Correlation Matters: A Stock Price Predication Model Based on the Graph Convolutional Network

Chengkun Xin, Qian Han, Gang Pan

Abstract: In the financial markets, accurate prediction of stocks is crucial for formulating investment strategies. Previous research predominantly relied on a stock's historical information for prediction, but overlooked the cross-effects between stocks. However, stocks are closely connected rather than independent of each other. This work introduces a deep learning framework named StockGCN for stock prediction, which can be easily extended by integrating other modules. By constructing a stock graph structure, the model transforms the prediction of individual stocks into the prediction of the entire graph. Experiments show that StockGCN effectively captures comprehensive spatio-temporal correlations through modeling multi-scale stock networks and consistently outperforms state-of-the-art baselines on real-world stock datasets.