Predicting Negative Day-Ahead Electricity Prices Across 12 European Bidding Zones: A Gate-Closure-Audited Explainable Machine Learning Framework

Abstract

The growing penetration of variable renewable energy (VRE) is increasing the frequency of very low and negative prices, although these events also depend on demand, transmission capacity, price-regime persistence and flexibility resources. This study examines which pre-auction and diagnostic variables are associated with negative day-ahead prices across European bidding zones, and whether these relationships remain stable over time and transferable across markets. More than 736,000 observations from 12 bidding zones in 2019–2025 were analysed, with sample coverage varying by data completeness. A gate-closure-audited Extreme Gradient Boosting (XGBoost) model achieved moderate risk-ranking performance and positive probabilistic skill in 2024–2025. The strongest pre-auction signals were completed-auction price history, calendar features and structural load relationships. Leave-one-market-out validation showed partial and heterogeneous transferability, supporting local calibration. A separate diagnostic layer revealed market-specific, non-linear associations involving VRE forecasts, residual load and cross-border exchange. Negative prices are therefore interpreted as screening signals for market configurations potentially associated with limited surplus absorption, rather than direct evidence of a flexibility shortfall. The framework separates operational pre-auction prediction from later market diagnosis and provides a reproducible basis for local early-warning applications. All SHAP, ALE and scenario results are interpreted as predictive associations and model diagnostics, not as causal effects or direct measures of physical flexibility.