Beaten: lower error than both, Diebold-Mariano p < 0.05.
Energinet's published forecast seems to miss part of the fleet; scaled up, it is what a trader could get by correcting it with the last 60 settled days.
On the dashed line the forecast was exactly right.
Error by how much we forecast, as a share of the fleet.
A wind farm that sells its forecast day-ahead settles the miss at the imbalance price. The average bill is about the same for all three: it depends on which way the grid leans, not on the size of the miss.
Standard deviation of the daily bill for the whole fleet. A smaller miss means a steadier bill.
| 2026, mean absolute error | Ours | Energinet | Energinet scaled | Beaten |
|---|---|---|---|---|
| Offshore wind | 170 MW | 360 MW | 210 MW | yes |
| Onshore wind | 180 MW | 194 MW | 154 MW | no |
| Solar | 120 MW | 97 MW | 94 MW | no |
The same rules and the same weather were used for all three. Onshore was lower than Energinet's published forecast but not significantly (p = 0.06), and behind it scaled to the fleet. Solar was behind both. Energinet knows every wind farm and solar plant and forecasts later, so only offshore, where its forecast misses part of the fleet, runs live.