Biases in estimating bank loan default probabilities

Thomas Mählmann

Abstract

ABSTRACT

Whether motivated by risk management or regulatory needs, banks are keenly interested in accurately estimating the probabilities of default (PDs) associated with their internal rating grades. The quality of these estimates depends critically on the availability and completeness of time series of default and rating data. This paper studies PD estimation when ratings are missing for some obligors and periods in internal data sets. We argue that a within-bank incentive conflict can result in ratings systematically missing for obligors in financial distress. Using internal rating data from a German bank, it is shown that the widely used “cohort” PD estimator, ie, the long-run average of one-year realized default rates, which simply ignores missing ratings, yields downward-biased PD estimates. Unbiased PD estimators, which account for different missingness mechanisms, are proposed.