Forecasting carbon price with CEEMDAN-SE-transformer model

The consensus regarding the impact of carbon dioxide emissions on the environment has been widely acknowledged, with market-based mechanisms highlighted in international frameworks like the Paris Agreement. China has actively advanced carbon market development by establishing transparent carbon trading mechanisms to address emission challenges. Accurately forecasting carbon emission rights prices is crucial for supporting enterprise and government decision-making and promoting a low-carbon economy. However, the non-stationary and nonlinear nature of carbon price time series makes precise forecasting difficult. To address this issue, this paper proposes a new method for forecasting the prices of carbon emission rights with the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) method, which provides a reconstruction method from the original signal and a better spectral separation of the Intrinsic Mode Functions (IMFs). Based on correlation analysis, this study identified Brent crude oil prices, EU carbon prices, natural gas prices, thermal coal prices, and the Frankfurt DAX index as key influencing factors. To demonstrate and validate the model, simulations were conducted utilizing empirical data. The Guangdong carbon trading market serves as the research object, with historical data from 2015 to 2022 used for model training. Several comparative models—CEEMDAN-Transformer, CEEMDAN-SE-BPNN, and CEEMDAN-SE-LSTM—were also developed using the same dataset. Empirical results indicate that the CEEMDAN-SE-Transformer combined model outperforms other models, exhibiting lower forecast error and higher accuracy of each evaluation index. The results of this study can serve as an effective reference for carbon market traders' decision-making and for related regulatory departments.