The Suppressed Signal

28 July 2026 · [M] measured · [C] computed · [P] public accounts

On Sunday, the 11th of May 2025, at one o’clock in the afternoon, the German day-ahead electricity price reached −250.32 euros per megawatt-hour. Not zero. Minus two hundred and fifty. Producers were paying to put power onto the grid, and Germany’s neighbours were being paid — at moments up to 480 euros per megawatt-hour — to carry the surplus away.

The price was doing its job. It was screaming a single instruction with maximal clarity: stop producing, store this, or move consumption here, now. The price is the visible part. What is suppressed is the response — because at that moment German solar plants kept producing at full output, and kept being paid, in many cases hundreds of euros per megawatt-hour, for electricity whose market value was deeply negative. The signal fired. The design had disconnected the wire that carries it to the people who could act on it.

This essay asks a simple question with an uncomfortable answer: what does it cost, per year, to pay producers to ignore the market’s clearest signal? The question began, for me, as a back-of-envelope remark from my father about that day’s prices. Answering it properly meant computing the number from raw public data — day-ahead prices and actual generation, nothing modelled — and the answer is roughly two billion euros a year. But the number is the smaller finding. The larger one is what its composition reveals: the legislator has spent a decade repairing this exact rule, and the bill is almost entirely untouched by the repairs, because it lives in a fleet the repairs were written not to reach.

One clarification before anything else. This is not an argument against renewable energy. The economics of solar photovoltaics are among the great industrial achievements of this century, and a system built on cheap solar plus storage is very plausibly the lowest-cost system available to us. The argument is narrower: a subsidy design that decouples remuneration from price does not merely waste money. It suppresses the exact signal that would summon the storage, the flexibility, and the grid the transition needs — and so perpetuates the condition it pays for.

I. Why a price goes below zero

Wholesale markets clear by merit order: the last plant needed to meet demand sets the price for all. Wind and solar have marginal costs near zero — no fuel — so on a sunny, windy midday, when they alone can cover demand, the clearing price collapses toward zero.

It goes below zero because some producers will pay to keep generating. This is not madness; it is arithmetic. A plant receiving a guaranteed premium of sixty euros per megawatt-hour still profits at a market price of minus fifty, and will rationally bid down to minus sixty, where the subsidy is exhausted. Inflexible plants — waste incinerators, biomass on support contracts, thermal units avoiding costly restart cycles — join them for reasons of their own. The price stops falling only when enough give way.

So a negative price is not a malfunction; it is the market working — a surplus signal of maximal clarity — and its depth and frequency measure how many producers have been rendered deaf to it. In Germany the count has become a curve worth staring at: 139 negative-price hours in 2021; 301 in 2023; around 457 in 2024; 576 in 2025 (public reporting: 573). A fourfold rise in four years, tracking the photovoltaic build-out — roughly 100 gigawatts installed by end-2025, heading toward 215 by 2030. The surplus signal fires more and more often. The question is who is listening.

The three-hour disagreement with the public count is worth one sentence, because it is measurable rather than a matter of trust: of the 576 hours, four clear zero by less than a cent and 95 by less than five, with no natural cliff anywhere near the boundary — so a three-hour difference is the same order as the arbitrariness of the threshold itself, and nothing in what follows turns on it. (That is not an hour-by-hour reconciliation of 576 against 573, and is not offered as one: subtracting exactly the sub-half-cent hours would land on 573, but choosing a threshold because it produces the answer you want is precisely the move this essay is about.)

II. The number, and where it lives

I wanted one specific figure: how much EEG remuneration was paid to solar producers during the hours when the price was negative — hours in which every additional megawatt-hour had negative value to the system.

The computation uses only two raw public series — ENTSO-E day-ahead prices and actual generation for DE/LU across all of 2025 — and no model. In those 576 hours (the deepest, −250.32 €/MWh, reproducing the public record’s value and timestamp exactly), solar produced 17.94 terawatt-hours. Weighting that energy by the vintage structure of the German PV fleet and by the suspension rules that apply to each vintage — with the one modelled assumption, the vintage-weighted remuneration, carried as a pre-registered band rather than a point estimate — the central estimate is about 2.1 billion euros, with an honest range of 1.3 to 3.1 billion. That is roughly one-eighth of the entire federal EEG subsidy for the year, paid for production during hours when the production destroyed value. Separately, the system spent about 0.4 billion euros disposing of the negative-priced surplus: the producer was paid, and the system paid to get rid of the power.

The full method, sources, and pre-registered band are in the replication file that accompanies this essay; every figure here is reproducible from the two public series by a script that never touches my simulation.

But the composition of the 2.1 billion is the finding, because it overturns the intuitive diagnosis. The intuitive story is “too much new solar.” The data says otherwise. The pre-2016 vintages — roughly a third to two-fifths of installed capacity — carry guaranteed tariffs of 300 to 570 euros per megawatt-hour, negotiated when solar was expensive, and they face no suspension during negative hours. The vast new fleet earns 58 to 90 euros and loses most of its negative-hour remuneration to the suspension rules. Under the pre-registered central assumptions, the pre-2016 fleet carries about 86 percent of the bill — and evaluated at the same low and high corners, between 68 and 94 percent. The hours are made by the new fleet’s volume; the bill is made by the old fleet’s contracts.

Two known biases both push that central share down, and I state their direction rather than re-tune the parameters. First, I use the Bestand’s capacity share as a proxy for its energy share, but pre-2016 panels have degraded at roughly half a percent a year and now yield some 5 to 7 percent less per installed watt than their nameplate implies — so their true energy share, and thus their share of the bill, is lower than the capacity figure. Second, I proxy the market-premium regime by vintage, but EEG 2014 already made direct marketing mandatory above 500 kilowatts from August 2014 and above 100 from 2016 — so a slice of the pre-2016 stock is already in the §51 regime and does lose payment in negative hours, again trimming the old fleet’s share. Both corrections point the same way; neither disturbs the qualitative split. Two different problems wear one headline — and, as we shall see, the law’s decade of repairs was aimed at the first and written to spare the second.

III. The law’s own repairs — and the fleet they were written to miss

Begin with the statute as it now reads, because it has already conceded everything. §51 Abs. 1 in its current form carries no consecutive-hours threshold at all — “Für Zeiträume, in denen der Spotmarktpreis negativ ist, verringert sich der anzulegende Wert auf null.” For any period in which the spot price is negative, the support drops to zero, from the very first negative quarter-hour. The old rule of “six consecutive hours,” then four, then three, has not merely been loosened; it has been deleted from the core of the statute and survives only in the grandfathering provision, §100. On its own terms the law is now maximally strict. And that is precisely why the entire remaining bill lives, by legislative design, in the stock the grandfathering shields.

The path to that strictness is a decade of admissions. The negative-price suspension has been tightened relentlessly: six consecutive negative hours before remuneration stops (EEG 2014), then four (EEG 2021), then a phased walk-down under EEG 2023 — four hours in 2023, three in 2024–25, two in 2026, one from 2027 — and finally the Solarspitzengesetz of February 2025 (BGBl. 2025 I Nr. 51, in force 25 February 2025), which rewrote §51 to the first-quarter-hour rule above. Each tightening is an admission that the signal matters and that paying through it is wrong.

But the reform guts its own logic twice, and the statute is explicit about how. First, the suspended periods are not cancelled. Under §51a the zeroed quarter-hours are appended to the end of the twenty-year term as a contingent paid out afterward — and for solar the deferral carries a ×0.5 factor, so the extended portion is halved. A real cut, then, but a prospective one, reaching only plants already exposed to the first-hour rule.

Second, and decisively: that rule applies only to new installations — commissioning or tender date on or after 25 February 2025 (§100 Abs. 46 EEG). Every earlier vintage keeps its old, looser regime; existing plants may opt in only against a sweetener of +0.6 ct/kWh. So the hundred-plus gigawatts of legacy fleet — the pre-2016 installations with the 300-to-570-euro tariffs that carry 86 percent of the 2.1 billion — remain, by the statute’s own architecture, untouched. Paid in full at minus two hundred and fifty as at plus fifty.

The reform thus aims at the plants that are not the problem and exempts the plants that are. The marginal new producer now hears the signal; the marginal cost producer still does not. This is not a loophole; it is a choice, and a defensible one — you do not rewrite twenty-year contracts lightly. But it means the headline number is structurally insulated from the reforms, and will stay roughly where it is until the legacy contracts roll off the twenty-year cliff of their own accord. The suppression persists not despite the repairs but because the repairs were built around the stock that causes it.

IV. Ordnung and Prozess — the principle at stake

The frame is older than the energy transition. Walter Eucken, the central figure of the Freiburg school of ordoliberalism, drew a distinction worth rescuing from the seminar room: the state’s task is the Rahmenordnung — the framework of rules within which the market process runs — and emphatically not the steering of the Prozess itself. Plan the forms; never steer the process. Prices are the process’s nervous system, and a state that guarantees outcomes severs the nerves.

Guaranteed remuneration decoupled from price is a textbook case of what Eucken called interventionism: an outcome-steering measure that mutes the signal and whose costs compound in ways that summon further intervention. Consider what the muted signal would otherwise do. A negative midday price is precisely the invitation to build storage — the noon-to-evening spread is the battery business case — to shift industrial load, to site electrolysis, to build transmission where surplus is trapped. By guaranteeing producers against the negative price, the design defunds the very arbitrage that would absorb the surplus. Note too the perverse feedback: as more solar depresses midday prices, the market value of EEG power falls, so the per-unit subsidy — the gap between promise and market — rises. The state pays more precisely because the power is worth less.

There is a constructive version. The honest unit of renewable investment in a high-solar system is not the panel but the panel plus its firming — and forced to face the price, solar-plus-battery outcompetes naked solar exactly where the midday glut is worst, because the price structure pays it to. The claim is falsifiable: if remuneration stays decoupled, negative hours and the subsidy line should keep rising superlinearly with the build-out; if instead Germany’s battery build-out absorbs the midday surplus under current rules within two or three years, the mechanism I describe is weaker than I believe. Watch the 2026 and 2027 data.

V. The mechanism, corroborated

Is the causal story even right — or are the negative prices weather, or demand, or European coupling, and the subsidy design incidental? The headline numbers do not depend on the answer; they are measured from real prices. But I can offer an independent second route. I maintain a simulation of the European power market with the institutional rulebook represented explicitly and separately from the market mechanism. Give it the remuneration rules — tariff tables by vintage, suspension regimes, the must-take status of supported plants — and withhold the prices entirely. If the design story is right, it should generate Germany’s pathology on its own.

It does. Fed only the rules, the simulation produces negative-price hours with the correct fingerprint: concentrated in the midday solar peak, clustered in the sunny months, with the characteristic sawtooth in which deep runs trip the suspension rules and prices recover toward zero. With the full supported fleet represented — solar, wind, biomass, waste, and the cycling of thermal plants — it produces 537 hours against the real 576, peaking at the same hour of day, and its endogenous estimate of premium paid during negative hours lands in the same band as the model-free figure. Two methods — one from data, one from causes — arriving at the same place. The pathology is not weather. It is design. (The simulation is a corroborating exhibit; the numbers this essay stands on are the model-free ones of §II.)

VI. The strongest objections, taken seriously

“The premium is recoverable, so the number overstates the loss.” In the spirit of the Lucas critique: much of what looks like waste is transfer, not deadweight — the producer keeps the euro the taxpayer sends. Correct, and important: the 2.1 billion is not 2.1 billion of savings available to a reformer. It is the measured size of a distortion — the wedge between what the system values a megawatt-hour at (negative) and what the producer is paid to make it (positive). The deadweight is the second-order consequence — the storage not built, the load not shifted — and it is larger than the transfer, not smaller. The number is a thermometer, not an invoice.

“Two billion is a rounding error, and who even pays it?” Since 1 July 2022 the EEG-Umlage that once sat on every consumer’s bill has been abolished (permanently, via the EnFG); the differential is now an explicit line in the federal budget, about 16.5 billion euros in 2025, of which our hours are one-eighth. That the money is on-budget is exactly why the diagnostic matters: it is visible, appropriated, and defensible only if the design that generates it is. Fiscal smallness is not the point; the number grows superlinearly with the build-out, and it is the visible tip of the flexibility that was not financed.

“Solar isn’t even the sole cause of the hours.” Conceded, and the data forces it. In the negative hours, inflexible must-take biomass and waste supply about 7 percent of generation, and the near-zero residual-load band is tipped over the edge by must-run lignite and CHP that stay on for their own reasons. The negative hours are a whole-fleet phenomenon; my own simulation cannot reach 576 on solar-plus-wind alone. But this broadens the authorship of the hours; it does not touch the authorship of the bill, which §II located overwhelmingly in the grandfathered solar Bestand. Different question, same answer: the design that pays through the signal is the common root.

“The real failure is missing flexibility — de-risking was the point.” This is the strongest defence, and substantially correct. Fixed remuneration was not an accident; it was de-risking by design — twenty-year certainty is what financed the build-out at all, and negative prices reflect a system that built generation faster than the storage, grids, and demand response to absorb it. The premise holds; the conclusion does not follow. The flexibility must indeed be built — and the price signal is precisely the mechanism that finances it. Storage is the flexibility, and it gets built exactly when and where investors face the spread. Suppressing the signal is a standing subsidy against its own business case. De-risking generation and de-risking flexibility are not the same policy — and the second is the one the price was about to pay for. Which points at what a production-neutral reform would do: keep the certainty that finances capacity, but attach it to availability and delivery rather than to undifferentiated output, so the contract no longer rewards a megawatt-hour the system is paying to refuse.

VII. What restoring the signal looks like

Not “abolish support.” The ordoliberal position is more precise: move support from outcome guarantees to rule design. Full market exposure for new capacity — conceded in principle by the 2025 law; make it real by removing the deferral mechanics that preserve support volume through the back door. Let the spread stand: the midday-to-evening differential is the storage industry’s revenue, and capping it away is a tax on flexibility. Make the signal locationally honest: a single national price zone hides where surplus and scarcity sit, and smaller zones — the direction of the European bidding-zone review — turn congestion into a visible investment case for grids and storage at the right nodes. And for the legacy fleet: respect the contracts, stop growing the exempt stock, and keep the annual cost of the exemption on the visible budget line. The claim beneath the particulars: a market allowed to send honest signals assembles its own least-cost system. The signal is not a problem to be managed; it is the mechanism that solves the problem.

VIII. The bridge: one rule’s cost, and the general case

The two billion euros is the measured cost of one rule — a single remuneration mechanism, in a single country, for a single technology. Its interest is that it is a clean, isolable instance of a general proposition: that a nation’s electricity outcomes are shadows cast by its rulebook, traceable to their causes with discipline about what is mechanism, what is world, and what is rule.

The next piece in this sequence makes the general case directly. It decomposes Germany’s delivered-price gap against a low-cost benchmark — why industrial and data-centre power costs so much more here — into an institutional part (marginal-pricing rent, grid fees, levies — the residue of outcome-steering rules like this one) and a structural part (the physical cost of firming a variable resource through a northern winter). Measured on the same engine, the split is roughly 84 percent institutional, 16 percent structural: most of the gap is rulebook, closeable by reform, and only a residual firming premium of around 47 dollars per megawatt-hour for round-the-clock load is a fact of German geography that survives every reform. The negative-hour subsidy is one institutional item in that 84 percent.

A final reason to trust the method: the same engine — mechanism frozen, not a line changed — was pointed at North America with American fleets, demand, and rulebook (production tax credits, Texas scarcity pricing, California and Northeast carbon prices). It produced California’s midday “duck curve” and — the detail I find most persuasive — the opposite pattern in the wind-dominated Great Plains, where negative prices cluster at night, when the wind blows, not at noon, when the German sun shines. Same physics, same mechanism; different rules, different fingerprints, each correct. Market pathologies are not weather; they are the shadows of rulebooks. The state’s task, Eucken said, is to keep the framework honest and let the process speak. The process is speaking at minus two hundred and fifty euros an hour. The framework should stop paying us not to hear it.

Sources & Method

Every figure is labelled [M] measured (raw public data), [C] computed (a rule applied to measured data, with stated assumptions), [P] public accounts (official published totals), or [S] simulated (the corroborating model, §V — never the source of a headline number).

Full legal citations with verbatim German text and retrieval dates: docs/analysis/eeg_legal_ground_truth.md. Replication script + pre-registered band: scripts/analysis/replication_suppressed_signal.py → docs/analysis/REPLICATION_suppressed_signal.md. Corroborating simulation: docs/analysis/EEG_NEGATIVE_HOURS_2025.md. The argument is offered in the spirit of its own method: falsifiable. Corrections welcome.