Prediction of Financial Crisis Events Using Graph Neural Network Model : Based on Inter-Industry Spillover Information

This paper contributes to the financial crisis literature by developing a deep learning model for predicting financial crisis events. We propose a novel gated graph neural network (GGNN) early warning model based on information spillover networks. The spillover network allows us to clarify the dynamic characteristics of risk spillover and identify the transmission path of the crisis within the whole industry system. Then we propose a two-classification crisis early warning model based on GGNN, which can automatically extract features from inter-industry risk spillover networks. In the empirical application, we construct two types of information (risk and volatility) spillover networks of 28 Chinese industries and compare the prediction performance of the GGNN and other machine learning models. We discover that inter-industry linkage is an important risk factor to whole economy. The GGNN model displays a good out-of-sample performance in predicting financial crises, especially model based on risk spillover networks. Our results highlight the importance of considering inter-industry linkage in forecasting framework