Adequacy & Capacity Markets

The missing-money problem

Price caps, out-of-market interventions and the political impossibility of letting scarcity prices run their course can leave peaking and flexible resources unable to recover fixed costs from energy revenues alone. Capacity mechanisms exist because someone concluded this gap is real; their critics argue the gap is mostly self-inflicted.

Stoft's version of the argument cuts deeper than “caps are too low.” The textbook theorem that competitive short-run prices induce optimal investment quietly assumes supply and demand always intersect — that is, it assumes shortage never happens. But adequacy is precisely about the hours when near-inelastic demand goes vertical and no market-clearing price exists; in those hours the price is set by an administrator, a cap, or market power — never by competition. Expecting the energy market to price its own failure moments is the incoherence at the heart of the energy-only position. And the marginal adequacy investment makes the exposure concrete: it is a peaker that must recover its entire fixed cost from a handful of super-peak hours — on average perhaps tens of minutes a year, in practice a few extreme hours and then nothing for a decade. Revenue that rare and that spiky is very expensive risk capital, whatever its expected value.

Reliability standards

The modern EU approach derives the standard rather than asserting it: estimate the value of lost load (VoLL), estimate the cost of new entry (CONE), and set the loss-of-load expectation (LOLE) where the two curves cross. A standard of three hours per year is not a law of nature — it is an economic statement about how much consumers value not being curtailed.

Probabilistic assessment

The European Resource Adequacy Assessment (ERAA) and its national counterparts simulate thousands of weather-and-outage years against the projected fleet. The methodology choices — weather correlation across borders, forced-outage draws, demand flexibility assumptions — move the results as much as the fleet itself does.

Scarcity pricing as the alternative

The purist answer to missing money is to fix the energy market instead: administrative scarcity adders tied to reserve shortfall (the Texas ORDC being the canonical example) let prices rise smoothly toward VoLL. Whether politicians can commit to not intervening is the real design constraint.

The theory is elegant: set the shortage price at the value of lost load and both timescales come right at once — in the short run the scarce megawatt-hours are rationed to whoever values them most, and in the long run capacity gets built exactly when its cost falls below expected shortage hours times VoLL. The catch is that VoLL cannot be discovered by the market it is supposed to discipline — estimates span €1,000 to €100,000 per megawatt-hour, and no trade reveals which is right, because during shortage there is no clearing price to observe. Scarcity pricing is therefore always administered pricing wearing market clothing; the honest version of the debate is not “market versus regulation” but which administrator, setting which number, with which commitment device against interference — and with market power at the super-peak (where withholding a few hundred megawatts can move the price a hundredfold) policed by what means.

Firmness is relative

A megawatt of nameplate capacity is not a megawatt of adequacy. De-rating factors translate wind, solar, storage duration and demand response into firm-equivalent contribution — and the chosen de-rating curve quietly decides which technologies a capacity mechanism actually buys.

Adequacy is now cross-border

Interconnected systems share scarcity. EU rules require capacity mechanisms to admit foreign capacity and to reflect interconnector contribution — which makes national adequacy a coordination problem: every country counting the same neighbour's surplus is the classic failure mode.

Centralised capacity auctions

The system operator forecasts demand, builds a sloped demand curve and buys firm capacity years ahead (typically a main auction around T-4 refined by a T-1 top-up). Great Britain and Poland run the archetype. Strength: bankable multi-year contracts for new build. Weakness: the central forecast becomes the market.

Decentralised obligations

Suppliers must hold capacity certificates matching their customers' contribution to system stress — the model France ran for a decade before replacing it with centralised T-4/T-1 auctions in 2026. The market decides how much capacity each retailer contracts; the state decides only the obligation formula. Elegant in theory; in practice certificate-market liquidity and stress-period definition carried the design, and its principal implementation has now moved on.

Reliability options

Capacity providers receive a fixed premium and return energy revenues above a strike price — a one-way contract for differences on scarcity, used in Italy and Ireland. The consumer buys a hedge, not just steel: high prices still signal scarcity operationally, but their rents flow back.

The design's genius, worked out by Stoft two decades ago, is what the hedge does not change. A generator that fails to produce during a spike still forgoes the market price and still owes the option payout — its marginal incentive to perform is identical to facing the full uncapped price. What changes is the shape of revenue: the lottery ticket of rare spike income is converted, through the capacity auction, into a steady annual payment of the same expected value, collapsing the risk premium investors charge. And withholding to manufacture a spike becomes pointless, because the withheld capacity owes hedge payments on the very price it inflated. Stoft's accounting even quantified the stakes: the cost of slightly-too-much reliability is small — peakers are cheap — but the cost of spike-risk capital and the political blowback of uncapped prices is not. The hedge attacks the expensive problem and leaves the cheap one alone.

Strategic reserves

Keep the mechanism out of the market: contract retiring plants to sit idle, activated only when the market fails to clear, as in the German and Swedish reserves. Minimal market distortion, but capacity in the reserve is capacity the energy market can never price again — the slippery-slope argument writes itself.

Targeted vs. market-wide

Paying only new entrants (or only at-risk plants) is cheaper per megawatt but discriminates within the merit order and invites gaming at the boundary. Market-wide mechanisms are cleaner and much more expensive. State-aid discipline has historically pushed toward market-wide, technology-neutral designs.

Demand-side and storage participation

A capacity product defined as always-on availability excludes precisely the resources cheapest at covering rare stress events. Duration limits, testing regimes and penalty structures determine whether batteries and demand response are genuine competitors or decorative participants.

Insurance or subsidy?

A well-designed mechanism buys insurance the energy market cannot sell; a badly designed one becomes a permanent side-payment to incumbent generation. The test is counterfactual: would the procured capacity exist anyway? Payments to must-run plant fail it.

The demand curve is the design

In centralised auctions, the administratively drawn demand curve — its cap, its slope, its target level — determines prices more than competition does. A vertical curve at a generous target guarantees expensive procurement; a sloped curve concedes that adequacy is a continuous good, not a cliff.

The sloped version is built from three anchors. The spine is Net CONE — the annualised cost of the cheapest new entrant, net of what it expects to earn in the energy and ancillary markets: by construction, when the auction procures exactly the target, the price lands at Net CONE and the marginal new plant just breaks even, which is precisely the missing money the mechanism exists to pay. Left of the target the curve rises to a cap (typically around 1.5× Net CONE) so shortage years pay more and pull entry forward; right of it the curve slopes to zero a few percent past the requirement, because capacity beyond the need is worth little. The slope itself is the point: with a vertical curve, one retiring plant can swing the price from zero to the cap, and a large incumbent can engineer that swing by withholding — a gentle slope makes the price move smoothly with the surplus, stabilises revenues across years, and shrinks what withholding can earn.

Constructing a sloped capacity demand curve (VRR-style, as in PJM or ISO-NE): capped left of the target, anchored at Net CONE at the reliability requirement, falling to zero just beyond it. The grey vertical shows the alternative — a fixed target where price is either cap or zero.

Emission limits inside adequacy

EU rules attach CO2 intensity limits to capacity-mechanism eligibility, making the capacity market a decarbonisation instrument by the back door. The design tension is real: excluding the highest-emitting flexible plant tightens the very adequacy the mechanism exists to buy.

Market power at the exit door

Adequacy auctions concentrate exactly where exit looms: the marginal supplier deciding between closing and bidding knows its own pivotality. Auction monitoring, price caps on price-setters and mothballing rules are the standard defences; none is fully satisfying.

Cost allocation and stress incentives

Charging capacity costs pro rata to consumption is simple; charging them to consumption during system stress turns every consumer into an adequacy resource. Peak-coincident allocation is quietly one of the strongest demand-response instruments available.

Sunset discipline

The hardest test: mechanisms should shrink as scarcity pricing, storage and demand flexibility mature — but every procured megawatt creates a constituency for the next auction. Credible periodic review against the energy-only counterfactual is the design feature most often promised and least often exercised.

Nine questions to ask of any mechanism

The trade-offs above are cross-cutting; the choice in front of a government is not. It is a choice between instrument families, and the honest way to compare them is to fix the questions first and then answer them the same way for each. Does the mechanism actually finance new build, or only pay for what exists? Who controls what consumers end up paying, and is that control a market or an administrative parameter — and, separately from control, what does it cost in practice? Does it leave the energy market able to do its own job? Can demand response, storage and interconnection compete on their merits, and does the design fit a system whose adequacy increasingly comes from portfolios of small, duration-limited resources rather than from large plants? How exposed is it to the market power of the very plants deciding whether to close? How much administrative machinery does it take to run, and how many parameters can be wrong without anyone noticing? And can it be wound down when the problem it was built for has gone?

No type answers all six well. The grid below is a judgement, not a measurement — a scheme's national implementation can beat or miss its row — but the pattern it shows is real and it is the pattern that matters: every design buys adequacy by conceding something, and the concessions differ by family rather than by country.

Centralised capacity auction

The workhorse, and the one to choose when the gap is large, structural and needs steel that does not yet exist. Nothing else finances new build as reliably: a fifteen-year contract at a known £/kW, awarded in a descending clock auction four years ahead, is a bankable instrument in a way that no energy-market forecast has been since liberalisation. Great Britain, Poland, Italy and Belgium run versions of it, and France abandoned a decade of decentralised obligation to join them in 2026.

What it concedes is control. The demand curve is drawn administratively, so the price is set by the buyer's judgement about the reliability standard and Net CONE rather than by competition — as the construction above makes explicit, the auction's job is to land at Net CONE at the target, which means the target and the curve, not the bidders, determine the bill. It concedes the energy market too, and more subtly than the textbook admits: the mechanism does not cap scarcity prices, but once capacity revenue exists, the political tolerance for high energy prices falls further, regulators cap harder, and the energy market's own investment signal atrophies — which makes the next capacity auction more necessary than the last. And it concentrates market power precisely where the mechanism is most exposed, at the exit margin, where a supplier deciding between closing and bidding knows its own pivotality better than the monitor does. Its openness to demand response has been contested in court rather than in economics: the British scheme was suspended for almost a year after the EU General Court annulled its state-aid clearance in 2018 over exactly that question. De-rating factors, not eligibility rules, decide whether a battery is a competitor or a decoration.

Decentralised obligation

The most market-consistent design on paper and the least bankable in practice. Suppliers are obliged to hold certificates covering their customers' contribution to system stress; the state sets only the formula, and the market decides how much capacity gets built and at what price. Nothing is procured centrally, so no administrator has to guess Net CONE, and the obligation can be dialled down to zero without breaking any contract — the best exit route of any volume-based mechanism.

It failed on financing. A certificate market with a short horizon and thin liquidity produces a price signal that a project finance committee cannot use, and the developer who needs a fifteen-year revenue line gets a series of annual ones instead. France ran the model for a decade and has now replaced it with centralised T−4 and T−1 auctions, which is the strongest available evidence about the design: its principal implementation, in a large system, concluded that market consistency was worth less than bankability. That is a verdict on the instrument, not on France.

Reliability option

The best design in the family, and the page has already explained the mechanism; the evaluation is what it does to each of the six questions. It finances as well as a plain capacity auction, because the premium is the same fixed annual payment, and slightly better, because revenue is hedged in both directions and the investor's residual risk is smaller. It is the only mechanism that bounds what adequacy costs consumers in total rather than only what the capacity leg costs: pay the premium, receive the scarcity rent above the strike back. It protects the energy market better than anything except a reserve, because the spot price stays uncapped and continues to allocate dispatch — only the rent changes hands. And it is the only design with market-power mitigation built into its own arithmetic, since a generator withholding to manufacture a spike owes the payout on the price it just inflated.

What it adds is a parameter, and parameters are where mechanisms fail. The strike price has to be set: too low and it claws back rents that were legitimate scarcity signals for genuinely flexible plant, too high and the hedge is theoretical. The reference price has to be chosen, and it must be the price at which scarcity actually appears — which in several European systems is the imbalance price rather than the day-ahead price, so an option written on day-ahead can leave both sides mis-hedged. Non-delivery penalties have to be calibrated against the option payout, or the two obligations interact perversely. A country that cannot answer those three questions credibly is better off with a plain capacity auction it understands than an elegant one it does not.

Strategic reserve

A bridging device, and the right answer to a temporary, identified deficit: contract plant that would otherwise close, hold it outside the market, activate it only when the market fails to clear. It is the cheapest mechanism because it is the smallest, and the least distorting because it is not in the market at all. Germany, Sweden and Finland use it, and it is the design most compatible with the EU's insistence that mechanisms be temporary.

It is the wrong answer to a structural deficit, for two reasons. It finances almost nothing new — it pays for existing steel, which is the point, but it means the reserve cannot solve a gap that requires investment. And it is quietly corrosive at the edges: the activation threshold becomes a de facto price cap, because the market knows the reserve will appear before prices reach the levels that would justify entry, and every megawatt inside the reserve is a megawatt the energy market may never price again. Grow the reserve and the energy-only market shrinks; the slippery slope is not a rhetorical device but a description of the incentive facing the next plant deciding whether to compete or to be paid to stand still.

Administered capacity payment

Set a price per megawatt, pay it to whoever is available, and take whatever volume arrives. Spain, Portugal and Greece ran versions of it and have largely retired them. It fails the counterfactual test that defines the whole field: most of the money goes to plant that would have stayed anyway, which makes it a transfer rather than insurance. It has no volume control by construction, so consumer cost is unbounded in the one direction that matters. And with no auction, there is no competitive discipline on the level and no information produced about what adequacy actually costs.

Targeted tenders — paying only new entrants, or only identified at-risk plants — are the cheaper cousin and better in one respect, since something is competitively procured. But they discriminate inside the merit order, which distorts exactly the exit decisions the mechanism is trying to influence, and they reward the threat to close. State-aid discipline and Article 22 of the Electricity Regulation have pushed both designs out of favour, and the same article now conditions every mechanism in the Union: an adequacy concern identified in the European assessment, a temporary character, and since 1 July 2025 an emissions limit that excludes existing plant above 550 gCO₂/kWh from capacity payments altogether.

What it actually costs

The European bill is now large enough to be a policy object in its own right. ACER counted almost €11 billion spent across the Union in 2024 on some forty separate security-of-supply measures, €6.5 billion of it on capacity mechanisms — more than double the 2020 figure. Only 29% of that support went to low-emission technologies; the remaining 71% implicitly backed fossil generation, with gas expected to cover roughly 30% of peak demand through 2035 on contracts already signed. And capacity auction prices vary by more than a factor of ten across the Union for what is nominally the same product, which is the strongest available evidence that the number a capacity market produces is manufactured by the reliability standard and the demand curve rather than discovered from any underlying cost of adequacy.

Great Britain shows the same thing inside one country. The T−4 auction for 2028/29 cleared at £60/kW‑year across some 43 GW — on the order of £2.6 billion for a single delivery year — and the very next T−4, for 2029/30, cleared at £27/kW as the procurement target fell. Same market, same fleet, one year apart, less than half the price. Nothing about the physical adequacy of the British system changed by that much; the curve did.

Against which sits the most uncomfortable number in the field: across the EU in 2024, outages averaged under two hours per consumer and none was attributable to inadequate supply. That is either the mechanisms working exactly as intended or €6.5 billion a year buying a margin that was already there, and no amount of outturn data can separate the two. It is why the counterfactual test — would this capacity have existed anyway — matters more than any cost figure, and why a mechanism that cannot answer it is spending money it cannot justify.

Complexity is a design constraint, not a detail

Rank the five by administrative burden and the ranking inverts. A strategic reserve is a contract and an activation rule. An administered payment is a price and a payer. A centralised auction needs prequalification, de-rating tables by technology and duration, a demand curve calibrated on Net CONE, testing and penalty regimes, a secondary market and a monitor. A reliability option needs all of that, plus a strike, a reference price index and the settlement machinery to claw back against it. The designs that perform best are the ones that demand the most from the institution running them.

Complexity is not merely an administrative cost; it is failure surface. Every parameter is somewhere the mechanism can be wrong for years before anyone notices, and every one is a lobbying target with an identifiable beneficiary and a diffuse loser. The de-rating table is the clearest case: whether a one-hour battery counts as 20% or 40% firm is a technical judgement worth hundreds of millions, taken inside a methodology most participants cannot audit and few politicians could describe. So the question facing a government is not which design the literature prefers but which design its institutions can actually parameterise, monitor and defend. A badly struck reliability option is worse than a competently run capacity auction, and both are worse than an honest strategic reserve where the gap really is temporary.

Fit to a decentralised system

Every one of these instruments was designed to procure adequacy from a fleet of large dispatchable plants, and every one is now being asked to procure it from portfolios of small, duration-limited, weather- and behaviour-dependent resources. The mismatch shows up in three specific places, and they are worth separating because they have different fixes.

The product is availability across a defined window, which rewards anything that can sit still and wait. De-rating converts a duration-limited asset into a fraction of a firm megawatt, which is defensible in principle and is, in practice, where storage economics are decided by administrative judgement rather than by competition. The horizon suits a gas turbine: a four-year-ahead auction awarding fifteen-year contracts matches a thermal development timeline and fits poorly with aggregations that form, grow and dissolve on annual cycles. This is the one dimension where the decentralised obligation genuinely wins, because the obligation sits on suppliers, who aggregate by nature. And the penalty is sharpest under reliability options: a payback obligation is a good match for a plant that can run through a scarcity event and a poor one for a two-hour battery, which can owe the difference on hours it is physically unable to cover unless the option is written against a duration-matched volume. That is soluble in product design, but it is not solved by default.

The aggregate evidence is unflattering. Fewer than a third of the euros spent on European capacity support in 2024 reached low-emission technologies, and the long-term contracts already awarded will keep gas covering a large share of peak demand into the mid-2030s. Whatever these mechanisms are currently procuring, it is not a decentralised system — and because they procure years ahead and contract for years after that, the mismatch is not self-correcting. A mechanism that fits the fleet of 2015 will still be paying for it in 2035.

Choosing between them

The decision rule is shorter than the literature. If the gap is a known, temporary deficit during a specific transition, use a strategic reserve and write the sunset date into the instrument. If it is structural and needs new capacity, use a market-wide volume mechanism — and prefer a reliability option to a plain capacity payment wherever a credible reference price and strike can be defined, because it is the only version that caps the total bill and disarms withholding at the same time. If the binding constraint is political exposure to scarcity prices rather than adequacy itself, that preference becomes decisive. If the state wants minimal administrative discretion and can accept that less will be built, a decentralised obligation is coherent, provided nobody pretends it will finance a new fleet. And never run an administered payment without a volume control, because it is the one design that reliably pays for what would have happened anyway.

Two caveats sit above all of it. The mechanism is a second-best instrument for a first-best failure — an energy market that cannot price scarcity credibly — so the question of whether the mechanism is shrinking as scarcity pricing, storage and demand flexibility mature is not a technicality but the test of whether the design was insurance or subsidy. And the choice of type matters less than the parameters inside it: a well-parameterised capacity auction beats a badly struck reliability option, and a de-rating table decides more about who competes than any eligibility rule in the statute. Add the three criteria this section has introduced and the rule tightens once more: choose the most sophisticated design your institutions can genuinely parameterise and audit, cost it against the counterfactual rather than against the reliability standard, and check that its product, horizon and penalty fit the resources you expect to be buying in ten years rather than the ones you bought in the last decade.