The busiest batteries are the worst at following the price, and nobody can tell us why

Everybody assumes a battery charges when power is cheap and sells when it is dear. It is the whole premise of the asset class. It is also checkable, half hour by half hour, and some sites do it fifteen times more tightly than others. The thing that predicts which is not size, not duration and not what they earn. It is how often they change direction. Why they change direction is a question we have now failed to answer six different ways, so we are publishing the question.

What was measured

For twenty one days we recorded where every battery in the balancing mechanism actually was, half hour by half hour. Not what it said it would do: the level it was instructed to run at where the control room had instructed it, and the level it notified where it had not. Reading notifications alone gives the wrong answer, and we have written up why.

Then each site's position was set against the wholesale reference price in the same half hour. A site that sells hard into the dear half hours scores high. A site whose position has nothing to do with the price scores near zero.

The first thing it measures is whether a site moves at all

Twelve sites moved in fewer than 15 per cent of half hours over three weeks. They score near zero, and that says nothing about how well they are traded, only that they were not doing much. So they are set aside.

That leaves 66 sites, 3,088 megawatts, moving in at least 40 per cent of half hours. Among those the spread is wide and it is real.

Fifteen times the difference, between equally busy sites

Whitebirk and Newport are moving four days in five, more than almost anything else on the list, and their position tracks the price barely at all. Burwell and the Kilmarnock South units move slightly less often and track it fifteen times more closely.

Four explanations that do not survive

It is not that they are selling reserve instead. Splitting the 66 into thirds by score, the share of money from reserve auctions runs 28, 33 and 40 per cent. Wrong direction and far too small. One of the most responsive sites earns nothing from reserve at all; one of the least earns 62 per cent of its money there.

It is not duration. We now hold a storage duration for 138 of the fleet, published for 105 and measured for 33. Against price response the correlation is +0.08. One hour and two hour batteries score about the same.

It is not size. Correlation +0.11.

It is not how hard they push when they move. Correlation −0.08. The unresponsive sites are not making timid little moves. They push about as deep as anyone.

What does explain it

How often a battery changes direction. Correlation −0.67.

That is much the strongest relationship in the data, and it runs the opposite way to intuition. The more often a battery switches between charging and discharging, the less its position has to do with the price.

Half an explanation, and we are not going to pretend it is a whole one

A battery trading the price curve has a small number of decisions to make in a day. Fill up when power is cheap, empty out when it is dear, perhaps twice. That is a handful of direction changes. Twenty three is not a set of trading decisions. Nobody has twenty three opinions about the price in a day.

So something else is moving these assets, and it moves them constantly. That much the data shows clearly. What that something is, we have not established.

The obvious answer, tested twice and refused twice

The natural explanation is response work: an asset nudged up and down around a set point by system frequency, which has nothing to do with the price of power. We tested it two different ways and it failed both.

By money. The share of income from reserve auctions runs 28, 33 and 40 per cent across the thirds. Wrong direction, and far too small a gap.

By capacity held. Money is a poor proxy, because reserve pays for availability rather than energy, so an asset could spend most of its time on it for a modest share of its income. So we measured the megawatts each site actually had committed to reserve, half hour by half hour, over fifty days. Against how often a battery changes direction, the correlation is +0.05. Against price response, +0.05. The batteries that flip constantly hold no more reserve than the ones that do not.

That is a clean refusal and we accept it. An earlier version of this page offered frequency response as the likely answer. It was wrong and this is the correction.

Two other explanations, also tested

Co-location. A battery behind the same connection as a wind or solar farm is not free to follow the price; it is firming whatever the generator is doing. We cannot test this properly and we will not pretend otherwise. The planning database says which projects are co-located but its coordinates and ours agree only loosely, and proximity alone cannot tell a shared connection from a solar farm in the next field. Ten of our assets have a generator within 500 metres and 105 have one within three kilometres, which tells you how little the distance settles. This is the most promising untested explanation on the list.

Balancing the trader's own book. A company selling power to customers can use a battery to shave its own imbalance rather than to trade the curve. Splitting the sites by whether their trader has a large book of its own gives median scores of +0.21 against +0.23. No difference we can see, though our reading of which companies have their own book is a judgement rather than a fact.

Where this leaves it

We can describe the difference precisely and we cannot yet explain it. The relationship between direction changes and price response is strong, consistent and survives every control we have been able to apply. Every candidate cause we can measure has come back flat.

That is an honest place to stop and an obvious place to ask. If you operate one of these and know what is moving it, the methods section says how to tell us, and we will publish the correction with your name on it if you want it there.

One number worth knowing while you think about it: across these sites, most of the fleet's power is committed to reserve for most of the time. That is measured across stacked services, so the level is overstated where a site sells the same megawatt into more than one product, but it is high on any reading. Free capacity to trade with is scarcer than the headline megawatts suggest.

One more thing, which may be nothing

Great Britain's largest battery is Coalburn, five balancing units and about 512 megawatts. Over the same three weeks each of those units moved in between 6 and 11 per cent of half hours. It is entirely possible that is a commissioning programme rather than a business decision, and we say so rather than guess. But it is the biggest thing on the system and it was mostly still.

Every unit has its own page, with what it sold and what it bought against the market index over the same half hours.

Twenty one days to 19 August 2026, 94 sites with enough half hours to score, 998 of 1,008 half hours carrying a price.

See it on the map