Predicting Credit Rating and Credit Rating Changes: A New Approach

In this paper, we propose a hazard rate model for studying credit rating and credit rating changes. Theoretically the hazard rate model is more appropriate than the previous static models. Yet it is difficult to estimate hazard rate model, especially when the covariates are time-varying. This paper extends the results of Shumway(2001) and shows that a multiple-state hazard rate model can be estimated through the standard logistic methods. This finding greatly simplifies the application of hazard rate model to our credit rating studies, and other potential areas. Applying the hazard rate model, we find that: Different from previous findings, the duration effect on credit rating changes is not monotonic. This solves previous conflicting findings on duration effect; Further, we document that duration effect on downgrade is mainly caused by the downgrade momentum effect; We also provide additional support for the argument that credit rating agencies were adapting more stringent rating policy over time, yet this trend broke down around year 1997; Moreover, our empirical results suggest that our hazard rate model does outperform the traditional models in terms of eight predictability measures used in this paper.