The Consumer Loan’s Payment Default Predictive Model: an Application of the Logistic Regression and the Discriminant Analysis in a Tunisian Commercial Bank

Consumer credit scoring is often considered as a classification task where borrowers receive a good or a bad credit status. The current paper attempts to uncover the issue of allocating credits to bad borrowers. In this respect, consumer credit scoring is a measure taken to overcome challenges encountered by Tunisian banks in the process of granting credits. These challenges stand as initiatives to enable banks to predict the future performance of borrowers, to determine the optimal credit limit with regard to the applicants' repayment behavior, and to ensure their efficiency through automating the credit-granting decision process. To reach this end, we use the logistic regression as well as the discriminant analysis in order to develop predictive models that distinguish between "good" and "bad" borrowers. The data have been collected from a commercial Tunisian bank over a 3-year period, from 2010 to 2012. These data consist of four selected and ordered variables. By comparing the respective performances of the logistic regression (LR) and the discriminant analysis (DA), we notice that the LR model yields a 99 % good classification rate in predicting customer types, and then, a significantly low error rate (1 %), as compared with the DA approach (where the good classification rate is only equal to 68.49 %, leading to a significantly high error rate, i.e., 31.51 %). Though the study is limited to a Tunisian commercial bank, it remains an attempt to minimize the rate of nonperforming loans.