A utilização de modelos de duração para gerenciar risco de crédito

Puggina, Wladimir A.
Título da Revista
ISSN da Revista
Título de Volume

Hazard models, also known as time-to-failure or duration models, have been used to examine what independent variables have higher explanatory power in predicting bankruptcy. It is an alternative approach to binary logit, probit and discriminant analysis. Duration models should yield more efficient results than discrete choice models, since it takes into consideration the surviving time in the estimation ofthe instantaneous probability of failure for a given set of observations on the independent variables. Discrete choice models typically ignore information about the timing of failures, and provide an estimate only ofthe probability offailure within a given interval oftime. The question I address in this work is how to use hazard models to project default rates and construct migration matrix conditioned on the state ofthe economy. The concept of the model is pretty closed to the historical default and mortality rates used in the literature. I test the Cox proportional hazard model in Brazilian companies and fmd that the default probability reduces significantly after the third year after issuance date. Default probabilities mean and standard deviation are also affected by economic cycles. I discuss how to incorporate the Cox proportional hazard model to the four most popular credit risk models: CreditRisk +, KMV, CreditPortfolio View and CreditMetrics, and improve existing features.

Área do Conhecimento