Forecast
Share of perfect foresight (Ensemble)
94%
327 days
Earned in total, after costs
57,761 €
177 € a day, before costs 69,475 €
Days with a loss
11
Worst: -24 €
What the battery would have earned after costs, added up
Each forecast, the same battery
| Forecast | Days | After costs | Share of perfect | Loss days | vs ensemble, a day |
|---|---|---|---|---|---|
| Ensemble | 327 | 57,761 € | 94% | 11 | |
| BMT-X | 327 | 56,079 € | 91% | 16 | -5.1 € (p < 0.001) |
| LEAR-X | 327 | 56,987 € | 92% | 13 | -2.4 € (p 0.008) |
| XGBoost-X | 327 | 57,478 € | 93% | 12 | -0.9 € (p 0.363) |
Difference to the ensemble per day, p: Newey-West, Holm for the three.
Share of perfect foresight by month (Ensemble)
Share of perfect foresight (BMT-X)
91%
327 days
Earned in total, after costs
56,079 €
171 € a day, before costs 67,984 €
Days with a loss
16
Worst: -62 €
What the battery would have earned after costs, added up
Each forecast, the same battery
| Forecast | Days | After costs | Share of perfect | Loss days | vs ensemble, a day |
|---|---|---|---|---|---|
| Ensemble | 327 | 57,761 € | 94% | 11 | |
| BMT-X | 327 | 56,079 € | 91% | 16 | -5.1 € (p < 0.001) |
| LEAR-X | 327 | 56,987 € | 92% | 13 | -2.4 € (p 0.008) |
| XGBoost-X | 327 | 57,478 € | 93% | 12 | -0.9 € (p 0.363) |
Difference to the ensemble per day, p: Newey-West, Holm for the three.
Share of perfect foresight by month (BMT-X)
Share of perfect foresight (LEAR-X)
92%
327 days
Earned in total, after costs
56,987 €
174 € a day, before costs 68,724 €
Days with a loss
13
Worst: -35 €
What the battery would have earned after costs, added up
Each forecast, the same battery
| Forecast | Days | After costs | Share of perfect | Loss days | vs ensemble, a day |
|---|---|---|---|---|---|
| Ensemble | 327 | 57,761 € | 94% | 11 | |
| BMT-X | 327 | 56,079 € | 91% | 16 | -5.1 € (p < 0.001) |
| LEAR-X | 327 | 56,987 € | 92% | 13 | -2.4 € (p 0.008) |
| XGBoost-X | 327 | 57,478 € | 93% | 12 | -0.9 € (p 0.363) |
Difference to the ensemble per day, p: Newey-West, Holm for the three.
Share of perfect foresight by month (LEAR-X)
Share of perfect foresight (XGBoost-X)
93%
327 days
Earned in total, after costs
57,478 €
176 € a day, before costs 69,061 €
Days with a loss
12
Worst: -23 €
What the battery would have earned after costs, added up
Each forecast, the same battery
| Forecast | Days | After costs | Share of perfect | Loss days | vs ensemble, a day |
|---|---|---|---|---|---|
| Ensemble | 327 | 57,761 € | 94% | 11 | |
| BMT-X | 327 | 56,079 € | 91% | 16 | -5.1 € (p < 0.001) |
| LEAR-X | 327 | 56,987 € | 92% | 13 | -2.4 € (p 0.008) |
| XGBoost-X | 327 | 57,478 € | 93% | 12 | -0.9 € (p 0.363) |
Difference to the ensemble per day, p: Newey-West, Holm for the three.
Share of perfect foresight by month (XGBoost-X)
Days below the 1 in 20 floor (Ensemble)
7.3%
24 of 327 days, target 5%
Kupiec test
p = 0.07
Is the share of days below the floor 5%?
Average expected
165 €
Per day
Average earned
177 €
Per day
Expected revenue, the 1 in 20 floor and what the battery earned (red: below the floor)
Days below the 1 in 20 floor (BMT-X)
10.7%
35 of 327 days, target 5%
Kupiec test
p = 0.00
Is the share of days below the floor 5%?
Average expected
176 €
Per day
Average earned
171 €
Per day
Expected revenue, the 1 in 20 floor and what the battery earned (red: below the floor)
Days below the 1 in 20 floor (LEAR-X)
6.1%
20 of 327 days, target 5%
Kupiec test
p = 0.37
Is the share of days below the floor 5%?
Average expected
161 €
Per day
Average earned
174 €
Per day
Expected revenue, the 1 in 20 floor and what the battery earned (red: below the floor)
Days below the 1 in 20 floor (XGBoost-X)
10.1%
33 of 327 days, target 5%
Kupiec test
p = 0.00
Is the share of days below the floor 5%?
Average expected
169 €
Per day
Average earned
176 €
Per day
Expected revenue, the 1 in 20 floor and what the battery earned (red: below the floor)
Before costs
70,205 €
1.89 MWh sold a day
Costs
11,714 €
17% of the revenue
After costs
57,761 €
177 € a day
Days it doesn't trade
16
Loss days 22 → 11
Fees and wear, 18 Nov 2025 to 10 Oct 2026, ensemble plan
| Costs | Plan | Costs | After costs | MWh sold a day | Days no trade | Loss days | In the plan vs after (per day) |
|---|---|---|---|---|---|---|---|
| No costs | no costs | 0 € | 70,205 € | 1.89 | 0 | 2 | |
| Fee + wear 20 €/MWh | after the fact | 12,461 € | 57,745 € | 1.89 | 0 | 22 | |
| Fee + wear 20 €/MWh | in the plan | 11,714 € | 57,761 € | 1.78 | 16 | 11 | +0.05 € (p 0.916) |
| Fee + wear 10 €/MWh | after the fact | 6,266 € | 63,939 € | 1.89 | 0 | 10 | |
| Fee + wear 10 €/MWh | in the plan | 6,161 € | 63,913 € | 1.86 | 4 | 7 | -0.08 € (p 0.720) |
| Fee + wear 40 €/MWh | after the fact | 24,850 € | 45,356 € | 1.89 | 0 | 68 | |
| Fee + wear 40 €/MWh | in the plan | 20,664 € | 46,820 € | 1.58 | 50 | 25 | +4.48 € (p 0.001) |
| Fee + wear 20 €/MWh + Energinet's tariffs | after the fact | 23,926 € | 46,279 € | 1.89 | 0 | 64 | |
| Fee + wear 20 €/MWh + Energinet's tariffs | in the plan | 20,085 € | 47,547 € | 1.59 | 48 | 24 | +3.88 € (p 0.002) |
After the fact: costs taken off afterwards. In the plan: it only trades when the spread pays for them. At 20 €/MWh wear the total is the same, but loss days halve. The higher the costs, the more it pays to plan with them. Wear is an assumption, sources in the README. p: Newey-West.
Revenue given up for each loss limit
Loss limit backtest, 18 Nov 2025 to 10 Oct 2026, ensemble plan
| Limit | Days it binds | Revenue | p | Loss days | Worst day |
|---|---|---|---|---|---|
| None | 0 | +0.0% | 2 | -22 € | |
| 100 € | 5 | -0.3% | 0.192 | 2 | -22 € |
| 50 € | 24 | -1.3% | 0.017 | 2 | -22 € |
| 25 € | 57 | -2.8% | 0.001 | 2 | -14 € |
| 0 € | 102 | -7.5% | < 0.001 | 0 | 0 € |
The limit is on the average of the worst 5% of scenarios. It costs more than it saves: losses are rare, and the days it hits are the volatile ones, which pay the most. Before fees and wear.