aFRR and mFRR, with the numbers

Two acronyms that decide what imbalance costs. Most explanations stop at what the letters stand for. This one goes to how often they are actually activated, what they are paid, and which endpoint gives you the data.

The reserve family in one table

Grid frequency has to stay at 50 Hz. When generation and consumption drift apart, reserves close the gap — in layers, each slower and cheaper than the one before it.

FCR stops the fall, aFRR restores the frequency, mFRR relieves aFRR so it is free again. That sequencing is the whole point: the fast product is scarce and expensive, so the system keeps handing the load down to slower ones.

How often they are really activated

Definitions make aFRR sound like an emergency measure. It is not — it runs almost constantly. These are the Czech control area figures for August 2026, 2 977 quarter hours:

Both directions of aFRR are active in roughly three quarters of all quarter hours, and often both within the same quarter hour — the imbalance flips sign inside fifteen minutes and the controller follows it. Anyone modelling a quarter hour as "short" or "long" is already simplifying.

mFRR is a different animal: upward five times more often than downward. Downward mFRR means telling plants to produce less, and systems are short far more often than long.

What it costs

Czech regulating energy prices for the same month, 44 641 minute values:

The median is unremarkable and the maximum is fifty times higher. That gap is the entire economics of balancing. Reserve revenue is not a steady stream with occasional noise — it is a flat line with a handful of hours that pay for the year.

The same shape shows up in the imbalance settlement price, which is built from these activations. We measured that on five years of data: the tail is 18.6 % of quarter hours but 35.9 % of total magnitude.

The European platforms

Until recently every TSO balanced alone. Now the energy side of each product clears on a shared European platform, which is why a German activation can be served by a Czech unit:

Practical consequence: your local activation price is no longer a local variable. It is set by the marginal bid across a coupled area, which is why prices in neighbouring zones move together until the border congests, and then abruptly do not.

Where to get the data

Every TSO publishes this, but nowhere the same way.

# Czech aFRR and mFRR activation, no API key needed
from ceps_ote import Ceps

ceps = Ceps()
akt = ceps.balancing_activation("2026-08-01", "2026-09-01")

for t, row in list(akt.items())[:3]:
    print(t, row["aFRR+ [MW]"], row["mFRR+ [MW]"])

That client is on GitHub: ceps-ote, one file, no dependencies. For everything else there is entsoe-quickstart and the ENTSO-E API guide.

Three traps worth knowing

1. Capacity price and energy price are different numbers. A revenue model that adds them without knowing which hours were actually activated will overstate income by a wide margin.
2. Published prices are provisional. Most TSOs revise balancing data after the fact. If you train a model on the final value and claim you could have traded on it, that is look-ahead bias — the single most common way a balancing backtest lies to you.
3. Sign conventions differ per TSO. Downward activation is negative in some feeds, positive with a direction flag in others. Check against a quarter hour you can reason about before trusting a series.

Frequently asked questions