1. According to China’s 2025 Government Work Report, preventing and mitigating local government debt risk, financial institution risk, and real estate risk are among the top three issues in maintai...
I. Countries around the world face significant challenges in managing government debt, with local government debt risk (LGDR) being particularly prominent in developing countries (Gao, Ru, and Tang...
With the rapid development of digital finance, online public opinion has become an important factor influencing the systemic risks of small and medium-sized financial institutions. This study integrates the DeepSeek and FinBERT large language models to construct a multi-dimensional public opinion index system covering investor attention, emotional disposition, and opinion divergence. It measures risk exposure using ARMA-GARCH-Copula-CoVaR and conducts early warning analysis through deep learning models. The research findings show that the introduction of public opinion indicators can significantly improve the accuracy of early warnings. Among them, the BiGRU-Attention model has the best performance and can effectively capture the dynamic risks. Further analysis reveals that investor attention is the most sensitive leading indicator, and its predictive ability is significantly better than that of the emotion and divergence indicators. The research results provide a new perspective for financial risk prevention and control in the digital era, and offer important inspirations for building an intelligent monitoring system and implementing differentiated supervision.