Price and Volatility Forecasting in Electricity with Support Vector Regression and Random Forest

Liberalized electricity market players all over the world face a significant challenge due to the volatile and uncertain nature of these markets. Therefore, price and volatility forecasting in those markets are as remarkably of interest as other commodity markets. There exists a recent and increasing tendency in the literature to apply machine-learning methodologies to those markets’ data and various methods have been proven effective to produce highly accurate forecasts. The Turkish electricity market is one of the recently liberalized and emerging markets. In this research, we aim to carry out price and volatility forecasting for the Turkish day-ahead electricity market with Support Vector Regression (SVR) and Random Forest (RF) to observe the effectiveness of the methods. A rolling forecasting scheme is proposed and experimented with using hourly prices between 2013 and 2019. The performance metrics of the SVR model are compared with those of naive and RF estimations. Furthermore, the sensitivity of the proposed model to feature reduction is also discussed. Overall, the results reveal SVR as an effective tool for electricity price forecasting in the Turkish electricity market, whereas RF modeling is found to be slightly better in volatility forecasting.