Quantifying Credit-scoring Performance

Abstract

Recently, the consumer credit industry has experienced a sizeable growth, where scoring techniques have grown to outperform the traditional, judgmental manner of assessing credit risk. Using the data of a Belgian direct-mail company offering consumer credit, the authors have shown a clear improvement in comparison with its current credit evaluation system, constructed by an international company specialized in consumer credit scoring, and identify the size of this performance increase due to population drift versus model improvement. Considering the crucial impact of the accuracy of the score on the cost side (lowered credit risk) as well as on the revenue side (increased accepted applications), in this study, we highlight the importance of introducing different performance measures to quantify credit risk performance. Hence, instead of reporting predictive performance on the total sample, we have quantified the predictive performance in more detail into a graphical overview that can be used for effective decision making. More specifically, we have augmented the traditional performance measurement by evaluating credit risk as well as credits-scoring profitability in an entangled manner because of the intertwined nature of the outcomes of an adaptation of the current credit-scoring algorithm.