ESIS2: Information Value Estimator for Credit Scoring Models

Information value is widely used to assess discriminatory power of credit scoring models, i.e. models that try to predict a probability of client’s default. Moreover it is very often used to assess the discriminatory power of variables that enter into these models. This means that the Information value is used as a filter for variable selection. However, empirical estimate using deciles of scores, which is the common way how to compute it, may lead to strongly biased results. The main aim of this paper is to give an alternative estimator of the information value, named ESIS2, which leads to lowered bias and mean square error. The implication of this is better credit scoring model. And what is essential, the direct consequence of having better credit scoring model is significantly higher profitability of credit business. Copyright Springer Science+Business Media New York 2015