Industry credit spread forecasting based on time series data and industry relationships

Credit spread is an important index in the financial market, as it captures the risk for the bond issuer and industry. As there are several types of connections among different industries, through which the risk may propagate from one industry to another, it may be helpful to take these connections into account for bond spread forecasting. In this paper, we propose a novel industry bond spread forecasting method based on LSTM and the temporal graph neural networks, which exploits the relations among different industries. The experimental results demonstrate that introducing industry relations can improve the prediction accuracy of the prediction.