CDS 스프레드의 결정요인에 대한 연구

Estimating and explaining the credit risk has long been a central research question in finance. Structural and reduced-form models are the two most popular approaches in credit risk literature. Both approaches have advantages and disadvantages : while explicitly explaining the mechanism of default, structural models do a poor job in explaining the prices of corporate bonds. On the other hand, the reduced-form models lack of economic rationale for determining the default process despite of the empirical success in pricing the corporate bonds. As an alternative, Collin-Dufresne et al.(2001) identify the determinants of the credit spreads from the structural models, and regress the credit spreads on these determinants to examine the explanatory power of these variables. Following the approach used by Collin-Dufresne et al. (2001), and employing credit default swap (CDS) spread as a proxy of credit risk, this paper investigates the determinants of CDS spreads by regressing the CDS spreadsin level (CDS) and spreads differences (ΔCDS) on the variables such as leverage ratio (Lev), equity volatilities (Vol), jump risk (Jumps), past stock returns (Ret), absolute bid-ask spread (Liq), risk-free rate (Macro),and CDS premium on foreign exchange stabilization bonds issued by Korean government (Macro). Specifically, we use the following regressions.Our main findings can be summarized as follows : First, the estimated signs for the explanatory variables are consistent with the theory, which justifies the use of CDS spread as a proxy for the credit risk. For example,while CDS spreads covary positively with equity volatilities, leverage ratios,absolute bid-ask spreads, the estimated signs for the past returns on stocks and risk-free rate are negative. Second, equity volatilities, leverage ratios,absolute bid-ask spreads, and CDS premium on foreign exchange stabilization bonds issued by Korean government are the most powerful variables:they account for approximately 95% of the CDS spreads, and explain 21%of the differences in CDS spread. The reduced explanatory power in difference regressions is consistent with Ericsson et al. (2009), who document that the maximum R2 is less than 100% given the presence of noise in the CDS data. Third, the estimated signs for the sub-sample regressions are also consistent with the theory. Sub-sample regressions are performed since the Chow (1960) test implies that there is a structural break in parameters in Oct 24, 2008 when the KOSPI 200 index declined more than 10%. Finally, the estimated coefficients of the volatility, leverage ratio, absolute spread, and CDS premium on foreign exchange stabilization bonds issued by Korean government remain positive when we investigate the stability of the regression coefficients over time using rolling-window regressions. The key distinction in this paper is that we use the credit default swap (CDS) spread as a proxy of the credit risk, while most researchers have used the corporate bonds in credit risk literature. The advantages of CDS spreads over corporate bond spreads are as follows. First, CDS spreads measure the credit risk better than corporate bond spreads because the CDS spreads are traded on standardized terms while corporate bond spreads varies from the coupon rates, seniorities. Second, CDS spreads are independent of the specification of risk-free rates. Finally, CDS spreads may better reflect the short-term credit conditions since it is known that the CDS spreads lead corporate bond spreads (Zhu, 2006). We perform the regressions using the CDS spread levels and CDS spread differences. While Campbell and Taksler (2003), Cremers et al. (2004) conduct the level regressions, Collin-Dufresne et al. (2001) perform the regressions in spread differences. Pedrosa and Roll (1998) argue that high R2 may be attainable with level regressions since the CDS spreads are highly persistent. However, given that the power of unit root test is quite weak,we conduct both regressions following Ericsson et al. (2009).