A combined model of Hawkes-ICEEMDAN-Catboost-Holt for interval-valued carbon price forecasting

Abstract

Accurate carbon price forecasting is crucial for efficiently operating both carbon trading and financial markets. This study proposes a Hawkes-ICEEMDAN-Catboost-Holt combined interval forecasting model to improve the accuracy of carbon price forecasts. First, the t-SNE-LLE second-order dimensionality reduction was performed on the collected search index data to quantify the attention of netizens. Subsequently, the attention series was converted into an interval series using the Hawkes exponential decay model, and further decomposed into radius and center series. In addition, the radius and center series were decomposed using ICEEMDAN and then reconstructed. Finally, the carbon price interval combined forecast was based on the Catboost–Holt model. The proposed model exhibits high accuracy and strong stability in comparison to other forecast models. The Hawkes exponential decay of netizens’ attention is creatively included in the influence mechanism of carbon price fluctuations, effectively improving forecast accuracy. The interval decomposition and reconstruction methods provide an innovative development path for interval forecasting research. The residual and trend series obtained from interval reconstruction can be forecasted using Catboost and Holt, respectively, thus significantly improving the accuracy and stability of interval forecasts and enhancing the decision-making process for carbon trading and financial markets.