The Graph Convolutional Networks Framework for Predicting Pandemic Impact on Stock Prices

Covid-19 has dealt an unprecedented hit to the global economy and all industries, with varying degrees of decline from retail to real estate. This volatility is most evident in stock prices. Previous stock price forecasting methods typically used historical data for each stock as a separate input into the system. This paper proposes an attention-based parallel graph convolutional network framework, which consists of two parallel GCNs. The first GCN takes stock features as input, and the second GCN takes other industry features as input, and sets an attention model to reflect the pairwise interactions between networks. Experimental results on selected stock data show that the model outperforms both the LSTM model and the GCN model in accuracy and F1 score.