Risk Analysis of Local Government Bonds Based on LSTM and Logistic Regression Models

DOI: https://doi.org/10.62517/jse.202611205

Author(s)

Siyao Song

Affiliation(s)

Beijing Normal-Hong Kong Baptist University, Zhuhai, Guangzhou, China

Abstract

The local government bond is a crucial component of the financial market, and the accuracy of the risk assessment for the bond is of great significance for optimizing investor decision-making. However, the existing traditional nonlinear model, such as logistic regression, is not able to effectively grasp the nonlinear characteristics of bond risk, while there are limitations in balancing "dynamic risk capture" and "logical rule interpretability" for the existing deep learning model. This paper builds a research dataset by data integration and feature engineering, through multidimensional time-series data such as the volume of local government bond issuance and GDP growth rate, to develop a separate LSTM model, a logistic regression model, and a hybrid "LSTM + logistic regression" model. The study compares the advantages of the three models for risk prediction accuracy and interpretability. Additionally, it adds SHAP value analysis to improve the explainability of the model. The results show that the hybrid model achieves "dynamic feature capture" and "logical rule interpretation," thereby providing local government debt regulators with accurate risk warning tools and providing a scientific basis for investors to allocate assets and optimize bond pricing mechanisms. This research attempts to fill the gap in existing research in the comprehensive model application of local government bond risk assessment for financial risk management innovation.

Keywords

Local Government Bond Risk; LSTM Model; Logistic Regression; Model Fusion; Risk Assessment

References

[1] Chen, X. B., Wu, J., & Xu, M. D. (2021). Analysis of bond default risk based on LSTM and MCM. New Finance, (6), 54-59. [2] Du, H., Song, Y., Piao, T., & Shi, H. (2024). Research on Financial Risk Intelligence Monitoring and Early Warning Model Based on LSTM, Transformer, and Deep Learning. Journal of Organizational and End User Computing, 36(1), 1–24. [3] Li et al., "Research on Financial Risk Early Warning Based on Deep Neural Network and Logistic Model," 2024 IEEE 2nd International Conference on Sensors, Electronics and Computer Engineering (ICSECE), Jinzhou, China, 2024, pp. 1143-1149. [4] Liu, Q. K. (2019). Research on China's Credit Bond Default Early Warning Mechanism: An Empirical Analysis Based on Random Forest and Logistic Regression Models [D]. Shanghai: Shanghai University of Finance and Economics. [5] Maher, J. J., & Sen, T. K. (1997). Predicting Bond Ratings Using Neural Networks: A Comparison with Logistic Regression. International Journal of Intelligent Systems in Accounting, Finance & Management, 6(1), 59-72. [6] Nunes, M., Gerding, E., McGroarty, F., & Niranjan, M. (2020). Long short-term memory networks and laglasso for bond yield forecasting: Peeping inside the black box. arXiv. [7] Sun, C. T., & Chen, H. C. (2024). Research on Bond Default Early Warning and Proactive Rating Systems. Far East Credit Rating Co., Ltd. [8] Zhao, M. (2022). Research on bond credit risk of commercial and retail enterprises in the post-pandemic era. Research on Commercial Economics, (19), 132-135.