Default probability estimation in small samples-with an application to sovereign bonds

In small samples and especially in the case of small true default probabilities, standard approaches to credit default probability estimation have certain drawbacks. Most importantly, standard estimators display high variability and tend to underestimate the true default probability, which are clearly undesirable properties from the perspective of prudent risk management. As an alternative, we present an empirical Bayes approach to default probability estimation and apply the estimator-which is capable of multi-period predictions-to a comprehensive sample of Standard & Poor's rated sovereign bonds. By means of a simulation study, we then show that the empirical Bayes estimator is more conservative and more precise under realistic data-generating processes.