中国保险业系统性风险的评估与预警研究——基于Attention-LSTM模型的分析

师荣蓉1; 杨娅2

中国保险业系统性风险的评估与预警研究——基于Attention-LSTM模型的分析

引用本文：师荣蓉1，杨娅2.中国保险业系统性风险的评估与预警研究——基于Attention-LSTM模型的分析[J].财经理论与实践,2025,(2):26-34

摘要点击次数: 99

全文下载次数: 65

作者	单位
师荣蓉1，杨娅2	(1.西北大学经济管理学院，陕西西安 710127；2.西北大学数学学院，陕西西安 710127)

中文摘要:基于保险业系统性风险传导机制和预警机制的理论分析，利用CoVaR方法评估保险业系统性风险，从微观保险机构和宏观经济环境构建Attention-LSTM模型对保险业系统性风险进行预警分析。研究发现：当遭遇重大事件冲击时，系统重要性保险机构对保险业的风险溢出增加；将金融压力指数纳入风险预警体系，其预测平均绝对误差、均方根误差和平均绝对百分比误差分别降低8.59%、7.27%和4.55%；Attention-LSTM模型能捕捉风险间的关联性和传染性，在预测准确性、泛化能力和时间稳定性方面均优于传统机器学习模型。鉴于此，应建立保险业风险分区管理体系，融合深度学习模型多维度构建保险业系统性风险预警机制。

中文关键词:保险业系统性风险评估预警 Attention-LSTM模型

Research on Systematic Risk Assessment and Early Warning in China’s Insurance Industry: Analysis Based on Attention-LSTM Model

Abstract:Based on the theoretical analysis of the systemic risk transmission mechanism and early warning mechanism in the insurance industry, this study uses CoVaR method to evaluate the systemic risk of the insurance industry, and constructs Attention-LSTM model from the micro insurance institution and macroeconomic environment to carry out early warning analysis of the systemic risk in the insurance industry. The results indicate that when the insurance industry is hit by major events, systemically important insurance institutions have an increased risk spillover effect on the insurance industry. By incorporating the financial stress index into the risk warning system, the mean absolute error, root mean square error, and mean absolute percentage error of predictions are respectively reduced by 8.59%, 7.27%, and 4.55%. The Attention-LSTM model can capture the correlation and contagion between risks, and it outperforms traditional machine learning models in prediction accuracy, generalization ability, and time stability. In view of this, it is necessary to establish a risk zoning management system for the insurance industry, and integrate deep learning models to build a multidimensional systematic risk early warning mechanism in the insurance industry.

keywords:systemic risk of insurance industry assessment early warning Attention-LSTM model

查看全文 查看/发表评论 下载pdf阅读器