Modeling the Credit Risk Assessment of the Clients of the Export Guarantee Fund of Iran (EGFI) Using Machine Learning Methods and Neural Network

One of the most important issues that credit & financial institutions are faced with, is the probability of non-fulfillment of obligations by the receivers of credit facilities on due dates. With an accurate forecast of risks, we will not only be able to do the relevant estimation of the premium and the percentage of required collaterals more precisely, we will also witness a meaningful decline in the volume of claims related to non-fulfillment of obligations covered under credit guarantees, a decrease in the debt recovery costs and increased efficiency, which all together, will pave the ground for more competitive operation of credit institutions. This research has been conducted with the aim of modeling the credit risk assessment of the clients of the Export Guarantee Fund of Iran (EGFI) using Kernel Support Vector Machines (KSVM- SVM) and Group Method Data Handling (GMDH) . The results of this research indicate that in credit risk assessment of the applicants of credit facilities, the KSVM based modeling is more reliable and accurate in comparison with other studied models. Also with regards to the model’s output ratios, 5 variables such as history of company (years in operation since foundation), credit tenor (term of the credit received), average of export operations (amount of exports made by the company since foundation), payment records and the company’s turnover and its balance of debt (financial stability of the client) have the highest impacts on the client’s credit assessment.