Carbon emissions futures price forecasting with random forest

Prognozowanie cen emisji dwutlenku węgla z wykorzystaniem lasów losowych
Piotr Pawlowski

    Streszczenie
    The accurate carbon price forecasts are necessary for energy and financial market participants. However, nonstationary
    and nonlinear nature of carbon prices time series, makes it relatively hard to capture rapid price fluctuations.
    Literature concerning carbon prices forecasting has extended visibly during last decade, focusing on ARIMA, GARCH
    or hybrid models combing characteristics of linear and non-linear predictive methods. Development of machine learning
    techniques and widely available computing power made it possible to test more power consuming algorithms such as XGboost
    or Random Forest. In this article Random Forest model was used to predict carbon emissions futures price for day ahead, with
    additional parameter tuning. The final results evaluated on testing dataset indicate that the proposed model performs better
    than classic linear model and parameter tuning can additionally enhance model accuracy. Overall, the developed approach
    provides an effective method for predicting carbon price.
    Dostęp do pełnej wersji artykułu po zalogowaniu się na stronie.