GCNET: Graph-based prediction of stock price movement using graph convolutional network

The importance of considering related stocks data for the prediction stock price movement has been shown in many studies; however, advanced graphical techniques modeling, embedding and analyzing behavior inter-related have not widely exploited movements yet. main challenges this domain are to find a way modeling existing relations among an arbitrary set exploit such model improving performance those stocks. most methods rely on basic graph-analysis techniques, with limited power, suffer from lack generality flexibility. In paper, we introduce novel framework, called GCNET that models as graph structure influence network uses history-based infer plausible initial labels subset nodes graph. Finally, Graph Convolutional Network algorithm analyze partially labeled predicts next direction each is general framework can be applied fluctuations interacting based their historical data. Our experiments evaluations NASDAQ index demonstrate improves state-of-the-art algorithms terms Accuracy Matthew’s Correlation Coefficient by at least 1.5% 2%, respectively.