An event sentiment and herding-based stock price forecasting using graph neural networks

Stock price forecasting is considered difficult due to the volatility of financial markets. A reliable stock price forecasting model is critical for making sound investing decisions while managing financial risks and portfolios. In recent years, Graph Neural Networks (GNNs) have emerged as an effective approach for forecasting stock prices by capturing correlations among stocks. The primary research gaps are that they ignore the herding behavior of investors and event sentiments when designing a stock price forecasting model. As a result, in this study, an ESH-GNN (Event Sentiment and Herding GNN) has been presented. The proposed framework models the prediction of stock prices as a time series with multiple variables, incorporating a herding index that includes Cross-Sectional Absolute Deviation (CSAD) along with event sentiment as supplementary factors to establish connections between stocks. The graph edges are formed based on herding patterns and will lead to a herding-based graph model. Moreover, attention layers are also incorporated before the stacked LSTM and BiLSTM layers to make the model focus on important prior time steps. The stocks of the S&P500 index, five political events, five natural disasters, and four terrorist attack event sentiments were all used in the experiments. The ESH-GNN model’s findings demonstrate a MAPE error of 1.315%.