Spillover effect between oil prices, climate risks, and carbon prices in China: dynamic effects and key drivers

Against the backdrop of escalating global climate change and volatile international oil markets, the interdependence among energy prices, climate-risk information, and carbon markets has become increasingly prominent. Clarifying their dynamic spillover patterns is important for climate governance, energy security, and carbon-market risk management. This study examines the dynamic interactions among decomposed oil price shocks, physical and transition climate-risk indices, and China’s regional carbon price represented by the Guangdong pilot market. The research design is structured in three stages. First, oil price fluctuations are decomposed into supply, demand, and risk shocks following Ready (2018), and a time-varying parameter vector autoregressive (TVP-VAR) model is employed to capture evolving spillovers. Second, the Baruník–Křehlík (2018) frequency-domain spillover framework is used to distinguish short- and long-horizon connectedness. Third, the XGBoost–SHAP framework is used as a post-estimation predictive complement to identify nonlinear feature contributions to changes in spillover intensity; the resulting SHAP values are interpreted as feature-association evidence rather than structural causal estimates. The results show that: (1) oil price shocks, carbon prices, and climate risks exhibit substantial volatility spillovers, with short-term spillovers considerably stronger than long-term spillovers (baseline short-term TCI = 24.91 versus long-term TCI = 5.30, using a 1–10 trading-day short-term band); (2) physical and transition climate-risk information has strong bilateral spillover links with carbon prices, while oil-related risk shocks are more relevant in the short term and supply- and demand-related oil shocks display stronger long-horizon connections; and (3) SHAP-based variable-importance analysis indicates that exchange rate dynamics, China’s economic policy uncertainty, and market liquidity are important predictors of time-varying spillover intensity. Robustness checks using the Hubei pilot carbon market, together with additional frequency-domain evidence under the baseline 1–10 trading-day partition, confirm that the main conclusions remain qualitatively stable. The findings should be interpreted in light of the fact that regional pilots differ from the national ETS in liquidity, sector coverage, and allowance fungibility, and that the climate-risk indices are constructed from global media sources.