Evaluate the performance of cardholders' repayment behaviors using artificial neural networks and data envelopment analysis

In face of intense competitions and time pressure, card issuers need to not only engage in long-term credit relationship management but also evade potential risks of default in time. Since the databases that banks use for analysis of cardholders' payment behaviors is usually large and complicated, and the extant classification techniques do not offer high classification accuracy, prediction of cardholders' future payment behavior remains a difficult task in the present. Accordingly, most banks do not launch debt collection operations until occurrence of default and thus need to bear an unnecessary waste of costs. This paper attempts to construct a two-stage cardholder behavioral scoring model. Chi-square automatic interaction detector (CHAID) and artificial neural networks (ANN) are first applied to construct the first-stage classification models. By comparing the overall classification accuracy rate the optimal customer classification model is obtained. Later, the important variables selected by the classification model are used as the input and output variables in the second-stage data envelopment analysis (DEA). We then feed back the derived efficiency values to the classification model to verify its previously classified results. Unlike most literatures of DEA which focus on evaluation of overall management performance, we initiatively apply DEA to evaluation of individual behavior scoring to help banks reduce the cost of potential misclassification of customers. For inefficient customers, we can also provide directions on improvement of individual customers to further increase their efficiency and contribution.