Multifactor forecasting of carbon trading prices using machine learning

Abstract

The prediction of the prices in carbon emissions trading is a complicated issue. Carbon prices are affected by various factors such as macroeconomics, market emotions, and climate policy. Non-linearity, non-stationarity, and multi-frequencies are among the features of the carbon prices. Now, there is little carbon emissions trading price history data available, and a large number of affecting factors are present; hence, a prediction model should be established by considering all factors fully. First, we made the analysis of carbon emission trading price's affecting factors. Four factors are considered in our research. They are the stock market factor, international exchange rate factor, international energy commodity price factor, and international carbon emissions trading price factor. Through the use of LASSO algorithm, we have concluded that all the above factors have important impacts on carbon prices. Then, we used four methods to predict the prices of carbon emissions trading. They are random forest (RF), xgboost, back propagation (BP) neural network, and support vector regression (SVR). It turns out that RF model performs better than the other three models in predicting the carbon emissions prices. This methodology will be helpful for policymakers, market regulators, and investors who are involved in the carbon emissions trading mechanism.