Securities Price Movement Prediction Based on Graph Neural Networks

Abstract

Securities play a crucial role in the modern economy. However, securities price movement prediction, a type of time-series forecasting problem, remains challenging. This paper suggests that the underutilization of temporal structural information in the data hinders the improvement of accuracy in securities price movement prediction. Therefore, this paper proposes a securities price movement prediction method based on graph computing. By abstracting trading days and the temporal relationships between them as nodes and edges, the method transforms historical trading data into graph data. Subsequently, graph neural networks (GNN) are used to process the graph data and make predictions about stock price movements. Experiments show that the proposed method effectively improves the performance of price movement prediction. Thus, the proposed method is a simple, effective way to utilize time-series data and holds substantial value in securities price movement prediction.