Mean-reverting for a dream

: using stochastic and GARCH-based forecasting to detect and exploit electricity market inefficiencies in the German spot market

Abstract

This thesis develops and evaluates a hybrid forecasting model combining an Ornstein- Uhlenbeck (OU) mean-reverting process with an Exponential Generalized Autoregressive Conditional Heteroskedasticity (EGARCH) volatility specification to predict day-ahead electricity spot prices in the German market from 2015 to 2025. Addressing the challenges of non-stationarity, limited data transparency, and the need for economically relevant metrics, the model captures key market characteristics - mean reversion, seasonality, volatility clustering, and asymmetric shock responses - using only historical price data. Out-of-sample forecasts over 20-, 30-, 60-, and 90-day horizons are benchmarked against market futures and a naïve historical average, employing statistical metrics (MAE, RMSE, MAPE, directional accuracy) and economic measures (Sharpe ratio, trading returns). Results demonstrate that the model, particularly under a Generalized Error Distribution (GED), significantly outperforms futures benchmarks at short-to-medium horizons (20-30 days), generating positive trading returns (e.g., EUR 2,371.56 at 30 days) with high Sharpe ratios (e.g., 1.15) and win rates (up to 90%). Monte Carlo simulations and bootstrap confidence intervals confirm robustness, though performance weakens in low-volatility summer regimes. The thesis contributes to the literature by integrating mean-reverting dynamics with asymmetric volatility, testing against real futures prices, and demonstrating economic value through a forecast-based trading strategy, offering a replicable framework for traders and risk managers in volatile, non-storable commodity markets.

Date of Award	17 Oct 2025
Original language	English
Awarding Institution	Universidade Católica Portuguesa
Supervisor	Alberto Manconi (Supervisor)