Benchmarking Pre-Trained Time Series Models for Electricity Price Forecasting

Paper: arXiv 2506.08113

Authors: Timothée Hornek Amir Sartipi, Igor Tchappi, Gilbert Fridgen

Abstract

Accurate electricity price forecasting (EPF) is crucial for effective decision-making in power trading on the spot market. While recent advances in generative artificial intelligence (GenAI) and pre-trained large language models (LLMs) have inspired the development of numerous time series foundation models (TSFMs) for time series forecasting, their effectiveness in EPF remains uncertain. To address this gap, we benchmark several state-of-the-art pretrained models–Chronos-Bolt, Chronos-T5, TimesFM, Moirai, Time-MoE, and TimeGPT–against established statistical and machine learning (ML) methods for EPF. Using 2024 day-ahead auction (DAA) electricity prices from Germany, France, the Netherlands, Austria, and Belgium, we generate daily forecasts with a one-day horizon. Chronos-Bolt and Time-MoE emerge as the strongest among the TSFMs, performing on par with traditional models. However, the biseasonal MSTL model, which captures daily and weekly seasonality, stands out for its consistent performance across countries and evaluation metrics, with no TSFM statistically outperforming it.

Complexity vs Empirical Score

• Math Complexity: 2.0/10
• Empirical Rigor: 7.0/10
• Quadrant: Street Traders — practical and empirical, lighter on theory

Why this score: The paper is focused on empirical benchmarking of pre-trained models, using real-world data, standard evaluation metrics, and statistical tests like Diebold-Mariano, but the mathematical formulation is minimal, relying more on established libraries and methodologies rather than novel theory.

Research Flowchart

  flowchart TD
  A["Research Goal: Benchmarking TS Foundation Models<br>vs. Traditional Methods for EPF"] --> B["Data Selection<br>2024 Day-Ahead Auction Prices<br>Germany, France, Netherlands, Austria, Belgium"]
  B --> C["Modeling Process<br>Forecast Daily Prices (1-day Horizon)"]
  C --> D{"Statistical/ML Models<br>e.g., MSTL (Bi-seasonal)"}
  C --> E{"Pre-trained TS Foundation Models<br>Chronos-Bolt, Time-MoE, etc."}
  D --> F["Evaluation & Comparison<br>Performance Metrics across Countries"]
  E --> F
  F --> G["Key Findings & Outcomes"]
  G --> H["TSFMs: Strongest (Chronos-Bolt, Time-MoE)<br>Comparable to Traditional Models"]
  G --> I["Overall Best: MSTL Model<br>Consistent High Performance<br>No TSFM Statistically Superior"]

Cited References (251)

1. TimeGPT -1 (2024) — link
2. Bridging Smart Meter Gaps : A Benchmark of Statistical , Machine Learning and Time Series Foundation Models for Data Imputation (2025) — link
3. A decoder-only foundation model for time-series forecasting (2024) — link
4. Chronos: Learning the Language of Time Series (2024) — link
5. Large Language Model -based Bidding Behavior Agent and Market Sentiment Agent - Assisted Electricity Price Prediction (2024) — link
6. Financial Fine -tuning a Large Time Series Model — → 2412.09880
7. Unified Training of Universal Time Series Forecasting Transformers (2024) — link
8. Time- MoE : Billion - Scale Time Series Foundation Models with Mixture of Experts (2024) — link

3 of 251 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.