Use of Survival Models in credit scoring Modeling

Abstract

Many statistical models were used in credit scoring problems to obtain improvements in risk reduction and increased profitability. Among them, discriminant analysis, logistic regression, neural networks and survival analysis among other techniques.In this project new models of discrete fragility-induced survival are proposed for the modeling of univariate and bivariate survival data to predict the risk of default by borrowers in a financial system. The univariate survival model results in assuming that the fragility distribution is the Generalized Poisson distribution. Consequently the bivariate model is constructed considering the generalized bivariate Poisson distribution of fragility. For the proposed models we intend to develop inferential procedures from a classical and Bayesian perspective. In the context of classical (or frequentist) inference we intend to use the methodology of maximum likelihood and in the Bayesian approach Monte Carlo methods via Markov Chains (MCMC). The proposed models will be applied to a set of data collected from a portfolio of clients of a Brazilian bank. (AU)