Acceptance and profitability modelling for consumer loans

Abstract

This thesis explores and models the relationships between offers of credit products, credit scores, consumers' acceptance decisions and expected profits generated using data that records actual choices made by customers and their monthly account status after being accepted. Based on Keeney and Oliver's theoretical work, this thesis esti¬ mates the expected profits for the lender at the time of application, draws the iso-profit curves and iso-preference curves, derives optimal policy decisions subject to various constraints and compares the economic benefits after the segmentation analysis.This thesis also addresses other research issues that have emerged during the explo¬ ration into profitability and acceptance. We use a Bivariate Sample Selection model to test the existence of sample selection bias and found that acceptance inference may not be necessary for our data. We compared the predictive performance of Support Vector Machines (SVMs) vs. Logistic Regression (LR) on default data as well as on accep¬ tance data, without finding that SVMs outperform LR. We applied different Survival Analysis models on two events of interest, default and paying back early. Our results favoured semi-parametric PH-Cox models separately estimated for each hazard.