Cancelling the €600bn the Eurosystem holds is not a debt write-off. It is a bet on the term premium, and it is priced.
Jean-Luc Mélenchon says it would cost no one anything. Matthieu Pigasse says it has no economic or financial impact. Olivier Blanchard says it is simply false. All three are arguing about the coupon, which is worth €6.5bn a year and is almost entirely offset by the central-bank income the State stops receiving. The tribunes that started this in 2020 were arguing about something else — the principal — and that is where the money is. This page prices both, and shows every number.
Why cancelling the debt is not the simple, free operation it sounds like
“The ECB is holding €600bn of French debt. Cancel it — nobody is out of pocket.” Said quickly, that sounds obviously true. It is not, and the reason is that the debt did not stay where it started.
Where the €600bn actually is
Four ordinary steps. The one that matters is the last: what the bank is left holding.
Quantitative easing did not destroy the debt, it swapped it. Where the State once owed a bank for ten years at a fixed rate, it now owes its own central bank — and the bank holds a deposit at that central bank instead. That deposit is a reserve, and it earns the ECB’s policy rate, 2.25% today, paid every year for as long as it exists. This is the part the slogan skips: cancelling the bond does not cancel the reserve. The liability stays, and so does its interest bill.
Three things you can actually do
Every figure here is the model’s own, at its default settings. The levers in §02 let you move them.
None of that means the idea is stupid. There is a serious argument for cancellation, and it is not about the interest at all — it is that the principal would never have to be repaid or refinanced. This page prices that too: worth roughly €6.3bn a year, and bought with permanent interest-rate risk the State can never exit. It is a trade with a price, not a free lunch — which is why the rest of this page has levers instead of a headline.
01
What is actually on the table
Three different operations are being discussed as if they were one. They are not the same legally, and only one of them changes the published debt ratio — but economically all three are identical.
The figure decoder
Every number in this debate — €469bn, €600bn, €640bn, €700bn — is defensible under some perimeter and wrong under the others. The denominators are being swapped mid-argument, and the €469bn is the clearest case: it is 16.3% of the €2,881.91bn of negotiable State debt, not the 18% usually attached to it, which comes from a €2,613bn denominator that is not the negotiable total. Here is what each one actually counts.
The stock is falling on its own.Eurosystem holdings of French public-sector paper peaked at €839bn in January 2023 and stand at €600bn today — quantitative tightening has already retired €239bn of it, at roughly €68bn a year, without anyone voting on anything. On the current run-off the portfolio is largely gone by the mid-2030s. Cancellation is a proposal to stop that unwind, not to remove a debt that would otherwise persist.
Eurosystem holdings of French public-sector securities
€ billion, amortised cost, monthly. PSPP from March 2015, PEPP from March 2020. APP reinvestments ended July 2023; PEPP reinvestments ended December 2024.
The French State owns 100% of the Banque de France, so there are two books to keep, and the debate is about whether you may look at only one. There are also two periods. Phase 1 runs while the cancelled bonds would still have been outstanding — about six years, since the portfolio's weighted average maturity is 6.4 years and nothing is reinvested. Phase 2 is everything after that, and it is where the case actually lives. Move the levers; every number recomputes.
Propagation of a cancellation, per year
€ billion a year, phase 1, years 1–6, effect on the French Treasury. Each bar is a step; the final bar is what is left.
Adds to the TreasuryTakes from the TreasuryNet
How much spread widening erases the gain
€ billion a year to the Treasury, against a permanent widening of France's funding spread applied to the debt that is not cancelled. The gold line is the arithmetic used in public. The teal line adds the Banque de France offset.
Coupon only, no offsetFull accounting
03
Where the money actually goes
Nothing is destroyed by a write-off and nothing is created. A claim moves from one balance sheet to another, and the only question worth arguing about is whose. These cards track the central case as you left it above.
The circuit
Follow the euro. The Treasury stops paying a coupon to the Banque de France; the Banque de France stops remitting it back to the Treasury. The loop closes on itself — except for the slice on the ECB's own books, whose income is pooled across the Eurosystem and shared by capital key. One mechanical detail worth knowing: for pooling purposes the Eurosystem values these securities at the main refinancing rate, currently 2.40%, not at their actual coupon. The Banque de France is therefore already contributing to the pool at 2.40% on bonds yielding about 1.08% — a drag of roughly €7bn a year that explains much of its recent losses, and that a write-off would make permanent.
The residual is not a saving. It is a transfer.Whatever is left after the Banque de France offset comes from one place only: the 10.5% of the portfolio that sits on the ECB's own books, whose losses are shared across the Eurosystem by capital key. France's key is 19.77%, so France keeps a fifth of that slice and the other member states absorb four fifths. That is the entire net gain — a one-off €50bn and about half a billion a year, taken from German, Italian, Spanish and Dutch taxpayers. It is not a free lunch. It is somebody else's lunch, and they have a veto.
04
The reserve trap
The one lever that would make cancellation genuinely pay is to stop remunerating bank reserves. It is the missing half of the proposal — and for France specifically, it goes the wrong way.
Quantitative easing did not extinguish the debt. It swapped long-dated fixed-rate bonds for overnight central-bank reserves, remunerated at the deposit facility rate. Write off the bond and the reserve is still there, still costing 2.25% a year. So the only route to a real cash saving is to stop paying it — which every serious participant, including its proponents, concedes is a tax on banks. Andrew Bailey put it most sharply: "removing remuneration on reserves is akin to a tax on banks. It is only appropriate that such a tax be imposed by the elected Government of the day."
Here is the part nobody in the French debate has run. Reserve remuneration is set by the Governing Council and applies uniformly across the euro area; a member state cannot tax only its own banks. And French banks hold 21.4% of the remunerated euro-area reserve base while France's capital key is 19.77%. (Minimum reserves have earned 0% since September 2023, so they are already outside the base.) Under Article 32.5 of the ESCB Statute monetary income is pooled and redistributed by that key. So France pays more of the tax than it collects of the proceeds.
Ending reserve remuneration: what France pays, what France gets
€ billion a year at the current deposit facility rate of 2.25%, against the share of euro-area reserves left unremunerated. No base erosion assumed, so these are upper bounds on both sides.
Paid by French banksReceived by the French TreasuryNet for France
Who actually pays it
The evidence on incidence is one-sided and it does not favour the proposal. Gray (IMF, 2011) frames it as a ratio rather than an addition: with a share RR of deposits parked in non-interest-bearing balances, the ratio between credit and deposit rates has to be roughly 1/(1−RR), so the wedge widens with the requirement. McCauley and Pinter's reading of the eurodollar natural experiment is blunter: mobile, wealthy and corporate depositors relocate, "leaving smaller, less wealthy depositors to pay the tax." The zero lower bound that protected French retail savers in 2014–19 is not binding at a positive policy rate, so the household channel that was shut then is open now. No study found supports the view that shareholders bear it.
The objection that is weaker than claimed
"It would break monetary control" is decisive against blanket unremuneration and largely a straw man against tiering. The ECB ran a two-tier system from October 2019 to September 2022 with about €855bn exempted at a 50bp differential, and the Banque de France's own study found it had not led to a sustained rise in money-market rates. What ended tiering was the deposit rate turning positive, not any finding that it damaged transmission. But the precedent is weaker than it looks and we should say so: that tier exempted reserves from a negative rate, so it was a subsidy. Ending remuneration at a positive rate is a tax, and the arbitrage runs the other way — banks try to shed reserves and lend below the policy rate. The honest objection is anyway legal, not technical: a reserve decision taken for a stated fiscal purpose is monetary financing by another route.
05
What QE already did to French debt
The OBR publishes this calculation for the United Kingdom every year. Nobody publishes it for France, so we have derived it — and it reframes the whole question.
France's negotiable debt has an average maturity of 8 years and 5 months, one of the longest in the G7, and the Agence France Trésor is rightly proud of it. But that figure counts the €600bn the Eurosystem holds as though it were ordinary long-term debt. It is not: it has been paid for with overnight reserves. Consolidate the Treasury and the Banque de France, take the bonds out and put the reserves in, and the picture changes.
Average maturity of French public debt
Years. Headline as published, versus consolidating the Banque de France.
Share repricing within one year
Per cent of the debt stock. Short bills alone, versus bills plus central-bank reserves.
The proposal is a duration decision wearing a debt-relief costume.QE was a €600bn fixed-for-floating swap that France has already entered into. Quantitative tightening is unwinding it: as the bonds mature, the Treasury refinances at market rates and the reserves are extinguished. Cancellation stops the unwind and makes the swap permanent. That is worth something — the difference between what France pays on a ten-year OAT and what it pays on an overnight reserve — but it is a term-premium harvest, not a debt reduction. The obvious retort — that France could have the same thing today by issuing bills — is only half right, and worth stating carefully. €600bn of bills must be re-placed with private buyers every few months at whatever spread they then demand; central-bank reserves are a closed system whose aggregate quantity the holders cannot refuse. So cancellation buys a genuinely better version of short funding. What it does not buy is an escape from the risk that short funding is priced for: a 200bp move in the deposit rate costs €10bn a year on the same stock, and there is no maturity date at which the exposure ends. The Agence France Trésor keeps the maturity long on purpose. This proposal shortens it, permanently, and calls the compensation a saving.
06
Is the central bank then insolvent?
This is the question the Vandeweyer–Dufrêne exchange of 24–28 August turns on, and both men argue it qualitatively. It is a quantity, and it is computable from published figures. One caution before the names: the thread is public and section 07 links two of its posts, but we could not fetch it to check the wording, so the quotations are reported rather than verified — see the provenance note there.
Quentin Vandeweyer (London Business School, previously Chicago Booth and the ECB's research directorate) argues that once the asset is written off the central bank has three options and two of them are inflationary: keep remunerating every reserve and the stock snowballs at the policy rate with nothing to pay it, which is Ricardo Reis's Ponzi definition of central-bank insolvency; stop remunerating and the policy rate falls below the natural rate, which is Wicksellian inflation; or tier, which preserves rate control but is a tax on banks that recapitalises the central bank by the back door. Nicolas Dufrêne replies that the snowball already exists without any cancellation, because reserve interest is already paid by crediting reserve accounts.
Reis distinguishes three insolvencies — period, rule and intertemporal. The operative one here is the third: a central bank is solvent while its net worth plus the present value of its seigniorage is non-negative. So there are two tests, a flow test and a stock test, and they give different answers.
The flow test: does the reserve stock outgrow the economy?
Banque de France liabilities to credit institutions, % of French GDP, projected 30 years. The gap between what it pays on reserves and what its remaining portfolio earns is funded by creating more reserves.
At today's 2.25%At 2023–24's 4.00%
The stock test: solvency against the discount rate
Net worth after the write-off, plus the present value of banknote seigniorage, € billion. Solvent above the line, insolvent below.
Net worth + PV of seigniorage
Neither side is right, and the reason is interesting.On the flow, Dufrêne wins: at a 2.25% deposit rate against 2.79% nominal growth, the reserve stock shrinks relative to the economy with or without a write-off. Nothing explodes. But cancellation lowers the policy rate at which it starts compounding faster than the economy by 0.72 points. The level of that threshold depends on where you draw the balance sheet and the difference does not: excluding the Banque de France's TARGET2 liability it moves from 3.74% to 3.02%, including it from 2.81% to 2.09%, and the gap is 0.72 points on either perimeter. The difference is the number to quote; the level is a perimeter choice. The deposit rate stood at 4.00% for nearly nine months in 2023–24, and was above the 3.02% threshold as recently as December 2024. Vandeweyer's mechanism is real; it is simply not live at today's rates. On the stock, neither wins: the write-off takes Banque de France net worth from +€283bn to −€253bn, and solvency then rests entirely on the present value of banknote seigniorage. That breaks even at a net discount rate of 2.64%. Below it, solvent; above it, insolvent. Whether the Banque de France is intertemporally solvent after this operation is not a fact — it is a choice of discount rate, which is exactly what Reis warned about when he showed the same seigniorage flow is worth 5% or 94% of GDP depending on the rate you pick.
And the seigniorage arrives exactly when it is least useful.The flow that is supposed to rescue the balance sheet is the interest saved by funding assets with banknotes, which pay nothing. It is therefore proportional to the policy rate: €6.7bn a year at 2.25%, and zero at the zero bound. So it vanishes in precisely the deflationary states where a central bank needs capital, and it is largest in the inflationary states where it must raise rates — which is also when the reserve bill explodes. The two move together. That is why the knife-edge is sharper than either party allows, and it is the strongest technical argument against the operation that neither of them makes.
07
The exchange, line by line
Every statement below is scored against the model on this page and against the primary sources. Green means the claim survives checking, orange means it is half-right or right under conditions it does not state, red means it does not survive. Nobody's side wins uniformly, which is the point.
Survives checkingHalf-right, or right only under unstated conditionsDoes not survive
Provenance — read this before the verdicts. The thread is public and you can read it yourself: the opening post of 24 August and a later instalment of 28 August. What we could not do is fetch it: X blocks automated access, so the wording quoted below reached us as pasted text and we have not been able to check it character by character against the posts. Treat the quotations as reported rather than verified, and follow the links if a verdict matters to you. For the Dufrêne replies we have no permalink at all. Two consequences we have acted on. Passages we hold in substance but not in wording are marked reconstructed and are not presented as quotations. And the one red verdict that rested on this exchange — the Article 32.4 argument — has been moved onto the Le Monde tribune of 26 May 2020, which makes the same argument, is public, and is linked in the sources, so the red mark rests on a document anyone can read rather than on a sentence only we have seen. The page's other red verdict, on the claim that inflation is harmless provided wages keep up, does not come from this thread at all: it is a recurring claim from the wider debate, scored on INSEE and ECB statistics, and it carries no provenance problem.
08
“Inflation is fine if wages keep up”
The most dangerous sentence in the debate, because the first half is testable and false, and the second half describes a loop that runs in both directions. Both halves are computed here.
The claim comes in two versions. The soft one says inflation is a tolerable price for erasing debt provided wages follow. The hard one says inflation is the erasure — that a State with its own currency never repays, it inflates. Neither is stupid, and the second is closer to how public debt has actually been retired historically. But both leave out the return leg, and in the French configuration the return leg is large, published and easy to find.
Start with the wages, because that half is not a model at all — it is a measurement. France ran the experiment between 2021 and 2024.
What actually happened to French real wages
Real change in the average net private-sector wage per full-time equivalent, per cent a year. INSEE.
Wages did not keep up, and the shortfall is measurable.INSEE puts the real average net private-sector wage at −1.3% in 2022 and −1.0% in 2023, recovering +0.8% in 2024 — leaving purchasing power barely back at its 2019 level after five years, against an average gain of +0.6% a year over 1996–2019. Divide the shortfall by the excess inflation and the implied pass-through is 0.59 in 2022 and 0.66 in 2023. That is the number this page uses as its default, and it is not an assumption: it is what France did. In fairness to the other side, the IMF's study of 79 wage-price episodes — the first identified in 1973 and the last in 2017; the 1960s belong to its separate 100-episode manufacturing sample — finds that such episodes rarely turn into sustained spirals, and it is right about that. Across all of them it finds real wage growth broadly unchanged afterwards, which means the level loss is not recovered so much as levelled off from a lower base. The strongest form of the other side is stronger than that, and it deserves to be stated rather than dodged: in the sub-sample that matches recent conditions — falling real wages with tight labour markets, which is precisely the 2021–23 configuration — the same paper finds that "similar past episodes were followed by a period of declining inflation while nominal wage growth increased thus allowing real wages to catch up." That is the best available case for "wages keep up", and it is a serious one. Our answer is that France ran that experiment and the French series is the test: real wages did not regain their 2019 level until 2024, and INSEE's own verdict on the rebound is that « Ce regain en 2024 permet à peine d'atteindre le niveau de 2019 en euros constants ». Catching up after five years is a different claim from keeping up, and only the second one makes the inflation costless.
And an average that did keep up would still not make it neutral.Inflation is not a single rate. At the October 2022 peak the ECB measures euro-area low-income households facing an effective inflation rate 0.55 points higher than affluent households, because energy and food weigh more in their basket; by November 2023 the gap had reversed as services inflation took over. Over 2022 as a whole the ECB puts the cost of inflation at 12.28% of current income for the lowest quintile against 5.69% for the highest. A pass-through of 1.0 on the average is consistent with a large transfer inside the average. Anyone who says wages will keep up is making a claim about the mean and calling it a claim about everyone.
Now the debt. This is the half that is usually asserted rather than computed, and it has three loops.
The three loops
One arrow reduces the ratio. Three return it. Figures are for one point of extra inflation.
Debt ratio, with and without the inflation shock
Per cent of GDP. The gap is the whole of the argument: positive means inflation helped.
No shockWith the shock
The gap, year by year
Points of GDP the inflation is worth. Above the line it is helping; below it, the loops have taken the gain back.
On a like-for-like perimeter, the published rate sensitivity understates the true one by a factor of about 2.4.The Senate's budget report gives France's exposure to a permanent 100 basis-point rise as €3.2bn after one year, €23.5bn after five and €33.5bn after nine. Those are the right figures for the perimeter they use — the State's own negotiable debt, where 7.6% reprices within a year. On the consolidated perimeter, where the Eurosystem's €600bn is correctly counted as the overnight reserves that paid for it, 26.1% reprices within a year, so the first-year cost is €7.27bn. Which multiple that is depends on what it is set against. Like for like — the same module run on the State's own perimeter, which returns €3.04bn — it is 2.4×. Against the Senate's published €3.2bn, built on a different perimeter by a different method, it is 2.3×. The like-for-like ratio is the one to quote. This is the same finding as section 05, arriving from the other direction, and it is what makes the inflation argument fragile: the State is far more exposed to the rate response than its own published sensitivity shows. Cancellation would make that exposure permanent.
The configuration in which the claim is true — and why nobody can choose it.Set the central-bank response to zero and the wage pass-through to 100% and the model agrees with the claim exactly: the debt is inflated away and wage-earners are held whole. That is not a trick, it is the claim stated precisely, and it is worth stating precisely because three things then follow that the model cannot draw for you. First, in that configuration the inflation does not stay where you put it: full indexation with no policy response is the one setting in which the wage-price mechanism the IMF finds rare becomes likely, because it is the setting its 79 episodes did not have. Second, the bond market prices the reaction function rather than the current rate, so a central bank known not to respond is paid for in term premium — which is the spread lever in section 04, and 28 basis points is all it takes. Third and decisively, France does not own the dial. The response coefficient is set in Frankfurt for twenty countries under a price-stability mandate written into Article 127(1) TFEU. Every version of “inflation will handle it” is a plan that requires the ECB to abandon its primary objective, which is the same institutional obstacle the cancellation proposal runs into — arriving, once again, from the other direction.
What the model is honest about.It holds the primary balance and the stock constant, so new deficits are not issued at the higher rate. That is a deliberate understatement: on the State's own perimeter the module reproduces the Senate's one-year sensitivity almost exactly (€3.04bn against €3.2bn) but falls short at five and nine years (€15.2bn and €24.4bn against €23.5bn and €33.5bn), and the gap is precisely the deficit-financed growth of the stock. It also assumes the indexation uplift accrues at the headline rate and stays in the principal, which is how the OATi and OAT€i contracts work — on the €311bn of indexed principal the AFT publishes, not on the smaller nominal sum of the fiches titres. The index-linked block reprices at (taylor−1)·π rather than taylor·π, because those bonds pay a real coupon: charging them the full nominal move on top of the indexation uplift would count the same inflation twice, and an earlier version of this module did. And it says nothing about output: an inflation that raised real growth would help through the denominator in a way not modelled here.
09
The arguments, scored
Both camps carry claims that do not survive contact with the balance sheet. Sorted by who makes them, not by which conclusion they support.
For cancellationAgainst cancellationAll
Survives checkingHalf-right, or right only under unstated conditionsDoes not survive
10
Method, and what we are unsure about
Everything above is an accounting identity plus a small number of explicitly-labelled behavioural assumptions. The identities are not opinions; the assumptions are, and they are all on the levers.
The identity the whole debate turns on
Consolidate the Treasury and the central bank. The consolidated public sector's liability to everyone outside it is the sum of three things — bonds held by the public, central-bank reserves, and banknotes:
L = Bprivate + V + H
Lt = (1 + i) · Lt−1 − St − σt−1
σ = i·H + (i − iv)·V ← seigniorage: on banknotes, and on reserves
Bonds held by the central bank, Bcb, do not appear. That is the formal content of the claim that cancellation changes nothing — and it is exactly right. It follows that the operation can change the consolidated position only through one of five routes, and the model prices the first three:
Every parameter
Every input the model takes, with its provenance. These are not controls — the controls are the levers in §02 and §08; this is where each of their starting values comes from. Observed means published by a statistical authority. Derived means computed here from published inputs, with the arithmetic stated. Assumption means a judgement you can and should move. Where it lands points at the section whose figures change when the parameter does; three rows say context instead, because they feed no number on the page — they are scale, an anchor for a lever, or a published benchmark this model is checked against rather than fitted to.
What this model does not do
11
Sources
Primary sources wherever one exists. Where a claim rests on our own arithmetic rather than a publication, it is tagged DERIVED above and the inputs are given.
