Unveiling high-dimensional time-varying extreme risk spillovers: AI-driven warning signals in the global energy market

This paper investigates extreme risk spillovers in global energy markets using the enhanced high-dimensional time-varying parameter vector autoregressive spillover (HD-TVP-VAR-SP) model. We employ the Long Short Term Memory (LSTM) model to develop an energy risk warning system, identifying key factors in risk contagion. Our findings reveal robust connectivity in global energy market risks, characterized by high-dimensional complex networks with marked temporal variations. The Americas region emerges as the leading contributor to systemic risk shocks, primarily through positive spillovers in its energy markets. The LSTM model demonstrates superior extreme risk prediction compared to other machine learning models like Gradient Boosting Machines, Random Forest, and Decision Trees. The oil market is identified as a critical driver of risk contagion in the energy sector. These insights provide valuable guidance for effectively identifying and managing global energy market risks and enhancing risk warning systems.