Thresholds for Ratings' Forecast Default Probabilities: A mean squared error based approach

by Guido Bichisao of the European Investment Bank,
E. Grillo,
Massimo Marchesi of the European Commission, and
Claudio Zucca of the European Investment Bank

December 12, 2005

Abstract: Given the increased importance rating agencies have assumed in the determination of credit institutions' capital requirements according to the Standardised Approach in the Basel II framework, the Basel Committee for Banking Supervision has proposed a procedure to "map" the ratings of different rating agencies into the risk weights that determine credit institutions' capital requirements. This procedure is built upon Cumulative Default Rates (CDRs) and is based for each rating class on reference values and upper admissible thresholds for observed CDRs. In fact, whenever the upper thresholds are exceeded, supervisors are encouraged to enquiry over the rating agencies' rating assignment standards. Thresholds are calculated on the basis of Monte Carlo simulations and assigned using the 99th and the 99.9th percentile of the distribution so obtained.

Some observers notice that this procedure seems to lead to very broad ranges, and worry that this could prevent the prompt detection of on-going weaknesses in the rating assignment standards followed by some rating agencies. In this paper, an approach to determine thresholds for ratings' forecast default probabilities using a test statistic based on the Mean Squared Error (MSE) over realized CDRs is presented. Such approach could represent a potential improvement of the "mapping" procedure suggested by the Basel Committee as the MSE can take directly into account the reference value suggested by the Basel Committee for each rating class, and it can also test the hypothesis of that reference value being correct.

Download paper (147K PDF) 18 pages