Industry-Specific credit risk identification for bank lending risk monitoring: An explainable machine learning approach based on chinese listed firms

Abstract

Corporate credit risk assessment is central to bank lending because banks need to evaluate borrowers’ future cash-flow and debt-servicing capacity. Although prior studies show that financial ratios, governance quality, audit information and financing pressure are useful credit risk signals, less attention has been paid to organising these predictors into economically meaningful risk-source categories, examining whether their importance differs across industries, and linking interpretable signals to lending review and risk monitoring. This study develops an explainable machine learning framework for industry-specific credit risk identification using Chinese A-share listed firms. Based on 25,222 firm-year observations from manufacturing, construction and information technology firms, a forward-looking ST/*ST-based credit-risk proxy is constructed by matching firm-level indicators in year t with ST/*ST status in year t+1. Logistic Regression and XGBoost are used for prediction, and SHAP is applied to interpret the trained XGBoost model. The results show that XGBoost outperforms Logistic Regression. Group-level SHAP results indicate that financial and operating deterioration risk is the dominant source of predicted credit risk, followed by governance and audit information risk and financing pressure risk. Variable-level SHAP results further reveal industry-specific risk channels. The findings suggest that explainable machine learning can support differentiated lending review and post-loan monitoring.