Accurate prediction of carbon allowance price movements is essential for informing environmental policy and strengthening market-based regulatory instruments. Sophisticated statistical and machine learning methods enable policymakers to calibrate carbon tax structures, improve the operational performance of emissions trading systems, and direct investment toward low-carbon initiatives with increased confidence. The present research analyses Chongqing’s emissions trading scheme (CQTS) in China — a pioneering instance among national carbon markets launched under the overarching decarbonisation agenda — and introduces a novel forecasting framework utilising Gaussian process regression (GPR) whose hyperparameters are determined through Bayesian inference. Through dynamic adaptation to latent market dynamics and unobserved structural changes, this approach responds more effectively to shifting trading patterns. The empirical analysis employs daily settlement data for Chongqing emission allowances from June 9, 2015 to March 23, 2021 — a period characterised by significant regulatory revisions, market maturational stages, and evolving participant behaviour as the scheme became incorporated into China’s national carbon pricing mechanism. Model assessment is conducted using an out-of-sample interval from January 22, 2020 through March 23, 2021, producing key performance indicators: a relative root-mean-square error (RRMSE) of 8.1950%, a root-mean-square error (RMSE) of 1.9930, a mean absolute error (MAE) of 1.5904, and a correlation coefficient (CC) of 97.146%. As far as we are aware, this constitutes the inaugural application of GPR within the context of Chinese carbon trading platforms. Beyond contributing to the theoretical discourse on price discovery in emerging emissions markets, the methodology offers a versatile analytical blueprint that may be readily extended to similar cap-andtrade frameworks globally.
Jin et al. (2025) studied this question.