Data Analytics for Developing and Validating Credit Models

The development of credit risk assessment models in the context of credit scoring and rating, is a data-intensive task that involves a considerable level of sophistication in terms of data preparation, analysis, and modeling. From a data analytics perspective, the construction of credit scoring and rating models can be considered as a classification task, that requires the development of models differentiating the borrowers by their level of credit risk. The model fitting process can be implemented with various methodological approaches, based on different types of models, model fitting criteria, and estimation procedures. This chapter presents an overview of different analytical modeling techniques from various fields, such as statistical models (naive Bayes classifier, discriminant analysis, logistic regression), machine learning (classification trees, neural networks, ensembles), and multicriteria decision aid (value function models and outranking models). Moreover, performance measurement issues are discussed, focusing on the presentation of various popular metrics for evaluating the predictive power and information value of credit scoring and rating models.