Reproduce “Foundation models for electricity price forecasting and battery arbitrage: Can they replace market-specific forecasting models?” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Arkadiusz Lipiecki, Rafał Weron
Researchers from Wrocław University of Science and Technology and Aarhus University evaluated foundation models for day-ahead electricity price forecasting and battery arbitrage across diverse European markets. Their work found that while tabular foundation models like TabPFN statistically outperformed specialized benchmarks, economic profitability was contingent on trading strategy and risk tolerance, with different models excelling in varying risk profiles.
Key insights
- Statistical forecast accuracy does not directly translate into economic performance, as evidenced by differing model rankings when evaluated using statistical metrics versus BESS arbitrage profits.
- Tabular foundation models, particularly the TabPFN family, consistently demonstrated superior statistical accuracy for both point and probabilistic forecasts across all markets compared to time series FMs and benchmarks.
- The most profitable forecasting model for BESS arbitrage is highly dependent on the trading strategy and the decision-maker's risk tolerance, with different models excelling under risk-averse versus riskier scenarios.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.