StockDGAT: Dynamic Graph Attention Network for Stock Price Movement Prediction

Financial markets are vital to the global economy, and accurately predicting their behaviors is crucial for traders and investors. However, the volatile nature of these markets complicates predictions, with traditional methods often failing to capture the complex interrelationships among stocks. This paper presents a novel framework that utilizes Graph Attention Networks (GATs) to improve stock price movement predictions. By creating a dynamic graph representation of stock interrelations over time and leveraging historical data, our approach uncovers the nuanced dependencies that influence price changes. Experimental results on NASDAQ stock data demonstrate that our GAT framework significantly outperforms state-of-the-art techniques in both accuracy and computational efficiency. With a lower-than-quadratic complexity, our model achieves better prediction accuracy, outperforming other models with quadratic complexities. This combination of reduced complexity and improved accuracy emphasizes the effectiveness and practical applicability of our approach, marking an advancement in graph-based methodologies for financial forecasting.