Modelos para análise de risco de crédito: aplicação numa instituição financeira

Abstract

Credit risk is one of the main concerns of financial institutions, which in recent years have witnessed an evolution in credit risk assessment and management techniques, with an increasing use of quantitative models, particularly credit scoring models. This work aimed to create customer approval models (application scoring) and a behavioral one (behavioral scoring) by applying logistic regression to assess credit risk in a financial institution. This institution provided demographic and behavioral information on a sample of its customers from January to December 2003. The forward stepwise method was used to choose the variables that compose the model, subsequently generating the score for each client, based on the coefficients of the chosen variables. The models were considered efficient in predicting the situation of customers, as with a correct classification percentage above 80%, they presented an excellent adequacy in relation to the discrimination between good and bad payers. Therefore, credit scoring models provide subsidies to the institution, helping it to prevent and reduce defaults and, consequently, reduce operating costs.