Unveiling the drivers of high-frequency carbon price dynamics: A nonlinear fusion approach with irregular events and mixed-frequency data

Carbon price—as a critical component in the functioning of carbon market mechanisms—plays an indispensable role in policy formulation, market development, and societal progress. Thus, accurately predicting carbon prices is of paramount importance. This study aims to comprehensively investigate the impact of unconventional events (e.g., political conflicts and extreme weather) and mixed-frequency data (e.g., daily high-frequency financial information and monthly low-frequency macroeconomic data) on carbon price forecasting; to this end, it introduces the novel Prophet-Backpropagation Neural Network-Reverse (Unrestricted) Mixed Data Sampling model, which innovatively integrates the following three key advantages: the quantification of irregular events using Prophet, nonlinear pattern recognition through a back-propagation neural network, and frequency alignment via reverse mixed data sampling. Applied to daily carbon price prediction in the Hubei carbon market, this model is statistically validated to significantly outperform other models, as demonstrated by the Diebold-Mariano test. This study's results underscore the model's superior predictive capability and elucidate the key drivers of carbon prices and their nonlinear impact mechanisms.