Combining Inverse Optimization and Modeling-to-Generate-Alternatives to Reveal Epistemic Uncertainty in Merit-Order Recovery: A Proof of Concept for the Iberian Electricity Market

Source abstract

Understanding how electricity prices are formed is important for both market analysis and forward-looking decision-making, especially when the underlying bids and offers cannot be directly observed. This paper develops a three-layer framework that combines market-data clustering, inverse optimization (IO), and modeling to generate alternatives (MGA), enabling the recovery of merit orders from observed market outcomes and the generation of plausible alternatives that capture parametric and structural uncertainties in the recovered market model. The framework addresses four methodological needs: (i) transforming heterogeneous market disclosures into model-ready inputs through two clustering algorithms suited to different data-availability settings; (ii) incorporating electricity-gas price coupling directly into the recovery process to better represent gas-fired bidding behavior; (iii) explicitly modeling multi-zone market clearing so that recovered merit orders capture zonal price formation and interzonal interactions; and (iv) extending the conventional single-solution IO setting with an MGA layer that generates alternative merit orders. As a proof of concept, the framework is demonstrated using 2024 data from the Iberian electricity market, covering Spain and Portugal. Results show that preserving generation-technology information during clustering improves the accuracy of zonal price reproduction by up to 26.97% relative to price-only clustering, while electricity-gas price coupling further improves the accuracy of zonal price reproduction by up to 5.15%. The MGA layer is used to generate alternative merit orders that reveal the uncertainty on the bids of gas-fired generators. These alternative merit orders are subsequently tested under out-of-sample high-gas-price conditions. Overall, the proposed framework extends merit-order recovery beyond static estimation into a more flexible tool for analyzing electricity price formation and exploring the epistemic uncertainty in forward-looking analyses.