Using a neural network-based methodology for credit–risk evaluation of a Tunisian bank

الملخص EN

Credit–risk evaluation is a very important and challenging problem for financial institutions.

Many classification methods have been suggested in the literature to tackle this problem.

Neural networks have especially received a lot of attention because of their universal approximation property.

This study contributes to the credit risk evaluation literature in the MENA region.

We use a multilayer neural network model to predict if a particular applicant can be classified as solvent or bankrupt.

We use a database of 1100 files of loans granted to commercial and industrial Tunisian companies by a commercial bank in 2002 and 2003.

Our main results are : a good capacity prediction of 97.1% in the training set and 71% in the validation set for the non cash-flow network.

The introduction of cash-flow variables improves the prediction quality to 97.25% and 90% respectively both in the in-sample and out-of-sample sets.

Introduction of collateral in the model substantially improves the prediction capacity to 99.5% in the training dataset and to 95.3% in the validation dataset.