Uma metodologia para o gerenciamento de modelos de escoragem em operações de crédito de varejo no Brasil

Resumen

The development of scoring models for credit risk management and behaviour analysis of clients is quite important in the routine of commercial banks, financing companies, credit card operations, marketing units and large department stores. Statistical techniques such as logistic regression, discriminant analysis, decision trees etc., have dominated the literature when the subject is related to credit modelling. An important point remains ignored, however: tracking the performance of theses models and validating their predictive ability after their implementation. In this article we present a methodology to track the stability of credit and behaviour scoring models, checking their predictions through time, in order to avoid problems when granting credit for clients. Several examples are presented in order to illustrate the practical use of our proposal. Our proposal seeks to generate actions to avoid possible operational and modelling failures related to these models, suggesting in some cases that these models be replaced by new models.