The scale of the non-bank financial sector
The FSB's global monitoring exercise puts non-bank financial institutions at roughly $250 trillion at end-2023, 49.1% of total global financial assets, having grown about 130% between 2009 and 2023. In 2023 alone the sector grew 8.5%, more than double the banking sector's 3.3%. The FSB's "narrow measure" — the entities it thinks actually do credit intermediation in ways that could go wrong — rose 9.8% to $70.2 trillion, the highest ever recorded in the exercise. The following year's report has NBFI growth of 9.4% in 2024 and the sector's share at 51.0% of global assets, which if right means non-banks passed the halfway and nobody held a ceremony.
The label is barely older than the last crisis. Paul McCulley, then an economist at PIMCO, coined "shadow banking" in 2007 to describe non-bank intermediaries doing bank-like maturity and liquidity transformation without deposit insurance or a lender of last resort behind them. At the time it mostly meant off-balance-sheet conduits and structured investment vehicles parked next to banks. It does not mean that any more.
The post-2008 reform programme did what it said: banks are better capitalised, and the risk is somewhere else. Those are the same sentence.
The market makers are not banks
The firms that make prices in equities and options are no longer, mostly, the banks. 's net trading revenue in 2025 was $39.6bn, close to double the $20.5bn it made in 2024. Goldman Sachs, by comparison, reported record equities net revenues of $13.4bn for full-year 2024, with combined FICC and equities trading revenue of roughly $18bn that year. Citadel Securities did $9.7bn in 2024 and a record $12.2bn in 2025, up 25%. Hudson River Trading posted $11.4bn of trading revenue in the three months to June 2026 with $7.4bn of net profit, a record. In 2025 Jane Street, Hudson River and Citadel Securities together took in more than $60bn, beating the trading desks at JPMorgan and Goldman Sachs.
None of these firms is a bank. None is subject to bank capital rules. All of them disclose close to nothing.
That would be a governance curiosity if the balance sheet doing the market-making were only doing market-making. It is not. Jane Street lost $15bn in July 2026 — its first losing month in a decade — on , and in its latest debt offering described its strategy as evolving to include longer-term positions, in line with a hedge fund. We wrote that up on 19 August and again on 20 August. The point is not that the firm lost money. It is that the entity you need to quote a price in a selloff is the entity taking losses in the same selloff. Banks are capitalised and supervised precisely to stop that overlap. Nobody sees it here until a month like July.
The basis trade and the leverage nobody totals up
The largest single pool of identified hidden leverage sits in US Treasuries. Federal Reserve research published in June 2026 put the at roughly $830bn as of September 2025, about double its early-2020 peak. Market commentary during 2026 has put it nearer $1 trillion and then noted more than $200bn coming out of it. The typical leverage described is 50x or more, because the cash leg is financed in repo and the spread being harvested is tiny.
The Dallas Fed in May 2026 flagged the basis trade and swap-spread arbitrage as pure funding demand, and the New York Fed published Staff Report 1170 on the subject in November 2025. Which is to say: the official sector knows. Knowing is not the same as being able to see it in real time, and the $830bn figure was nine months stale on the day it was printed.
The precedent is Long-Term Capital Management. In early 1998 LTCM had about $4.7–$5bn of equity supporting roughly $125bn of assets — about 25:1 — and $1 trillion or more of notional derivative exposure off balance sheet. Its lenders could see the balance sheet. They could not see the total, because the dangerous part was in OTC derivatives and repo financing spread across counterparties who each saw their own slice. On 23 September 1998 the New York Fed convened fourteen banks and broker-dealers who put in about $3.6bn for roughly 90% of the fund. The structural lesson survived; the reporting infrastructure did not.
We wrote on 19 August about the selling off everywhere at once, with the 30-year Treasury touching 5.33%, its highest since 2007. The marginal holder of that is a repo-financed relative-value trade, not a pension fund.
Collateral plumbing and the double-pledge
Funding leverage is also migrating to where the least is visible. The BIS reported in August 2026 that non-bank financial institutions now account for more than 40% of non-centrally cleared euro-denominated cash-borrowing repo, up from under 30% at end-2020, and expects the share to keep growing. Cleared repo has standardised , netting and an aggregate a regulator can look at. has a negotiated haircut, private terms, and no consolidated view of how much one borrower has taken from five lenders — all of whom will call on the same morning. We covered that shift on 20 August.
What happens when nobody re-verifies collateral turned up twice in one week, in unrelated markets. The SEC sued three former executives of lender Tricolor over its $1.9bn collapse, alleging they raised money on faked documents and used the same vehicles as collateral multiple times. "These defendants defrauded investors based on bogus collateral and violated the integrity of our private credit markets," said enforcement director David Woodcock. Days later Singapore police confirmed they were looking into iron ore trader , which has been under pressure over allegedly falsified documents; Vitol and Cargill stopped trading with it at the end of July, and the DOJ, CFTC and FCA are variously examining or monitoring. Radiant World has not been accused of wrongdoing. Our write-ups: 19 August and 20 August.
Two markets with nothing in common, one mechanism: the collateral was a document, and the check on the document was a spreadsheet from the borrower. That is the machine's weakest joint, and it scaled faster than anyone's verification did.
Why it matters to this crash
Because the and the leverage system are now the same story, and one of them is disguising the other.
Banks spent the summer carrying exposure to leveraged AI vehicles and to Situational Awareness, the AI fund that blew a hole in Jane Street, and they hedged the jump risk by buying semiconductor options. That hedging demand is what made chip volatility expensive. When the exposure vanished, so did the hedges. Implied volatility on the VanEck Semiconductor ETF fell from 65% a month earlier to 40% on Monday 17 August, the lowest since February, and the ratio of open puts to calls dropped to 1.89, the most call-heavy since early April. "Now they don't need those hedges, and I believe unwinding of those hedges has made volatility in the sector inexpensive," Zed Francis of Convexitas told CNBC. We took that apart on 19 August and 20 August.
Read the sequence slowly. A leveraged fund fails. The dealers exposed to it stop buying protection. Protection on the most crowded sector in the market gets cheap. Positioning turns more bullish than it has been in four months. The failure made the system look calmer while leaving it less hedged. Cheap volatility is being read as an all-clear when it is partly the residue of a blow-up.
The size of that blow-up is itself contested, which tells you something about disclosure at these firms. Bloomberg reported a $15bn net loss for July; the FT, ranking it against history's biggest trading losses, put the figure at $35bn and attributed it to 's leveraged AI stock bets; Reuters at one point put the Situational Awareness component at $1.5bn. The numbers are not reconcilable from public information. Nobody outside the firm knows.
What would make this dangerous
The failure mode is not a loss. It is a margin spiral: a price triggers collateral calls, meeting the calls requires selling, the selling moves the price, and round it goes. Three things would tell you it has started.
Haircuts widening in bilateral repo. In 2007–08 the run started in asset-backed commercial paper and moved into repo, and the transmission was lenders demanding more collateral against mortgage-linked paper until borrowers had to unwind. With non-banks now past 40% of non-cleared euro repo borrowing, a general haircut increase is the same lever with more people hanging off it. Watch also , which sat at 3.65% and had not moved as of our 19 August note on ; a persistent print above the target range is the front end telling you funding is tight.
A fast move in long yields with the basis trade still on. The UK's 2022 LDI episode is the clean template. Pension funds ran leverage of roughly 2x to 4x capital, a median around 2.5x, via gilt repo and receive-fixed swaps. When the mini-budget moved yields, the collateral calls between 23 and 28 September 2022 came to about £66bn, and the Bank of England had to announce purchases with capacity up to £65bn, of which about £19.3bn was used. That was a modest amount of leverage in a small market. The is put at $830bn on 50x.
A market maker stepping back from quoting. If a firm like Jane Street or Citadel Securities materially reduces its footprint in ETFs or options while carrying directional losses, the depth that everyone assumes is permanent turns out to be discretionary. There is no capital requirement obliging them to stay.
And one that is harder to observe: whether reported exposures are real. Lehman's examiner, Anton Valukas, found in his report released on 11 March 2010 that Repo 105 transactions let the firm move up to $50bn of assets off its balance sheet at quarter-end. Tricolor and Radiant World are allegations about documents that nobody independently checked. The pattern to watch for is creditors quietly withdrawing before any regulator arrives, because the creditors always know first.