Day-Ahead Electricity Price Forecasting Using Merit-Order Curves Time Series

We introduce a general, simple, and computationally efficient functional data analysis framework for forecasting day-ahead supply and demand merit-order curves, and the resulting electricity price.

We conduct a rigorous empirical comparison on data from the Italian (GME), German (EPEX-DE-LU), and French (EPEX-FR) day-ahead markets over the 2023-2024 period, analyzing curve forecasting performance, price forecasting performance, and the relationship between the two.

We find that strong curve forecasting performance does not necessarily translate into strong price forecasting performance, with important implications for curve model evaluation and selection when price forecasting is among the objectives.

We also show that this functional data representation approach consistently outperforms the original discretization-based approach of Ziel and Steinert (2016) on price forecasting across all three markets.

Finally, the proposed curve-based approach is competitive with state-of-the-art price-based models for two out of three markets (GME and EPEX-FR), and substantially improves accuracy during midday hours (when prices frequently drop due to high renewable generation) with MAE reductions of up to 27% in those windows.

For EPEX-DE-LU, however, price-based models retain a clear and significant advantage.