risk in Japan that they used can explain a large portion of carry trade returns. Brunnermeier and Pedersen (009) show that market liquidity changes as the market goes up or down, as funding conditions change along with the market. 1

22 6. Conclusions This paper extends the current literature on foreign currency carry trades by examining the first and second moment interactions between carry trade returns for two carry trade portfolios of Asia- Pacific currencies and sovereign risk premia measured by changes in the CDS spread index. Using a bivariate VAR-EGARCH model proposed by Koutmos and Booth (1995), we model market interdependence and information transmission asymmetric effects. From the analysis of two distinct periods (one which encompasses the 008 global financial crisis period) we are able to deduce that with reference to the series means, bidirectional causality exists between returns to the CT1 portfolio and changes in the Markit itraxx SovX Asia Pacific index spread. We also find that the two markets are significantly correlated and that correlations between the markets are higher when the crisis period is included in the sample. With regard to the volatility spillover between carry trade returns and Asia-Pacific CDS spreads, which is found to be mostly significant, once again the portfolio constructed using three high-yielding and three low-yielding currencies (CT3) behaves differently than that constructed using the highest- and lowest-yielding currencies (CT1). There is significant volatility spillover from carry trades to the sovereign CDS index, but spillover from CDS to carry trades is significant only in the CT1 portfolio. The impact of negative innovations of carry trade returns on CDS spread volatility is larger than the impact of positive innovations, while positive changes in sovereign spreads (higher sovereign risk) affect carry trade conditional volatility more than negative changes. We also investigate whether the spillover arises from the currency or interest rate component of carry trades. Our findings show higher conditional correlation between the currency component and spreads, as well as significant volatility spillover between the currency components of CT3 and sovereign spreads. We also look into the global versus idiosyncratic risk component

23 of the Markit itraxx SovX Asia Pacific index spread and find that when the crisis period is included, volatility goes from both regional- and global risk to spreads. When the crisis is excluded, CT1 return volatility is more affected by regional risk, while CT3 volatility is not influenced by either. Lastly, examining the interaction between individual currency carry trade returns and sovereign spread changes shows significant spillover when the crisis is included, but no clear pattern of spillover when the crisis is excluded (low- versus high-yielding currencies). Overall, this study demonstrates that information flows between carry trade returns (whether one or three funding and investment currencies are used) and the sovereign CDS index in the Asian market. The results from the bivariate VAR-EGARCH model further show that while these two markets are integrated, carry trades play an important role in information transmission. As such, investors and fund managers should consider this information when investing in the Asian markets. 3

24 References Abhyankar, A., Gonzalez, A., & Klinkowska, O. (011). Salvaging the C-CAPM: Currency carry trade risk premia and conditioning information. Retrieved from Attinasi, M. G., Checherita-Westphal, C. D., & Nickel, C. (009). What explains the surge in Euro area sovereign spreads during the financial crisis of ? (December 17, 009). ECB Working Paper No Avramov, D., Chordia, T., & Goyal, A. (006). The impact of trades on daily volatility. Review of Financial Studies, 19, Berndt, E. K., Hall, B. H., & Hall, R. E. (1974). Estimation and inference in nonlinear structural models. Annals of Economic and Social Measurement, 3 (4), : National Bureau of Economic Research. Black, F. (1976). Studies of stock market volatility changes, Proceedings of the 1976 Meetings of the American Statistical Association, Bollerslev, T. & Wooldridge, J. (199). Quasi-maximum likelihood estimation and inference in dynamic models with time-varying covariances. Econometric Reviews, 11, Brunnermeier, M. K., Nagel, S., & Pedersen, L. H. (008). Carry trades and currency crashes. Working Paper No Retrieved from National Bureau of Economic Research website: Brunnermeier, M. K., & Pedersen, L. H. (009). Market liquidity and funding liquidity. Review of Financial Studies,, Burnside, C., Eichenbaum, M., Kleshchelski, I., & Rebelo, S. (011). Do peso problems explain the returns to the carry trade? Review of Financial Studies, 4, Carr, P., & Wu, L. (007). Theory and evidence on the dynamic interactions between sovereign 4

25 credit default swaps and currency options. Journal of Banking & Finance, 31, Christiansen, C., Ranaldo, A., & Söderlind, P. (011). The time-varying systematic risk of carry trade strategies. Journal of Financial and Quantitative Analysis, 46, Christie, A. A. (198). The stochastic behavior of common stock variances: Value, leverage and interest rate effects. Journal of Financial Economics, 10, Ebrahim, S. K. (000). Volatility transmission between foreign exchange and money markets. Working Paper No Retrieved from Bank of Canada website: Eichengreen, B., Rose, A. K., Wyplosz, C., Dumas, B., & Weber, A. (1995). Exchange market mayhem: the antecedents and aftermath of speculative attacks. Economic Policy, 10, Fama, E. F. (1984). Forward and spot exchange rates. Journal of Monetary Economics, 14, Ferreira Filipe, S., & Suominen, M. (014). Currency carry trades and funding risk. Working paper. AFA 014 Philadelphia Meetings. Available at SSRN: Figlewski, S., & Wang, X. (000). Is the 'Leverage Effect' a leverage lffect? Retrieved from Fontana, A., & Scheicher, M. (010). An analysis of euro area sovereign CDS and their relation with government bonds European Central Bank working paper (No. 171). Frankel, J. A., & Rose, A. K. (1996). Currency crashes in emerging markets: An empirical treatment. Journal of International Economics, 41, French, K. R., Schwert, G. W., & Stambaugh, R. F. (1987). Expected stock returns and volatility. 5

26 Journal of Financial Economics, 19(1), 3-9. Granger, C. W. J. (1969). Investigating causal relations by econometric models and crossspectral methods. Econometrica, 37, Huang, H., & MacDonald, R. (013). Currency carry trades, position-unwinding risk, and sovereign credit premia. Retrieved from Hui, C.-H., & Chung, T.-K. (011). Crash risk of the euro in the sovereign debt crisis of Journal of Banking & Finance, 35, Koutmos, G., & Booth, G. (1995). Asymmetric volatility transmission in international stock markets. Journal of International Money and Finance, 14, Koutmos, G. (1996). Modeling the dynamic interdependence of major European stock markets. Journal of Business Finance & Accounting, 3, Kumar, M., Moorthy, U., & Perraudin, W. (003). Predicting emerging market currency crashes. Journal of Empirical Finance, 10, Longstaff, F.A. (010). The subprime credit crisis and contagion in financial markets. Journal of Financial Economics, 97, Longstaff, F. A., Pan, J., Pedersen, L. H., & Singleton, K. J. (007). How sovereign is sovereign credit risk? Working Paper No Retrieved from National Bureau of Economic Research website: Lustig, H., Roussanov, N., & Verdelhan, A. (011). Common risk factors in currency markets. Review of Financial Studies, 4, Menkhoff, L., Sarno, L., Schmeling, M., & Schrimpf, A. (01). Carry trades and global foreign exchange volatility. The Journal of Finance, 67,

27 Pan, J., & Singleton, K. J. (008). Default and recovery implicit in the term structure of sovereign CDS spreads. The Journal of Finance, 63, Pindyck, R. S. (1984). Risk, inflation, and the stock market. The American Economic Review, 74, Pu, X., & Zhang, J. (01). Can dual-currency sovereign CDS predict exchange rate returns? Finance Research Letters, 9, Tse, Y., & Zhao, L. (01). The relationship between currency carry trades and US stocks. Journal of Futures Markets, 3, Wang, J., & Yang, M. (009). Asymmetric volatility in the foreign exchange markets. Journal of International Financial Markets, Institutions and Money, 19,

28 Table 1. Descriptive Statistics Panel A. Descriptive Statistics Full Sample (9/3/008 8/16/013) Statistics CT1 Regional CT1FX CT1Int CT3 CT3FX CT3Int Spread VIX Risk Mean Std. dev Skewness Kurtosis K-S N/A N/A LB(1) for RR tt LB(1) for RR tt Panel B. Descriptive Statistics Excluding Crisis (7/01/009 8/16/013) Statistics CT1 CT1FX CT1Int CT3 CT3FX CT3Int Spread VIX Regional Risk Mean Std. dev Skewness Kurtosis K-S N/A N/A LB(1) for RR tt LB(1) for RR tt Notes: The table reports descriptive statistics and correlations for carry trade returns (CT1 and CT3) in percent, and sovereign CDS index spread in percent. The statistics for the full sample and the subsample excluding the global financial crisis period are reported separately. symbolizes change. Spread represents the sovereign risk premia. CT1 is the carry-trade portfolio with one lowest- and highest-interest rate currencies and CT3 is the carry-trade portfolio with three lowest- and highest-interest rate currencies. The Kolmogorov-Smirnov (K-S) test is a test for normality. The Ljung-Box (LB) statistic for 1 lags is distributed as χχ. 8

29 Table. VAR Granger Causality CT1, CT3 and Change in Spread CT1 Spread CT3 Spread Full Sample (9/3/008 8/16/013) Excluding Crisis (7/01/009 8/16/013) df CT1 Spread df CT1 Spread 1.45 (0.00) (0.000) CT1 4 Spread (0.017) (0.003) df CT3 Spread df CT3 Spread (0.000) (0.000) CT3 4 Spread (0.001) (0.035) Notes: symbolizes change. Spread represents the sovereign risk premia. CT1 is the carry-trade portfolio with one currency and CT3 is the carry trade portfolio with three lowest- and highest-interest rate currencies. Values presented are the χχ (Wald) statistics for the joint significance of each of the other lagged endogenous variables in that equation. Values in parentheses refer to the p-values. All represents the χχ statistic for joint significance of all other lagged endogenous variables in the equation. 9