Carbon emissions allowances trade amount dynamic prediction based on machine learning

Abstract

With the increasing threatening of the greenhouse effect, taking some measures to reduce carbon emissions is extremely essential. Carbon pricing policy is one of the most powerful tools for reducing emissions. However, most researches focused on carbon emissions allowances trade price prediction, only few investigations paid attention to carbon emissions allowances trade volume prediction. In order to better understand carbon emissions of carbon emissions, this study performed long-term prediction and short-term prediction models to predict carbon emissions trading volume. Linear regression, Decision tree, Random forest, Extreme gradient boosting and Support vector machine were chosen to perform long-term prediction. In these five long-term prediction models, Random Forest performs best with the smallest mean absolute error (24.42). What’s more, a time series forward multi-step hybrid intelligent prediction model (consists of reinforcement learning model, hidden Markov model and neural network) was used to perform short-term prediction, which improves the accuracy and effect of the prediction. The mean absolute error of short-term prediction is 6.79. Our study shows there is a nonlinear relationship between the lag of volume/frequency of transactions and price in the future. The machine learning-based Carbon emissions allowances trade prediction enables traders and environmental organizations to observe the overall carbon emissions trading volume trends and changes as well as judge the short-term trading situation and make judgments.