Credit scoring: aplicando técnicas de regresión logística y modelos aditivos generalizados para una cartera de crédito en una entidad financiera.

Abstract

The purpose of this degree project is to study a statistical methodology based on association measures, separation measures and present the Generalized Additive Model as a promising alternative to Logistic Regression, which are applied in the construction of a statistical model, that allows estimating the probability of default (PD) of a portfolio of some type of granted credit.

Logisticregressionisthemostusedstatisticalmodelinthecreditratingindustry.Despi­te its advantages in easy interpretation and low computational cost, logistic regression is under the criticism of the difficulty of modeling the non­linear characteristics of the predictors' effect on the dependent variable and, therefore, could lead to Unsatisfactory results.

The present study proposes the use of a technique known as Generalized Additive Model introduced by Hastie and Tibshirani (1990), which provides the ability to detect the nonlinear and non­monotonic relationship between the dependent variable and the non-sacrificing predictors Interpretability.

The performance of the models are evaluated and compared using the Kolmogórov­ Smirnov statistic value, the area under the curve (ROC) and the GINI test.