基于复杂网络方法的中国金融市场联动分析

Analysis of Interconnectedness in China’s Financial Markets Based on the Complex Network Approach

摘要: 本文基于2017~2025年中国股票、外汇、债券、货币、黄金及期货市场高频数据，设计动态波动识别调整时间窗口，结合线性格兰杰因果检验与Hiemstra-Jones非线性检验构建因果关系网络，并运用核回归与VAR脉冲响应函数量化传染强度。引入图注意力网络模型，通过余弦退火学习率调度和混合损失函数实现风险传染路径的动态拓扑建模。实证发现：中国金融市场风险传导呈现显著非对称性，外汇市场为核心枢纽，尤其美元兑在岸人民币汇率与美元兑离岸人民币汇率具有显著双向联动关系；股票与期货市场自解释性较高。但股票市场仍受外汇、债券及期货市场外溢冲击；货币市场作为共同影响因素广泛关联各子市场，债券市场对外汇冲击敏感，黄金市场则承受多市场联合冲击。

Abstract: Based on high-frequency data from China’s stock, foreign exchange, bond, money, gold, and futures markets spanning 2017~2025, this paper designs a dynamically adjusted time window driven by volatility identification. It constructs a causal relationship network by integrating the linear Granger causality test and the Hiemstra-Jones nonlinear test, and employs kernel regression and the VAR impulse response function to quantify contagion intensity. Additionally, the Graph Attention Network (GAT) model is introduced, and dynamic topological modeling of risk contagion paths is achieved through cosine annealing learning rate scheduling and a hybrid loss function. Empirical results reveal that risk transmission in China’s financial markets exhibits significant asymmetry: the foreign exchange market acts as a core hub, with particularly prominent two-way interconnectedness between the USD/onshore RMB exchange rate and the USD/offshore RMB exchange rate. The stock and futures markets demonstrate high self-explanatory power, yet the stock market remains susceptible to spillover shocks from the foreign exchange, bond, and futures markets. As a common influencing factor, the money market maintains extensive connections with various sub-markets; the bond market is sensitive to foreign exchange shocks, while the gold market endures joint shocks from multiple markets.