What Does It Take to Generalize in Electricity Price Forecasting?

Jochen Cremer – TU Delft

Electricity price forecasting has made rapid progress, yet strong performance in one market or setting often fails to transfer to another. So what actually generalizes? In this talk, we explore this question across two complementary lenses. First, at the intraday level, we show how rich orderbook features reveal an asymmetric generalization effect: models trained in more liquid markets transfer well to less liquid ones, but not vice versa. Second, at the continental scale, we introduce a foundation-model perspective that explicitly models spatial interdependencies across European power markets. Together, these results suggest that generalization must be designed, informed by market structure, scale, and uncertainty.