An Investigation to Assess the use of Multiple-outcome Logistic Regression Techniques in Credit Scoring

Abstract

Traditionally credit scoring systems have aimed to reduce the risk in a lender’s portfolio by predicting whether a customer would default on their account. However, the current models offer limited depth, as they only use binary outcomes. It has been suggested that by utilising multiple outcomes more predictive models could be produced. This paper investigates the use of multiple-outcome logistic regression models in the context of credit scoring. These multiple-outcome models are then compared to a binary logistic model; a widely used model in the credit industry, to offer a comparison in discriminatory power. Finally the paper discusses the impact that multiple outcome models would have if implemented into a credit scoring solution.