eThesis - School of Business electronic theses

» hse_ethesis_12299.pdf   size:664 KB (679910)

This thesis uses empirical observations on consumer credit behavior to study the determinants of default in Finland. The main objective is to investigate if both socio-demographical and behavioral variables have effect on default. In the thesis I construct three different models to show which variables have predictive power the most. The models are compared in terms of efficiency and power to discriminate between low and high risk customers. The purpose of this study is also to provide practical information for credit companies to create more up-to-date and reliable credit scoring models. I also illustrate how such a model can be constructed to achieve the strategic objectives of the credit institution.

The data set of this paper is from an anonymous consumer credit company who offers loans to retail customers. I have 14 595 observations of customers of which 29% turned out to default their loan. All the applications were received between May 2008 and September 2009 and the default information was captured in December 2009. Out of 30 explanatory variables 23 were socio-demographical and the rest, 7 were behavioral. There are several unique and important features of this data set that enables me to test the impact of both socio-demographical and behavioral variables. The analyses are performed using logistic regression, forward and backward stepwise analysis and several tests with SPSS program.

The main findings are that both socio-demographical and behavioral variables have a notable effect on default. Consistent with previous literature the most significant socio-demographical variables are income, time since last moving, age, possession of credit card, education and nationality. Some behavioral variables seemed to have even more predictive power. Those are the amount of scores the customer obtained, loan size and the information if customer has been granted a loan earlier from the same company. Interestingly, the results have variation to some extent when excluding few of the variables outside the model. The predictive power of all three models is adequate and thus can be employed as a reliable credit scoring model for the credit institutions.