Bayesian Hierarchical Model With Classification And Regression Tree In Predicting Loan Default

ABSTRACT Bayesian modelling as well as decision tree methods are some of the efficacious classification methods in credit scoring applications. Application of these methods to credit scoring provides several advantages, which are highlighted in the literature. In this research, a Bayesian hierarchical model using the latent variable approach coupled with the classification and regression tree ‘CART’ approach is applied to classify customers who applied for loans into potential defaulters and non-defaulters. The contribution of this paper is two-fold. First, we apply a Bayesian hierarchical model, it is assumed that there is a latent variable relationship between applicant’s credit amount, age, credit duration and credit history from which predictive posterior estimate is acquired. Subsequently the predictive posterior mean of each observation coupled with the CART is used to generate decisions. Secondly, we assess the effectiveness of Bayesian Hierarchical model with the Classification and regression tree approach in predicting loan defaulters. The finding from the research shows that the Bayesian Hierarchical model with the CART approach has a predictive accuracy of 91% as compared to the singular use of the CART model. The Bayesian Hierarchical model with classification and regression tree is therefore built to improve upon the credit granting decisions of financial institution