Comparacion de dos modelos para predecir la probabilidad deincumplimiento (Modelo Logit-Modelo GAM)

Abstract

One of the most important needs of lending institutions is to have reliable criteria to determine who should grant the credit. In this sense, methodologies such as credit granting models should be implemented as a solution to these needs. In this project, two methodologies that can be used in Credit Risk are purchased, the data provided to this project are from a Colombian financial institution, as part one describes the previous studies that responded to the problem, the methodologies used to develop a model of origin, as a first step for this construction is the grouping of characteristics into categories and its predictive capacity is calculated using prediction measures, such as the weight of the evidence and the information value, the variables that have little predictive capacity are eliminated of the model. Once the models whose variables have a good predictive level are obtained, the model selection stage is followed using techniques such as the AUC criterion and the confusion matrix and finally some recommendations are made in order to implement the best model in this entity.