A novel framework for carbon price forecasting with uncertainties

Abstract

Carbon price prediction is a key issue in the field of carbon market research. However, the existing methods of carbon price forecasting mostly regard carbon price series as a certain time series, but pay less attention to the uncertainty implied by carbon price series. Based on this, this paper proposes a forecasting model framework considering the uncertainty of carbon price series, which represents the carbon price series as a time series of probability density function to deal with the uncertainty. A prediction model based on the probability density recurrence network of carbon price is constructed with the help of the recurrence network construction technology of data and link prediction, and the advantages of the model are verified by numerical simulation. The empirical analysis is made by using the EU carbon price data. In the whole sample interval, we build 123 time windows. The results indicate that our prediction model has better level forecasting precision in 90.69% time windows and directional prediction accuracy in 67.82% time windows than the prediction model built based on deterministic network, which can improve the level prediction accuracy and directional prediction accuracy by 25.76% and 5.09% on average. By comparing with the accuracy results of existing carbon price prediction models, we discuss the feasibility of improving the prediction effect by increasing the number of similar nodes.