Multivariate kernel discrimination for bank loans

Abstract

The purpose of this dissertation is to obtain a kernel discriminant model to classify bank loans and determine which loans are at risk of default. This study starts by introducing the concept of kernel density estimation, which is a widely used non-parametric technique to obtain an estimate for the probability density function. This procedure is based on two main parameters: the kernel function and the bandwidth, the latter being the crucial parameter. The main focus of this dissertation is the multivariate case of kernel density estimation and this nonparametric technique is later applied to discriminant analysis to obtain kernel discrimination. This is a method which classifies observations into a predetermined number of distinct and disjoint classes. Finally, we apply multivariate kernel discriminant analysis to a sample of bank loans in order to determine which loans can be actually classified as defaulted.