Renewable Penetration and Residual Load Predictability in the Romanian Power System

This thesis investigates how increasing variable renewable energy (VRE) penetration reshapes the distribution of residual-load forecast errors and whether the resulting uncertainty is reflected in day-ahead electricity prices. Using hourly data from the Romanian power system (2019–2025) and a fixed Prophet forecasting baseline, the empirical strategy employs threshold regression, quantile regression, and extreme value methods to map the penetration–uncertainty relationship. The results reveal three primary mechanisms. First, the relationship between renewable penetration and forecast uncertainty is fundamentally convex; an endogenous threshold regression identifies a structural break at approximately 15.0% penetration, beyond which the marginal effect on error magnitude accelerates. Second, the distribution of errors exhibits a regime-dependent directional shift, moving from a shortage bias at low penetration to a severe surplus bias (−1,053 MW) at high penetration, indicating that the dominant deviation shifts from under-forecasting to systematic over-forecasting of residual load. Third, this convex amplification is heavily concentrated in the tails of the distribution. The Generalised Pareto Distribution (GPD) scale parameter more than doubles between low and high regimes, substantially increasing the probability of extreme deviations. A complementary pricing analysis examines whether day-ahead prices incorporate this heightened uncertainty. While prices exhibit a diurnally varying, negative association with conditional volatility—driven by the merit-order effect—the implied uncertainty premium remains negligible. These findings suggest that day-ahead prices primarily reflect the merit-order suppression associated with VRE output but do not embed compensation for the convex forecast-error escalation documented in the distributional analysis. The decoupling of forecast-error severity from day-ahead price signals is consistent with the missing-money problem discussed in the electricity market design literature and motivates further investigation into whether balancing-market data or alternative market design interventions may be needed to properly reflect ex-ante forecast uncertainty in market outcomes.