Semiparametric Estimation of a Credit Rating Model

This paper develops a semiparametric, ordered-response model of credit rating in which ratings are equilibrium outcomes of a stylized cheap-talk game. The proposed model allows the choice probability to be an unknown function of multiple indices permitting flexible interaction, non-monotonicity, and non-linearity in marginal effects. Based on Moody's rating data, I use the estimated model to examine credit rating agencies' (CRAs) incentive to bias ratings when the CRA's shareholders invest in bond issuers. I find the degree of Moody's rating bias varies significantly for both rating categories as well as the institutional cross-ownership between Moody's and the bond issuer. To obtain the statistical significance of these results, I prove a U-statistics equivalence result that implies asymptotic inference for a large class of semiparametric models