Integrating data-driven and fundamental models: multivariate black-box and probabilistic estimation of the merit order for electricity price forecasting

Paul Ghelasi, University of Duisburg-Essen

Co-authors: Florian Ziel, University of Duisburg-Essen

Abstract: Power prices can be forecasted using data-driven or fundamental models. Data-driven models learn from historical patterns, while fundamental models simulate electricity markets. Traditionally, fundamental models have been too computationally demanding for intrinsic parameter estimation or frequent updates, which are essential for short-term forecasting. We propose a novel data-driven fundamental model that combines both approaches. By estimating the parameters of a fully fundamental merit order model from historical data, similar to data-driven models, we eliminate the need for fixed technical parameters or expert assumptions. This allows direct calibration to observations. The model is efficient enough for rapid parameter estimation and forecast generation. Applied to German day-ahead electricity prices, it outperforms both classical fundamental and purely data-driven models. It captures price volatility and sequential price clusters, which are increasingly important with growing renewables. Additionally, the framework enables probabilistic forecasting by modeling entire distributions of input data. By fitting probability distributions to all input variables and sampling from them, the model can generate a range of possible price outcomes, effectively capturing forecast uncertainty.