Gated Fusion Enhanced Multi-Scale Hierarchical Graph Convolutional Network for Stock Movement Prediction

TL;DR

The paper tackles stock movement prediction by modeling dynamic inter-stock relationships and intra-stock attribute patterns across multiple time scales. It proposes MS-HGFN, a framework that combines a Multi-Scale Sampling Module, a Temporal-Transformer–driven hierarchical GNN, and a top-down gated fusion mechanism to preserve both coarse- and fine-grained features. The method demonstrates up to 1.4 percentage points improvement in accuracy and improved MCC and backtesting stability on US and Chinese market datasets, outperforming several strong baselines. This work provides a robust, scalable approach for multi-scale, spatio-temporal stock forecasting with practical implications for trading strategies and risk management.

Abstract

Accurately predicting stock market movements remains a formidable challenge due to the inherent volatility and complex interdependencies among stocks. Although multi-scale Graph Neural Networks (GNNs) hold potential for modeling these relationships, they frequently neglect two key points: the subtle intra-attribute patterns within each stock affecting inter-stock correlation, and the biased attention to coarse- and fine-grained features during multi-scale sampling. To overcome these challenges, we introduce MS-HGFN (Multi-Scale Hierarchical Graph Fusion Network). The model features a hierarchical GNN module that forms dynamic graphs by learning patterns from intra-attributes and features from inter-attributes over different time scales, thus comprehensively capturing spatio-temporal dependencies. Additionally, a top-down gating approach facilitates the integration of multi-scale spatio-temporal features, preserving critical coarse- and fine-grained features without too much interference. Experiments utilizing real-world datasets from U.S. and Chinese stock markets demonstrate that MS-HGFN outperforms both traditional and advanced models, yielding up to a 1.4% improvement in prediction accuracy and enhanced stability in return simulations. The code is available at https://anonymous.4open.science/r/MS-HGFN.

Figures (4)

• Figure 1: The schematic diagram of correlation between two stocks by considering inter-attributes and intra-attributes
• Figure 2: Overall structure of MS-HGFN. (a) Average pooling is used for K-scale sampling. (b) Temporal dependencies are explored using transformer backbone. (c) Adjacent matrices are created at each scale to dynamically model spatial dependencies using patterns learned from intra-attributes and features learned from inter-attributes. (d) Top-down multi-scale feature fusion is performed at different scales to balance coarse- and fine-grained features. Finally, the fused features are sent to the prediction module.
• Figure 3: Ablation study
• Figure 4: Performance diagram of backtesting for all methods.