Abstract:

Accurate prediction of carbon emissions trading prices is the basis for stabilizing the sustainable development of the carbon financial market. When analyzing carbon emission price forecasts， existing studies often use a single modeling approach to construct forecasting models， ignoring the characteristics of changes in carbon emission prices in different dimensions. Based on the coherent and year-on-year evolution characteristics of time series， these two traditional “unit processing” modeling ideas are defined as vertical processing and horizontal processing for the first time. At the same time， these two “unit processing” modes are combined into a “dual processing” mode through the idea of integration. Then an ensemble learning model is built based on the using of Empirical Mode Decomposition （EMD）， eXtreme Gradient Boosting （XGBoost）， Extreme learning machine （ELM）， Fractional Accumulation Order Seasonal Grey Model （FSGM）， Random Forest （RF） and their combinations Model. Then， 6 domestic and foreign carbon trading markets are selected as research objects， and Shenzhen Carbon Exchange is taken as an example to conduct special analysis. To verify the effectiveness of the proposed model， the prediction results of this model are further compared with Holt Winters （HW）， Support Vector Machine （SVM）， Long Short-Term Memory Network （LSTM）， FSGM， RF and EDM-XGB-ELM. The research results show that the prediction model proposed in this paper has better prediction performance than the benchmark model， and this modeling paradigm based on dual processing level also has good application prospects in other fields.

Key words: carbon finance, EMD-XGBoost-ELM, FSGM, RF, dual processing