Gamma–lindley regression cure model for corporate credit default prediction

Abstract

Predicting the firm’s default over time is marked persistently with certain flaws for firms seeking to avoid their exposure to credit risk and losses induced by defaults. From this perspective, a Gamma–lindley cure model, relying on a Gamma–lindley distribution and the survival analysis framework, was introduced to model the default probability and time of corporate credits referring to their inference ability once the firm is likely to default. In addition, it extends the standard mixture cure model allowing more flexibility for modeling lifetime data through relaxing the independence assumption of the probability and the time of default. Furthermore, it handles the missing defaulting labels as a latent variable and identifies three classes of default probability distribution, namely short, medium, and long-term through its potentially accurate classification method. A relevant parameter-learning procedure was described resting on the Maximum Likelihood approach. The results of an empirical analysis on a firms dataset reveal that the advanced Gamma–lindley cure model has a rather effective performance in terms of default time prediction compared to Exponential, Gamma, Weibull, and Cox proportional hazards regression modeling specifications, which yield a meaningful solution for timely credit risk management and cluster firms defaulter in three classes, enabling a deeper analysis of firm’s defaults situation. The numerical experiments were conducted with Python language.