Zero-inflated Cure Rate Regression Modelling for Financial Time-to-Default Data

Francisco Louzada (ICMC-USP)

Abstract: We introduce a methodology based on zero-inflated long-term survival data in order to deal with fraud rate estimation in bank loan portfolios. Our approach enables us to accommodate three different types of loan borrowers, i.e., fraudsters, those who are susceptible to default and finally, those who are not susceptible to default. Regarding to the survival analysis framework, an advantage of our approach is to accommodate zero-inflated times, which complement the classical cure rate model. The parameter estimation is reached by maximum likelihood estimation procedure. To illustrate the proposed method, a real dataset of loan survival times is fitted by the zero-inflated Weibull cure rate model. (Joint work with Mauro de Oliveira Jr and Fernando Moreira).