Extended decomposition ensemble framework based on full data analysis and optimized combination with relaxed boundary for carbon price forecasting

The carbon price forecasting is a challenging and meaningful task which can help investors, manufacturers, and policymakers to make decisions. The decomposition ensemble framework is widely employed in the short-term carbon price forecasting, but there is still room to improvement. The existing literature about it ignores thorough data analysis, the inherent limitations of single prediction model, and over simple ensemble operation. In this study, a comprehensive statistical and chaos analysis is conducted to estimate the characteristics of raw data and preprocessed data such as linearity, stationarity, and chaotic significance. In addition, our proposed model incorporates an optimized combination with relaxed boundary for fostering distinct models’ strengths and circumventing their weaknesses. For further improving the precision of forecasting model, a magnitude correction strategy is introduced in the ensemble operation for flexible conditioning weights and achieving better prediction performance. Finally, five evaluation indicators were used to measure the prediction performance of the proposed model and six comparative models with the application to three cases, systematically proving the superiority of our proposed model. From the empirical results, our proposed model is at least 5% lower in mean average percentage error compared to the one that performs best in benchmark models.