Benchmarking Pre-Trained Time Series Models for Electricity Price Forecasting

2025-06-09

AI Summary

Accurate electricity price forecasting (EPF) in day-ahead markets is critical for market decision-making, yet the real-world efficacy of pretrained time series foundation models (TSFMs) across multinational settings remains systematically unvalidated. This study conducts the first cross-national empirical evaluation of six TSFMs—including Chronos-Bolt, TimesFM, and Time-MoE—against classical benchmarks (MSTL, SARIMA, XGBoost) on 2024 day-ahead auction data from five European countries (e.g., Germany, France), using single-step-ahead forecasting. Results show that while Chronos-Bolt and Time-MoE achieve top performance among TSFMs, neither significantly outperforms the dual-seasonal MSTL model. MSTL demonstrates superior robustness and stability across countries and evaluation metrics. This work fills a critical gap in the empirical assessment of TSFMs for cross-regional electricity markets and reveals the enduring competitiveness of lightweight statistical models—particularly under high noise and nonstationarity inherent in electricity price series.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Planning, Routing, and Scheduling: Planning with Language Models
• Multiagent Systems: Other Foundations of Multi Agent Systems

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

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.

Problem

• Evaluating pre-trained time series models for electricity price forecasting
• Comparing TSFMs with traditional statistical and ML methods
• Assessing model performance across multiple European electricity markets

Innovation

• Benchmarking pre-trained time series models for EPF
• Comparing TSFMs with traditional statistical methods
• Evaluating models using multi-country DAA electricity prices