A Distillation-based Future-aware Graph Neural Network for Stock Trend Prediction

TL;DR

This work tackles stock trend prediction under distribution shifts between historical and future data, noting that prior GNNs focusing only on historical patterns underperform in volatile markets. It introduces DishFT-GNN, a distillation-based, future-aware framework with a teacher model that encodes historical spatiotemporal patterns and encoded future trends, fused via an attention-based multi-channel mechanism to produce future-aware embeddings. The student model is trained to learn these future-aware representations through HSIC-guided distillation, combining a standard prediction loss with a distillation term. Empirical results on S&P 100 and NASDAQ 100 datasets show consistent accuracy and MCC gains, with backtesting profits that widen over time, demonstrating the practical value of modeling historical–future distribution correlations for robust stock forecasting. Overall, the approach provides a principled way to inject future-oriented supervision into GNN-based stock predictors, yielding state-of-the-art performance and actionable profitability improvements.

Abstract

Stock trend prediction involves forecasting the future price movements by analyzing historical data and various market indicators. With the advancement of machine learning, graph neural networks (GNNs) have been extensively employed in stock prediction due to their powerful capability to capture spatiotemporal dependencies of stocks. However, despite the efforts of various GNN stock predictors to enhance predictive performance, the improvements remain limited, as they focus solely on analyzing historical spatiotemporal dependencies, overlooking the correlation between historical and future patterns. In this study, we propose a novel distillation-based future-aware GNN framework (DishFT-GNN) for stock trend prediction. Specifically, DishFT-GNN trains a teacher model and a student model, iteratively. The teacher model learns to capture the correlation between distribution shifts of historical and future data, which is then utilized as intermediate supervision to guide the student model to learn future-aware spatiotemporal embeddings for accurate prediction. Through extensive experiments on two real-world datasets, we verify the state-of-the-art performance of DishFT-GNN.

Figures (3)

• Figure 1: An example shows that events, such as a corporation’s annual report, cause sudden structural changes between historical and future stock data.
• Figure 2: The framework of DishFT-GNN
• Figure 3: Back-testing illustration.