Estimating Credit Rating Transition Probabilities for Corporate Bonds

Abstract

In this paper we take the task of motivating and exhibiting the potential of conditioning on economywide state variables in improving the forecasting of the Credit Rating Transition Probability (CRTP) Matrix. The improvement in CRTP matrix forecasting accuracy by utilizing state variable information is significant, both statistically and economically, in in-sample and out-of-sample experiments. As a byproduct of examining the one-step ahead forecasting power of state variables in the information set of the researcher we undertake a study of the variation of credit migration (which includes default risk) over the business cycle. We find that an increase in nominal short and long and real rates, a lower equity return and a higher equity return volatility are associated with higher relative downgrade intensities. We compare our results with the predictions of credit risk models. We deal with the implications of possible non-Markovian behavior of the credit rating process. Issuer-specific information is introduced in the form of time dependence and rating momentum. We examine whether the information provided by the state variables is subsumed or is augmented by incorporating various parametric or nonparametric types of time and occurrence dependence in the credit rating transition process. Even though there is significant evidence of time acceleration and downward rating momentum the pattern of state variable sensitivities remains mostly intact. Taking into consideration industry classification reveals the different sensitivity of certain sectors to term structure variables with forecasting, relative value, possibly hedging and correlation implications. The financial sector in particular is more sensitive to interest rates. Finally we examine unobservable to the econometrician heterogeneity which turns out not to alter the conclusions in an appreciable fashion.