Evaluating Credit Risk Based on Combined Model of Neural Network of Pattern Recognition and Ants’ Colony Algorithm

Abstract

A great amount of potential financial losses arise from borrowers’ abstaining from refunding their debts calls and the development and improvement of credit risk measurement techniques in the financial literature in order to decrease such losses has transformed into an intevitable subject. The purpose of bankruptcy forecasting models is to estimate the probability of a company or a person’s abstaining during a certain period of time. This research used the data gathered from a sample of 218 active companies in Tehran Stock Exchange Market as well as Over-The-Counter for the period between 1990 and 2016. Moreover, ants’ colony algorithm was used to determine the most effective factors of credit risk and also pattern recognition neural network technique was applied to classify and evaluate the precision of bankruptcy forecasts. As a result, such ratios as profit before interests and taxes to total sale; total benefits of shareholders to debts; and current ratio, cash ratio and shareholders’ benefits ratio to total assets are the most effective factors. Finally, the presented model which employs data belonging to one, two and three years before the intended year is able to forecast the credit condition of companies with higher precision as compared to the average precision of current models.

Keywords

• credit risk
• bankruptcy probability
• neural network
• pattern recognition algorithm
• ants’ colony aalgorithm.