双层动态金融网络中心性与资产配置效率——基于MT-HITS算法的实证研究

Two-Layer Dynamic Financial Network Centrality and Asset Allocation Efficiency—An Empirical Study Based on the MT-HITS Algorithm

摘要: 金融市场中资产间的关联不仅包含线性风险传导特征，还包含复杂的非线性信息交互机制，但传统基于相关系数构建的金融网络已难以全面刻画市场真实的运行规律。本文基于沪深300成分股的日交易数据，采用滚动窗口法构建融合Granger线性因果关系与Transfer Entropy非线性信息流的双层动态金融网络，并使用六阶张量融合时间维度与层间关系，再用MT-HITS算法计算出股票网络的Hub中心性与Authority中心性，并在此基础上，构建网络中心性选股策略，并结合CVaR模型进行投资组合优化。结果表明：金融市场具有明显的动态变化特征，使用低Authority中心性策略的投资效果显著优于基准策略，即可说明网络核心股票往往承担较高的系统性风险，而处于网络边缘的股票则具有更优的风险收益特征。

Abstract: The interconnections among assets in financial markets encompass both linear risk transmission characteristics and complex nonlinear information interaction mechanisms. However, traditional financial networks constructed based on correlation coefficients are inadequate for fully capturing the true operational patterns of the market. This paper utilizes daily trading data of CSI 300 constituent stocks and adopts a rolling window approach to construct a two-layer dynamic financial network that integrates Granger causality for linear risk transmission and Transfer Entropy for nonlinear information flow. Furthermore, a sixth-order tensor is employed to integrate the temporal dimension with inter-layer relationships, and the MT-HITS algorithm is applied to compute the Hub centrality and Authority centrality of stocks within the network. On this basis, a stock selection strategy based on network centrality is developed and combined with the CVaR model for portfolio optimization. The results show that the financial market exhibits significant dynamic evolution characteristics. The investment strategy based on low Authority centrality significantly outperforms the benchmark strategy, indicating that core stocks in the network tend to bear higher systemic risk, while stocks at the network periphery demonstrate superior risk-return profiles.