The following article presents an approach to non-parametrical estimation of recovery rate's (RR) probability distribution for credit exposures to defaulted obligors. The described algorithms allow to deal with the obstacle of limited available data about past recovery rates. The analyses based on Markov process assumptions make use of information about amounts recovered from defaulted obligors in relatively short time. The methods presented in the theoretical section are then implemented in order to analyze probability distributions of recovery rates in four different credit portfolios. The obtained probability distributions are bimodal as the highest probabilities have been observed for very low or relatively high recovery rates. The results are in accordance with ones presented by other authors and suggest that the common practice of implementing beta distribution to models of recovery rates should be regarded as not fully justified.

Keywords: recovery rate, loss given default, LGD, credit risk.

  Łukasz Kozłowski, Piotr Osiński, Non-parametrical Estimation of Recovery Rate's Probability Distribution for Credit Exposures to Defaulted Obligors - plik pdf; (317 KB)