Resonance and dispersion mechanisms of tail risks in energy markets based on explainable machine learning: A time-varying network perspective

Energy markets exhibit complex tail risk spillover characteristics under extreme events, and traditional linear models often fail to capture nonlinear correlations and time-varying dynamics. This paper proposes the CAViaR-XGBoost-SHAP-TENET model, which integrates explainable machine learning with complex network theory, to analyze the transmission mechanisms of tail risk in energy markets from the dual perspectives of risk resonance and risk dispersion. The results show that external shocks significantly amplify risk resonance effects while greatly weakening risk dispersion capacity. Traditional fossil fuels such as crude oil, natural gas, and coal act as key amplifiers during most crises. The electricity market plays a dual role in absorbing and buffering risks, serving as a crucial stabilizer of the system. Clean energy and carbon markets demonstrate certain vulnerabilities amid the green transition. The model leverages XGBoost to capture nonlinear relationships across markets, applies SHAP values to disentangle the direction and magnitude of risk contributions, and employs the TENET framework to construct a time-varying risk matrix—for the first time distinguishing between resonance and dispersion network topologies. This study overcomes the limitations of traditional models, reveals the bidirectional nature of risk transmission, and offers a novel framework for dynamic early warning, targeted regulation, and optimized investment strategies, thereby contributing to global energy security and a sustainable energy transition.