Carbon price forecasting based on secondary decomposition and feature screening

With the increasing development of China’s carbon market, the prediction of carbon prices has become a popular research topic. Reasonable carbon price forecasts are critical to ensure the smooth operation of the carbon market. This study presents a novel hybrid forecasting model based on secondary decomposition and three-stage feature screening to predict carbon prices in Hubei, Guangdong, and Shenzhen. Two algorithms, the improved complete ensemble empirical mode decomposition with adaptive noise and the discrete wavelet transform, constitute a secondary decomposition strategy for the decomposition of the carbon price time series. Support vector regression and multi-layer perceptron are used to predict subsequences with different complexities. In the prediction of the low-frequency component, not only the historical data but also the external influencing factors are considered, and a screening analysis is performed. Finally, the forecasting results of the proposed model are derived by integrating the predicted values of all the subsequences. Empirical studies illustrate that the model outperforms other benchmark models. The proposed model combines the advantages of the secondary decomposition strategy, considers and screens external factors, and uses a hybrid machine learning model to effectively improve the prediction accuracy of carbon prices, thus providing a new approach for carbon price prediction.