Abstract
Recent credit risk literature has proposed (i) sample selection models for dependencies between the one-year Probability of Default (PD) and Loss Given Default (LGD), and (ii) multi-year approaches which are limited to default risk. This paper provides a model for the simultaneous prediction of continuous default times and multi-year LGDs. These measures are paramount to predict term structures ...
Abstract
Recent credit risk literature has proposed (i) sample selection models for dependencies between the one-year Probability of Default (PD) and Loss Given Default (LGD), and (ii) multi-year approaches which are limited to default risk. This paper provides a model for the simultaneous prediction of continuous default times and multi-year LGDs. These measures are paramount to predict term structures of LGDs and Lifetime Expected Losses for the revised loan loss provisioning framework of IFRS 9 and US GAAP (current expected credit loss, CECL). The model includes a variation of copulas and corrects for sample selection bias of LGDs, which are only observed given a default event. We find empirical evidence that bonds which default closer to origination tend to generate higher LGDs. The model enables more precise estimates of Lifetime Expected Losses and prevents a severe underestimation in contrast to more restricted credit risk models. (C) 2018 Elsevier B.V. All rights reserved.
Altmetric