What Does It Take to Generalize in Electricity Price Forecasting?
Date:
Electricity price forecasting is rapidly evolving, yet strong performance in one market or context does not automatically transfer to others. In this talk, I will present insights into what aspects of price models actually generalize, covering both intraday dynamics and continental-scale dependencies.
I will highlight how rich feature representations and market liquidity influence transferability across markets. Building on recent work with foundation models for probabilistic electricity forecasting, we will also examine how explicit modeling of spatial interdependencies across multiple regions (such as European markets) can improve probabilistic forecasting performance.
The talk will argue that generalization must be designed, not assumed — guided by market structure, data characteristics, spatial scale, and uncertainty quantification.