A new use of stochastic dominance: Measuring default prediction accuracy of bond credit ratings

This paper presents a new application of stochastic dominance. The distributions of initial credit ratings assigned by the credit rating agencies are compared with the distributions of future defaults to measure default prediction accuracy. Rules are derived for first-degree and second-degree dominance of the default predictions of one agency versus another. To illustrate their use, these rules are then applied to the ratings of Moody’s and Standard & Poor’s for corporate bonds and Residential Mortgage-Backed Securities. Stochastic dominance rules provide more discriminatory power compared to the procedures typically used by the credit ratings firms.