气候风险的金融传染：来自跨行业复杂网络与机器学习的证据

摘要

Identifying the financial linkages associated with climate risk is crucial for preventing systemic risk. Based on annual reports of listed firms, this paper constructs physical and transition climate risk indices reflecting firms' attention and disclosure of climate risks, and characterizes their interdependencies with cross-industry financial risks. We adopt complex network analysis to map risk-linkage networks and measure industry resilience, apply econometric models to examine linkage characteristics, and use Random Forest (RF), XGBoost and CatBoost to predict systemic risk contagion. The results indicate that brown industries feature stronger index-financial risk associations than green industries. Physical risk linkages are short-term, whereas transition risk linkages are more persistent. Linkage characteristics vary with industries' positions in the industrial structure and are moderated by climate policy. Disaster losses, management's climate perception and investor sentiment drive the observed heterogeneity. The RF model delivers the best prediction performance for risk contagion. In conclusion, there is a significant systemic association between China's climate risk indices and industry-level financial risks, and risk prevention efforts should focus on industry-level structural linkage features.