Understanding Default Risk Through Nonparametric Intensity Estimation

This paper investigates instantaneous probabilities of default implied by rating and default events. We propose and apply an alternative measurement approach to standard cohort and homogenous hazard estimators. Our estimator is a smooth nonparametric estimator of intensities, free of bias and unambiguously more accurate. It also avoids the Markovian framework and takes care of censoring. Using Standard & Poor’s ratings database we then show that intensities vary both with respect to calendar time and ageing time. We deeper investigate the behaviour of through-the-cycle default probabilities, update and complement knowledge on documented non Markovian patterns. Results do not support associated timeliness problems but indicate a low reactivity of ratings in terms of magnitude. Because of their target horizon, they indeed integrate the mean reverting feature of default intensities.