The rapid development of China’s financial markets, particularly the expansion of the ETF ecosystem, has made dynamic sector rotation strategies increasingly feasible and cost-effective for investors. At the same time, the growing availability of diverse and high-frequency data has created new opportunities and challenges for developing more accurate and adaptive investment models. Traditional models often rely solely on historical trading data and fail to capture complex inter-sector relationships and heterogeneous information sources. To address these challenges, this paper proposes a novel Temporal Heterogeneous Attention Network (Temporal HAN) for dynamic sector rotation. The model integrates mixed-frequency heterogeneous datasets and captures both temporal dynamics and structural dependencies among sectors. A key innovation is the spatio-temporal encoder, designed to overcome representational bottlenecks in cross-frequency data fusion, and an adaptive relation fusion layer that selectively aggregates information from multiple edge types. Extensive experiments using CSI Level 1 sector data demonstrate that our approach significantly outperforms state-of-the-art baselines in terms of predictive accuracy and portfolio returns, enhancing the efficiency of sector-level risk pricing and offering a robust data-driven multi-asset allocation framework in China.
C.P. Wang (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: