Evaluating Intervention Effectiveness during Extreme Price Events: A Counterfactual Analysis of Romania’s 2024 Energy Crisis

Extreme electricity price events pose significant challenges for energy policymakers, yet the question of which interventions most effectively mitigate such crises remains under-explored. This paper develops a counterfactual analysis framework to evaluate intervention effectiveness during Romania′s June–July 2024 electricity price crisis, when day-ahead prices reached 1,022 euros per megawatt-hour. Using a dataset of 59,160 hourly observations from the European Network of Transmission System Operators for Electricity Transparency Plat-form, we trained a Random Forest model on 70 engineered features and generated 147 counter-factual scenarios across three intervention categories: renewable generation increase, demand reduction, and combined approaches. To address the fundamental limitation that tree-based models underpredict extreme events by 68 percent, we validated all findings through crossmodel comparison with Ridge regression, achieving a Spearman rank correlation of 0.73 between intervention rankings across both architectures. Results reveal that demand response of 10 percent load reduction is the robust first-tier intervention, achieving 70 percent price reduction with unanimous crossmodel consensus and full operational feasibility. A critical finding is that the peak event occurred during nighttime hours when solar generation stood at 0.2 percent of capacity, rendering solar-based interventions physically impossible and identifying battery energy storage as critical enabling infrastructure. No feasible intervention achieves normal price levels, reflecting the genuine severity of the crisis. All scenarios were assessed against operational constraints grounded in eight academic references covering forecasting accuracy limits, demand response capacity, and activation warning times. The paper contributes a methodological framework for counterfactual analysis under model uncertainty, demonstrating that meaningful policy guidance can be derived from relative effectiveness rankings even when absolute predictions are unreliable.