Abstract: This paper applies the dynamic semiparametric quantile regression model to construct network structure, clarifying the extreme risk contagion effects among financial institutions. We consider the market sentiments as conditional variables in the network model. The results show that the systemic risk contribution and exposure as well as tail risk contagion degree hold remarkable different information and characteristics. The network contagion degree ranking of financial firms, the tail risk receiver ranking, and tail risk emitter ranking are significant heterogeneous. During the economic and financial turmoil, the network contagion effects within three financial sectors (banking, securities, and insurances) have risen sharply, and the interconnectedness within securities is significantly higher and more volatile than between banks. The financial network model based on the rolling window width selection criteria optimizes the time-varying identification process of risk contagion relationships among financial institutions. The supervision on the highly contagious financial institutions has become a crucial policy issue. Therefore, exploring the network dependence of extreme risks is crucial for financial regulators to improve the supervision efficiency of the financial system.

Key words: systemic risk, CoVaR, semiparametric regression, network contagion, market sentiment