Many studies have shown that observable covariates are very useful in predicting default. Shumway (2001) demonstrates that firm specific variables such as the excess stock return, stock return volatility, the ratio of net income to total assets, and the ratio of total liabilities to total assets can explain the probability of default. Duffi e, Saita and Wang (2007) use distance to default, the firms stock return, the three month Treasury bill yield, and the one year trailing S&P 500 index return as explanatory variables to estimate the probability of default. Given these studies on default prediction under the natural probability measure, it is reasonable to ask whether observable covariates are also key determinants of the prices of credit risky securities. The main contribution of this paper is to answer this question using data on corporate credit default swaps (CDS), derivatives contracts contingent on a firmos default.
When pricing corporate bonds and CDSs, it is necessary to estimate the loss distribution under the pricing probability measure. There are several approaches to identifying the observable determinants of credit spreads for corporate bonds and CDSs. One approach uses structural models of default, following Merton (1974). In these models, the observable covariates are determined by the underlying theory. For example, in the simplest structural models suggested by Merton (1974) and Black and Cox (1976), credit spreads are determined by interest rates, firm asset volatility, and firm leverage. However, several authors have come to the conclusion that structural models do a poor job of explaining credit spreads for corporate bonds and CDSs. These findings cast some doubt on the value of observable covariates for explaining credit risk.
A related literature uses linear regressions to investigate credit spreads. This literature is extensive and a wide variety of empirical results for CDSs and corporate bonds are available, using different sample periods, data frequencies, and explanatory variables. Interpreting the results from this literature is not straightforward. Using a linear regression of covariates to explain either the level or change in CDS spreads represents an approximation, given that theory implies a non linear relationship. Results seem to critically depend on the statistical framework. Some studies regress credit spread levels on observable covariates, others use credit spread differences as dependent variables. The latter regressions often generate rather low R'squares, while the R'squares for the levels regressions are much higher. Overall, this strand of research has failed to provide a consensus on the explanatory power of observable covariates for credit spreads.
The evidence from linear regressions, together with that from structural models, suggests a disconnect between the literature on default prediction under the natural measure, where observable covariates are highly successful in explaining default, and the literature under the pricing measure, where observable covariates are much less useful for explaining default and credit spreads.
Download
On Pricing Credit Default Swaps With Observable Covariates
