Forecast performance
Daily forecasts are stored with a time stamp before the auction and then compared with the prices that cleared; yearly figures come from a walk-forward backtest. Here is how — and one real day as an example.
Example: Germany–Luxembourg, 13 Nov 2026
The latest day with cleared auction prices, not a hand-picked one. Forecast created 12 Nov 10:34 CET, before the 12:00 CET auction on the day before delivery.
Forecast (median)10–90 % rangeActual price
How we evaluate it
- 15-minute granularity. Every quarter-hour of the delivery day is forecast and checked.
- Stored before the auction. Each morning’s forecast is saved with a time stamp and a hash before the 12:00 CET auction on the day before delivery; it is never changed afterwards.
- Continuous evaluation. After the auction we compare every forecast with the cleared price and with a simple benchmark — the price at the same time one day earlier.
- Walk-forward testing. Backtests run day by day and only use data that was available before that day’s auction, the same way the live forecast does.
One year in Germany
Every day of the last year for Germany–Luxembourg, month by month, with a CSV of every day: forecast accuracy in Germany. What it means for a battery: case study of a 10 MW / 20 MWh battery.
See the same comparison for your zones.
Request a Demo