Nasrin Motedayen, Rafik Nazarian, Marjan Damankeshideh, Roya Seifi Pour,
Volume 12, Issue 45 (11-2021)
Abstract
Credit risk is the probability of default of the borrower or the counterparty of the bank in fulfilling its obligations, according to the agreed terms. In other words, uncertainty about receiving future investment income is called risk, which is of great importance in banks. The purpose of this article was to estimate the credit risk of Mellat Bank's legal customers. In this study, the statistical information of 7330 real customers was used. In this regard, the results of neural network model and support vector machine model have been compared. The obtained results have shown that the components considered in this study based on personality, financial and economic characteristics had significant effects on the probability of customer default and credit risk calculation. Also, the results of this study showed that the application of control policies at the beginning of the repayment period suggests facilities that have the highest probability of default with long life and high repayment. Comparing the results obtained from the prediction accuracy of different models, it was observed that the explanatory power of the support vector machine model and the use of the survival probability function was higher than that of the simple neural network model for the studied groups of real customers.
Seyed Ahmad Ameli, , ,
Volume 16, Issue 59 (5-2025)
Abstract
By designing an efficient loan management system, banks can increase efficiency and reduce the probability of non-repayment of principal and sub-loans. In this paper, the efficiency of logistic regression models, artificial neural network, was examined to predict the credit risk of real customers or in other words, applicants for microloans, which include a large group of customers in the country's banking system. Given the imbalance of the number of data, the optimal threshold was calculated using two sensitivity and detection curves, and the credit risk of each model was extracted from this method. In logistic regression, the compensated maximum likelihood method was used to estimate the coefficients considering the small number of bad customers instead of the maximum likelihood method. Finally, the accuracy and precision of each model was examined with multiple criteria. Using the Rock curve, the resolution of the models was examined, where the neural network model had the best resolution. Then, by comparing the MSE, RMSE and MAE errors, the efficiency of the methods was compared, and the performance of MPLE logistics and neural network is almost the same. Finally, considering the bank's goal in three scenarios of minimum credit risk, identifying good customers and separating customers, neural network, MPLE logistics, and in the third scenario, neural network and MPLE logistics simultaneously have been selected as the best models.
