Abstract
Amid global climate governance transformations, accurate carbon price prediction and causal analysis have become critical. Existing studies predominantly examine linear impacts of single factors (e.g., policy or energy prices), neglecting complex nonlinear interactions among financial markets, climate variables, market heterogeneity, and behavioral responses. This study addresses these gaps by developing an integrated analytical framework incorporating four key drivers (policy, market, climate, behavioral indicators) for China's Hubei carbon market. We propose a novel three-stage methodology involving feature selection, predictive model optimization, and causal identification. In the first stage, 66 variables from eight categories (macroeconomic indicators, financial markets, climate data, etc.) are narrowed to nine predictors via hybrid selection. In the second stage, we compare six machine learning models and find that Random Forest (RF) (RMSE = 2.06) and XGBoost (RMSE = 2.31) significantly outperform traditional time-series models. Finally, we use a double machine learning (DML) approach to partial out high dimensional confounding and to obtain orthogonalized estimates. Under the identification assumptions of the DML framework and supported by statistical evidence, the estimates indicate that key financial indices and cryptocurrency related variables have statistically significant causal effects on the carbon price in the Hubei market. These empirical findings provide valuable guidance for regulators in developing cross-market risk warning systems and improving the efficiency of carbon allowance allocation.
Jiang, H.; Cao, R.; Chen, X.; Song, M.; Du, J. 2026. Predictability and causal identification of carbon prices using interpretable variables: Evidence from China's Hubei carbon market. Energy economics : Elsevier. ISSN 0140-9883. eISSN 1873-6181. 158, 109330, p. 1–19. DOI: 10.1016/j.eneco.2026.109330. [Scopus; Social Sciences Citation Index (Web of Science)].