Machine Learning Forecasts of Tail‐Risk Spillovers in Carbon and Energy Markets

ABSTRACT This study examines extreme downside risk spillovers between the global carbon market and major energy markets and evaluates their predictive value. Using daily data from 2013 to 2024, we estimate tail‐risk dependence with the MVMQ‐CAViaR model and quantify the dynamic transmission of extreme shocks via pseudo‐quantile impulse responses. Our results document strong and asymmetric spillovers between carbon and major energy markets. A bidirectional forecasting framework using Quantile Regression Forests, Quantile Gradient Boosting, and Quantile Regression Neural Networks results in substantial out‐of‐sample gains, confirmed by Diebold–Mariano tests. These findings suggest that regulators and market operators should integrate carbon–energy tail‐risk linkages into early‐warning systems and cross‐market surveillance frameworks, so materially enhance the detection of extreme risk events. The results also highlight the value of adopting machine learning‐based quantile models in policy settings where timely assessment of systemic risk is essential.