EP-GAT: Energy-based Parallel Graph Attention Neural Network for Stock Trend Classification

2025-07-10

AI Summary

Existing graph neural network (GNN) approaches for modeling stock correlations rely on static or handcrafted priors, failing to capture time-varying dependencies and hierarchical, dynamic characteristics of individual stocks. To address this, we propose a Dynamic Stock Graph Modeling framework: (1) an adaptive temporal graph construction mechanism leveraging energy differences and the Boltzmann distribution to model evolving stock relationships; (2) a parallel graph attention network that jointly captures intra-stock multi-scale temporal patterns and inter-stock interaction dynamics; and (3) a hierarchical feature fusion module integrating representations across granularity levels. Evaluated on real-world stock datasets from five countries (China, the U.S., the U.K., etc.), our model consistently outperforms five state-of-the-art baselines. Ablation studies and hyperparameter sensitivity analyses confirm the effectiveness and robustness of both the dynamic graph construction and the parallel attention architecture.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Reasoning under Uncertainty: Graphical Models

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Models for Web evolution

Abstract

Graph neural networks have shown remarkable performance in forecasting stock movements, which arises from learning complex inter-dependencies between stocks and intra-dynamics of stocks. Existing approaches based on graph neural networks typically rely on static or manually defined factors to model changing inter-dependencies between stocks. Furthermore, these works often struggle to preserve hierarchical features within stocks. To bridge these gaps, this work presents the Energy-based Parallel Graph Attention Neural Network, a novel approach for predicting future movements for multiple stocks. First, it generates a dynamic stock graph with the energy difference between stocks and Boltzmann distribution, capturing evolving inter-dependencies between stocks. Then, a parallel graph attention mechanism is proposed to preserve the hierarchical intra-stock dynamics. Extensive experiments on five real-world datasets are conducted to validate the proposed approach, spanning from the US stock markets (NASDAQ, NYSE, SP) and UK stock markets (FTSE, LSE). The experimental results demonstrate that EP-GAT consistently outperforms competitive five baselines on test periods across various metrics. The ablation studies and hyperparameter sensitivity analysis further validate the effectiveness of each module in the proposed method.

Problem

• Model dynamic inter-dependencies between stocks effectively
• Preserve hierarchical intra-stock dynamics accurately
• Improve stock trend classification performance significantly

Innovation

• Dynamic stock graph with energy difference
• Parallel graph attention mechanism
• Hierarchical intra-stock dynamics preservation