An hourly day-ahead forecast for DK1 and DK2, published every morning before the 12:00 auction, plus a quarter-hour read on whether the system will run short or long over the next four hours.
than "same as yesterday": 21.3 vs 32.9 €/MWh average error, out of sample.
than "same as yesterday": 22.2 vs 34.8 €/MWh average error, out of sample.
Walk-forward test from 2025-12-01 to 2026-11-17: each month scored by a model that had never seen it.
The intraday read is newer and its edge is smaller; its own backtest is shown alongside it.
After a 7-day free trial. Cancel any time from your billing page.
Start free trialIf, in a billing month, our day-ahead forecast doesn't cut the error of "same as yesterday" by at least half of what it does in testing, we refund 50% of that month's fee.
Measured on the forecasts actually issued before each auction, against the cleared prices. How it's measured.