Every year the Italian Treasury’s Rapporto sul Debito Pubblico carries a chart with a long title: “Composizione per tipologia di controparte degli ordini portati nelle aste dei BTP nominali da parte degli Specialisti in titoli di Stato” — the composition, by counterparty type, of the orders brought to nominal BTP auctions by the Specialists in government securities. It is a stacked monthly bar chart, seven investor categories in every edition but one, two calendar years per edition, and it shows who was buying around Italy’s nominal BTP auctions. In the 2025 edition it is Figura 5. In the 2016 edition it was Grafico IV.16. The Treasury has published it in every edition in between and has never released the numbers behind it as a table. I read them off the charts, put the result next to what the Treasury says about the same figure in its own prose, and then follow a footnote in the Treasury’s report to where the data come from, which turns out to be a return that fourteen euro-area sovereign debt offices receive in the same format.

What one issuer publishes

The 2025 Rapporto’s sentence beside Figura 5 is the natural place to start, because it gives the Treasury’s own reading of its own chart. In 2025, it says, the participation of banks fell to “una quota sul totale di poco inferiore al 15%”, just under 15% of the total; investment funds rose slightly to “intorno al 32%”, around 32%; and hedge funds rose slightly too, “pari a circa il 42%”, about 42%. The 2016 edition reads the same chart for that year: investment funds “circa il 40%”, banks “in media sul 30%”, on average around 30%, hedge funds “circa il 16%”, central banks “una media del 7%”. Same chart, same issuer, the Treasury’s own numbers at both ends, nine years apart.

The chart is a vector drawing inside a PDF, which means every bar is a filled rectangle with a height, and the height of each segment divided by the height of the column is that category’s share for the month. Decoding ten editions that way gives an annual series for hedge funds of 14.8% in 2015, 16.0% in 2016, then 31.0, 34.1, 40.5, 33.8, 37.9, 33.4, 37.3, 40.8, and 44.4% in 2025. Banks run the other way, from 28.7% to 15.6%. Asset managers are roughly flat, 41.1% to 34.0%.

The shape matters more than the endpoints. The largest single-year increase is a fifteen-point step between 2016 and 2017, which is about half of the whole rise from 14.8 to 44.4; the series then runs between 31 and 41% for six years and rises again, by eleven points, between 2022 and 2025. Two movements, then, and the second is not small. The Treasury describes that step itself, in the 2017 edition: a considerable increase in the presence of hedge funds compared with the two preceding years (”il considerevole aumento della presenza di hedge fund rispetto ai due anni precedenti”), matched by a fairly marked reduction in the bank component (”una riduzione abbastanza marcata della componente bancaria”), and it adds that the increase was not only an Italian phenomenon but affected most European countries (”non è stato un fenomeno solo italiano, ma ha interessato la gran parte dei paesi europei”). Two things about that step should be said before it is read as a rise in hedge-fund flow. It is a share, and a share cannot say whether the hedge-fund numerator rose or the bank component of the denominator fell; the 2017 edition does not say which, and 2016–17 was also the period of the Italian banking system’s own stress. The one bank decline the Treasury does explain is the following year’s, which the 2018 edition attributes, “presumibilmente” — presumably — to foreign credit institutions pulling back; it offers no such reading for 2017. And it is the one date at which a change in how dealers coded their clients, or a change in which dealers were in the Specialist panel that reports, would produce exactly this pattern; nothing in the public record rules either in or out. The Treasury’s remark that the rise was not only Italian is quoted as its remark; no second issuer publishes a series that could test it.

A reader who knows these reports will want two things checked before trusting a chart-derived series, and both can be checked. Consecutive editions overlap by a year, so nine of the eleven years have two independent readings; the largest disagreement between them is 0.7 percentage points. And the Treasury quantifies categories in its prose often enough to test the decode against it: the 2016 edition’s four figures above against decoded 38.2, 29.5, 16.0 and 7.2; banks at 16% for 2019 against 15.0; “circa il 24%” for 2020 against 23.8; “poco inferiore al 24%” and “intorno al 39%” for 2023 against 22.8 and 38.5; “poco inferiore al 25%” and “intorno al 29%” for 2024 against 24.8 and 30.0. The gaps are small and they do not all run the same way: the decode sits below the Treasury’s figure in 2019, 2020 and 2023, on both sides of it within 2016 and within 2024, and above it for all three categories in 2025, by up to 2.4 points in the hedge-fund figure. The Treasury describes its annual figures as shares “sul totale” (of the total) and “in media” (on average) for the year without saying how they are computed, and the decode gives simple means of monthly shares, so the cause of the residual is not established. The figure to quote for 2025 is the Treasury’s “circa il 42%” on the plain ground that it is the issuer’s own annual number; the decoded 44.4 is the reading of its chart. The annual values are in the table at the end and the monthly observations — 240 month-and-edition rows across ten editions, each carrying every category its edition draws, with the edition and window definition for every row — are released with this piece as a data file alongside the decoder, because a measurement piece that will not show its measurements is not one.

There is one methodology change in the series, and it belongs in the text. Between the 2017 and 2018 editions the Treasury altered the window over which it selects the dealer flows: from the auction announcement to the day after the auction, which included the supplementary tranche reserved for Specialists, to the day after the announcement through the day before settlement. The 2017 edition covers 2016 and 2017 on the old definition and the 2018 edition covers 2017 on the new one, so the same year is observed both ways. The two readings differ by 0.2 percentage points. That comparison confounds two things — a different edition and a different window — and cannot isolate the window on its own; what it shows is that the two together move the reading by less than the 0.7 points by which editions on the same window disagree, so whatever the window change did is bounded by ordinary edition noise. It says nothing about dealer panels or client coding at the same date. A documented change with a measured effect is still better evidence than a series with no known change, because it is the difference between an assumption and a test.

Three limits, stated plainly. This is a demand proxy and not a record of bids. The Treasury selects, from its dealers’ returns, the client flows in a window around each auction, and calls the result, in its own words, no more than “una buona proxy della domanda in asta da parte degli investitori finali”: a good proxy for auction demand from final investors. Second, where a category’s flows in a month “siano risultati nulli o negativi”, null or negative, its contribution is set to zero for the chart. That floors a category at zero in the months it is a net seller, which relative to a signed treatment raises it and lowers everyone else. Which category the floor binds on most often is not published. Hedge funds are the plausible candidate, since the Treasury reports their net purchases across its dealers’ whole client book as negative or near zero in most recent years (an overall contribution that “è risultato negativo (-4%)” in 2025) and says they tend “a mantenere un posizionamento neutro alla fine di ciascun periodo contabile”, to hold a neutral position at the end of each accounting period; but that is behaviour over the year, and the incidence of negative months inside the auction windows themselves is not reported. If the floor binds on them, it works toward overstating their share, and the size of the effect cannot be recovered from what is published. Third, Eurosystem purchases are not in the series at all. The reporting format the Treasury draws on excludes “transactions with central banks that are executed within the framework of their monetary policy (i.e. PSPP and PEPP)” from reporting altogether, so the denominator is dealer client flow with the Eurosystem removed.

Where the number comes from

The Treasury’s own footnote says where the flows come from: “le comunicazioni secondo lo schema EMAR, il formato armonizzato adottato dal Sottocomitato ESDM (European Sovereign Debt Markets) del Comitato Economico Finanziario” — the returns under the EMAR schema, the harmonised format adopted by the ESDM sub-committee of the Economic and Financial Committee. Follow it and you leave Rome for the Economic and Financial Committee, whose Sub-Committee on EU Sovereign Debt Markets has run a dealer-reporting standard called the Harmonized Reporting Format since a decision of 8 December 2004.

The format is quoted here at length because the institutional point rests on it and on nothing else. Every primary dealer appointed by a participating debt office files a monthly XML return with that office, one row per transaction. The row carries the trade date, “according to the calendar trade date (T)”; the ISIN; the direction, B or S, from the dealer’s perspective; the nominal quantity, to two decimals; the counterparty’s country of incorporation; the trading system, from a list of fourteen; the value date; and the counterparty type. The documented 2022 version of the format lists thirteen counterparty type codes: Inter Dealer Broker, Bank-Primary Dealer, Bank-Inter Dealer, Bank-Connected Entity, Bank-Customer, Debt Management Office, Public Entity, Pension Fund, Insurance Company, Fund Manager, Hedge Fund, Retail, Corporate. Hedge Fund is code 12. Code 2 is missing from the sequence because “Bank-Own Account” became obsolete in January 2012. That is why this note says the architecture dates from 2004 rather than that the category has been measured continuously for twenty-two years: the taxonomy has changed at least once.

Which rows are which also needs stating, because a supervisor will ask. Under code 7, Debt Management Office, the reported trades “include all primary market transactions (auctions, taps, syndications) and buy backs” and “all the amounts bought by the reporting PD in the framework of auctions, including non-competitive bids”. Those rows are the dealer’s own purchases from the issuer. The Italian series is not built from them; it is built from the dealer’s trades with its clients, falling inside a date window around each auction, and what makes that construction possible is simply that the trade date and the counterparty type sit on the same row. How the Treasury folds the format’s thirteen codes into its chart’s seven categories is not published, and the correspondence used here is inferred from the names: “Fondi Hedge” is taken to be code 12, “Fondi gestione” code 11, “Fondi pensione” and “Compagnie assicurative” codes 9 and 10, “Corporate & Retail” codes 13 and 14. Which of the format’s four bank codes make up “Banche”, and where the chart’s “banche centrali ed altre entità pubbliche” sits in a format whose public-sector codes are Debt Management Office and Public Entity and which names no central bank, are the two places the inference is weakest, and neither touches the hedge-fund category.

The sub-committee does something with these returns, and it is public. Each quarter it publishes the Euro Market Activity Report, covering “14 sovereign and 2 supranational securities markets in the euro area”, built from turnover “reported by the dealers to the participating Debt Management Offices in their respective markets”, collected “on a trade by trade basis”, with primary-market activity inside the totals. The report’s first table is headed “Counterpart type vs counterpart region”. It has two rows: Inter Dealer, and Customer. Where the dealers’ own primary-market purchases from the issuer sit between those two is not visible from the report, only that they are inside the totals.

So the Italian series is not a special collection. It is one national cut of a return that every participating debt office receives, and the public euro-area aggregate of that same return distinguishes dealers from everyone else and stops. What I do not know is where the reduction happens: whether the sub-committee receives the transaction-level returns and aggregates them, or receives an already-reduced two-category submission from each office. The published aggregate is what is visible. I have a query with the report’s own mailbox, and will report the answer or its absence.

What can be said about the other offices is narrower than it might sound. Each of them appears to hold the fields needed to construct a similarly structured national cut for its own market, subject to its own dealer coverage, its choice of window, its treatment of negative flows and its instrument scope. Whether two such cuts would be comparable across markets is a separate question and this note does not answer it.

What would have to be true

The principal limitation goes first, and it concerns the category. Throughout, the result is the share the Italian Treasury assigns to code 12. The format describes no validation of the counterparty type a dealer assigns against the legal form of the client, and none is visible from outside, so a rising code-12 share is consistent with dealers recoding relative-value accounts as well as with a change in who is actually trading, and nothing in the public record separates the two. Three things would: the coding instructions or validation procedures dealers work to; the dispersion of the code across dealers; or a second issuer publishing the same cut. The one documented change in the series has been measured at 0.2 points, but that is evidence about the window and not about the category. I do not upgrade “the share assigned to code 12” into a claim about the underlying investor population, and a reader should not either.

The second question is whether a window around the auction measures demand at the auction or trading near it. The Treasury publishes a comparison series in the same reports, in the section on dealers’ trading by counterparty type, and it is a share of gross turnover — “quota di volumi negoziati”, a share of traded volumes, purchases plus sales — where the auction-window series is a floored net flow, so the two are compared here for direction and not for level: the hedge-fund share of its dealers’ trading with final clients was 17.4% in 2017 and above 19% in 2018 (both stated in the 2018 edition), 16% in 2020 and 18% in 2021 (2021 edition), 27% in 2022, 30% in 2023, 32.5% in 2024 and 33.3% in 2025, each in its own edition; the 2019 figure is not stated in those terms. The two series are not mechanically linked. In 2022 the trading share rose nine points while the auction-window share fell four and a half; in 2025 trading rose under one point against three and a half or more in the window. Across 2017 to 2025 both rise; the turnover share rises further, but gross turnover and floored net flow are different enough constructions that the proportional comparison should not be pressed, and annual figures cannot separate demand that is about the auction from activity that is merely around it. I claim nothing about hedge funds moving toward auctions, and the chart below is there to show the limit rather than to resolve it.

What would make the note wrong, by proposition. The documentary finding fails if the format’s codes are not collected as specified, or if a granular breakdown is published somewhere in the euro-area aggregate that this reading missed. The measurement fails if the decode is wrong, which the table at the end exposes to anyone who wants to check, or if the series is unstable to how the window is drawn, which the 0.2-point test partly addresses. The interpretation of code 12 fails on evidence that dealers assign the category inconsistently.

Where you cannot run this at all

Nine sovereign issuers other than Italy were checked, plus the European Union as a supranational, on the documents examined as of September 2026: France, Germany, Spain, the Netherlands, Belgium, Austria, Ireland, Portugal and Finland. Eight of the nine publish the order book as a scalar and nothing more — a size, an account count, an oversubscription ratio — and every investor-type series they publish is an allotment, a holding, or secondary turnover. Germany is in the eight on the Finanzagentur’s own word: its Transaction Reviews carry an uncaptioned investor-type pie for each syndicated Bund, and asked which side it shows, the agency’s press office replied in September 2026 that the charts refer to the allocation and not to the order book, and that the composition of an order book is not published. For the ninth, Spain, the same is true of everything except one chart whose label does not settle whether it shows allocation or demand; a query is with the Tesoro, and until it answers Spain is unresolved rather than negative. For each of the nine the documents examined were the annual report or review, the funding outlook or programme, the per-syndication results notices, and the investor presentation where one is published; quarterly bulletins and national-language variants were not exhausted, and the document list is in the working note released with this piece. Agence France Trésor is representative: total demand as a single number, then the breakdown of what was allocated. Belgium is the most explicit. Its deal summary for the ten-year OLO 97 of January 2023 reports an order book “in excess of EUR 51 billion” from “over 300 investors”, then two splits of the allocation: by geography, with “85% of the total allocation going to customers outside of Belgium”, and by investor type, with fund managers and bank treasuries at 31% each, central banks and other public entities 17%, banks 11%, hedge funds 5%, pension funds 3% and insurers 2%. Everything after the order-book sentence is allocation.

That finding has a boundary. The Euro Market Activity Report covers fourteen sovereign markets and its published documentation does not name them, so with ten sovereign issuers examined here — Italy and the nine above — at least four of the fourteen are outside this survey, and nothing is claimed about them either way. The formulation is the only equivalent cut found in these ten issuers’ document sets as of this date, and not that no issuer publishes one. Spain’s reading rests on the Treasury’s Chart Pack dated 9 December 2022 — which is the file its website served at the Chart Pack address on 4 September 2026, with a server modification date of 1 March 2023 — and on the Kingdom of Spain presentation of December 2023, whose investor-base section is holdings.

The near misses should be named, because a reader who knows these markets will think of them. Belgium publishes net purchases by investor type. The Netherlands publishes DSL turnover across six dealer-reported categories. Germany’s creditor-structure table is estimated from dealers’ purchases and sales to end investors. All three are dealer-reported, and all three are secondary or stock measures with no anchor to the primary market; whether any of them is drawn from the harmonised return rather than from a national collection is not stated in their documentation, and I do not assume it. What separates Italy’s is the date window, and the fact that its source is named.

What it would be for

A request that a public body publish more because the fields exist is not an argument, so here is the question the categories would answer, stated at the level the data can bear. What is the composition, by investor type, of the client flow dealers report around euro-area sovereign auctions, and how does it move with market conditions? Someone already wants to know. In September 2024 an ECB blog reported an ad-hoc survey of large primary dealers run through the Bond Market Contact Group, finding that hedge funds “now represent over 50% of the total requests received by dealer banks around public debt auctions” in Germany, France, Italy and Spain. Requests received are a different quantity from trades executed, and the returns carry the second; for the same year the Italian auction-window reading of executed net flow is 40.8%, which is what one would expect to sit below a share of requests if requests are inflated relative to fills, though the two are not the same object and the comparison is offered as a juxtaposition. The survey is evidence that authorities want to know the investor composition around auctions, and the returns are the nearest thing to it that already exists every month. A counterparty aggregate with the categories intact could answer that question at whatever level of aggregation confidentiality permits; Inter Dealer versus Customer cannot, because that split puts a pension fund and a hedge fund in the same row. The Italian series is the demonstration that the question has content: the share moves, by a lot, and the two-row aggregate could not show that it moves. What the categories would not deliver should be said too. They are a share of a proxy, which is what the Italian chart is, and neither bids, nor leverage, nor a measure of concentration.

The likeliest reason the published aggregate has two rows is not oversight. It is some combination of cell confidentiality across fourteen markets of very different sizes, and the comparability of a category that each dealer assigns for itself — which is the same objection that makes the Italian series a measure of code 12 rather than of hedge funds. Those constraints are real. A note that pretended otherwise would deserve the answer it got.

The question

So the ask is a question, and it is addressed to the sub-committee as the committee of debt management offices: what is the constraint? Thirteen codes in the reporting format. Two in the published report. One national cut that keeps the categories, published every year for a decade by an office that receives the same returns as the others. If the reduction is confidentiality, the useful answer says at what level of aggregation it binds. If it is comparability, the useful answer says what would have to be true of dealer coding for the categories to be published. Either is more than the current position, in which the question has not been put in public.

The number every issuer does publish is the one that carries least. After each syndicated bond, euro-area debt offices report how much of it was allotted to each investor type. A hedge fund allotted 1% of a French syndication — the figure in each of Agence France Trésor’s seven syndication results releases from January 2025 to April 2026, dated 21 January, 12 February, 25 March and 29 April 2025 and 14 January, 11 February and 14 April 2026 — or 3.4% of an Italian one, the simple mean of the hedge-fund row across the eight nominal BTP syndications in the Rapporto’s Tabella III.5, may have asked for a great deal or for very little, and the published record does not say which. There is no published denominator. The reporting format the fourteen offices receive carries the fields from which a proxy of the kind Italy builds for its auctions could be built, and nine of the ten offices examined here print nothing of the sort. The tenth prints a different cut of the same return instead. That is a proxy for auction demand, which is what has been shown; no equivalent has been shown for syndications. Whether the two supranational issuers in the report receive the same transaction-level file is not stated in the public documents, and nothing here is addressed to them.

Annual values, decoded

Paweł Fiedor — The Macro Prudential View

Views are my own. This article is a personal analytical piece based on public sources. It does not represent the views of the European Systemic Risk Board, the Central Bank of Ireland, the Eurosystem, or any other institution.

Italian Treasury

Ministero dell’Economia e delle Finanze, Dipartimento del Tesoro, Rapporto sul Debito Pubblico, editions 2016 to 2025 — the auction-order chart (Grafico IV.16 in the 2016 edition, Figura 5 in the 2025 edition) and the prose beside it in every edition; the footnote naming “lo schema EMAR”; the window definition in the 2017 and 2018 editions; the zeroing rule; the hedge-fund share of Specialists’ trading with final clients from the section on trading by counterparty type (2018, 2021, 2022, 2023, 2024 and 2025 editions); Tabella III.5 of the 2025 edition for the syndication allocations. All quotations in Italian are from these documents; translations are mine.

The reporting format and the published aggregate

EFC Sub-Committee on EU Sovereign Debt Markets, HRF 2022 — Harmonized Reporting Format — the decision of 8 December 2004, the fields of the monthly return, the thirteen counterparty type codes, Hedge Fund at code 12, the Bank-Own Account code made obsolete in January 2012, the definition of code 7, the exclusion of PSPP and PEPP transactions.

Euro Market Activity Report, Q1 2025 to Q1 2026 — “14 sovereign and 2 supranational securities markets”, the trade-by-trade basis, and the “Counterpart type vs counterpart region” table with its two rows.

Other issuers

Agence France Trésor, syndication results releases of 21 January, 12 February, 25 March and 29 April 2025, and 14 January, 11 February and 14 April 2026 — total demand, then the allocation by investor type, hedge funds at 1% in each.

Belgian Debt Agency, OLO 97 deal summary, January 2023 — the order book, the geographic split and the investor-type split of the allocation.

Tesoro Público, Chart Pack — the file served at that address on 4 September 2026, dated 9 December 2022 — and the Kingdom of Spain presentation of December 2023.

Dutch State Treasury Agency, Outlook 2026, Figure 12; Deutsche Finanzagentur, Transaction Reviews and the creditor-structure table at deutsche-finanzagentur.de, and the agency’s press office by email, September 2026, on what the Transaction Review charts show; the annual reports, funding programmes, syndication notices and investor presentations of the OeBFA, NTMA, IGCP, the Finnish State Treasury and the European Commission, as listed in the working note.

ECB

Ferrara, Linzert, Nguyen, Rahmouni-Rousseau, Skrzypińska and Vaz Cruz, Hedge funds: good or bad for market functioning?, The ECB Blog, 23 September 2024 — the Bond Market Contact Group’s ad-hoc survey of large primary dealers and the “over 50%” figure for Germany, France, Italy and Spain.

Data and code

github.com/pafiedor/tmv-data/code12 — the 240 monthly observations with edition and window for each row, the annual table, the window-versus-turnover series, the decoder, and the survey note.

Calculations. The chart is decoded from the PDF’s vector content: each bar segment is a filled rectangle, and its height divided by the column’s height is the category’s share for the month. Annual values are simple means of the twelve monthly shares, with a category that draws no segment in a month counted at zero, which is the Treasury’s own treatment of null or negative months. Where two editions draw the same year the later edition is used and the earlier reading is shown beside it. The Treasury’s stated annual figures, where it gives them, are the figures of record; the decoded values are readings of its chart. The 3.4% syndication figure is the simple mean of the hedge-fund row across the eight nominal BTP syndications in Tabella III.5 of the 2025 Rapporto. Translations from the Italian are mine.

The Uses of Slack left one question open on purpose. The ECB’s fund statistics show that Austrian open-ended real estate funds borrowed their way through a four-year redemption run — a loan stock of €627m in mid-2022 and €1.6bn at its peak in the third quarter of 2024, with the largest quarterly draw, €361m, in the last quarter of 2023, when redemptions peaked — and a sector series cannot say whether the funds that borrowed were the funds that faced the redemptions. A handful of vehicles can. Austria’s public-offer property funds publish audited annual reports and unaudited half-year reports under § 13 of the Immobilien-Investmentfondsgesetz, each with a balance sheet, a flow statement and a management review, and there are five of them. This note assembles that record from 2020 to 2026 and puts each fund’s borrowing next to its own flows.

The short answer is that the cross-section sorts nobody. Every public fund bled, and every public fund borrowed. The interesting material is within each fund — what changed between one reporting date and the next — and the record there is heterogeneous enough that the note keeps it so.

Five funds, four calendars

The population is small. Five public-offer funds under the ImmoInvFG were in existence through the run and reported on it: ERSTE IMMOBILIENFONDS and ERSTE RESPONSIBLE IMMOBILIENFONDS (ERSTE Immobilien KAG), REAL INVEST Austria (Bank Austria Real Invest), immofonds 1 (Union Investment Real Estate Austria) and LLB Semper Real Estate (LLB Immo KAG). The FMA’s register also lists a sixth public fund, launched in January 2025 and outside everything below except the residual, one fund for professional investors, and five Immobilienspezialfonds with no public report at all. “The five” in this note means the five incumbents. The five report on three different year-ends — 30 April, 30 September, 31 December — which is why nothing below is a quarterly series and why the reconciliation to the sector at the end is a set of bounded points.

Borrowing throughout means the credit line of the fund’s own Vermögensaufstellung — Kreditverbindlichkeiten, Verbindlichkeiten aus Krediten, Verbindlichkeiten aus Fremdfinanzierung — as the fund draws it. Where a report shows property companies as participations with loans of their own, as at REAL INVEST and ERSTE IMMOBILIENFONDS, that debt is outside the line; where a report presents a property partnership transparently, as ERSTE RESPONSIBLE does with one Vienna partnership, its €30m bank loan sits on the fund’s own long-term credit line and is counted here because the fund counts it. That is the fund’s statutory balance sheet, which is also the unit the ECB’s investment-fund statistics collect, under Regulation 1073/2013 to November 2025 and the recast Regulation 2024/1988 from December 2025. The comparison with the sector series later on is therefore a comparison of concept; there is no published line-by-line mapping, and the note says so where the comparison is made. Part 1 checked that every change in the sector’s loan stock since the peak equals its transactions, which leaves no room for a reporting-population break at the changeover. Every figure below was read on its page in the fund’s report; sales volumes are the ones the management reviews state, and where a report records sales without a volume the chart says so.

Who bled

REAL INVEST Austria is the largest fund and took the largest outflow. Its own review puts net outflows since June 2022 at “rd. EUR 1,5 Mrd.” by September 2024 and “rd. EUR 2 Mrd.” by September 2025; the flow statements give gross redemptions of €668m, €1,010m and €578m in the three fiscal years to September 2025, against fund assets that went from €4.4bn to €2.6bn. ERSTE IMMOBILIENFONDS ran a net outflow of €530m in the year to April 2024, then €413m and €240m. LLB Semper Real Estate took €380m of redemptions in the ten months of 2023 before it suspended on 23 October — its report says the first half was institutional money, the second half retail — on a fund of €1.1bn. immofonds 1 lost €258m net in the year to September 2024, on €1.3bn. ERSTE RESPONSIBLE, the smallest, lost about €100m in its worst year.

The OeNB’s Financial Stability Report recorded the sector’s 2024 net outflows at €1.6bn, some 17.6% of assets; the fund reports add the other side of each balance sheet.

Who borrowed, and in what order

At two funds the reporting dates fall where an order can be seen, so they come first.

ERSTE IMMOBILIENFONDS held €781m in bank balances in April 2022 and no credit. By April 2023 the balances were €368m; by October 2023, €212m; still no credit. The half-year report at that date records ten Vienna residential properties sold “im Herbst 2023” and announces that the management company will draw credit “bei Bedarf”, because “mit der Aufnahme von Fremdfinanzierung … wollen wir die Liquiditätssituation der Immobilienfonds stärken”. The first borrowing appears in the year to April 2024 — €190m at the year-end, 9.41% of fund assets before deduction — and rose to €262m by October 2024, 16.04% on the same basis. That was the year of the €530m outflow and of fifteen property sales the review values at around €140m, ten of them the autumn ones. The following year the fund sold twenty properties for around €403m and reduced the credit; a year later, nine for €239m, and the credit stood at €130m — €100m long-term, plus €30m short-term drawn again after the short-term lines had been repaid in June 2025. The wording moved with the years: “Stärkung der Liquiditätsquote” in the 2023/24 report, credit “zur Bewirtschaftung der Immobilien” in 2024/25 and 2025/26.

immofonds 1 reads similarly, and its record arrived last, from the management company itself, because the fund’s website carries only the current year. Liquidity ran from €349m in September 2022 to €200m in September 2023 with credit unchanged at €27m and, the report says, no sales that year. Between October 2023 and March 2024 the fund drew €125m, taking credit to €152m, in a half-year of €174m net outflows; one forward-funded purchase agreed in 2022 also completed in that half. Five properties, “rund EUR 135 Mio.”, were sold in the fiscal year — one transferred in March 2024, inside the half-year of the draw, the other four between June and September — and the credit was then reduced to €113m, where it has stayed. The 2023/24 review says redemptions were met “aus den liquiden Mitteln” and lists, as its liquidity-risk measures, “die Aufnahme von zusätzlichem Fremdkapital und die Vorbereitung und Umsetzung von Immobilienverkäufen”; the 2024/25 review describes liquidity as steered “über drei Säulen … Netto-Zuflüsse, Einsatz von Fremdkapital und Objektverkäufe”.

At the other three the rundown and the drawing overlap, and the note describes the overlap and claims no order.

REAL INVEST Austria lifted its credit from €83m in September 2022 to €259m by March 2023 while still holding €972m in bank balances and securities, then to €465m by March 2024 and a peak of €539m in March 2025, before repaying to €379m. Its liquidity holdings fell from €1.06bn to €262m over the same three years. The review’s account of the borrowing is that properties bought with equity were “mit langfristigem Fremdkapital refinanziert” while the fund’s voluntary securities reserve was sold down to meet redemptions, and it gives a broader leverage figure that includes the property companies’ debt — “rd. 17 %” of property volume in 2024, “rd. 20 %” in 2025 — which is a different quantity from the fund-level line charted here. It also sold: eighteen properties with a book value of around €536m in the year to September 2024, twenty more for around €516m the year after, and by its own count 43 properties with a book value of around €1.2bn between October 2023 and November 2025, which is more than twice what its credit ever rose.

LLB Semper Real Estate entered the run already borrowing, at €160m in December 2021 and €217m a year later, and drew hardest in the first half of 2023: €389m by June, €447m by December, on a fund whose assets had fallen to €676m. The 2023 report is the plainest of the five about purpose. Credit lines were drawn “zur Erhöhung der Liquidität”, and a short-term loan was taken “zur Bedienung von Anteilscheinrücknahmen in zwei Tranchen”. Liquidity was 4.45% of fund assets at the end of 2023 and 8.84% at the end of 2024, both below the 10% floor of § 32, and the report says so. The fund suspended redemptions on 23 October 2023, its management company gave the six months’ notice of § 15 on 23 April 2025, and the fund has been in liquidation since 24 October 2025 with about €300m of credit still on the balance sheet.

ERSTE RESPONSIBLE, which had no credit before 2024, had about €62m by September 2024 — €32m of short-term credit drawn under Pfandbestellungsurkunden of up to €60m, and a €30m bank loan at a property company the fund consolidates — and has since moved between €36m and €51m. Its report calls the credit Fremdmittel “zur Bewirtschaftung des Immobilienportfolios” and the sale of its main stake in a Vienna campus property a step “zur Sicherung der Liquidität”.

Five funds, then, and five ways of describing the same balance-sheet line: redemptions, liquidity, refinancing, management, a pillar. Whether these are one function in five legal shapes is my inference, and it is labelled as one. What the balance sheets establish is narrower and firmer: at every fund, between the start of its outflows and the end of its worst outflow year, cash fell and credit rose; at two of the five the reports place the cash rundown before the borrowing; and borrowing-before-sales is visible nowhere: at ERSTE the autumn-2023 sales came before the first credit, and at immofonds 1 the March-2024 transfer sits inside the draw half-year.

Two routes to the same line

The Austrian statute gives a real estate fund two ways to raise money against its assets, and the structure of the second says which one the record is looking at.

The ordinary route is § 5(2). A fund may borrow and encumber its properties, “unbeschadet des § 11”, where the fund rules provide for it, where it is “im Rahmen einer ordnungsmäßigen Wirtschaftsführung geboten”, and where the depositary consents to the borrowing and the encumbrance “weil sie die Bedingungen … für marktüblich erachtet”; borrowing and encumbrances together may not exceed 50% of the market value of the fund’s real-estate assets. The other route is § 11(2), which is written for exceptional circumstances. Where bank balances and the proceeds of securities do not suffice for redemptions and orderly management, the fund is to sell assets; redemption may be refused for up to a year, two if the fund rules say so; and only after that period may the management company encumber the fund’s assets “wenn das erforderlich ist, um Mittel zur Rücknahme der Anteilscheine zu beschaffen”, unwinding the encumbrance as soon as it can and notifying the FMA at once.

The two are concurrent powers — the “unbeschadet” says so — but the second opens only after a fund has refused redemptions under § 11(2) and the refusal period has run. Four of the five funds never refused a redemption: their flow statements show every year’s redemptions paid, and no report of theirs records a refusal or a notice to the FMA. For them the period never started, the § 11(2) power never opened, and their borrowing can only have been § 5(2) borrowing, consented to by the depositary in the ordinary course. LLB Semper refused redemptions from 23 October 2023 with €389m borrowed at the June 2023 date before the gate and €447m at the December one; the refusal period could not have run before October 2024, its credit stock has fallen since June 2024, its 2024 and 2025 reports describe no encumbrance taken to fund redemptions, and redemptions never reopened. For LLB that is an absence of any sign, since gross draws are not observed; for the other four it is the statute. None of this is a claim that any order was reversed or any rule breached: the fund rules are FMA-approved, the reports are audited, and the legislative materials of 2003, 2021 and 2026 say nothing about whether credit that meets redemptions falls within “ordnungsmäßige Wirtschaftsführung”. No public FMA statement on the point has been found. The relationship between the two provisions is left, on the record, unresolved.

One governance observation belongs here, stated exactly. The security behind both ERSTE funds’ credit is disclosed property by property, and it reconciles to the euro. At 31 October 2024 ERSTE IMMOBILIENFONDS’s €262m was €162m drawn under Pfandurkunden of up to €240m “zugunsten der Erste Group Bank AG”, a €70m mortgage in favour of the same bank, fully drawn, and €30m under a €36m Simultanpfandrecht in favour of Erste Bank der oesterreichischen Sparkassen AG. ERSTE RESPONSIBLE’s €62m at 30 September 2024 was €32m under Pfandbestellungsurkunden of up to €60m in favour of Erste Group Bank AG and a €30m “Bankkredit der Erste Bank der oesterreichischen Sparkassen AG” at a consolidated property company. Erste Group Bank AG is the depositary of both funds and the parent of their management company; Erste Bank der oesterreichischen Sparkassen AG is its Austrian banking subsidiary. So the pledge beneficiary and the consenting depositary are the same group bank, and where a lender is named it is the group’s bank. For the Pfandurkunden facilities the reports name the pledgee and not the lender, and the note does not infer one from the other. The other three funds’ reports name neither; Part 1 found 86% of the sector’s loans owed to domestic banks.

What the pairing cannot show

A fund-by-fund record invites a test, and the honest version of the test is thin. The reports give twenty-two fund-years. Five are inflow years, all before the run. Two are the zero-flow years of a gated fund. Fifteen are outflow years, and seven of those carry a material increase in borrowing — one at each of the five funds, plus a second at REAL INVEST and at LLB. No fund increased borrowing materially in an inflow year. That is a period effect in a window that contains one shock, and the empty cell says more about the window than about behaviour.

Nor can the annual and half-year dates resolve the quarter Part 1 flagged. The sector’s largest draw, €361m, came in the last quarter of 2023. Between their autumn-2023 and spring-2024 reporting dates four funds drew: ERSTE IMMOBILIENFONDS €190m, REAL INVEST €203m, immofonds 1 €125m, LLB €58m in the second half of 2023. The windows overlap the quarter without matching it, and the split stays open.

The larger limit is use of proceeds. Money is fungible and a balance sheet records positions, so the same year-end can show an outflow, a smaller cash pile, a larger loan and a list of sales, and cannot say which paid for which. REAL INVEST’s 2022/23 year is the clean illustration: liquidity fell by €487m and credit rose by €179m against a net outflow of €479m, with property up €59m and distributions of €50m — the changes over-explain the outflow, because income, purchases, capital expenditure, valuation and settlement timing are all moving too. Chart 1 groups the four changes instead of stacking them for that reason, and the phrase “the money came from” does not appear in this note. That the public accounts cannot identify use of proceeds is one of the note’s results.

Distance to the cap

Part 1 could say only that nothing showed the statutory ceiling binding. The reports allow the ratio to be stated on the statute’s own base.

LLB Semper’s Fremdfinanzierungsquote, as it reports it, was 41.88% at the end of 2023, 43.10% at the end of 2024 and 42.40% at the end of 2025, against 50%. immofonds 1 prints its own ratio too: 2.46% in September 2023 and 14.0% six months later. ERSTE and REAL INVEST print the rule and not the ratio, so for them the chart’s line is credit over property value — a lower bound, since the statutory numerator also counts encumbrances — and it stayed at or below about 17%. At the floor, two funds disclose breaches of § 32’s 10% liquidity minimum in their own words: LLB in 2023 and 2024, and ERSTE IMMOBILIENFONDS “zeitweilig” in 2023/24 and “vorübergehend” in 2025/26, its liquid assets at 10.4% and 11.22% of fund assets on the two April year-ends. That is what the funds disclosed. The note does not claim the ceiling or the floor bound anyone’s conduct, and the sentence from Part 1 stands: fund rules, lender covenants and credit supply bind long before a product-law maximum does.

The rest of the sector

The five public funds are most of the Austrian sector’s borrowing, and the reports let that share be stated at reported dates. At each 31 March and 30 September from 2023 to 2026, three funds report on the day; LLB Semper and ERSTE IMMOBILIENFONDS do not, and each is taken as the range between its two adjacent reporting dates. Summed that way, the five stood at 55–74% of the ECB loan stock in March 2023, 59–64% in September 2023, 67–78% in March 2024, 78–85% in September 2024, 76–86% in March 2025, 84–90% in September 2025 and 84–88% in March 2026. Each range assumes only that a fund’s credit on the intermediate date lay between its two adjacent reported values.

The remainder is the five Spezialfonds, the professional-only fund and, from 2025, the newly launched sixth public fund, and it is a residual — the ECB stock minus the public sum — so it absorbs every difference of concept between a fund’s statutory balance sheet and the ECB item, every reporting gap and the changeover of regulation in December 2025. It moved, though: from €232–403m in March 2023 to €418–476m in September 2023, then down to €246–349m by September 2024 and €136–179m by March 2026, while the public five were adding €540–700m between September 2023 and September 2024. The arithmetic is reported because it is part of the reconciliation, and it is indicative: a residual of this kind can move for accounting reasons as well as borrowing ones. Nothing in it says what the non-public funds did or why, and the note claims nothing about them.

What is left

What the public layer cannot say is what it could not say at the start: collateral beyond the pledge ceilings two funds disclose, recourse, the gross draws and repayments behind the net year-end figures, and the behaviour of the unobserved sixth of the sector. With five funds and one shock it estimates nothing; it is a case assembly, said once.

Two things it does say. The Austrian episode in Part 1 was the sector; it is now the funds, and it is all five that were there for it. And the credit on their balance sheets through the run went through the ordinary-course door of the statute at the four funds that never refused a redemption, with the depositary’s consent, and shows no sign of the other door at the fifth. Against the 50% ceiling, the two funds that print the statutory ratio were at 42–43% (LLB) and 14% (immofonds 1); the other three print no statutory ratio, and on credit alone were at 17% of property value or less, which bounds their headroom from one side only. What it was for is what the funds say it was for, and they do not say the same thing: LLB’s report ties its short-term loan to redemptions; ERSTE’s tie credit to liquidity and then to management; immofonds 1 lists it as one of three pillars; REAL INVEST, cash-rich when it started drawing, calls it the refinancing of equity-bought property. At four funds the stock has come down from its peak and at none is it back where it started. At the fifth it is part of the liquidation. LLB Semper’s reports are a public account of an Austrian property fund unwinding under § 17, where distribution to unit-holders comes “erst nach Erfüllung der Verbindlichkeiten des Immobilienfonds”: ten properties sold in 2025 for close to €158.7m, credit down to €300m, a first partial distribution of €70m in December 2025. Whether the borrowing lengthened the fund’s life or the unit-holders’ wait is not something its reports can settle, and the note does not.

The legislator’s answer arrived before the run. The 2021 amendment that takes effect on 1 January 2027 rewrites § 11(1) for existing funds: a twelve-month minimum holding period, twelve months’ notice of redemption, redemption dates at least quarterly. It is the instrument Germany adopted in 2013 with half the holding period — the German rule is twenty-four months’ holding and twelve months’ notice — so the verdict in Twelve years of evidence does not transfer as it stands. The Austrian act that transposed AIFMD II in July 2026 then added, on its own initiative and not the directive’s, an allowance for holders on the books at the end of 2026: €20,000 a year in 2027 and 2028 and €10,000 in 2029 and 2030, redeemable without the notice period, suspended while a liquidity tool is active. § 5 was left untouched. ERSTE has already written gating and notice extension into its fund rules, in April 2026. So a run in 2027–30 meets a twelve-month notice period, a retail allowance running beneath it, and an unchanged credit power; after 2030, the first and the last.

The question Part 1 left — whether the funds that borrowed were the funds that bled — has an answer for the public funds that were there for the run: all five of them, on their own published numbers, audited at each year-end. The prior work on this record, a 2025 TU Wien thesis on the liquidity-management toolkit, names the funds and the § 32 floor and does not pair borrowing with redemptions; the pairing is above. It shows one path under a common shock, taken by five funds in five somewhat different ways. Whether the next set of rules produces the same path is a question for the next run.

Paweł Fiedor — The Macro Prudential View

Views are my own. This article is a personal analytical piece based on public sources. It does not represent the views of the European Systemic Risk Board, the Central Bank of Ireland, the Eurosystem, or any other institution.

Fund reports (all figures read on the cited pages; EUR; fiscal years as stated)

ERSTE Immobilien KAG — ERSTE IMMOBILIENFONDS, Rechenschaftsberichte FY2020/21–2025/26 (30 April) and Halbjahresberichte (31 October); autumn-2023 sales and the liquidity wording HJB 2023/24 p. 4; 9.41%, floor wording and Pfandurkunden RB 2023/24 pp. 4, 20; 16.04% figure and the three security packages HJB 2024/25 pp. 5, 14; short-term repayment and “zur Bewirtschaftung” RB 2024/25 p. 4; 2025/26 floor breach and rule amendment RB 2025/26 p. 4. ERSTE RESPONSIBLE IMMOBILIENFONDS, Rechenschaftsberichte FY2020/21–2024/25 (30 September) and Halbjahresberichte to 31 March 2026; credit composition, wording and stake sale RB 2023/24 p. 4; Pfandbestellungsurkunden p. 18; the property-company bank loan p. 39. Both at ersteimmobilien.at.

Bank Austria Real Invest — REAL INVEST Austria, Rechenschaftsberichte FY2020/21–2024/25 (30 September) and Halbjahresberichte to 31 March 2026; narrative on outflows, refinancing and the 17% figure RB 2023/24 pp. 6, 8; “rd. EUR 2 Mrd.”, “rd. 20 %” and the 43-property figure RB 2024/25 p. 6. At realinvest.at.

LLB Immo KAG — LLB Semper Real Estate, Rechenschaftsberichte 2021–2024 (31 December), Abwicklungsbericht 2025, Halbjahresberichte to 30 June 2026; credit-line and loan wording, Fremdfinanzierungsquote and liquidity ratio RB 2023 pp. 10, 12 and RB 2024 pp. 11–12; 2025 sales, first distribution and 42.40% AB 2025 pp. 10–12. Suspension, termination and liquidation dates from LLB investor information.

Union Investment Real Estate Austria — immofonds 1, Rechenschaftsberichte FY2019/20–2024/25 (30 September) and Halbjahresberichte to 31 March 2026; back years supplied by the management company, September 2026; liquidity-measures wording RB 2023/24 p. 5, sales p. 9; “drei Säulen” RB 2024/25 p. 5. Current reports at union-investment.at.

Legal texts

ImmoInvFG (BGBl. I Nr. 80/2003), §§ 5, 11, 13, 15, 17, 32 and 43a, consolidated text on RIS; § 11(1) as amended by BGBl. I Nr. 198/2021 (applicable to existing funds from 1 January 2027 under § 43a(2)); § 43a(2a)–(2c) and § 11(1a) inserted by Austrian BGBl. I Nr. 57/2026 (AIFMD II transposition, in force 29 July 2026). Legislative materials: RV 97 d.B. XXII. GP (2003), RV 1100 d.B. XXVII. GP (2021), RV 502 d.B. XXVIII. GP (2026), all at parlament.gv.at.

Regulation (EU) No 1073/2013 (ECB/2013/38), repealed with effect from 1 December 2025 by Regulation (EU) 2024/1988 (ECB/2024/17) — investment fund statistics: fund-by-fund reporting under both, the concept the reconciliation uses on both sides.

KAGB § 255 — the German twenty-four-month holding and twelve-month notice rule for retail open-ended property funds.

Official reports and prior work

OeNB, Financial Stability Report 49 (June 2025), p. 37 — the 2024 outflow figure; Reports 50 and 51 carry no real-estate-fund passage.

FMA, Fondssuche — the register of ImmoInvFG funds, read 1 September 2026.

Giller, Liquiditätsstrategien der österreichischen Immobilienfonds nach der Zinswende 2022, master’s thesis, TU Wien, submitted 20 February 2025 — the prior document analysis of the liquidity-management toolkit.

Data

ECB Investment Fund Statistics (IVF), 2026-Q2 vintage — the Austrian loan stock (loans and deposits received, all maturities and counterparts), as in The Uses of Slack.

Calculations. The fund-year table, the disposals file and the reconciliation are compiled by The Macro Prudential View from the reports above; every value carries the PDF page it was read on. Liquidity holdings are bank balances plus securities at the reporting date. Changes are between consecutive fiscal year-ends. “Material” borrowing increase means at least 1% of opening fund assets. Coverage at each 31 March and 30 September takes REAL INVEST, ERSTE RESPONSIBLE and immofonds 1 at their own figures on the date and ERSTE IMMOBILIENFONDS and LLB Semper as the range between their adjacent reporting dates (30 April/31 October and 30 June/31 December respectively); no interpolation is used anywhere. Translations from the German are mine

A leverage ratio is a fraction, and a fraction can rise for reasons that have nothing to do with each other. Between mid-2022 and mid-2026, the credit leverage of euro area real estate investment funds — loans and deposits received on the funds’ own balance sheets, as a share of total assets, from the ECB’s investment fund statistics — rose in Austria from 5% to 15%, in Portugal from 13% to 19%, in France from 4% to 9%, and in Finland from 21% to 25%. A league table built on those endpoints would tell you almost nothing. Decomposed into what actually moved — net borrowing transactions, other changes in the loan stock, and the balance sheet underneath — the same four numbers turn out to describe three distinct mechanisms, and one country mixing them.

Austria borrowed. The sector’s loan stock rose two-and-a-half-fold between mid-2022 and its peak in late 2024, and the increase matches the cumulated net loan transactions exactly: all of it is new credit rather than a valuation or reporting artefact. The ratio had help — the shrinking balance sheet underneath contributes about a quarter of the ten-point rise, the borrowing the rest — but the numerator moved on transactions alone. Finland did the opposite. Its loan stock fell, but total assets fell faster — a mix of net asset-side transactions (roughly 40% of the ex-reclassification decline) and revaluations and other changes (roughly 60%) — so the ratio drifted up while the numerator shrank. And Portugal’s rise mostly is not a transaction story: 92% of the increase in its loan stock arrived without net borrowing, in jumps concentrated in fourth quarters — the signature of reclassifications and reporting-population changes rather than of funds levering up, an inference from the pattern, since the statistics do not label their adjustments; what the transaction split does establish is that the jumps were not recorded as net borrowing by the funds in the population — the borrowing history of whatever entered it is another matter. Portugal sits out the rest of this note. France is the mixture: roughly a third of its rise is net borrowing, with most of the remainder arriving through the same kind of non-transaction additions.

That leaves the interesting case, and a question I want to be careful with.

Vienna, quarter by quarter

Austrian open-ended real estate funds spent 2022 to 2026 in a redemption shock. Net fund flows turned negative in 2022-Q3 and stayed negative for every quarter since — €5.5bn of cumulative net outflows against a sector that started at €12.6bn of total assets. Gross redemptions climbed from around €200m a quarter to €423m in 2022-Q4 and €748m in 2023-Q4. In October 2023, LLB Semper Real Estate suspended redemptions; its management company gave notice in April 2025, and the fund has been in liquidation since 24 October 2025.

The sector’s borrowing moved over the same window. The loan stock went from €627m in 2022-Q2 to €1.6bn at its 2024-Q3 peak, and the quarterly pattern is worth stating precisely: net borrowing was largest in 2023-Q4 (+€361m), the quarter in which gross redemptions peaked and net outflows were deepest at once; the net increase equalled 53% of the quarter’s net outflows. That is a peak coincidence, not a lockstep — borrowing dipped in 2023-Q2 while redemptions stayed above €500m, then jumped in 2023-Q3 as redemptions eased. Two further facts sharpen the picture. The borrowing sits entirely in open-ended funds: closed-ended Austrian property funds report loans of at most €1m in any quarter on record, and one per cent of the sector’s gross redemptions over the period. And the lenders are almost entirely domestic banks, whose loans to the sector went from €429m to €1.4bn at peak and stand at €1.07bn of the sector’s €1.24bn total loan stock now, the stock declining since early 2025. The decline is net repayment — from the 2024-Q3 peak onwards, every quarterly change in the sector’s loan stock equals its net transactions exactly, leaving no room for a reporting-population break.

What these aggregates cannot say is what any of the borrowed money was for. The funds that drew credit need not be the funds that faced the redemptions; a sector series would look identical if unstressed vehicles borrowed for acquisitions, or took debt that had sat at property level onto their own balance sheets at refinancing — the years of the European property repricing were heavy with such rollovers — while stressed ones sold assets and gated. New net credit it is either way: the transaction match rules out mere repricing of loans already on the books. The timing, the open-ended concentration and the lender identity make the liquidity reading plausible, and it is the one I lean to — with a caveat the calendar forces: the best-known stressed vehicle spent the peak quarter gated, so the redemptions that peaked around it were mostly other funds’, and whose the borrowing was, the aggregate cannot say. They do not rule out acquisition or onboarding borrowing, and nothing in this note should be read as a claim about the use of proceeds.

What can be said turns out to close, once the statistics’ own conventions are respected. Net outflows of €5.5bn are the issues-less-redemptions measure, gross of the income the funds kept earning through the run. In the transaction accounts, retained income is booked as if paid to holders and reinvested in their units, so the shareholder outflow net of it — measured as transactions in fund shares — is €2.45bn; the roughly €3.0bn between the two measures is that retained-income adjustment, an accounting attribution, and nothing in this note treats it as a liquidity tool anyone reached for. It is on that measure that the identity closes to within €17m: the deposit book gave up €1.3bn — 92% claims on banks, 91% of it short-maturity — net disposals of everything else came to about €550m, gross sales and purchases unobservable behind the netting, and €614m of net new borrowing came in. Sizes of sector aggregates, not funding shares of any fund. Against the €5.5bn cash measure, borrowing is 11%; against the €2.45bn financial-account counterpart, about a quarter, with the deposit book near half — different denominators answering different questions, and neither a claim about use of proceeds. Two timing facts sit on top. The deposit book ran down early, some €900m of it by mid-2023, before net borrowing accelerated — though sector timing cannot show a single fund spending cash before borrowing. And the net borrowing was concentrated: equal to 53% of net outflows in 2023-Q4 — the deepest outflow quarter and the gross-redemption peak at once — 29% of cumulated outflows to the loan stock’s 2024-Q3 peak, and 11% over the window; the gross loan draws and repayments behind those net figures are unobserved. The path since is the shape a buffer would produce: the flexible instrument drawn hardest in the single worst quarter, then run down — fallen in six of the seven quarters since the peak, with a renewed €114m net draw in the latest. Whether that is a buffer working or something else is a hypothesis, and this note treats it as one.

One more discipline check belongs here. Across eight countries and twenty-nine quarters, no average contemporaneous relationship between property funds’ net borrowing and their net fund flows is detectable — the panel coefficient is indistinguishable from zero, though a null in eight clusters is weak evidence of absence — and in Germany and Finland the correlation is positive, borrowing accompanying growth rather than outflows. Restricting to outflow quarters produces the expected negative sign, though a split chosen that way is a descriptive exercise rather than a test. The Austrian episode is distinctive. It is a case, not a law.

The long view, and the boring case

Stretch the window back a decade and the cross-country picture reorganises itself around a different fact: since the shock, the jurisdiction with the highest-profile leverage policy has been the quietest line on the chart — quiet on the fund-balance-sheet series charted here, which for Ireland sits furthest from the total-debt measure its policy is written in, so this is a statement about the series, not about the policy variable’s path.

Irish property funds built their credit leverage in the mid-2010s — peaking at 37.7% of total assets in 2017-Q3, on the loans-and-deposits measure used throughout this note — and have been grinding down since, to 26.6% now. The decline predates the Central Bank of Ireland’s November 2022 measure by five years, so the chart is not evidence that the limit caused the deleveraging. What the measure did do, as I wrote in Ireland’s emerging macroprudential framework: two cases, one template, is codify a perimeter and a direction of travel: a 60% total-debt-to-total-assets limit for Irish property funds under Article 25 of the AIFMD, with a five-year phase-in that completes on 24 November 2027. Total debt is a broader concept than the loans and deposits charted here — shareholder loans have long rivalled bank loans on Irish property fund balance sheets — so the Irish line on the chart is not the ratio the limit is written in, and the two should not be read against each other. On its own measure, the CBI reports sector leverage of 47% at end-2024, down from 49% a year earlier: the gap between the two definitions, in one pair of numbers. The calibration was explicit about where it sat in the distribution — the CBI’s framework document notes the proposed limit “would lie above the 90th percentile of leverage across all EU real estate funds”, while the most leveraged tenth of Irish funds stood above 93% — and explicit about the pace, with the implementation period lengthened “to facilitate a gradual and orderly adjustment to the measures”. Ireland, in other words, made two deliberate choices: a limit above the bulk of the distribution, and a long adjustment path for the leveraged tail the limit was built to bind. That matters for how to read everyone else.

Austria’s line is the mirror image: a decade parked at 5%, then the near-vertical segment of 2023. Finland’s drifts up as the sector shrinks. Germany’s barely moves — 16% to 18% — but the level hides the largest absolute borrowing in the sample: German open-ended funds — an aggregate that nets retail outflows against Spezialfonds inflows, so none of what follows attributes to stress — added €9.2bn of net loans between mid-2022 and mid-2026, alongside €10.1bn of net outflows and asset-side transactions of −€6.5bn. I am deliberately not assembling those three numbers into a story. They do not form a closed identity (valuations, cash and other positions move too), and the German aggregate mixes retail vehicles with the institutional Spezialfonds that dominate it — roughly speaking, the retail funds tracked by Scope run some €95bn of fund assets on a NAV basis against the sector’s €433bn of balance-sheet total assets — different measures, but the mix rather than the ratio is the point, and the BVI’s segment statistics put the retail funds in net outflow for a second consecutive year in 2025 while the Spezialfonds continued to take money in. The asymmetry with Vienna is earned: Austria’s aggregate is a single vehicle class — the open-ended funds are, to within rounding, the sector — with an identified domestic-bank lender base and a loan stock that reconciles to transactions exactly, while the German co-movement is an artefact of the mix until the mix is split. Until that split can be done properly on one statistical basis, Germany is a set of components worth watching, and the note leaves it there.

What the caps actually say

Each of these jurisdictions writes a borrowing limit for at least part of its fund population, and the limits are less comparable than any cross-country chart implies — different numerators, different denominators, different vehicle populations. Austria’s real estate fund law caps borrowing and encumbrances together at 50% of the market value of the fund’s real-estate assets — the base excludes the liquidity portfolio. Germany runs two regimes: retail open-ended funds may borrow up to 30% of the value of their properties, while the Spezialfonds that hold most of the sector’s assets sit under a separate provision allowing property loans up to 60% of property value. Finland’s cap — borrowing of at most half of the fund’s assets, four-fifths for company-form rental-housing vehicles — was consolidated into its AIFM Act at the end of 2025, and the same amendment extends it beyond the company-form vehicles it grew up with: a special investment fund investing mainly in real estate must now observe the borrowing rule too, which brings the open-ended funds in the statistical population inside the cap. Portugal’s CMVM regulation sets 25% of total assets for open-ended vehicles and 50% for closed-ended ones offered publicly, with privately placed vehicles uncapped. France’s numerical cap — 40% of the property assets, plus 10% of the rest — applies to the retail OPCI segment; the SCPIs that make up much of the French retail population borrow under ceilings their shareholder meetings are required to fix, with no statutory number. Ireland’s 60%, alone in this list, is a macroprudential measure adopted under EU law rather than a product rule — the AIFMD itself imposes no numerical cap, and Article 25’s power to set one has been exercised for property funds by one authority; the only other uses recorded in ESMA’s leveraged-AIF assessments — Ireland’s and Luxembourg’s Article 25 restrictions on sterling liability-driven funds, written as minimum yield-buffer requirements — are a different instrument for a different problem.

I am not going to line these numbers up against the sector ratios above, because the comparison would be false precision: the ratios in this note are loans over total assets, and most of the caps are not — different bases, different populations. Nor can sector averages say whether any individual fund approached its own limit; averages can hide a distribution’s tail, and the distribution is not published. What is observable is narrower still: whether any single fund neared its own cap, no one outside the supervisory returns can say — and nothing here shows the statutory ceiling was ever the operative constraint, either: fund rules, lender covenants and credit supply bind long before a product-law maximum does. And the borrowing that did occur — in Austria’s case, raised hardest in the worst quarter of a redemption shock, the stock then falling in six of the seven quarters after its late-2024 peak — appears in none of these frameworks’ statistical or supervisory disclosures — the funds’ own annual reports are its only public trace.

The data that would answer it

Which is the point on which this note ends, because the question the Austrian episode poses is answerable, just not from where I sit.

What the borrowing did — cushion the funds, increase the banks’ exposure, or, in the way of secured lending, both at once — depends on facts that sit below the aggregates: which funds drew the credit, against what collateral, at what valuations, and whether the same funds faced the redemptions. Collateral, recourse and loan-level valuations are not public. Who borrowed and who bled largely is: Austria’s public-offer property funds are a handful of vehicles whose annual reports show borrowing and flows fund by fund, and assembling that layer — who borrowed, who bled, fund by fund and year by year — is the natural next piece of work, deliberately not attempted in a note built on the quarterly aggregates. What no assembly of public reports would yield is the cross-country distribution, and much of what would build that is collected, though not in one place or on one measure. Fund-level leverage flows to supervisors under AIFMD Annex IV on the directive’s own measures — gross and commitment leverage, rather than the ratios the national caps are written in; those sit, where they are collected, in national returns, as in Ireland’s Article 25 collection, whose aggregates are published. The ESRB’s 2025 NBFI Risk Monitor devoted a special feature to EU banks’ credit to real estate funds, drawing on supervisory data on bank–fund exposures that sits outside the published statistics. The distribution of distance-to-cap — the one chart that would locate the slack, fund by fund, and show how close anyone came to a limit — could in principle be assembled from what supervisors hold — how much joining of Annex IV returns with national collections it would take, and, for the caps written against property values, how much valuation data besides, only the authorities can say. On its own it would still leave the use of proceeds open; what it would do is replace inference about headroom with a distribution. To my knowledge, no one has published it.

There are observations worth naming in advance, as things to watch rather than tests that select a reading — because the readings are not mutually exclusive. Secured borrowing that met redemptions cushioned the funds and increased the banks’ exposure to the same property in the same motion; how much risk actually moved depends on collateral, recourse and loss incidence, none of which the aggregates carry. The things to watch are the loan stock’s path as the liquidations complete, Austrian banks’ provisioning against fund exposures, and whether drawn leverage persists once outflows end — the last of which could as easily mean property-level debt brought onto fund balance sheets as redemption funding. The record to date — a loan stock down in six of seven quarters, then a fresh net increase in the latest one — is short, commercial property valuations are still finding their level, and a note built on the distinction between what aggregates show and what they suggest should not end by blurring it.

The redemption side of this story — gates, notice periods and the rest of the liquidity toolkit — is its own subject, and one where the German record already has a published verdict (see Twelve years of evidence: did Germany’s OEIF macroprudential template work?); Austria follows the same route in 2027, when an amendment already on the books introduces a minimum holding period and a twelve-month redemption period for its real estate funds. The leverage side, on the evidence here, resolves into something plainer than a warning: the ratios moved for different reasons in different places, and only in Austria was new borrowing the main driver of the ratio — 11% of the window’s cash outflows, about a quarter of the financial-account outflow net of retained income, half of the single worst quarter’s. What remains open is not whether the slack was used. It is who it served, and the numbers that would tell us are already sitting in the returns.

Paweł Fiedor — The Macro Prudential View

Views are my own. This article is a personal analytical piece based on public sources. It does not represent the views of the European Systemic Risk Board, the Central Bank of Ireland, the Eurosystem, or any other institution.

Legal texts

ImmoInvFG § 5 — the Austrian combined borrowing-and-encumbrance limit, read with § 21 for the base.

KAGB § 254 and § 284 — the German retail 30% and Spezial-AIF 60% property-loan provisions.

Finnish AIFM Act (162/2014), ch. 16a § 6 and ch. 18a § 18, as amended 30 December 2025 — the half-of-assets cap and its extension to real-estate special investment funds.

CMVM Regulation 7/2023, Arts. 19–20 — the Portuguese 25% / 50% endividamento limits.

Code monétaire et financier, Arts. L214-39 and L214-40 — the OPCI 40% and 10% borrowing limits; AMF General Regulation, Art. 422-225, for SCPI borrowing ceilings.

Directive 2011/61/EU (AIFMD), Art. 25, and Delegated Regulation (EU) No 231/2013, Annex IV — the leverage-limit power and the supervisory reporting template.

Regulation (EU) No 1073/2013 (ECB/2013/38) repealed with effect from 1 December 2025 by Regulation (EU) 2024/1988 (ECB/2024/17), which governs the last three quarters of the vintage used here — investment fund statistics: solo, fund-by-fund reporting, which is why property-company debt enters the series only when onboarded.

Official reports

Central Bank of Ireland (24 November 2022), The Central Bank’s macroprudential policy framework for Irish property funds — calibration quotes at pp. 8 and 21 — and Financial Stability Review 2025:II for the total-debt leverage figures.

ESMA (May 2026), Annual risk assessment of leveraged AIFs in the EU — the Article 25 inventory, pp. 12 and 22.

ESRB (September 2025), EU Non-bank Financial Intermediation Risk Monitor 2025, Special Feature 2.4.

OeNB, Financial Stability Report 49 — the Austrian redemption wave and the 2027 holding- and redemption-period amendment.

LLB investor information — the Semper Real Estate suspension, termination and liquidation dates.

BVI segment statistics for the German fund flows; Scope Fund Analysis for the German retail-fund volume (NAV basis).

Data

ECB Investment Fund Statistics (IVF), 2026-Q2 vintage — all balance-sheet series: loans and deposits received (stocks and transactions, all maturities and counterparts), total assets, fund-share issues, redemptions and transactions, and asset-side deposits and loan claims; the open-end/closed-end split uses the IVF sub-sector series (4A/4B). The leverage numerator is the liability item “loans and deposits received” as reported on fund balance sheets — investment funds take no deposits in any ordinary sense, and 86% of the Austrian stock is owed to domestic banks, so the text calls it loans.

Calculations. All decompositions, reconciliations and shares are computed by The Macro Prudential View from the IVF series. Chart 1’s decomposition is exact and additive: Δ(L/A) = F/A₁ + (ΔL−F)/A₁ + L₀(1/A₁−1/A₀), with F the cumulated net loan transactions; no residual. Net fund flows are the IVF net-issues measure; measured as ESA transactions in fund shares — which treat retained income as reinvested — the Austrian shareholder outflow is €2.45bn, which is how the identity in the text closes (residual €17m); the €3.0bn between the two measures is the retained-income imputation (ESA D.443), a statistical attribution rather than an observed cash flow. The panel estimate regresses quarterly net loan transactions on net fund flows, both scaled by lagged total assets: eight countries (Italy excluded for semi-annual reporting), 2019-Q2 to 2026-Q2, country fixed effects, standard errors clustered by country; with eight clusters the t-statistics are indicative, so the null was re-checked with time fixed effects and an exact wild cluster bootstrap (p≈0.4 either way), and lagged fund flows at one and two quarters show no relationship either. Series keys, per-pull checksums and all computations are in the article repository; working files available on request.

Most people who cite a monitoring framework never read its annex. That is not a criticism — annexes are long, and the point of a framework is that you can use its output without reconstructing its construction. But the construction is where the judgements live, and in the case of the ECB–ESRB geoeconomic report the workstream wrote enough of them down that you can follow the reasoning from 38 candidate indicators to the handful that actually appear.

What follows is that walk-through. It is descriptive. I am not going to argue that the framework should have done something else, partly because the case for that is weaker than it first looks, and partly because the interesting thing here is how much of the design turns out to be visible once you go looking — and how much doesn’t.

The occasion for looking was a small question I could not answer. On 31 July the Federal Reserve Bank of New York sold euros to buy yen on behalf of the US Treasury, in the first joint US–Japan operation to support the yen in nearly thirty years. Japan’s Ministry of Finance confirmed three days later that it had purchased yen in coordination with the US Treasury that day. The Financial Times reports that the ECB was told after the trade had been executed, and that Christine Lagarde and Treasury Secretary Bessent spoke the following day; it also reports senior ECB officials describing the use of euros as an unprecedented breach of longstanding conventions between western monetary authorities. A Treasury spokesperson told the paper that decisions on allocating the Exchange Stabilization Fund are made by the Treasury, are not coordinated with foreign authorities, and that reserve assets within the fund had been reallocated that week.1 No transaction amount has been published.

Set the diplomacy aside — others are better placed to weigh it, and I have no standing to. What it left me with was a narrower and duller question. If the euro’s position in the international monetary system is the kind of thing that turns out to matter on a Friday afternoon, what does the euro area’s own geoeconomic framework say about it? Specifically: what share of allocated official foreign exchange reserves is held in euro, and which way is it going?

The framework has a series for that. It sits in the report’s main indicator table. It appears in neither of the two derived outputs the report produces. Following that led into the selection process, which turned out to be the more interesting object.

The episode, briefly, and why it does not carry weight

Across the fixes from 29 July to 3 August the yen appreciated 4.5% against the dollar and 3.1% against the euro, with EUR/USD up 1.4%.2 Those are large moves and they are broad yen appreciation rather than anything euro-specific. A euro sale would, other things equal, push EUR/USD down rather than up; that it rose tells you nothing either way without the transaction size and the other flows in the market that week, which is roughly the point.

The reported rationale for using euros rather than dollars is that dollar sales could have been read as an attempt to weaken the currency, against the Treasury’s stated strong-dollar policy. Analysts have separately estimated, from provisional Bank of Japan data, that Japan alone may have spent around ¥13.8tn over two days; no figure has been published for the US side.

Attribution of the price action is beyond what daily data can support, and the fixing calendar is the reason. ECB reference rates come from a concertation procedure normally conducted at 14:15 Central European Summer Time, which is 08:15 in New York — the start of the New York session rather than through it. An execution for the Treasury on Friday 31 July would more naturally appear in Monday’s fix than Friday’s. Monday is indeed the larger single move. None of that identifies anything, and I will not pretend otherwise. The episode is the reason I opened the annex, not evidence for anything in it.

Thirty-eight candidates, two outputs

The ECB and ESRB published Financial stability risks from geoeconomic fragmentation on 22 January 2026, with a separate technical annex. The report classifies geopolitical risks into six categories — general, military conflict, infrastructure, trade, capital and finance, and politics and society — and assembles 38 indicators across them, drawn from news-based indices, model-based estimates, official statistics and event counts.

From that pool, two things get built, and they are parallel rather than sequential. Box 2 recommends sixteen indicators for closer monitoring, and Figure 4 displays fifteen of them as a heatmap, the drop of one being the local-perception geopolitical risk indicator, set aside as capturing country-specific rather than regional or global risks while being retained in the recommended set. Separately, the workstream augments the ESRB growth-at-risk apparatus with geopolitical variables, using a LASSO-regularised quantile regression to select among candidates, with a benchmark specification taking one indicator per topical category — five indicators covering five of the six, with infrastructure unrepresented. An indicator can be in one without being in the other; neither is a filtered version of the other.

These are exhibits in a report recommending a monitoring set, not a live dashboard someone opened on 31 July. Worth saying plainly, because it changes what the omissions mean.

Both outputs are thinner in official statistics than the pool they draw on, and they are thinner to very different degrees. Official statistics are eight of the 38, two of the sixteen recommended for monitoring, and none of the five in the growth-at-risk benchmark. News-based indices run sixteen, nine and four. The monitoring set keeps some official statistics; the modelling benchmark keeps none.

The growth-at-risk end of that has a straightforward candidate explanation. The machinery estimates the conditional distribution of future GDP growth, so relevance for GDP tail risk is the criterion. Why news-based indices dominate the result is a further question the annex does not settle: they move fast and load on the same news that moves markets, but regularised selection among correlated candidates can also be sample-dependent, and a set of news indices proxying one latent factor could crowd out slower series without any of them being economically preferred. A measure of how much euro is held as reserves abroad is not built to forecast GDP four quarters ahead.

The shortlist end is more interesting, because the report shows its reasoning.

What Box 2 actually says

Box 2 justifies each recommended indicator in turn, and thirteen of the sixteen are argued on predictive relevance at a stated horizon — the report’s own term, and worth keeping rather than upgrading to forecasting performance, since what the annex reports is relevance within its estimation exercise rather than out-of-sample evaluation. The Economic Policy Uncertainty Index has predictive value at short horizons of up to one year. The World Uncertainty Index shows predictive strength up to a year. The Trade Policy Uncertainty Index is a strong predictor at the one-year horizon. The JLN Financial Uncertainty Index has predictive value up to a year. The Migration Fear Index has predictive relevance up to eight quarters, the Sanctions Intensity Index over the same horizon, the Elections EMV Tracker up to a year. The Common Volatility Index has weaker but still notable relevance at quarterly horizons. The Global Supply Chain Pressure Index showed predictive relevance only when the COVID-19 period was excluded from the estimation sample, which is a candid thing to write down. Four more are argued the same way: the JLN Real Uncertainty Index provides short-term signals for tail-risk assessment, the local-perception geopolitical risk indicator has predictive strength most evident at a year and longer, the National Security EMV Tracker shows more relevance at one-year horizons than in the short term, and the Financial Fragmentation Index shows some predictive strength at one year in first differences.

Three are not argued that way, and the Box says so openly.

The Geopolitical Risk Index is retained while the Box notes it does not show high predictive power for economic activity, possibly because its data origins lie in Anglo-Saxon news sources. The Significant Cyber Incidents Indicator is retained while the Box notes its empirical use is limited owing to its short time series, on the ground that it captures an increasingly important dimension of modern conflicts. And the Trade Openness Indicator is admitted with no horizon and no predictive claim at all, described as a widely used official statistics measure that reflects both the vulnerabilities of countries to geopolitical events and the reactions to these events.

So predictive relevance is not a hard gate. Indicators enter this shortlist when the workstream judges the thing they measure to matter, and it says which ones those are and why. That is a defensible way to build a monitoring set, and it answers an objection a purely statistical selection would run into: a set chosen only on measured relevance drops exactly the risks that have not yet materialised. It is also, as documentation goes, unusually honest.

The series that isn’t there

Which brings back the question I started with.

The Allocated Foreign Exchange Reserves indicator tracks reserve currency composition from IMF COFER data. It is in the report’s main indicator table. It is quarterly, so the frequency reasoning the annex applies to annual series — military expenditure, the sanctions count, capital restrictions, UN voting alignment — does not reach it. On the published record it appears in none of the growth-at-risk screening tables: not the multicollinearity summary, not the unit root tests, not the LASSO relevance table, not the model-fit ranking. The annex does not state whether it was among the candidates screened at all. And it is not among the sixteen.

I cannot tell you why, and neither can anyone else working from the published documents. What I can do is set out the reasons a workstream might reasonably have left it out, because they are not hard to find.

A COFER share is a stock whose period-to-period changes mix valuation with transactions: the euro share can fall because holders sold euro, or because the dollar rose and revalued everyone’s holdings. It identifies no holder, no instrument, no counterparty. It is a global aggregate, not a euro-area exposure. Its denominator is allocated foreign exchange reserves — the portion of foreign exchange reserves whose currency composition reporting authorities disclose. Unallocated reserves are foreign exchange too, just undisclosed; gold, SDRs and the IMF reserve position sit outside foreign exchange reserves altogether. Those are plausible considerations. What the record does not show is their weight, or whether any of them was applied, or whether the series was put to the shortlist test at all. Absence from a published table shows non-appearance, not rejection.

One candidate explanation can be ruled out, which is worth noting for what it says about the others. A display of levels cannot show a risk that materialises as a decline. True — and true of trade openness, which is on the heatmap anyway, handled in ordinary dashboard practice by showing the change alongside the level. Whatever kept the reserves series out, it was not that.

For the record, the series itself gives a straightforward answer to the question I started with. The euro is a shade over 20% of allocated official foreign exchange reserves as at the first quarter of 2026, and it has sat between roughly 19% and 20.6% every quarter since 2017 — remarkably flat through the pandemic, through 2022, and through everything since. Whatever else is true, the number is not doing anything dramatic. That is worth knowing, and the framework would not have told you.

What the walk-through does show is how much the answer depends on judgement that a reader outside the workstream cannot reconstruct. Thirteen are in on reported predictive relevance, and the Box gives the horizons. Three are in because someone decided the thing they measure matters, and the Box says so. The rest are out, and for most of them the record is silent about which of those two tests they failed.

Coverage and timing

Two descriptive patterns are worth having in front of you, offered as description rather than diagnosis.3

The first is timing. One indicator updates daily, a model-derived common volatility factor. Eighteen update monthly, twelve of those news-based indices. Fifteen are quarterly, four annual. No series drawn from official statistics updates more often than quarterly, and the three that record an instrument being applied — capital restrictions, military spending, the count of active sanctions — are annual.

That association between source type and frequency is real and unsurprising: news and market data are generated continuously, official statistics are compiled on a calendar. It is not evidence of a selection preference by itself, and the annex is explicit that frequency constraints informed which indicators were excluded from the final empirical subset. Annual frequency genuinely is a problem for a quarterly-horizon model, and saying so is not a criticism.

The second is coverage. Military conflict carries nine indicators in the main table, trade six, sanctions two, capital and finance eight. Of the capital-and-finance eight, four reach the heatmap: two uncertainty and volatility measures and two cross-border flow ratios. The two official-statistics series in the category — the Capital Restriction Index and the reserves indicator — reach neither derived output.

One caution about reading counts. The annex reports that most military and infrastructure indicators were not selected at any horizon in the growth-at-risk work, adding little beyond the core controls. The richest-covered category contributes the least. Coverage and usefulness are different axes, and a bigger row count is not a better category.

The thing in the annex

The most interesting object in either document is not in the indicator set at all.

Box A.1 constructs a financial position-weighted effective exchange rate, following Lane and Shambaugh, weighting bilateral exchange rates by their importance in cross-border financial positions rather than in trade. The construction matters as much as the result. Because the weights come from cross-border positions, the exercise can be run separately for the asset side and the liability side, and separately by financial instrument and by holding sector. That is what turns an exchange rate into something you can say things about institutions with.

Applied to German data, around 40% of the asset side and 20% of the liability side of the international investment position are denominated in foreign currency, with the dollar accounting for more than half of both. Germany is a net creditor with a large share of foreign-denominated assets, so the asset-based indicator is far more sensitive to an exchange rate shock than the liability-based one. Splitting by sector, the net foreign-currency creditor position comes primarily from non-bank financial intermediaries, while the monetary financial institutions’ balance sheet appears currency matched.

Then the instrument breakdown does something the sector aggregate could not. The box reads the high volatility of the indicator for short-term bonds held by MFIs as revealing their short-term funding needs in foreign currency — a conclusion that only becomes visible once the measure is cut by instrument rather than by sector. On the non-bank side, the higher variability of asset-weighted rates, driven by equity holdings, may amplify the transmission of a geopolitical shock. Set alongside evidence that non-banks have been accelerating their hedging of these exposures through banks using short-term FX derivatives, the box says this supports calls for MFIs to have arrangements in place for swift access to central bank cross-currency swap lines.4

That is a different kind of object from anything in the indicator set. An index tells you a level. This starts from a currency movement, attaches it to positions, splits those by sector and by instrument, and arrives at a specific institutional exposure and the facility that would matter if it bound. It is partial — it does not identify a state’s decision to act, or quantify losses — but it runs from a quantity to who is exposed to it, which no row on a heatmap does.

It was built by the authors of the report, demonstrated on one country’s data, and left in an annex.

Extending it is not one project but two, and they should not be quoted at the same price. A macro version at aggregate or sectoral level would draw on international investment position currency breakdowns and existing securities-holdings and portfolio-investment statistics; whether those suffice is a data-gap question I have not tested. A holder-level map, which is what would actually attach a currency move to identifiable institutions, is the harder object and the one where look-through and harmonised reporting bind. The German case is unusually well served by its statistics. The method is written; the feasibility of each version beyond Germany is an exercise someone should run.

What the walk-through leaves you with

There is a supervisory exercise that runs the other way round from a scenario test, and it was published on the same day as the reported operation. On 31 July 2026 ECB Banking Supervision released the results of its thematic reverse stress test on geopolitical risks, covering 110 euro area banks under its direct supervision. Banks were given a fixed adverse outcome — 300 basis points of CET1 depletion — and asked what geopolitical path would get them there. That is the inverse of the biennial EU-wide exercise, where every bank faces the same scenario and the question is the size of the capital impact. Structure enters as the transmission each bank has to specify to make its own scenario work, not as the starting point.5

Banks produced narratives rather than readings: military conflicts, trade and energy disruption, sanctions, macroeconomic shocks, cyber incidents, each tailored to a bank’s own business model. The ECB found the scenarios generally economically meaningful, and also found weaknesses — among them, that several banks produced only a muted response in their liquidity metrics despite the significant capital decline the scenario specified, which the ECB will follow up because solvency and liquidity stress are closely intertwined in a crisis.

I offer that as an existence proof rather than as an argument. The euro-area official sector does ask structure-to-outcome questions; it is not the case that only gauges exist. What is not evident from the published material is any connection between that exercise and the indicator framework — between the scenarios banks wrote in July and the series the January report recommends monitoring.

Which is where the walk-through actually lands. The report’s own conclusion says gaps remain, that indicator availability and comparability vary across countries and over time, and that it could be worthwhile to complement indicator-based monitoring with more elaborate scenario analyses involving specific risk materialisations. Read against the annex, that is a more pointed sentence than it looks, though it is worth being precise about what would answer it. Box A.1 is not scenario analysis; it is a structured exposure measure, a third kind of object alongside the indicator set and the July supervisory exercise. The report asks for scenarios. The annex demonstrates an exposure metric. The reverse stress test elicits bank-specific narratives. Three different instruments, none of which is a longer indicator list, and only one of which the January report actually built.

The broader point is duller and, I think, more useful. A monitoring framework is a sequence of judgements, most of which never reach the people who read its outputs. This one wrote down more of them than most — the horizons, the failures, the three admissions on grounds other than performance, the reason a country-specific index was dropped from a global display. Anyone using the output is inheriting those judgements whether or not they have read the annex. That is not an argument for reading annexes. It is an argument for the workstream having written this one the way it did.

Paweł Fiedor — The Macro Prudential View

Market data are as at 10 August 2026.

The author works at the Central Bank of Ireland and in the ESRB Secretariat. This piece is written in a personal capacity; views are the author’s own and should not be attributed to the Central Bank of Ireland, the ESRB, or the Eurosystem. It draws exclusively on publicly available sources and takes no position on whether the measures discussed should be adopted.

Sources: ECB and ESRB, Financial stability risks from geoeconomic fragmentation, 22 January 2026, in particular Box 2, Figure 4 and its note, and Table 1 — https://www.esrb.europa.eu/pub/pdf/reports/esrb.report202601_financialstabilityrisks.en.pdf; and the accompanying technical annex, including Table A.1, the growth-at-risk selection tables and Box A.1 — https://www.ecb.europa.eu/pub/pdf/other/ecb.report202601_financialstabilityrisks_annex.en.pdf; European Central Bank, euro foreign exchange reference rates, daily series — https://www.ecb.europa.eu; International Monetary Fund, Currency Composition of Official Foreign Exchange Reserves (COFER) — https://data.imf.org/en/datasets/IMF.STA:COFER; Japan Ministry of Finance, statement on coordinated yen purchases, 3 August 2026 — https://www.mof.go.jp/english/public_relations/statement/others/20260803073000.html; Reuters, “US Treasury Secretary Bessent will do whatever it takes to support Japan”, 4 August 2026 — https://www.reuters.com/world/asia-pacific/us-treasury-secretary-bessent-will-do-whatever-it-takes-support-japan-2026-08-04/; Financial Times, “US euro sale to prop up yen blindsided ECB”, August 2026 — https://www.ft.com/content/d9922d0b-51a0-48be-811b-42e08f90985a; ECB Banking Supervision, “ECB publishes results of 2026 geopolitical risk reverse stress test”, 31 July 2026 — https://www.bankingsupervision.europa.eu/press/pr/date/2026/html/ssm.pr260731~93964644b0.en.html; P. R. Lane and J. C. Shambaugh, “Financial exchange rates and international currency exposures”, American Economic Review, 2010. Chart data as cited in each figure.

Every Parameter But the Number

Rehden, in Lower Saxony, is Germany’s largest gas storage facility. Its operator puts its working gas capacity at around four billion cubic metres, close to a fifth of all German storage. On the gas day of 9 September the larger of the two AGSI records covering it, 35.7 TWh of working gas, was 9.3% full.

German law requires each named facility to reach a level of its own, and for Rehden that level is 45% on 1 November. The requirement quoted in winter previews is 90%. That one is also in force and also applies to Germany, as directly applicable EU law. The two have coexisted since May 2025, and I have not found a public document that reconciles them.

One caution before any of it. The legal floor attaches to the facility; the public data reports marketed products. Rehden’s second record is pooled with another site, so its stock cannot be separated out, and nothing below establishes where the facility Rehden stands against its 45%. Every Rehden number in this piece is the larger record’s.

Two instruments

The EU obligation sits in Article 6a of Regulation (EU) 2017/1938 as amended last year. It requires each Member State to reach 90% of the aggregated capacity of the underground storage facilities “located on their territory”, at any point between 1 October and 1 December. The unit of obligation is the Member State. The EU aggregate figure, 67.50% on 9 September, is a security-of-supply indicator and not the legal test of anything.

The German instrument is the Gasspeicherfüllstandsverordnung of 5 May 2025, made by the economics ministry under § 35b(3) of the Energiewirtschaftsgesetz and expiring on 31 March 2027. It sets no national figure at all. It requires that “in each of the named storage facilities” a level be held, expressed as a share of that facility’s own working gas volume, at two dates:

On 1 November, 80% at every facility except six — Bad Lauchstädt, Frankenthal, Hähnlein, Rehden, Stockstadt and Uelsen — which owe 45%. On 1 February, 30% everywhere except Bierwang, Breitbrunn, Inzenham-West and Wolfersberg, which owe 40%. The six at 45% are carved out, in the regulation’s own words, on account of “their markedly reduced injection and withdrawal rates and their geographic location”. The four at 40% are given no reason at all.

Set that against what the statute itself says and the shape of the change appears. Section 35b(1) puts Germany’s own levels at 80% on 1 October, 90% on 1 November and 30% on 1 February — the November figure being the EU figure. The decree deviates from all of it, and not in one direction: November comes down, October disappears, and February goes up for four sites.

Reporting converts this into a German national target of about 70%. No such number exists in the instrument, and the conversion is not innocuous. Each facility owes its own floor, so a full site does not discharge an empty one, and the weighted average binds nobody. I can bracket it between 68% and 73%, depending on how two pooled AGSI records divide between named and unnamed facilities, but I cannot reproduce a point estimate, and neither, as far as I can tell, has anyone who has published one.

The two instruments do not conflict, because they bind different people. The EU obligation sits on the Member State. The German one runs through § 35b of the Energiewirtschaftsgesetz, which puts the duty on the operator of each facility — but the duty is to write contract terms creating the conditions for the level to be reached, and the facilities “should” reach it. Working gas is held by storage users, not by the operator, so the obvious question is what an operator does when its users have not delivered.

The statute answers it. Under § 35b(5), where the levels cannot be reached because a user is not using working gas volume it has firmly booked, the operator must hand those unused capacities to the market area manager, pro rata to the extent of non-use, along with a proportional share of injection and withdrawal capacity. The user goes on paying for the capacity it booked, minus the variable injection charges. That is the plumbing connecting a facility floor to a state buyer, and it is what stops the German floors being a mere exhortation.

Read § 35b(1) before any of this and the sequence sharpens. The default German levels are 80% on 1 October, 90% on 1 November and 30% on 1 February. The 1 November figure is the EU figure. The Gasspeicherfüllstandsverordnung deviates from all of it: 80% on 1 November, 45% at the six named sites, and no 1 October checkpoint at all.

What EU law permits

Two things moved in 2025 before any derogation is reached. The target may now be met at any point between 1 October and 1 December rather than by a fixed 1 November. And the filling trajectories that functioned as intermediate checkpoints became “a series of indicative intermediate targets” that Member States “strive to follow”, with no binding obligation at EU level.

The first is not a technicality this year. Hold the realised EU injection rate flat and the same physical path is 8.1 points short of 90% on the old fixed date and 0.1 points short at the close of the new window. On this year’s numbers the window is not a technicality at all: it is the difference between a Member State that plainly misses and one that arrives within a rounding error, having done nothing differently.

That 0.1 is a floor rather than a forecast. Holding a September rate through November is generous in a way the calendar makes obvious: injection slows as reservoirs fill, and in most years the last weeks before December add little or turn negative. The gap between the two dates is real; its size on any particular path is not.

Then the derogations. The terms matter here, because they are routinely summarised wrong.

Paragraph 5a lets a Member State decide to deviate by up to ten percentage points where “difficult conditions” limit its ability to fill. The European Parliament, describing its own text, put the trigger as difficult market conditions with “indications of speculation hindering cost-effective storage filling”. Paragraph 5b adds five more and says expressly that it applies “in addition to” 5a, so those cumulate to fifteen. It is available where national production exceeds consumption, or where “specific technical characteristics of an individual underground storage facility with technical capacity above 40 TWh” require an injection period beyond 115 days. Paragraph 5c adds a final five, granted by the Commission for a single filling season by delegated act. No such act has been adopted for 2026.

None of that is automatic. The target remains 90% unless a Member State decides to deviate and justifies the decision to the Commission, and 5c requires the Commission to legislate. The floor a Member State can reach if it takes every step available this season is 75%. The floor that applies until it does is 90%.

Nor does 90% mean the same thing everywhere. Article 6a(2) reduces the target, for Member States whose storage is large relative to their demand, to a volume worth 35% of average annual gas consumption over the preceding five years. The Netherlands is the obvious candidate in this region, and on that reading it owes a volume rather than a percentage, which converts to something below 90% of its capacity. I have not reproduced the five-year series the article specifies, so I am not putting a number on it. What survives without one: two Member States under a single headline figure, owing different fractions of their own storage before either has injected anything.

Note where 5b’s second limb points. It needs two things of the same site: technical capacity above 40 TWh, and a slow injection rate producing a period beyond 115 days. Rehden appears to clear both, though not from any single source. Its operator puts the facility at around four billion cubic metres of working gas, about 44 TWh. On the injection side, AGSI’s own declared figure for the larger Rehden record is 282 GWh/d against 35.7 TWh of working gas, which is 127 days to fill it from empty.

That second calculation is mine, not the operator’s, and “injection period” is not defined in the article. But the capacity limb is a problem for anyone reading the public data. AGSI reports Rehden as two marketed products, the larger 35.7 TWh, and neither crosses 40 on its own. The German carve-out and the EU derogation rest on the same physical fact about the same asset, and the public data reports that asset in pieces.

Paragraph 5f requires a Member State taking 5a or 5b to consult the Commission and justify itself immediately, so visibility across the stack runs from a private justification to a published delegated act. Nothing in the public record shows that Germany has taken either. The Commission invited Member States to use the flexibility, and an invitation is not evidence that anyone accepted it. The loosening that has demonstrably happened sits outside this ladder altogether.

The Commission asked

Three weeks after the Middle East disruption began, the flexibility stopped being theoretical. The Commission announced on 23 March 2026 that Commissioner Jørgensen had written to EU energy ministers. Its Gas Coordination Group note of 26 March records the ask: the letter invited “EU countries to make use of the flexibility provisions in the Gas Storage Regulation and to consider reducing their filling target to 80% as early as possible in the filling season”, to “provide certainty and reassurance to market participants”. The April note refers to refilling “to a minimum of 80% by 1 November depending on LNG supply availability”. The letter itself has not been published.

A mandatory buyer with a published target is a buyer the market can position against, and softening the target early reduces the size of a forecastable bid. That is the argument for releasing a countercyclical buffer at the onset of stress rather than waiting to see whether the stress binds. It is often right, and it is always available.

Where Germany actually is

German storage was 55.16% full on 9 September, and the country is injecting about 566 GWh/d, measured as the change in stock over the preceding thirty days. Hold that rate and Germany reaches 67.3% on 1 November and 74.1% on 1 December.

That leaves it 15.9 percentage points short of the EU’s 90% at the close of the window. Getting to 90% by 1 December means covering 34.8 points from where it stood on 9 September, which over those 83 days is 184% of the rate it is managing now. Uniper’s chief executive, speaking on 11 August, put it in commercial terms: “We can still get there, but we do have to see the market prices move to incentivize the filling of the storage.”

On 20 August the economics ministry said reserves at 60% to 70% would be sufficient for winter, conditional on extra imports. That is a further number for the same winter, from the department that wrote the facility list.

I would not defend the decimal places. A thirty-day August rate extrapolated forward is crude, injection slows as reservoirs fill, and a cold snap before the window closes would make it worse. What follows does not depend on the extrapolation.

What the aggregate hides

Because each German facility owes its own floor, the quantity that binds is the sum of the individual shortfalls. Gas held above the floor at one site does not discharge the obligation at another — it could physically be withdrawn and re-injected elsewhere at a cost, but it does not count where it sits, and the arithmetic of a national percentage silently assumes it does.

Take the 44 facilities whose floor is unambiguous — 209.8 TWh of working gas, 85% of German capacity, leaving aside two AGSI records that pool a named 45% facility with unnamed others. From 9 September to the 1 November date:

• 30 of the 44 sit below their own floor

• the shortfalls sum to 42.8 TWh

• facilities above their floor hold 4.4 TWh more than they need

• a national percentage reports the difference, 38.4 TWh

The aggregate understates the requirement by 4.4 TWh, about 12%. The bias is structural and always runs the same way, because a percentage lets a full site pay for an empty one and the law does not. It has doubled since late August, and for the reason you would expect: as the country fills, more sites rise above their own floor, and every tonne of that surplus is netted off a requirement it cannot discharge.

The Rehden record alone accounts for 12.8 TWh of the 42.8, thirty per cent of the shortfall across this sample. The sample is 84.8% of German working gas volume rather than all of it, so that share is not a national figure — but it is close to a third of the measured requirement sitting in one record, and that record has a lower floor than any unnamed facility in Germany precisely because it fills slowly. Its share has risen since August, because the rest of the country has been filling and it has not.

The distribution behind the 55.16% is stark, and has grown starker. The four cleanly identifiable facilities on the 45% floor hold 48.9 TWh of working gas between them and were 19.7% full on 9 September. Everything else was 63.8% full. Frankenthal was still at zero. Over the three weeks to 9 September the rest of German storage gained five points and this group gained under two, so the sites given the lower requirement are not merely the empty ones — they are falling further behind.

Four facilities that would have to beat their own declared rate

Run each German facility at the injection capacity it declared on 9 September, every day until 1 November, and four of them still fall short of their floor.

Breitbrunn was 10.5% full and tops out at 42.6%, thirty-seven points below its line. Wolfersberg was at 0.4% and reaches 49.2%. Inzenham-West was at 11.9% and reaches 59.7%. Those three owe 80%. Frankenthal, which owes 45%, was at zero and reaches 35.3%.

Three weeks ago the same calculation caught three sites and put their ceilings eleven to seventeen points higher. Nothing about the facilities changed. The calendar did: there are eighteen fewer days left to inject in.

The assumption carries the result, so it belongs in the open. Declared injection capacity is held at its 9 September value for fifty-three days, and real rates move with reservoir pressure over a season in a direction I have not established. Read as arithmetic on a declared rate, this holds. Read as a claim that these sites cannot physically comply, it does not, and I am not making that claim.

The facility view answers a question the aggregate cannot. Germany’s installed injection capacity is 4,294 GWh/d and the country is using 566, an eighth. Across only the facilities that are short, the required rate is 808 GWh/d against 2,705 installed, which is 30%. Nationally there is slack. At these four sites there is none at the rate they are declaring. Both are true, and only the facility data distinguishes them.

Three of the four sit among the Bavarian facilities that the regulation singles out in February, granting them a higher floor than everyone else, 40% against 30%. It is tempting to read that as the instrument recognising a constraint at one date and forgetting it at the other. The regulation gives no reason for the February carve-out — unlike the November one, which it justifies expressly — and a higher end-of-winter floor for a regional cluster has an obvious reading as a withdrawal-season standard. The coincidence is real and it will not carry an inference.

If not capacity, then price

Jack Sharples and Ricky Hill, writing for the Oxford Institute for Energy Studies on 27 July, report seasonal spreads “consistently negative” through 2026, against 1.50 to 2.00 EUR/MWh from May to August 2025. They do not name the contract months, so neither will I. The trade that fills European storage, buying summer gas to sell in winter, has been booking a loss on entry all summer.

The cross-year record is at least suggestive, and less so than it was a month ago. In 2021 EU storage entered the same thirty-day window at 60.6% full and took 3,457 GWh/d. In 2026 it entered at 59.6% and took 3,045, about an eighth slower from an almost identical starting point. On the equivalent window ending in August the gap was a quarter, so it has been closing.

One matched pair is not a control, and I should not dress it as one. Storage fees, capacity bookings, financing costs and fuel losses all sit between a headline spread and a filled tank; outage patterns and country composition differ year to year; and an EU aggregate fill percentage conceals the facility distribution behind it, which is this piece’s own complaint turned against its own evidence.

It is also a claim with a shelf life, and some of it has expired. Through August Europe was filling an unusually empty system no faster than it had filled nearly full ones in 2023 and 2024. Through early September it is filling faster than either, and faster than 2025. Whatever was binding in high summer has loosened. What remains is narrower: the emptiest system in the record is still being refilled more slowly than 2021 managed from the same point, and a persistently negative spread is the most obvious candidate for why.

A common claim is that only about 76% of German storage capacity is commercially booked, and that this caps what can be filled. It caps what the market fills. Section 35c(2) of the Energiewirtschaftsgesetz, in force until March 2027, lets the market area manager book the capacity it needs at a formula-set fee. The booking rate is a constraint on traders, not on the Member State.

How Germany filled its tanks last time

In 2022 Europe went into the filling season facing the loss of Russian pipeline gas at prices several times today’s, and filled anyway — Germany reached about 95% by 1 November. It is tempting to read that as proof that price never stops a determined buyer.

What it shows is narrower. Most of that gas was bought by the market; the part that closed the gap was not. Trading Hub Europe, the market area manager, was tasked by the federal government with buying close to 50 TWh of gas for around €9 billion, mostly on the spot market at record prices, on volume the market had not booked. Fifty terawatt hours against a national working gas volume of 247 is a fifth of the country’s storage, procured by the state. The cost was recovered through the Gasspeicherumlage, a levy on consumers that began at €0.59/MWh and reached €1.86/MWh by January 2024.

Both halves of that arrangement were changed in 2025, and in opposite directions. The purchasing power survives: section 35c(2) of the Energiewirtschaftsgesetz, in force until 31 March 2027, still empowers the market area manager, with ministry approval, to undertake “the acquisition of physical gas and its injection” where the market’s own filling falls short. The financing does not. Section 35e now runs only to the end of 2025, and since 1 August 2025 it excludes from the levy any cost beyond the manager’s own material, staff, advisory and financing costs — which is to say, the cost of the gas itself.

So Germany still has the power it used in 2022, and the particular route it used to pay for it has been closed. Whether another exists — an appropriation, a guarantee, borrowing — I have not established, and the question has not been put in public. That change and the facility floors arrived in the same year, both before the disruption that made either matter. The Gasspeicherfüllstandsverordnung is dated 5 May 2025; the Commission wrote to energy ministers in March 2026; the ministry’s remark about 60% to 70% came last week. The instruments were in place before the winter they now govern.

So the two episodes compare on something firmer than price. In 2022 Germany bought a fifth of its storage and socialised the bill. By 2026 it had lowered the floor its operators owe and closed the account the 2022 purchase was charged to. Both are defensible responses to a changed supply picture. Neither shows up in the headline target, and only one of them shows up in a fill percentage.

The reading in which nothing has gone wrong

There is a version of these facts in which this is all sensible, and it deserves more than a caveat.

On that reading, 90% was calibrated in 2022 against the abrupt loss of Russian pipeline gas, where storage was the only buffer between Europe and a supply cliff. The risk now is different. LNG is the marginal supply, it arrives on ships that can be redirected, and a persistently negative summer-winter spread is the market’s forecast that winter gas will be available and affordable. Filling to 90% then buys expensive insurance against a contingency the market has already priced away, and the Commission’s 80% invitation is a rational recalibration to a different supply regime. Its April language points the same way, conditioning 80% on “LNG supply availability”, which is a supply-contingent standard rather than a fixed one.

The German facility carve-outs have the strongest version of this behind them. The regulation says the six sites at 45% cycle slowly, the ceiling arithmetic above is consistent with that, and a rule that ignores injection physics is a worse rule than one that accommodates it.

Two things stop me settling on it. If this is a recalibration, it has never been stated as one: no impact assessment lowers the 90%, no delegated act has been published, and the headline figure is unchanged. And a forward curve prices what the market expects, which is not the same as what a buffer is for. Storage is held against the case the curve does not contain. A negative spread is evidence that the market does not expect to need it; the target exists because expecting not to need it has never been the test.

The parameter that gets reported

Count the numbers now attached to one requirement for one winter. Ninety in the EU regulation. Eighty if a Member State takes the ordinary derogation, and the level the Commission asked for. Seventy-five if it also qualifies for the structural one. Seventy-three for the Netherlands under the consumption cap. Eighty in Germany for most facilities, forty-five at six of them. Sixty to seventy as sufficient, per the German ministry on 20 August. Thirty in Germany on 1 February, and forty at four Bavarian sites.

The first is the one that travels.

The reason is not conspiratorial. Lowering a stated requirement is a public act with an identifiable constituency against it. Widening a window, softening a trajectory to indicative status, legislating a derogation exercised on a private justification, or setting facility-level floors in a ministerial regulation are governance acts of a different order, and they attract less scrutiny for a comparable effect on the buffer that actually exists.

None of it was concealed. The EU flexibilities were legislated in the ordinary way and presented as measures that would bring prices down; the German regulation was published in the Bundesgesetzblatt; every fact here came from a public document. A requirement can be adjusted in the open, repeatedly, and still leave the published comparisons unchanged, because those comparisons track the parameter that did not move.

The macroprudential parallel is close, and I want it stated narrowly. Article 136(5) of the Capital Requirements Directive sets the timing of increases in the countercyclical buffer rate: institutions must apply an increase no later than twelve months after it is announced, and a shorter deadline than that has to be justified by exceptional circumstances. Nothing imposes an equivalent lag on reductions; 136(6) asks only that an authority reducing the rate indicate a period during which no increase is expected, and says that indication does not bind it. The asymmetry is deliberate and defensible. It also means an announced increase can sit a year away from applying, which a comparison reporting the currently applicable rate does not show, and which a comparison carrying the announced rate and its effective date would.

For anyone building resilience metrics, the German case makes a sharper point than “headline rates are incomplete”. Aggregating facility-level floors into one percentage understates the requirement structurally, in the direction of comfort, and the understatement grows as the system fills — 6% in late August, 12% three weeks later, because every site that rises above its own floor adds surplus the law will not let it spend. The aggregate also says nothing about which sites are furthest from what they owe, which is the part that decides whether the shortfall is recoverable.

What happens next

Europe will spend the autumn quoting 90%. Whether Germany meets anything turns on a window widened in 2025, derogations that may or may not be exercised, a delegated act that has not been adopted, a facility list that expires in March 2027, a purchasing power whose financing was withdrawn in the same year it was last needed, and whether the seasonal spread turns before October.

Not one of those was decided this winter, and every one of them is decided before this winter is scored. On its present national trajectory Germany misses the EU number, and thirty of the forty-four facilities I can measure sit below the floor their own regulation sets. Whether they close it is a question about forty-four separate injection paths, which is exactly the question the published figure is not built to answer.

Paweł Fiedor — The Macro Prudential View

The views expressed are the author’s own and do not necessarily reflect those of any institution with which the author is or has been affiliated.

Sources: Regulation (EU) 2017/1938 as amended by Regulation (EU) 2025/1733, Official Journal — https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32025R1733; Regulation (EU) 2022/1032, for the Article 6a(2) five-year consumption basis — https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32022R1032; Gasspeicherfüllstandsverordnung of 5 May 2025, BGBl. 2025 I Nr. 130 — https://www.gesetze-im-internet.de/gasspf_llstv_2025/BJNR0820A0025.html; Energiewirtschaftsgesetz §§ 35b, 35c and 35e — https://www.gesetze-im-internet.de/enwg_2005/__35b.html, https://www.gesetze-im-internet.de/enwg_2005/__35c.html and https://www.gesetze-im-internet.de/enwg_2005/__35e.html; Directive 2013/36/EU, Article 136 — https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32013L0036; European Commission, gas storage — https://energy.ec.europa.eu/topics/energy-security/gas-storage_en, and the Gas Coordination Group notes of 26 March and April 2026; Jack Sharples and Ricky Hill, European Storage Refill in Summer 2026: Key Countries, Their Stocks, and Replenishment Mechanisms, Oxford Institute for Energy Studies, 27 July 2026 — https://www.oxfordenergy.org/publications/european-storage-refill-in-summer-2026-key-countries-their-stocks-and-replenishment-mechanisms/; SEFE Storage GmbH, Rehden facility — https://www.sefe-storage.de/en/storage-locations/rehden-storage-facility. Storage data: AGSI+ (Gas Infrastructure Europe) — https://agsi.gie.eu — EU aggregate to gas day 9 September 2026, the German national series to the same day, and the German facility-level series for gas day 9 September. Author's calculations throughout.

What Was Holding the Buffers Up

In June, writing about the European Commission’s money market fund review, I described the United Kingdom as proposing “to raise weekly liquid assets toward 50%” and as having “confirmed the opposite instinct” to Brussels. That was already wrong when it published. The FCA had set out a revised position on 8 June, a fortnight earlier: it intends to keep the existing minimum liquidity requirements in rules, and to place 40% for stable NAV funds and 20% for variable NAV funds in supervisory guidance. Those are the Commission’s two numbers. The error is worth correcting slowly, because the piece it sits in was about the difference between a requirement and an expectation, and the UK has moved across that line since.

What the FCA announced

The FCA plans a new rule requiring “that all MMFs hold sufficient liquidity for adequate resilience”: a binding duty with no figure in it. Beneath that rule will sit guidance carrying a “strong supervisory expectation” that stable NAV funds hold 40% weekly liquid assets and variable NAV funds 20%, “in order to meet the new resilience requirement”. The numerical minima stay where the retained UK regulation left them, at 30% weekly and 10% daily for low-volatility and public debt constant NAV funds. The 2023 proposal to raise the weekly and daily minima to 50% and 15%, for all money market funds rather than the stable NAV ones alone, has gone; on daily liquid assets the FCA will retain the 10% floor and does “not plan new guidance on DLA levels”.

None of this is yet in force. The FCA has said it will publish interim final guidance on weekly levels; the statement page still carried its 8 June date when this was checked on 18 August, with neither the guidance nor the policy statement issued. The Treasury, announcing the reform jointly with the FCA on 14 May, expects the new regime in place by the fourth quarter of 2026, subject to Parliamentary approval. What follows is about a decision taken, not an instrument issued.

The package also includes delinking: removing the regulatory connection between a stable NAV fund’s liquidity level and the board’s obligation to consider fees or gates. That connection is the cliff edge in the run literature, and removing it is the most substantial thing in the package.

Where forty came from

Three months before the FCA moved, a Bank of England staff working paper asked what the buffer should be. Rishabh Kumar’s A simulation framework for sterling money market funds estimates redemption capacity and failure probability across a range of redemption profiles and market-liquidity conditions. It finds that “the reduction in failure risk is concentrated in the transition from 30% to 40% WLA; beyond 40%, additional increases in WLA typically generate comparatively smaller marginal improvements in survival across horizons”.

Failure there is a term of art. A fund fails, in this framework, when its daily liquid assets fall below 10% of total assets: a breach of a regulatory floor rather than a gate, a suspension or a loss to investors. It is a measurement device for whether a fund got into trouble, and the headline reads more dramatically than the thing it counts.

The paper describes its result in the language of the rulebook. What it models is “increasing the requirement to around 40%”, and the gain it finds comes from “increasing the current regulatory minimum to a higher buffer”. That establishes less than it appears to. The simulations vary the liquid assets a fund holds; they do not model legal instruments, so they cannot show that a binding minimum is the only way to get funds to 40%, or that guidance would fail to. Whether an expectation holds the lower tail of the distribution as reliably as a numerical floor is an open question, and neither authority has published anything on it. The chart below sets the three decisions against the band the simulations identify.

There is a baseline against which it could be answered, and the Commission has published it. Its own data put the 25th-percentile cutoff for EU low-volatility funds close to the 40% line for two years, and then stop: the series ends before either instrument exists. That is an EU population and the benchmark in question is the EU one, so it is a baseline for the Commission’s figure alone, and the holdings predate the benchmark in any case. It is the starting line, and someone should be measuring from it, as the next chart shows.

The 30-to-40 result comes from a specific counterfactual: runs under historically extreme shocks, with the 30% threshold effect switched off. Removing that effect, the paper finds separately, itself “yields sizeable resilience gains by reducing cliff-edge behaviour”. It is the same effect the FCA is now removing, so the two halves of the package belong together.

The cliff is narrower than it is usually made to sound. Article 34 is a joint condition rather than a single line, it reaches only stable NAV funds, and the chart below sets out both of its limbs. The 30% line does not bite in calm markets. It bites in precisely the conditions in which a manager can least afford it, and that is a reason to hold enough that a run cannot carry you across it.

The Commission’s series shows EU low-volatility funds holding weekly liquid assets in the forties on average against a floor of thirty from 2020 to 2025, with the 25th-percentile cutoff close to the 40% line from 2024. Whether the threshold is why, I cannot show, and neither can anyone else from public data. For sterling funds the distribution is reported to supervisors and never published. Article 37 requires every UK money market fund to report at least quarterly, and the reporting template in the FCA’s own Handbook asks, at field A.4.6, for the percentage of assets qualifying for the weekly liquidity buffer. Equivalent returns reach the Central Bank of Ireland and the CSSF for the funds that hold the other nine tenths. The supervisory picture exists in three places and is public in none of them. Some of that is unavoidable: the FCA authorises three stable NAV funds, and a distribution over three funds identifies them. But an aggregate is publishable — the FCA published one in 2023 — and none has appeared since, in any jurisdiction, while both authorities were naming forty. Ratings methodologies, investor mandates, internal limits and plain post-2020 caution all push the same way. The test that would separate them — whether variable NAV funds, which face no Article 34 cliff, also sit well above their own 15% minimum — needs a fund-type breakdown the Commission does not publish. Treat the cliff as one channel among several.

What does not depend on the mechanism is what was previously proposed. CP23/28 put delinking and a higher numerical floor in the same package: remove the threshold link, and raise the weekly minimum to 50%. I am not going to claim the FCA said the one was there to replace the other, because it did not say so. The observable fact is that the earlier package contained both and the announced one contains delinking with the number in guidance. It may hold. Nothing published shows that it will.

The older modelling, which points the other way

Annex 4 of the FCA’s 2023 consultation reported that holding 40% weekly liquid assets “would be insufficient to provide resilience against the largest historical outflows in sterling MMFs without resorting to what could be fire sales of assets”, with 50%, 60% and 70% covering progressively more. Read flatly, that is the FCA settling in 2026 on a level its own consultation rejected in 2023.

It does not read flatly. Annex 4 asks whether a buffer covers a given outflow without selling assets; the 2026 paper asks where the marginal reduction in breach probability is largest once the cliff-edge effect is gone. The two metrics answer different questions, so the results do not conflict.

It is also a reconciliation nobody has published, and the FCA says as much. Its June statement cites the exploratory scenario as the reason for coming off 50%, and then undertakes that “our policy statement will provide more detail on the updated proposals and the modelling on which they are based”. Until that arrives, no public document takes the scenario’s results and derives forty rather than fifty, or forty-five.

The case for the FCA

That case is stronger than the shape of this piece implies.

The stated basis for coming off 50% is the exploratory scenario, which “indicates that, in some scenarios, outflows from MMFs may be somewhat lower than in previous stress episodes”. Revising a calibration when the evidence moves is what a supervisor is for. And guidance beneath a binding rule is a real constraint. Firms manage to supervisory expectations, and supervisors engage on them. The resilience requirement gives the figure a hook a free-standing expectation would lack.

Then there is the argument the FCA has not made in public, which may be the strongest available to it. The chart below has the sterling sector as the FCA last published it: the regulator supervises about a tenth of it by assets, and the stable NAV expectation reaches a fraction of that tenth. Nine tenths sit in EU-domiciled funds, mainly Ireland and Luxembourg, against a 30% floor, a split the 2026 paper independently reports for its own sample.

A binding UK 40% would therefore have fallen on a handful of funds while their competitors carried on unaffected. The cost lands on the funds the FCA authorises, and redomiciliation into the regime they are competing against is one response available to them. Whether it is the cheapest response I cannot say — the trade against simply absorbing the yield cost of a larger buffer depends on portfolio structure and on legal and operational costs that nobody has published — but it is on the table in a way it would not be for a regulator covering the whole market. This is a hypothesis. The FCA has published no impact assessment on it and has not offered it as a reason.

There is a second constraint, and it is the more concrete of the two. The minima sit in retained UK legislation that FSMA 2023 lists for revocation but which has not been commenced, and the Treasury has said most requirements for UK money market funds will end up in FCA rules and guidance. The FCA’s own words are that it intends to “retain in rules the current minimum WLA requirements as set out in UK MMFR”, which reads as the regulator restating the figures in its own rulebook and therefore able to choose them. Whether it could have chosen differently without further Treasury action is not spelled out anywhere public. Some Treasury action is coming in any event, since the timetable is subject to Parliamentary approval; its scope is not stated.

Two instruments

The UK will have a qualitative rule requiring adequate liquidity, with 40% carried in guidance underneath it and the numerical minimum left at 30%. The EU has 40% in a Commission review report: a review-and-report exercise that may still be followed by a legislative proposal and until then changes nothing. The binding weekly minimum for an Irish or Luxembourg LVNAV remains 30%, where it has sat since the regulation began to apply in 2018.

There is an obvious question the UK construction leaves open, and it cannot be answered yet because the guidance has not been written. What happens to a fund sitting at 34% — above the floor, below the expectation, and in breach of nothing numerical? Some supervisory conversation presumably follows, and its force is the whole of the difference between this arrangement and a rule. Until the interim guidance appears, the supervisory trigger and its consequences are unstated.

So the larger part of the sterling sector faces no 40% of any kind, only a figure in a report and whatever its national supervisor makes of it — and national supervisors, along with ESMA and the ESRB, retain instruments of their own that do not require amending the regulation. Those funds are not unregulated from a UK standpoint — they need recognition or permission to market here, and UK conduct requirements travel with that — but their liquidity rules are set in Dublin, Luxembourg and Brussels, and their access to UK investors is itself unsettled. The Government has said it intends to extend the Temporary Marketing Permissions Regime “with a view to establishing a longer-term solution on market access”, which is one route by which a UK liquidity condition could reach them, if one were ever wanted.

Locking in

The Financial Policy Committee took this up in July. Its record says the Committee “judged that it remained important to lock in the higher levels of resilience that had been held by MMFs since the ‘dash-for-cash’ in 2020, as liquidity mismatch in these funds remained a key potential source of vulnerability”, and that it “welcomed recent statements by HM Treasury and the FCA on their plans to enhance MMF resilience”.

The object of that sentence is levels held, not levels required, and the distinction has some history. In December 2023, on the day the proposals appeared, the FPC recorded that it “welcomes proposals by UK authorities to increase the resilience of UK-based money market funds, which have been published today” — those being the proposals for a binding 50% weekly and 15% daily. In July 2026 it welcomed the statements that withdrew both. The wording of the Committee’s support is consistent across a change in the substance it was supporting, which is not an accusation of anything; it is what a committee’s welcome is worth as evidence.

Preservation is a coherent objective, and an expectation can serve it. What preservation assumes is that the buffers stay where they are. One of the things that may have been holding them up is being removed, and the number that would have replaced it is in guidance rather than in the rules. The most recent published analysis of the sterling sector models funds at around forty breaching the daily floor materially less often than funds at thirty. A staff working paper is not committee policy and its author’s views are his own. Nobody has argued against the finding either.

Two decisions sit behind this. The FCA is restating the numerical minima in its own rulebook and restating them unchanged; whether it could have restated them at a different figure without Treasury action is unresolved on anything public. The Commission has so far stopped short of proposing an amendment to the money market fund regulation, which is the most it could have done alone, since raising the EU minimum would then have needed the Parliament and the Council, and its report leaves a later proposal open. What the two outcomes share is only this: the figure ends up somewhere other than the rules. The Commission gave a reason for declining to raise the minimum, that doing so would not be proportionate given the heterogeneity of the sector and its fund types. The FCA has not said why forty sits in guidance rather than in the rules it is writing.

Relocating the seam, on 31 August, asked whether the powers to act on fund-sector systemic risk move together with the supervisor. This is what happens to a power that has not moved anywhere.

Paweł Fiedor - The Macro Prudential View

Positions described are as at 18 August 2026.

The author works at the Central Bank of Ireland and in the ESRB Secretariat. This piece is written in a personal capacity; views are the author’s own and should not be attributed to the Central Bank of Ireland, the ESRB, or the Eurosystem. It draws exclusively on publicly available sources and takes no position on whether the proposals should be adopted.

Sources: Financial Conduct Authority, Update on reforms to the UK Money Market Fund Regulation, 8 June 2026 — https://www.fca.org.uk/news/statements/reforms-uk-money-market-fund-regulation; FCA, CP23/28, Updating the regime for Money Market Funds, December 2023, including the Bank of England modelling at Annex 4 and the sector figures at 2.29, 2.30 and 5.4 — https://www.fca.org.uk/publication/consultation/cp23-28.pdf; Rishabh Kumar, A simulation framework for sterling money market funds: estimating redemption capacity and evaluating liquidity requirements, Bank of England Staff Working Paper No. 1,177, March 2026 — https://www.bankofengland.co.uk/-/media/boe/files/working-paper/2026/a-simulation-framework-for-sterling-money-market-funds.pdf (the views expressed in a staff working paper are the author's and not necessarily those of the Bank of England or its committees); Bank of England, Record of the Financial Policy Committee, 7 July 2026, paragraph 49 — https://www.bankofengland.co.uk/-/media/boe/files/financial-policy-committee-record/2026/fpc-summary-and-record-july-2026.pdf; Bank of England, Financial Stability Report, December 2023 — https://www.bankofengland.co.uk/financial-stability-report/2023/december-2023; European Commission, Report on the functioning of Regulation (EU) 2017/1131 on money market funds, COM(2026) 350 final, 11 May 2026, Chart 1 — https://ec.europa.eu/finance/docs/law/260511-money-market-funds-report_en.pdf; HM Treasury, Reforms to Money Market Fund Regulations, written ministerial statement and accompanying publication, 14 May 2026 — https://www.gov.uk/government/publications/reforms-to-money-market-fund-regulations; Regulation (EU) 2017/1131, Articles 24, 25, 34 and 37, and the retained UK version — https://www.legislation.gov.uk/eur/2017/1131; and the Article 37 reporting template, Commission Implementing Regulation (EU) 2018/708, Annex, as it stands in the FCA Handbook — https://handbook.fca.org.uk/technical-standards/provision/s135c1153sn0p1713. Chart data as cited in each figure.

On 29 May 2026 the finance ministers of Germany, France, Italy, Spain, the Netherlands and Poland agreed to back more centralised supervision of EU capital markets. The industry answer was already on record: for EFAMA, turning ESMA into an extra supervisor for asset managers is “unwarranted, a major distraction, and runs contrary to the EU competitiveness and simplification agendas” — and the version in the Parliament’s draft report is worse than the Commission’s. Since then, the public argument has mostly run on those rails: cost and competitiveness on one, institutional design on the other.

Read the legislative texts, though, and the divide that matters most for the system-wide risk monitor sits elsewhere: in whether the powers to act on fund-sector systemic risk move together with the supervisor. The texts do not carry equal weight. The Commission’s December proposal has initiative status; the rapporteur’s June draft report for the Parliament’s economic affairs committee is one pole of an amendment process, not yet voted in committee and not a Parliament position; the Council had adopted no position at the time of writing; and the ECB supports integration while calibrating, for asset managers specifically, short of what the rapporteur proposes. This piece maps what each text does to the macroprudential machinery and where the open questions sit. It takes no position on what should be adopted.

What the texts do

The Commission’s Market Integration and Supervision Package (4 December 2025) does not centralise the supervision of asset managers. It creates a recurring review (new Articles 47a and 47b of the AIFMD and 110b and 110c of the UCITS Directive): ESMA, in cooperation with home authorities, would review at least annually each EU group of management companies and AIFMs with aggregate EU-wide net asset values above €300 billion and a multi-Member-State footprint, with escalation powers running as far as suspending a group’s cross-border activity. AIFMD Article 25 — the provision under which national authorities impose leverage limits and other restrictions on fund managers — is left untouched.

The rapporteur’s draft report for ECON (Eero Heinäluoma, 12 June 2026) restructures this into two tiers. The annual review would apply to groups between €50 and €150 billion. Above €150 billion, with cross-border operations, ESMA would designate a group significant and, in the words of the inserted Article 44a of the AIFMD, “assume the supervisory tasks and duties assigned under this Directive to the competent authority of the home Member State, including for their authorisation and supervision.” The review article confirms the tiering: it applies to groups “that do not fall under the direct supervision of ESMA pursuant to Article 44a.”

The macroprudential part of the draft sits next door. New Articles 25a and 25b would write a toolkit into the AIFMD itself: powers to require notice periods, holding periods and redemption-frequency limits for open-ended AIFs, applicable to whole categories of funds where systemic risk is identified; and leverage limits at fund or category level, with real-estate funds calibrated on total debt to total assets — the design the Central Bank of Ireland used in 2022, generalised. ESMA would draft the technical standards for both, jointly with the ESRB on leverage. The draft also puts depositaries under direct ESMA supervision, mirrored across both directives (new Articles 101a of the UCITS Directive and 54a of the AIFMD) — with a duty to coordinate with the banking supervisors, since the large fund depositaries are credit institutions supervised under the SSM or nationally. It is a measure of how far beyond the Commission text the draft reaches. The rapporteur’s explanatory statement is explicit about the logic: a more integrated supervisory framework for funds should be accompanied, as the ECB has argued, by amendments to the macroprudential framework.

Two design questions sit in the texts themselves. The first is the perimeter. Both thresholds are group aggregates: they add up the net assets of the management companies and AIFMs inside a group, which measures how much a manager runs, not the footprint its funds have in markets. A group can sit under the line while managing funds whose gross exposure is a multiple of their net assets — the profile behind most leverage-driven fund episodes. The drafting compounds it: Article 44a sets the threshold in net asset values while its definition paragraph speaks of assets under management, and for leveraged managers the two differ by multiples. That mismatch is in the legislative text, and technical standards cannot repair it.

The second question is timing. The rapporteur frames Articles 25a and 25b as ex-ante instruments, but the text does not confine them to calm conditions, and a category-wide notice period announced into stress sharpens the first-mover incentive that liquidity tools exist to dampen. Fund-level gates carry the same anticipation problem in miniature; scaling them to whole categories scales it too. That one is a calibration question, and it does land on the technical standards.

The layer that does not move

Every text on the table leaves the fund layer national. Authorisation of UCITS and AIFs, product supervision, day-to-day oversight and enforcement in national courts stay with the domicile authority under the Commission review and under the rapporteur’s direct-supervision tier alike. That layer is where the concentration sits: Luxembourg and Ireland domiciled €11.7 trillion of fund net assets at end-2025, 55% of the EU total. The management-company population is concentrated in the same places, though not universally — several large groups anchor their EU management companies elsewhere — which is why the chart is evidence about the fund layer, and only that. The cross-border character of the sector is structural either way: funds sold outside their domicile have grown from 49% of European fund assets in 2015 to 55% in 2025 — on EFAMA’s European aggregate, a wider perimeter than the EU figure above, which is an own calculation excluding the UK, Switzerland and the other non-EU domiciles. Any move at group level, under either proposal, would sit on top of a product and enforcement layer that is organised nationally and concentrated where the chart shows.

The same four questions

The honest way to compare the three architectures — the one in force, the Commission’s, the rapporteur’s — is to hold four questions constant: what information improves, which decision rights move, who leads when a large manager is under stress, and where the power to activate macroprudential tools sits. “Leads” means the authority a supervisor, a counterparty or a finance ministry calls first, which is a different thing from the authority that can impose a limit. The matrix carries the full cells; the prose here stays on the points where the columns diverge.

Information is the narrow difference. AIFMD reporting already flows to ESMA, UCITS reporting follows from April 2027 under Directive 2024/927, and the real gaps are group-level consolidation and the comparability of filings made under twenty-seven national interpretations. The Commission’s review forces consolidation for the groups above its threshold; the rapporteur’s significant-group tier converts that into full supervisory access.

The Commission has priced the difference, and the numbers are worth sitting with. Its impact assessment budgets ESMA 10 to 20 additional staff, €2–4 million a year, to coordinate supervision of around 10 to 15 large cross-border asset management groups. For comparison, in the same document, directly supervising the trading venues belonging to 8 to 11 groups is costed at 90 to 105 staff; five to eight CCPs at 30 to 50; ten to thirteen CSDs at 30 to 40. The gap between coordinating a review of a dozen groups and supervising a dozen entities is roughly a factor of five in people, on the Commission’s own estimates — which is one way of seeing how much lighter the review is than the word ‘supervision’ suggests, and what the rapporteur’s tier would be asking ESMA to build.

Decision rights are the wide difference, though not a clean one. The Commission’s review does move a decision right: escalation running as far as suspending a group’s cross-border activity is a power to act, not a power to look. What it leaves untouched is macroprudential activation — authorisation, supervision and Article 25 all stay national.

The rapporteur’s draft goes at the anchor itself, and here the drafting matters more than the intent. Article 44a would have ESMA assume the home authority’s tasks and duties, and Article 25(3) assigns the imposition power to exactly that authority. But the AIFMD defines competent authorities as the national authorities of the Member States, and the draft does not amend that definition — so whether ESMA could exercise powers the directive addresses to a category it does not belong to is an open legal question rather than a plain reading. The matrix marks two further cells open. The first is horizontal: 25a and 25b allow measures across whole categories of funds, a power addressed to “competent authorities” at large, which does not map onto the home-authority assignment the transfer clause uses. The second runs along the UCITS side: the significant-group definition counts UCITS management companies and AIFMs together, and the draft mirrors its depositary powers across both directives, but the transfer clause for significant groups sits in the AIFMD; without a UCITS counterpart, a designated group would answer to ESMA for its alternative-fund business and to its home authority for its UCITS business. A plain reading is not an operational answer, and the distance between the two is where the trilogues will spend their time.

Even on the reading where everything transfers, ESMA would be imposing entity-level measures on funds that remain authorised and governed under national law. How such a measure would be decided, challenged and enforced — through ESMA’s own decision and appeal machinery, through national procedure, or through both — the drafting does not trace. The measures in force show what that means: the property-fund limit bites an Irish market under Irish fund law; the yield buffer was built around the plumbing of funds in two jurisdictions. A limit calibrated in an ESMA technical standard would still have to bind, and be enforced against, specific funds in specific legal systems. This is the double layer EFAMA warns about, relocated to the macroprudential level; it cuts both ways, because that layer is where enforcement power and supervisory expertise both live. The Parliament’s own scrutiny unit reaches the same point from the other direction, and puts it more precisely: supervision in asset management is organised around the nexus between a management company and the funds it runs, since judgements about a fund rest on what the authority knows about the manager’s governance and risk arrangements. Splitting the two, on that reading, risks fragmenting layers that are operationally intertwined rather than resolving anything.

One part of the in-force baseline is easy to miss, and it sets the bar the new tools would have to clear. Directive 2024/927, applicable since 16 April 2026, already requires open-ended AIFs and UCITS to select and maintain liquidity management tools from a defined list. The power it gives authorities, though, is narrower than the toolkit it mandates: under the amended Article 46(2)(j) of the AIFMD, a home authority may require a manager to activate or deactivate suspension of subscriptions and redemptions — Annex V, point 1, and only that one — in exceptional circumstances, after consulting the manager, where risks to investor protection or financial stability warrant it. Notice periods, holding periods and redemption-frequency limits are tools the manager chooses and operates; no authority can order them today.

That is what Articles 25a and 25b would change, and in two directions at once: from the on-off switch to the calibrated instruments, and from a case-by-case direction to standing terms set across a category in advance. The Commission’s text leaves both the Article 25 powers and the 2024/927 powers as they are.

The UCITS side carries a home-host detail that bears on all of this. A host authority may ask the home authority to exercise its supervisory powers, informing ESMA and, where the stability of the financial system is in question, the ESRB — but the request mechanism excludes the very point covering activation of suspension. The tool that bites hardest in a run is the one a host authority cannot ask for.

The 2022 episode shows what the open cells would have required the texts to answer in real time: UK pension schemes, running liability-driven strategies through funds domiciled in Ireland and Luxembourg, selling into a falling gilt market. Under the architecture in force, the response ran through two national authorities coordinating bilaterally, with ESMA advising. Under the Commission’s review, ESMA’s information would plausibly have been better consolidated; the activation mechanics would have been identical. Under the rapporteur’s tier, if the managers involved had crossed the designation threshold and the funds were AIFs rather than UCITS, ESMA would have been the group supervisor — adding a Paris counterpart for the Bank of England alongside Dublin and Luxembourg rather than replacing them, since the product layer stays national under every text here. That dialogue ran on bilateral supervisory relationships, including memoranda of understanding with the UK authorities, and how they would map onto an additional EU-level counterpart the drafting does not settle. None of this says how the episode would have ended; it says which questions would have needed answers in the middle of a run.

The strongest institutional case for moving in this direction sits with the ECB, and even it stops short of the draft report. In its formal opinion on the package, the ECB welcomes the annual review of large groups as proportionate, warns that centralising functions within a group must not create supervisory blind spots or single points of failure, and states the condition the rapporteur later picked up: more integrated supervision of funds and managers should be “accompanied by a review of the Union macroprudential framework through targeted amendments to the UCITS and AIFM Directives”. Its Occasional Paper “One market, one supervision?” judges direct supervision of asset managers unlikely in the short term, proposing a stronger convergence role for the ten to fifteen largest cross-border groups. The Eurosystem argued in the same direction in the 2024 NBFI consultation, and the ESRB asked for European oversight of systemic cross-border actors — oversight and monitoring, which is not the same instrument as the direct supervisory mandate Article 44a would create. All of it was said before the draft report appeared, and neither institution has published a view on its specific articles. On the positions of record, the rapporteur’s text goes further than anything the official sector has proposed.

EFAMA’s October 2025 report carries the case against: ESMA would need supervisory capacity that some national authorities have accumulated over decades; splitting management-company supervision from product supervision creates a double layer; and collective decision-making is slower than a single authority in the moments that matter. The first and third are claims about institutional performance that the record cannot settle. The second is structural, and it survives every variant on the table.

A proportionality argument runs alongside it, and it does not come from the industry. The Parliament’s scrutiny unit notes that the sector has not historically produced the distress patterns seen elsewhere in finance, that UCITS in particular operate under binding constraints on leverage, liquidity and diversification, and that an added supervisory layer might therefore be questioned against the risks actually identified — while allowing that the Commission’s approach can be read as forward-looking, addressing coordination problems that past episodes have not yet surfaced. Whether that history is a guide or a lagging indicator is close to the whole disagreement.

The record, then the reach

The one body of codified fund-level practice is Article 25 itself, used in two episodes — three measures, counting authorities — on clocks that have nothing to do with each other. The structural case is the Irish property funds: a consultation opened in November 2021, the 60% debt-to-assets limit followed in November 2022, and existing funds were given five years to comply, with full phase-in due in 2027. That measure addressed domestic commercial real estate and had been in preparation for a year before the gilt market moved.

The crisis case is sterling LDI. Within two months of September 2022, the Central Bank of Ireland and the CSSF had set a 300–400 basis point yield-buffer expectation by industry letter. Converting that expectation into a binding rule took seventeen months more: aligned national consultations in November 2023, notification to ESMA and the ESRB in March 2024, supportive advice, measures applying from 29 April 2024. Part of that is consultation and notification procedure an EU-level activator would face too. Part — two aligned national consultations to bind funds in two legal systems, having acted through letters while the codified route caught up — is specific to the current design, and it is the part centralisers cite. Between them, the two cases show the current architecture running a measure in both modes, and what each one costs.

What the record cannot show is reach, and reach is where the geography matters.

Take Irish-domiciled funds — the second-largest EU domicile after Luxembourg, and the one that publishes this split — and read the chart in order. Holdings first: 2.5% of the securities Irish funds hold are Irish-issued, or 9.6% on the broadest cut, counting deposits, holdings of other funds and all other assets. Holders second: 10.2% of the fund shares in issue are held by Irish residents; the ECB’s own securities-holdings work puts the foreign share of the investor base in Irish and Luxembourg funds at roughly 90%. Booking chain third: 41% of the total is recorded against UK-resident holders. That figure identifies where holdings are registered, not who ultimately owns them — much of it is nominee and platform intermediation — which is why it matters operationally: the registered chain is the one a redemption instruction travels down. Portfolio management, meanwhile, is routinely delegated to London and New York. That boundary is operational, not only geographic: an Article 25 measure addresses the EU manager, while the execution it forces — the unwinding, the margining, the sales — runs through delegates, desks and market infrastructure regulated elsewhere, which the measure does not reach directly. However the Union allocates supervisory competence internally, the registered holder chain and much of the day-to-day risk-taking sit outside it. Every column of the matrix operates inside that boundary.

Scale, and who answers

Fund net assets stand at 69 times GDP in Luxembourg and 8.6 times in Ireland. France, the next-largest fund domicile, sits at 0.9; no other domicile in the chart exceeds one. The chart measures scale, not leverage or exposure. Ireland gets a second denominator because its GDP is a measured distortion: against 2025 GNI*, the CSO’s de-globalised benchmark, the ratio is 16.5 times — and no other country in the chart has an official alternative denominator, which is why no other country gets one.

The chart belongs in a supervision debate because of accountability. ESMA’s founding regulation places financial stability in its headline objective — “contributing to the short, medium and long-term stability and effectiveness of the financial system” — and instructs it to pay particular attention to systemic risk. The gap is not in the statute; it is in the architecture around it. Today, ESMA’s macroprudential instrument for funds is advice under Article 25(6) and (7), backed by comply-or-explain under 25(8); the ESRB warns and recommends; the binding tools are national. The rapporteur’s draft would change that materially, through the 25a and 25b technical standards and the 44a transfer. At that point, the question of who answers for error becomes concrete — and it is an accountability question, not a bailout question. Losses from a miscalibrated measure would fall on fund investors, who are mostly non-resident, and the market effects would land where the funds invest, which the geography above shows is mostly not the domicile either. What the domicile carries is the industry: the employment, the fees, the tax base built on them — and, for the funds that do hold domestic assets, the local market too, which is what the 2022 property-fund limit was about. The calibration would be made at Union level — though ESMA’s Board of Supervisors is composed of the heads of the national authorities, so the change is one of decision procedure rather than of who sits in the room: the domicile keeps its seat and can be outvoted. What it would not keep is the ability to set the number itself. The same chart is the centralisers’ exhibit on the distance between where decisions would sit and where consequences land, and the decentralisers’ exhibit on why supervisory expertise accumulated where it did.

What the precedent does and does not license

The debate keeps reaching for precedent, so the record deserves precision. ESMA directly supervises credit rating agencies (since 2011), trade repositories (2013), securitisation repositories and Tier 2 third-country CCPs (2021), critical and third-country benchmark administrators and data reporting service providers (2022), consolidated tape providers as they are authorised, and EU Green Bond external reviewers from June 2026. EU CCPs are not on the list: they remain nationally supervised, with ESMA’s CCP Supervisory Committee coordinating. Fifteen years of direct supervision, in other words, covers market infrastructure and data services, plus systemically important CCPs from outside the Union.

A large asset-management group would be the first supervised population whose dominant failure mode is redemption-driven market transmission and leverage dynamics at scale, rather than the operational integrity of a utility. The record licenses one inference: ESMA can build supervisory capacity in bounded, rule-dense domains, and has. It licenses no inference, in either direction, about supervising market-facing redemption risk in the fund sector — and the failure-mode contrast above is a hypothesis about that population, not an established finding. Anyone claiming the precedent settles the question — for or against — is claiming more than fifteen years of evidence contains.

What to watch

Three things from this mapping are worth carrying into the autumn. The variants on the table differ most substantively on macroprudential architecture, and the difference is visible in the texts themselves. Centralising group-level scrutiny and relocating the power to act on systemic risk are separable choices: the Commission’s text takes the first without the second, while the rapporteur’s draft creates the tools and moves the supervisor, and leaves the join between them unresolved. And wherever the powers land, part of the sector’s geography — its delegation chains, its booking chains — sits beyond any intra-EU allocation.

The perimeter is narrower than the debate’s temperature suggests: the Commission’s impact assessment puts the population at around 10 to 15 large cross-border groups, and the ECB proposed a convergence role for the ten to fifteen largest. What the rapporteur’s €50bn and €150bn thresholds would capture, and what share of EU fund assets that is, nobody has published. The questions for the remaining parliamentary and Council process, and for any trilogue that follows, come straight off the mapping: whether Articles 25a and 25b survive Council scrutiny, are stripped out, or migrate into the Commission’s separate macroprudential work on non-bank finance; who the “competent authority” of a significant group is for activation purposes; how the technical standards treat leverage that a NAV threshold does not see; and how an EU-level measure would be executed through national legal systems and third-country delegation chains. The timetable is set: the “One Europe, One Market” roadmap of April 2026 targets agreement on the package by the end of this year. The positions on record give the questions their shape. The direct-supervision tier has one institutional sponsor so far: the rapporteur who drafted it, whose report still has to survive committee and plenary before it becomes anything Parliament negotiates on. The macroprudential strengthening has more company — the ECB, the Eurosystem and the ESRB have all argued for it, which is support in principle for the direction, not for these articles as drafted. The Commission consulted on macroprudential policy for non-bank finance in 2024 and proposed no legislation after its March 2025 summary report. The rapporteur’s draft is, in practice, the first legislative text to move on it.

Paweł Fiedor - The Macro Prudential View

Positions described are as at 17 August 2026.

The author works at the Central Bank of Ireland and in the ESRB Secretariat. This piece is written in a personal capacity; views are the author’s own and should not be attributed to the Central Bank of Ireland, the ESRB, or the Eurosystem. It draws exclusively on publicly available sources and takes no position on whether the proposals should be adopted.

Sources: European Commission, Market Integration and Supervision Package — the Communication “Further development of capital market integration and supervision within the Union” (COM(2025) 940 final) and the legislative proposals (COM(2025) 941, 942 and 943 final), 4 December 2025 — https://finance.ec.europa.eu; European Parliament, ECON draft report by rapporteur Eero Heinäluoma, ECON-PR-789867, 12 June 2026 — https://www.europarl.europa.eu/doceo/document/ECON-PR-789867_EN.pdf; Directive 2011/61/EU (AIFMD), Articles 16, 25 and 46(2)(j) and Annex V, and Directive 2009/65/EC (UCITS), Article 98(3) and (4), as amended by Directive (EU) 2024/927 of 13 March 2024 (OJ 26 March 2024; applicable from 16 April 2026, reporting provisions from 16 April 2027) — https://eur-lex.europa.eu; Regulation (EU) No 1095/2010 (ESMA Regulation), Article 1(5) — https://eur-lex.europa.eu; joint statement and letter of the finance ministers of Germany, France, Italy, Spain, the Netherlands and Poland on the MISP and centralised capital markets supervision, 29 May 2026 — https://www.gov.pl/attachment/60b5fb71-a937-4d3a-8883-c8ec86d7700b (statement), https://www.gov.pl/attachment/56f1ed87-e749-4331-9347-271d491be331 (letter); Opinion of the European Central Bank of 9 April 2026 (CON/2026/13) — https://eur-lex.europa.eu/eli/C/2026/2837/oj; Carmassi, Dumora Lemaire, Evrard, Gati, Milea, Parisi, Rouveyrol and Spolaore, “One market, one supervision”, ECB Occasional Paper No 383 (2026), and ECB Blog posts of 13 February and 30 March 2026 — https://www.ecb.europa.eu; European Commission targeted consultation on macroprudential policies for non-bank financial intermediation (May 2024) and summary report (March 2025), with the Eurosystem response (November 2024) and the ESRB response — https://finance.ec.europa.eu, https://www.esrb.europa.eu; European Parliament, Economic Governance and EMU Scrutiny Unit, "MISP: A review of selected technical issues", PE 784.036, April 2026 —https://www.europarl.europa.eu; EFAMA, “Asset management supervision: why passporting remains the best supervisory model for Europe” (October 2025) and response to the Parliament draft report (2026) — https://www.efama.org; Central Bank of Ireland, macroprudential measures for Irish property funds (November 2022) and for GBP-denominated liability-driven investment funds (April 2024), with the CSSF’s parallel measures and ESMA’s advice under Article 25 AIFMD (April 2024) — https://www.centralbank.ie, https://www.cssf.lu, https://www.esma.europa.eu; ESMA, Supervision — https://www.esma.europa.eu; EFAMA Quarterly Statistical Release No. 104 (Q4 2025); Eurostat, EDP first notification, 22 April 2026; CSO, Annual National Accounts 2025 (modified GNI, 2 July 2026); Central Bank of Ireland, investment fund statistics, publication tables (2026Q1) and holders of investment fund equity by geography (Q3 2025). Chart data as cited in each figure.

Two readings of five percent

On 24 July the ECB published the third-quarter round of its Survey of Professional Forecasters. Among much else, the panel attaches a probability of 5.4% to negative euro-area GDP growth over the year to the first quarter of 2027.

In July 2008 the same survey, asking the same question about the year to 2009Q1, produced 5.0%. Growth over that year came in at −5.7%.

Each forecast in this piece is dated to the target quarter it concerns, but was formed roughly two quarters earlier at the one-year horizon. The distinction matters, because the question here is what the panel could have known at the time, not whether its distributions resemble revised data after the fact — asked of a record that contains three episodes of negative annual growth — the events the survey’s question actually defines, which are not the same object as a dated recession chronology. Outcomes are measured on today’s GDP vintage except where real-time releases are cited. A disclosure before the record: I work at the Central Bank of Ireland and in the ESRB Secretariat; this piece is written in a personal capacity, on public data only, and the full statement sits at the end.

The whole distribution

The mechanics briefly, because every caveat in this piece lives in them. Respondents attach probabilities to growth bins — half a percentage point wide in normal times, two points in the COVID-era grid — alongside their point forecasts. I rescale each usable density to sum to one and average across respondents with equal weights, the ECB’s own convention; between 24 and 58 densities survive that filter per round at the one-year horizon. Stacked across 111 rounds, the result is Figure 1: in the major episodes, the distribution shifts materially only after weakness is visible in the information available to respondents. The one emphatic early call is the post-lockdown rebound — 88% of the October 2020 round’s mass above +6% for 2021Q2, which delivered +15.3% — and it was aided by a base effect already embedded in the observed collapse, which makes it weak evidence that the survey can anticipate a downturn nobody has yet seen in the data.

One number in the survey’s favour before anything critical. On a Brier score for the probability of negative growth, the survey scores 0.095 against 0.121 for a constant benchmark set at the sample’s unconditional frequency and 0.132 for an expanding-window climatology using only information available at each origin; the ranking is unchanged excluding the 2020–21 reversal, though a paired bootstrap on the differential does not separate it cleanly from zero (method notes). That gain does not establish early-warning value: a Brier score rewards probability accuracy across all quarters, whereas the macroprudential question is whether the density adds enough information before rare, persistent downside states. Everything that follows is about where the survey’s information sits — in time, and in the distribution.

How to read the charts. A survey round is the forecast origin; fieldwork for the 2008Q3 round, for example, ran from 16 to 23 July 2008. Each round carries a rolling one-year target (annual growth to a quarter roughly two quarters beyond the round, defined as one calendar year beyond the latest available data) and a two-year target roughly six quarters out. Charts are indexed by target quarter, so each forecast sits at the outcome it concerns; the round that produced it, and what it could know, are stated in the annotations. Outcomes are measured on the current GDP vintage (20 July 2026) except where real-time releases are cited.

The warning arrives late

The record across the three recessions reads as follows. July 2008: 5.0% — with published national accounts covering only the first quarter of 2008; Eurostat’s first estimate showing the euro area contracting arrived weeks after fieldwork closed, and even that release put annual growth at +1.4%. October 2008: 40.8%. Two things reached the panel in between. Eurostat’s August flash showed the euro area contracting on the quarter for the first time, though annual growth in that release was still +1.4%; and Lehman Brothers filed on 15 September. No third-quarter national-accounts print existed yet. The repricing followed a financial collapse and a first contraction print together, not a downturn visible in the annual data the forecast concerned. January 2009: 92.8%, after recessionary data had become unmistakable, but while the eventual depth remained far beyond the distribution’s centre. The consensus repriced hard, but not before the year it was forecasting had begun, and it still understated the depth. The 2011–13 recession is the harder case, because it arrived gradually and was debated extensively in real time: the survey’s probability never exceeded 41.7% across six consecutive quarters of negative annual growth. The magnitudes matter for reading that number: outcomes ran between −0.2% and −1.3%, and a 40% probability attached to a mild contraction is not obviously miscalibrated. What the densities did not do is move their centre: the mean of the distribution stayed positive in every one of those rounds, falling from +1.5% to +0.2% but never crossing zero, while the outcomes it was forecasting were negative throughout. October 2019 put 8.3% on the year to 2020Q2, which realised at −13.9%; the pandemic stays in every headline number below and is labelled for what it is, an extreme exogenous shock with little plausible pre-event predictability.

Then the other direction. The sample’s highest one-year probability of negative annual growth — 60.6%, in the October 2022 round, for the year to 2023Q2 — was the largest ex-ante probability in the record not followed by the event, an outcome to which the forecast itself assigned meaningful probability. Revisions could in principle have erased a contraction the forecasters correctly saw. They did not: growth over the year to 2023Q2 was positive in the initial Eurostat release (+0.6%), the subsequent estimate (+0.5%), and today’s vintage (+0.55%). Quarterly growth through late 2023 was weak on every vintage, but the negative annual-growth event to which the probability referred never occurred.

The horizon result completes the picture. At two years the series never opens: its 27-year maximum is 15.0%, set in January 2009, when the recession was already in the published data. In October 2022 the same forecasters who put 60.6% on a negative outcome one year out put 9.1% two years out — not one shock seen at two distances, since the targets are different years and a panel expecting a short contraction followed by recovery would produce exactly that split. What the two-year series establishes is a full-sample property: across 27 years and three downturns, it never opened, whatever the state of the world at the time of asking. For the probability of negative growth at one year, in the downturns this record contains, the tail is a nowcast.

Coverage, honestly bounded

Realised growth fell below the survey’s one-year 10th percentile in 24 of 107 realised target quarters; excluding the four 2020 targets, 21 of 103. Inference has to respect the dependence first: overlapping annual windows make consecutive breaches dependent by construction, some clustering is mechanically induced by that overlap alone, and the effective sample is closer to the number of downturn episodes than to 107. A moving-block bootstrap puts a 95% interval of roughly 10% to 38% around the 22.4% point estimate — the interval excludes neither calibration nor coverage far worse than nominal. What the sample does locate is where the misses sit: in the downturn episodes, twice in runs of six consecutive quarters, rather than scattered through normal times. Because overlap mechanically induces dependence, this is evidence of an episode pattern, not a formal estimate of state-conditional coverage.

Conditioning on the survey’s own probabilities gives a more useful description than the raw rate:

The 5–15% cell contains four negative outcomes in fourteen observations — suggestive, and far too small to define a stable risk band. The relationship is non-monotonic in the middle bands, largely because elevated pessimism persisted after the 2011–13 downturn and again after the 2022 episode. And the 0–5% row, which looks consistent with calibration on frequency, contains a sting: its two events were the onset quarters of the two cyclical downturns, the years to 2008Q4 and to 2012Q1. Two of the 24 breaches are marginal — 2003Q3 by nine basis points, 2018Q4 by seven — and all of this is measured against today’s vintage rather than the data forecasters faced. Frequency calibration is not magnitude protection.

Two episodes, and what the panel knew

The two episodes compress the whole argument. In the top row the densities chase the outcome down, round after round, and the instrument itself limits what they could say: in July 2008 the lowest bin on offer was “below 0%”, so a respondent could express that a contraction might happen but not distinguish a mild one from the −5.7% that arrived. By January 2009 the panel had piled its mass into a spike near −1% while the year to 2009Q3 came in at −4.4% — direction finally right, depth still wrong. In the bottom row the mass walks below zero through the winter of 2022–23 while the outcome stays above it in every panel.

The panels invite a question the aggregate cannot answer: was early warning present in individual views and averaged away? The threshold decides the answer, so here is the sweep. Requiring the 90th percentile above 20% while the aggregate sits below 10%, the screen fires four times across 111 rounds, three resolved, and the only one followed by negative growth is October 2019 — the pandemic quarter, which no growth density plausibly foresaw. Lower the bar to 15% and July 2008 is caught, as it should be; but the same screen then fires twelve times, nine resolved, three followed by negative growth, two of those three again the pandemic. Lower it to 10% and four of fourteen resolved cases are followed by negative growth, three of them pandemic quarters. At that same threshold the upper decile crossed together with the aggregate in 2008 and in 2022, and one round earlier in 2011. July 2008 remains the sharpest exhibit: the median respondent attached zero probability to negative growth, the 90th-percentile respondent 15%, the most pessimistic 40%. The aggregate summarises that spread as 5% and does not publish the dispersion; no threshold in that range turns the dispersion into a rule: the cuts loose enough to catch 2008 also fire through years in which nothing followed. Other quantiles, trimmed or weighted aggregates, and panel-composition effects are not tested here. A fuller cross-sectional treatment, including threshold sensitivity, follows in Wednesday’s Note.

Mechanisms, and the Growth-at-Risk toolkit

Three mechanisms fit the record, at the strength the evidence allows. Aggregation: the equal-weight average is the ECB’s convention, not the only object one could build, and it compresses disagreement; in 2008, dispersion widened before the aggregate moved substantially, though on the screen above that pattern did not generalise into a rule, and whether alternative aggregations would have moved earlier is not tested here. Updating: a 5% ex-ante probability on an event that then occurs is not, in one draw, miscalibration; the late-arrival pattern rests on the two cyclical downturns, since the pandemic is not evidence either way; two episodes bound what a pattern can prove. The instrument: a fixed bin grid limits how finely severity can be expressed once probability mass reaches an open-ended tail bin, and the July 2008 grid is the exhibit.

The ESRB’s recent tail-risk work is the right reference point. Andersen et al. (2025) present a growth-at-risk model and the SPF density as complementary perspectives on euro-area downside risk — parallel lenses, not a weighted composite — and they note that the tighter survey distribution aligns with evidence that the SPF tends to underestimate the tails of the distribution, citing Clements (2011), Rich, Song and Tracy (2012) and the ECB’s own 25-year retrospective (Allayioti et al. 2024). On their own readings the two lenses diverged over 2025, the model’s 10th percentile falling to between −1.7% and −2.6% while the survey’s moved to about −0.4% — figures produced under their specifications and conditioning, not aligned here to a common origin and target.

The retrospective evaluates the densities with probability integral transforms. A PIT-based assessment asks whether realised outcomes are plausibly drawn from the forecast distribution over the full record; this piece asks a different question — whether the survey’s lower tail remained protective during the small number of persistent downturn episodes. The two exercises are complements, not competing verdicts. For instruments whose calibration, legal process and transmission require a horizon materially longer than the survey’s short-horizon information, the record here makes the aggregate density a weak candidate for a stand-alone tail-risk signal. Whether financial, market-based or balance-sheet indicators would have done better on the same dated information sets is not tested here — the comparison a user would most want, and the one this piece does not make. A lagging density that correctly identifies an economy already in a bad state retains monitoring value; the scope of this piece is ex-ante information for slow-moving tools. Whether any of this should change how such inputs are weighted inside a framework is a question for the framework’s owners, and outside this piece.

The current read

The Q3 2026 round expects growth of 0.6% this year, after a downward revision of 0.4 percentage points, and 1.2% in 2027. It attaches 5.4% to negative annual growth one year out and 5.1% at two years — a spread not worth interpreting. A 5.4% probability is low. The historical record assembled here provides no basis for treating a low survey probability as stand-alone reassurance, particularly for a user concerned with slow-building downside risk.

Across 1999–2026, then: the equal-weight aggregate SPF density scored better than two naive benchmarks, descriptively and by a margin not cleanly separated from zero, and repriced sharply as conditions deteriorated. But its record provides limited evidence that it identifies euro-area downturn tails early enough to stand alone for slow-moving macroprudential decisions. Its two-year tail barely opens; its lower-tail misses, on the current vintage, concentrate in the few persistent downturn episodes — though that small number of effectively independent episodes also prevents a decisive calibration verdict; and respondent disagreement, at every threshold tested here, does not supply a stable remedy.

Paweł Fiedor — The Macro Prudential View

Data described are as at 28 July 2026 (SPF round of July 2026, released 24 July; GDP vintage of 20 July 2026).

The author works at the Central Bank of Ireland and in the ESRB Secretariat. This piece is written in a personal capacity; views are the author’s own and should not be attributed to the Central Bank of Ireland, the ESRB, or the Eurosystem. It draws exclusively on publicly available sources and takes no position on whether the policy measures discussed should be adopted.

Method notes. ECB SPF individual-forecast microdata, 111 rounds, 1999Q1–2026Q3. Individual densities rescaled to sum to one where reported probabilities total between 90 and 110, otherwise dropped; equal-weight averaging across respondents (the ECB’s convention). Bin labels read as contiguous half-open intervals (upper label extended by 0.1pp); open-ended bins treated as one interior-bin-width wide for percentiles, means and density heights. Rolling targets identified by their gap from the round (roughly two and six quarters). Realised growth: euro area (EA20) chain-linked volumes, Eurostat via FRED, vintage 20 July 2026; outcomes evaluated on the current vintage except where real-time releases are cited. Exceedance interval: moving-block bootstrap, 8-quarter blocks, 10,000 draws; the interval is a resampling diagnostic rather than a confidence interval, and is stable across block lengths (4q: 11–36%; 6q: 10–37%; 12q: 11–38%; 16q: 11–37%). Dropping the two marginal breaches leaves 22 of 107 (20.6%); both sit inside larger clusters, so the episode pattern is unaffected. The GDP section of every round since 1999Q1 carries rolling one- and two-year targets alongside calendar-year questions — the 2008Q3 file, for instance, lists target periods 2008, 2009, 2010 and 2013 together with 2009Q1 and 2010Q1; the rolling densities are the object used here throughout, and the July 2008 round records 57 density responses for the year to 2009Q1. Percentiles are insensitive to the open-tail convention: the p10 falls inside the open bottom bin in 3 of 107 rounds, and the breach count is 24 of 107 whether that bin is assumed one, two, four or eight interior widths deep. Outcomes use the current-composition EA20 series; membership changed over the sample, so the geographical aggregate is not identical to the one every historical round forecast. The constant Brier benchmark uses the full-sample frequency and is therefore in-sample; the climatology benchmark uses only data available at each origin; excluding 2020–21 targets, the ranking is unchanged (survey 0.072, constant 0.099, climatology 0.111). The climatology benchmark starts once twelve quarters of history are available (first evaluated target 1999Q4). A paired block bootstrap on the survey-minus-climatology score differential gives a mean of −0.037 with a 95% interval of [−0.095, +0.005]: the survey is ahead in 95% of resamples, but the differential is not cleanly separated from zero, so the ranking is reported as descriptive. As a descriptive diagnostic only: excluding 2020–21, the mean of the aggregate density correlates 0.83 with annual growth over the year ending at the survey round and 0.67 with the growth it forecasts; overlapping annual windows induce persistence in both series, so these are raw cross-correlations, not a formal timing test. Cross-sectional statistics are computed over respondents with usable densities under the same rescaling rule; the panel ranges from 24 to 58 respondents per round at the one-year horizon, and changes in composition may contribute to measured dispersion. Derived series and scripts available on request.

Sources: ECB, Survey of Professional Forecasters — Q3 2026 report (24 July 2026) and individual-forecast microdata, 1999Q1–2026Q3 — https://www.ecb.europa.eu; Eurostat via FRED, euro area (EA20) real GDP, chain-linked volumes, series CLV10MNACB1GQSCAEA20Q, vintage 20 July 2026 — https://fred.stlouisfed.org; Eurostat, euro-indicators releases for 2023Q2 GDP: preliminary flash of 31 July 2023 (2-31072023-BP), flash of 16 August 2023 (2-16082023-AP) and estimate of 7 September 2023 (2-07092023-AP) — https://ec.europa.eu/eurostat; Andersen, F., M. Andersson, S. G. Cecchetti, R. Giuliana, U. Pelca, J. Rice, C. Sarchi and A. de Vries (2025), “Navigating Tail Risks: Assessing Euro Area Economic Growth and Equity Market Vulnerabilities”, Macroprudential Commentaries, Issue 9, European Systemic Risk Board, December, and the accompanying VoxEU column of 16 December 2025 — https://www.esrb.europa.eu, https://cepr.org; Allayioti, A., R. Arioli, C. Bates et al. (2024), “A look back at 25 years of the ECB SPF”, ECB Occasional Paper No 364; Clements, M. P. (2011), Journal of Money, Credit and Banking 43(1); Rich, R., J. Song and J. Tracy (2012), Federal Reserve Bank of New York Staff Report No 588; Adrian, T., N. Boyarchenko and D. Giannone (2019), “Vulnerable Growth”, American Economic Review 109(4). Chart data as cited in each figure.

The proposition

Every euro that backs the stablecoin EURCV sits as cash at Société Générale. The issuer, SG-FORGE, is an e-money institution licensed by the ACPR, and Société Générale is its own parent bank. There is about 129 million euro of it. That is a small number, and nothing in what follows suggests that a stablecoin this size threatens a globally systemic bank.

But EURCV is one of the three largest euro-denominated stablecoins, and the only ones publishing reserve disclosures detailed enough to read against each other. Smaller euro tokens exist; on the figures reconciled below these three are about nine-tenths of the euro-pegged market by value. Together they make a useful small map of how a private euro connects to the banking system. The connection is closer, and more varied, than their combined half-billion-euro footprint would suggest.

Set the scale first. Euro stablecoins are a rounding error in a market that is almost entirely dollar. The euro-pegged tokens are worth something over half a billion euro between them; the whole stablecoin market is around 300 billion dollars. The euro’s share of it has never held above four-tenths of a percent, and sits near a fifth of a percent today. What follows is a question about structure, at a size that is not yet a question about stability.

That structural question is timely because Europe is, at the same moment, building a public euro: in July the European Parliament handed its negotiators a mandate on the digital euro, a retail central-bank instrument still some years from issuance. The proposal in front of the legislators is a retail instrument, aimed at the card rails; the Eurosystem’s work on settling tokenised transactions in central-bank money is a separate, wholesale track. Neither is aimed at the crypto-native, on-chain retail euro, which is left, for now, to private issuers under MiCA.

For most of what that leaves in private hands, the supervisory map is drawn. MiCA sets who may issue these tokens, who supervises them, and how their reserves must be held, and the European Banking Authority has written the recovery and redemption playbooks. One seam is less settled, and it is the one I want to sit with: what happens to a private euro’s reserve when the bank holding it is the one in trouble, and how a promise to redeem the token at par meets a reserve that a bank resolution can freeze.

Three reserve structures

Begin with what the three share. Each holds its reserve as cash at commercial banks rather than as a book of securities. For the two larger issuers that is documented. Circle’s EURC held 373 million euro at the end of May, all of it cash at regulated European banks, on its monthly attestation. EURCV’s 129 million is likewise all cash, on SG-FORGE’s own reporting. EURI, the smallest at around 34 million euro, holds its reserve in a segregated fiduciary structure; its issuer describes the assets as cash and cash-equivalents but does not itemise them. For the two documented cases there is no securities book to sell into a falling market; a redemption is met by drawing down a bank balance.

The differences are in whose bank, and on what terms.

Circle is a licensed e-money institution rather than a bank, and it spreads EURC’s cash across several regulated institutions in the EEA at arm’s length. If one of those banks fails, EURC has others.

EURCV concentrates. Its cash sits, in full, at Société Générale, which is also the parent of the issuer. The white paper does not hide this; it tells holders they carry “predominant SG credit exposure.” Reserve and issuer share one point of failure by construction.

EURI differs in its legal structure. Its funds sit in a segregated, bankruptcy-remote fiduciary structure under Luxembourg’s 2003 safeguarding law, with EURI holders named as the beneficiaries; what the structure holds is cash and cash-equivalents, whose allocation and custody are not itemised, and the claim on it is built to survive the issuer’s failure. That, with the undisclosed composition, is why it sits here as a contrast rather than a third instance of the same thing.

So the three do not share a reserve model. EURC diversifies its bank exposure, EURCV concentrates it inside its own group, and EURI segregates the reserve away from the issuer.

None of this needs a contrarian reading of the reserves. When the ESRB examined crypto-to-bank links last October, it treated EURCV’s backing as bank deposits for the plain reason that the cash is at a bank and the token is a claim that runs through it.

Three names are not a market. These are the euro stablecoins large enough to disclose reserves worth reading; smaller issues, and anything that changes its arrangements after this is written, sit outside the argument. Three is enough, though, to fix the feature the rest turns on. For the two documented cases the reserve is a commercial-bank deposit, whole and plain: a redemption is a bank withdrawal, and there is nothing else to sell. EURI’s safeguarded structure may behave differently, and its composition is undisclosed, so I hold it apart. For the deposit cases, at least, the bank is not a detail of the structure. It is the structure.

A two-way interface

The mechanism here is a deposit story, so take the two cases whose reserves are documented as deposits, EURC and EURCV. EURI sits out, and for a reason worth stating: with its composition undisclosed, a run on it might work through the sale of whatever assets sit in its fiduciary structure rather than the drawdown of a bank balance. That is a market-liquidity channel, not a deposit one, and the disclosure does not let me follow it. Between the two cases that remain, stress moves in both directions, and unequally.

The direction that matters at today’s sizes runs from the bank to the token. If a reserve bank comes under stress — a funding problem, or a failure — the issuer’s access to the cash behind its token is suddenly in question, and the promise to redeem at par becomes a matter of timing and liquidity rather than a certainty. This is the credible leg. It does not depend on the stablecoin being large. It depends on the bank.

The reverse leg runs from the token to the bank: a wave of redemptions obliges the issuer to pull its reserve cash out, which lands on the reserve bank as a concentrated wholesale deposit outflow. This one is already discounted. A non-operational deposit from a financial customer attracts the highest outflow assumption in the bank’s thirty-day liquidity rules, feeding the bank’s aggregate requirement rather than earmarking assets against this one balance; and at half a billion euro across the three, with 129 million the largest single-bank balance any of them discloses, the outflow would not normally trouble a large bank. This is the weaker leg, and it grows teeth only at a scale the tokens have not reached.

The two legs meet in one place. Where the reserve sits at the issuer’s own group — EURCV at Société Générale — a shock to the bank and a shock to the issuer are closer to being the same event, and each can feed the other. For EURC, spread across arm’s-length banks, they stay largely separate.

The one time something like this played out, it played out in dollars. In March 2023 roughly eight percent of the reserve behind USDC was stuck at Silicon Valley Bank when it failed, and the token broke its peg, trading down to about 87 cents before recovering over the following days, as the issuer set out its reserve position and the US authorities invoked a systemic-risk exception to make all SVB depositors whole. The depeg travels to the euro case; the recovery does not. European resolution is built not to extend that kind of public make-whole to uncovered wholesale deposits of financial institutions. That episode illustrates the bank-to-peg direction cleanly enough, though it understates the concentrated case: USDC’s exposure was a partial slice of a diversified reserve, and EURCV’s would be the whole of it. There is no euro precedent here, only a dollar illustration and a euro structure more concentrated than the illustration.

Two qualifications carry through the rest. The subject throughout is the peg and the timing of redemption. A principal-loss outcome is not assumed, and what a holder can claim, and how quickly, depends on the structure. EURI’s segregated beneficiaries stand differently from a holder’s ordinary claim on a parent bank. And the magnitude stays honest: at these sizes the concern is the bank-to-peg direction, and the single-name concentration that would bite harder if issuance ever grew.

What is settled, and what is left implicit

Most of the supervisory questions these tokens raise have answers. Under MiCA a euro e-money token must be issued by a licensed e-money institution or a bank; a national competent authority authorises the issuer and supervises its conduct and prudential compliance; the reserve must be held and safeguarded to a standard the regulation sets. The EBA has gone further and written the recovery and redemption plans every issuer must keep, whatever its size, and those guidelines already ask issuers to account for the size, composition and concentration of the assets behind the token.

Above this sits the significant-token regime, for issues large enough to matter system-wide. Designation is not a single trigger: it requires meeting at least three of a set of statutory criteria, one of which is that the value issued, market capitalisation or reserve size exceeds five billion euro, alongside holder numbers, daily transaction activity, cross-border reach and interconnectedness. Once designated, the EBA supervises the issuer’s compliance with the significant-token requirements, while the home authority keeps the entity’s own authorisation. On size, no euro token is near: EURC’s 373 million euro is about a thirteenth of the five-billion mark, and the three together about a tenth of it. Holder counts and transaction activity can be read off the chain, and would need their own study; on the strength of the size test alone these tokens are an order of magnitude short, and none has been designated.

The assigned map is real, then, and worth stating plainly before pointing at what it does not cover. What it does not cover is a specific moment, and the moment is nameable.

The reserve behind one of these tokens is, at the bank, a deposit placed by a financial institution, and the deposit guarantee does not straightforwardly cover it. The scheme excludes deposits placed by financial institutions, and whether an e-money institution’s safeguarding account sits inside that exclusion has been read differently across member states, a divergence the EBA documented in opinions in 2019 and 2021. The 2026 reform of the framework, part of the wider crisis-management package, harmonises the treatment: client funds held in segregated safeguarding accounts can carry protection up to 100,000 euro per identified client.

Two questions sit behind that protection, and only the second is a matter of plumbing. The first is legal: whether someone who bought the token on a secondary market is a client of the safeguarding arrangement at all. Look-through of this kind is built around the account-holder’s own clients, and whether it reaches a subsequent transferee of a bearer-style claim is not settled on the face of the reform. The second, if it does reach them, is operational: a deposit guarantee pays against a register, matching each protected client to an amount within the few days a failed bank allows, and a self-custody wallet address is not that. Both run through the technical standards due in 2027. Until they are written, what a dispersed on-chain holder is owed by a deposit guarantee, and how it would reach them, is unresolved. Deposit guarantee assumes an institution that knows its clients. A bearer instrument on a public chain is built so that, past issuance, its issuer largely does not.

Set against that is the reserve bank’s own resolution, where the proximate risk is timing. If a reserve bank is failing, a pre-resolution moratorium can suspend payment out of the account for a short window. Whether that holds up redemptions depends on what else the issuer can reach: cash at other banks, committed lines, or a buffer held outside the frozen account. For a diversified reserve the suspension may be bridged; for a reserve held at one bank there is nothing else to draw on. None of the three publishes its contingency arrangements, so the size of that cushion is not something a reader can check. An actual write-down of the deposit is a further and more remote matter. The deposit is an ordinary unsecured claim within the scope of bail-in. It gets no help from the deposit preference in the creditor hierarchy, which lifts covered deposits and the eligible deposits of individuals and smaller firms above ordinary claims and does not reach a deposit placed by a financial institution. What sits in front of it is the stack of own funds and eligible liabilities a large bank must maintain, and resolution authorities keep discretion to spare liabilities whose write-down would spread contagion. An intragroup balance is a weaker candidate for that discretion than a third-party deposit, since internal loss-absorbing capacity exists to be used inside the group. That buffer is thinner at a smaller reserve bank than at a G-SIB. The MiCA duty to redeem at par does not displace any of these resolution powers, and I have found no public, stablecoin-specific protocol setting out how a redemption obligation and a bank moratorium are meant to act on the same deposit at the same time. Resolution planning is not all public, so this is an absence in the published framework rather than proof that no authority has considered it.

That is the seam, and it is a narrow one. The legal hierarchy is settled and the reserves are supervised. What is not yet settled is operational: how a par-redemption promise is honoured when the deposit behind it can be frozen by a bank moratorium, and how far that protection reaches when the holders who matter sit out on the chain, outside the register a deposit guarantee works from. Neither is a supervisory failing. Both wait on rules and playbooks now being written.

Two features of the three cases sharpen this. On the financial side, concentration is the risk, and EURCV carries the most of it: its entire reserve stands or falls with one bank, which happens to be its own parent, so an idiosyncratic problem at Société Générale would hit reserve and issuer together. EURC, spread across arm’s-length banks, is the less exposed case, though its per-bank allocation is not published. The institutional side offers EURCV no offset. It is tempting to read the in-group structure as tidy — one bank, one supervised group, one resolution authority — but a resolution authority’s mandate runs to continuity of critical functions and financial stability, and an intragroup deposit from a small subsidiary is exposed to how that mandate is exercised. Its legal rank is not lowered by the affiliation. What changes is that one event hits reserve and issuer together, and the discretion over that deposit belongs to an authority weighing the group. EURC’s reserve, spread across third-party banks a non-bank issuer does not control, is instead the clean illustration of the cross-perimeter problem: the duty to redeem and each bank’s resolution answer to different authorities on different balance sheets. The concentrated structure is the more exposed, and the diversified one is where the coordination question shows most plainly. Neither removes the seam. EURI, issued by a bank rather than an e-money institution, falls in a different exclusion category again, one more reason to treat it apart.

One clarification forecloses an easy objection. It might seem that a bank leaning on a large, redemption-sensitive deposit is carrying a hidden funding risk that supervisors have missed. The prudential rules already discount it. An at-call, non-operational deposit from a financial customer attracts the highest outflow assumption in the thirty-day liquidity rules and little or no credit as stable funding in the one-year rules, so it flatters neither ratio and will not support longer assets. That treatment runs through the bank’s aggregate position rather than setting assets aside against this one deposit, so it reduces the concern without erasing it. The sharper residual sits with the token: a par promise exposed to a moratorium, and holder protection that is unresolved.

Scale

Return to scale, because the seam above is a structural feature and not a live exposure, and the difference matters.

The euro-pegged stablecoins are, collectively, about a fifth of a percent of the combined euro- and dollar-pegged market. The share is small, and it has not grown. It reached about four-tenths of a percent in early 2022, on the back of tokens that have since wound down, and it sits below that level now, after MiCA’s stablecoin rules took effect in 2024. Non-compliant euro tokens were delisted and the market consolidated around compliant issuers, and through all of it the euro’s on-chain share stayed below where it had been four years earlier. Read the chart as a level and a date. It does not show that MiCA held the share down; the peak came first, and the flat line came after.

A word on the numbers, since two measures sit close together in this piece. The market-cap series counts every euro-pegged token against every dollar-pegged one, in dollars, from public on-chain data; the reserve figures earlier are the disclosed euro reserves of three named issuers, on their own reporting dates. At the euro’s level over the June-July window, about 1.14 dollars, the three issuers’ half-billion euro is roughly 610 million dollars, against a euro-pegged total near 660 million. The difference is other, smaller euro tokens and the usual gaps of date and source. The three are a large majority of the euro-pegged universe, not the whole of it.

The scale does not dismiss the structure; it places it. A monitoring seam on a market this size is a thing to specify before it grows, and the reform timetable in the last section is, in that light, well-judged: the rulebook is being written while the market is still small. The mistake would be to read the structure as a present danger. Nothing here is a present danger.

The public euro

One question has hung since the opening: why is any of this a public-versus-private story, rather than simply a story about three private tokens? The answer is the digital euro, and it is worth being careful about what the public project is built to do.

The case for a public retail euro is, in large part, a sovereignty case, and the sovereignty case is about the card rails. On the ECB’s figures for 2022, published in its 2025 report on card schemes, international schemes — Visa and Mastercard — carried around 61 percent of euro-area card payments measured within the area, and 13 euro-area countries rely on international schemes entirely, with no domestic card scheme of their own. That is the dependence the digital euro is designed to answer. (The figure is a card-scheme share rather than a claim about the whole acquiring and settlement stack, and it is dated; treat it as context.)

What the public project is built for shapes what it leaves to others. On the architecture now before the legislators, the digital euro addresses retail payments, and the Eurosystem’s separate DLT-settlement work addresses wholesale infrastructure. Neither is primarily designed for the crypto-native, on-chain retail euro: the euro that settles a trade on a public chain or sits in a self-custody wallet. That space is, for now, private, and filled by the three tokens this piece has been about. The division may prove lasting, or may be an artefact of the current design and timetable; either way, it is why a supervisory question about private euro reserves is also a question about where the public euro’s own perimeter is drawn.

What the three show

None of the three tokens is big enough to be a problem for the banking system. That is the frame the argument sits in, and it holds all the way through. What the three show is smaller and more particular than a systemic risk. It is an interface.

A euro stablecoin’s holder is promised redemption at par. Behind that promise, in two of the three cases, is the issuer’s access to a cash deposit at a commercial bank — spread across several for EURC, concentrated at its own parent for EURCV; behind the third is a safeguarded fiduciary claim whose composition its issuer does not disclose. The gap that can move a peg is the fast one. A resolution authority can suspend payment out of that deposit for a short window, and a par promise that cannot be met on demand is a broken peg long before any question of compensation arises. Behind it sits a slower question about the holders: whether a deposit guarantee, capped and paid against a register of identified clients, reaches people who acquired the token on-chain and never entered that register. The first is a stability question and the second a matter of holder protection. The public framework has not yet settled the operational answer to either.

That is a narrow finding, and it should stay narrow. Europe’s on-chain euro is supervised, and half a billion euro of tokens is no threat to anything by itself. The finding is only this: a small part of the euro’s money, private and on-chain, connects to the banking system through its reserves in three different ways, and meets that system at a join the rulebooks are still, on their own timetable, closing.

Paweł Fiedor — The Macro Prudential View

Positions described are as at 17 July 2026 unless a different disclosure date is given; reserve figures carry their own attestation dates and are not a single point-in-time snapshot.

The author works at the Central Bank of Ireland and in the ESRB Secretariat. This piece is written in a personal capacity; views are the author’s own and should not be attributed to the Central Bank of Ireland, the ESRB, or the Eurosystem. It draws exclusively on publicly available sources and takes no position on whether the policy measures discussed should be adopted.

Sources: Circle, EURC reserve attestation, 29 May 2026 — https://www.circle.com/transparency; Société Générale-FORGE, EURCV white paper and daily reserve disclosure — https://www.sgforge.com; Banking Circle, EURI white paper — https://www.bankingcircle.com; DefiLlama, stablecoin market capitalisation by peg type — https://stablecoins.llama.fi; CoinGecko and CoinMarketCap, EURCV and EURI market data, July 2026; ESRB, “Crypto-assets and decentralised finance” and accompanying recommendation, October 2025 — https://www.esrb.europa.eu; ECB, “Report on card schemes and processors”, February 2025 (data as at February 2024), and material on the digital euro and wholesale DLT settlement (Pontes, Appia) — https://www.ecb.europa.eu; European Parliament, digital euro negotiating mandate, July 2026 — https://www.europarl.europa.eu; Regulation (EU) 2023/1114 (Markets in Crypto-Assets), including the significant-token criteria (Articles 43 and 56) and the recovery- and redemption-plan requirements (Articles 46 and 55); European Banking Authority, Guidelines on recovery plans under Articles 46 and 55 of MiCAR (EBA/GL/2024/07), Guidelines on redemption plans, and Opinions on the treatment of client funds under the DGSD (2019, 2021) — https://www.eba.europa.eu; Directive 2014/49/EU (Deposit Guarantee Schemes Directive), Articles 5 and 6; Directive (EU) 2026/804 amending the DGSD (Article 8b), Directive (EU) 2026/806 amending the BRRD, and Regulation (EU) 2026/808 amending the SRMR (Crisis Management and Deposit Insurance package); Directive 2014/59/EU (Bank Recovery and Resolution Directive), Articles 33a and 69; Single Resolution Board, resolution tools — https://www.srb.europa.eu. Chart data as cited in each figure.

A new BIS paper finds that bank–sovereign risk co-movement has become more closely associated with banks’ non-bank exposures. The European holder data show why the transmission chain cannot be reconstructed from any published dataset.

In mid-July, the French ten-year briefly traded through the Italian one. The OAT yield reached 3.94% on 15 July, its highest since June 2009; at monthly close, France’s spread over Bunds was near 80 basis points, against Italy’s 83. The same week, the Bank for International Settlements published a working paper — Avdjiev, Hardy and Jager, “The evolving nexus: sovereigns, banks and NBFIs”, dated 14 July — arguing that the sovereign-bank doom loop of the euro area crisis has not closed so much as changed its plumbing. And the Financial Stability Board’s Nonbank Data Task Force, chaired by Andrew Bailey, is due to report by mid-2026 on its test case: leveraged trading strategies in sovereign bond markets, chosen precisely because authorities cannot currently see who runs them, at what leverage, financed by whom. As I write, the measurement problem this piece describes has outlived the task force’s own timetable.

This piece takes the BIS result and asks where its channel lives in the European data. The answer is uncomfortable in a specific way: the published datasets cannot tell us which leg of the channel matters most once risk passes between sovereign holders, non-banks and their bank counterparties.

The paper, and the map

The BIS paper does something clean. Using EBA bank-level data and the BIS consolidated banking statistics, it shows that before 2016, banks whose sovereign holdings were larger co-moved more tightly with their sovereigns — the classic result, the mechanism the euro area crisis made famous. After 2016 that relationship weakens significantly. In its place, banks’ exposures to the financial sector — and, where the data allow NBFIs to be separated, to NBFIs — become a significant determinant of the co-movement, strongest for high-risk counterparty countries, even as direct bond holdings matter less. The authors find the same pattern at the country level; find the post-2016 financial-exposure link stronger for periphery banks holding more liquid assets, a pattern the paper itself reads as consistent with short-term repo lending to leveraged NBFIs; and find it on the liability side too, through the deposits NBFIs place with banks, in a shorter sample beginning in 2019 when those data start.

Call the first arrangement the first geography: the domestic feedback most legible in the post-crisis disclosure and sovereign-exposure metrics — banks holding their own sovereign. The paper finds that the measured association has shifted away from direct holdings toward NBFI-related exposures.

The paper is also candid about what it cannot see. Its NBFI exposure measure is proxied from financial-sector exposures, refined with a non-performing-loan share; its NBFI-sovereign leg exists only for domestic holdings, because cross-border data on who holds what, levered how, does not exist in usable form. That confession is the door this piece walks through.

The next sections test how far published holder and balance-sheet data can trace those dashed edges, and where they stop.

Where the bonds went

Start with the intuitive version of the story: banks sold their sovereign bonds, non-banks bought them, the risk moved house. The euro-area holder data refuse to cooperate.

The ECB’s who-to-whom accounts let us decompose who holds the general government debt securities of the old periphery — Spain, Greece, Ireland, Italy, Portugal — from 2015 onwards. Domestic banks did step back: from 22% of the total to 15.5% by early 2026. But domestic non-bank financial institutions stepped back further, from 19% to 10.5% — and not only in shares. Italian domestic NBFIs held roughly €453bn of Italian government paper at end-2015 and hold about €327bn now.

The eleven-year window contains two distinct ownership configurations. During QE, the Eurosystem’s share rose above 30% as the non-resident residual declined from around 41% to 31%. Since 2022, as Eurosystem holdings have run off, the non-resident share has rebuilt to roughly 39%. The aggregate series cannot identify the marginal buyers behind that change.

The holder data therefore rule out the simple story. Domestic non-banks did not replace banks as the larger holders of periphery debt. That is worth being precise about, because it does not rule out domestic NBFIs as a transmission channel: they still hold three hundred billion euros of Italian paper, and the BIS result works through the interaction of existing holdings with sovereign risk, not through purchases. What the shares refute is replacement. If the BIS channel operates in Europe, it runs less through visible domestic cash holdings than through financing, collateral and cross-border intermediation — which is where the published data begin to thin.

One flow number sharpens the stock picture, though it concerns France rather than the periphery sample above, and it is a flow, not a holding: the French Treasury’s financing programme for 2026 provides for €310bn of medium- and long-term issuance net of buybacks — a record — landing while the Eurosystem’s holdings run off. The arithmetic implies substantial net absorption by private investors, whoever they turn out to be.

Two geographies

So who does hold the risk, and where? Plotting each euro-area sovereign by its spread over Bunds and the share of its debt securities held by non-residents produces a map that inverts the intuitive one.

Foreign ownership tends to be high where risk is low. Germany’s debt is 62% foreign-held, Finland’s 76%, Austria’s 66%. At the other end sit the highest-spread sovereigns with the lowest non-resident shares in this sample: Italy at 34% foreign, Greece at 21%. The middle is genuinely mixed — Portugal at 40% and Spain at 48% do not fit a tidy home-held grouping — but the clustering is visible, and it is home bias: the configuration in which sovereign stress lands first on domestic bank balance sheets, which is exactly why the post-2011 apparatus watches it.

France does not fit either cluster. It is an Italian-sized market — nearly €3tn of government securities outstanding — that in July briefly traded at periphery-like spread levels, on a foreign-held base of about 59%. The comparison is about the coexistence of similar headline spread levels and sharply different holder geographies; it is not an assertion that French and Italian credit risk, market liquidity or institutional backstops are equivalent. The fiscal backdrop, in one sentence: a 5.1% deficit in 2025, debt at 115.6% of GDP and projected by the Commission to keep rising, and four rating actions since 2024, most recently Moody’s to Aa3.

The question France poses is the interesting one, and it is not really a question about France. What does repricing look like when a large issuer’s debt is held mainly by investors beyond the issuer country’s own holder map, across datasets that do not connect positions to financing, leverage or liquidity terms? Belgium, at 67% foreign-held and a 55 basis point spread, sits on the same line at smaller scale. The old configuration made a larger part of the bank–sovereign link visible in domestic balance-sheet data. For issuers such as France, the visible holder base is dispersed across jurisdictions and broad sectors, while the transmission chain remains only partly observable.

The observable linkage

If sovereign risk reaches banks through their non-bank counterparties, the pipe should show up on bank balance sheets. It does — and its most striking property is how little it moves.

Euro-area banks’ claims on NBFIs — loans plus debt securities, with reverse repo inside the loans line — have run at 6–7% of total bank assets since the counterpart data begin in 2014, and stand at 6.6% now. Funding taken from NBFIs, mostly deposits, is slightly larger at 6.9%. Together the two directions sum to roughly €5.8tn. To be clear about the accounting: that is a gross sum of claims and funding, a measure of two-way balance-sheet linkage, not of net credit exposure or loss-absorbing capacity. The point it establishes is scale and permanence. This is standing infrastructure, not a boom.

A flat pipe and a changing risk association are not a contradiction, and it is worth saying why. The BIS result concerns what exposures transmit, not how large they are: co-movement conditional on a given exposure can strengthen while the exposure itself is unchanged — through the shift out of the zero-rate regime, through the volatility of the collateral moving across it, or through changes in who sits inside the aggregate and how levered they are. Whether the composition inside the pipe shifted is precisely what the aggregate cannot show. The euro-area statistics capture euro-area counterparts only; the counterparty, collateral and cross-border composition of the linkage is not published.

Direct evidence on the unobserved euro-area leg is limited, and what can be said is comparative rather than conclusive. Where the leveraged non-bank presence can be observed, it is rising: the BIS’s Annual Economic Report documents hedge funds’ US sovereign exposures more than doubling since 2022 and their share of euro-area electronic government-bond trading increasing markedly — a turnover measure, not a leverage one. The Bank of England’s July Financial Stability Report finds the leverage concentrated in “a small number of hedge funds pursuing similar trading strategies across jurisdictions”. The relevant mechanism is nevertheless not uniquely American: the ESRB’s EU Non-bank Financial Intermediation Risk Monitor identifies leverage, liquidity mismatch and interconnectedness as material vulnerabilities in EU non-bank intermediation. Whether, and how far, those vulnerabilities sit in the French sovereign-holder chain is precisely what published data cannot determine.

The FSB’s February report on government bond-backed repo markets supplies the mechanism in general form: repo is how non-banks lever sovereign positions, how that leverage depends on short-term funding, and how it interconnects with the dealer banks that provide it. Which returns us to the question the holder data leave open: who, concretely, is on the other side of France’s roughly $2.6tn of foreign-reported sovereign holdings?

Who the foreigners are

The honest answer comes in two layers, from two datasets that do not quite meet.

The first layer is historical and granular. The Fang–Hardy–Lewis dataset — built at the BIS, the successor to the Arslanalp–Tsuda investor-base series the nexus paper itself relies on — decomposes France’s foreign holders into official institutions, banks and private non-banks through 2018. In it, non-banks became the largest foreign holder class around 2012 and reached 56% of foreign holdings by 2015, roughly double the official share and five times the banks’; the published version in the Review of Financial Studies adds that within euro-area private non-banks, investment funds dominate, not insurers or pension funds. The second layer is current but coarser. The IMF’s CPIS lets us split France’s foreign-reported holdings — about $2.6tn at mid-2025 — into official reserve holdings and everything else; “private” here means simply non-official, and includes foreign banks. The official share is around 14% and falling; the private share has risen from 82% to 86% since 2017. The creditor geography offers one further cut, to be read with its limitation first: about a fifth of foreign-reported holdings is reported from Luxembourg and Ireland. That is consistent with the importance of fund domicile and custody chains, but CPIS cannot identify the ultimate investor or establish a fund-sector share. The data establish, in order: that banks stepped back; that domestic non-banks did not replace them; that France’s foreign base is overwhelmingly private rather than official, with classification caveats; that within private, non-banks were the dominant class as far as the sector detail runs, which is 2018. What remains a hypothesis is the final step — that this foreign non-bank money is the channel through which the next episode of sovereign stress transmits.¹

The 2018 break is substantive. Current publications confirm that non-bank intermediation matters in sovereign markets, but they do not restore a current France-specific breakdown of foreign creditors into banks, funds, insurers and other private investors. And the breakdown is what stress analysis would need: an insurer, an index fund, a levered relative-value fund and a foreign bank treasury hold the same bond with entirely different redemption structures, leverage, repo dependence, hedging behaviour and dealer relationships. Knowing that the foreign base is “private” tells an authority almost nothing about how it behaves when spreads gap.

The composition still matters for a further reason. In the Fang–Hardy–Lewis sample, private non-banks absorb a disproportionate share of increases in sovereign debt — roughly 70% — and the strongest yield-responsiveness result concerns emerging-market sovereigns, with elasticities much lower across investor groups in advanced economies. That does not identify the current marginal buyer of French debt, nor does it establish a French stress-elasticity. It does show why the distinction between official, bank and non-bank foreign demand is not merely descriptive — and why losing the sector detail after 2018 is consequential.

The toolkit and the join problem

The first geography is comparatively legible in existing sovereign-exposure, bank-disclosure and home-bias metrics — the instruments exist because the euro area crisis taught us where to point them. The cross-border chain is only partly observable as a connected system, and not for want of attention; the FSB built a task force for it. The specific gap is this: no authority holds an integrated, instrument-level map linking sovereign holders, their leverage and liquidity terms, and the bank counterparties that finance them across borders. The ECB’s securities statistics do not sectorise non-resident holders. AIFMD reaches EU-domiciled funds but does not join their reporting to holdings data. The marginal holder of French debt may file with the SEC, the FCA, or no one — and the datasets that each capture one link in the chain do not connect.

The holder geography and the public backstop are assessed through different institutional frameworks: one concerns the structure of private intermediation, the other a separate monetary-policy assessment. The public data do not connect the two.

On policy, restraint is warranted, starting with what the paper does and does not do. It does not calibrate instruments or identify a uniquely European response; it changes the object that any such response must be able to observe. It strengthens the case for treating sovereign-market intermediation as a bank–NBFI resilience issue rather than solely a bank-sovereign concentration problem. The ESRB has already located part of the live policy debate in exactly these markets: in its response to the Commission’s consultation on macroprudential policy for non-bank financial intermediation, it wrote that the Commission “should consider introducing margin requirements in bilaterally cleared government bond cash and repo transactions and ways to facilitate the central clearing of such transactions”. The relevance here is not that the BIS paper proves those instruments right; it is that the paper changes the intermediation chain such instruments would need to address — and, one might add, that minimum haircuts, central clearing and NBFI leverage limits are also the closing policy list of the paper itself. Its final sentence, though, belongs to the fiscal side: none of this substitutes for sustainable fiscal trajectories.

The unfinished map

What this piece has traced is not a second geography fully mapped, but the boundary at which the published map breaks: cross-border holders, financing relationships and collateral chains that existing datasets do not join. The coastline is measured — a large foreign-reported stock of French debt, mostly private; a large, persistent balance-sheet linkage between euro-area banks and NBFIs. The interior — who within it is levered, against what collateral, financed by which dealer, subject to which liquidity terms — is unmapped at the instrument level: a join problem, not an absolute absence. Aggregate holder shares and aggregate bank–NBFI links cannot identify the transmission chain an authority would need to see during a sovereign-market stress event.

Three things to watch. The FSB task force’s report, whenever it lands, will show how much of the interior the official sector believes it can map. The French budget round this autumn will be a key event for whether July’s repricing persists. And the arithmetic of quantitative tightening against record issuance will keep making the identity, financing and liquidity terms of the marginal buyer more important — and still only partly observable. Mapping that chain is the unfinished task.

¹ A note on definitions. Holder residence follows the issuer’s perspective: “foreign” means non-resident of the issuing country, which includes other EU domiciles — a Luxembourg fund is foreign to France and inside the EU’s regulatory perimeter. “Official” in Fang–Hardy–Lewis is estimated from reserve-allocation data; in CPIS it is the voluntary SEFER survey, which undercounts reserve managers — hence the level gap where the two series overlap. CPIS “private” is everything non-official, including foreign banks; the bank/non-bank split exists only in Fang–Hardy–Lewis, to 2018. CPIS records holdings by the economy of the reporting holder, so custody and fund domicile shape the geography. Fang–Hardy–Lewis is face value, annual; CPIS is market value in dollars, semiannual.

Paweł Fiedor - The Macro Prudential View

Positions described are as at 7 August 2026.

The author works at the Central Bank of Ireland and in the ESRB Secretariat. This piece is written in a personal capacity; views are the author’s own and should not be attributed to the Central Bank of Ireland, the ESRB, or the Eurosystem. It draws exclusively on publicly available sources and takes no position on whether the policy measures discussed should be adopted.

Sources: Avdjiev, Hardy and Jager, “The evolving nexus: sovereigns, banks and NBFIs”, BIS Working Papers No 1369, July 2026 — https://www.bis.org/publ/work1369.htm; Fang, Hardy and Lewis, “Who holds sovereign debt and why it matters”, BIS Working Papers No 1099 (2023) and Review of Financial Studies 38(8) (2025), with the accompanying dataset — https://www.bis.org/publ/work1099.htm; Arslanalp and Tsuda, “Tracking global demand for advanced economy sovereign debt”, IMF Economic Review 62 (2014); FSB, “Vulnerabilities in Government Bond-backed Repo Markets”, 4 February 2026, and “FSB Workplan to Address Nonbank Data Challenges”, July 2025 — https://www.fsb.org; Bank of England, Financial Stability Report and FPC Record, July 2026 — https://www.bankofengland.co.uk; BIS, Annual Economic Report 2026, Chapter II — https://www.bis.org; ESRB, response to the European Commission targeted consultation on macroprudential policies for non-bank financial intermediation (”A system-wide approach to macroprudential policy”, November 2024), and EU Non-bank Financial Intermediation Risk Monitor — https://www.esrb.europa.eu; ECB Quarterly Sector Accounts and Balance Sheet Items statistics, ECB Data Portal — https://data.ecb.europa.eu; IMF Coordinated Portfolio Investment Survey — https://data.imf.org; Agence France Trésor, indicative State financing programme for 2026, 30 December 2025 — https://www.aft.gouv.fr; INSEE, national accounts first estimate, March 2026; European Commission, Spring 2026 Economic Forecast (France); ten-year government bond yields via investing.com. Chart data as cited in each figure.

At the Fund’s spring briefing in April, Tobias Adrian described the job with unusual economy. The task for policymakers, the IMF’s financial counsellor said, is “not in and of itself to predict shocks”; it is to make sure the system’s vulnerabilities are understood and contained, and that something can be done if instability arrives.

This is a coherent doctrine, and for the most part a correct one. No one can forecast the next geopolitical rupture, and an institution that tried would mostly embarrass itself. Buffering the vulnerability rather than divining the shock is a reasonable division of labour, and it is the one the whole financial-stability apparatus is built around.

But it draws a line, and something important sits on the far side of it. A doctrine organised around containing vulnerabilities treats the shock as exogenous — as weather that arrives, of uncertain timing and force, against which you hold capital and liquidity. That is the right posture when the shock really is weather. It is an incomplete one when the shock is a move: chosen and timed against the places where you are thinnest. Buffering a vulnerability and anticipating a move are not the same exercise, and the reviews, at least in what they publish, do almost none of the second.

Consider the cleanest recent example. Recently the Riksbank published a careful staff memo, “The transmission of geopolitical shocks to bank lending in Sweden,” by Cristina Cella. Using loan-level data, it shows that after Russia’s 2022 invasion Swedish banks expanded credit overall while pulling back from energy-intensive and financially weaker firms, with the banks most exposed to geopolitical risk tightening hardest. It is good empirical work, and it is entirely representative. It reaches for the Caldara–Iacoviello geopolitical-risk index, names “geopolitical” dozens of times, and uses not one term from the vocabulary strategic studies has built to reason about exactly this kind of risk. The title states the frame: the invasion is a shock, and the task is to trace how it transmits.

That pattern — reaching for salience and for transmission while leaving strategy untouched — is not Cella’s choice. It is the genre’s. In the core geopolitical sections of the flagship reviews, geopolitical risk is measured along two axes the field owns comfortably: how loud it is, and what it does once it lands. The third axis — what the actor wants, whether the threat is credible, where the escalation runs — is the one the reviews barely enter. This piece is about that blank cell, and about what reading only the populated ones costs a supervisor.

Three cells, and an empty one

Picture the problem as a two-by-two.

One axis is how the risk is treated: as an exogenous shock that arrives, or as an endogenous move inside a system the analyst is part of. The other is what the analysis captures: the realised outcome, after the fact, or the intention behind it, before.

Three of the four cells are populated, to varying depth. The bottom-left — exogenous shock, realised outcome — is home ground: the salience indices, the news-intensity measures, the stress tests, the buffers sized to historical loss. This is where the reviews are strongest. The top-left — intention, but still treated as something handed in from outside — is thinner: severity read off an external index and dropped into a model as a given. The bottom-right — endogenous, but again only the outcome — is where the field has genuinely moved in the past three years: fragmentation chapters that record the world’s financial linkages splitting along geopolitical lines. That cell is no longer empty, and it matters that it isn’t.

The empty one is the top-right: endogenous and intentional. The move, anticipated rather than recorded. This is the cell strategic studies was built to occupy — Schelling on deterrence and bargaining, Fearon on when bargaining breaks down — and it is the one the reviews do not enter.

A word on scope, because the claim is narrow and the narrowness is the point. I mean the core geopolitical sections of the flagship reviews: the GFSR chapters, the FSR special features, the staff memos that translate them. Pockets of real strategic reasoning exist elsewhere in the official sector — sanctions-impact units, cyber-resilience scenarios, bespoke work that never reaches the flagship. The gap is not that no one anywhere thinks strategically. It is that strategic interaction is not built into the buffer and scenario logic at the same level, with the same machinery, that salience already is.

That the empty cell belongs to the genre and not to one author is visible in the prose itself.

Across twenty-six consecutive editions of the ECB FSR and IMF GFSR since 2020, the vocabulary of strategic studies — deterrence, coercion, bargaining, brinkmanship, commitment problem — is exactly zero in twenty-three of the twenty-six editions, and never rises above a single stray mention in the other three, even as “geopolitical” climbs edition over edition. The starkest cases are the dedicated ones: a single GFSR fragmentation chapter used “geopolitical” nearly three hundred times and the strategic vocabulary not once.

A word count proves absent words, not absent thinking, and I do not want to overclaim from it. So take the strongest of those chapters at its best. When the GFSR maps the fragmentation of cross-border finance, it measures the thing as an outcome to be monitored — a drift in the data, plates sliding. Read with the empty cell in view, the same numbers are something else: a deliberate severing of financial infrastructure, with a timing and an asymmetry that a drift cannot have. The measurement is unchanged. The question it answers moves — from “how far has it drifted” to “who moved it, when, and whether they will move it again.” I will come back to that chapter with the mechanics in hand; for now it is enough that the reframing is available and changes the read.

The obvious objection is that the blank cell is deliberate, and proper. Assessing an adversary’s intent, credibility and escalation ladder is the work of intelligence and defence ministries; an apolitical financial-stability institution should not be naming aggressors and forecasting their moves, and its restraint is a feature. I think that is half right, and the right half is worth conceding cleanly: naming the actor and adjudicating intent is genuinely out of lane.

But the core outputs the reviews produce — buffer adequacy and its timing, adverse-scenario design, the identification of wrong-way risk in backstops and facilities — already require conditional reasoning about how strategic interaction moves exposures. A supervisor running a liquidity stress on a bank with heavy correspondent or custody exposure to the rival bloc is already obliged to ask what a targeted move against that specific node would do to the unwind path, and whether the backstop reaches the severed side. It is the same second-round exposure analysis the reviews already perform for every other risk, run conditionally.

The only difference from a conventional stress test is where the severity comes from. A standard exercise assumes, say, a twenty-per-cent disruption to cross-border clearing as a parameter. The strategic version asks instead which single node in the clearing chain is the point of failure, and assumes a counterparty with the means to close it picks the worst funding window to do so. The logic of a targeted move replaces the logic of a random one.

What keeps that move strategic rather than merely operational is that its feasibility is conditional: a node is exposed only where control over it sits on the far side of a possible fracture, which is read off the same bloc-alignment taxonomy the fragmentation chapters already use. Mapping which nodes are attackable in that sense, and whether the backstop reaches them, requires no forecast of the move and no name for the actor — only the exposure logic the reviews already apply to every other risk.

Declining to name the aggressor is a boundary, and a proper one. Treating the shock as if intention were irrelevant to how it transmits is not a boundary; it is a modelling assumption — an exogeneity built into the published frame, and one the post-2022 record increasingly contradicts.

What the salience index measures

Start with the measure the reviews actually cite.

The Caldara–Iacoviello geopolitical-risk index — the GPR, the series that appears across the fragmentation chapters and the staff memos — and the ECB’s euro-area systemic-stress indicator, the CISS, barely move together: the contemporaneous correlation across 2014–2026 is 0.05.

It would be cheap to read that as an indictment, and it isn’t one, so let me grant the honest reading first. A near-zero correlation is close to what the official sector’s own account predicts. Most geopolitical events do not become systemic stress, because facilities and buffers absorb them; the reviews would say, correctly, that the low correlation is a sign the machinery works. The chart is no embarrassment to macroprudential policy.

But granting that closes one door and opens another. Buffers absorbing the realised salience of past events tell you nothing about how the system is positioned against a move timed against a specific chokepoint — which is the case the rest of this piece is about. The low correlation does not vindicate the framework; it relocates the question. And it leaves the narrower point standing: salience and systemic stress are different objects. The GPR is an excellent gauge of how much attention geopolitics is drawing. It is not a measure of where the structural vulnerability sits. The reviews rely on it because it is the available, citable series for geopolitical intensity, in a way that a direct measure of strategic exposure is not — but intensity is not structure, and a model that regresses on the one is not capturing the other.

There is one more reason the blended index misleads, and it points straight at the next section. The GPR is a single number that folds together two very different things: the phase in which a threat is made, and the phase in which it is carried out. Averaged into one series, the attention measure obscures the one window that might actually be informative.

The phase that would lead

Split the index into its parts.

Caldara and Iacoviello publish the GPR in two components, a threats sub-index and an acts sub-index, and they behave like two regimes. Threats rise in the bargaining and brinkmanship phase, when something is being signalled and not yet done. Acts rise on realisation, when it is. The build-up to February 2022 shows up in threats; the Paris attacks of 2015 show up in acts, with no threat phase at all.

If a forward-looking signal lived anywhere in this series, it would live in the threat phase — the bargaining moment is, in Schelling’s account, where intentions are revealed before they are executed. It is worth checking whether it does.

It does not, or not in a way that survives a robustness check. Correlate each sub-index with euro-area CISS across horizons of nought to six months and a threat-phase lead does appear in the full sample, climbing to roughly 0.23 to 0.28 at three to six months while the acts series stays flat. But it is one episode. Drop the 2022 pairs and the threat lead does not merely weaken — it turns slightly negative, the acts series sits near zero, and neither phase shows a stable lead. On rank correlation, which is not hostage to a handful of joint-high months, the threat-phase lead over the full sample is indistinguishable from zero. The apparent channel is 2022 doing the work: a threat phase in January and February that happened to precede the energy-and-rate stress of that year.

That failure has a mechanical cause, and the cause matters. The GPR is built from newspaper text; it measures how much the press writes about geopolitical risk. Financial brinkmanship does not read like that. Quietly probing a clearing system, or shifting custody out of a jurisdiction before it can be frozen, generates exposure rather than headlines. 2022 registers because that coercion was unusually physical and public — armies on a border — while the financial moves that matter most are the ones the press cannot see. So the chart does not show that the strategic window is imaginary; it shows that a media-intensity index is the wrong instrument for finding it. The strategic literature gives good reason to expect a bargaining-phase window; this series cannot identify it, and no supervisor should expect it to.

That leaves the cost where it actually falls. A two-to-six-month strategic window is far too short to move a structural or countercyclical capital buffer, whose governance and phase-in run to several quarters; no one should be setting capital off it, and a framework that declined to would be right. The window belongs to the fast, discretionary instruments — liquidity readiness, a review of collateral eligibility, swap-line and operational preparation against specific chokepoints. A framework that reads only realised salience keeps the slow tool in view and tends to leave the fast one un-cued, and the fast one is the only one a timed move would test. The cue for it cannot come from a public index either; it has to come from exposure mapping and supervisory judgment, which is the structural question the rest of this piece is about.

The network is the weapon

The deeper claim has nothing to do with timing. It is about what kind of thing the system is. In the exogenous picture, the financial system is the thing a shock hits. In the strategic picture, the financial system is the terrain the contest is fought on, and its chokepoints — the points where flows concentrate and can be cut — are the weapons. Henry Farrell and Abraham Newman named this weaponised interdependence: the same network that makes the system efficient hands whoever controls its hubs a coercive instrument. Once that is true, a “shock” is not weather. It is a move, made by someone, against a node they chose because you depend on it.

The banking book is already re-routing around exactly this logic, slowly.

On BIS locational data, the share of global cross-border bank claims on the China- and Russia-aligned bloc has roughly halved since 2014, from about four per cent to under two. The China component is a gradual, decade-long decoupling. The Russia component is something else: a near-vertical collapse after 2022, from a small share to almost nothing, as the system was deliberately cut off. No non-aligned bloc has absorbed the difference in cross-border bank claims — the connector share is flat, though friend-shoring would surface sooner in trade and direct investment than in claims this sticky. The de-linking is asymmetric: the Western-and-offshore core holds at some ninety per cent of the book throughout, and the bloc construction here is a first-pass one, with Hong Kong and Macao kept as financial centres rather than folded into China. But the Russia line is the cleanest recent demonstration of weaponised interdependence on offer: a chokepoint, identified and closed.

One caution about which way it points. The Russia line shows the West using its own network centrality against an adversary; it does not show the West on the receiving end. For a European supervisor the defensive question is therefore the narrower one: which nodes run the other way, where both the dependence and the control over it sit on the far side of a possible fracture, including dependence on infrastructure controlled inside the Western bloc itself. The exhibit proves the mechanism is live; it does not locate your own exposure for you.

The reason this matters for supervision is that it converts into observables. Farrell and Newman’s argument, put into the language a risk authority already speaks, says to track a particular set of exposures: funding-currency concentration; custody and jurisdiction concentration; dependence on payment and market infrastructure — correspondent banking, and the central counterparties and securities depositories that sit in sanctioning jurisdictions; and backstop eligibility conditional on bloc alignment. None of these requires naming an adversary. All of them are balance-sheet or plumbing facts a supervisor can map today. And the banking book is only the baseline — capitalised, supervised, slow to fragment; the same concentrations compound in the non-bank sector, where leverage is opaquer and the backstop slower, which is where they bite first.

The point is not the list but the rule that orders it, and the rule is what a review would act on. Rank these exposures by where they compound: flag the positions at which collateral location, legal jurisdiction and payment access could fracture together, because that is the configuration in which standard liquidity support fails — the asset is sound, but it sits where the home backstop cannot reach it at the moment it is needed. Put formally: map the concentrations at which a severance is both feasible for the counterparty and non-offsettable by the home backstop within the bargaining window in which it would be used.

The sharpest case is the interaction of two of these exposures. A reserve asset held in a custody node on the wrong side of a sudden fracture loses its eligibility as central-bank collateral at the precise moment a liquidity run makes that eligibility matter most. For an official holder the Fed’s FIMA repo facility now reaches part of this — Treasuries custodied at the New York Fed can be turned into dollars without a fire sale — which is precisely why the binding exposure is the one the facility does not cover: the commercial bank, and above all the non-bank, holding the same asset in the same place with no such line. Custody-jurisdiction concentration and backstop eligibility are each tolerable alone; together, and outside the reach of a standing facility, they are wrong-way risk of the purest kind. Fearon supplies the timing — the dangerous windows are the commitment-problem moments, when a bargain that cannot be credibly held breaks down, which is as good a one-line description of February 2022 as any. Kindleberger supplies the question that closes the loop: does the backstop span the fracture? The Fed’s swap lines are a Western-bloc good. A backstop that reaches only your own side of a severed network is itself a wrong-way property.

So return to that GFSR fragmentation chapter, now with the mechanics in hand. Measured in the bottom-right cell, it is a fact about drift: the share of claims on one bloc has fallen. Measured in the top-right cell the chapter does not occupy, the same fall is a deliberate move: a chokepoint someone closed, in a particular month, with a backstop that did not reach the far side. The data point is identical. The supervisory question it answers is not, and only one of the two tells you where to look next.

The policy world is, in places, already ahead of the analysis. The EU’s Anti-Coercion Instrument exists precisely to raise the cost of economic coercion — which is an institutional acknowledgement that the strategic layer is real and legible. The financial-stability analogue would be to map the chokepoint exposures and ask the instrument’s own question in reverse: given our specific dependencies, what is the least-cost move against this particular node?

There is a reason this work belongs in supervision rather than in the flagship reviews. Some of the empty cell is not an analytical gap at all. A review answerable to a multilateral board has understandable reasons not to print a liquidity scenario conditioned on a named member severing infrastructure, and the restraint on the publication is neither unreasonable nor likely to lift. But a constraint on what can be printed is not a constraint on what can be examined, and the examination has an obvious home: supervisory scenario design and liquidity-preparedness review, where a conditional path can be run against the exposures without a flagship’s need to name anyone. That is the institutional home the argument is asking for. Collateral policy and operational contingency mapping follow from it; they do not need a separate venue.

What it costs

Put the whole thing in the form a risk officer will recognise. A liquidity or capital stress test today dials geopolitical severity as an exogenous parameter: assume conditions worsen by some amount, and check that the buffers hold. The unmeasured quadrant asks a different question. Given our specific exposures to dollar clearing, reserve custody, and correspondent banking with the rival bloc, what is an adversary’s least-cost way to force a disorderly unwind — and in which bargaining window would they choose to execute it? The first question is about how bad the weather can get. The second is about what an opponent does when they can see where you are thin. They produce different stress tests, and only the second is conditioned on the thing that has actually changed about the world.

The gap bites hardest where the backstop is slowest. In the non-bank system, funds and insurers carry liquidity promises on top of the same chokepointed infrastructure, but the central bank’s reach there is the most discretionary and the least automatic — so a strategic move would pick a funding or collateral node precisely because no standing facility covers it, landing as a margin call or a collateral freeze before a discretionary backstop could be authorised.

None of this is a case for central banks to become strategists, or a correction of any particular report. Adrian’s doctrine holds: the job is to contain vulnerabilities, not to predict shocks. The argument is only that one of those vulnerabilities is now the system’s own exposure to a deliberate move against its chokepoints, and that this vulnerability lives in the one cell the reviews do not yet measure. The shock itself cannot be predicted, and the reviews are right not to try. The board it would be played on — its chokepoints, and the backstops that may not span them — can be mapped. That is the cell they still leave unmeasured.

Paweł Fiedor — The Macro Prudential View

The views expressed are the author’s own and do not necessarily reflect those of any institution with which the author is or has been affiliated.

Sources: Tobias Adrian’s remarks, IMF Global Financial Stability Report press briefing, Spring Meetings, 14 April 2026 — https://www.imf.org; Cristina Cella, “The transmission of geopolitical shocks to bank lending in Sweden: Evidence from Russia’s full-scale invasion of Ukraine,” Sveriges Riksbank Staff Memo, June 2026 — https://www.riksbank.se; geopolitical-salience and strategic-studies vocabulary counts across twenty-six editions of the ECB Financial Stability Review and IMF Global Financial Stability Report, 2020–2026, own text analysis; Geopolitical Risk (GPR) index and its threats and acts sub-indices, Caldara and Iacoviello — https://www.matteoiacoviello.com/gpr.htm; euro-area New Composite Indicator of Systemic Stress (CISS), European Central Bank Data Portal (series CISS.D.U2.Z0Z.4F.EC.SS_CIN.IDX) — https://data.ecb.europa.eu; cross-border bank claims by counterparty, Bank for International Settlements Locational Banking Statistics — https://data.bis.org; illustrative bloc construction based on UN General Assembly voting alignment, following Bailey, Strezhnev and Voeten (2017) and the IMF’s geoeconomic-fragmentation analysis (GFSR, April 2023; Aiyar et al., 2023), with UN Resolution ES-11/1 (March 2022) as a robustness check; theoretical anchors: Thomas Schelling, The Strategy of Conflict and Arms and Influence; James Fearon, “Rationalist Explanations for War” (1995); Charles Kindleberger on the lender of last resort; Henry Farrell and Abraham Newman, “Weaponized Interdependence” (2019); EU Anti-Coercion Instrument, Regulation (EU) 2023/2675 — https://eur-lex.europa.eu. Chart data as cited in each figure.

On 4 March 2025, Germany’s incoming coalition announced that it would loosen the constitutional debt brake and stand up a large infrastructure and defence fund. The bond market did not wait for the legislation. Within the week the ten-year Bund yield rose by around 43 basis points, the sharpest weekly move since the 1990s, and French and Italian yields moved almost exactly in step. Spreads between the member states held roughly steady, so the repricing landed on the common benchmark rather than on the gaps between sovereigns: a parallel lift in the level of the curve that anchors euro-area pricing, set off by a fiscal decision taken in one capital and transmitted across the bloc in days. The Bundestag and Bundesrat ratified the change a fortnight later, on 18 and 21 March. By then the market had long since moved.

The episode is a clean window onto a fact that is easy to state and easy to lose sight of: the conditions macroprudential policy is asked to manage are set, to a first approximation, by others. Sovereign supply, the path of policy rates, the term premium, the willingness of a central bank to stand behind a market in stress — that is the terrain. Macroprudential policy operates on top of it. It does not grade it.

The residual claimant

It is worth being precise about that subordination, because the strong version of the claim is wrong and the defensible version is narrower.

Macroprudential policy operates downstream. Its mandate is the residual financial-stability risk that monetary and fiscal policy leave behind, and the influence runs both ways: bank buffers change loss-absorbing capacity, and a well-capitalised system gives monetary policy more room to ease in a downturn without fearing for stability. So the claim is not that the policy is powerless or merely passive.

The defensible claim is narrower. Macroprudential policy is the arm of stabilisation most exposed to conditions it does not set. It is also the least able to make its own countercyclical instrument credible before it is tested. Three constraints sit behind that, each demonstrable on its own. It does not set the monetary-fiscal mix that drives the cycle it has to contain; the March repricing is one view of that. It does not control the public backstop the system leans on when stress crystallises, which I come to below. And it struggles to make its ex-ante countercyclical buffer bind through anticipation. The first two constraints are structural and largely uncontested. The third is where the argument has to do work, because it is the one place where the policy is acting on an instrument it genuinely owns.

A jurisdictional edge runs through all of this. The core toolkit — bank capital buffers — covers the regulated banking core. A growing share of the risk it is meant to contain has moved outside that perimeter. That is why the third constraint matters as much as it does, and it is where the rest of the piece begins.

Even its own tool sat idle

Start with the revealed preference, which is more informative than any account of what macroprudential authorities can in principle do. The countercyclical capital buffer is the one instrument built to lean directly against the financial cycle: to rise in the expansion, accumulating releasable capital, and to be cut in the downturn. Across the euro area, through the last expansion, it did neither.

Most members held the buffer at zero through the upswing; as late as 2019, only four euro-area countries had set a positive rate at all, and the handful that had built anything released it in 2020. There was little to release. The European Central Bank’s own post-pandemic review is blunt about the arithmetic: of the capital relief the banking system drew on, only a small slice came from releasable macroprudential buffers, because so little had been accumulated beforehand. Authorities, in its phrase, had “little ammunition.”

The reasons were layered. The standardised guide meant to flag the build-up phase — the credit-to-GDP gap — rarely flashed in a decade of low growth and low rates, and raising capital requirements into an expansion is in any case costly to announce and easy to defer. The point is not that supervisors were negligent; it is that the cyclical instrument, as designed and as governed, did not build the releasable resilience it was meant to. The move from 2023 toward a “positive neutral” buffer — a positive rate set in normal times rather than waiting for a gap to open — is best read as the system’s own recognition of that limitation.

That dormancy matters more than a quiet decade suggests, because the risk the buffer was built to cushion was in the meantime migrating outside the perimeter the buffer can reach, while private intermediation capacity was being capped on purpose.

The sector outgrew the core

The migration is not subtle. Non-bank financial intermediation — investment funds, money-market funds, insurers and pension vehicles, the broad universe the Financial Stability Board tracks — has grown into the larger half of the euro-area financial system.

On the FSB’s measure, euro-area non-bank assets crossed above the banking sector around 2013 and now stand at roughly €59 trillion against the banks’ €38 trillion, about one and a half times the regulated core, up from barely half its size two decades ago. The non-bank share of euro-area financial assets has risen from around 15 to 22 per cent over the period.

A buffer binds on the entities it is written for. The countercyclical buffer sits on banks; it does not sit on the open-ended fund running a liquidity mismatch, the insurer reaching for yield, or the leveraged fund crowded into the same trades as its peers. As intermediation has moved beyond the banking perimeter, the buffer’s direct reach has narrowed to a shrinking share of the system. That reach is not nil: banks still fund, warehouse for, lend to, clear for, and run prime brokerage against the non-bank sector, so a capital tool on the core still bites the perimeter indirectly, through the price and availability of bank balance sheet. But indirect and incomplete reach is a weaker thing than a buffer that sits where the transformation happens — and it is one reason the credibility of the bank buffer still matters even after the risk has moved, a point the mechanism section returns to.

Capacity capped by design

The other half of the migration story is who absorbs the stress when it arrives. Here the cleanest evidence is American, and I want to be careful about what it does and does not establish.

Indexed to the pre-crisis peak, the US Treasury coupon market is around six times larger than it was in 2007, while the balance-sheet capacity of the dealers meant to intermediate it has, in aggregate, round-tripped to roughly where it began. Dealer capacity relative to the market it must absorb has fallen roughly sevenfold. The shape of that gap is not an accident or a loss of nerve. It is, in large part, the intended consequence of post-crisis bank regulation — the leverage ratio and the capital rules that made banks safer by making it more expensive for them to warehouse large inventories of even low-risk assets. The capacity was capped on purpose.

The chart is US data and the euro-area market is built differently, so this is an analogy for a general constraint rather than direct evidence for the bloc. The point worth importing is narrow: post-crisis balance-sheet rules have made private intermediation capacity more state-contingent relative to the size of the markets it clears. The United States leans on standalone and non-bank primary dealers; euro-area sovereign market-making runs overwhelmingly through banks, so the binding constraint here is the optimisation those bank-dealers perform under the Capital Requirements Regulation — the leverage ratio, risk-weighted-asset targets, and internal liquidity transfer pricing that together govern how much sovereign inventory a desk will carry and how fast it steps back when balance sheet turns expensive.

There is an awkwardness worth owning. The leverage ratio that caps that capacity is itself a prudential rule, set within the regulatory family macroprudential policy belongs to, so this stretch of the terrain is not external weather in the way the monetary-fiscal mix is. But it is a microprudential safety choice whose systemic side-effect — thinner market-making in stress — lands on the macroprudential authority without that authority having set the rule. The downstream pattern holds even here: macroprudential policy is left managing a constraint calibrated for other ends.

The public balance sheet fills the gap

When a market that has outgrown its private intermediation capacity comes under stress, something has to absorb the imbalance. Repeatedly, across jurisdictions, that something has been the public balance sheet.

The episodes are heterogeneous, and the chart is illustration rather than proof: three different tools, in three settings, scaled by GDP. In the September 2019 US repo dislocation the Federal Reserve’s balance sheet expanded by around 1.9 per cent of GDP through repo operations and bill purchases. In the March 2020 dash-for-cash it deployed Treasury purchases worth roughly seven per cent of GDP within weeks. In the 2022 gilt and liability-driven-investment episode the Bank of England bought £19.3 billion of gilts — under a facility framed at up to £65 billion — over thirteen days, and had unwound the position by the following January. Small in that last case, and decisive: the credibility of the backstop did more work than its volume.

One honest qualification. We observe the interventions, not the stresses that buffers quietly absorbed without one, so the claim is about the marginal and tail cases, not the whole distribution. And the backstop arrives because the central bank cannot credibly promise, in advance, that it will not. Once a dislocation threatens to become disorderly, intervening is usually the right thing to do after the fact, which is exactly what makes the promise not to intervene hard to believe beforehand. The effect on private incentives is at the margin: where support is anticipated in precisely the states where resilience would matter most, the incentive to hold costly resilience ex ante is weakened there. That is the more modest claim the record will bear. The FSB has framed its own non-bank agenda in compatible terms — strengthening resilience partly in order to reduce the system’s reliance on extraordinary central-bank intervention, which is the language of an authority working on a commitment problem from the inside.

The Eurosystem is not exempt from the structure. Its pandemic asset purchases included an explicit market-stabilisation phase, and the flexibility in how those holdings are reinvested is the form the same liquidity-backstop logic takes inside the monetary union. I am narrowing the comparison to market-function support deliberately, because the distinction carries weight: the public balance sheet reliably backstops liquidity in a dash-for-cash; it is not aimed at a parallel duration repricing of the sort that opened this piece. The March 2025 move was a level repricing rather than a liquidity event, and the liquidity backstop is not aimed at offsetting that kind of shift. The backstop covers one kind of tail well; it leaves macroprudential policy facing the rest.

The credibility discount

Return to the third constraint. If macroprudential policy owns the countercyclical buffer outright, why is it so hard to use?

Call it a credibility discount: the wedge between the cyclical force an instrument is meant to exert and what its setter can credibly commit to in advance, given how observable the instrument is. The property I want to isolate is observability. An instrument whose triggers rest on unobservable, model-based, judgement-laden inputs is hard to make bind through anticipation. Two things follow from that property — discretion makes the setter’s future action hard for anyone to predict, and model-dependence makes tightening contestable in real time — but both run through observability, and it is observability that is worth holding onto.

The fiscal side gives the cleanest foothold. Recent work by Cuesta Bartolomé and Larch finds that sovereign markets price compliance with the deficit and debt rules — nominal, observable, unambiguous in real time — but barely price compliance with the cyclically-adjusted, structural-balance rules that hang on an output-gap estimate. Markets discipline what they can verify. The observability condition carries to the macroprudential side; the pricing channel does not, because no market prices the countercyclical capital buffer. What enforces a macroprudential buffer is supervisory resolve and the anticipation of regulated agents, not a yield. The common factor across the two domains is observability, not the disciplining mechanism.

What does the discount do in practice? At the build-up, if banks expect a model-based buffer to be raised only rarely — the signal contestable, the act costly for the supervisor — they will not price an expected future buffer into capital planning during the expansion, so the instrument restrains little just when it is meant to restrain most. At the release, freed capital helps only if banks believe it will not be quietly clawed back once conditions normalise; doubt about that promise blunts the release too. The discount shows up as weak transmission at both ends of the cycle.

The move toward a positive-neutral buffer since 2023 is often read as the fix, and for part of the instrument it is. A static baseline rate, set in normal times, is observable and rule-like; to that extent it belongs with the credible anchors, and it is a sensible answer to the credit-gap signal’s failure. But the baseline is not the cyclical instrument. The countercyclical function — raising the buffer above the baseline as risk builds, releasing it in stress — remains the discretionary, model-dependent act, and that part stays where it was. Positive-neutral makes the floor credible. It leaves the active, countercyclical margin above the floor as exposed to the discount as before.

Borrower-based limits look like a counterexample, observable and seemingly countercyclical at once, but in practice supervisors rarely modulate them over the cycle, because tightening a loan-to-value cap mid-boom to keep first-time buyers out of the market is politically very hard. They function as static, nominal limits that bite harder in a boom mechanically, which places them alongside the leverage ratio among the credible anchors. They confirm the pattern rather than breaking it: the macroprudential tools that command credibility are the ones kept observable and out of cyclical hands.

Calibration, conditional

None of this argues against backstops, and it does not call for bigger buffers as such. Backstops are often the system working as intended, and a buffer earns nothing for size alone. The narrower point is about conditionality. Macroprudential calibration is set inside a regime it does not control — the mix that moves the terrain, the backstop that underwrites the liquidity tail — and the part of its own toolkit meant to lean against the cycle is the part hardest to make credible before it is tested. Positive-neutral has made the floor credible; the active margin above it is the gap that remains.

For the euro area the stakes are concrete. The non-bank sector that can, in a tail liquidity stress, end up leaning on the public balance sheet is now half again the size of the banking core. The fault line the public balance sheet defends runs through the sovereign-bank nexus that March’s repricing lit up. The European Central Bank and the European Systemic Risk Board have, in their joint work and in the spread of positive-neutral buffers across the bloc, already named the under-build and begun to act on it, and the diagnosis here is meant to sit alongside that work rather than second-guess it. The residual claimant does not get to set the terms of the regime it operates in. What it can work on is the credibility of the one instrument those terms still leave in its hands — which, on the evidence of how that instrument has been used, remains the harder half of the problem.

Paweł Fiedor — The Macro Prudential View

The views expressed are the author’s own and do not necessarily reflect those of any institution with which the author is or has been affiliated.

Sources: Ten-year government bond yields for Germany, France and Italy, Stooq, with the German Bund cross-checked against Deutsche Bundesbank (BBSSY_REN_EUR_A630) — https://stooq.com, https://www.bundesbank.de; euro-area non-bank and bank financial assets, Financial Stability Board, Global Monitoring Report on Non-Bank Financial Intermediation 2025 (to end-2024), converted to euro at the FSB’s published euro-area exchange rate — https://www.fsb.org; US broker-dealer total financial assets, Federal Reserve, Financial Accounts of the United States (Z.1), via FRED (BOGZ1FL664090005A), and marketable Treasury notes and bonds, US Treasury Monthly Statement of the Public Debt — https://fred.stlouisfed.org, https://fiscaldata.treasury.gov; Federal Reserve balance sheet (H.4.1), total assets and Treasury securities held outright, via FRED (WALCL, TREAST); US gross domestic product, Bureau of Economic Analysis via FRED (GDP); UK gross domestic product, Office for National Statistics via FRED (UKNGDP); Bank of England temporary gilt purchase operations, September–October 2022 — https://www.bankofengland.co.uk; applicable countercyclical capital buffer rates by member state, European Systemic Risk Board — https://www.esrb.europa.eu; composition of the pandemic capital release, European Central Bank, Macroprudential Bulletin, and European Central Bank and European Systemic Risk Board, joint analysis of positive neutral countercyclical capital buffer rates (2025) — https://www.ecb.europa.eu; fiscal-rule pricing evidence, Cuesta Bartolomé and Larch, VoxEU/CEPR (2026) — https://cepr.org. Chart data as cited in each figure.

The mandate

In December 2023 the US Securities and Exchange Commission adopted a rule that will push much of the US Treasury market — both cash and repo — into central clearing through the Fixed Income Clearing Corporation (FICC), the only clearing house that does this work. After an extension granted in February 2025, the compliance dates now fall at the end of 2026 for cash and 30 June 2027 for repo. Neither is fully settled: the Investment Company Institute has asked for repo to slip to 2028, and in April 2026 Acting Chairman Uyeda floated relief for inter-affiliate transactions. Treat the timeline as directional rather than fixed.

The direction is what matters. The mandate takes a market that is a little under half-cleared today and removes, for most of the rest, the option not to clear. The case for doing so is the standard post-2009 case for central clearing, and it is largely correct: multilateral netting shrinks gross exposures, a central default-management process replaces a scramble of bilateral close-outs, and the clearing house sees positions the bilateral market does not.

This piece is about what the framework does around that move — how it prices the resulting exposures, where it ranks them in a failure, and what it asks of their holders in a stress — and about a misalignment that sharpens as more of the market passes through one counterparty.

Two privileges

Cleared and collateralised claims enjoy two distinct forms of favourable treatment, worth keeping separate because they come from different parts of the rulebook and bind on different activities.

The first is capital, and which part of the capital framework bites depends on the trade. For cleared derivatives, the headline is the risk weight: a bank’s trade exposure to a qualifying central counterparty carries a 2% risk weight under Article 306(1)(a) of the EU’s Capital Requirements Regulation, mirroring the Basel standard (CRE54), against at least 20% for the bilateral equivalent — a reduction of more than 90% on the trade-exposure leg. For Treasury repo, the risk weight was never the binding constraint; repo against government collateral carries little of it to begin with. What binds there is the leverage ratio, which is risk-blind and therefore bites hardest on low-risk, balance-sheet-intensive business. It was amended in 2019 to recognise client-cleared collateral, lowering the leverage cost of clearing, and it is now being relaxed further (below). So the right reading is not “derivatives are capital-light”: the standardised approach to counterparty credit risk (SA-CCR), live in the EU since June 2021 and in the US and UK since January 2022, raised capital on many bilateral books. It is that cleared exposures are treated more cheaply relative to bilateral ones, on whichever constraint actually binds. The framework steers traffic toward the clearing house, on purpose.

It helps to separate the exposures this produces, because they are often run together: the trade exposure to the CCP (risk-weighted at 2% for a QCCP), the default-fund contribution (a mutualised, loss-absorbing claim with its own, much higher charge), the leverage exposure (the balance-sheet footprint), and the liquidity cost (the cash needed to keep the position margined). The first three are priced. The fourth, at the level that matters here, is not.

The second privilege is priority. In a dealer or bank failure, cleared and collateralised claims do not wait in the ordinary creditor queue. Repo and derivative counterparties can net their positions and enforce their collateral outside the automatic stay — the US Bankruptcy Code’s safe harbours, the EU’s Financial Collateral Directive — so they are satisfied largely outside the insolvency estate rather than ranked within it. The clearing house sits one step further out: FICC’s margin, held on a bankruptcy-remote basis and doubling as its default fund, is carved out even of the resolution stay. The US and EU have converged here. Both impose a short stay on close-out in resolution — until 5pm the next business day under the US regimes, until the end of the next business day under Article 71 of the EU’s Bank Recovery and Resolution Directive — and both carve CCP margin out of it. None of this is new, and the mandate does not create it. What the mandate does is enlarge the population of claims that enjoy it.

The capital treatment is sometimes called a subsidy. That is the wrong frame, and the one a critic would most enjoy taking apart. The low risk weight reflects a real reduction in counterparty credit risk: the clearing house interposes itself, margins the position daily, and mutualises the tail. On counterparty credit risk, the framework is pricing about right.

Nor is it true that the framework ignores liquidity. The Liquidity Coverage Ratio and Net Stable Funding Ratio price liquidity risk directly — banks hold high-quality liquid assets against potential repo and derivative outflows, and the LCR treats cleared repo comparatively favourably. What those rules price is each institution’s own liquidity position: its own outflows, its own buffer. They do not price the liquidity demand that daily margining imposes on the system when the same call arrives everywhere at once. The exposures the framework treats as safest to hold and most senior to be paid are also the ones that generate the largest correlated demand for cash, exactly when cash is dear — and that correlated demand sits outside what any of these instruments is sized against.

The channel is already filling

Two facts about the market the mandate is enlarging, and a word on who actually bears the call.

First, the denominators, because two of them are easy to confuse. As of the third quarter of 2025, central clearing accounts for about 45% of Treasury repo on the OFR’s measure — a little under half. But the mandate does not target all of the uncleared remainder: once the carve-outs are applied, the Treasury Borrowing Advisory Committee reckons around 58% of the volume actually subject to the mandate already clears. So: ~45% of the whole Treasury repo market, ~58% of the mandate-eligible subset. The slice the rule actually forces in is therefore at most the remaining ~42% of that eligible set, and less once the carve-outs apply — a real enlargement of the cleared population, but not the doubling that “half the market is uncleared” might suggest.

Second, the buy-side leg has been growing on its own economics, and mostly before the rule could have caused it. FICC’s sponsored service — the route through which hedge funds and money funds reach clearing — expanded from roughly $1.1 trillion of combined borrowing and lending at the end of 2023 to about $2.9 trillion at the end of 2025, more than 150% on the end-of-period measure. The steepest growth is concurrent with mandate preparation, but the trade driving it predates the mandate and would exist without it: the borrowing leg is leveraged investors funding the cash–futures basis trade, the lending leg is money market funds placing cash. The mandate is not what built this channel. What it does is make the cleared share near-universal, mandatory and permanent, and close the bilateral route that today still carries the other half.

The plumbing decides who is actually called in a stress, and it is also why dealers use the service. Two distinct benefits meet at the dealer, and they should not be run together. The first is an accounting benefit, available only because the trade is cleared: novating matching sponsored repo and reverse-repo legs to FICC lets a dealer net them on its balance sheet under US GAAP (ASC 210-20-45, the former FIN 41), which shrinks the reported leverage exposure of the activity. The second is the regulatory leverage relief discussed below, which is a separate change, applies more broadly than cleared repo, and operates on leverage capital rather than reported exposure. The accounting netting is what has driven the growth in the chart above — DTCC puts the balance-sheet capacity freed at well over a trillion dollars on peak days — not the regulatory relief alone. But netting the balance sheet is not the same as shedding the risk. The dealer-sponsor still guarantees the sponsored member’s performance to FICC, still posts FICC’s value-at-risk margin, and still carries a clearing-fund and contingent-liquidity obligation against the position. Clearing transforms the dealer’s exposure rather than retiring it, which makes the dealer-sponsor the load-bearing node: when margin is called, the binding questions are whether the hedge fund can fund it, whether the money fund will keep rolling, and whether the dealer can carry the position while it does.

The synchronised call

Margin is procyclical whoever holds it. Variation margin tracks mark-to-market moves; initial-margin models re-estimate risk upward as volatility rises; a single shock raises both at once. This is not special to clearing. Bilateral repo has variation margin, discretionary haircuts and collateral calls that tighten in stress too — and being discretionary, they can tighten harder and faster, with no anti-procyclicality requirement on them at all. A regulated CCP, by contrast, runs a transparent, rule-bound margin model subject to explicit anti-procyclicality rules. Per dollar of exposure, it is not obvious that cleared margin is more procyclical than the bilateral arrangement it replaces; it may well be less.

What clearing changes is not the timing of the calls — a severe shock synchronises bilateral calls too; panic moves every desk at once — but their character. A bilateral market applies many margin schedules, many haircut conventions, and discretion: a dealer can delay or soften a call to a valued client to keep it from dumping collateral into a falling market. A central counterparty applies one model and one parameter set to everyone, marked at fixed times, with no discretion to grant. It replaces a heterogeneous, partly negotiable set of calls with a single uniform, mechanical, non-negotiable one — and the mandate pushes the share of the market inside that one call toward near-universal. The distinctive feature is not the synchronisation of the clock; it is the uniformity of the model and the absence of forbearance at the moment forbearance is what would arrest a fire sale.

That cuts both ways. Discretionary forbearance is exactly what post-2008 reform set out to remove: the bilateral discretion that can soften a call is also the discretion that conceals losses, delays recognition, favours some clients over others, and amplifies the run when it finally comes. Mechanical uniformity strips out the stabilising forbearance and the loss-concealing kind together, and that uniformity was a deliberate regulatory objective, not an accident. The net effect on fire-sale risk is therefore ambiguous in theory. The concern here is narrower than “discretion is good”: it is that removing the release valve, while concentrating the population inside one model, removes a margin of adjustment exactly where the system has least slack.

The Treasury-specific evidence sits in March 2020. As the dash-for-cash took hold, leveraged funds running the cash–futures basis trade — the same trade now being routed into FICC — were forced to unwind, and their selling of Treasuries to meet margin and funding pressure fed the dysfunction in the world’s deepest sovereign market. That is the transmission this piece is about, observed before the mandate concentrated the trade any further. The cleared figures themselves are not, in calm times, large in proportion: FICC’s Government Securities Division held on the order of $78 billion of initial margin in late 2025 and runs about $12 billion of variation margin a day, peaking near $19 billion — small against roughly $4.4 trillion of cleared Treasury repo. The industry makes that point: the World Federation of Exchanges has argued that peak crisis margin calls were 2.5% or less of participants’ liquidity resources. On average, true. But averages are not where this risk lives.

The mechanism that takes it into the tail is specific to the basis trade, and worth stating precisely, because the obvious analogy understates it. The 2022 UK gilt episode is the existence proof: liability-driven investment funds faced margin and collateral calls Bank of England staff put in excess of £70 billion, concentrated on a specific, liquidity-poor set of holders, and those calls forced fire sales — against which the Bank stood up a backstop of up to £65 billion. But LDI was a directional trade, hurt by a parallel move in rates; its margin call was roughly linear in the size of the rate move. The Treasury basis trade is a relative-value trade — long the cash bond, short the future — and a parallel shift in the curve does not break it; a widening of the spread between cash and futures does. The cleared-Treasury failure mode runs through that spread: a mechanical margin call on the futures leg forces deleveraging of the cash leg, the forced selling widens the cash–futures basis, the wider basis raises the margin again, and the loop feeds itself. That is the negative-convexity tail the averages do not see, and it lands on the most leveraged holders in the market.

Put the increment honestly, because the argument does not need to overstate it. With ~58% of mandate-eligible repo already cleared, the volume the rule actually forces in is at most the remaining ~42% of that eligible set, less after carve-outs. The case does not rest on that number being large; it rests on where it lands — on the most leveraged segment of the market, intermediated through dealers whose leverage capacity is being freed at the same time, with the bilateral escape valve closed. How much additional same-day margin that marginal volume would generate under a given volatility shock is not something public data lets anyone estimate, and this piece does not pretend to a figure. It is a design argument about where the rule concentrates risk, not a forecast of the size of the next call.

This is the fallacy of composition, in the sense Mark Roe (2011) and Patrick Bolton and Martin Oehmke (2015) gave it for derivative and repo priority — but the clearing house changes the argument, and the change has to be conceded. Roe and Bolton-Oehmke were writing about bilateral super-priority in opaque markets, where the safe-harbour claim weakens each creditor’s incentive to monitor and accelerates a race to the collateral. A CCP mutualises default, runs a disciplined waterfall, and is transparent to its regulator; it dampens precisely those bilateral dynamics. Where the frame still bites is one level up: each participant’s decision to hold the senior, collateralised, capital-cheap, centrally cleared claim is individually rational and individually risk-reducing, and the aggregate of those decisions is a market whose demand for collateral and cash is uniform and synchronised by construction. Privately optimal; in aggregate, something else.

What the optimists are right about

Four counterarguments deserve to be met rather than waved past.

The first is that clearing reduces systemic risk on net. It does. Netting genuinely shrinks gross exposures; a disciplined default waterfall genuinely beats a bilateral scramble; and cleared margin is not obviously more procyclical per dollar than what it replaces. The claim here is narrower: that concentrating these exposures into one uniform call raises a liquidity demand the framework does not price, and that the bill falls due in exactly the states of the world where it is hardest to pay. The net sign — more netting and central default management against a more uniform, less forgiving liquidity call — is genuinely uncertain, and I am not claiming it is negative. I am claiming one side of it is unpriced.

The second is that the Federal Reserve has built the backstop. The Standing Repo Facility, made permanent in 2021, exists to cap repo rates in a dash-for-cash, and the FIMA repo facility extends the logic to foreign official holders. This is the strongest objection, and it is largely right. Lending cash against pristine collateral at an administered rate is textbook Bagehot, not a giveaway, and a credible backstop does blunt the procyclical spiral. The point that survives is about ex-ante incentives: a backstop that is standing and anticipated, rather than an emergency improvisation, is priced in by participants in advance, and that shifts behaviour toward the structure it protects. It also makes the central bank the residual liquidity provider for a structure the prudential framework treats as private and safe.

The third is that FICC already prices this. It runs a Capped Contingency Liquidity Facility — a rules-based, pre-committed facility under which each member’s cap is set by the liquidity its own activity generates, attested annually by two of its officers and carried in its liquidity plan. It is, very much, a line item, and a real one. But what the CCLF prices is the liquidity FICC needs to settle the obligations of a defaulting member — a single-name event — not the correlated, market-wide variation-margin call that lands on everyone at once with no one in default. And its own mechanism is a synchronised draw: in a CCLF Event the surviving members must repo cash to FICC up to their caps, supplying liquidity precisely when they are scrambling for their own. The one committed facility turns a member default into a mutualised liquidity demand on the survivors; it sits inside the synchronisation rather than outside it. FICC itself expects the mandate to raise these obligations — it told the SEC so when it added a $10 billion commercial-paper programme in 2026 to diversify its default liquidity — which makes the adequacy of the post-mandate sizing an open question, not a settled one.

The fourth is that the tools already exist more broadly. CCPs have recovery and resolution regimes; EMIR sets anti-procyclicality requirements for margin models; the BCBS-CPMI-IOSCO follow-up work in 2025 pushed on initial-margin responsiveness and liquidity preparedness. These are real, and also discretionary, partial, and largely untested at the scale the mandate implies. They applied to the pre-mandate world too; the mandate is what scales the population they have to work on.

Loosening the leverage ratio

In the same window in which it is mandating clearing, the US is loosening the leverage ratio — the capital constraint that bears most directly on balance-sheet-intensive Treasury activity. The enhanced supplementary leverage ratio was finalised in November 2025 and took effect on 1 April 2026 (banks could adopt it from 1 January); it is now in force.

The agencies’ case is not frivolous. The leverage ratio is risk-blind, and a risk-blind constraint bites hardest on the lowest-risk, most balance-sheet-intensive business — which is exactly Treasury intermediation and repo. The FDIC put the rationale plainly: the relief is meant to give banks more capacity “to engage in low-risk activities, such as U.S. Treasury market intermediation and repo financing.” And an unconstrained dealer balance sheet is also the shock absorber — the capacity that takes Treasuries off forced sellers when the call lands. There is a coherent and serious view in which loosening the eSLR is what a larger cleared market needs: more balance sheet to intermediate it, and more to absorb the fire sales the uniform call can trigger. I do not dismiss it.

What complicates that view is where the relief lands. The rule’s own dissenting governor made the point: about 28% — roughly $219 billion — of the tier-1 reduction falls at the GSIBs’ depository-institution subsidiaries, and only 1.4% (around $13 billion) at the holding-company level, while Treasury-market intermediation happens mainly at the broker-dealer. Barr was, in his words, “skeptical that it will achieve the stated objective of improving the resiliency of the Treasury market.” If the relief does not reach the intermediating broker-dealer, then the agencies’ stated purpose — more capacity to intermediate Treasuries and finance repo — is not met, while group-level leverage discipline is loosened anyway. And if it does reach the dealer, then the one quantity constraint on the leveraged build-up is being eased as the build-up is required. There is no comfortable reading: either the relief misses the entity it was sold to help, or it removes the brake on the very activity the mandate concentrates. Under both, a quantity constraint on leverage is loosened, and no liquidity-specific tool is tightened in its place.

Europe is contemplating the same move

This is not only a US story, and the European version has a twist worth sitting with.

The EU has no Treasury-repo clearing mandate. But its own macroprudential authority is pointing the same way. In November 2024 the European Systemic Risk Board published A system-wide approach to macroprudential policy, its response to the European Commission’s targeted consultation on macroprudential policy for non-bank financial intermediation. It singles out clearing as one of three priority activities — alongside asset management and lending — and, on government-bond cash and repo, says it “sees merit in incentivising the central clearing” of those markets, together with margin requirements on the bilaterally cleared remainder. The ESRB, in other words, sees merit in incentivising the migration the US is mandating. The same report reaches for the same frame this piece does: in a footnote it invokes the fallacy of composition as it works through the system-wide approach. The body most alert to the aggregation problem is also the one encouraging the activity that creates it.

Then nothing moved: the Commission published a summary of the consultation responses in March 2025, and there is no legislative package. So Europe sits where the US sits, with a clear direction of travel, a clear set of cautions, and no settled framework reconciling them.

The market structure differs in a way that changes the risk rather than removing it, and the difference runs on two axes that point in opposite directions. The US clears Treasury repo through a single counterparty: FICC is a monopoly, which means extreme concentration but one model and one resolution regime. Europe clears government-bond repo across several CCPs — LCH SA in Paris, Eurex Clearing in Frankfurt, Euronext Clearing in Milan, BME in Madrid, LCH Ltd in London — over heterogeneous sovereigns, with roughly half of euro repo already cleared. On the uniformity axis, that fragmentation cuts against the US problem: multiple CCPs running different models make the cleared euro market less perfectly uniform than a single-FICC market, which blunts the one-model-for-everyone effect. But on the netting axis it cuts the other way, and harder: positions held long at one CCP and short at another do not net, so a dealer pays gross margin to both, and collateral posted at one cannot be redeployed to meet a call at another. Fragmentation across multiple CCPs therefore raises the aggregate, gross demand for margin per unit of exposure, and adds cross-CCP and cross-border coordination and resolution risk that a single regime is spared — with a London CCP still central to euro repo after Brexit. The EU does not inherit the same call. It replaces one form of synchronisation risk with a different and largely untested set: less single-model uniformity, more gross collateral demand and more coordination fragility. EMIR 3.0’s active-account requirement nudges derivative clearing into the EU but does not touch repo; the Bank of England, for its part, opened a discussion of gilt-repo resilience and clearing in September 2025.

The gap

The framework does one thing well and leaves another unpriced. It prices the counterparty credit risk of cleared exposures correctly, and ranks them where their collateralisation and netting earn them. It prices each institution’s own liquidity through the LCR and NSFR, and it prices FICC’s single-member-default liquidity through the clearing fund, the CCLF and the new commercial-paper programme. What none of these prices is the system-wide demand for cash when one uniform margin call lands on the whole cleared population at once. Stated narrowly, so that it is hard to rebut: no instrument prices or constrains the combination this piece has traced — a uniform, model-driven margin demand from the most leveraged segment of the repo market, holding super-senior cleared claims, intermediated through dealers whose main quantity constraint is being eased at the same moment — and the closest committed facility, the CCLF, is itself a synchronised draw on the survivors.

None of this is an argument against central clearing, and none of it forecasts that the next dash-for-cash breaks the Treasury market. It is a structural argument about incentive design, not an empirical claim that the mandate is the dominant source of the next stress. The backstops may hold; the dates may slip; the margin tools may mature. The narrower point is harder to dismiss: the framework treats these exposures as safe one at a time, and the macroprudential question is what happens to all of them at once. The body that comes closest to owning that question, the ESRB, is also the one encouraging the activity that makes it sharper.

Paweł Fiedor - The Macro Prudential View

The views expressed are the author’s own and do not necessarily reflect those of any institution with which the author is or has been affiliated.

All figures from public sources (OFR, TBAC, FICC GSD Public Quantitative Disclosures and SEC rule filings, BCBS-CPMI-IOSCO, Bank of England, FDIC and Federal Reserve, ESRB, ICMA, CRR/Basel framework). Nothing here draws on non-public data, and forward-looking characterisations of ESRB and Commission intent are attributed to the cited public documents.

The May 2026 Financial Stability Review contains a box that anyone worried about equity valuations should read. Its subject is euro-area investors’ exposure to highly valued markets — US technology and AI above all. Cyclically adjusted price/earnings ratios sit well above their own history, US indices well above European ones. Euro-area investors have quadrupled their holdings of US equities over the past decade while only doubling their total equity book, and the fund sector is the single largest holder of both. By the ECB’s own decomposition, roughly 70% of that rise came from valuation effects rather than fresh purchases, and currency moves contributed little. The box builds toward a now-standard worry: thin cash buffers, pockets of leverage, flows that reverse hardest in the most expensive segments, and second-round effects on household wealth and consumption as fund ownership reaches through insurance and pension wrappers.

I am not going to dispute any of that. The valuation read is sound and the transmission story is the right one. What I want to add is the question the supervisory lens leaves implicit — overvalued in what? — and then to follow one thread in the box’s own data, the currency the exposure is denominated in, a little further than the box does. The first runs through most of the piece. The second is where it lands.

A price is a number with a unit inside it

Most valuation arguments smuggle in a unit and a benchmark, then talk as if the result were absolute. CAPE, the measure in the box’s first chart, already does part of the cleaning: it averages real earnings over a decade, stripping inflation out of the denominator, so the FSR is not making the naive nominal-earnings mistake. But the numerator is still a price, and a price is denominated in something. For a euro-area investor that something is the euro. Change the unit, and the level moves.

Here is the S&P 500 — the index those dollar holdings largely track — since January 2015, rebased to 100 and shown in four units. In dollars it is up almost fourfold. In euros, almost exactly the same: across the full decade the exchange rate did little of the work, which is the FSR’s own point. In inflation-adjusted euros it is up about 2.8 times. In gold it is up barely a tenth. The four lines are one price series; none is the index “really” doing what the others are not. No price series comes with a privileged measuring rod attached. A valuation is conditional on a unit — and, the practitioner will add, on a benchmark, a liability structure, and a horizon. Hold the liability structure; it returns at the end.

Note the small gap between the dollar and euro lines. Over ten years it nets to almost nothing, but it is not nothing within the decade: the euro investor ran ahead of the dollar number while the dollar was strong into 2022, then gave it back as the euro firmed. That gap is the exchange rate.

The unit does more than colour the read. How expensive the market looks — in the unit investors and risk managers actually use — can feed position sizing, leverage appetite, and crowding: a market that reads as merely full rather than extreme invites a larger book. On that view the numéraire-dependent read of “expensive” helps shape how much nominal claim gets built in the first place — the stock of leverage and fund assets the acute trigger later acts on. The box’s finding that flows reverse hardest in the most expensive segments is at least consistent with this kind of valuation-sensitive crowding, even if it is evidence about the unwind rather than direct proof of how the position was sized.

The risk premium, measured two ways

The textbook version of the units problem is older than the debate. Modigliani and Cohn argued in 1979 that investors discount real cash flows at nominal rates and misprice the difference; its everyday descendant is comparing an equity earnings yield to a nominal bond yield — subtracting a nominal number from a real one and calling the gap a premium.

Compute the euro-area equity risk premium against the nominal ten-year and against the real ten-year, and the difference between the two readings is, by construction, expected inflation: about 2.8 points at the end of 2022, a little above two now. It is the same lesson as the first chart in a different metric — “stocks look expensive relative to bonds” turns on which yield you put on the other side.

Is some unit better?

If the unit is a choice, the natural reach is for a stable one, and it usually ends at gold. The data ends the reach quickly.

In constant dollars, gold’s January 1980 peak was not durably regained for forty-five years: the nominal price cleared the old high in 2006, but in real terms not until early 2025. A rod that loses much of its real value and needs a generation and a half to recover is no steadier than the currencies it is meant to discipline. The distinction is between a valid unit and a stable one — gold is a perfectly good thing to quote a price in, and its own instability is exactly why switching to it moves the reading. “Cheap in gold” is not firmer measurement, only measurement against a rod that happens to be long right now.

What actually binds

So far this is a piece about measurement, and measurement is a problem for the valuation debate. Whether it is a problem for stability needs the build-up and the trigger held apart.

The build-up metrics are unit-sensitive, in both senses above: the read is conditional on a rod, and the read helps size the position. The acute trigger usually is not. A fund meets redemptions in euros; a margin call is denominated in euros; leverage is a nominal liability the borrower cannot reprice when repricing would help. The dash-for-cash of March 2020 and the 2022 gilt/LDI episode were nominal liquidity spirals, and neither turned on the correct numéraire. So the unit critique bites hardest where nothing acutely binds — what a portfolio is worth, whether an index is dear, how rich a household feels — and barely touches the nominal plumbing that does the forcing.

The clearest place the measuring rod can become part of the margin call is the cross-currency balance sheet, and it is the one the box’s own holdings data draws. A euro fund holding the dollar index has euro liabilities and dollar assets. Redemptions and margin fall due in euros; the assets sold to meet them are priced in dollars. In the calm build-up this is, exactly as the box found, a currency footnote — FX did little. In stress, whether the currency leg helps or hurts depends on the kind of stress. In a classic flight-to-quality the dollar is the haven: it rises as risk assets fall, and for the euro holder the stronger dollar cushions the euro value of the loss. The mismatch bites in the other regime — a dollar-confidence shock, where US assets and the dollar fall together. There the euro fund is hit twice, the asset drawdown and a falling dollar widening the euro loss at the same moment redemptions and margin come due in euros. The numéraire, a footnote in the build-up, becomes the exposure in precisely the stress that is about the unit itself. That state-contingent flip is what the title names.

How far this binds is conditional, and the conditions are what separate it from a generic point about currency mismatch. It binds to the extent the currency exposure is left open rather than hedged; to the extent margin and redemptions fall in the home currency while the assets do not; and to the extent collateral or liquidity transformation leaves the fund short the currency of its liabilities when it most needs it. A fully hedged book with home-currency collateral does not have this problem. An imperfectly hedged one, meeting euro redemptions out of dollar assets under a hedge that itself calls variation margin in stress, does. The seam is specific enough to locate in a given fund’s book.

The map is analytical judgement, not measurement, and is drawn that way. Two questions: does the unit critique change the reading, and does the channel bind acutely? The acute, unit-invariant triggers sit bottom-right — domestic leverage and margin, open-ended fund redemptions. The unit-sensitive items that mostly do not acutely bind sit top-left — equity “overvaluation” as a level, household real wealth — though the level is not a pure non-binder, since, as above, the unit-dependent read may scale the exposure that later does. Cross-currency balance-sheet mismatch sits top-right, the one place both hold at once. Pension and insurance solvency sits between: real or inflation-linked liabilities make nominal-versus-real change the economic-solvency read, and while the channel is slower and better buffered than an FX margin call, a joint inflation-and-FX stress — the scenario where the unit matters most — can move it rightward toward binding faster than the baseline suggests.

The gap, named without a remedy

There is a blind spot here worth naming without pretending to close it. EU fund liquidity stress testing, under the ESMA guidelines in force since 2020, requires managers to stress both assets and liabilities — redemptions above all — and is deliberately principles-based: the manager selects the risk factors the fund’s profile warrants. Asset-side liquidity is the natural focus — how fast holdings can be sold, and at what haircut. The currency-translation leg sits at the seam between market-risk and liquidity-risk testing: a euro fund’s dollar assets, sold into euro redemptions with the exchange rate moving inside the same stress, is a cross-risk interaction that can be under-weighted in practice, especially where the FX hedges themselves call collateral at the wrong moment. Whether a given manager models it is left to the firm. “FX did little in the build-up” — the box’s finding — and “FX does little in stress” are different claims, and the distance between them is the blind spot.

The box ends on household second-round effects — wealth, consumption, the real economy — and that is the right place to end, because it is unit-dependent in just the way this argument has been. A wealth effect is a change in what a portfolio is worth, and worth is denominated in something; the consumption response runs on real purchasing power, a numéraire-sensitive quantity and a slow one. That is why the household channel sits where it does on the map — the rod matters, and it binds gradually if at all. The thing that forces the selling, when selling is forced, is in euros. The clearest place the rod can become part of the margin call is the cross-currency balance sheet the box’s own data describes, under conditions specific enough to look for. Whether it matters in the next cycle I don’t know. It is the part of the map I would watch.

Paweł Fiedor — The Macro Prudential View

The views expressed are the author’s own and do not necessarily reflect those of the Central Bank of Ireland, the European Systemic Risk Board, the Eurosystem, or any other institution with which the author is or has been affiliated.

Sources: European Central Bank, Financial Stability Review, May 2026, box “Drivers of investor behaviour in highly valued equity markets” (Baudino, Bosio, Dieckelmann, Kaufmann and Puga) — https://www.ecb.europa.eu/press/financial-stability-publications/fsr/focus/2026/html/ecb.fsrbox202605_04~fd7b91dd22.en.html; S&P 500, EUR/USD and COMEX gold front-month, Yahoo Finance (^GSPC, EURUSD=X, GC=F); euro-area HICP, Eurostat (prc_hicp_midx) — https://ec.europa.eu/eurostat; forward earnings yield from the NTM price/earnings ratio of the iShares MSCI EMU ETF (EZU), Koyfin / S&P Capital IQ consensus; German ten-year nominal yield, FRED / OECD (IRLTLT01DEM156N) — https://fred.stlouisfed.org; German ten-year expected real rate, Deutsche Bundesbank (BBSEI.M.ERZ.GVB.DE._Z.R10XX) — https://www.bundesbank.de; long-run monthly gold price, DataHub.io — https://datahub.io; US consumer prices, FRED (CPIAUCSL); ESMA, Guidelines on liquidity stress testing in UCITS and AIFs (ESMA34-39-882), in force 30 September 2020 — https://www.esma.europa.eu/document/guidelines-liquidity-stress-testing-in-ucits-and-aifs; F. Modigliani and R. Cohn, “Inflation, Rational Valuation and the Market,” Financial Analysts Journal (1979); J. Y. Campbell and T. Vuolteenaho, “Inflation Illusion and Stock Prices,” American Economic Review (2004). Chart data as cited in each figure.

Macroprudential policy has a working theory of what makes a financial system safe. Keep household leverage from outrunning income. Keep the external position from depending on financing that can reverse overnight. Hold down currency mismatches on unhedged balance sheets. Lean against the procyclical build-up of all three. The toolkit — countercyclical capital buffers, loan-to-value and debt-service limits, restrictions on foreign-currency lending — is a set of instruments for producing a balance sheet that can take a shock without amplifying it.

Poland arrived at the 2008–09 crisis with much of that balance sheet already in place, and largely before it had the tools. The modern Polish framework — the systemic risk board, the activated buffers, the borrower-based measures — came mostly after 2011, after the events that made Poland look resilient. The resilience came first. My argument is that it came in significant part from the composition of Polish growth: not the headline rate of expansion but what the expansion ran on and how it was financed. By the time the crisis arrived, Poland already had several of the balance-sheet characteristics that macroprudential policy later seeks to preserve.

It did not have all of them. The same growth that kept aggregate leverage moderate sat on top of a Swiss-franc mortgage stock larger, in absolute terms, than Hungary’s. That is not a footnote to the argument; it is the argument’s other half. Poland’s growth composition was strong on two counts and weak on a third, and the count it failed on produced a concentrated vulnerability that the favourable aggregate picture could not see and that fifteen years of resolution has still not cleared. Aggregate resilience and concentrated fragility on the same balance sheet — that recurring pattern is the through-line of this piece. It appears before the crisis in the foreign-currency mortgage book, and it appears today in a deeply negative credit gap sitting over a flagged housing-market risk. The aggregate flatters; the concentration bites; the distance between them is where the macroprudential questions live.

A word on what this is. It is a structured comparison across a handful of European economies through one shock, not an identification strategy. The things I am separating — leverage, external financing, currency mismatch, real-economy structure, the exchange-rate regime — co-moved in the countries that had them, and I make no claim to have isolated any one. To keep the central term from swallowing everything, I use growth composition narrowly, in three parts: the leverage intensity of the expansion, the currency in which that leverage is denominated, and the liability structure of the external financing behind it. Two things sit deliberately outside that definition. A country’s structural endowment — domestic-market size, trade openness, sectoral concentration — is largely inherited rather than chosen, and it turns out to matter as much. The exchange-rate regime is a monetary choice that transmits the balance sheet into outcomes rather than forming part of it. Keeping these three layers apart is what stops “composition” from becoming a synonym for “everything that went right,” and it is what lets the argument concede, later, that composition was not always the decisive layer.

One thing this definition leaves out is worth naming rather than smuggling past. The sectoral destination of credit — whether the borrowing built tradable-sector capacity or financed property and consumption — is a genuine determinant of long-run resilience, and two economies identical on my three counts could differ sharply on it. I scope it out here on purpose: the evidence in this piece is about the financial structure of the expansion and the balance sheet it left behind, not about where the credit was ultimately deployed, and that allocation question deserves its own treatment rather than a fourth column bolted onto this one.

The puzzle

Poland was the only EU economy to avoid a recession in 2009. The point worth pressing is comparative, because Poland was not the only fast-growing economy in the region beforehand. Latvia, Hungary and the rest of the CEE bloc were converging on Western European incomes at speed; Spain was doing the same inside the euro area. They were all growing. The chart shows the growth resolving into a spectrum of outcomes rather than a clean binary, from Poland’s continued expansion through Czechia’s and Hungary’s moderate dips to Latvia’s collapse.

Headline GDP conceals the difference because it counts output regardless of what produced it. An economy expanding on a property and consumption boom financed by reversible foreign capital and one expanding on productivity and employment post similar numbers and carry different risks. The macroprudential question is whether the difference is legible before the shock rather than only in the wreckage.

One country should make the argument nervous from the start, so I will name it now. Czechia had a floating currency and low household leverage — much of what I am about to credit Poland with — and still contracted close to 5% in 2009. Hold that case. It is the test the argument has to pass, and I return to it once the mechanism is visible, because Czechia does not break the thesis so much as fix its limits.

Aggregate leverage

By 2010 Polish household credit stood near 36% of GDP, Czechia’s at 30%, Hungary’s at 25%. Spain’s was 73% and had peaked near 86%. These are not points on the same distribution. Spain and the euro-area periphery had undergone a financial deepening of a different order, much of it into residential property; the CEE economies were building leverage from a lower base and more slowly. Polish credit grew, but in proportion to an economy itself expanding fast — leverage added to a denominator that was genuinely enlarging.

This is a leverage measure, not a decomposition of growth into productive and unproductive parts, and it should not be asked to carry more than it can. It does not establish that Polish growth was productivity-led and Spanish growth was not. It establishes something narrower: entering the crisis, Polish and Czech household balance sheets carried far less debt relative to income than Spanish ones, and were correspondingly less exposed to debt-deflation and to a rise in debt-service costs. That is the balance-sheet outcome a borrower-based macroprudential regime exists to produce, and Poland had it without yet having the regime. On the first of composition’s three counts — leverage intensity — Poland scored well.

The Hungarian line does something the others do not. It jumps, from around 37% in 2009 to 41% by mid-2010, then slides. That movement is not new lending; it is the forint depreciating and revaluing a stock of foreign-currency mortgages upward in domestic-currency terms. Ordinary credit does not behave like that. The line is carrying a currency mismatch, which is the cue to look at the mismatch directly — because Poland was carrying a larger one, and that is the count on which it did not score well.

The mismatch underneath

Poland’s household FX loans — overwhelmingly Swiss-franc mortgages — peaked near €50bn in July 2011, larger in absolute terms than Hungary’s €21bn, the exposure usually cited as the cautionary CEE case. This is the second count of composition, currency denomination, and Poland scored badly on it. The point needs stating carefully, because it is easy to overstate in either direction. The mismatch did not mean Poland’s aggregate leverage was secretly high: the €50bn book was around 13% of GDP, roughly a third of a household-credit ratio that was itself low, and stripping it out would leave Polish leverage lower still, not higher. Cheap Swiss-franc credit expanded mortgage access; it did not disguise a domestic-currency binge that would otherwise show. Low leverage and a concentrated currency mismatch genuinely coexisted. But a leverage ratio that is low and a third unhedged foreign currency is worth less as reassurance than the ratio alone suggests, which is exactly why currency denomination belongs inside the definition of composition and not outside it. Poland’s composition was mixed: strong on volume, weak on denomination.

Where did the denomination failure come from? Partly from the growth model, but not as cleanly as a tidy story would want. A fast-converging economy runs structurally higher inflation and therefore higher policy rates than the core, and the gap between złoty and Swiss-franc rates was wide and persistent; that differential created the pricing incentive for households to borrow in francs. The convergence process that shaped Poland’s macro balance also helped create the incentive for the mismatch. But the incentive was not the scale. Czechia and Romania were converging floaters with rate differentials of their own and did not build Poland-scale franc books; the differential was necessary, not sufficient. What turned an incentive into €50bn was supervisory tolerance of unhedged foreign-currency lending to households and the competition among foreign-owned banks to write it. The growth model lit the incentive; supervision and bank competition let it run. That distinction matters, because it locates the failure precisely where a tool could have acted — a cap or ban on foreign-currency lending to unhedged borrowers — and Poland’s tools were absent or weak through the build-up.

How the two stocks were resolved is a separate question from how they were built, and it runs along a different axis again: not growth composition but the capacity and willingness to act. Hungary extinguished its FX mortgage book almost completely through a forced conversion in 2014–15, converting borrowers to forint by statute, pushing the currency risk onto the banks, the exposure gone in roughly a quarter. Poland did not. Its stock has come down slowly, through amortisation and, since around 2020, through individual litigation, with Polish courts and the Court of Justice of the EU annulling Swiss-franc contracts on consumer-protection grounds. By mid-2022 the ESRB recorded almost 30% of CHF contracts in litigation; about €9bn remained outstanding in early 2026, with provisioning against the litigation a continuing drag on bank profitability.

On the narrow metric macroprudential authorities weigh most heavily — the speed with which a systemic mismatch leaves bank balance sheets — Hungary did better, doing in a quarter what Poland has not finished in fifteen years. Whether it did better overall is genuinely unsettled. Forced conversion carried a cost that does not appear as a balance-sheet line: a state that rewrites financial contracts by statute may raise the premium it later pays for bank and sovereign funding, with second-round effects on credit supply, though how large that cost was, and whether it exceeded the drag of Poland’s fifteen-year litigation overhang, is not something the evidence here settles. What the comparison does show is cleaner than a verdict. Hungary had the worse growth model and the greater resolution capacity, and used it; Poland had the better model and let the mismatch sit. Resolution capacity and growth composition are different things, and a country can be strong in one and weak in the other.

With both halves on the table — composition strong on leverage, weak on denomination, and resolution a separate axis entirely — the rest of the piece concerns the third count, external financing, and why Poland’s aggregate position held when the shock came.

What financed the deficit

Latvia ran current account deficits above 21% of GDP in 2006 and 2007; when the financing stopped in late 2008 the deficit swung to surplus within a year, but only because imports collapsed and output fell by a fifth. Spain, inside the euro and so without the same risk of abrupt currency-driven flight, ran deficits near 9% and adjusted slowly through years of internal grind. Poland’s deficits stayed in the mid-single digits — not dramatically smaller than Hungary’s, and at points not far from Spain’s. The size of the deficit is not what set Poland apart. What differed was the liability side: how the deficit was financed, and how reversible that financing was.

The contrast is the point, and it is one of scale and reversibility rather than of one pure financing type against another. Latvia financed its deficit overwhelmingly through short-term cross-border bank funding, the bulk of it from Swedish parents — Swedbank and SEB — lending into their Baltic subsidiaries and onward into real estate. That funding ran past 37% of GDP at its 2007 peak, and it reverses on a trigger: in 2009 it went sharply negative as the parents pulled back. The Riksbank’s own retrospective identifies the Baltic crisis as a combination of fixed exchange rates, foreign-currency debt and heavy parent-company funding of bank subsidiaries; the ECB describes those banks coming under acute pressure when intragroup liquidity seized. Poland’s bank-funding component never approached that scale — it peaked around 7% of GDP — and it did not reverse in 2009. Poland’s inflows were spread across direct investment, portfolio investment and bank funding rather than concentrated in the one category that runs.

Three honest qualifications keep this from overreaching, and none of them dissolves the contrast. First, the categories are not airtight: portfolio investment is not a stable flow, and over the period it was at times Poland’s largest single inflow, so the claim is not that Poland leaned on sticky FDI but that it leaned far less on the specific reversible funding — short-term parent-bank lines — that turned the Baltic deficit into a sudden stop. Second, a meaningful share of Poland’s FDI went into the banking sector itself, foreign parents capitalising their Polish subsidiaries to fund exactly the retail-lending expansion, including the Swiss-franc mortgages, that produced the mismatch above; the liability structure governs rollover and sudden-stop risk, and on that it protected Poland, but it does nothing about the asset-side currency mismatch the same capital helped fund. Third, Poland’s own mix shifted under stress, with the bank-funding component rising in 2006–08, visible in the chart. The defensible claim is bounded and is enough: Poland’s external financing was structurally less reliant on the funding that can be cut in a quarter, which is why it avoided the sudden stop that crushed Latvia — and that is a different protection from the one the FX book needed and did not get.

What absorbed the shock

The exchange-rate regime is a monetary choice, not a feature of how Poland grew, and the two should not be merged. The złoty’s roughly 30% depreciation between mid-2008 and early 2009 said nothing about Polish growth composition; it reflected a decision to float. What the float did was transmit the balance sheet Poland brought to the crisis into the output it recorded through it. For an economy with a large tradable sector and a household balance sheet not catastrophically short foreign currency, depreciation works as a shock absorber: exports cheapen as domestic demand weakens, and the balance-sheet damage from the weaker currency stays contained.

That second condition is why the easy version — floaters absorb, pegs cannot — needs qualifying, and the qualification runs straight back through the FX chart. Depreciation absorbs a shock only where private balance sheets are not fatally short foreign currency. Hungary floated, and for Hungarian households with forint incomes and Swiss-franc mortgages the depreciation deepened the damage rather than cushioning it; the same currency move that helped Polish exporters tightened the screw on Hungarian and many Polish FX borrowers. The pegged Baltics sat at the other pole, unable to depreciate at all, and took the entire external adjustment through internal devaluation — falling wages, prices and employment, the channel that manufactures non-performing loans and the reason Latvian output fell by a fifth. The float did not create Poland’s resilience. It transmitted a balance sheet that was resilient at the aggregate level, and only to the extent that it was.

Czechia, and the part composition leaves out

Now the case held back. Czechia had a float and low household leverage — good marks on composition as I have defined it — and still contracted close to 5% in 2009. On composition alone the thesis would not survive it.

It survives because resilience required more than good composition, and Czechia is where the structural endowment I set aside earlier does its work. Czechia’s recession came through trade: a small, extremely open economy wired into German manufacturing supply chains, hit hard when those chains contracted in late 2008. A floating koruna helped at the margin but could not offset a collapse in foreign demand for the goods Czechia made. Poland’s larger domestic market and lower trade openness gave the złoty’s depreciation more to work with — more domestic demand to defend and proportionally less exposure to the single channel that carried the shock into Czechia.

I separated endowment from composition at the outset precisely so this would not read as the thesis stretching to absorb its counterexample. Domestic-market size and trade structure are not features of how a country chose to grow; they are largely inherited, and in Czechia’s case the inheritance was decisive. But the right conclusion is not that endowment beats composition and the article should pack up. It is that composition and endowment are distinct failure modes, and in this episode they produced damage of different depth. The countries undone by composition took the deep hits: Latvia, financed by funding that vanished, lost more than a fifth of output; Spain, carrying household leverage near 86% of GDP, lost most of a decade. The country undone by endowment took a shallower, recoverable one: Czechia contracted around 5% and was back above its pre-crisis level by 2010. I would not build a general law out of three outcomes from a single crisis. What this configuration shows is narrower and still useful: when the shock hit, the largest collapses sat with the financing and leverage failures, while a trade-endowment shock produced a normal recession Czechia walked out of. Composition was not the only thing that determined whether a shock was survived, but here it was the layer that travelled with the outcomes a supervisor most wants to avoid.

What I am not ruling out

Poland’s resilience is overdetermined, and honesty requires conceding that some of the omitted factors may be first-order rather than additive. EU structural funds were not merely a helpful inflow; they were a non-reversible source of external financing during precisely the window when private capital retreated, and a stabiliser of that kind weakens how far the Polish case generalises to converging economies without an EU anchor. Foreign-bank ownership, common across the region, interacted with Polish supervision and home-host arrangements differently than in the Baltics, in ways that are about supervisory practice as much as funding structure. Pre-crisis house-price excess was milder than in Spain or the Baltics. Fiscal space let the automatic stabilisers run. The absence of US securitisation exposure spared Poland the first-round hit. A large domestic market — which I have filed under endowment — may simply be one of the principal reasons Poland could absorb the external shock at all.

The claim is therefore not that growth composition was the cause, nor that it can be isolated from the bundle it travelled with. It is that composition is a coherent and observable organising lens for part of that bundle, that it sits upstream of the specific vulnerabilities macroprudential authorities track, and that it was legible in the balance sheet before the shock — which is more than can be said for “Poland was lucky,” a description of the outcome rather than the structure that produced it.

Where Poland stands now

The credit-to-GDP gap is the indicator the Basel framework uses to signal whether a countercyclical buffer should be switched on; above +2 percentage points is the conventional activation level. Poland’s gap has been below that for over a decade and now sits around −17pp, having reached −23pp in 2023. Read literally, there is no credit boom in Poland and the economy is accumulating macroprudential space rather than spending it, which is why Poland’s authorities have kept the buffer rate inactive for most of the period.

The literal reading is the one to resist. A deeply negative gap is a descriptive residual — credit has grown more slowly than output — and a residual has several possible causes. Polish credit has been suppressed on the supply side as much as moderated on the demand side: a bank asset tax since 2016 that penalises lending relative to holding government securities, and capital tied up against the CHF litigation described earlier. Nominal output outrunning credit is consistent with a safe, well-balanced economy; it is equally consistent with a banking sector taxed and litigated into caution. The BIS gap is also known to be unstable in real time, sensitive to the trend it is measured against. I read it as a description of where credit sits relative to output, not as a verdict that Poland is safe.

The sectoral picture is where the verdict lives, and it returns to the through-line. The ESRB’s most recent residential real estate assessment still classifies Poland as medium stock risk, citing accelerating house prices, elevated debt-service ratios on new lending, a high share of variable-rate loans and the unresolved Swiss-franc legacy. A deeply negative aggregate gap sitting over a flagged housing vulnerability is the same configuration as the pre-crisis balance sheet: aggregate moderation on top of a concentration the aggregate cannot see. Fifteen years ago the concentration was a foreign-currency mortgage book; today it is the housing market and what remains of that same book. The instruments differ in resolution, and the coarser one flatters.

What this implies

For most of the period that matters here Poland barely had a macroprudential framework in the modern sense; the institutions and tools arrived largely after 2011, after the events that made Poland look resilient. The resilience came first, from the structure of the economy rather than from prudential policy applied to it.

That has an implication for the people who run these frameworks, and it is sharper than “monitor the growth model,” which is not an operational instruction. The lesson is about which tools sit where in the framework. Poland’s Swiss-franc book built up while aggregate credit stayed moderate, which means the credit-to-GDP gap — the cyclical indicator that anchors the countercyclical buffer — would never have flagged it: a concentrated currency mismatch is not a volume boom, and the gap is built to see volume booms. The borrower-based measures that catch this class of risk — caps on debt-service ratios, limits or bans on unhedged foreign-currency lending — are not cyclical switches a supervisor flips when a dashboard turns red. They are structural underwriting standards, and the Polish case is an argument for treating them that way: in place as a permanent baseline rather than held in reserve for a cyclical signal that, by construction, will not arrive in time for a risk that is not also a boom. Composition metrics earn their place a step below that: not as automatic triggers, but as the early evidence that justifies heightened scrutiny and informs how tightly the structural measures are calibrated. The currency denomination of household leverage, the share of external financing in reversible bank funding, the concentration of the real economy on single external channels: all were legible in Poland’s and Latvia’s data well before 2008, and all speak to risks the cyclical anchor cannot register. The contestable claim is not that supervisors should add three indicators to a dashboard; many already track them. It is that the framework should not depend on the cyclical gap to call for the structural tools, because for this class of risk the gap is the wrong instrument and will stay silent while the exposure builds.

The boundary on all of this is the FX book itself. Good composition on two of three counts did not stop Poland accumulating a Swiss-franc stock larger than Hungary’s; the growth model lit the incentive and supervision let it run; and neither composition nor surveillance resolved it, which is why the stock and the housing risk beside it are still on the ESRB’s list. Reading the terrain earlier is worth doing. It is not a substitute for holding the borrower-based instruments that catch what the terrain leaves exposed, nor for the supervisory willingness to use them — the count on which Hungary, with the worse model, outperformed.

Poland’s 2009 outcome was the right one, and the structure that produced it was partly legible in advance. The unresolved Swiss-franc book is the wrong one, on the same charts, fifteen years on. A resilient aggregate is worth having. It is not, on its own, the same thing as a safe one.

Paweł Fiedor — The Macro Prudential View

The views expressed are the author’s own and do not necessarily reflect those of any institution with which the author is or has been affiliated.

Sources: Eurostat, national accounts — GDP and main components, chain-linked volumes (nama_10_gdp) — https://ec.europa.eu/eurostat; Eurostat, balance of payments by country, current account and financial account (bop_c6_a), and GDP at market prices (nama_10_gdp), author calculations; BIS, Total Credit Statistics — credit to households and NPISHs, % of GDP, adjusted for breaks — https://www.bis.org/statistics/totcredit.htm; BIS, credit-to-GDP gaps (private non-financial sector) — https://www.bis.org/statistics/c_gaps.htm; Basel Committee on Banking Supervision, Guidance for national authorities operating the countercyclical capital buffer (December 2010) — https://www.bis.org/publ/bcbs187.htm; ECB, Statistical Data Warehouse — euro reference exchange rates (EXR) and MFI balance sheet items, loans to households in foreign currency (BSI) — https://data.ecb.europa.eu; Z. Zimny, Inward FDI-Related Challenges to Poland's Further Economic Progress (2015) — https://www.davidpublisher.com/Public/uploads/Contribute/56d3febf33bd9.pdf; Sveriges Riksbank, Financial crisis preparedness in the Nordic-Baltic region, Economic Commentary (2025) — https://www.riksbank.se; ECB Banking Supervision, Financial integration in the Baltics: lessons in resilience and transformation (4 October 2024) — https://www.bankingsupervision.europa.eu/press/speeches/date/2024/html/ssm.sp241004~9588c58d39.en.html; Peterson Institute for International Economics, Baltic Protests and Financial Meltdowns — https://www.piie.com/commentary/op-eds/baltic-protests-and-financial-meltdowns; European Systemic Risk Board, Follow-up report on vulnerabilities in the residential real estate sectors of the EEA countries (February 2024) — https://www.esrb.europa.eu/pub/pdf/reports/esrb.report.vulnerabilitiesresidentialrealestatesectors202402~df77b00f9a.en.pdf; European Systemic Risk Board, national macroprudential measures database — https://www.esrb.europa.eu/national_policy/; M. Piątkowski, Europe's Growth Champion: Insights from the Economic Rise of Poland (Oxford University Press, 2018). Chart data as cited in each figure.

The claim

The thing that makes a risk macroprudential is not that it is large. It is that it is correlated — that many institutions fail in the same way at the same time because they share an exposure. A maturity mismatch at one bank is that bank’s problem; the same mismatch across a funding market, snapping shut on the same morning, is a system’s problem. Supervision is organised around finding those shared exposures and stress-testing what happens when they crystallise together.

There is a class of shared exposure the framework does not look at this way, because it files it under operational risk and operational risk under IT. The financial system runs on a thin, concentrated layer of common substrate: a handful of core-banking platforms, a few cloud providers, a small set of security-tooling vendors, one public-key infrastructure underneath authentication almost everywhere. A flaw in any of those is a correlated exposure across every institution standing on it, and that correlation is the systemic content. Two developments in 2026 made the point concrete, and both are currently treated as engineering questions rather than stability ones. That treatment is the categorisation error this piece is about.

One disciplining caveat governs everything that follows, and I would rather state it once, plainly, than dress it up. In both cases the threat is, today, better evidenced on paper than in the world: demonstrated in controlled settings, or estimated on the page, but not executed against hardened financial infrastructure at scale. I mark claims as established, plausible, or speculative throughout, because the distance between those three is the analysis. The two cases are also not symmetric, and I will not pretend otherwise — the first is the live, uncertain one everyone is talking about; the second is slower-moving but far more certain in direction. They are weighted accordingly.

Case 1: offensive capability is outrunning the cadence of supervisory testing

In the first week of April 2026, Anthropic published a model it declined to sell. Claude Mythos Preview was presented not as a product but as a capability judged too sensitive for ordinary release: a system that finds previously unknown software vulnerabilities and writes working exploits for them with limited human direction. Rather than ship it through an API, Anthropic restricted it to a vetted consortium of around fifty organisations under a programme called Project Glasswing, and reported that in its first month it had surfaced more than ten thousand high- or critical-severity flaws. For nearly two months not one European bank, regulator or government could use it; on 1 June, after the European Commission flew officials to San Francisco, Anthropic agreed to admit the EU’s cybersecurity agency, ENISA, to the programme. The manner of that admission — a bilateral concession in response to a direct ask, rather than the exercise of any European instrument — is itself part of the argument here, and I will come back to it.

The precise capability claim matters, because the overclaimed version invites dismissal and the defensible version does not. What is established: the UK’s AI Security Institute, evaluating the model independently, found it the first to complete a thirty-two-step simulated intrusion — reconnaissance through to full network takeover — that AISI estimates at around twenty expert-hours, and found it able to discover and exploit vulnerabilities autonomously in controlled settings. What AISI states with equal clarity, and what bounds the claim: those environments were small, weakly defended, and lacked the active defenders, segmentation and endpoint tooling of a real enterprise network. The defensible statement is therefore narrow. A model has shown benchmark-tier offensive capability against soft targets. It has not been shown to defeat a hardened, actively defended bank at scale. Everything systemic downstream is conditional on that benchmark-to-production gap closing, which is plausible, not established.

Set against this is the one long-run resilience series the field has.

Mandiant’s global median dwell time — the days an intruder sits undetected inside a network after breaking in — has fallen from 416 in 2011 to the low teens. It is worth being exact about what that shows, because it is easy to misread. Dwell time is measured after a breach succeeds; it captures how fast defenders notice an adversary already inside. Autonomous vulnerability discovery acts before the breach, compressing the work of finding and weaponising a flaw, which raises the probability and frequency of initial compromise. Those are different phases of an attack, and a model that writes a zero-day in minutes changes nothing about dwell time once an attacker is in. So the series says nothing about attacker-side compression. What it does show is narrower and still useful: defence has a measurable cadence, won over a decade against human adversaries at human speed. The open question is whether the pre-intrusion phase is about to move on a timescale that cadence was never set against.

There is a specific supervisory exposure here, though it is narrower than earlier versions of this argument implied, and worth scoping precisely. The Digital Operational Resilience Act’s threat-led penetration testing regime, built on TIBER-EU, is intelligence-led and largely tests the post-compromise contest: given a foothold, can the institution’s defenders detect lateral movement, privilege escalation and exfiltration. A faster initial breach vector does not, by itself, degrade that — the red team can simply begin from the assumed breach, and the blue-team detection it certifies stays comparable. So the honest version is not that AI breaks threat-led testing. It is twofold and smaller. First, AISI’s own ranges show these models executing the whole chain autonomously, including the lateral movement and escalation that TLPT exists to test, which raises the question of whether a red team that lacks frontier tooling is exercising the blue team against the speed and breadth the real adversary now brings. Second, and this is the durable point, an EU red team operating below the offensive frontier — because the most capable tooling is access-restricted — tests against a weaker adversary than the one that exists, and certifies detection against that weaker adversary. That is a calibration concern about the realism of the exercise, not a claim that the instrument is broken; it is one supporting strand, not the load-bearing one.

What actually carries the case is the trend, and here the evaluator is explicit. AISI found a second model, from a different developer, reaching the same tier weeks after the first, and concluded the capability is emerging as a byproduct of general improvements in autonomy, reasoning and coding — with further increases expected “in quick succession.” That is the systemic content: not one fenced tool, but a frontier that is moving, broadening across developers, and only contingently access-restricted.

The scope of the claim has to be drawn carefully, because the obvious rebuttal is correct as far as it goes. Most of a European bank’s defensive patching is inherited: global vendors fix a flaw once and push it everywhere, and several of those vendors — Microsoft, Google, AWS, CrowdStrike, Cisco — sit inside the Glasswing consortium. A better red-team tool does not let a bank rewrite a vendor’s source code; it cannot fix what only the vendor can fix. So the margin that matters is not patch distribution, where Europe is largely fine. It is the residual window between a frontier-grade flaw being exploitable and the vendor’s fix arriving — the interval in which detection, containment and response, not patching, are the only defence. That residual window is exactly where first-party capability and testing realism decide the outcome. The claim is not that Europe cannot defend itself. It is that the supervisory exercise certifying it can detect a frontier intrusion may be calibrated, through restricted access to the best offensive tooling, to an adversary slower and narrower than the one that now exists.

The access picture sits behind this, and it should be stated as documented participation rather than absolute exclusion. Per Wall Street Journal and Bloomberg reporting: the consortium is mainly US-based; the NSA had access through earlier arrangements; Anthropic’s proposal to widen the group from around fifty to around a hundred and twenty organisations was blocked by the Trump administration, citing misuse risk and concern that broader access would dilute the federal government’s own use. Euro-area finance ministers met in Brussels on 4 May to discuss the lack of European access; the negotiations that followed ran for weeks without result, until the Commission sent officials to San Francisco and, on 1 June, Anthropic agreed to admit ENISA. That outcome is worth reading carefully rather than as simple closure. Europe obtained access to a capability that can be used against its own infrastructure not through any instrument it holds — the AI Act regulates deployment inside the Union and cannot compel an American firm to grant access to a model that sits outside it — but through a bilateral concession secured by travelling to ask for one. A single agency inside a testing consortium is also not a supervisory regime with standing access. So the asymmetry was real and it was contingent: it lasted under two months, then bent to a flight and a request, which says both that the fence was never structural and that Europe’s leverage ran through diplomacy rather than through the regulatory apparatus meant to govern operational resilience.

That reading was overtaken within weeks: in mid-June the US, citing national-security authorities, issued an export-control directive suspending all foreign-national access to Mythos and Fable (its guardrailed public release), forcing Anthropic to disable both for every customer, ENISA’s new access included — Anthropic disputes the basis and says it is working to restore service. Inside six weeks European access went from withheld, to granted by concession, to suspended by a US national-security order that no European instrument could reach or reverse.

The hinge: transient or structural?

That question — passing condition or standing feature — is where the macroprudential weight of Case 1 turns, and I do not think it is currently answerable. The evidence points both ways.

The case that the frontier keeps tilting starts with compute. Frontier training compute has grown roughly four-to-five-fold a year since 2010; my fit to Epoch AI’s frontier-model series returns 4.6×. Extended to 2030, that reaches about 10²⁹ floating-point operations — close to where Epoch’s own analysis puts the ceiling of what electricity can deliver, power binding ahead of chips and data. So one reading is an established trend running into a plausible physical ceiling: the offensive frontier plateaus, and the asymmetry is transient.

The counter-case is not weak. DeepSeek-V3 reached near-frontier capability at perhaps a tenth of the compute of its contemporaries, for a reported final-training-run cost of around $5.6 million — a figure that excludes prior research and ablations and has been widely misreported as total project cost. If algorithmic efficiency keeps improving — Epoch estimates roughly a halving of the compute needed for fixed capability every nine months — a power ceiling on training need not cap capability, because capability can be bought with cleverness instead of electricity. The qualification cuts back toward transience: efficiency has its own diminishing returns, and a power limit is physical in a way ingenuity is not. Neither side is settled.

And DeepSeek is Chinese, which dissolves the clean “fenced by Washington” story rather than complicating it. If frontier-adjacent offensive capability can be reached for single-digit millions outside the US, the consortium’s control is contingent and probably temporary, which is the reason I will not call the asymmetry structural. The fencing is real now; efficiency diffusion is why it may not last; the implication is multipolar rather than US-favouring. That contradiction is better made the argument than hidden: Case 1 is acute and quite possibly transient, and should be sized as such.

One corroboration keeps the near-term reading alive. AISI evaluated OpenAI’s GPT-5.5 on the same battery and placed it in the same tier — the second model to complete the thirty-two-step intrusion, marginally ahead on some narrow metrics. Two labs reaching the same capability independently is harder to contain than one. The same caveat governs throughout: these are benchmark environments without active defenders, so what is established is benchmark-tier capability while the operational threat stays speculative. The gap between the two is the room in which red-team standards and model-assisted defence have time to adapt, slowly but not at zero.

Case 2: the loss is created at capture, not at decryption

The quantum threat is mostly hype, and for most of its advertised uses the skepticism is earned. The physicist Sabine Hossenfelder has put the sharp version: across finance, logistics, chemistry and materials, quantum computing has consistently failed to convert a theoretical advantage into a practical one, and the lone clear exception is codebreaking. For a financial-stability reader that is the load-bearing simplification — it sets aside the portfolio-optimisation claims that rarely survive their implementation costs and isolates the one capability with a systemic payoff: breaking the public-key cryptography that protects stored data and underpins authentication.

That capability splits into two channels, and collapsing them — or waving the threat away as “just privacy” — misses half of it. Google’s March 2026 post setting its own migration deadline draws the line precisely: the threat to encryption is present today through harvest-now-decrypt-later, while the threat to digital signatures and authentication is a future one that must be fixed before a cryptographically relevant machine exists.

The first channel is confidentiality.

An adversary who captures encrypted traffic or an archive today can store it and read it whenever a capable machine arrives; for data whose confidentiality must outlast that arrival, the loss is booked at capture, not decryption. This is the one place a time-separation genuinely does analytical work that concentration alone does not — the exposure and its crystallisation are split across an interval no one can date. A risk officer will rightly push back, and the answer concedes before it holds: not everything is harvested, storage and targeting cost money, and much traffic already uses forward secrecy and short-lived keys. So the channel is real but selective — the marginal exposure concentrates in high-value, long-horizon, poorly-rotated stores: custody records, certain pension and mortgage books, sovereign archives. That is narrower than “the loss is fully booked,” and it is the defensible claim.

The second channel is integrity and availability, and the obvious version of it is wrong. It is tempting to say a quantum computer could forge clearing instructions and inject fraudulent payments. It could not, at least not that way: core interbank messaging and real-time gross settlement rely heavily on symmetric message authentication, which Shor’s algorithm does not break and which a larger key handles. Mass-forging a TARGET2 instruction is not the threat. The exposure is one layer down, in the asymmetric public-key infrastructure that authenticates endpoints, establishes secure channels, issues the certificates counterparties trust, and signs the software updates institutions install. That PKI is pervasive across financial infrastructure — it is what the sector’s own post-quantum migration guides catalogue first — and it is what a cryptographically relevant machine breaks.

The realistic systemic event is not silent, prolonged impersonation; it is more likely a forced halt. Once a certificate hierarchy is known to be breakable, the rational response is to revoke trust in the compromised roots, at which point connections fail closed: secure channels drop, handshakes are refused, and institutions fall back to out-of-band verification and manual settlement. How far that propagates depends on which trust anchors are shared across how much critical infrastructure — existing revocation procedures, segmented trust and out-of-band fallbacks are precisely the controls that would bound it — so the plausible range runs from messy-but-contained disruption to something much wider, rather than an automatic continent-scale outage. But the direction is a denial of service on the trust layer rather than a quiet looting of the system, which is the more credible worry: a system that has lost confidence in its own certificates stops rather than continues corrupted. The integrity exposure lives in the narrower, nastier windows around that halt — the period before the break is recognised, and the migration itself, in which a maliciously signed software update or a forged endpoint certificate could be accepted as genuine before trust is rebuilt on quantum-safe foundations. Either way the point holds against the “just privacy” dismissal: this is an attack on the availability and integrity of market infrastructure, not a data-protection matter, and it is why the migration cannot wait for the machine, even though, unlike harvesting, this channel only opens once the machine exists.

The hardware reality is the counterweight to any sense of imminence.

The estimated physical-qubit cost of factoring a 2048-bit RSA integer has fallen roughly a thousandfold in thirteen years, to under a million in Craig Gidney’s 2025 analysis — and the whole fall is algorithmic and error-correction progress at fixed hardware assumptions, not hardware improving. Yet demonstrated hardware is nowhere close, and the gap is not the one the headlines imply. Industry quotes qubit count, on which Quantinuum’s ninety-four logical qubits look within reach of the fourteen hundred required. But those ninety-four are error-detected — a distance-two code that spots an error and discards the run rather than correcting it and continuing — while the attack needs roughly fourteen hundred error-corrected qubits at code distance near twenty-five, holding integrity through a week of computation. On count and code distance together, every demonstration sits in one corner and the region the attack requires is empty.

The two windows are not the same width. The quantum gap — distance-two detection against distance-twenty-five sustained correction — is wide, plausibly many years. The cyber gap is narrow: the capability exists now, and only operational generalisation is in question. A supervisor faces a fast near-term exposure in Case 1 and a slow but directionally one-sided one in Case 2 — directional not in the sense that Q-Day has a knowable date, which it does not, but that the case for migrating strengthens monotonically as resource estimates fall and embedded dependencies stay hard to inventory. They call for different instruments.

What makes the slow exposure pressing despite the wide window is the migration, not the machine.

Past cryptographic migrations took nine to thirteen years from first warning to broad completion — and those were comparatively easy, near-drop-in replacements for algorithms already known broken. SHA-1’s full federal removal lands in 2030, twenty-four years after the first warning. Post-quantum migration is harder on every axis: larger keys, protocol changes, hardware dependencies, and the unglamorous problem of finding where cryptography is embedded at all. Measured from the 2024 NIST standards, the EU roadmap allows six years to its 2030 high-risk deadline and eleven to full transition in 2035, against a historical base rate north of a decade for an easier task. The honest qualification is that history is a guide, not a forecast: post-quantum migration is also better coordinated, more standardised and more policy-salient than those earlier episodes, which could compress it — or the embedded-cryptography problem could prove worse than any precedent. The base rate frames the risk; it does not settle it.

That a builder of quantum computers is migrating early is the revealed-preference tell, handled carefully. In March 2026 Google set its own internal post-quantum deadline at 2029, ahead of the EU’s high-risk target, citing hardware progress, error-correction advances and the falling factoring estimates. This is not a prediction that a cryptographically relevant machine is near; Google makes none, and neither will I. It is evidence of how much margin a sophisticated actor wants between its defences and a threat it cannot date — conservative private timing under deep uncertainty, not a forecast of Q-Day.

Why the frameworks miss both

The apparatus that should hold these exists and is, on paper, substantial. DORA governs operational resilience and incident reporting; the TIBER-EU regime tests it; the ESRB runs a systemic-cyber framework and a pan-European coordination arrangement for systemic cyber incidents; and the coordinated post-quantum roadmap sets 2026, 2030 and 2035 milestones. Most importantly for the argument here, DORA’s central third-party pillar is already a concentration instrument: the European Supervisory Authorities published their first list of designated Critical ICT Third-Party Providers in November 2025 — nineteen of them, hyperscale cloud, core-banking platforms, data-centre operators — and took direct oversight powers over each, with designation turning explicitly on systemic impact, concentration of reliance, and substitutability. So the framework can see provider concentration, and a piece claiming otherwise would be arguing against its own evidence. But CTPP designation operates at the level of the provider — whether a given critical supplier is systemically important and runs its own house well. It is not built to capture what happens to that provider’s clients together, nor to see the cryptographic layer at all. The gap is not that the framework is blind to concentration. It is narrower, and it lies in two places the CTPP regime does not reach.

The first is the difference between provider concentration and correlated lag — and getting the mechanism right means being precise about where lag actually lives, because not all shared dependencies have it. When a flaw is in a hyperscaler’s managed platform — its hypervisor, a managed database — the provider patches it on the back end and the fix propagates across every client at once, with no client-side action and no window to exploit. That layer concentrates risk but does not lag. The correlated-lag exposure lives in a different layer: distributed software and security agents that each client must deploy itself, and on-premise packages that must be patched one estate at a time. There the same flaw sits exposed across every institution running the dependency, on a correlated clock, during the window before each deploys the fix. The shape of the danger is not hypothetical: the July 2024 CrowdStrike incident, in which a single faulty update to a near-ubiquitous endpoint agent grounded airlines, hospitals and banks within the same hours, was a correlated-failure event through exactly this channel — a shared software dependency, deployed everywhere, failing synchronously. CTPP oversight supervises whether the provider behind such an agent runs its own house well. It does not measure the synchronous exposure of all that agent’s financial-sector clients in the window a frontier-grade flaw is live, because that exposure is a property of the system’s topology rather than the provider’s governance, and it falls between the provider-facing CTPP lens and the firm-facing supervision of each client, seen whole by neither. The second gap is that none of this apparatus has remit over the cryptographic substrate. CTPP criticality is about ICT-service continuity, not about which signature schemes underpin which certificate hierarchies, nor about sequencing a migration so that shared trust anchors do not all reach their deadline unmigrated together. The post-quantum roadmap sets dates but does not own cross-institutional cryptographic concentration. So the formulation is not that no one is looking — it is that provider concentration is supervised, while correlated lag in the distributed-software layer and cryptographic concentration, the two properties that actually synchronise a failure, are currently owned by no single mandate.

That is what makes this macroprudential rather than a pile of firm-level IT problems, and it is a coordination-and-visibility failure across mandates rather than a total blind spot. Firm-by-firm supervision does not, on its own, reliably reveal a system-wide exposure: each institution inventories its own dependencies, tests against its own scenarios, and migrates its own cryptography on its own schedule, and the property that matters — concentration plus correlated lag across the whole estate — is the one no single filing contains. The framework already knows this shape from funding markets. It has built half the apparatus to see it in the substrate, in the CTPP regime, and not yet the half that would catch the synchronous failure.

What a supervisor should measure differently — and what would prove me wrong

The practical content is not a forecast. It is a sequence, and the order matters, because the obvious objection to everything above is fair: if this is really macroprudential, where is the operational equivalent of a capital buffer or a liquidity coverage ratio — the hard instrument, not another reporting form? The answer is that the instruments exist in adjacent form and the binding constraint on using them is measurement. A liquidity coverage ratio was possible only once liquidity could be measured against a defined stress; the buffer came after the metric, not before. The substrate has no agreed metric yet, which is why the first asks are measurement asks — not as a substitute for hard tools, but as their precondition.

So, in order. First, require threat-led tests to record the realism gap they now carry: a TLPT attestation should state whether the red team had access to frontier-grade offensive tooling comparable to the assessed threat, so a certification of detection is read against a known adversary level rather than an unstated one. Second — the prescriptive one — require the ESRB’s systemic-cyber scenarios to carry an explicit autonomous-capability parameter and, for the quantum channel, a code-distance-progress assumption, with a public justification when either is omitted; operationally this means the scenario-design committee adds two named variables to an existing process rather than inventing a new exercise. Third, extend the CTPP regime’s concentration lens from the provider to the correlated-lag exposure of its clients in the distributed-software layer: cross-institutional dependency inventories and cryptographic inventories, the latter tagged by confidentiality duration and signature dependency, held by the authority whose mandate is the system rather than the firm.

And then the direction hard instruments would take — named so the thesis is not just dashboards, but flagged as direction rather than ready policy, because the trade-offs here are real and partly adverse. Once correlated-lag exposure is measurable, the analogue to a large-exposure limit is an operational concentration limit — a ceiling on how much of a critical function the system routes through a single dependency. The honest objection is that infrastructure is not fungible the way capital is: you cannot tell a bank it has “hit its hyperscaler limit” without telling it to stop, and a naive mandate to run core systems actively across two clouds can worsen resilience by doubling the attack surface and multiplying configuration error. So the feasible version is not forced active multi-provider; it is measured concentration thresholds that trigger heightened oversight and tested exit and fallback capability — closer to a supervisory add-on than a hard cap — and it applies most cleanly to the distributed-software layer where substitutability is real, not to the hyperscale oligopoly where it largely is not. The point of leading with measurement is that which of these is even feasible, and where, cannot be known until the concentration-plus-lag exposure is quantified. A limit set against an unmeasured exposure is arbitrary, and arbitrary limits are how good instruments get discredited.

And the kill criteria, because a piece arguing for supervisory attention should say what would retire the argument — in metrics a supervisor can actually read. The thesis weakens materially if a tracked frontier-model evaluation against hardened, actively defended environments (not benchmark ranges) shows the capability failing to generalise; if the share of tier-one institutions’ critical services concentrated in a handful of designated providers turns out low rather than high once the CTPP registers are mapped against each other; if post-quantum migration progress, measured against the historical nine-to-thirteen-year baseline, runs ahead of rather than behind schedule; or if the empty region in the logical-qubit chart stays empty while access to offensive models diffuses widely. Each is discrete and reportable. If they come back reassuring, the right conclusion is that the framework was adequate and the substrate belonged in the engineering domain after all.

The summary is plain. Two correlated exposures surfaced in 2026 that the resilience framework does not yet hold as stability concerns — one live and uncertain, one slow and directionally one-sided — and both turn systemic through a property the framework half-sees: it supervises provider concentration through the CTPP regime, but not the correlated lag or the cryptographic concentration that would actually synchronise a failure. In Case 1 the supervisor’s window is narrow and may be closing as the capability generalises or diffuses. In Case 2 it is wide, but the migration clock inside it is already short against the historical base rate. The cost of recognising this late is specific and not abstract: a single flaw in a shared dependency detected across the estate in the same week, or trust in a common certificate hierarchy failing closed during a stress window, is exactly the correlated event prudential supervision exists to prevent — and the one the substrate is currently arranged to produce. The instruments to measure it are a scenario parameter and a register away. The instruments to limit it are the ones the framework already owns, pointed at a layer it has not yet pointed them at. What is missing is not the tools but the decision to treat the substrate as load-bearing before the event that proves it was.

Paweł Fiedor — The Macro Prudential View

The views expressed are the author’s own and do not necessarily reflect those of any institution with which the author is or has been affiliated.

Sources: Anthropic, Project Glasswing announcement and Claude Mythos Preview, 7 April 2026 — https://www.anthropic.com/glasswing; UK AI Security Institute, Our evaluation of Claude Mythos Preview’s cyber capabilities, April 2026 — https://www.aisi.gov.uk/blog/our-evaluation-of-claude-mythos-previews-cyber-capabilities; UK AI Security Institute, Our evaluation of OpenAI’s GPT-5.5 cyber capabilities, April 2026 — https://www.aisi.gov.uk/blog/our-evaluation-of-openais-gpt-5-5-cyber-capabilities; Wall Street Journal, White House Opposes Anthropic’s Plan to Expand Access to Mythos Model, 30 April 2026 — https://www.wsj.com/tech/ai/white-house-opposes-anthropics-plan-to-expand-access-to-mythos-model-dc281ab5; Bloomberg, reporting on the reported private-forum access leak, April 2026 — https://www.bloomberg.com/news/articles/2026-04-21/anthropic-s-mythos-model-is-being-accessed-by-unauthorized-users; Bloomberg, Anthropic to Give Mythos Access to EU Cybersecurity Agency ENISA, 1 June 2026 — https://www.bloomberg.com/news/articles/2026-06-01/anthropic-to-give-eu-s-cybersecurity-agency-access-to-mythos; Eurogroup, statement following the meeting of 4 May 2026 — https://www.consilium.europa.eu/en/meetings/eurogroup/2026/05/04/; Mandiant, M-Trends 2025 (global median dwell-time series, 2011–2024) — https://cloud.google.com/security/resources/m-trends; Mandiant, M-Trends 2026 (2025 dwell-time figure) — https://cloud.google.com/security/resources/m-trends; Epoch AI, frontier-models database and Can AI Scaling Continue Through 2030? — https://epoch.ai/data/notable-ai-models and https://epoch.ai/blog/can-ai-scaling-continue-through-2030; International Energy Agency, Energy and AI, April 2025 — https://www.iea.org/reports/energy-and-ai; DeepSeek-AI, DeepSeek-V3 Technical Report, arXiv:2412.19437 — https://arxiv.org/abs/2412.19437; C. Gidney, How to factor 2048-bit RSA integers with less than a million noisy qubits, arXiv:2505.15917 (2025) — https://arxiv.org/abs/2505.15917; C. Gidney and M. Ekerå, How to factor 2048-bit RSA integers in 8 hours using 20 million noisy qubits, Quantum 5:433 (2021) — https://quantum-journal.org/papers/q-2021-04-15-433/; Quantinuum, Helios system results, March 2026 — https://arxiv.org/abs/2602.22211; Google Quantum AI, Quantum error correction below the surface code threshold (Willow), Nature, December 2024 — https://www.nature.com/articles/s41586-024-08449-y; Harvard/QuEra et al., Logical quantum processor based on reconfigurable atom arrays, Nature, December 2023 — https://www.nature.com/articles/s41586-023-06927-3; NIST, post-quantum cryptography standards FIPS 203/204/205, August 2024 — https://csrc.nist.gov/projects/post-quantum-cryptography; NIS Cooperation Group, Coordinated Implementation Roadmap for the Transition to Post-Quantum Cryptography, 2025 — https://digital-strategy.ec.europa.eu/en/library/coordinated-implementation-roadmap-transition-post-quantum-cryptography; Google, A timeline for post-quantum cryptography migration, 25 March 2026 — https://blog.google/innovation-and-ai/technology/safety-security/cryptography-migration-timeline/; European Supervisory Authorities (EBA, EIOPA, ESMA), first designation of Critical ICT Third-Party Service Providers under DORA, 18 November 2025 — https://www.eba.europa.eu/publications-and-media/press-releases/european-supervisory-authorities-designate-critical-ict-third-party-providers-under-digital; ESRB, systemic-cyber framework and EU-SCICF — https://www.esrb.europa.eu/news/pr/date/2022/html/esrb.pr.220127~f1548f677e.en.html; S. Hossenfelder, video commentary on quantum computing and cryptanalysis, 2025–26 — https://www.youtube.com/watch?v=N-9muK0mv5w&pp=ygUbc2FiaW5lIGhvc3NlbmZlbGRlciBxdWFudHVt & https://www.youtube.com/watch?v=qV7hQEtr3ic&t=2s&pp=ygUbc2FiaW5lIGhvc3NlbmZlbGRlciBxdWFudHVt. Chart data as cited in each figure; dwell-time series from M-Trends as above; logical-qubit and code-distance points from the Gidney, Quantinuum, Google and QuEra sources above.

On 11 May the European Commission delivered its report on the functioning of the Money Market Fund Regulation. The EU’s €1.95 trillion sector, it found, has grown by close to half since 2019 and came through both the March 2020 dash for cash and the 2022 gilt crisis without a single fund gating or suspending redemptions. Funds hold liquidity comfortably above the regulatory minima. The appropriate response, the Commission concluded, was not to reopen the Regulation but to set a non-binding supervisory benchmark: weekly liquid assets of around 40% for the stable-value funds, against a binding floor of 30%.

On 14 May, HM Treasury and the FCA confirmed the opposite instinct. Having consulted on this since 2023, the UK announced it will legislate to raise its binding liquidity requirements and, more consequentially, to sever the regulatory link between a fund’s liquidity level and the tools that let it slow or stop redemptions. The new regime is expected to be in place by the end of 2026.

The two documents describe the same instrument, draw on the same two crises, and read off much the same data; they reach opposite conclusions about what to do.

There is a third party to the disagreement, and it had already cast its vote. Fifteen months earlier, in February 2025, the European Systemic Risk Board graded the Commission materially non-compliant with the recommendations it had issued in 2021 on precisely this question. The assessment was neither gentle nor fringe: the team included Richard Portes and was chaired from within ESMA. On three of the four headline recommendations — the ones that would have changed the funds’ structure and their liquidity — the Commission had taken no action, and its justifications were judged insufficient.

That grade measured the Commission’s response against the ESRB’s recommendations: it records that the action fell short of what was asked, not that the sector failed a test. Even read that narrowly, it means the Commission reached its reassurance over the objection of the authority it sits beside on systemic risk, on a reform agenda that authority still judges unfinished. Working out who has the better of it means being precise about what a liquidity buffer measures, and what lies beyond its reach.

A short word on the LVNAV

European regulation offers three kinds of money market fund. Public-debt constant-NAV funds hold government paper and maintain a fixed share price. Variable-NAV funds let their price float with the value of their assets. Between them sits the low-volatility NAV fund, the LVNAV, which is the one that matters here and the largest single category in Europe at roughly 46% of the sector.

The LVNAV is a hybrid by construction. It values its assets at amortised cost and rounds its share price to a stable €1.00, but only while its true mark-to-market value stays within twenty basis points of par. Step outside that collar and the fund must price at its real value, like a VNAV. The stability is therefore conditional, and the condition is invisible to the investor until the moment it fails. That is the feature corporate treasurers pay for: cash that looks and settles like a deposit. It is also where the fragility sits. The Commission understood as much once — its own 2013 impact assessment warned that a constant share price “triggers false incentives and exacerbates runs.”

Why a money fund runs

The collar builds in a cliff. A fund whose assets are quietly losing value still transacts at €1.00, right up to the twenty-basis-point edge, at which point the price drops to meet reality. An investor who redeems while the fund is inside the collar gets out at par. An investor who waits until it breaks absorbs the loss the early movers left behind. The collar holds the price flat and then moves it in a single step, so the deterioration that a floating fund would show gradually arrives here all at once.

The Regulation adds a second cliff beside the first. When a fund’s weekly liquid assets fall below 30%, its board is obliged to consider imposing fees or redemption gates. The trigger is discretionary, not automatic. But investors do not need it to be automatic; they need only to anticipate it. Watching the buffer approach 30%, a holder knows a breach may bring restrictions, so the moment to leave is before the threshold is reached, while redemptions still clear freely and at par. The discontinuity that matters is in the behaviour, not in the text of the rule.

Put the two together and the redemption decision becomes a coordination game with two equilibria, one of them bad.

If you expect other investors to sit tight, you sit tight too: the fund holds at par, you keep the yield, and redeeming would only cost you a little of it. If you expect them to run, the calculation inverts, and redeeming becomes the one way to get out at par, ahead of the step and ahead of any gate. Holding and running are both equilibria, and which one obtains depends on what each investor believes the others will do.

The fragility is in the second equilibrium. The calm one holds only while everyone trusts that it will hold. Any event that shifts enough investors toward expecting a run makes running the correct response for each of them individually, and the expectation then confirms itself. The collar and the threshold are what give the run its edge, because they put a discrete prize on being early: exit at par, before the step. This is the first-mover advantage the ESRB identified in 2021. It does not require panic or misjudgement; it follows from the payoffs.

There is a corollary the Commission’s own evidence supports and its conclusion passes over. If drawing on a buffer is what signals distress and trips the threshold, managers will be reluctant to draw on it. The ESRB found exactly this when it examined March 2020: funds suffered heavy outflows but did not run their weekly liquid assets down to match, which the assessors read as “low levels of usability” for the buffers. A buffer that a manager will not spend, for fear that spending it invites the run, is worth less in an event than its headline size suggests.

The constraint outside the fund

The report measures resilience as a fund’s ability to meet redemptions out of its own liquid assets without forced selling. On that measure the sector looks sound, because funds do hold liquidity well above the minima. The measure is a single-fund one, though, and money funds are tested by system-wide events: in a stress, every fund holds similar paper and reaches for cash at the same moment.

Start with how a fund actually raises that cash, because the intuitive version of the short-side argument goes wrong here. Weekly liquid assets are not mostly a stock of sellable securities; they are largely paper maturing within the week. A fund under redemption pressure meets it first by letting that paper mature and collecting the proceeds, and by declining to reinvest. It does not have to sell commercial paper into a thin market to find the cash. For the individual fund, the buffer works more or less as designed.

The strain does not disappear; it moves to the issuer. When a money fund collects its maturing paper and declines to roll, the issuer that sold that paper — a bank or a large corporate funding itself week to week — has to refinance somewhere else. The natural buyers of the new paper are money funds, and in a system-wide event they are all doing the same thing at once: husbanding cash rather than extending it. So the issuer cannot roll at the old size or the old price. The pressure the fund relieved on its own balance sheet reappears on the issuer’s, at the rollover margin. And it does not stay wholly on the issuer’s side. A fund meets this week’s redemptions from maturing paper, which pays at par, so the buffer does its job there. The credit strain lands instead on the same issuer’s longer-dated paper, still on the fund’s books: it reprices as the issuer’s spread widens, and in the extreme, if the issuer defaults, even maturing paper stops paying in full. That is what moves the shadow NAV toward the collar. Holding more weekly liquid assets reduces this exposure, since it leaves a smaller spread-sensitive long book, though it does nothing for the dealer-capacity limit that decides whether the issuer can refinance at all. None of this puts issuer-side strain in a separate regulatory silo: it returns to the funds through their own asset marks.

Dealers are meant to bridge exactly this kind of gap, standing in to buy paper that cannot immediately be placed. Their capacity to do so is fixed in the short run by leverage and capital rules, and when the whole sector turns at once it is quickly exhausted. Price impact becomes severe, and at the margin some issuance cannot be placed at a price the issuer can pay. The adjustment then runs through quantities — issuance abandoned, rollovers refused — and not through price alone. A fund’s weekly liquid assets tell you whether that fund can find cash. They tell you nothing about whether the market can absorb every fund finding cash together. Raising every fund’s buffer improves the first and relocates the second rather than resolving it: each fund is individually safer, while the aggregate strain shifts onto issuers and a dealer sector that cannot grow to meet it. A higher buffer changes the timing and size of that strain; it does not create the capacity to absorb it.

None of this is new, and the piece is better for not pretending otherwise. The role of dealer balance-sheet capacity as the binding constraint on absorbing forced sales, with quantity rationing taking over from price once that capacity is reached, is the standard reading of the 2020 dash for cash. The Financial Stability Board’s Holistic Review of the March Market Turmoil set it out at the official level in November 2020, and it runs through the central banks’ own accounts of why they intervened and the wider market-based-finance literature. The Federal Reserve stood up two facilities in March 2020, not one: a money market fund liquidity facility to take funds’ assets, and a commercial paper facility to backstop issuance. Both channels had seized — the fund-level and issuer-level constraints are one constraint seen from two ends.

The Commission sees this. Its report accepts that dealer liquidity provision is more constrained and less responsive than before 2008, characterises much of the problem as external to the funds, and points to the right remedies: deeper short-term funding markets and access to central-bank facilities. Having located the binding constraint outside the Money Market Fund Regulation, it concludes that the Regulation itself needs no substantive change. That is where the diagnosis and the conclusion stop fitting together. An external constraint is a reason a fund-level buffer cannot fix the system; it is not a reason to leave every other lever untouched, including the levers, such as the run-inducing threshold and the stable share price, that sit squarely inside the Regulation. This is not a call for a bigger buffer to do a buffer’s impossible job. A higher, usable floor buys fund-level resilience while the systemic gap still needs the external tools the Commission names; the complaint is that the Commission declined the structural levers it did control and answered with a benchmark no fund is obliged to meet.

Three regimes, three answers

Set the three big regimes beside each other and the EU’s choice looks lonelier than its report admits.

The United States went first and went furthest. In 2014 the SEC required institutional prime and institutional municipal funds — the ones most prone to run — to abandon the stable share price and float their NAV, a change that took effect in 2016. It removed the deposit illusion from the funds where it was most dangerous. Then in 2023, in response to March 2020, it raised weekly liquid assets to 50% and daily to 25% across the board, and cut the automatic link between breaching a liquidity threshold and imposing fees. The American settlement is a floating price for the riskiest funds and high, usable buffers for the rest.

The United Kingdom proposes to keep the stable-value structure and the collar but to raise weekly liquid assets toward 50% and to sever the link between the liquidity threshold and the redemption tools. The FCA’s reasoning is explicit: a 40% requirement would not cover the largest plausible sterling outflows, 50% would, and a dash-for-cash event of the 2020 kind would call for 60 to 70%. The effect is to make the buffer both higher and usable, so that a fund can spend it without that act alone triggering the restrictions that begin the run.

The EU’s answer is a binding floor it left untouched at 30%, a non-binding benchmark of 40%, and one genuine reform: the requirement, delivered through the 2024 overhaul of the AIFMD and UCITS directives rather than the MMFR itself, that funds make at least one liquidity-management tool available. That reform is real, and it is the single ESRB recommendation the Commission was graded as having substantially met. It is also the cheapest of the four. The two costly ones, floating the LVNAV and raising binding liquidity, it declined.

The bars are not strictly like-for-like, and the chart’s footnote says so: the regimes define liquid assets differently, and the ESRB’s 45% is a composite that counts public-debt holdings the others treat separately. The ranking survives the caveat. The EU’s binding floor is the lowest of the group, and the one number it added is the only one in the field that no fund is obliged to meet.

The benchmark that binds no one

Return to the 40% benchmark, because it is the centre of the Commission’s actual policy. Two things about it matter.

The first is that it is not binding. It is a supervisory expectation, not a regulatory minimum, so a fund that sits below it has breached nothing and owes no remediation. In the legal sense the benchmark binds no one, which is the sense the title of this piece intends.

The second is where the level sits. Forty per cent is the level the FCA examined and rejected as too low for the largest sterling outflows; both are pure weekly-liquid-asset figures, so that comparison is like-for-like. It is also at or below where the weakest quarter of EU low-volatility funds has held its liquidity through much of the past five years. The 25th percentile ran through the mid-40s for most of 2020 to 2023, touched 40% as early as the third quarter of 2021, and has sat in a narrow band straddling the 40% line since the start of 2024.

On the Commission’s own chart the 25th percentile reaches 39% in the second quarter of 2025, the only quarter in the five years plotted in which it falls below 40% outright. The report’s text says the figure never fell below 40%; the chart a couple of pages on shows 39%. The discrepancy itself is minor. The 25th percentile sits exactly on the 40% line in four other quarters and has stayed within a point or two of it since the start of 2024, so a binding 40% floor would have bound on part of the weakest quartile across much of the recent period rather than in one isolated quarter. A non-binding version of it, pitched at that level, describes where the sector already operates rather more than it disciplines it.

That is the substance of the objection. The ESRB asked for four things in 2021: make the LVNAV’s price fluctuate, repeal the threshold that manufactures the first-mover advantage, raise binding liquidity, and improve the data on who holds what. Against that list, a non-binding number pitched where the weakest quartile already sits falls well short of a reform, which is the reading the ESRB itself recorded when it graded the response.

What the Commission gets right

The honest objection to all of this is that the Commission is not simply wrong, and the argument is stronger for saying so.

The stress in money funds is not spread evenly across the sector. It concentrates in a tail. And the tail is exactly where the coordination game described earlier actually fires. A run needs a correlated investor base — holders who will reach for the exit on the same signal at the same time. Funds with diversified, operational cash investors do not supply one; funds whose investors are a single strategy, or a single type all exposed to the same shock, do.

The 2022 gilt crisis showed the pattern. The median sterling LVNAV saw a one-week net outflow of around 9% — heavy for a cash vehicle, but well inside the liquidity the rules require it to hold, and several funds with diversified cash-management investors took money in over the same weeks. The damage was concentrated in a small set of funds used as collateral vehicles by liability-driven investment strategies, whose investor base was about as correlated as an investor base can be. The worst one per cent lost more than a third of their assets in a single week. The weekly liquidity rules are sized to a week of stress, so the comparison is a fair one: a one-week outflow above a third of assets exceeds the 30% the rules require a fund to hold in total, which means that for those funds the buffer would have been fully spent and the fund still short. March 2020 had the same shape — most LVNAVs saw modest outflows while a small fraction of US dollar funds lost over 30% of their assets in a week. The European sector is not a bystander to this. Its own correlated tail is the US-dollar-denominated prime and LVNAV funds concentrated in Ireland and Luxembourg, held by investors who cluster by currency and mandate; that segment carried the 2020 strain on the European side, and it sits inside the EU rules.

A tail problem is a poor match for a sector-wide instrument. Raising the floor on every fund to reach a concentrated minority charges the prudent majority for resilience they already carry, and leaves the concentration that creates the tail untouched. The Commission’s wariness about a blanket increase has a real argument behind it, and the LVNAV does serve a cash-management function for which, in its words, there is “no obvious economic alternative.”

The trouble is that this argument points to a different policy, not to none. If the danger lives in particular funds with particular investor structures, the response is to find and constrain them: concentration and investor-type limits, granular data on who holds what, and buffers a manager can actually draw down under stress without that act alone starting the run. The first two are close to the monitoring and reporting recommendation the ESRB made and the Commission declined; the third is the core of the UK reform. The case against a blunt floor is a case for sharper tools; the Commission adopted neither.

What the resilience rests on

There is a reason the Commission can read the same record as reassuring where the ESRB reads a deferred problem.

In March 2020, and again in the 2022 gilt episode, the constraint that would otherwise have bound was relieved from outside, though by different routes. In 2020 the ECB’s pandemic asset purchases — and, for the EU’s dollar funds, the Federal Reserve’s facilities — stood behind short-term funding directly, at the moment dealer capacity ran out. In 2022 the Bank of England’s gilt intervention worked a step upstream: buying long-dated gilts broke the margin spiral that was forcing liability-driven investors to sell, which stopped the redemptions draining the sterling money funds. Either way a public balance sheet stood in where private capacity had run out, and no EU fund had to gate. That is the fact the Commission leans on. What the record cannot show is how the funds would have fared without those interventions, because the interventions are the reason the counterfactual never ran.

This is the part the Commission half-states when it lists central-bank facilities among the answers. The resilience it observes is real, but it is conditional, and the condition is not priced. An LVNAV offers its investors something close to a bank deposit — a stable value, settlement like cash — at a yield above a deposit. A bank that offers a deposit pays for the public backstop behind it through deposit-insurance premiums, capital requirements, and supervised access to the central bank’s standing facilities. The LVNAV has been able to rely on public support in stress — less formal than deposit insurance, and discretionary rather than guaranteed — while carrying none of those costs. That specific gap, rather than the general truth that all liquidity transformation ends somewhere at the central bank, is what the report leaves unaddressed.

Providing the LVNAV with that implicit backstop may well be the right policy. It is defensible if it is acknowledged, and if the implied support is either priced or contained by the kind of structural limits the ESRB asked for. It is harder to defend when the support is read as evidence of the funds’ own soundness, because it is discretionary: the willingness to deploy it again, at the same scale, may not be there when it is next needed. That is the gap the ESRB chair pointed to in urging the European Parliament’s economic committee, last December, not to wait for the next crisis to settle the framework. The Commission’s report establishes that the funds met their redemptions. It is quieter on what allowed them to.

Paweł Fiedor — The Macro Prudential View

The views expressed are the author’s own and do not necessarily reflect those of any institution with which the author is or has been affiliated.

Sources: European Commission, Report on the functioning of Regulation (EU) 2017/1131 on money market funds, COM(2026) 350 final, 11 May 2026 — https://ec.europa.eu/finance/docs/law/260511-money-market-funds-report_en.pdf; HM Treasury and Financial Conduct Authority, money market fund reform announcement, 14 May 2026 — https://assets.publishing.service.gov.uk/media/6a059b2622977ebc82cb3f5d/The_Government_and_FCA_announce_plans_to_reform_UK_Money_Market_Fund_Regulations..pdf; European Systemic Risk Board, Recommendation ESRB/2021/9 on the reform of money market funds — https://www.esrb.europa.eu/pub/pdf/recommendations/esrb.recommendation220125_on_reform_of_money_market_funds~30936c5629.en.pdf; European Systemic Risk Board, assessment of compliance with Recommendation ESRB/2021/9, February 2025 — https://www.esrb.europa.eu/pub/pdf/recommendations/esrb.compliancereport202502_1~cfa5aff4bd.en.pdf; Regulation (EU) 2017/1131 on money market funds (MMFR) — https://eur-lex.europa.eu/eli/reg/2017/1131/oj/eng; European Commission, impact assessment accompanying the 2013 money market funds proposal — https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:52013SC0315; Directive (EU) 2024/927 (AIFMD II), amending the AIFMD (2011/61/EU) and the UCITS Directive (2009/65/EC) to require liquidity-management tools — https://eur-lex.europa.eu/eli/dir/2024/927/oj; Financial Conduct Authority, CP23/28, Updating the regime for Money Market Funds — https://www.fca.org.uk/publications/consultation-papers/cp23-28-updating-regime-money-market-funds; U.S. Securities and Exchange Commission, Money Market Fund Reforms, Release No. 33-11211 (2023) — https://www.sec.gov/files/rules/final/2023/33-11211.pdf, and the 2014 institutional prime and municipal floating-NAV reforms — https://www.sec.gov/newsroom/press-releases/2014-143; Financial Stability Board, Holistic Review of the March Market Turmoil, 17 November 2020 — https://www.fsb.org/2020/11/holistic-review-of-the-march-market-turmoil/; Federal Reserve, Money Market Mutual Fund Liquidity Facility and Commercial Paper Funding Facility, March 2020 — https://www.federalreserve.gov/monetarypolicy/mmlf.htm; European Central Bank, Pandemic Emergency Purchase Programme, March 2020 — https://www.ecb.europa.eu/mopo/implement/pepp/html/index.en.html; Bank of England, gilt market operation announcement, 28 September 2022 — https://www.bankofengland.co.uk/news/2022/september/bank-of-england-announces-gilt-market-operation; Bank of England, Financial Stability Report, July 2023 — https://www.bankofengland.co.uk/financial-stability-report/2023/july-2023; European Systemic Risk Board, chair’s remarks to the European Parliament Committee on Economic and Monetary Affairs, December 2025 — https://www.europarl.europa.eu/doceo/document/P-10-2025-004833_EN.html. Data: regime liquidity requirements per the regulatory sources above; the 25th-percentile and cross-sector-average weekly-liquid-asset series read from COM(2026) 350 final, sterling LVNAV redemption data for the 2022 gilt episode — https://ec.europa.eu/finance/docs/law/260511-money-market-funds-report_en.pdf.

On 2 April 2025, the US administration announced a wall of tariffs, and the market did what markets do when the rules of global trade are rewritten overnight. The VIX spiked to 52 — its second-highest intraday peak since the COVID crash. Treasury volatility, measured by the MOVE index, jumped to levels last seen during the 2023 banking crisis. European equity volatility followed. For a few days the screens looked the way they look before a central bank picks up the phone.

And then nothing happened. No emergency facility. No unscheduled meeting. No statement promising to do whatever it takes. The Federal Reserve, which was at that moment still shrinking its balance sheet, carried on shrinking it. The market repriced, found a level, and moved on.

This is the kind of episode that invites a comforting story. The comforting story goes like this: modern markets are electronic, deep, and fast; algorithmic market-makers provide liquidity continuously and reprice risk in milliseconds; we have, through some combination of technology and hard-won experience, built markets that absorb shocks on their own. Two wars and a trade war in four years, and the system held. Perhaps the machines have made us anti-fragile.

I want to take that story seriously, because a version of it is true. But the cleaner reading of the evidence is that it is incomplete about why — and getting the why wrong has real consequences for anyone whose job is to keep the system standing.

The calm is selective

Start with the texture of the last few years, because “markets were calm” is too coarse a claim to be useful.

The point is not that one market was calm and another wild. It is that “the system” is not one thing. Each of these markets has its own microstructure, its own participants, its own plumbing — and they broke, or didn’t, for different reasons. The shock that froze the gas market in 2022 left equities largely untouched; the shock that drove the MOVE index to a banking-crisis high in March 2023 was a banking crisis, not a vol event. If we want to know whether markets have become more resilient, we have to ask the question market by market, mechanism by mechanism — not wave at an index and call it calm.

It is worth holding one distinction in mind from the outset, because the whole argument turns on it. There is a difference between a repricing shock — where asset prices move violently as the market digests new information — and a funding shock, where the problem is not the level of prices but the inability to finance positions at any price. The two can look similar on a volatility chart. They are not similar in the plumbing, and as we will see, that difference is what separates a market that bends from one that breaks.

A decade of rising rescues

Here is the fact that the comforting story has to explain away.

For most of the post-crisis period, the official sector was called in with rising frequency. The chart above plots the major liquidity-event episodes since 2010 — the 2010 Flash Crash, the October 2014 Treasury flash rally, the August 2015 ETF dislocation, the 2016 sterling flash crash, 2018’s “Volmageddon,” the September 2019 repo spike, the March 2020 Treasury dysfunction, the 2022 LME nickel suspension and retroactive trade cancellation, the September 2022 gilt/LDI doom loop, the March 2023 banking failures, and the August 2024 yen carry unwind — and beneath them, the cumulative count of episodes that drew an official response.

The line steepens. Six events in the first decade; five more in the four years after. And the nature of the response escalated alongside the frequency. The early episodes drew rule changes — circuit breakers after the Flash Crash, reporting requirements after 2014, product delistings after Volmageddon. No public money. From 2019 onward, every episode drew the balance sheet: emergency repo operations, unlimited quantitative easing, the Bank of England’s gilt purchases, all deposits backstopped at Silicon Valley Bank, the Swiss authorities’ rescue of Credit Suisse. The system moved from adjusting the rules to deploying the central bank’s own balance sheet. That is a meaningful escalation, and it is the part of the record that should make anyone cautious about declaring victory.

And then — the cumulative line goes flat. After the March 2023 banking crisis, three shocks of comparable scale arrived, and none of them drew a backstop. The yen carry unwind in August 2024, which sent the Nikkei down 12.4% in a single session — its largest one-day fall by points in history — drew nothing more than verbal guidance from the Bank of Japan. Liberation Day in April 2025 drew nothing. The Iran/Hormuz shock in February 2026 drew nothing. Three loud shocks, no balance sheet.

I have classified the yen carry episode as “absorbed” rather than as an intervention, and that is a judgement worth defending openly. The Bank of Japan did say something — its deputy governor signalled that the Bank would not raise rates while markets were unstable. But it committed no public balance sheet, opened no facility, and bought nothing. For a chart whose vertical axis is the deployment of official liquidity, jawboning belongs on the “absorbed” side of the line. A purist could disagree; I have noted it rather than hidden it.

What the central bank had to do — and didn’t

The intervention timeline makes a claim; the cleanest way to test it is to look at the backstop itself.

The chart above shows the weekly change in the Federal Reserve’s Treasury holdings during two windows, on a single shared scale. The left panel is the spring of 2020. The bars climb a wall: the Fed added more than $360 billion to its Treasury portfolio in a single week, and roughly $1.7 trillion over a few months, simply to keep the market clearing. This was not stimulus. It was plumbing — the central bank becoming the buyer of last resort because private dealers could no longer absorb the flood of selling.

The right panel is the spring of 2025, on the same axis. It is almost flat. Through Liberation Day, the largest weekly move is a reduction of about eighteen billion dollars — because the Fed was running quantitative tightening, draining its balance sheet at a deliberate, pre-set pace, and a market shock that would once have hijacked monetary policy did not so much as interrupt the runoff. The Treasury runoff cap had been reduced to $5 billion a month from April 2025; the balance sheet was barely moving by design, and the shock did not change that. (Runoff finally ceased entirely on 1 December 2025, for reasons that had nothing to do with any acute event.)

That contrast is the argument. In 2020, the shock forced the central bank to reverse course violently. In 2025, a comparable shock did not register as a reason to pause a trickle — not because the stress wasn’t real, but because nothing about it required a balance-sheet response.

It is worth pausing on just how deep the 2020 dysfunction ran, because it sets the bar for what “the plumbing breaking” actually looks like. In normal times the New York Fed transacts in a narrow band of liquid, on-the-run Treasury issues — a few dozen distinct securities on a busy day. In the dash-for-cash of March 2020, that number exploded to more than 250 distinct securities in a single day. The Fed was not fine-tuning; it was buying across the entire curve, absorbing the old, illiquid, off-the-run issues that dealers would no longer hold. The breadth of what the central bank had to hold is itself a measure of how completely private intermediation had seized. (This is a 2020-only series — the transaction-level data is released with a two-year lag — so it cannot speak to the recent episodes. It is here as a yardstick for the depth of genuine dysfunction, not as part of the resilience comparison.)

The shape of a market that bent without breaking

So the recent shocks were absorbed. But “absorbed” can mean two very different things, and the distinction matters enormously. It can mean the shock was small. Or it can mean the shock was real and the market took it without seizing. Gas is the cleanest place to see the difference, because Europe has now been hit twice.

The chart above tracks the Dutch TTF gas benchmark — front-month price on top, open interest (the number of outstanding futures contracts, a proxy for how many participants are actually in the market) below. In 2022, the price reached €339 per megawatt-hour, and open interest collapsed — falling 54% from its pre-invasion peak to its trough. This is what a participation crisis looks like: as prices and margin requirements spiralled, leveraged hedgers could no longer fund their positions and were forced out of the market precisely when hedging was most valuable. The market didn’t just get expensive; it got thin, shedding the very participants who give it depth.

By February 2026, gas prices roughly doubled on the Iran/Hormuz shock — a sharp move, peaking near €62. But look at the open interest. It barely flinched. Participation fell only about 23% from what was, by then, an all-time high — a high more than four times the 2022 trough. A larger relative price move, from a vastly higher base of participation, produced less than half the dislocation. The 2026 decline was an orderly de-risking from a record high, not a margin-driven exodus. The market bent. In 2022, it had broken.

I think this is the single most important contrast in the whole picture, and it generalises. The question is never simply “how big was the shock?” It is “did the shock force participants out of the market, or did they choose to step back and then return?” The first is dysfunction. The second is resilience. And the difference between them lives in the plumbing — in margin, in clearing, in who is obligated to keep providing liquidity when it stops being profitable to do so. This is not a parochial commodities concern: when a margin spiral forces hedgers out of a cleared futures market, the strain runs straight through the clearing house to its members, and a commodity dislocation becomes a systemic one — which is precisely the channel the gilt crisis exposed in a different market months later.

Where the dysfunction lived

If resilience lives in the plumbing, it is worth asking where, exactly, the 2022 dysfunction concentrated — because the answer turns out to complicate the obvious “European energy crisis” reading.

The chart above indexes open interest across four gas benchmarks to January 2022. Three of them collapsed together through 2022: the Dutch TTF (continental Europe), the UK’s NBP, and — crucially — Asia’s JKM. The fourth, North American gas, did not move. It sat flat through the entire period while the other three lost between 44% and 60% of their participation.

The obvious story — that this was a European energy crisis — does not survive that fourth line. JKM is an Asian benchmark; if the dysfunction were geographic, it should not have collapsed in lockstep with the European hubs. What the three falling benchmarks share is not geography. It is that they are the LNG-linked contracts, traded by a globally mobile set of participants — utilities, commodity trading houses, LNG portfolio players — who hedge across all of them at once. When a price and margin spiral in one drained the collateral those participants had available, it drained their capacity to hold positions in the others. North American pipeline gas, with its different and more domestic participant base, was insulated.

There is a further pattern worth flagging, though I want to be careful not to overclaim it. The three benchmarks that collapsed all clear at the same central counterparty — ICE Clear Europe — while North American gas clears elsewhere. That is consistent with a story in which procyclical margin models at a shared clearing house amplified the contraction across all the products it cleared. But it is also consistent with the simpler explanation that these markets just share participants. The data here cannot separate the two, and I would treat the clearing overlap as an intriguing pattern that deserves investigation rather than a settled mechanism. Either way, the lesson is the same: the dysfunction respected the boundaries of participants and plumbing, and ignored national ones — which is exactly why a venue-by-venue regulatory perimeter can miss it.

Two kinds of liquidity

We can now name the thing the comforting story gets wrong. It assumes that the liquidity you see in calm markets is the liquidity you will have in a crisis. It is not.

There are, roughly, two kinds of people providing liquidity in a modern market. The first are committed dealers — banks, intermediating on their own balance sheets, backed by capital and bound by obligation and regulation. In calm markets they are present but capital-constrained; they do not provide the tightest prices. In stress they retrench and widen their quotes, but they stay in the book. They degrade gracefully.

The second are what the microstructure literature calls endogenous market-makers — the principal trading firms and high-frequency market-makers whose capital is uncommitted. In calm markets they are abundant: they provide the tightest spreads and the deepest top-of-book, and they are the reason trading has never looked cheaper or smoother. But their capital is uncommitted by definition. In stress, they withdraw — fast — and the depth that looked so reassuring evaporates exactly when it is needed. This is the cell where systemic liquidity risk concentrates: the flash crashes, the 2014 Treasury flash rally, the dash-for-cash of March 2020.

The structural shift over the last fifteen years has been toward the second kind of provider. Markets in calm conditions are deeper and cheaper than they have ever been precisely because uncommitted, fast, electronic capital dominates the order book. But that same shift means that the liquidity is more fragile in stress — more of it can simply leave. The machines did not make markets robust. They made calm-state liquidity abundant and stress-state liquidity flightier. Those are not the same thing, and macroprudential policy cares about the second one.

So what absorbed the last three shocks?

The tectonics underneath

If it was not the microstructure — if anything, the microstructure got more fragile — then the resilience has to have come from somewhere else. The honest answer is that it came from the slow, unglamorous accumulation of reforms layered on top of that fragile microstructure, compensating for it.

I want to resist the temptation to crown a single reform as the cause, because the evidence does not support it and the truth is almost certainly a composite. But the strongest candidates are visible. Central counterparties have been pushed toward less procyclical margin models, so that a price spike is less likely to trigger the kind of self-reinforcing margin call that drove hedgers out of gas in 2022. The gilt crisis prompted liquidity-buffer requirements for liability-driven investment funds, so that the specific doom loop of 2022 is harder to set off. And the Federal Reserve’s Standing Repo Facility — established in 2021 as a backstop for funding markets — has quietly matured into a more usable tool, with daily morning operations added in June 2025 and, by December 2025, its aggregate cap removed in favour of full allotment. It is telling that the facility was actually used — it lent over $10 billion in a single day at the end of June 2025, its largest draw since inception — quietly, without drama, which is rather the point. A backstop that works at the routine, technical level is one that may never need to be deployed at the dramatic level.

These are the shifting tectonic plates of the regulatory landscape, not a single silver bullet. And not all of the adjustment was regulatory: some of it was behavioural. The desks that lived through the margin spirals of 2022 and the funding scramble of 2020 now manage collateral and position-sizing differently, whether or not a rule required them to — institutional memory is itself a buffer, if a perishable one. But one prominent reform is conspicuously not yet among the formal causes: the SEC’s Treasury central-clearing mandate, which will require central clearing of cash Treasuries from the end of 2026 and repo from mid-2027, has not yet taken effect. Whatever resilience we have seen in 2024–2026, it cannot be credited to a rule that is still prospective. That matters, because it means the most significant structural reform to the largest market in the world is still ahead of us — and untested.

The test that hasn’t come

Here is where I have to be honest about the limits of the argument, because the temptation to over-read three quiet shocks is strong and the cost of giving in to it is high.

The three shocks that were absorbed without intervention — yen carry, Liberation Day, Iran/Hormuz — were all, fundamentally, the repricing shocks I flagged at the outset. Asset prices moved violently as the market digested new information about tariffs, rates, or geopolitical risk. They were not funding shocks. Nobody’s collateral became unfundable; no clearing member faced a margin call it could not meet; the repo market did not seize. The dysfunction of 2020 and the gilt spiral of 2022 were different in kind: they were crises of funding and collateral, where the problem was not the level of prices but the inability to finance positions at any price.

The reforms I have described are aimed precisely at funding and collateral fragility — at margin procyclicality, at liquidity buffers, at the availability of a funding backstop. But they have not yet been tested by the kind of shock they were designed to withstand. A repricing shock that the market takes in stride is encouraging. It is not proof that the next funding shock will be absorbed the way the last three repricing shocks were. The true test of the reinforced plumbing is a stress that hits the plumbing directly, and we have not had one since the reforms went in.

So the cleaner reading of the evidence is this. Markets did not become resilient because the machines learned to provide liquidity in a crisis — they did not, and the structural shift toward uncommitted liquidity provision means the order book is in some ways more fragile than before. Markets coped because a decade of dysfunction produced a slow accumulation of reforms that compensated for that fragility, and because the recent shocks happened to be of the type those reforms address. That is a real achievement, and the people who built those reforms deserve more credit than the algorithms are getting. But it is a conditional achievement, resting on a structure of buffers that has to hold, and on a particular kind of shock continuing to be the kind we get.

The market learned to bend. Whether it has learned not to break, we will only find out when something pushes on the part that bent last time.

Paweł Fiedor — The Macro Prudential View

The views expressed are the author’s own and do not necessarily reflect those of any institution with which the author is or has been affiliated.

Sources: SEC & CFTC, Findings Regarding the Market Events of May 6, 2010 — https://www.sec.gov/news/studies/2010/marketevents-report.pdf; Joint Staff Report, The U.S. Treasury Market on October 15, 2014 — https://home.treasury.gov/system/files/276/joint-staff-report-the-us-treasury-market-on-10-15-2014.pdf; Federal Reserve, FOMC statement of 23 March 2020 — https://www.federalreserve.gov/newsevents/pressreleases/monetary20200323a.htm; Office of Financial Research, The London Metal Exchange Nickel Market Crisis (WP 24-09) — https://www.financialresearch.gov/working-papers/files/OFRwp-24-09_central-clearing-and-trade-cancellation.pdf; Bank of England, gilt market operation announcement, 28 September 2022 — https://www.bankofengland.co.uk/news/2022/september/bank-of-england-announces-gilt-market-operation; Swiss National Bank, Credit Suisse liquidity assistance statement, 19 March 2023 — https://www.snb.ch/en/mmr/reference/pre_20230319/source/pre_20230319.en.pdf; Bank of Japan, Deputy Governor Uchida speech, 7 August 2024 — https://www.boj.or.jp/en/about/press/koen_2024/ko240807a1.htm; Federal Reserve, FOMC statement of 19 March 2025 (QT taper) — https://www.federalreserve.gov/newsevents/pressreleases/monetary20250319a.htm; Federal Reserve, FOMC statement of 29 October 2025 (end of runoff) — https://www.federalreserve.gov/newsevents/pressreleases/monetary20251029a.htm; Federal Reserve Bank of New York, Standing Repo Facility operating policy, 28 May 2025 and 10 December 2025 — https://www.newyorkfed.org/markets/opolicy/operating_policy_250528 and https://www.newyorkfed.org/markets/opolicy/operating_policy_251210; U.S. Securities and Exchange Commission, Treasury Clearing implementation — https://www.sec.gov/securities-topics/treasury-clearing; Bank of England, Financial Stability Report, July 2023 (LDI) — https://www.bankofengland.co.uk/financial-stability-report/2023/july-2023. Volatility data: CBOE (VIX), ICE BofA (MOVE), Qontigo/STOXX (VSTOXX). Gas price and open-interest data: ICE Endex (TTF) via Investing.com; ICE Global Natural Gas Futures (TTF, NBP, JKM, North American hubs). Fed balance-sheet data: Federal Reserve H.4.1 via FRED (TREAST). CUSIP breadth: Federal Reserve Bank of New York, Outright Purchases and Sales of Treasury Securities.

PIK is not an anomaly

Imagine a sponsor-backed mid-market borrower. The original 2022 senior secured term loan was clean: cash interest paid quarterly, fixed and floating spreads, standard maintenance covenants. By late 2024 the borrower’s cash flow has deteriorated. Rather than mark the loan, restructure it, or trigger a covenant breach, the lender and the sponsor agree to convert a portion of the interest from cash-pay to PIK — payment-in-kind, where the interest doesn’t actually leave the borrower as cash but accrues to the principal balance instead. The headline default rate is unaffected. The loan continues to be marked at or near par. The BDC holding the loan reports interest income, including the PIK component, in its quarterly results. The cash distribution to LPs goes out on schedule, funded partly by other loans and partly, increasingly, by new equity raises or draws on the fund’s credit lines.

No one has crystallised a loss. The borrower’s principal is growing. The reported numbers look stable, for now. Because valuations in private credit are GP-determined and infrequently tested against arms-length markets, the “for now” can persist much longer than common sense expects.

This is not a small phenomenon. Based on my own aggregation of the top fifteen exchange-traded BDCs’ EDGAR filings, PIK income has risen from roughly 4% of total interest income at end-2022 to about 6% by Q1 2026 on a four-quarter rolling basis. The more striking shift is where PIK now sits: at Ares Capital Corporation, the largest exchange-traded BDC and one of the most conservatively-run, the first-lien share of PIK-paying investments (measured by fair value) has risen from 60% in 2021 to 76% by year-end 2025. PIK was historically a feature of distressed mezzanine and second-lien paper. It now sits in senior secured paper at the most benchmark-quality BDC in the market.

The FSB’s Report on Vulnerabilities in Private Credit, published in May 2026 after a year-long investigation, made the data limitations explicit. The report states that private credit “remains untested in a prolonged economic downturn” and that authorities face “significant data challenges” in monitoring the bank-fund interconnection. It also flagged “the rising use of payment-in-kind arrangements” as one indicator of deteriorating credit conditions. After a year of work involving every major macroprudential authority in the developed world, the official sector is still working around structural data blind spots — unable to map, with anything approaching precision, how much PIK is outstanding, who holds it, or what these valuations would survive in a real downturn.

This is not a small market. The FSB estimates the global private credit asset class at $1.5–2 trillion, with direct bank credit lines to private credit funds running in the hundreds of billions. The interconnection is large enough that the data gaps matter.

I want to be careful about what kind of claim I am making here. PIK is not a fraud. It is not even, on any individual loan, a problem. PIK toggles have a legitimate role in financing structures where cash flows are lumpy or where a sponsor needs flexibility through a specific investment phase. What is interesting is not any single PIK toggle. What is interesting is the equilibrium — the fact that every participant in the capital structure has private incentives pointing the same way.

That equilibrium is the subject of this essay. Private credit’s PIK creep is the contemporary illustration, but the underlying pattern is much older. It is, I will argue, the central problem in financial stability — and the one that macroprudential policy, for reasons that have to do with how the toolkit was designed rather than with anyone’s incompetence, is least well-equipped to address.

Why the equilibrium holds

Walk the stack of incentives, one participant at a time.

The borrower wants survival. Cash interest converted to PIK is, for a stressed borrower, the difference between a default and a non-default. Even if the principal is compounding at 12% — and that is roughly where current PIK rates sit — the alternative is worse. The borrower is not making this trade strategically; it is making it under duress.

The sponsor’s interest aligns. Injecting fresh equity into a stressed portfolio company crystallises a down-round, recalibrates the fund’s mark on the position, and hits the sponsor’s IRR. A PIK arrangement protects all three. It is the sponsor’s own version of smoothing, and the GP-LP interests align on this — neither party benefits from a marked-down NAV at this stage of the fund’s life.

The General Partner of the credit fund holds the loan at fair value, which is its own determination, audited but not market-priced. As long as the loan continues to pay interest (whether in cash or in kind) and as long as no covenant trips, the GP can defend a near-par mark. The GP earns management fees on assets under management and performance fees on returns above a hurdle. Both are smoother if losses don’t crystallise. Beyond the direct fee economics, the GP is in a competitive market for institutional capital where marked-down NAVs become a fundraising problem within months. The incentive to extend rather than mark is structural.

The Limited Partner — pension fund, insurance company, sovereign wealth fund — receives a cash distribution on the fund’s reporting schedule, and the distribution does not distinguish between cash interest received from healthy borrowers and PIK income accrued from stressed ones. The LP’s allocators are evaluated on relative performance against a benchmark, and the benchmark is constructed from the same fund universe with the same valuation conventions. There is no LP that has a strong unilateral incentive to demand a fair-value crystallisation.

For the BDC structure specifically, the regulatory framework adds an additional twist. US Regulated Investment Company (RIC) tax status requires the BDC to distribute at least 90% of its taxable income annually to maintain that status. That taxable income includes PIK. The BDC therefore must pay cash dividends on income it did not receive in cash. The cash for those dividends comes from other loans, from credit-line draws, from new equity raises, and — as the PIK share grows — from increasingly complex funding manoeuvres to sustain the dividend policy. iCapital argues that there is natural pressure on BDCs to keep PIK income below roughly 10% of total interest income, because above that level cash-distribution stress becomes structural. Prospect Capital, one of the more PIK-heavy listed BDCs, operates at roughly one-third of net investment income based on its filings — well past the threshold.

The mechanical consequence deserves to be stated plainly. When a BDC raises new equity, draws on a subscription line, or rolls a credit facility to fund a cash distribution against PIK income, real cash is leaving the system. The fund is burning actual liquidity — investor capital, bank lending capacity, dividend reserves — to sustain the appearance of a smooth, non-defaulting portfolio. The equilibrium is not a free lunch. It is a slow drawdown of liquid resources to defer the recognition of a credit problem, financed by the rotation of new capital in to pay out old.

The bank — and there is usually a bank somewhere in the structure — earns commitment fees on the subscription line or NAV-backed facility extended to the fund (typically collateralised by fund NAV and short-dated), and avoids any direct credit exposure to the underlying borrower. The bank’s balance sheet looks safer than it would have if the bank had made the loan directly. The economic exposure is largely the same, but it sits in a structure that doesn’t show up on the bank’s lending book or in its regulatory disclosures.

The point of walking through the stack is that nothing in it is wrong on its own terms. Every participant is responding rationally to its own incentive structure. The contractual fix — converting a cash-flow problem into a balance-sheet item — works at the level of each individual contract. What fails is the aggregation. The system as a whole is taking on credit risk that is not visible to any single participant, including the prudential supervisor.

This is often called “extend and pretend,” and it is recognised, in retrospect, as a feature of essentially every credit cycle for at least four decades.

The same machine, different fuel

In my reading of the post-1980 history of financial regulation, one of the most consistent patterns is not crisis followed by reform. It is innovation followed by smoothing, where smoothing means the introduction of some contractual or structural innovation that converts a cash-flow or credit problem into something that doesn’t trigger immediate recognition. Every cycle generates a new technology for doing this.

US savings and loans in the 1980s used regulatory accounting principles — RAP, distinct from GAAP — to defer recognition of losses on residential mortgages whose market value had collapsed when interest rates spiked. The RAP allowed S&Ls to amortise losses over the remaining life of the loan rather than mark them at issuance. By the time the system was forced to recognise the underlying insolvency, the bill was several hundred billion dollars. In each case I describe in this section, the binding choice was not whether losses existed, but when — and in which set of accounts — they would be recognised.

Mortgage securitisation in the 1990s and 2000s separated the originator of a loan from the bearer of its credit risk. The originator earned a fee on issuance and passed the cash flows to investors via structured vehicles whose ratings depended on internal models that the rating agencies themselves did not fully understand. The CDO bid in 2005–2007 produced a market in which the demand for AAA-rated mortgage paper exceeded the underlying supply of AAA-quality mortgages, with the difference made up by financial engineering. The losses crystallised in 2008.

Leveraged loans and CLOs in the 2010s migrated the same originate-and-distribute model into the corporate credit space, with two refinements. First, covenant-lite documentation removed most of the maintenance triggers that historically forced workouts before losses became catastrophic. Second, the CLO buyer base — primarily insurance companies and bank trading desks searching for AAA-rated paper — provided structurally inelastic demand at the top of the capital structure. By the early 2020s, over 85–90% of new leveraged loans were covenant-lite, and the global CLO market had grown to around $1–1.4 trillion.

Private credit and PIK in the 2020s are, in my read, the next iteration — the same equilibrium I described earlier, refracted through different financial-engineering vocabulary. The contractual innovation is the PIK arrangement; the structural innovation is the move from public bond markets to private fund vehicles where valuations are GP-determined and disclosure is limited to LPs. Each generation looks technical and specific enough that the previous generation’s lessons “don’t apply.” Each generation is the same equilibrium — converting cash-flow stress into something that doesn’t have to be recognised today, because everyone in the stack is happier that way.

This is not the only path innovation can take. Plain-vanilla interest-rate swaps, simple pass-through securitisations, and CCP-cleared derivatives have all become standard infrastructure without obvious crisis episodes. The pattern I am identifying applies to the subset of innovation that combines leverage, opacity, and contractual smoothing of credit stress — which is, on the evidence of the last forty years, the consequential subset.

What is common across the consequential cases is not a failure of regulation in the narrow sense. Capital requirements existed in 1985, in 2005, in 2015, in 2025. They have generally gotten tighter over time, not looser. What is common is that the regulation operates on quantities — how much capital, how much liquidity, how concentrated the exposure — while the innovation operates on incentives — who recognises losses when, who captures the upside, who bears the downside if the bet doesn’t work. The architecture of post-2008 macroprudential policy was built to manage quantities. It works for that. It does not, by its own design, address the contractual and structural margins where the next equilibrium is being built.

This is the central argument of this essay. The rest is detail.

What macroprudential policy actually does

Before going further it is worth being precise about what the post-2008 macroprudential toolkit consists of, because the argument is not that macroprudential policy is bad — it is that the toolkit’s design defines its perimeter, and that perimeter is largely orthogonal to the incentive structures that drive the kind of equilibrium described above.

The bank-side toolkit, organised by category, has six main families. Capital tools — the Capital Conservation Buffer, the Countercyclical Capital Buffer, the Systemic Risk Buffer, the G-SII and O-SII buffers — together raise the equity-to-assets ratio of regulated banks above the Basel III minimum, in some jurisdictions substantially. Risk-weight constraints — output floors, IRB restrictions, sectoral risk-weight floors — limit the extent to which banks can use internal models to reduce the denominator of their capital ratio. The leverage ratio provides a non-risk-based backstop against the gaming of risk weights, and in Europe has been binding for a subset of large institutions since CRR2 took effect in 2018. Liquidity tools — the Liquidity Coverage Ratio and the Net Stable Funding Ratio — require banks to hold sufficient high-quality liquid assets to survive a 30-day stress scenario and to maintain a stable funding profile over a one-year horizon. Large exposure limits and concentration rules cap the credit risk a bank can take to a single counterparty or sector. Borrower-based measures — LTV, DTI, DSTI, maturity limits on mortgages — operate on the demand side of the lending market.

Beyond banks, the macroprudential perimeter has expanded into non-bank financial intermediation, with three main vectors. AIFMD Article 25 gives national authorities the power to impose leverage limits on alternative investment funds. EMIR margin and CCP resilience requirements address counterparty risk in derivatives markets. The MMF reform agenda restructured the prudential treatment of money market funds after the 2008 and 2020 runs.

Every item in this toolkit operates on a quantity: amount of capital, ratio of capital to assets, share of liquid assets in the balance sheet, size of exposure to a single counterparty, leverage of a fund vehicle. None of them operates directly on a contract — on how interest is structured, how covenants are written, how valuations are determined, how compensation is linked to outcomes. (Supervisors do issue expectations on underwriting standards through guidelines and SREP letters, but these are not primary macroprudential tools, and they operate at the institution level rather than at the market level.) Bail-in and MREL come closest to touching incentives, in that they were explicitly designed to reset bondholder expectations about loss-absorption capacity, and CRD V’s remuneration rules — deferral, malus, clawback — reach into compensation structure. But as we will see, the latter has been substantially arbitraged and the former applied unevenly. SREP capital add-ons under Pillar 2 give supervisors institution-specific discretion, but the discretion is exercised through the same quantity-based instruments as the rest of the toolkit.

There is a defensible objection at this point: that capital and MREL do reach incentives indirectly, through the cost-of-capital channel. Higher required capital makes risk-taking more expensive at the margin; higher MREL makes bail-in-eligible debt price the risk of loss; rising buffers compound through retained earnings into discount rates. This is true. But what the price channel adjusts is the price of risk-taking. It does not adjust the contractual terms through which risk is masked. A bank facing higher capital costs may demand a higher spread on a leveraged loan; it does not, through the price channel, get an instrument to constrain whether that loan can convert from cash-pay to PIK once made. The price channel changes the marginal economics of activities the framework can see. It does not bring activities the framework cannot see into the framework.

This is not a critique of the toolkit. The macroprudential framework was designed to address the failures most clearly diagnosed in the aftermath of 2008 — banks were undercapitalised, illiquid, and concentrated. The post-2008 reforms made banks better capitalised, more liquid, and less concentrated. The empirical literature on the effectiveness of macroprudential tools — Galati and Moessner’s BIS survey, the IMF’s iMaPP studies, the ECB’s Macroprudential Bulletin series — generally finds that the tools work for what they were designed to do.

The toolkit has also expanded. AIFMD Article 25 leverage limits, MMF structural reform, CRD V remuneration rules, and SSM Pillar 2 discretion all reach further than the framework did in 2010. Yet these expansions remain largely quantity-based, or — in the case of remuneration — have been substantially arbitraged, as the bonus-cap evidence later in this piece will show.

The argument is structural, not evaluative. The toolkit’s design constrains its reach. The kind of equilibrium described above — where PIK creep happens because every participant has private incentives pointing the same way — operates at margins the toolkit does not directly touch. The regulatory architecture has the tools to constrain how much bank capital sits behind a leveraged loan, but no comparable tools to constrain how that loan is contractually structured once made.

The deeper question, and the subject of the rest of this essay, is whether macroprudential authorities can — or should — develop tools that reach the contractual and incentive margins directly. The right place to start is where the post-2008 reform agenda itself started: the capital framework.

Capital alone hasn’t worked, but not for the reason most people think

After 2008, the regulatory response was dominated by capital. Basel III roughly tripled the minimum CET1 requirement; CRD IV and CRR layered the Capital Conservation Buffer, Countercyclical Capital Buffer, Systemic Risk Buffer, and G-SII/O-SII surcharges on top. The trajectory has been almost monotonically upward: average CET1 ratios for large international banks have risen from approximately 7% pre-crisis to comfortably above 14% by 2024. By any reasonable measure, large internationally-active banks are better capitalised today than at any point since the 1980s.

This was the right starting point. The 2008 crisis revealed that major banks were running with equity ratios of 2–3% of unweighted assets, supporting maturity transformation and credit creation that depended on continuous wholesale funding access. When that funding froze, the banks did not have the equity buffer to absorb mark-to-market losses while remaining solvent. The capital response is, in retrospect, the clearest success of the post-2008 reform agenda.

But it has limits. Two of them deserve attention.

The first limit is the one Haldane diagnosed in his 2012 “Dog and the Frisbee” speech. Regulatory complexity, beyond a certain point, becomes a tool of gaming rather than a constraint on it. The more granular the risk-weighting framework, the more parameters available for a sophisticated bank to optimise against. The internal ratings-based (IRB) approach was, in its design, supposed to align capital requirements more tightly with actual risk. In implementation, it produced significant divergence between banks holding similar portfolios — divergence that the EBA’s IRB benchmarking exercises have documented year after year. The output floor introduced under Basel IV is the multilateral response, but a partial one: it limits how far banks can use internal models to reduce risk weights, without addressing the underlying complexity that makes the gaming possible. Tarullo, in his April 2017 “Departing Thoughts,” made adjacent points from the regulatory side: the framework had become too complex for supervisors themselves to verify in real time.

The second limit is more fundamental and gets less attention. Capital regulation operates on the denominator of the loss-absorbing layer. Increase capital requirements, and equity holders bear more of any given loss. Useful — but it does not change who decides whether to take the loss-generating risk in the first place. That decision is made by management, on the basis of an incentive structure that allocates the upside of risk-taking to equity (via options, restricted shares, performance metrics that scale with ROE) and the downside to depositors, bondholders, and ultimately the sovereign. Capital regulation does not touch that asymmetry directly. It just changes the size of the buffer between management’s incentive and the taxpayer.

Admati and Hellwig pushed hard on the level of capital in The Bankers’ New Clothes (2024 edition) — arguing that bank capital should be substantially higher than current Basel standards, and that the conventional industry case against higher capital is built on flawed arguments about funding costs. Their analysis is rigorous, and the response from the official sector and from academic critics has, in my read, mostly conceded the basic point while disagreeing on the operational margin. Their argument is about level. The argument I make next — the Bebchuk-Spamann argument — is about structure.

The same logic, refracted, applies to private credit. The capital that sits behind a bank’s subscription line to a private credit fund is regulatory-light because the line is structurally senior, collateralised by fund NAV, and short-duration. That is a quantity argument that holds up under standard regulatory analysis. But the bank’s decision to extend the line, and the fund’s decision to use it for cash dividends to LPs rather than for borrower workout, is driven by incentive structures that no quantity tool reaches.

The incentive mechanism: a call option on bank assets

The cleanest formal statement of the bank-comp problem is Bebchuk and Spamann’s 2010 Georgetown Law Journal article, “Regulating Bankers’ Pay.” Their argument has a single moving part. A bank’s capital structure is overwhelmingly debt-financed: equity is typically 5–7% of assets, with deposits (50–60% of liabilities) and various forms of senior and subordinated debt filling the rest. Executive compensation is heavily linked to equity returns — through restricted stock, performance share units, ROE-based bonuses. The structure of the executive’s claim, then, looks like a long call option on the bank’s assets: substantial upside if asset values rise (equity is the residual claimant), capped downside at zero (because the executive does not personally fund losses below their own equity holdings, which are typically a fraction of one year’s compensation).

The mechanics are straightforward. With a thin equity layer at the top of a heavily-levered capital stack, small changes in asset values translate into large changes in equity value. A 1% increase in asset values, on a balance sheet where equity is 5% of assets, is a 20% increase in equity. Conversely, a 1% decline in asset values wipes out 20% of equity — but the executive’s exposure is asymmetric, because their pay is linked to equity returns when those returns are positive (vesting of options, performance bonuses paid in good years), while the executive does not personally absorb negative returns beyond their own holdings. The result is an executive payoff profile that is convex in asset volatility: the executive is, in option theory terms, long vega.

This is what the chart above shows. Compensation tracks equity at the top of the stack; losses cascade downward through layers that share none of the upside. The depositor at the bottom of the cap structure is the residual loss-absorber, with explicit protection via deposit insurance schemes funded by the rest of the banking system and, in extremis, the sovereign. The implicit taxpayer backstop at the very bottom is not a flourish. It is what makes the option valuable. Without an implicit floor, the executive’s claim would not be a call — it would be a position in a residual claim with genuinely unbounded downside, and the optionality that drives the compensation argument would disappear.

The empirical evidence is consistent with what theory predicts. Cheng, Hong, and Scheinkman, in their 2015 Journal of Finance paper, showed that US banks with higher residual pay (the unexplained portion of total executive compensation) were more likely to fail in 2008, controlling for size, leverage, and asset composition. Bebchuk, Cohen, and Spamann had earlier documented, in The Wages of Failure (Yale Journal on Regulation, 2010), that the senior executives of Bear Stearns and Lehman Brothers extracted some $2.4 billion in cash bonuses and equity sales between 2000 and 2008 — money that did not flow back to creditors or to the FDIC when the firms failed. The compensation system, in their phrasing, was structured so that executives “made money during the build-up of risk that crashed their firms.” Cerasi, Deininger, Gambacorta, and Oliviero (BIS Working Paper 630, 2017) found that across a large international panel, banks with more equity-linked CEO compensation had higher CDS spreads and higher Z-score volatility — consistent with the option-like payoff inducing more risk.

The structural question this poses for regulation is whether the asymmetry can be addressed by any policy tool other than direct intervention in compensation structure. The Bebchuk-Spamann answer is essentially no. Capital requirements reduce the option’s intrinsic value but do not eliminate the convexity. Bail-in shifts more loss-absorption onto debt holders but does not change executive incentives. Deferred compensation rules push the payoff out in time but do not address its option-like nature. The only structural fix is to change what executives are paid with — to give them claims that share in the downside, not just the upside.

The parallel to today’s PIK structures is not hard to draw. The GP of a private credit fund has a similar option-like payoff: 20% carried interest above a hurdle (the upside), management fees on AUM (a base layer that does not decline with NAV losses below a certain band), and limited personal capital at risk in the fund. The compensation asymmetry is structurally identical to the bank-executive case Bebchuk and Spamann formalised. Only the capital-structure terminology differs.

What the EU has actually tried

The European post-crisis architecture has gone further than the US on incentives. This is not a partisan claim — it is the empirical record of legislation and implementation.

The BRRD (Bank Recovery and Resolution Directive, 2014) and SRMR (Single Resolution Mechanism Regulation, 2014) established a framework that placed loss-absorption onto a hierarchy of bondholders before any taxpayer support could be deployed. Article 18 of the SRMR specifies the three tests that must be met for SRB-led resolution: (a) the bank is failing or likely to fail; (b) no private or supervisory alternative exists; and (c) resolution is in the public interest. The framework was designed to make TBTF support politically and operationally costly enough that managers and creditors would price the bail-in risk into their decisions.

MREL (Minimum Requirement for Own Funds and Eligible Liabilities) operationalised the framework: each significant bank must hold a portfolio of bail-in-eligible debt sufficient to cover potential losses and a recapitalisation amount. The SRB’s MREL Dashboard reports that the banks under its mandate had built up MREL stocks averaging close to 28% of risk-weighted assets by end-2024 (SRB MREL Dashboard, end-2024) — well above pre-crisis subordinated debt levels.

CRD IV’s remuneration framework went further. From 2014, variable pay for identified staff was capped at 100% of fixed pay, extendable to 200% with shareholder vote. Deferral periods of 4–5 years became mandatory, with malus and clawback provisions for risk-management failures discovered after vesting. EBA Guidelines (EBA/GL/2021/04) provide the implementation detail. The CRD V update (2019) tightened the regime further on smaller and mid-sized institutions, addressing earlier proportionality exemptions.

ECB-led supervision via the SSM, established in 2014, provided centralised micro-prudential oversight of significant institutions, with SREP capital add-ons (Pillar 2 Requirement and Pillar 2 Guidance) giving the supervisor institution-specific discretion to layer capital requirements on top of Pillar 1.

This is, on paper, a more ambitious regulatory architecture than the US framework. Dodd-Frank’s bail-in framework (Orderly Liquidation Authority) has never been invoked since enactment in 2010, and is subject to 2018 Treasury guidance that narrowed the practical path to OLA, emphasising bankruptcy as the preferred route. The US compensation rules under Dodd-Frank Section 956, which would have imposed deferral and clawback requirements on incentive compensation at large banks, were proposed in 2011 and again in 2016 and have still not been finalised and implemented, despite being re-proposed in 2024. The Volcker Rule, while implemented, was substantially weakened in the 2020 amendments that loosened its application to proprietary trading and merchant banking.

What Europe attempted, the US did not try — or tried and abandoned. This is worth saying directly, because it sets up the question the next section takes on: where the European architecture has been tested in real cases, what did the results actually show?

Two natural experiments: bail-in and the bonus cap

Two events in 2017 provide the cleanest natural experiment we have on bail-in. Both involved failing banks meeting Article 18(1)(a) and (b) of the SRMR — failing or likely to fail, with no private alternative. Both required the SRB to decide whether to invoke resolution under (1)(c)’s public interest test. The two outcomes diverged completely.

Banco Popular was declared failing or likely to fail by the ECB on 6 June 2017, after a deposit run and an ELA collateral shortfall that left the bank unable to reopen on 7 June. The SRB triggered resolution within hours, applied the bail-in framework, and sold the bank to Santander for €1 on the morning of 7 June. The cap structure was applied as designed: shareholders 100% wiped, AT1 written down to zero, Tier 2 converted to equity which was then sold to Santander for €1, depositors preserved. Direct taxpayer cost: €0.

Veneto Banca and Banca Popolare di Vicenza were declared failing or likely to fail by the ECB on 23 June 2017, with the SRB issuing a same-day determination that resolution was not in the public interest under Article 18(1)(c). The reasoning given was that the banks’ functions were “not critical,” interconnectedness was “low,” and Italian insolvency proceedings would achieve resolution objectives to the same extent. The banks were placed into national liquidation. The European Commission approved Italian state aid for the resolution under Commission Decision SA.45664 within 48 hours.

The state aid package was substantial. The Commission decision authorised cash contributions to Intesa Sanpaolo of €4.785 billion (€3.5bn cash, €1.285bn restructuring lump sum), plus state guarantees of up to ~€12 billion on the funding gap and NPL transfer. Subsequent retail-bondholder compensation through the Fondo Indennizzo Risparmiatori, established by Italian Budget Law 145/2018, added another €1.575 billion over 2019–2021.

The reading I take from these two events — and others may read them differently — is that the bail-in framework works when it works, and the question is when it works. The Article 18(1)(c) public-interest test is the discretionary margin. In the Italian case, the public-interest test was applied in a way that allowed a national-law liquidation with state aid, effectively bypassing the SRB’s bail-in regime and shifting losses back to the sovereign balance sheet. The framework is the same in both cases; the application diverged on a single discretionary judgment.

This matters for incentives because what bondholders price is expected loss-absorption, not legal loss-absorption. If the expectation is that medium-sized banks in politically-sensitive jurisdictions can secure exemptions, the disciplining effect of the framework on bondholder behaviour is correspondingly diluted. This is not a critique of the framework or of the people applying it. It is an observation that the framework is structured to permit discretion at exactly the point where the structural-reform alternatives — which we will come to shortly — are structured to remove it. The CMDI (Crisis Management and Deposit Insurance) proposal of April 2023 is, in effect, an admission that the framework as it stands does not systematically work for mid-sized banks, and an attempt to plug the gap by expanding DGS use and resolution scope.

The bonus cap provides a different kind of natural experiment. EBA’s annual benchmarking of high-earner remuneration in EU banks reports the data needed for a clean comparison: 2013 (the last year before the cap, since CRD IV took effect for performance year 2014) and 2014 (the first year of binding cap).

The headline numbers are unambiguous. The average total compensation of EU high earners (≥ €1 million total compensation) was essentially unchanged: €1.89m in 2013, €1.91m in 2014. The composition flipped: average fixed pay rose from €454k to €840k (+85%), while average variable pay fell from €1.44m to €1.07m (−26%). The variable/fixed ratio collapsed from 317% to 127% — exactly as the cap mechanically required. The sample of EU high earners grew from 3,178 to 3,865 over the period as reporting harmonised, so the per-individual averages are the right metric for isolating the composition effect from sample growth.

Colonnello, Koetter and Wagner formalised the empirical pattern in their 2023 Journal of Accounting and Economics paper, using difference-in-differences identification on a larger panel of European banks. They documented what the literature calls the indemnification effect: banks responded to the variable-pay constraint by raising base salaries to maintain total compensation, with the level of total pay roughly preserved across the cap transition.

The economic interpretation matters. The cap was designed on the theory that variable pay drives risk-taking, and reducing variable pay relative to fixed pay should reduce risk incentives. The mechanical response of banks was to raise base salaries to maintain total compensation. This is exactly what the Bebchuk-Spamann framework would predict: if executive talent is competitively priced in a labour market, a constraint on one component of compensation will be offset by adjustments in others.

There is a second, less-discussed consequence worth flagging. By converting variable pay (which is cuttable in stress) into fixed pay (which is structurally rigid), the cap raised the operating leverage of EU banks’ compensation cost base. In a systemic downturn, variable pay can be reduced quickly to preserve capital; fixed salaries cannot. The cap, on this reading, did not just fail to discipline risk-taking — it made bank cost structures less flexible in exactly the scenario the regulator most cares about.

The UK provides the other end of the experiment. In October 2023, the PRA and FCA jointly removed the bonus cap entirely (PS9/23), citing exactly the arbitrage argument: the cap had pushed compensation toward fixed pay without reducing total compensation, and fixed pay is less responsive to performance — meaning the cap may have actually weakened the link between pay and risk discipline. Whether the UK was right to abandon the rule, or whether the abandonment trades one set of incentives for another, is a separate question worth its own essay. But the trajectory is striking: the UK first implemented the rule (as part of CRD IV transposition), saw a decade of evidence that the implementation did not achieve its goals, and abandoned it.

The same indemnification logic appears in today’s PIK structures, refracted into different vocabulary. The cap on cash distributions imposed by the RIC tax framework pushed BDCs to maintain headline distribution rates while their underlying cash receipts deteriorated. The constraint on one margin produced an adjustment on another: PIK accruals expanded, the dividend was funded by other means, and the headline numbers held. The system finds the path of least resistance — not because anyone is corrupt, but because every participant has private incentives pointing toward that path. The bonus cap and PIK are two manifestations of the same dynamic: rules that operate on one margin while the binding incentive operates on another.

Both cases illustrate the deeper pattern the rest of this essay develops. When regulation constrains one margin of the incentive structure, markets adjust on another — often in ways the original rule never anticipated. The question is whether the macroprudential framework can be designed to anticipate the adjustment, and whether the political-economy conditions exist to do so.

Goodhart’s boundary problem

The classical statement is Goodhart’s law: “Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes.” Goodhart published this in the context of monetary aggregates in the 1970s, but the form is general. Any quantity that becomes a regulatory target ceases to be a reliable indicator of the underlying phenomenon, because the regulated population reorganises its activities to optimise against the target.

The macroprudential framework runs into a specific version of this problem. The framework is built on a regulatory perimeter — entities classified as banks, alternative investment funds, money-market funds, central counterparties — and within each perimeter operates with quantity-based instruments. The problem is that financial activity does not stay inside the perimeter when the regulatory cost of being inside it rises. The activity migrates outward, to non-bank vehicles, to off-balance-sheet arrangements, to jurisdictions and contractual structures where the regulator has less reach.

The empirical record on this is striking. iMaPP data — the IMF’s database of macroprudential actions covering 2010–2024 — show that European jurisdictions cumulatively implemented around 21 tightening actions per country on average, with the United States implementing around 13. Across the same period, the global private credit asset class grew roughly twelvefold, from $91 billion outstanding in 2010 to $1.246 trillion by 2024 (BIS data). The growth was not uniform — much of it concentrated in the latter half of the period — but the trajectory is unmistakable.

The pattern is not coincidence. As the quantity tools tightened on regulated balance sheets, the same credit risk migrated to the least-regulated edge — exactly as Goodhart’s boundary would predict. I want to be careful about what this correlation does and does not show. The chart does not establish that macroprudential tightening caused the migration to private credit. The growth of private credit is over-determined: yield-seeking institutional investors after a decade of low rates, the post-2008 retreat of banks from leveraged lending, the rise of private equity as the dominant ownership form in mid-market US business, regulatory arbitrage at the bank-fund interface. All of these mattered. The point is that the macroprudential framework was tightening throughout the period and did not, by its design, constrain the migration.

This is Goodhart’s boundary problem in the macroprudential context. The framework can tighten capital, tighten liquidity, tighten leverage limits for banks and for AIFs within its perimeter. It cannot, by direct instrument, prevent activity from migrating to entity types and contractual structures outside the perimeter. Each tightening creates marginal pressure for migration; the cumulative effect, over fifteen years, is a substantial transfer of credit-creation activity to vehicles where the prudential supervisor has less visibility.

The natural response — and the one that has been pursued throughout the period — is to expand the perimeter. AIFMD II in 2024 extended leverage limits and reporting requirements to a broader population of alternative funds. The FSB’s NBFI work programme since 2022 has tried to identify systemic activities across non-bank intermediaries and bring them within prudential reach. The ECB’s Macroprudential Strategy 2024 review specifically called out NBFI as a strategic priority.

Each expansion buys some visibility. But the expansion has a logical limit: every entity type the regulator brings inside the perimeter incentivises the next iteration of innovation outside it. The PIK arrangement in a BDC subscription line is a contemporary illustration. AIFMD II expanded the perimeter; the contractual structure that combines BDC distribution policy with sponsor IRR protection with GP carry economics sits at the contractual margin that no expansion of the entity perimeter directly reaches.

The deeper question is whether expanding the perimeter is the right response, or whether the regulatory architecture needs a different theory of where to act — one that operates on contract design rather than entity type. The next section takes on the three approaches that have been proposed and the frictions each runs into.

Three approaches, three frictions

If the macroprudential toolkit, as currently designed, cannot directly reach the incentive structures that drive the kind of equilibrium described earlier, what are the alternatives? There are essentially three, and each runs into a different binding constraint.

The first approach is structural reform: removing the activities or organisational forms that generate the problematic incentives in the first place. The intellectual genealogy runs through Glass-Steagall, Vickers (the UK’s Independent Commission on Banking, 2011), Liikanen (the EU’s High-Level Expert Group, 2012), and the Volcker Rule (the US’s 2010 Dodd-Frank provision). The proposals differed in detail but shared a common logic: ring-fence retail banking from investment banking, or prohibit certain trading activities entirely, or break up entities deemed too complex to supervise. By altering the structure rather than the parameters, structural reform aims to remove the underlying option-like payoff that drives the Bebchuk-Spamann mechanism.

The friction is political economy. The benefits of structural reform are dispersed (general financial stability, fewer crises) while the costs are concentrated on specific institutions and their employees. Mancur Olson’s collective-action logic predicts exactly which side wins in this configuration. The UK’s ICB recommendations were implemented but in diluted form: the ring-fence applies only to the very largest banks and the operational separation has been progressively softened. Liikanen was formally withdrawn by the European Commission in 2017. Volcker survived but was substantially weakened in 2020. The structural-reform path has not failed for intellectual reasons. It has failed for political-economy reasons that were predictable in advance.

The second approach is direct incentive regulation: rules that target compensation structure, contract design, or specific forms of risk-taking directly. CRD IV’s bonus cap is the cleanest example. The UK PRA’s Senior Managers Regime, which makes senior managers personally accountable for prudential failures, is another. EBA’s remuneration guidelines, which constrain deferral periods, malus, and clawback, fall in this category. So do underwriting standards expressed in supervisory letters, which try to constrain how loans can be structured.

The friction is arbitrage. Pay structures adapt around the rule. The bonus cap pushed compensation toward fixed pay without reducing total compensation. The Senior Managers Regime has changed reporting lines without obviously changing behaviour. Underwriting guidelines on covenant-lite loans were largely ignored because they were not formally binding. Where rules are binding, the regulated population adapts; where they are not binding, the rules do not have force. The UK’s decision to abandon the bonus cap in October 2023 is the most direct admission of this dynamic.

The third approach is perimeter expansion: bringing new entities and activities under the prudential framework as they emerge. AIFMD, MMF reform, the FSB’s NBFI agenda, ESRB activity-based recommendations all sit here. The logic is to chase the activity wherever it goes, rather than to constrain what activity can do.

The friction is Goodhart’s boundary problem — the issue covered in the previous section. Each tightening creates marginal pressure for migration outward. The perimeter expands; the next iteration of innovation appears at the new margin. The expansion is necessary — without it, activity migrates faster than the framework adapts — but it is not sufficient, because the dynamic does not converge to an equilibrium where the framework catches all systemic activity.

Each of these approaches has had its serious advocates. Each is genuinely useful at the margin. None has been able to address the core problem at scale, because each runs into a different binding constraint. The structural-reform proposals are diluted by political economy. The direct-incentive rules are arbitraged. The perimeter expansion is chased by Goodhart’s logic.

(A separate debate, which I am consciously not engaging with here, concerns whether the European architecture can be made more credible by completing Banking Union — common deposit insurance, full single rulebook, removal of national exemptions on resolution. That conversation is important, but it is about credibility within the existing framework, not about the framework’s reach. This essay is about the latter.)

What this leaves is a macroprudential framework that has done what it was designed to do — capitalised banks, made them more liquid, made them less concentrated — but that has not addressed and probably cannot address the kind of equilibrium I described at the beginning. The PIK creep, on this reading, is not the result of regulatory failure in the conventional sense. It is the natural expression of a framework that operates on quantities while the binding incentives operate on contracts.

What might move the needle

Against this backdrop of institutional frictions, what, if anything, can macroprudential authorities do to move the equilibrium? I am going to be modest about claims here. I do not think there is a single instrument that resolves the problem this essay has been describing. What I do think is that the framing of the policy debate has, over fifteen years, settled into asking the wrong questions — and that reframing the questions is, in itself, useful.

Three angles deserve more attention than they have received.

The first is transparency and data infrastructure. The FSB’s May 2026 report on private credit is candid about the data gaps. The most actionable response is not to wait for harmonised disclosure regimes, which are slow to negotiate and easy to dilute. It is to make existing reporting infrastructure work harder. AIFMD reporting templates could be extended to require loan-level disclosure on PIK accruals, covenant amendments, and valuation methodologies. Bank reporting on subscription lines and NAV-backed facilities could be made granular enough that supervisors can see the underlying exposure rather than just the credit-line commitment. The MMF reform agenda showed that targeted disclosure requirements, well-designed, can have substantial effects on behaviour. The same logic should apply to private credit.

The second is the contractual margin. The cleanest insight from the bank-comp literature — that compensation should track the cap structure as a whole rather than just equity — is implementable. The “inside debt” literature, including Edmans and Liu (2011, Review of Finance) and the operationalisations proposed in Bebchuk-Spamann themselves, argues for compensating bank executives partly in subordinated debt or other loss-absorbing instruments (extensions of the deferral and malus logic already in CRD V, rather than departures from it). Various deferred-pay vehicles linked to bank-wide loss-absorption performance, rather than equity returns, are another path to the same end. The structural problem with the current framework is that it constrains the level of variable pay without constraining its composition. Constraining composition — requiring a portion to be linked to subordinated claims or loss-absorbing instruments — addresses the option-like payoff directly. For private credit, the analogue is GP carry structures that share in NAV declines as well as gains. Whether this is achievable through fund-prospectus disclosure requirements or through LP fiduciary standards is a separate question, but the structural fix is identifiable.

The third is credibility on the resolution framework. The Banco Popular/Veneto comparison is not an argument for abolishing the public-interest test. It is an argument for narrowing the discretion that the test permits, or for separating the politically-sensitive judgment from the technical resolution decision. The CMDI proposal moves in this direction by expanding the scope of resolution and clarifying the DGS-funded path for mid-sized banks. Whether the final version of CMDI, when negotiated, will narrow the discretion enough to change bondholder behaviour is an open question. But the direction of travel — narrowing rather than expanding the room for case-by-case judgment — is the right one if the framework is to have ex-ante disciplining effect.

I am not putting these forward as a programme. I am putting them forward as the angles that follow from the analysis. Each of them runs into one or more of the frictions described above. Transparency proposals face the same political-economy resistance that structural reform faces, but at lower intensity. Contractual-margin reforms face industry pushback comparable to compensation regulation in general, but with the advantage that the literature supporting them is robust. Resolution-framework credibility faces sovereign-political resistance, but the CMDI process suggests there is at least appetite for incremental progress.

What none of these does is dispose of the underlying problem. The configuration I described in the PIK case is, in my reading, a permanent feature of financial systems with private capital allocation, competitive funding markets, and a generalised expectation of public support in extremis. The point is not to dissolve it. The point is to slow its growth and to make its manifestations more visible to supervisors before they crystallise.

What we have established

This essay has tried to make a single argument with several supporting claims.

The argument is that macroprudential policy, as currently designed, operates on quantities — capital, liquidity, leverage, exposure — while the equilibria that drive financial fragility operate on incentives. The post-2008 reform agenda has made banks better capitalised, more liquid, and less concentrated, and that is a real and substantial achievement. But it has not addressed the contractual and structural margins where the next equilibrium is being built, and the design of the toolkit does not permit it to do so directly.

The supporting claims are five.

First, the PIK creep in private credit is not an isolated phenomenon. It is the contemporary expression of a pattern that runs back through cov-lite leveraged loans, through mortgage securitisation, through the S&L crisis: contractual or structural innovation that converts cash-flow stress into balance-sheet items that do not require immediate recognition.

Second, the equilibrium holds because every participant in the capital structure has private incentives pointing the same way. This is not corruption or incompetence. It is the natural product of the contractual and fiduciary structures within which the participants operate.

Third, the post-2008 capital framework, while a genuine success at what it was designed to do, has not addressed the structural asymmetry that Bebchuk and Spamann identified — the option-like payoff that bank executives hold on a thinly-capitalised cap structure. Capital regulation changes the size of the buffer between management’s incentive and the taxpayer, but does not change the asymmetry itself.

Fourth, the European post-crisis architecture has gone further than the US on incentive-side regulation, and the natural experiments — Banco Popular versus Veneto, the bonus-cap implementation between 2013 and 2014 — are genuinely informative about what direct incentive regulation can and cannot do. The bail-in framework works when the political conditions for invoking it exist; the bonus cap was arbitraged through fixed-pay adjustments that left total compensation roughly unchanged.

Fifth, each of the three approaches to addressing incentives directly — structural reform, direct incentive regulation, perimeter expansion — runs into a different binding constraint. Structural reform is diluted by political economy. Direct incentive regulation is arbitraged. Perimeter expansion is chased by Goodhart’s boundary problem. None of these is a fatal objection; all of them are reasons why progress has been incremental.

The macroprudential scorecard in the chart above is the synthesis of this argument. Most cells in the matrix are grey — pairings of policy tools with incentive types where the tool does not, by its design, reach the incentive. The dominance of grey is the visual restatement of the argument: most policy tools do not apply to most incentive structures. The handful of green cells indicate genuine matches, mostly in the leverage-ratio backstop against model gaming and the liquidity-tool response to retail wrapper mismatches. The yellow cells are partial matches, often through AIFMD’s reach into fund-side activities. The red cells — risk-weight insensitivity to compensation convexity, the role-based-allowances arbitrage of the bonus cap, selective application of the public-interest test, the documentation opacity of PIK and covenant-lite — are the explicit cases of arbitrage where the rule was tried and the market adjusted around it.

Which brings us back to where we started. PIK in first-lien private credit is the latest expression of this incentive pattern. It is not a fraud, it is not a bubble, and it is not the product of regulatory failure in the narrow sense. It is the natural equilibrium of a system in which every participant — borrower, sponsor, GP, LP, BDC, bank — has private incentives pointing the same way, and in which the macroprudential framework, as currently designed, cannot reach the contractual margin where the equilibrium forms. Whether the next downturn crystallises the underlying credit problem in private credit, in some adjacent corner of the financial system, or — most likely — in a form that nobody is currently watching, the structural argument is the same. The fault line is at the contractual margin. The toolkit operates on the balance sheet.

That is the wrong margin. And the structural reason it is the wrong margin is not that the macroprudential framework has failed. It is that the framework was never designed to act there in the first place.

Paweł Fiedor - The Macro Prudential View

Views expressed are personal and do not represent the institutions with which the author is affiliated.

Sources and references

Chart 1 (PIK creep timeline) compiled from author’s aggregation of SEC EDGAR 10-K and 10-Q filings for the top fifteen exchange-traded BDCs, four-quarter rolling basis. Ares Capital Corporation first-lien PIK share computed from ARCC Schedule of Investments (measured by fair value). Chart 5 (bail-in case comparison) drawn from SRB Decision SRB/EES/2017/08 on Banco Popular Español S.A. (7 June 2017); SRB Decision SRB/EES/2017/11 and SRB/EES/2017/12 on Veneto Banca S.p.A. and Banca Popolare di Vicenza S.p.A. (23 June 2017); European Commission State Aid Decision SA.45664 (25 June 2017); and Italian Budget Law 145/2018 establishing the Fondo Indennizzo Risparmiatori. Chart 6 (bonus cap natural experiment) computed from EBA Report on Benchmarking of Remuneration Practices and Data on High Earners 2013 (Annex I) and EBA-Op-2016-05, Report on Benchmarking of Remuneration Practices at EU Level and Data on High Earners (data as of end-2014) (Annex I), averages per individual. Chart 7 (private credit growth vs macroprudential tightening) uses private credit outstandings from Avalos, Doerr and Pinter, The global drivers of private credit (BIS Quarterly Review, March 2025) and cumulative macroprudential actions from the IMF integrated Macroprudential Policy (iMaPP) Database.

Cited reports and primary documents: FSB, Vulnerabilities Associated with Private Credit (May 2026); iCapital, Painting a PIKture: The Benefits and Risks of PIK in Private Credit (2025); SRB, Minimum Requirement for Own Funds and Eligible Liabilities (MREL) Dashboard, Q4 2024; European Commission, Proposal for a Regulation amending the BRRD and SRMR — Crisis Management and Deposit Insurance framework (18 April 2023); PRA, PS9/23: Remuneration: Ratio between fixed and variable components of total remuneration (’bonus cap’) (October 2023); ECB Banking Supervision, ECB Macroprudential Bulletin (various issues).

Cited academic literature: Admati and Hellwig, The Bankers’ New Clothes: What’s Wrong with Banking and What to Do about It (Princeton University Press, expanded 2024 edition); Bebchuk and Spamann, “Regulating Bankers’ Pay”, Georgetown Law Journal 98 (2010); Bebchuk, Cohen and Spamann, “The Wages of Failure: Executive Compensation at Bear Stearns and Lehman 2000-2008”, Yale Journal on Regulation 27 (2010); Cerasi, Deininger, Gambacorta and Oliviero, “How post-crisis regulation has affected bank CEO compensation” (BIS Working Paper 630, 2017); Cheng, Hong and Scheinkman, “Yesterday’s Heroes: Compensation and Risk at Financial Firms”, Journal of Finance 70 (2015); Colonnello, Koetter and Wagner, “Compensation regulation in banking: Executive director behavior and bank performance after the EU bonus cap”, Journal of Accounting and Economics 76 (2023); Edmans and Liu, “Inside Debt”, Review of Finance 15 (2011); Galati and Moessner, “Macroprudential Policy — A Literature Review” (BIS Working Paper 337, 2011); Goodhart, “Problems of Monetary Management: The U.K. Experience”, Papers in Monetary Economics (Reserve Bank of Australia, 1975); Haldane, “The Dog and the Frisbee” (speech at the Federal Reserve Bank of Kansas City Jackson Hole Symposium, August 2012); Tarullo, “Departing Thoughts” (speech at the Woodrow Wilson School, Princeton University, 4 April 2017).

The Financial Stability Board’s first dedicated report on private credit, published 6 May, lands at a moment when European regulators are visibly turning their attention to the asset class. Frank Elderson, vice chair of the ECB’s supervisory board, said in an interview reported by Barron’s that European private credit “bears watching.” Earlier this month, the Eurosystem published its own roadmap for strengthening the macroprudential framework around non-bank financial intermediation — and private credit features prominently.

What stands out about the FSB report, though, is that the most interesting thing in it isn’t private credit. It is the structural picture of bank–nonbank interconnection that emerges, almost by accident, from the report’s data-availability annex. Read carefully, the report is best understood not as an asset-class warning but as a diagnostic of broader NBFI supervision. Banks and NBFIs increasingly operate as what the European NGO Finance Watch has called “communicating vessels,” and supervisors only partially see the connections between them. Private credit matters not because it is unique, but because it makes that wider problem legible. The same circles of risk — leverage layering, fund-bank financing loops, synthetic risk transfer, cross-border intermediation, opaque valuation — appear across the NBFI ecosystem. They are easier to see in private credit right now because it has grown fast enough to attract dedicated supervisory attention. The Eurosystem’s May paper makes the same point in regulatory language: data gaps in private credit are part of a wider NBFI data strategy still being built.

It is worth saying at the outset what the FSB also says: direct bank lending to private credit funds, where it can be separately identified, is below 0.5% of total bank assets in the FSB member data; private credit can support economic activity, diversify lending across the system, and provide tailored financing to underserved borrowers. The argument here is not that direct exposures are obviously systemic. It is that, even where direct exposures look manageable, the intermediation web and the data gaps are the supervisory issue the report points to.

The rest of this piece reads the FSB report through that lens, drawing also on the ECB Financial Stability Review, the EBA Risk Assessment Report, the OFR’s recent brief on counterparty exposures, and the Eurosystem’s May paper. Five charts walk through the architecture, the supervisory range of uncertainty, the cross-border asymmetry that defines European private credit, the SRT loop that closes some of the circles, and finally the data-availability heatmap that explains why all of this remains poorly observed.

The architecture

Start with the structural picture. The FSB’s ecosystem diagram shows banks financing private credit funds through multiple facilities — subscription lines, NAV facilities, fund portfolio financing — while those same funds lend onward to corporate borrowers. The vertical axis of the schematic above replicates that: euro area banks lend to EA-managed private credit vehicles, which lend on to borrowers either inside or outside the area. The horizontal axis is the harder part. Three further flows complicate the picture, and the FSB names each of them.

First, the destination of euro-area fund portfolio finance is not euro-area corporates. About 70% of EA fund portfolio finance is directed to non-EU borrowers, the FSB reports — most of them US-linked. The euro area’s private credit fund infrastructure is, in significant part, intermediating credit to firms outside the bloc.

Second, euro-area corporates’ own private credit borrowing largely doesn’t come from euro-area lenders. The ECB’s May 2024 Financial Stability Review found that only around a fifth of private credit borrowed by EA companies is provided by EA-managed funds. The other four-fifths comes from outside the area — predominantly US managers operating cross-border.

Third — and this is where the FSB’s analysis gets most interesting — banks may remain indirectly exposed to credit risk they thought they had transferred. Through synthetic risk transfers (SRTs), banks transfer the credit risk of their loan books to investors while keeping the underlying loans on their balance sheets. The EBA reports that private credit funds are the largest single investor group in EU bank SRTs. And, as the FSB itself acknowledges, it remains unclear how far banks fund the very NBFIs that buy their SRT risk. When that loop closes, the economic risk may remain partly within the bank–NBFI perimeter even as the regulatory capital requirement against the underlying loans has been reduced.

The FSB calls these “circles of risks.” The phrase is appropriate. It is also a structural description that applies well beyond private credit.

What we don’t know about size

Here is the most uncomfortable finding in the FSB report, and the one most easily missed. The report’s headline estimate of bank lending to private credit funds — about $220bn drawn and undrawn — is based on regulatory data from FSB member jurisdictions. But the report also references commercial and academic estimates that range up to $546bn for what is broadly a related perimeter.

The chart above shows where the four most-cited serious estimates land. The FSB member figure ($220bn, end-2024, five jurisdictions) is the lowest. The HKMA’s reconstruction from commercial data ($268bn, global, end-2023) sits modestly higher. Moody’s analysis of US bank filings ($300bn, drawn only, late 2025) is higher still. The FSB’s own summary of private-sector composite estimates puts the upper bound at $546bn — combining $300bn drawn loans, $155bn undrawn commitments, and $91bn from EU branches operating in the US, drawn from a body of research including Berg and Lee’s recent OFR brief (which itself, more narrowly, finds $74bn utilised and $123bn committed in US Y-14 data for participating banks).

The honest reading is not that supervisors see less than half of true exposure. It is that the available estimates measure different things — different geographies, different dates, drawn versus undrawn commitments, different scopes of what counts as “private credit” — and there is no harmonised perimeter to compare against. The spread is therefore not estimation error around a single number. It is a measurement of how much methodological choice drives the headline figure.

This is the data deficit story in one chart. It is also a leading indicator of the wider NBFI surveillance problem that the Eurosystem’s May paper is explicitly trying to fix.

A cross-border story Europe is not at the centre of

Of the three FSB findings sketched above, the one that should preoccupy European policymakers most is the cross-border asymmetry. The chart above puts it in a single panel.

Euro area-managed private credit funds direct roughly 70% of their portfolio finance outside the EU, predominantly to US-linked borrowers — the top bar. Euro area corporates, meanwhile, get roughly 80% of their private credit from non-EA lenders — the bottom bar. The two figures come from different but complementary datasets, and they tell the same story from opposite ends.

Read together, they suggest that the euro area sits in the middle of a cross-border intermediation chain rather than being the destination of either flow. EA fund infrastructure is helping place credit risk on mostly US-located corporate balance sheets. EA corporates are mostly funded by non-EA private credit funds. The euro area, in private credit, is a transit node — not a self-contained system.

What does this mean for supervision? At a minimum, it complicates the EA’s regulatory perimeter. The corporate credit risks that matter most for the euro area economy are largely originated outside it. The fund-level vulnerabilities most easily observed by EU supervisors — concentration, liquidity mismatch, leverage layering — increasingly relate to non-EA exposures. Effective monitoring requires looking at things the EA framework was not originally built to see.

This is not a uniquely private-credit problem. The same cross-border intermediation pattern appears in repo, derivatives, and prime brokerage. But private credit is the segment where the asymmetry is most clearly documented in recent regulatory work.

The SRT loop is closing

Now to the loop that gives private credit its specifically circular structure.

Synthetic risk transfers — instruments that move the credit risk of a bank loan book to outside investors while leaving the loans on the bank’s balance sheet — have grown rapidly. The right-hand panel above shows currently-outstanding protected loan pools by origination vintage, rising from around €30bn per year in 2017–2020 to €82bn in 2021, €113bn in 2022, and €143bn in 2023. The 2024 H1 partial-year figure of €63bn suggests another year close to or above the 2023 level. Older vintages have partially matured, so these are stock figures rather than annual issuance.

The left-hand panel, drawn from EBA reporting, shows who is on the other side of those transactions. Private credit funds are reported as the leading investor group, named by roughly a third of responding banks. Other investment funds, insurance companies, and pension funds account for another 45% between them, with a residual ~22% spread across asset managers, sovereign wealth funds, and other investor types.

This is where the FSB’s “circles of risks” language earns its plural. Some banks are simultaneously: financing private credit funds through credit lines; transferring loan credit risk to those same funds through SRTs; and holding revolving credit facilities with the corporates the PC funds are lending to. The motive for the SRT itself is regulatory capital relief — by transferring the credit risk of a loan book to outside investors, the bank reduces the risk-weighted capital it must hold against those loans. But when the same fund holds a bank’s SRT and receives that bank’s subscription financing — and the FSB acknowledges this is plausible but unquantified — the capital has been freed up while the economic exposure has not entirely left the bank–NBFI perimeter. The risk has been re-routed, not eliminated.

The supervisory response to this is hard. It requires look-through data that doesn’t currently exist at the regulatory level. The Eurosystem’s May 2026 paper identifies legal entity identifiers (LEIs) and a dedicated AIFMD subcategory for private credit funds as priority data infrastructure precisely to begin closing this gap. Until those are in place, the question of how much SRT risk is sitting inside fund vehicles that the same banks also finance is — to use the FSB’s own framing — a question.

Why supervisors still can’t see most of this

The FSB report’s data annex contains what I think is the most important chart in the entire document, even though it sits near the back. Redrawn above, it shows what data is available to authorities in each major jurisdiction, by data category.

The story is not, as I initially expected, “US sees everything, EA sees nothing.” Look at the columns. The euro area can fully see aggregate drawn exposures (top row, green) but only via ad hoc collection. Total commitments — including undrawn lines — are missing. Facility-type breakdowns are partial. Subscription lines, the most common form of bank-to-fund financing, are only partially reported. Revolving credit facilities, lending to offshore funds, and granular time series are largely missing.

Look at the rows. The categories where private credit risk could be most plausibly hidden — subscription lines, revolving facilities, offshore fund exposure, granular time series — are red or yellow in most jurisdictions, including in the US. Switzerland is blind to most of the granular facility metrics, with five of seven categories missing. The UK and the US are the strongest reporters among the jurisdictions shown, but neither is comprehensive.

This is the supervisory deficit made concrete. The implication for the wider NBFI argument is direct: if this is what data availability looks like for the segment that has just received a dedicated FSB report, the surveillance situation for the broader NBFI sector — hedge funds, MMFs, securitisation vehicles, derivatives exposures — is no better, and in many places worse.

The Eurosystem’s May paper makes the same diagnosis. It proposes three priority interventions: targeted amendments to AIFMD, the UCITS Directive and related legislation to enable data sharing between the ECB, ESRB, ESMA, and national supervisors; a centralised data access mechanism, with ESMA exploring a data hub as part of an integrated reporting framework; and mandatory use of LEIs across NBFI reporting, including for private credit funds. The paper also proposes amending ESMA’s AIFMD guidelines to introduce a specific AIF subtype for private credit funds — directly addressing the residual-category problem that today makes EU private credit hard to identify in regulatory data.

These are not glamorous proposals. They are plumbing. But they are the right plumbing to fix, and the FSB heatmap above is the diagnostic that makes the case for them.

What the FSB report is really about

The FSB report can be read in two ways. The narrower reading treats it as a specific assessment of a fast-growing asset class — useful, careful, not particularly novel. The wider reading, which I find more interesting, treats it as a diagnostic of NBFI supervision: a worked example of what we know and don’t know about a single segment of market-based credit intermediation, written by people who clearly understand the limits of their own evidence.

Finance Watch’s recent report on systemic risk from shadow banking makes the wider case explicitly. Banks and NBFIs, they argue, function as communicating vessels — connected via leverage, funding, securitisation, derivatives, and liquidity channels that supervisors only partially see. The FSB’s circles-of-risks language is the same point in narrower form. So is the Eurosystem’s proposal to build a centralised NBFI data infrastructure. So is the ESRB Advisory Scientific Committee’s reported consideration of a formal private credit inquiry.

What’s notable is that these efforts are converging. The diagnosis is broadly shared: NBFI is now a large enough share of global financial assets that supervising the banking system alone is no longer sufficient. The prescriptions are still being assembled — data infrastructure first, then macroprudential tools, eventually possibly entity-level designations of systemically important NBFIs.

Private credit is where the architecture is currently easiest to see in Europe. That is partly because it has grown fast, attracted political attention, and become the subject of a dedicated FSB report. It is also because — paradoxically — private credit’s interlinkages with banks are more visible than those of other NBFI segments, precisely because banks are involved at every step. The FSB report is therefore both an asset-class document and, more interestingly, a structural map.

Read it as the structural map. The risk of circular exposures is real. The data to confirm or refute the more concerning interpretations is not yet available. The Eurosystem is building the surveillance infrastructure that would make better answers possible. The recent publications suggest these theoretical risks are increasingly being viewed as supervisory priorities rather than abstract concerns.

What we are watching, in short, is the start of a wider NBFI supervisory framework being assembled — with private credit as the most legible test case.

Paweł Fiedor - The Macro Prudential View

The views expressed are the author’s own and do not necessarily reflect those of any institution with which the author is or has been affiliated.

Sources: FSB, Report on Vulnerabilities in Private Credit, 6 May 2026 — https://www.fsb.org/uploads/P060526.pdf; Eurosystem, Strengthening the macroprudential lens in the regulation of non-bank financial intermediation, May 2026 — https://www.ecb.europa.eu/press/intro/publications/pdf/ecb.ebaecb202605.en.pdf; ECB, Financial Stability Review May 2024, Special Feature C — https://www.ecb.europa.eu/press/financial-stability-publications/fsr/special/html/ecb.fsrart202405_03~bc23a48dbc.en.html; EBA, Risk Assessment Report Spring 2025; Berg and Lee, Measuring Counterparty Exposures to Private Credit, OFR Brief Series, March 2026 — https://www.financialresearch.gov/briefs/files/OFRBrief-26-02-measuring-counterparty-exposures-private-credit.pdf; Finance Watch, Systemic risk from shadow banking — https://www.finance-watch.org/policy-portal/stability-supervision/report-systemic-risk-from-shadow-banking/; Barron’s, “Europe’s Regulators May Launch Private Credit Inquiry,” — https://www.barrons.com/articles/europe-private-credit-probe-c22ea2dc.