Guarantees Risk Management in a Financial Institution

Overdue claims resulting from the lending process can pose a significant credit risk to financial institutions. To mitigate this risk, institutions often acquire guarantees. However, borrowers may encounter challenges when providing adequate and valid guarantees, particularly guarantees with lower risk. The present research focused on loan credit risk, borrower utility, and liquidity risk of guarantees within a private fund. First, data mining and classification methods were applied to a dataset of loans. The random forest algorithm, with a prediction accuracy of 0.986, was found to be optimal for constructing a guarantees composition model. The guarantees composition involves using multiple types of guarantees to secure a loan. Two models were established to generate guarantee compositions with a maximum default rate of 10%. In testing scenarios, the average risk of total default for acceptable combinations stands at 3.94%, a significant improvement compared to the fund loans’ overall default rate of 6.3%. Furthermore, the proposed model increases borrower utility from 4.22 to 4.6, not only reducing the default rate but also enhancing borrower utility.IntroductionWhen providing loans to customers, banks require guarantees due to insufficient knowledge of customers and the default risk. Obtaining guarantees from borrowers is recognized as a solution to reduce default risk in banks, but its impact on risk reduction depends on various factors. The combination and type of guarantees are among these factors, which have received less attention in the literature.The current state of overdue bank claims in Iran is unfavorable, and if conditions persist, it will lead to significant monetary and financial crises with negative effects on various sectors of the economy. In recent years, the ratio of non-performing loans to total disbursed facilities in Iran has been consistently higher, averaging around 5% to 10% higher than the global average. Reduction of the default risk in loans can decrease the ratio of non-performing loans to total disbursed facilities.The present study first intended to create various combinations of guarantees for each loan, followed by predicting the probability of default for each combination. In line with their priorities, borrowers can then select their desired guarantee composition from a list of acceptable combinations.MethodologyTo address the research problem, the study identified common classification methods in data mining by relying on published articles in the field of credit risk. Then, a sample dataset of loans from a financial institution was examined, and the data mining process based on classification methods was applied to the dataset. The random forest method, with a prediction accuracy of 0.986, was ultimately chosen as the approach for constructing the guarantee composition model. Using the previous guarantee compositions, the study developed two models by relying on machine learning techniques. These compositions take into account the perspectives of both the financial institution and the borrower.Final ResultThe two models generate guarantee compositions with a maximum acceptable default rate of 10%. Considering their own priorities circumstances, borrowers can select their desired guarantee composition from the available combinations, which contributes to a reduction in the default rate in the financial institution.