Abstract:

Due to the complex dynamic evolution of correlations within financial systems， the diversified， multi-channel characteristics of financial risk contagion and its spillover effects have become increasingly prominent. Concurrently， the challenges associated with systemic financial risk prevention and control have intensified， making effective risk management a critical issue requiring urgent solutions. This study investigates China's money market， capital market， foreign exchange market， gold market， and real estate market. Firstly， we employ generalized forecast error variance decomposition and complex network analysis to examine risk spillover effects in China's financial markets from both static and dynamic perspectives. Subsequently， a Time-Varying Parameter Vector Autoregression （TVP-VAR） model is utilized to explore the impact of macroeconomic conditions， micro-level individual behaviors， and network topology on systemic financial risk spillovers. Finally， we enhance the prediction accuracy of systemic financial risk by optimizing BP neural network and Logit models through deep belief network architecture. The experimental results reveal three key findings （i） Risk spillover analysis demonstrates that cross-market spillover effects significantly surpass intra-market effects. Volatile economic conditions have substantially altered risk transmission pathways， with the stock market and real estate market emerging as primary risk transmitters and receivers. （ii） Impact effect analysis shows an inverse relationship between macroeconomic performance/micro-level expectations and systemic financial risk. Economic expansion and optimistic consumption expectations correlate with subdued risk spillovers， whereas economic contraction and pessimistic expectations amplify systemic risk propagation. Network structure exhibits complex nonlinear associations with risk spillovers. （iii） Risk early warning tests indicate that deep belief network-optimized models significantly improve systemic risk prediction accuracy， validating the inclusion of these indicators in financial risk warning systems. These findings provide substantial theoretical support for establishing systemic financial risk warning mechanisms， formulating risk prevention strategies， and developing macroeconomic regulation policies. The research holds significant practical value for maintaining stable economic growth and achieving dynamic equilibrium in financial risk management.

Key words: financial markets, impact effect, risk early warning, deep learning, deep belief network