Precise forecasting of fluctuations in carbon allowance valuations is critical for shaping environmental policy and for bolstering the effectiveness of market-based regulatory mechanisms. Advanced statistical and machine-learning techniques afford regulators the capacity to fine-tune carbon taxation schemes, enhance the operational efficiency of emissions trading frameworks, and steer financial resources toward low-carbon development projects with greater assurance. This study examines the China Emissions Trading Scheme (CHNTS)-one of China's pioneering carbon markets established under the broader national decarbonization strategy-and presents an innovative predictive model based on Gaussian process regression (GPR) whose hyperparameters are optimized through a Bayesian framework. By dynamically adjusting to latent market behaviors and unobserved structural shifts, this method adapts more responsively to evolving trading patterns. Our empirical investigation utilizes daily settlement data for China Emission Allowances spanning July 16, 2021, through April 9, 2025-a timeframe marked by key regulatory amendments, market maturation phases, and changing participant conduct as the scheme integrated into the wider national carbon pricing system. Model validation is performed on an out-of-sample window from June 28, 2024, to April 9, 2025, yielding notable performance metrics: a relative root-mean-square error (RRMSE) of 1.0771%, root-mean-square error (RMSE) of 1.0212, mean absolute error (MAE) of 0.6773, and a correlation coefficient (CC) reaching 98.604%. To the best of our knowledge, this represents the first deployment of GPR in the context of China's carbon trading exchanges. Beyond enriching theoretical understanding of price discovery in emergent emissions markets, the proposed approach provides a flexible analytical template that could readily be applied to analogous cap-and-trade systems worldwide.
Jin et al. (Thu,) studied this question.