Predictive Power of Credit Scoring Questions: Case of SMEs Loan Approvals

Abstract

This research paper aimed to examine predictive power of credit scoring questions being used for screening small and medium enterprises (SMEs) loan applicants. The questions and data of 619 applicants from a non-disclosed leading bank for SMEs in Thailand were explored. Firstly, the exploratory factor analysis was employed. The results showed that from the 12 questions being used in the credit scoring system, there were 5 extracted factors. Secondly, in order to prove predictive power of the questions, the 5 extracted factors were used as predictors or explanatory variables in a Binary Logit model for the non-performing loans (NPLs) classification, i.e. being NPLs or normal debt status. Using step-wise regression technique, 3 factors remained in the model. The 3 factors were "Time and Size", "Debt Records", and "Debt Burden". Consequently, the implication was that for screening SMEs application, credit providers should pay more attention to the questions related to the 3 factors. Meanwhile, as suggested by the model, for the excluded factors, i.e. "Income Reliability" and "Business Prospective", these factors may also still be considered but with comparatively less importance. Although the overall model predictive accuracy was at 69.3%, the percentage was at 31.5% for correctly predicted NPLs cases. Thus, more studies on the credit scoring questions and also more varieties of the models are needed.