Assessing the Accuracy of Electricity Price Forecasting Models, Before and After, the Impact of Energy Crisis Using Univariate and Multivariate Methods

Abstract

Forecasting wholesale electricity prices (EPs) is a highly challenging process, especially in an unstable environment. The electricity market is sensitive to crisis events, which can cause significant fluctuations in EPs. Meanwhile, the energy transition and the increasing interconnectedness of the EU's electricity markets add another layer of complexity and further complicate the modeling. This article aims to compare and evaluate several EP forecasting models and methods based on different time horizons, which have unique characteristics. The difference between the periods, reflects the impact of the energy crisis. Therefore, pre-(June 2019 – May 2021) and energy crisis (June 2021 – May 2023) periods were estimated based on best-fit univariate (exponential smoothing and ARIMA) and multivariate (ARIMAX and multiple linear regression) models, built on out-of-sample datasets and the results were assessed primarily with evaluation metrics, such as MAE, MAPE and RMSE. Our empirical results reveal that multivariate methods performed better in estimating monthly average EPs in the EU during pre-and energy crises periods, although the exact models varied between the datasets. Furthermore, regardless of the models utilized, the estimation for the pre-energy crisis period generally resulted in lower error values. Overall, we concluded that different conditions lead to diverse models being more effective.The unprecedented surge in Eps, during 2021-2023, underscores the importance of re-evaluating model performances over time and under different market conditions.