The Financial Review. The Drivers of Sovereign CDS Spread Changes: Local vs. Global Factors

Transcription

1 The Financial Review The Drivers of Sovereign CDS Spread Changes: Local vs. Global Factors Journal: The Financial Review Manuscript ID FIRE--0-0.R Manuscript Type: Paper Submitted for Review Keywords: Sovereign CDS, CDS Spread Determinants, quantile panel regression

2 Page of The Financial Review The Drivers of Sovereign CDS Spread Changes: Local vs. Global Factors Abstract We use daily data for a panel of countries to investigate regional differences in sovereign credit default swaps (CDS) spread determinants and the significance of local vs. global market factors. Similar to prior studies, we find a high level of commonality among CDS spreads, but our results show that this effect is stronger in Latin American CDS. The results of our quantile panel regression model show that while global forces drive spreads across the conditional distribution, changes in credit ratings are significant in explaining CDS spreads only in the upper quantiles. We also confirm the existence of regional differences in spread determinants.

3 The Financial Review Page of The Drivers of Sovereign CDS Spread Changes: Local vs. Global Factors. Introduction The recent debt crisis in the European Union (EU) underscores the importance of using credit instruments to hedge against losses from sovereign default or debt restructuring. Sovereign credit default swaps (CDS) are derivatives that offer protection against losses on sovereign debt from credit events. Sovereign CDS can be used for arbitrage, hedging, or speculation. CDS buyers can either hedge exposure in the underlying sovereign bonds or hedge exposure in other assets, e.g., bank bonds, in the reference country. Speculative positions in sovereign CDS are taken on a naked basis, i.e., without a position in the underlying asset. The sovereign CDS market has grown at a fast pace since 0, with the notional amount exceeding $ trillion in (Figure ), and later declining but remaining close to $ trillion in. The rapid growth in credit derivatives notional value has drawn questions about their possible destabilizing effects as well as what determines the changes in sovereign CDS spreads. The increased volatility in European debt markets after the worsening situation in Greece in 0 and prompted EU governments to express concerns about the role of CDS speculation in intensifying price declines and raising borrowing costs in already struggling EU periphery countries. As a result, naked sovereign CDS trading was prohibited by the EU in November. Recent reports by the International Monetary Fund (IMF), however, do not find the ban justified and warn about decreased market liquidity. The spotlight on sovereign CDS markets as a possible destabilizing factor has also prompted an increase in studies on the determinants of sovereign CDS prices. Longstaff, Pan, IMF Global Financial Stability Report, Old Risks, New Challenges, April IMF Global Financial Stability Report, Old Risks, New Challenges, April

4 Page of The Financial Review Pedersen, and Singleton () is among the few studies that decompose sovereign CDS drivers into country-specific and global macroeconomic factors. They show that credit spreads are more related to U.S. equity market and global financial variables than to country-specific economic measures. Longstaff, Pan, Pedersen, and Singleton find that the first principal component accounts for a large portion of the variability in sovereign spreads, and that in times of market turbulence this factor is highly (inversely) correlated with the Chicago Board Options Exchange (CBOE) volatility index (VIX), having a correlation coefficient of -%. Pan and Singleton (0) also show a strong association of the spreads of Mexico, Turkey, and Korea with the VIX. Other studies in the same strand of the literature discussing the importance of global determinants of sovereign spreads are those by Ang and Longstaff (); Augustin and Tedongap (); Aizenman, Hutchison, and Jinjarak (); Fender, Hayo, and Neuenkirch (); and, others. This study adds to the literature on country-specific vs. global drivers of sovereign CDS by focusing on the short-term determinants of sovereign credit spread changes. In particular, using daily data for a panel of countries, we examine whether there are regional differences in sovereign spread determinants (Europe vs. Latin America) as well as global and country-specific determinants of spreads. We contribute to the sovereign CDS literature by exploring the heterogeneity of the response of sovereign spreads to explanatory factors by using a quantile panel regression with fixed effects approach, which describes the conditional distribution of the dependent variable, as opposed to focusing just on the mean. Our approach is motivated in part by Pires, Pereira, and Martins (), who show that ordinary least squares (OLS) mean regression of the determinants of spreads are not precise in describing the center of the CDS spreads distribution and represent to a higher extent the results for the upper quantiles. The

5 The Financial Review Page of quantile panel regression methodology we use follows Koenker (0) and provides a more flexible and robust approach by consolidating quantile regression with panel data modeling. Our findings confirm a high level of commonality among CDS spreads. However, we show that this effect is stronger in Latin American CDS. While changes in credit ratings are significant in explaining CDS spreads in the upper quantile of the conditional distribution of spreads, global stock and bond market factors influence spreads across the entire conditional distribution. For example, whereas a change in the credit rating or outlook has no effect on spreads in the lower quantiles of the conditional distribution of Latin American sovereign spreads, a rating upgrade (or improvement in outlook) causes close to a two basis point drop in spreads for this region. Our results also show that Latin American spreads are mostly driven by global factors, while changes in sovereign credit ratings play an important role in determining European spreads in the upper quantile of the conditional distribution. Overnight rates have limited explanatory power compared to changes in long-term Treasury yields. The rest of the paper is organized as follows: section presents a brief review of the literature and outlines our research questions, and section describes the data and the explanatory variables used in the study. In section we describe the empirical methods used and the results. In section we offer concluding remarks.. Previous literature and research questions A number of articles investigate the determinants of sovereign bond spreads. While some studies look into the effect of country-specific factors on credit spreads (Edwards, ; Min, Lee, Nam, Park, and Nam, 0; Rowland and Torres, 0; Baldacci, Gupta, and Mati, ),

6 Page of The Financial Review others investigate global factors and spreads in emerging markets (Dailami, Masson, and Padou 0; Kamin and von Kleinst, ). Some of the country-specific factors included in these studies are the levels of political risk, fiscal policy, current account balance, international reserves, terms of trade index, and the real exchange rate. Georgoutsos and Migiakis () examine the Euro area sovereign bond yield spreads and show that spreads are related to market and economic sentiment indicators, whereas fiscal variables have less effect. A different strand of the literature takes a variance decomposition approach. For example, Dungey, Martin, and Pagan (00) decompose bond yield spreads into global and national factors, where the factors follow a GARCH-type process. Better related to our paper are those studies that examine the dynamics of sovereign debt instruments using CDS data. For instance, Longstaff, Pan, Pedersen, and Singleton () use a panel of sovereign CDS from countries before and during the 0 global financial crisis to show that a bigger portion of the variability in CDS spreads can be explained by global economic factors as opposed to local economic indicators. In the period before the crisis, the first principal component explains % of the variability in spreads, while in the 0-0 period, the first principal component explains up to % t. In a similar vein, Ang and Longstaff () use weekly sovereign CDS data from the United States and European countries from May 0 to January. They decompose sovereign spreads into systemic and sovereign-specific parts, Kose, Otrok, and Whiteman (0) also use a decomposition approach to find the common components of output, consumption, and investment across 0 countries and document world, regional, and country-specific factors.

7 The Financial Review Page of and show that global financial market variables play an important role. Ang and Longstaff also stress the significant negative relation of U.S. systemic risk with the VIX. In a broader study, Augustin and Tedongap () use data for countries to relate economic shocks to CDS spreads co-movements. They show that U.S. growth and consumption volatility are strongly associated with the variation in the first two principal components of the term structure of CDS spreads. Some recent studies focus on the determinants of CDS spreads in specific groups of countries or regions. For example, in an article on emerging market CDS spreads, Fender, Hayo, and Neuenkirch () use a GARCH model and show the importance of global factors as opposed to country-specific ones. Another paper involving emerging markets CDS is by Ismaliescu and Kazemi (0), who investigate the effect of credit rating change announcements on CDS premiums. They report stronger reaction to positive rating changes and weaker reaction to negative changes, speculating that most of the negative downgrade information is already incorporated in the CDS premiums. Aizenman, Hutchison, and Jinjarak () study Greece, Ireland, Italy, Portugal, and Spain and show that fiscal and selected macroeconomic factors are significant in explaining sovereign risk. Along this line of research, Afonso, Furceri, and Gomes () investigate the reaction of EU sovereign bond and CDS spreads to rating changes and show significant response to negative announcements from the rating agencies.

8 Page of The Financial Review Some of the research papers surveyed in this section discuss the impact of the 0 financial crisis on sovereign CDS. Fewer studies investigate the effects of the Greek debt crisis. For instance, Andenmatten and Brill () use data from October 0 to July 0 for countries to test for the contagion and interdependence of sovereign CDS spread premia and report that there were periods of contagion during the Greek debt crisis. Arghyrou and Kontonikas () also confirm contagion from Greece to other European countries. While Andenmatten and Brill () and Arghyrou and Kontonikas () use the Forbes and Rigobon (0) test and principal components analysis, respectively, information transmission in CDS markets is also modeled within a GARCH framework. For example, Wang and Moore () show increased integration in CDS markets after the Lehman collapse using dynamic conditional correlation. Tamakoshi and Hamori () apply a similar methodology to bank CDS. Building on the existing research in this area, our three main research questions are:. What is the relative importance of Global vs. Local Factors in explaining daily changes in sovereign CDS spreads?. Do the factors that affect sovereign CDS spread changes vary across the conditional distribution of CDS spreads?. Do the sovereign spread determinants have the same impact on spreads across different regions?

9 The Financial Review Page of Data and variable description. Sovereign CDS The daily CDS data in this paper are obtained from Thomson Reuters through Datastream. Our sample spans June 0 to July. We use the most actively traded contracts with five years to maturity. Datastream provides sovereign CDS data for reference countries for at least five years. After filtering the available CDS data for liquidity and imposing a filter of at least 0% nonzero spread changes over the sample period, our final sample consists of CDS premia for reference countries. Within our sample we have nine countries from Latin America (Brazil, Chile, Colombia, Costa Rica, Mexico, Panama, Peru, Uruguay, and Venezuela); countries from Europe (Austria, Belgium, Croatia, Cyprus, Hungary, Iceland, Ireland, Italy, Portugal, Romania, Russia, Spain, and the United Kingdom); and, others (China, Indonesia, Jamaica, Japan, Kazakhstan, Malaysia, Philippines, Republic of Korea, South Africa, Thailand, Turkey, and Vietnam). In Appendix A we provide current account balance, total government debt as a percent of GDP, currency reserves in USD, and GDP growth for each of the countries in our sample. These data are obtained from the World Bank s World Development Indicators. The countries with the highest levels of debt as a percent of GDP are Italy, Portugal, and Japan. The highest currency reserves are reported for China, Japan, and Russia. While the current account balance is negative for all countries in the Latin American group, most European and Asian countries in our sample have a positive balance. Appendix A also lists the Standard & Poor s (S&P) foreign currency rating for each country as of October. The highest-rated countries are Belgium, the UK and the Republic of Korea, while the lowest-rated is Venezuela with a CCC rating. Overall, there are significant differences across each indicator, over both the entire sample and across the regions.

10 Page of The Financial Review The dynamics of the average CDS spreads for all of the countries and separately for the Latin American and European countries over the sample period are plotted in Figure, while Figure presents the CDS spread levels for each country. Figure shows that on average, sovereign CDS spreads remained relatively flat from June 0 until about September. In late there is a notable spike in the average spreads across the European countries, before falling in late to pre-0 levels. The Latin American countries exhibit two other, albeit smaller, increases in spreads in and, but significant jumps after the second half of. The large jumps after mid- could be due to Venezuela s economic struggles as a result of lower crude oil prices, in addition to the impact of lower commodity prices in general on the economies of Latin American countries. The individual country plots in Figure confirm that the large spike in spreads observed between and is more notable across the European countries, which may be attributed to the European debt crisis. We report descriptive statistics for daily CDS spread changes over the sample period for each of the respective countries in Table. The countries are grouped by region. The mean CDS spread change is negative for out of the countries. The largest positive mean spread is observed for Venezuela, which also has the highest standard deviation. The CDS spreads of some of the European countries exhibit higher volatility too, with Cyprus and Portugal having very high standard deviations of.0% and.%, respectively. A large variation between the countries is noted in terms of skewness and kurtosis. The largest daily swings in spreads are noted in the case of Venezuela, followed by Cyprus. In Table, we report the correlations between the daily changes in spreads separately for the Latin American and European groups. High correlations are generally observed between some of the CDS spreads of countries within the same region. For instance, the correlation between changes in CDS spreads of Brazil and

11 The Financial Review Page 0 of Colombia is % and that between Belgium and Italy is %; both are significant at the % level or higher. In the European group, Cyprus appears to have the lowest correlations with the rest of the countries. In the Latin American group, Costa Rica and Venezuela each exhibit weaker linkages with the other countries.. Explanatory variables Motivated by prior studies in the literature, we employ a range of explanatory variables in our subsequent analysis. Our first group consists of two country-specific variables: () the daily return on the major stock index for each country (StReturn), and () a dummy variable Ratingd, which is negative one on days when there is a negative change in either the rating or outlook of the country s debt, one on days when there is a positive change, and zero otherwise. Our second group includes four variables that are meant to capture U.S. market conditions: () the daily change in the Chicago Board Options Exchange (CBOE) volatility index ( VIX); () daily changes in the 0-year U.S. Treasury yield ( Treasury); () the daily change in the TED spread ( TEDspread); and () FedFundsd, which is the change in the federal funds effective rate. Our third group consists of four variables that are meant to capture global market conditions. These are: () daily changes in the Emerging Markets Bond Index ( EMBIplus); () daily changes in the Euro Stoxx 0 Volatility Index, i.e., the European VIX ( VSTOXX); () daily changes in the German 0-Year Bond yield ( GermBund); and () ECBRepod, which captures changes in the European Central Bank overnight rate. For brevity we do not report the results between all countries, but the tables are available upon request. However, correlations between countries from different regions are much lower. For example, the correlation between Cyprus and Brazil is % (and statistically insignificant). 0

12 Page of The Financial Review Empirical methods & results.. Principal components analysis To start our analysis, we perform principal component analysis on the daily spread changes over the entire sample period. In Table, Panel A we report the results for all countries, while in Panels B and C we report results separately for the Latin American and European countries. The estimates in Panel A show that the first four components explain % of the variability in spread changes for our full sample. The results in Panels B and C show that this number is even higher at the regional level, with the first four components explaining over % of the changes in spreads of Latin American countries and.% of spread changes of European countries. These results are consistent with the finding of earlier studies that there are common factors that drive changes in the spreads of sovereign CDS at the daily level... Regression models Prior studies that investigate the determinants of sovereign spread changes largely employ either pooled OLS, panel models with fixed effects, or panel models with random effects. However, since we observe significant fluctuations in spreads over the sample period, we employ the quantile regression model of Koenker (0) to investigate any differential impact of the dependent variables across the conditional distribution of spread changes. Koenker starts with the standard random effects model generally given by, = + + () For our study, y is the daily change in the CDS spread, and t represents the daily observations of each country i. The α s are meant to capture country-specific sources of variability that are not An exception is Pires, Pereira, and Martins ().

13 The Financial Review Page of controlled for by the other covariates in the model. Koenker (0) proposes the estimation of a distributional shift for each country. The conditional quantile function of the response on day t for the i th country can then be written as: = + () where the α s have a location shift on the only conditional quantiles of the dependent variable. Effects of the independent variables, the x it are allowed to depend on the quantile q of interest. The model in is estimated for several quantiles simultaneously by solving the following expression:, () where ρ q (u) =u(k-i(u<0)) is the piecewise linear quantile loss function of Koenker and Bassett () and the weights w k control the relative influence of the p quantiles. Koenker (0) solves equation iteratively using a sequence of diagonally weighted least squares steps that employ a Cholesky factorization. The fixed-effects coefficients are penalized by a lasso penalty term with associated penalty parameter lambda, which shrink these coefficients toward zero. The penalized version of equation is given by:, + () as λ an estimate of the model purged of the fixed effects is obtained. Motivated by the results of our principal component analysis, we estimate three multivariate regression models. Each model contains the two country-specific variables, StReturn and Ratingd, which are in turn supplemented by the three groups of global variables meant to

14 Page of The Financial Review capture, respectively, U.S. Market Conditions only (Model ), U.S. and Ex-U.S. Global conditions (Model ), and Ex-U.S. Global Conditions only (Model ). We also include the oneday lag of the independent variable in each of our models to capture any persistence in CDS spread changes. : = : = : = where q is the quantile or linear model and it is the panel of countries denoted by i over days denoted by t, is the one-day lagged change in spreads in first difference. The other variables are described in Section. We first compare the results of a quantile panel regression of Model for all countries to the estimates of a panel model with fixed effects, a panel model with random effects, and pooled OLS estimates. Comparison of the results in the first six columns of Table (quantile FE panel regression) to the other six columns reveals that the pooled OLS, the panel model with fixed effects, and the random effects model present estimates that represent the lower quantiles of the conditional distribution, while the upper quantile of the distribution is not adequately described when these models are used. Another benefit of the quantile regression approach is that estimates are robust to outliers and allows for a quantitative comparison of the effect of a particular We thank an anonymous referee for this suggestion.

15 The Financial Review Page of variable across the quantiles. Based on these findings, we report only results of our quantile panel regression model for the remainder of our investigation. Table reports results of the three models for all countries in our sample over the entire period under study. The models control for heterogeneity across countries and show how the distribution of the dependent variable changes in response to the predictor variables. For all three models, the one-day lagged change in the independent variable is highly significant, confirming the persistence in spread changes. The results for the first model reported in Panel A show that the changes in credit ratings are significant only in the upper quantiles of the conditional distribution of spreads. A rating upgrade or an improvement in the credit outlook leads to a highly significant. basis point decrease in spreads in the upper quantile, while having no significant impact in the lower quantiles of the spread distribution. This impact is economically significant since all of the countries in our sample, except Venezuela, exhibit a mean daily change in spread of less than one basis point. Changes in U.S. stock and bond market conditions are highly significant across all quantiles. Higher yields in the U.S. Treasury markets lead to lower spreads, while increases in U.S. stock market volatility are associated with higher sovereign spreads. Unsurprisingly, spread changes are more sensitive to U.S. bond market changes than they are to U.S. stock market conditions. For example, within the lowest quantile of the spread distribution, a % increase in the U.S. stock market is associated with less than a tenth of a basis point drop in spreads, while a similar increase in Treasury yields leads to a seven basis point drop in spreads on the same day. Parente and Silva () also show that the quantile regression estimator of Koenker and Bassett (), which we employ in our empirical tests, is consistent and asymptotically normal when there is within-cluster correlation of the error terms.

16 Page of The Financial Review In Panel B of Table we report results of Model where Model is expanded to include changes in the TED spread and changes in the EMBI plus index as a proxy for perceived credit risk and liquidity. Positive changes in emerging bond markets tend to decrease sovereign spreads across the conditional distribution of spreads, while an increase in the gap between the Treasury and Eurodollar rates (i.e., lower liquidity in the interbank market) leads to an increase in sovereign spreads across all quantiles. Lastly, the third model reported in Table includes indicators for European stock and bond market conditions. Since our sample covers the European sovereign debt crisis, we examine how changes in the VSTOXX and German government bond rates affect CDS premia, in particular. The results in Panel C reveal a strong impact of European stock and bond market conditions on spreads, with each coefficient having the expected sign. That is, equity volatility is associated with higher sovereign spreads, while increase in the German bond yields decreases spreads in the lower quantiles. Changes in the European Central Bank (ECB) repo rate do not have explanatory power over the entire period when all countries are considered. Overall, based on the results for all countries in Table, changes in spreads appear more sensitive to global stock and bond market conditions than to country-specific variables. Our findings are consistent with Remolona, Scatigna, and Wu (0), who show that liquidity variables and the VIX, as well as average institution s ratings, are significant in explaining spreads... Regional differences To explore whether the determinants of spreads differ by geographical region, we separately estimate our three models for the Latin American and European groups. Table reports the results for the drivers of Latin American spreads, and Table reports similar results for the European countries. The results in Table indicate that spread changes in the Latin American

17 The Financial Review Page of group appear to be influenced by both local and global factors. The credit ratings dummy is significant only in the upper quantile of Latin American CDS spreads, while both U.S. and European stock and bond market conditions affect spreads across the entire conditional distribution. Changes in the overnight rate in the U.S. or Europe do not seem to have any effect and neither do changes in the TED spread, whereas emerging markets credit risk appears highly significant in determining spreads in the Latin American region. Our results extend the findings of Wang, Yang, and Yang (), who discuss the significance of the market volatility as the most important determinant of Latin American credit spreads. The results in Table show that domestic stock market conditions for European countries have significant influence on spreads in the lower and upper quantiles of the conditional distribution. In addition, rating changes appear important for the middle and upper quantiles of the distribution of spreads. In Panel B of Table the role of the VIX is insignificant in the lower quantiles of spread changes for this group, while the TED spread is significant throughout. Furthermore, increases in the Treasury and German bond yields lead to lower spreads, and this effect is more prominent in the lowest quantile. Again, overnight rates have limited explanatory power. Together, our findings can be interpreted from a macroeconomic conditions perspective. If the VIX is taken as a proxy for stock market conditions and Treasury rates indicate the strength of the economy, then the positive relation between VIX changes and credit spreads is expected since higher volatility reflects worsening equity market conditions. Conversely, rising

18 Page of The Financial Review long-term Treasury yields may be a sign of an improving economy and thus lead to lower sovereign spreads.. Conclusions We use quantile panel regressions to capture the drivers of sovereign spread changes across the conditional distribution of spreads. In contrast to the OLS regression, which examines the factors affecting spread changes on average, this approach enables us to uncover the differences of CDS spread drivers across quantiles. Focusing on the results for all countries in our sample, changes in spreads appear more sensitive to global stock and bond market conditions than to countryspecific variables. We also find regional differences in spread determinants. Latin American CDS spreads are influenced more by volatility and long-term bond rates, while European spreads are mainly affected by tightening credit (changes in the TED spread) and credit rating changes, in addition to volatility and long-term yields. Our empirical results may have implications for financial market participants and regulators. Global financial conditions have different influence on sovereign credit risk depending on the region; for example, in our case Latin American sovereign CDS spreads exhibited higher sensitivity. Financial markets may require an increased premium for the higher global risk exposure and as a result affect borrowing costs in these markets. Countries experiencing higher influence of the local factors can focus on improving fiscal health and political stability, among others. While we empirically investigate the drivers of daily sovereign CDS spread changes across the conditional distribution, further research can look into how the dependence of spreads We thank an anonymous referee for this suggestion.

19 The Financial Review Page of on other financial variables can be used for forecasting. Studying the distribution is critical for estimating Value-at-Risk (VaR), and further research could continue in the area of forecasting VaR and co-var along the lines of studies by Pires, Pereira, and Martins () and Adrian and Brunnermeier ().

20 Page of The Financial Review References Adrian, Tobias and M. K. Brunnermeier,. CoVaR, The American Economic Review 0, 0-. Afonso, António, D. Furceri and P. Gomes,. Sovereign credit ratings and financial markets linkages: application to European data, Journal of International Money and Finance, 0-. Aizenman, Joshua, M. Hutchison and Y. Jinjarak,. What is the risk of European sovereign debt defaults? Fiscal space, CDS spreads and market pricing of risk, Journal of International Money and Finance, -. Andenmatten, Sergio and F. Brill,. Measuring co-movements of CDS premia during the Greek debt crisis. Discussion paper, Department of Economics, Universität Bern. Ang, Andrew and F. A. Longstaff,. Systemic sovereign credit risk: Lessons from the US and Europe, Journal of Monetary Economics 0, -0. Arghyrou, Michael G. and A. Kontonikas,. The EMU sovereign-debt crisis: Fundamentals, expectations and contagion, Journal of International Financial Markets, Institutions and Money, -. Augustin, Patrick and R. Tédongap,. Real economic shocks and sovereign credit risk, Journal of Financial and Quantitative Analysis, -. Baldacci, Emanuele, S. Gupta and A. Mati,. Political and fiscal risk determinants of sovereign spreads in emerging markets, Review of Development Economics, -. Dailami, Mansoor, P. R. Masson and J. J. Padou, 0. Global monetary conditions versus country-specific factors in the determination of emerging market debt spreads, Journal of

21 The Financial Review Page of International Money and Finance, -. Dungey, Mardi, V. L Martin and Ad. R Pagan, 00. A multivariate latent factor decomposition of international bond yield spreads, Journal of Applied Econometrics, -. Edwards, Sebastian,. LDC's foreign borrowing and default risk: An empirical investigation. The American Economic Review, -. Fender, Ingo, B. Hayo and M. Neuenkirch,. Daily pricing of emerging market sovereign CDS before and during the global financial crisis, Journal of Banking & Finance, -. Forbes, Kristin J. and R. Rigobon, 0. No contagion, only interdependence: measuring stock market comovements, The Journal of Finance, -. Georgoutsos, Dimitris and P. M. Migiakis,. Heterogeneity of the determinants of euro-area sovereign bond spreads; what does it tell us about financial stability?, Journal of Banking & Finance, 0-. Ismailescu, Iuliana and H. Kazemi, 0. The reaction of emerging market credit default swap spreads to sovereign credit rating changes, Journal of Banking & Finance, -. Kamin, Steven B. and K. Von Kleist,. The Evolution and Determinants of Emerging Markets Credit Spreads in the 0s, Working paper, Bank for International Settlements. Koenker, Roger, 0. Quantile regression for longitudinal data, Journal of Multivariate Analysis, -. Koenker, Roger and G. Bassett Jr.,. Regression quantiles, Econometrica: Journal of the Econometric Society, -0.

22 Page of The Financial Review Kose, M. Ahan, C. Otrok and C. H. Whiteman, 0. International business cycles: World, region, and country-specific factors, The American Economic Review, -. Longstaff, Francis A., J. Pan, L. H. Pedersen and K. J. Singleton,. How Sovereign Is Sovereign Credit Risk?, American Economic Journal: Macroeconomics, -0. Min, Hong-Ghi, D.-H. Lee, C. Nam, M.-C. Park and S.-H. Nam, 0. Determinants of emerging-market bond spreads: Cross-country evidence, Global Finance Journal, -. Pan, June and K. J. Singleton, 0. Default and recovery implicit in the term structure of sovereign CDS spreads, The Journal of Finance,, -. Parente, Paulo M.D.C. and J. M.C. Santos Silva,. Quantile regression with clustered data, Journal of Econometric Methods, -. Pires, Pedro, J. P. Pereira and L. F. Martins,. The Empirical Determinants of Credit Default Swap Spreads - A Quantile Regression Approach, European Financial Management, -. Remolona, Eli M., M. Scatigna and E. Wu, 0. Interpreting sovereign spreads. BIS Quarterly Review March 0, -. Rowland, P. and J. L. Torres, 0. Determinants of Spread and Creditworthiness for Emerging Market Sovereign Debt: A Panel Data Study. Working Paper, Banco de la República. Tamakoshi, Go, and S. Hamori,. An asymmetric DCC analysis of correlations among bank CDS indices, Applied Financial Economics, -. Wang, Ping, and T. Moore,. The integration of the credit default swap markets during the US subprime crisis: Dynamic correlation analysis, Journal of International Financial

23 The Financial Review Page of Markets, Institutions and Money, -. Wang, Alan T., S.-Y. Yang, and N.-T. Yang,. Information transmission between sovereign debt CDS and other financial factors - The case of Latin America, The North American Journal of Economics and Finance, -0.

24 Page of The Financial Review Appendix A Country S&P rating as of Oct. Current Account Balance (% of GDP) / Total Government debt (% of GDP) Total Reserves (billion US$) GDP growth (annual %) Latin America Brazil BB Chile AA Colombia BBB Costa Rica BB Mexico BBB Panama BBB Peru BBB Uruguay BBB Venezuela CCC.. -. Europe Austria AA Belgium AA Croatia BB...0. Cyprus BB Hungary BBB-.... Iceland BBB Ireland A+.... Italy BBB Portugal BB Romania BBB Russian Federation BB Spain BBB United Kingdom AA Other Countries Turkey BB Kazakhstan BBB-... Jamaica B South Africa BBB China AA-.0.,.. Indonesia BB Japan A+..,. 0. Republic of Korea AA.... Malaysia A Philippines BBB Thailand BBB+.... Vietnam BB

25 The Financial Review Page of Table Descriptive statistics of daily changes in sovereign CDS spreads The sample period is June, 0 to July,. All values are in basis points. Mean Std Dev Minimum Maximum Skewness Kurtosis Latin America Brazil Chile Colombia Costa Rica Mexico Panama Peru Uruguay Venezuela Europe Austria Belgium Croatia Cyprus Hungary Iceland Ireland Italy Portugal Romania Russia Spain United Kingdom Other Countries Turkey Kazakhstan Jamaica South Africa China Indonesia Japan Republic of Korea Malaysia Philippines Thailand Vietnam

26 Page of The Financial Review 0 Table Correlations between daily changes in sovereign CDS spreads The sample period is June, 0 to July,. All estimates are significant at the % level or higher, except for that between Cypress and the following countries: Iceland, Portugal, Russia and the UK, which is insignificant. Panel A. Latin America Brazil Chile Colombia Costa Rica Mexico Panama Peru Uruguay Chile 0. Colombia Costa Rica Mexico Panama Peru Uruguay Venezuela

27 The Financial Review Page of 0 Table contd. Panel B. Europe Austria Belgium Cyprus Croatia Hungary Iceland Ireland Italy Portugal Romania Russia Spain Belgium 0. Cyprus Croatia Hungary Iceland Ireland Italy Portugal Romania Russia Spain United Kingdom

28 Page of The Financial Review Table Factor analysis: changes in sovereign CDS spreads The sample period is June, 0 to July,. Eigenvalue Difference Proportion Cumulative Panel A: All Countries PC PC PC PC PC Panel B: Latin America PC PC PC PC PC Panel C: Europe PC PC PC PC PC

29 The Financial Review Page of 0 Table Model comparison for all countries This table reports quantile panel regression, panel model with fixed effects, panel model with random effects, and pooled OLS estimates of the following model: : = where is the quantile or linear panel of daily changes in spreads of countries denoted by i over days denoted by t; is the one-day lagged change in spreads in first difference; StReturn is the daily return on the major stock index for each country; Ratingd is a dummy variable that is -/+ on days when there is a negative/positive change in either the rating or outlook of the countries debt and zero otherwise; VIX is the daily change in the CBOE volatility index; Treasury is the daily change in the 0-year U.S. Treasury yield; TEDspread is the daily change in the TED Spread; and FedFundsd is the change in the federal funds effective rate, and 0 otherwise. The sample period is June, 0 to July,. Parameter estimates in bold are significant at the % level or higher. Quantile Quantile FE Panel Regression FE Panel RE Panel Pooled OLS Parameter Parameter Parameter t -statistic t-statistic Estimate Estimate Estimate t -statistic Parameter Estimate t -statistic Parameter Estimate t -statistic Parameter Estimate Intercept CDSt StReturnt Ratingdt VIXt Treasuryt FedFundsdt t -statistic

30 Page of The Financial Review 0

31 The Financial Review Page of Table Determinants of sovereign CDS spreads This table reports estimates of the following three models: : = : = : = Where is the quantile of daily changes in spreads of countries denoted by i over days denoted by t; is the one-day lagged change in spreads in first difference; StReturn is the daily return on the major stock index for each country; Ratingd is a dummy variable that is -/+ on days when there is a negative/positive change in either the rating or outlook of the countries debt and zero otherwise; VIX is the daily change in the CBOE volatility index; Treasury is the daily change in the - year U.S. Treasury yield; TEDspread is the daily change in the TED Spread; FedFundsd is the change in the federal funds effective rate; EMBIplus is the daily change in the Emerging Markets Bond Index; VSTOXX is the daily change in the Euro Stoxx 0 Volatility Index, i.e. the European VIX ; GermBund is the daily change in the German 0-Year Bond; and ECBRepod captures changes in the European Central Bank overnight rate. The sample period is June, 0 to July,. Parameter estimates in bold are significant at the % level or higher. Estimate t-statistic Estimate t-statistic Estimate t-statistic Quantile Panel A. Model Intercept CDS t StReturn t Ratingd t VIX t Treasury t FedFundsd t Panel B. Model Intercept CDS t StReturn t Ratingd t VIX t Treasury t FedFundsdt

32 Page of The Financial Review TEDspread t EMBIplus t Table contd. Estimate t-statistic Estimate t-statistic Estimate t-statistic Quantile Panel C. Model Intercept CDS t StReturn t Ratingd t VSTOXX t GermBund t ECBRepod t

33 The Financial Review Page of Table Determinants of Latin American sovereign CDS spreads This table reports estimates of the quantile regression models described in the text and in Table for the Latin American countries only. The sample period is June, 0 to July,. Parameter estimates in bold are significant at the % level or higher. Estimate t-statistic Estimate t-statistic Estimate t-statistic Quantile Panel A. Model Intercept CDS t StReturn t Ratingd t VIX t Treasury t FedFundsd t Panel B. Model Intercept CDS t StReturn t Ratingd t VIX t Treasury t FedFundsdt TEDspread t EMBIplus t Panel C. Model Intercept CDS t StReturn t Ratingd t VSTOXX t GermBund t ECBRepod t

34 Page of The Financial Review Table Determinants of European sovereign CDS spreads This table reports estimates of the quantile regression models described in the text and in Table for the European countries only. The sample period is June, 0 to July,. Parameter estimates in bold are significant at the % level or higher. Estimate t-statistic Estimate t-statistic Estimate t-statistic Quantile Panel A. Model Intercept CDS t StReturn t Ratingd t VIX t Treasury t FedFundsd t Panel B. Model Intercept CDS t StReturn t Ratingd t VIX t Treasury t FedFundsdt TEDspread t EMBIplus t Panel C. Model Intercept CDS t StReturn t Ratingd t VSTOXX t GermBund t ECBRepod t

35 The Financial Review Page of 0 0 0,00,000,000,000,00,000,000,000,00,000,000,000 00, Source: BIS Statistics Figure Sovereign CDS notional amount outstanding (in millions USD)

36 Page of The Financial Review Europe and All CDS levels Figure 00,00 00,0 00,0 00, Europe Average CDS Latin America Average CDS All Average CDS Mean sovereign CDS levels Latin America CDS levels

37 The Financial Review Page of 0 Panel A: Latin America BRAZIL CHILE COLOMBIA COSTA RICA PERU MEXICO URUGUAY PANAMA VENEZUELA,000 0,000,000,000,000,

38 Page of The Financial Review 0 Figure contd. Panel B: Europe AUSTRIA HUNGARY PORTUGAL,00, UNITED KINGDOM BELGIUM CROATIA ICELAND ROMANIA ,00, IRELAND RUSSIA CYPRUS,000,00, ITALY SPAIN

39 The Financial Review Page of 0 Figure contd. Panel C: Others TURKEY CHINA MALAYSIA Figure CDS levels by country KAZAKHSTAN JAMAICA,0, INDONESIA PHILIPPINES JAPAN THAILAND SOUTH AFRICA KOREA VIETNAM

40 Page of The Financial Review Figure

41 The Financial Review Page of Figure

42 Page of The Financial Review 0 Figure Panel A

43 The Financial Review Page of 0 Panel B

44 Page of The Financial Review 0 Panel C