Accurate carbon price predictions are vital for supporting the effective functioning of the carbon market. Most existing studies rely on point-valued modeling, thus failing to fully explore interval-valued data and mixed-frequency information. To address this limitation, this paper proposes a new interval-valued carbon price forecasting paradigm and presents a mixed-frequency data-driven stacking ensemble forecasting system. The data preprocessing module in this system was designed to remove noise through signal decomposition and reconstruction. Additionally, the mixed-frequency modeling module integrates a mixed-frequency model, statistical model, and artificial intelligence model, which can fully utilize the significant potential of mixed-frequency information and overcome the limitations that result from selecting only one type of basic model. Moreover, a stacking ensemble learning module is proposed to fully exploit the advantages of the mixed-frequency modeling module, thereby providing more accurate forecasting results. Comparative experiments were performed and discussed based on the real carbon market, proving that the developed mixed-frequency data-driven stacking ensemble forecasting system outperforms other advanced methods and could provide an effective technique for improving carbon market management.
Hao et al. (Sat,) studied this question.