Robust Inference in Rating Models

Various sources of inconsistency are identified in usual statistical rating models. Several semiparametric methods, which are more robust with respect to specification errors, are proposed. In particular, the Pseudo Maximum Likelihood Methods, the Generalized Method of Moments and the Asymtotic Least Squares Methods are used in a new approach of a priori and a posteriori rating. An empirical implementation, based on data from Groupe Monceau, is discussed.

Keywords: Inconsistency, Robustness, Misspecification, Semiparametric Methods, Pseudo Maximum Likelihood Methods, Generalized Methods of Moments, Asymptotic Least Squares Method, A Priori and A Posteriori rating