Does inter-industry risk spillover network predict financial crisis? Evidence from a gated graph neural networks approach

This study proposes a novel binary-classification model for systemic risk warning, utilizing inter-industry tail-risk spillover networks as input. These networks are constructed using the Tail-Event driven network (TENET) model, which captures high-dimensional and non-linear characteristics of risk contagion. The model leverages the Gated Graph Neural Network (GGNN) framework to uncover the ambiguous specification of crisis prediction. Applied to 11 key U.S. industry indices, the empirical results demonstrate that: (i) the topology of the risk spillover network is strongly correlated with financial crises during critical periods; and (ii) the GGNN model based on the TENET network provides superior reliability in early warning compared to traditional machine learning and other graph-based models.