(Non) consistency of the Beta Kernel Estimator for Recovery Rate Distribution
by Christian Gourieroux of CEPREMAP & the University of Toronto, and
Abstract: In this paper, we explain why a nonparametric approach based on a beta kernel [Renault, Scaillet (2004)] will lead to significant bias when applied to recovery rate distributions. This is due to a specific feature of these distributions, which admit strictly positive weights at 100% corresponding to full recovery (and also at 0% corresponding to total loss). Moreover, for distributions without point mass at 0% and 100%, the beta kernel approach features significant bias in finite sample. In large sample, the method is consistent, but other competing approaches presented in the paper provide more accurate results.
Keywords: Loss-Given-Default, Recovery, Credit Risk, Kernel Estimation.