The penalty is |imbalance − day-ahead| — in absolute value, because on any given slot the sign depends on which way the system deviated, not you. The magnitude is what you pay for being off schedule, either way.
The difference keeps its sign and says something else: whether the month was one where imbalance sat above or below the market price.
The median sits next to the mean everywhere, because the distribution is skewed. In 2026 the mean is about 1,6× the median: a month with a low mean can still contain slots of several thousand lei.
An example from the table below. In 2026-02, mean imbalance was 509 lei and the mean market price 508 — a difference of +1 lei. On each individual quarter-hour, though, the distance between them averaged 620 lei. It cancels out across slots. That is why the difference says which way the month leaned, and the penalty says what it cost.
| Month | Coverage | Day-ahead | Imbalance | Difference | Mean penalty | Median | Slots above DA | Min | Max |
|---|---|---|---|---|---|---|---|---|---|
| 2026-01 | 100% | 766 | 1.090 | +323 | 852 | 477 | 60% | -10.019 | +16.241 |
| 2026-02 | 100% | 508 | 509 | +1 | 620 | 427 | 51% | -14.829 | +14.638 |
| 2026-03 | 100% | 536 | 384 | -152 | 722 | 483 | 39% | -8.897 | +9.343 |
| 2026-04 | 100% | 487 | 121 | -366 | 906 | 511 | 34% | -7.524 | +6.956 |
| 2026-05 | 100% | 573 | 463 | -110 | 781 | 528 | 42% | -8.150 | +6.536 |
| 2026-06 | 100% | 645 | 834 | +189 | 722 | 454 | 60% | -10.368 | +17.865 |
| 2026-07 | 100% | 629 | 368 | -261 | 739 | 488 | 41% | -8.305 | +8.039 |
| 2026-08 | 100% | 794 | 756 | -38 | 534 | 425 | 47% | -8.618 | +6.552 |
Mean penalty by local hour, across 2026. It is the one dimension anyone can act on tomorrow: the same nomination error costs differently at 19:00 than at 04:00.
| Hour | Mean penalty | Median | Slots |
|---|---|---|---|
| 00:00 | 681 | 480 | 972 |
| 01:00 | 583 | 433 | 972 |
| 02:00 | 546 | 427 | 972 |
| 03:00 | 523 | 421 | 967 |
| 04:00 | 486 | 411 | 972 |
| 05:00 | 500 | 388 | 972 |
| 06:00 | 569 | 435 | 972 |
| 07:00 | 711 | 516 | 972 |
| 08:00 | 831 | 538 | 972 |
| 09:00 | 870 | 548 | 972 |
| 10:00 | 821 | 503 | 972 |
| 11:00 | 645 | 440 | 972 |
| 12:00 | 591 | 414 | 972 |
| 13:00 | 604 | 359 | 972 |
| 14:00 | 650 | 400 | 972 |
| 15:00 | 814 | 454 | 972 |
| 16:00 | 839 | 484 | 972 |
| 17:00 | 1.003 | 582 | 972 |
| 18:00 | 945 | 633 | 972 |
| 19:00 | 1.017 | 668 | 972 |
| 20:00 | 901 | 650 | 972 |
| 21:00 | 804 | 546 | 972 |
| 22:00 | 954 | 567 | 972 |
| 23:00 | 757 | 513 | 972 |
The published imbalance price is a single one per slot, so it does not say what it costs to be short versus long. Direction comes from aFRR activations: up = the system buys energy (someone delivered less than notified), down = the system refuses it.
| Direction | Slots | Mean price | Min | Max |
|---|---|---|---|---|
| up (system buys) | 23.267 | +857 | -1.016 | +57.525 |
| down (system refuses) | 23.267 | -30 | -7.999 | +1.600 |
Of the year's total penalty, the most expensive 1% of slots carry 9,6%, the top 5% carry 30,7%, and 10% carry 43,3%.
What that means concretely. That 1% amounts to 2,4 days a year, gathered from scattered quarter-hours. Within them the mean penalty is 7.030 lei/MWh — 15× what a typical quarter costs (471 lei). The rest of the year is cheap; a few days are not.
Why it changes what you do. A volume tax is reduced by deviating less, evenly: effort spread equally, gain spread equally. A tail risk is reduced differently — by not being wrong in the expensive quarters. The same hour of work spent on the evening peak is worth ten spent on the night.
And they are not random. The table above shows them clustering in the evening. A tail that has a time of day can be anticipated — which is why it deserves to be read apart from the mean.
For budgeting. It cannot be treated as a per-MWh cost. A month's mean says almost nothing about what a bad month costs: the same mean can hide a quiet month or two expensive days.