Forecasting Carbon Emission Allowance Prices With Deep Learning Models

Abstract

As the effects of the climate crisis deepen, the measures taken against it are increasing. These precautionary mechanisms significantly affect companies. One of these mechanisms is carbon trading systems, which limit companies’ carbon emissions. These systems create additional costs for companies, and predicting these potential costs is crucial today. For this reason, the importance of correctly estimating carbon permit prices has increased. To estimate carbon permit prices, the correct variables must first be determined. Many factors affect the carbon markets due to their complex nature. Using the correct estimation method is another important issue. Traditional econometric models cannot clearly explain the complex structure of the carbon market, resulting in low accuracy. In this direction, artificial intelligence-based estimation models have been developed to estimate carbon permit prices. This study estimated the prices of European Union Emissions Allowances (EUA) using energy and power market prices and the economic outlook. The estimation model utilized deep learning algorithms, which are feedforward neural networks and recurrent neural networks (Elman). As a result of the study, both models were successful, and strong accuracy was obtained. While the FFNN model provided more stable and low-variance results, the Elman model provided advantages in short-term predictions.