Using Dynamic Graph Neural Networks for Temporal Analysis in Financial Markets

In this study, we explore the use of Dynamic Graph Neural Networks (DGNNs) to predict stock price trends by modeling the temporal dependencies in financial data. A novel sliding-window graph approach is introduced to construct temporal graphs that capture historical stock price movements. Our approach utilizes a custom Graph Neural Network (GNN) architecture optimized for processing these graph-structured data, enabling the prediction of next-day normalized stock prices. The sliding-window method allows for localized temporal dependency modeling, enhancing the GNN's ability to capture short-term trends in financial markets. Extensive experiments demonstrate the effectiveness of the proposed model. DGNN achieves a 0.22% reduction in mean square error (MSE) compared to Recurrent Neural Networks (RNNs) and a 0.44% reduction in MSE compared to Linear Regression models. These results highlight the potential of graph-based learning to outperform traditional forecasting methods, providing valuable information on temporal relationships in financial forecasting tasks.