Credit risk is one of the main risks faced by commercial banks. Credit risk management includes risk identification, assessment, and early warning, among which risk assessment is fundamental and key. Currently, research on credit risk assessment in China is still in its developing stage, and the precision of measuring credit risk needs improvement. Among various evaluation methods, the KMV model has shown good practical application and is relatively suitable for the national conditions of China. However, it still has some flaws. To address the issue of insufficient external validity in the default point parameter settings of the KMV model, the PSO algorithm is used to optimize these parameters, and the PSO-GWO algorithm is integrated to construct the APSO-KMV model and the PSO-GWO-KMV model. Based on an empirical study comparing real data from 5,234 companies, it was found that the original KMV model had an AUC value of 0.7362, accuracy of 0.2610, and binary cross-entropy loss of 0.7006; the PSO-KMV model had a short-term debt coefficient α of 0.0496, a long-term debt coefficient β of 0.2508, an AUC value of 0.9994, accuracy of 0.9996, and binary cross-entropy loss of 4.1990; the PSO-GWO-KMV model had a coefficient α of 0.0496 and a value β of 0.2690, an AUC value of 0.9987, accuracy of 0.7603, and binary cross-entropy loss of 4.0804. The optimized KMV model showed a significant improvement in predictive accuracy.
Research Article
Open Access