Variance Stabilizing Transformations for Electricity Price Forecasting in Periods of Increased Volatility

Abstract

Accurate day-ahead electricity price forecasts are critical for power system operation and market participation, yet growing renewable penetration and recent crises have caused unprecedented volatility that challenges standard models. This paper revisits variance stabilizing transformations (VSTs) as a preprocessing tool by introducing a novel parametrization of the asinh transformation, systematically analyzing parameter sensitivity and calibration window size, and explicitly testing performance under volatile market regimes. Using data from Germany, Spain, and France over 2015-2024 with two model classes (NARX and LEAR), we show that VSTs substantially reduce forecast errors, with gains of up to 14.6% for LEAR and 8.7% for NARX relative to untransformed benchmarks. The new parametrized asinh consistently outperforms its standard form, while rolling averaging across transformations delivers the most robust improvements, reducing errors by up to 17.7%. Results demonstrate that VSTs are especially valuable in volatile regimes, making them a powerful tool for enhancing electricity price forecasting in today's power markets.

Figures (5)

• Figure 1: EPEX-DE (top), EPEX-FR (middle), OMIE-ES (bottom) hourly day-ahead prices for the period 1.1.2015-31.12.2024. The first vertical dashed line marks the end of the 1456-day calibration window for point forecasting models and the beginning of the 56-day period for selection/averaging methods. The second dashed line marks the beginning of the 2141-day out-of-sample test period.
• Figure 2: Visualization of the NARX network with five hidden neurons with hyperbolic tangent activation functions and one linear output neuron. Source: lip:uni:25
• Figure 3: Empirical distribution of MAE-optimal parameter values for each variance stabilizing transformation (VST). Results are shown for the German EPEX market using the longest calibration window of 1,456 days, with separate distributions for the NARX and LEAR models. Optimal parameter is selected separately for each of 51,384 hours in out-of-sample period (2,141 days $\times$ 24 hours). Analogous plots for the Spanish and French markets and for other calibration window sizes show similar patterns and are therefore omitted for brevity.
• Figure 4: Average MAE improvements relative to the baseline model without variance-stabilizing transformation for three electricity markets and two base models (NARX and LEAR).
• Figure 5: Improvements in forecast accuracy from variance stabilizing transformations relative to untransformed benchmarks, reported for the rolling averaging scheme ($\mathrm{AVG}_{\text{roll}}$). Results are shown for three subperiods (2019–2020, 2021–2022, 2023–2024), five calibration window sizes, and two model classes (NARX in blue, LEAR in orange).