The Impact of Variable Renewable Energy Sources Forecast Errors on Electricity Price Spreads and Asymmetric Volatility Dynamics
Author
Basse, Marcus
Term
4. term
Education
Publication year
2026
Submitted on
2026-05-26
Pages
66
Abstract
This thesis examines how errors in day-ahead forecasts of Variable Renewable Energy Sources (VRES) production affect both the average level of the electricity price spread (the difference between two related prices) and its time-varying volatility. Using hourly data, standard tests indicate the series are stable in level (stationary, I(0)). A simple regression and Wald tests reject perfect market efficiency (α=0, β=1), pointing to systematic risk premiums in the spread. VRES forecast errors have a clear linear effect on the spread level, but the static model explains little of the variation (R2=0.1445) and leaves residuals with heavy tails and clustered volatility. To capture these risk dynamics, the study fits GARCH, EGARCH, and TGARCH volatility models under a skewed Student’s t distribution. Results show trade-offs: a standard GARCH model is unstable (α1+β1>1), and TGARCH understates risk out of sample. By contrast, an EGARCH(1,1) model remains covariance-stationary, captures asymmetric leverage effects, and performs robustly out of sample, achieving 96.4% empirical coverage against a 95% target. A seasonal analysis reveals a physical–financial paradox: winter shows the largest absolute forecast errors but the lowest spread volatility, consistent with thermal baseload buffering. Guided by parsimony to avoid overfitting, the study rejects higher-order exogenous specifications. Overall, a baseline EGARCH(1,1) appears practical for risk management and for procuring ancillary services.
Specialet undersøger, hvordan fejl i dag‑forud prognoser for produktion fra Variable Vedvarende Energikilder (VRES) påvirker både niveauet for pris-spreadet (forskellen mellem to relaterede priser) på elmarkedet og dets skiftende volatilitet. Med timedata viser standardtests, at serierne er stabile i niveau (stationære, I(0)). En enkel regressionsanalyse og Wald-test afviser perfekt markedseffektivitet (α=0, β=1), hvilket peger på systematiske risikopræmier i spreadet. VRES-prognosefejl har en tydelig lineær effekt på spread-niveauet, men den statiske model forklarer kun lidt af variationen (R2=0,1445) og efterlader residualer med tunge haler og klyngevis volatilitet. For at indfange disse risikodynamikker estimeres GARCH-, EGARCH- og TGARCH-volatilitetsmodeller under en skæv Student’s t-fordeling. Resultaterne viser afvejninger: en standard GARCH-model er ustabil (α1+β1>1), og TGARCH undervurderer risiko ude af stikprøven. Derimod er EGARCH(1,1) kovarians-stationær, fanger asymmetriske leverage-effekter og er robust ude af stikprøven med 96,4% empirisk dækning mod et mål på 95%. En sæsonanalyse finder et fysisk-finansielt paradoks: vinteren har de største absolutte prognosefejl, men den laveste spread-volatilitet, i tråd med buffering fra termisk grundlast. Styret af princippet om enkelhed for at undgå overtilpasning afvises højere-ordens eksogene specifikationer. Samlet set synes en basal EGARCH(1,1) at være praktisk til risikostyring og indkøb af systemydelser.
[This apstract has been rewritten with the help of AI based on the project's original abstract]