This paper highlights the importance of accurate electricity price forecasting (EPF) for effective decision-making in the electricity spot market and evaluates the electricity price forecasting performance of recently developed time-series-based models (TSFMs) based on generative artificial intelligence (GenAI) and pre-trained giant language models (LLMs). State-of-the-art pre-trained models, including Chronos-Bolt, Chronos-T5, TimesFM, Moirai, Time-MoE, and TimeGPT, are compared and analyzed against existing statistical and machine learning (ML) methods. Using the 2024 Daily Futures Market (DAA) electricity price data from Germany, France, the Netherlands, Austria, and Belgium, one-day forecasts are performed. Results show that Chronos-Bolt and Time-MoE perform best among the TSFMs, achieving performance comparable to existing models. However, the bi-seasonal MSTL model, which considers daily and weekly seasonality, consistently outperforms the MSTL model regardless of country or evaluation index. No TSFM statistically outperforms the MSTL model.
