Analysis of an Intelligent Early Warning Model for Credit Risk in Listed Companies Based on Multi-Source Data Fusion

Yuqiao Shao

doi:10.54097/t6f91c45

Authors

Yuqiao Shao School of Finance, Tianjin University of Finance and Economics, Tianjin, 300222, China

DOI:

https://doi.org/10.54097/t6f91c45

Keywords:

Credit Risk Early Warning, Multi-source Data Fusion, XGBoost, Intelligent Risk Control.

Abstract

As global financial markets deepen their integration and accelerate digital transformation, the complexity and significance of credit risk management have become increasingly prominent. This study proposes an intelligent early warning model for listed companies' credit risk based on multi-source data fusion. By integrating financial, public sentiment, online, and compliance data across four dimensions, the model aims to enhance the accuracy and foresight of risk identification. Using non-financial A-share listed companies from 2017 to 2022 as the sample, the study constructed a fusion model based on the XGBoost algorithm and compared it with a benchmark logistic regression model using only financial data. Empirical results show that the fusion model achieved an Area Under Curve (AUC) value of 0.941 and an F1-Score of 0.742 on the test set, significantly outperforming the benchmark model. Feature importance analysis further reveals that alternative data features such as negative sentiment indices and litigation amounts possess predictive power comparable to traditional financial indicators, confirming the incremental informational value of multi-source data in credit risk assessment. This study provides theoretical support and practical reference for intelligent risk control, suggesting future integration with explainable AI technologies to further optimize model transparency and application scope.

Downloads

Download data is not yet available.

References

[1] Bardos, K. S., Zhu, J., & Zhang, M. (2020). The value of alternative data in credit scoring: Evidence from a large field experiment. Journal of Banking & Finance, 121, 105961.

[2] Cao, Y., Jiang, J., & Li, X. (2023). The role of supply chain networks in predicting corporate financial distress: A machine learning approach. Journal of Corporate Finance, 78, 102345.

[3] Chen, M., Liu, T., Chen, M., et al. (2020). Early warning model of enterprise default risk based on multi-source big data. Journal of Management Sciences in China, 23(5), 1–25.

[4] Chen, T., & Guestrin, C. (2016). XGBoost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining(pp. 785–794).

[5] Dou, E. X., & Xin, L. (2022). Application of RegTech in risk early warning: Theoretical framework and China practice. International Finance Research, (4), 12–15.

[6] Duan, T., Feng, J., & Li, Y. (2022). Beyond financials: The value of alternative data in corporate credit risk assessment. Journal of Financial Economics, 145(2), 507–527.

[7] Huang, Y., Li, X., & Zhang, H. (2021). Social media and corporate credit risk: Evidence from China. Pacific-Basin Finance Journal, 68, 101596.

[8] Lundberg, S. M., & Lee, S. I. (2017). A unified approach to interpreting model predictions. In Advances in Neural Information Processing Systems (pp. 4765–4774).

[9] Mai, F., Tian, S., Lee, C., & Ma, L. (2021). Deep learning models for corporate credit risk assessment with textual disclosures. Journal of Financial and Quantitative Analysis, 56(6), 2012–2045.

[10] Shimpi, P. A. (2017). The rise of alternative data: Data protection, privacy and other legal issues. Journal of Investment Compliance, 18(4), 42–47.

[11] Zhang, Y., Li, J., & He, J. (2023). Explainable AI in credit risk modeling: An application of SHAP to corporate loan defaults. Expert Systems with Applications, 213, 118887.