BIAS CORRECTION AND STATISTICAL TEST FOR DEVELOPING CREDIT SCORING MODEL THROUGH LOGISTIC REGRESSION APPROACH
A credit scoring model is a statistical model that uses empirical data to predict the creditworthiness of credit applicants. A simple but very powerful approach to developing a credit scoring model is to employ logistic regression. Due to the heterogeneity among the population, segmentation into reasonably homogeneous subpopulations is desirable to enhance model performances. However, one often needs to use unequal sampling ratios across the segments to extract the development sample. Hence, the models developed will be biased unevenly and needed to be adjusted to make score comparisons across different segments meaningful. In this paper, we focused on the topic of detection of uneven bias and its correction for segmented scoring models. A statistical test based on the large-sample theory is proposed for detecting the uneven bias along with its mathematical derivation and the simulation results of the test. When uneven bias over different segments has been detected, a formula to alleviate the effects of the uneven bias is suggested along with its heuristic derivation.