Journal of Accounting and Management Information Systems (JAMIS)

Credit–risk evaluation of a Tunisian commercial bank: logistic regression vs neural network modelling

Vol. 9, Nr. 1/2010 ,   p92..119


Keywords:   Banking sector, Accounting data, Credit risk evaluation, Default risk Prediction, Neural network Models

Abstract:   This paper addresses the question of default prediction of short term loans for a Tunisian commercial bank. We make  a comparative analysis of two different statistical method of classification (artificial neural network and linear logistic regression with panel data). We use a database of 1434 files of credits granted to industrial Tunisian companies by a commercial bank in 2003, 2004, 2005 and 2006. The results show that the best prediction model is the  multilayer neural network model  and the best information set is the one combining accrual, cash-flow and collateral variables. We got a good classification rate of  97% in the training data set and 89.8% in the validation data set. Moreover, our investigation shows that the bank could have reduced the percentage of loss (failure on short term credits) from 18.7% to 12% during 2007.