Reducing Electricity Use During Peak Hours (Demand Flexibility Service)
Introduction
The Demand Flexibility Service (DFS) was a market created by Great Britain’s National Energy System Operator (NESO) to buy voluntary reductions in electricity use at times of system stress. When the forecast system margin fell below an operational threshold, NESO called an event (typically one to one and a half hours during the evening peak) and retailers and aggregators bid to deliver a given quantity of reduction at a given price. Accepted bids during Winter 2023–24 ran from £150 to £6,000 per megawatt-hour, and providers chose how much to pass through to the customers whose behavior produced the reduction. Octopus Energy Limited, the partner utility in this study, offered its customers between £1.75 and £4 per kilowatt-hour. Households enrolled once in the provider’s flexibility program, then opted in ahead of each event. Payment was based on metered consumption against a customer-specific baseline set by the regulated P376 methodology.
Bernard, Metcalfe, and Schein (2026) evaluate the program with a randomized encouragement design across roughly 2.7 million households and businesses, withholding the sign-up invitation from a control group of about 120,000. Of the sixteen events called that winter (two live events called by NESO in response to system stress and fourteen coordinated test events), thirteen enter the paper’s demand analysis and twelve the welfare analysis. Because enrollment was voluntary and the encouragement randomized, the paper identifies effects for compliers: the households induced to sign up and opt in by the invitation email. Opting in cut consumption by 23.1 percent during event windows, a local average treatment effect of 0.1773 kilowatts per opted-in household. Every quantity in the welfare analysis is measured on these compliers, who delivered 295.21 megawatt-hours across the twelve events, so the MVPF below values a marginal expansion of DFS enrollment through an email encouragement rather than the program as a whole.
In addition, the paper estimates a second MVPF that captures the capacity adequacy benefit of DFS. That scenario asks what the reductions were worth when the alternative was not a cheaper balancing action but involuntary disconnection.
MVPF = 0.6
Net Cost
The budget in the denominator is that of the system operator, NESO, which procures DFS.
The payment to the flexibility provider for delivered demand reduction, taken from settlement records at the half-hourly settlement-period level, came to £686,400 across the twelve events (roughly £2,325 per megawatt-hour of complier reduction).
Against this the paper nets the balancing actions NESO did not have to take because demand fell. Each settlement period is assigned a marginal generator: the highest-priced accepted Balancing Mechanism action of at least 25 megawatts managing overall supply and demand balance. Each event’s reduction is valued at that event’s marginal price. Those prices ranged from about £58 to £900 per megawatt-hour, highest in the live November and December events, and imply a volume-weighted average of about £347 per megawatt-hour, or £102,385 of avoided balancing cost.
Netting the two gives £686,400 – £102,385 = £584,015.
The paper treats this payment net of avoided balancing cost as the relevant government cost and does not extend the denominator further: it reports that the program’s VAT, corporate tax and utility revenue effects were small relative to the system operator’s direct expenditure, and omits out on that basis.
Willingness to Pay
The estimated willingness to pay has three components.
Consumer surplus of £274,500: Eighty percent of the system operator’s payment was passed through to households. Treating every participating household as marginal, the standard Harberger triangle approximation under a linear demand curve values the marginal participant’s gain at half the payment received, so half of that household share is counted as surplus. The paper does not add households’ own bill savings separately: by the envelope theorem, the payment already captures the value of the consumption participants gave up.
Utility surplus of £68,600: The remaining 20 percent of the payment was retained by the retailer, split evenly between the logistical costs and overheads of delivering the program (a resource cost, generating no willingness to pay) and retailer surplus, which is counted.
Carbon abatement of £24,500: Reducing demand at peak displaced fossil generation at the margin. Valued at each settlement period’s marginal fuel emissions factor, the twelve events avoided 97.1 tonnes of CO2-equivalent, which the paper values at HM Treasury’s carbon value of approximately £250 per tonne.
The total is then £274,500 + £68,600 + £24,500 = £367,600.
The paper provides some evidence to support the 50 percent valuation of household payments: In a separately pre-registered sub-trial, 5,000 signed-up households were randomly offered £2.25 per kilowatt-hour for an upcoming event instead of the standard £1.75. Opt-in rose from 27.9 to 36.7 percent, an opt-in elasticity with respect to the incentive of approximately 1.10. Under a constant-elasticity participation supply, surplus is the payment divided by one plus that elasticity, so an elasticity of 1.10 implies that about 47.6 percent of the payment accrues as consumer surplus.
MVPF Summary
Dividing willingness to pay by the net cost to the system operator yields £367,600 / £584,015 = 0.63: the value of expanding DFS enrollment at the margin through an email encouragement.
The ratio falls below one because household remuneration was high relative to the balancing actions displaced: households were paid an average of about £1,860 per megawatt-hour, against avoided balancing actions worth about £347. The paper notes that these estimates omit learning-by-doing (the future cost reductions that come as platforms improve and consumers grow familiar with the program), which suggests this base-case figure may be a lower bound estimate of the MVPF.
The paper also reports a second scenario in which DFS is treated not as a substitute for routine balancing but as scarcity insurance, so that the counterfactual to a household’s reduction is involuntary disconnection. Carbon abatement and avoided balancing costs drop out, and each megawatt-hour is instead valued at the GB regulatory value of lost load (VoLL) of £6,000 per megawatt-hour, worth £1,771,260. Willingness to pay becomes £274,500 + £68,600 + £1,771,260 = £2,114,360 against a cost of £686,400, an MVPF of 3.08. This is best read as an upper bound, while the 0.63 above describes the conditions actually obtained in Winter 2023–24.
References
Bernard, Louise, Robert D. Metcalfe, and Andrew Schein (2026). “Critical Peak Rewards: Evidence from a Nationwide Demand Response Program.” NBER Working Paper 35485. http://www.nber.org/papers/w35485
Policy Features
- Category
- Environment
- Sub-Category
- Environmental Subsidies
- Beneficiary Type(s)
- Adults, Firms
- Country of Implementation
- United Kingdom
- Year of Implementation
- 2023-2024
- Empirical Method
- RCT
- Research Type
- Primary
- Peer Reviewed
- No
- MVPF Publication Link
- www.nber.org/papers/w35485