Removing non-physical traders would leave NESO with 3.1 to 3.9 TWh more imbalance to clear over 2021-2024, according to Cary and Morstyn. The arithmetic is defensible. It gives us less reason to trust a forecast of what banning the practice would do. And it tells us nothing about whether the trade still pays.

Who supplies the helpful imbalance?

Since November 2015, GB has settled all imbalances at a single settlement price (SP). A trader expecting the system to finish short buys power in the continuous intraday market without consuming any. It sells the surplus back at SP and profits if SP exceeds its purchase price. That purchase probably brought physical generation onto the system, reducing NESO's need for upward regulation. Usually, a smaller plant sells the power and ramps up to deliver it.

The authors call this speculative imbalance positioning, or SIP. Their definition covers only non-physical traders (NPTs), Elexon parties registered solely as traders. In the counterfactual, the generator supplying an NPT simply produces less. A supplier selling to that trader is assumed to have found a physical counterparty instead.

Elexon's party-level imbalance reports provide the half-hourly data. Removing 1,305 periods with missing observations leaves 68,823 settlement periods. The authors add aggregate NPT imbalance back to NESO's net imbalance volume (NIV), the net of its balancing actions in each period. They then re-dispatch the actual balancing-mechanism stack in merit order, holding constraint-management actions fixed. Their own re-implementation of Elexon's pricing rules supplies the recalculated SP. They use two definitions: AMV includes every NPT imbalance; ZMV includes only NPTs whose metered volume in that period is zero.

The headline figures compare the factual result with AMV and ZMV, respectively:

These figures describe system effects. We did not find an estimate of the traders' own earnings. The volume measure deserves care too. The introduction says "SIP reduced the total volume of energy balancing actions required by 3.1-3.9 TWh (16-20%)". That sounds like gross accepted volume. The table instead sums absolute NIV across periods, netting bids against offers within each half-hour. Gross accepted volume could behave differently.

The identity behind 3.1-3.9 TWh

NIV' = NIV + Imb requires every MWh of NPT position to displace exactly one MWh of physical output elsewhere. It also requires every other party's imbalance to remain unchanged. The authors acknowledge this, describing it as "one of the main limitations of our work." They argue that losing post-gate liquidity makes their assumption conservative. That defence addresses liquidity. Passive balancers remain a problem: these small batteries respond to the live system state. The authors concede that their behaviour would change and that they cannot observe it.

Holding the bid-offer stack fixed follows the authors' account that most speculative trading happens after gate closure. That is plausible. NESO correctly labels constraint actions at least 99.5% of the time (FY2024-25). Keeping those actions unchanged requires a separate assumption. The authors accept that NPT trades could ease constraints or worsen them.

We found no standard errors or confidence intervals for any difference.

The paper's most useful finding challenges a reading Elexon took in 2019. Elexon found in 2019 that most NPT imbalances share the system imbalance's direction. Under AMV, NPT imbalances oppose the factual system imbalance in just 43.1% of periods. Against the counterfactual imbalance, that share reaches 68.0%. Removing NPTs still increases absolute imbalance in 57.4% of periods. Tests based on matching signs therefore understate their contribution.

The capacity figures belong beside that result. At 198 MWh, the mean absolute AMV NPT position approaches the factual average absolute NIV of 233 MWh. Competing traders reverse the imbalance's sign in 25.0% of periods. Their book already operates on the scale of the residual it trades against. A new entrant would be entering a crowded spread.

Do AMV and ZMV bound the effects?

The paper presents AMV as an upper bound on SIP volume and ZMV as a lower bound. For included volume, this holds: ZMV traders form a subset. The measured effects behave differently. ZMV gives the larger volume effect, with 291 against 278 MWh average absolute NIV. AMV gives the larger supplier saving, £325m against £193m.

Absolute value need not move monotonically as positions are removed. NPTs with metered volume, including interconnector traders, can partly offset the zero-metered book. Adding their positions back can therefore reduce the measured effect. These ranges describe two separate classifications.

The percentages use different denominators as well. The 75% supplier reduction uses the counterfactual £431m as its base. So does the 5.9-6.7% balancing saving use a counterfactual base. Although the paper reports factual unflagged spend of £5.54bn, £396m divided by £5.936bn gives 6.7%, and £347m divided by £5.887bn gives 5.9%. The 12% and 21% increases in settlement cashflow to Elexon use the factual £853m. Against the factual base, the £325m supplier reduction would amount to about three times the factual £106m liability.

In 2022, less imbalance still meant a higher NESO bill

Under AMV in 2022, SIP reduced absolute volume while costing NESO money. The average imbalance with traders present was 15.0 MWh short. Removing them made it -49.5 MWh long. NESO paid for offers with traders present; without them, it would have collected on bids.

The authors identify 2022 themselves. They use it to argue that supplier imbalance revenue should be counted alongside balancing costs. Even so, the BM cost saving fails to hold year by year. Under AMV, 2022 brings NESO a net cost, and the benefits in 2024 are "significantly smaller" than in other years.

Over the full 2021-2024 sample, mean SP changes little. It moves +1.47 £/MWh under AMV and -0.62 under ZMV, in opposite directions. The distribution's shape accounts for the supplier saving. With traders present, supplier imbalance has a 0.40 correlation with the system, compared with 0.66 and 0.54 without them. When suppliers and the system are both short, SP is 168.89 £/MWh, compared with 189.23 and 180.74.

We could not test these results ourselves. Doing so requires Elexon's party-level imbalance reports, available only on request, along with the balancing-mechanism stack. Our data cover no GB power contracts.

Preserve intraday access, discount the £347-396m saving

The authors draw on these results to oppose moving intraday close back to gate closure. They acknowledge that the model leaves intraday price effects and constraint interactions unmodelled. NPTs average a net short position (-27.9 MWh under AMV). The authors describe SIP as "a transfer from generators to NPTs, the SO and suppliers."

Outlook Energy funds the DPhil studentship behind the work, and its team commented on drafts. The result that would change my view is specific: the 2022 AMV year turning positive when third-party imbalances can respond, rather than remaining fixed under the identity.