Variable Selection for Factors Affecting Credit Costs and Comparison of Estimate Model

Abstract

Earlier studies that estimated the credit cost of banks used explanatory variables and estimate models which had been already used in further earlier studies. This study explored some key explanatory variables from the mound of factors by using a multi-regression analysis and CART. Also this study assessed the suitability of a multi-regression analysis, neural network, and support vector machine as an estimate model. As a result of the study, the loan amount per customer and the deposit growth rate in the city where headquarters is located were selected as key variables. Also non-parametric method, such as neural network and support vector machine was proven suitable as a variable selection method and an estimate model on the ground that the relation between the credit cost and explanatory variables is non-linear and non-continuous.