Credit risk quantification: application of neural networks and fuzzy logic

Abstract

Credit risk is one of the key challenges faced by banks worldwide. Credit risk is the possibility that a borrower fails to meet agreed debt obligations. Banks need to maintain credit risk exposure within specified parameters in order to maximize the risk-adjusted rate of return. For most banks, loans represent the major source of credit risk. Banks need to predict the possibility of default of a potential customer before a loan is extended. In this project we have tried to apply Fuzzy logic technique to predict bankruptcy and hence quantify credit risk. We have used MATLAB® as the tool to implement fuzzy logic. While both the Mamdani and the Sugeno models are supported, we have implemented the Sugeno model which is more suitable for this application. Fuzzy logic is a multi-valued logic based on fuzzy sets that tries to mimic the judgment of a human expert and is useful in modeling nonlinear systems. Fuzzy sets classify elements as belonging to a set based on the extent of membership rather than an absolute membership. We have briefly explored the various developments in Credit risk Management over the last twenty years. We have compared the performance of the Fuzzy logic with neural network and classification tree models. We conclude that neural networks and classification tree models tend to perform slightly better than Fuzzy methods in predicting bankruptcy.