Short-term electricity trading on intraday markets is crucial for integrating variable renewable energy in the power system. For instance, it allows energy suppliers to adjust their market positions based on updated variable renewable energy and consumption forecasts, reducing their potential imbalances. In the case of Germany, the continuous intraday market allows trading from the day before delivery until several minutes before delivery. However, the complex market design and high price volatility make developing price forecasting models challenging. This paper lays a foundation for price forecasting by comparing baseline models used to benchmark rolling continuous intraday price forecasts. These baselines help develop price forecasting models as they serve as a reference for these models. We also adapt a price normalization approach from the literature to benchmark price forecasts in a volatile market environment. Our baselines include the generalization of two baselines used in literature and one new baseline. We benchmark our baselines throughout 2021 and 2022. Among other baselines, we find that the price average of the last four trades yields the lowest root mean squared error. Moreover, the analysis suggests that baseline errors are independent of the market price development through normalization.
Keywords machine learning, electricity price forecasting, continuous intraday market