Skip to content

FAQ

Day-ahead wholesale electricity prices for Spain, Portugal, France and Germany, for every 15-minute period, up to seven days ahead.

Twice a day. The day-ahead forecast for tomorrow is issued around 10:00 UTC, before the day-ahead auction results are published. The strategic forecast for D+2 to D+7 is issued around 15:00 UTC, after tomorrow’s prices are known.

Forecasts for Spain and Portugal (the Iberian market, OMIE), France and Germany. The app shows actual day-ahead prices for 45 European bidding zones (4 forecast).

15 minutes, for all four markets, matching how the European day-ahead market has cleared since October 2025.

A bidding tool on Spain’s real day-ahead and intraday auctions. Bids are virtual: they are placed before each real closing time and settled on the real prices, but no money changes hands. Every result is compared with EPF’s own recommendation, and the Balance view re-runs the same portfolios on naive forecasts since 1 January 2026.

In Spain and Portugal the day-ahead forecast is over 20% more accurate than every naive baseline in a 12-month walk-forward test: 23.5% in Spain (MAE 18.02 €/MWh) and 22.5% in Portugal (19.62). France (34.96, 11.2% behind) and Germany (38.79, 17.0% behind) are behind the naive today. Figures as of 24 Oct 2026; live production has trailed them. Every number, live and walk-forward, is on the Track record page.

Prices repeat strongly from day to day and week to week, so simple rules like “tomorrow will look like today” are hard to beat. A forecast is only useful if it beats them. See Walk-forward and baselines.

Germany has the largest swings, including frequent negative prices, and one key public input, an official day-ahead wind and solar forecast, is not published for it on the European transparency platform.

Why is live accuracy worse than the walk-forward?

Section titled “Why is live accuracy worse than the walk-forward?”

The live record includes earlier models and a period when some inputs were missing in production at forecast time. Those issues were fixed in September and October 2026. Spain’s live window is also short and fell in a high-price period, when errors are larger in absolute terms.

Because a forecast can have a low average error and still be useless for trading. EPF also reports direction accuracy, spread capture and spike recall. See Evaluation metrics.

Gradient-boosted decision trees: one model family per market and product, with a model for each group of horizons, each forecasting directly. See Forecasting approach and Design decisions.

Public data only: market prices, grid demand and generation, official renewable forecasts, weather, and fuel and carbon prices. See Data sources.

A candidate must beat the current model and the naive baselines in a 12-month walk-forward test, win on most individual days, and not worsen the economic measures. An automated research pipeline proposes and tests candidates; a human approves every promotion.

How does it handle negative prices and spikes?

Section titled “How does it handle negative prices and spikes?”

Both are in the training data and the models can forecast them. Very cheap and negative hours remain harder than average, and extreme spikes driven by events with no warning in public data cannot be anticipated. A forecast run stops rather than publish implausible prices.

What does the band around the forecast mean?

Section titled “What does the band around the forecast mean?”

It shows the typical error for each time of day. See Reading the uncertainty band.

Yes, in private beta. See API access.

Yes, free during the beta. API access on request: see API access.