Carbon futures return forecasting: A novel method based on decomposition-ensemble strategy and Markov process

Abstract

Accurate forecasting of carbon futures return is essential for global climate policy and financial markets, yet it remains a significant challenge due to the complex nature of these returns. This paper proposes a novel hybrid model for carbon futures return forecasting based on decomposition-ensemble strategy and Markov process. This model can effectively capture nonlinear, nonstationary and volatile features of carbon prices and provide accurate point and interval forecasts. We use the Moth–Flame Optimization (MFO) algorithm and the Average Maximum Envelope Entropy (AMEE) function to optimize the parameters of Variational Mode Decomposition (VMD), which is a data decomposition technique that can separate the regular components and the random temporal component (noise) in the time series. We use Support Vector Regression (SVR) models to forecast the regular subsequences obtained by VMD, and Markov processes to quantify and forecast the random temporal component. We combine these forecasts to obtain the final prediction results, which incorporate both the regular and random information in the data. The proposed model supports both point and interval predictions, which are derived by summing the results of all the sub-models. We conduct extensive experiments to validate our approach and compare it with other models. The results show that our model outperforms other models in terms of prediction accuracy, reliability, and economic value. Our model can provide significant economic value for carbon futures investors and help them make favorable decisions in investment.