Analytical and Empirical Features of Internal Ratings: An Empirical Consistency Test based on Statistical Models

type="main" xml:lang="en"> Internal ratings represent an important input to the new generation of ‘credit portfolio models’ and to the new capital requirements proposed by the Basel Committee on Banking Supervision; however, their reliability must be validated. In this paper, we first recall the main theoretical and practical aspects explaining the relevance gained by internal ratings in the banks’ operating practices and in the current reform proposals on bank capital regulation. We then show how some statistical models can be used to evaluate a rating system when no great amount on past data is available. To this goal, we first set up a binomial model to capture the main drivers of a borrower’s default; then, we build a separate multinomial model to ‘clone’ the judgements issued by human experts. Finally, we compare the two models: the degree of consistency between them looks reasonable, but is far from being perfect. The main causes of these disagreements are analysed, and some improvements of the current rating process are proposed. (J.E.L. G290, C250)