Sovereign credit risk dynamics in the European Monetary Union (EMU)

Abstract

The article provides evidence of the key determinants of the sovereign credit spreads by including some unique variables which proxy credit risk, country-specific risk, and international risk for 10 European Monetary Union (EMU) member countries. The findings suggest that though both country-specific and global risk factors significantly influence the sovereign yield spreads, the primary balance and not the forecast deficit is a key factor of credit spreads. Prior to the sovereign debt crisis, investors appear to focus mainly on the global risk-aversion factors. However, during the sovereign credit crisis only fiscal factors appear to affect the credit spreads while international risk factors have had little or no impact on the sovereign credit spreads.

Keywords:

• sovereign credit risk
• EMU
• sovereign debt crisis
• country risk
• global risk

JEL Classification:

• F34
• G15
• P34

Notes

¹ For a related literature review on the study of corporate credit spreads, see Manzoni (2002), and for emerging markets sovereign credit spreads see Hilscher and Nosbusch (2010).

² Geyer et al. (2004) and Favero et al. (2010) also suggest the use of French OATs as a benchmark for the short maturities in the EMU bond market.

³ For example, Bernoth et al. (2012), using a simple mean-variance model where investors aim to maximize utility, decompose yield spreads between a domestic and a foreign risk-free government bonds into the (a) the default risk premium, (b) the liquidity premium and (c) the country-specific risk premium that depends negatively on the percentage of the gross payment in case of default, positively with the variance of the default probability of the domestic issuer, the gross nominal return and the level of relative risk aversion of domestic and foreign investor.

⁴ A common approach in the empirical research is to include in the explanatory variables lagged values of credit spreads with the scope to capture spread persistence. When the term is included, it is significant and positive and its estimated value is typically in the area 0.95 when the dependent variable is the spread and in the area of 0.25 when the dependent variable is the difference in spreads.

⁵ For the period before EMU, government bond yields were reflecting the national currency interest rates. In order to set sovereign yield spreads before and after EMU on a comparable basis, sovereign spreads versus Germany were calculated on an asset swap basis, that is, sovereign yield differential versus Germany minus the Interest Rate Swap (IRS) differential of the country’s currency versus DEM.

⁶ In many cases, country and bond issue-specific variables are used in relative values versus the benchmark, that is, Germany. In addition, when forecasted values are available, the empirical investigation very often uses the most recent available forecasted values and not the actual values which are published with a relative delay.

⁷ Barrios et al. (2009), Haugh et al. (2009) and Bernoth et al. (2012) estimate the debt service ratio as a percentage of government revenues, while Alexopoulou et al. (2009) do so in terms of GDP.

⁸ Based on their data, Greece’s future pension liabilities were expected to increase by 12% points of GDP over the period 2010 to 2050. The results of Haugh et al. (2009) also suggest that although the inclusion of the country’s relative bank exposure in Eastern and Central Europe improve the fitness of the model for Austria, Greece and Belgium, it is not statistically significant.

⁹ A similar relationship has also been reported for corporate spreads by Campbell and Taksler (2003). Hilscher and Nosbusch (2010), using data for 31 emerging countries during the 1994 to 2007, find evidence that the longer the period since last default, the lower is the sovereign credit spread. Cantor and Packer (1996) suggest the exclusion of credit rating from the model when other credit-related variables are included. They examine the cross-sectional relationship between credit rating and sovereign spreads. When credit rating is the only independent variable in the regression model, it has considerable power in explaining sovereign yield spreads. When additional variables are included in the model, these remain statistically insignificant.

¹⁰ Oliveira et al. (2012) also report a significant positive relationship for the change in the lagged value of the current account ratio and the sovereign spread. A significant negative relationship with sovereign spreads is also reported by Alexopoulou et al. (2009) for the sum of imports and exports as a percentage of GDP (trade openness), and by Hilscher and Nosbusch (2010) for the change in country’s terms of trade.

¹¹ Fleming (2003) evaluated various liquidity measures in the US Treasury market. Their findings suggest that the bid/offer is a simple and effective measure for assessing and tracking the market liquidity. In addition, other related liquidity measures such as the quote and trade size, the off-the-run/on-the-run spread are assessed as modest proxies for the market liquidity.

¹² Exception to these findings is the evidence of Codogno et al. (2003) that reports that bid-ask spreads have no impact on spreads. When the liquidity proxy used is the on-the-run/off-the-run spread, Gómez-Puig (2009a, b) reports a significant positive relationship, while Geyer et al. (2004) find an insignificant relationship.

¹³ Barrios et al. (2009) emphasize the interconnection between liquidity and credit risk. A high level of debt might have a positive impact on liquidity, but negatively affects the creditworthiness of the country.

¹⁴ Similar spreads versus the German government bonds reported by Geyer et al. (2004) confirm the significant positive relationship.

¹⁵ Longstaff et al. (2011) examined the impact of the international risk factor CDS spreads of 26 developed and less developed countries using the spread of BBB US corporate bonds versus the BB high yield bonds. They also report a significant positive relationship for 16 of the 26 countries in the sample.

¹⁶ Cyprus, Estonia, Luxembourg, Malta, Slovakia and Slovenia are not included in the sample since they represent a relative small part of EMU-17 gross domestic product (2%) and almost 1% of the euro area countries’ government debt (based on Eurostat data).

¹⁷ The Dickey Fuller (DF) test is considered superior for time series with autoregressive structure compared with the unit root tests of Phillips and Perron (1988) and Choi and Phillips (1991) which suffer from serious distortions (Dejong et al., 1992).The present study concentrates on the augmented version of the DF test which is more reliable since it ensures that residuals are white noise. The appropriate number of lags is selected using Schwarz (1978) Information Criteria. Constants and trends have also been considered.

¹⁸ Im et al. (2003) tests evaluate the null hypothesis of unit roots while the Hadri (2000) test examines the null hypothesis that all of series in the panel are stationary.

¹⁹ The presence of a lagged dependent variable among the regressors and the potential endogeneity of the explanatory variables render both fixed-effects and random-effects estimations inconsistent. Specifically, Nickell (1981) has shown that fixed-effect estimator is biased (for large N and small T) because within transformation induces a correlation of order 1/T between the lagged dependent variable and the error term.