Enhancing carbon price robust forecasting: A text-driven method utilizing weighted interval-joint quadratic support vector regression

Accurate forecasting of carbon prices can effectively promote the low-carbon transition and has received increasing attention in recent years. Interval carbon prices forecasting incorporating online texts has emerged as a promising approach to improve the prediction accuracy and timeliness. However, existing forecasting models struggle with extracting correlation features from interval prices. The significant data noise and outliers in input variables further complicates the accuracy and stability of prediction results. Therefore, in this study we propose a novel text-driven method for interval carbon price forecasting based on weighted interval-joint quadratic support vector regression (WIJQSVR). First, we incorporate news text reflecting investor sentiment into the prediction factors, providing a more comprehensive consideration of factors influencing carbon prices. Second, we design a weight algorithm to evaluate the relative importance of each training point, thereby reducing the interference of outliers. Furthermore, a kernel-free WIJQSVR model is proposed by designing an inverse cumulative distribution function following the triangular distribution for representing interval carbon price. The model effectively handles the noisy data and learns the underlying interval interrelation of input values by generating the interval-joint quadratic hypersurface. To verify the performance of the proposed method, extensive computational experiments on Guangdong and Hubei interval carbon price forecasting are conducted. Results clearly support that the proposed method is superior to other benchmark methods in both forecasting accuracy and robustness, which will be an effective tool for carbon price forecasting.