On 22 July 2026, Aston Martin announced new loans led by HPS, the BlackRock-owned private lender. “This new £550m debt financing,” said Doug Lafferty, its finance chief, “significantly strengthens our liquidity.” The loans effectively bypassed existing bondholders, including a creditor group led by Arini Capital Management, Bloomberg reported, and Arini’s flagship fund lost about 8% that month.

Through 23 October the flagship was down about 26% for the year, Bloomberg reported. Bloomberg and FT headlines framed a rout: a “Star Trader’s Woes,” a “bloodbath” for credit investors.

The next suspect is leverage. Yet Sona Asset Management, a credit peer with a larger flagship, shares four of Arini’s five prime brokers, the banks that finance its trades. Sona files about the same gross assets, everything held before debts, per dollar of reported capital, and it was about flat in mid-October. What separated the two sat under that ratio: in what each fund held, and, the evidence suggests, in derivatives the filing does not count. That matters to everyone who holds the same bonds.

• Aston Martin’s July loans put new secured debt ahead of Arini’s bondholder group on some assets; Aston Martin’s notes kept falling, and Arini is preparing a claim.

• Arini and Sona sat on the same lender committee at Altice International, the telecom group, and in the US junk index, Janus Henderson found, cable and telecom drove much of the selloff in CCC, junk’s lowest rating band.

• Margin calls, banks’ demands for more collateral, made Arini sell its more liquid holdings, the WSJ reported, and some investors now hesitate to buy debt it holds in size, Bloomberg reported.

• The firm is sturdier than its flagship: money in its CLOs, loan pools funded by bonds, stays for each deal’s life, and investors added to the flagship.

• Below the paid line, both flagships’ filed ratios and a month by month split of the loss against the market.

• Then a model of when the flagship’s investors are made whole, with my odds.

Aston Martin went around them

I set Arini’s flagship beside Sona’s on their SEC filings, read Arini’s three years of audited UK accounts and Aston Martin’s own bond announcements, and split the loss month by month against index data anyone can download.

A falling US junk market explains much of Arini’s loss at the ratio on file. The rest came in two quiet months that only its names explain, and Aston Martin is the name with the clearest paper trail.

On 13 July, DCM Insider reported that BlackRock, Sculptor and Arini had formed a creditor group holding more than half of the carmaker’s 2029 notes, to guard against a liability management exercise. The notes then traded near 70 cents on the dollar.

Such an exercise moves value away from existing creditors: new debt ranked ahead of them, or assets moved into a unit they cannot claim.

Aston Martin is one of several losing names: its 2029 notes come to about $1.8bn after two more placings in 2024, against a flagship of roughly $8bn in January. They date from 13 March 2024, when Aston Martin priced $960m at 10.000% and £400m at 10.375%, both senior secured, backed by pledged assets.

On 22 July it announced a £450m term loan, borrowed for a fixed term, and a second, £100m loan, both led by HPS. They are secured on “certain of the Group’s assets situated in a newly incorporated subsidiary, together with certain other assets.” Of the money, £190m repaid two existing loans that already ranked ahead of the notes.

In early August, creditors owed £1.3bn, the 2029 noteholders among them, threatened to unwind the HPS deal and block a sale of 50.1% of the non-automotive branding rights to Authentic Brands, the FT reported. Arini is gathering evidence in New York for a UK claim, Just Auto reported, and Aston Martin’s lawyers argue the deal is lawful.

Bloomberg put the notes at about 55 cents at the start of October and 47 cents on 28 October, the day Aston Martin reported a narrower third quarter loss. At 47, the dollar notes’ 10% coupon, their annual interest, alone yields about 21%: the price assumes heavy losses for holders who now stand behind HPS’s secured loan.

The seat Sona shared

Sona sat on the same Altice committee as Arini and was down about 0.5% for the year in mid-October; Arini was down more than 21%. Neither fund had ever had a losing calendar year, the FT and Bloomberg reported.

Altice International is Arini’s other large fight. The telecom group’s drop-down in late November 2025 moved assets away from its secured lenders, and in December 2025 Octus, a credit research firm, listed Arini and Sona among the eleven funds leading them. The FT reported that Altice was the flagship’s largest loss in July.

Sona’s October losses included Virgin Media O2 and new Paramount Skydance debt, the FT and Bloomberg reported. Arini lost in July on equity tranches of credit default swaps providing protection against losses on Virgin Media debt, Bloomberg reported. A credit default swap insures against a borrower’s default; an equity tranche of a pool of such swaps takes the pool’s first losses. The WSJ’s account of Arini’s losses named Brightspeed and Optimum, two US telecom and cable borrowers.

Distressed debt like this trades in or near CCC and below, and that band fell. The ICE BofA US CCC and lower index lost 6.9% in total return from 31 December to 23 October, almost all of it from September; the broad euro high yield index lost 1.6%. Bloomberg reported European CCC spreads, the extra yield those bonds pay, at their highest since March 2020.

Janus Henderson wrote on 15 October that, in the US high yield index’s CCC bucket, the cable and satellite sector “is dominated by CSC Holdings, a subsidiary of Optimum,” and the wirelines sector by Altice. The two sectors were “responsible for much of the spread widening within CCC,” it wrote.

Diameter, another credit hedge fund, also slipped into loss in October.

The selloff explains October’s timing, when all three lost. In the first half of October the US CCC index fell 1.84%: Sona lost about 1.7 times that, Arini about 3.8 times. What the selloff cannot explain is a twenty-point gap for the year between two funds on the same committee.

Too big to sell

Hamza Lemssouguer, a former Credit Suisse high yield trader, founded Arini in 2021 and is its chief investment officer. His method is the large stake and the lead seat. A Bloomberg headline of 10 September 2025 read “Ex-Credit Suisse Prodigy Is Dominating Europe’s Distressed Scene.” In one such fight, lending to keep a company alive at Kem One, Arini was a rescue lender.

A lead seat buys control of a restructuring, and its price is a large share of the debt, which a fund cannot sell in a hurry. Some investors have become reluctant to buy debt in which Arini holds large stakes, Bloomberg reported. A seller the whole market can see gets a lower bid, and that bid then sets the value of the rest of the position.

Concentration has cost the fund in calm markets before. In March 2024 it lost 6.6%, Bloomberg reported from its letter, which blamed “significant moves in three very large capital structures”; the US CCC index rose 1.3% that month. It also lost in each of the last four months of 2025, Bloomberg reported, while the index moved -0.1%.

Leverage turns the problem into cash. The WSJ reported on 30 October that Arini had sold about $800m of debt in recent weeks after margin calls from its banks. The calls were tied to credit default swap indexes, traded baskets of those swaps, and to other leveraged positions. Arini met them by selling its more liquid holdings.

The liquid bonds go first, so what stays is the paper buyers avoid, in the names that caused the loss. Arini had earlier sold its Intrum and Ritz-Carlton Yacht Collection investments and had recently left Pathway Vet Alliance, the WSJ reported. In late October Bloomberg put the yacht debt below 75 cents; it had been above 92 at the end of September. Lemssouguer has reportedly cut new purchases to manage the positions he has. A Bloomberg headline on 30 October said Pimco and Oaktree had bid for Arini’s junk debt.

Arini’s own account is that the losses are marks, not a broken thesis, and its investor letter, as the FT and Bloomberg reported it, says why. “We do not believe it is an indicator of soured investments, but rather part of the process – being contrarian is a necessary condition for this opportunity set.” That view does not change what the fund must do while its banks want collateral.

Each forced sale sets a price for everyone else in those seats, from BlackRock and Sculptor in Aston Martin’s group to PGIM, Invesco, King Street, BlueBay and Fidelity on Altice’s committee.

The firm outlives the fund

From June on, each report found the flagship deeper in loss:

The firm is sturdier than that table. Arini managed $22bn at 31 August, its release of 21 October says, and much of that is money that cannot leave in a hurry.

Arini’s Form ADV, the adviser filing the SEC requires, lists 19 CLOs with $9.2bn of gross assets. Each CLO’s loans are financed by bonds sold to investors, so its money stays for the life of the deal.

On 21 October Arini held the final close of its first direct lending fund, with the money it has lent approaching $2bn.

Against the firm, the flagship’s loss is survivable. Bloomberg put the flagship at $7.5bn at the end of July, down about 7% for the year, by my count. The FT put it at around $7.3bn after a 15.7% loss to September. Together they imply about $8.0bn to $8.7bn at the start of the year, depending on whether the FT’s figure includes October’s new money. A 26% loss on that is about $2.1bn to $2.3bn, about a tenth of the firm.

Investors still added money. By early October Arini had raised $1.5bn of new commitments, promises to invest, for the strategy, Bloomberg reported. More than $400m arrived at the start of the month, about $425m by the FT’s count, and the rest comes by February. That money buys in at the flagship’s lower October value. The fund has climbed back before: after a drawdown, a fall from a peak, in early 2024 it gained roughly 29% over the next twelve months, the FT reported. Gavin Yates, Arini’s former US head, left in August and is expected to build a distressed team at Citadel, the WSJ reported.

Below the line I date when the flagship regains its high-water mark, the level a fund must recover before it can charge performance fees again, and put my odds on it. The split rests on the ratio on file, which about matches the flagship’s loss as a multiple of the CCC index’s fall in the weeks it fell; the date rests on where the flagship stands.

→ Credit default swap tranches, as in Arini’s July loss: Paulson & Co.’s 2007 ABACUS protection, rebuilt from the SEC’s complaint, Trade File No. 5 on Patreon.

How long does a fund down a quarter, and still holding the names that sank it, take to climb back?

Below the paid line:

• A month by month table of the loss against the lowest-rated bond index, rebuilt from public data.

• The two flagships side by side: gross assets and banks from their SEC filings, set against capital as reported.

• A three-outcome model of the recovery, with the odds I would bet at.

• Arini’s London company’s audited accounts, read for what they show, and four dated checks.

Read more

Leopold Aschenbrenner’s fund is offering its own clients a stake in Fluidstack, the company building data centres for Anthropic. Hedgeweek, relaying Bloomberg, reported on 29 October that the fund’s clients may “co-invest alongside it in Fluidstack” at a valuation of $40bn. Situational Awareness led Fluidstack’s January round at $7.5bn.

In other words, the fund’s new money goes where its old money already sits. Its biggest position is one bet on Anthropic’s data centre buildout, held three ways: the Anthropic shares TechCrunch counted in July as a cushion against its public losses, its January Fluidstack shares and this round. A client taking the offer adds to that bet; the 30 September 13F tests whether the fund rebuilt a public book beside it.

I read every Form D from Fluidstack and the fund, the fund’s 13Fs, every landlord 8-K on Fluidstack and Anthropic’s S-1 with its drafts, then summed the leases to $27.0bn across four campuses.

Read in order, they show four things, and the paid chapters a fifth:

• Citadel took most of the public book in July; the private stakes stayed, Anthropic’s at about half the fund.

• The fund led Fluidstack’s January round, and an outside fund had nearly doubled its Fluidstack mark by June.

• Fluidstack’s lease book, the base rent it owes three listed landlords at four campuses, sits inside Anthropic’s buildout, with Google guaranteeing part of it.

• Anthropic’s prospectus prices no commitment to Fluidstack, yet a delay in Fluidstack’s halls could cost Anthropic too.

• The paid chapters settle what a buyer of the round owns: the equity behind the rent, Google’s share and the rent against Anthropic’s.

Source: Anthropic’s S-1 (26 Oct 2026); TeraWulf, Cipher and Hut 8 8-K exhibits (Aug to Dec 2025); author’s sum · as of 30 Oct 2026

What Situational Awareness kept

Citadel bought most of the fund’s public book on 30 July, after the losses in it. Bloomberg then put the fund’s assets at about $10bn and its Anthropic stake at $5bn, as TechCrunch relayed: one source for both figures. Bradley Saacks of Business Insider reported on 14 October that the fund still holds the stake; its size since July is not public.

The last public snapshot, the fund’s 13F for 30 June, a month before the sale, showed $20.24bn. A 13F lists only long US listed securities at a quarter end, options at the underlying’s value, so it never shows a private stake.

Since the sale the fund has filed on two public stakes. A 13D/A of 4 August showed 4.4% of Core Scientific, after sales took it under 5%. A 21.1% block of SharonAI Holdings, a Nasdaq company, bought under a purchase agreement of 17 June, was still held in September.

Its private stakes reach beyond Anthropic and Fluidstack. Bloomberg reported, as Investing.com relayed on 8 August, a $500m investment in Source Foundry, a chip manufacturing startup, and Hedgeweek names a stake in MatX, another private company.

The clients are investors in two funds, Situational Awareness Partners LP in Delaware and Situational Awareness Offshore LP in the Cayman Islands. Their Form Ds report “amount sold”: cumulative capital sold since each fund’s first sale, never what it is worth and never net of redemptions. By 11 September they had sold $9.53bn to 187 investors, about a fifth of it by 10 March.

On July’s reports, about half of every dollar those clients had in the fund sat in Anthropic shares. Marina Temkin of TechCrunch wrote that many would view the stake as “an asset that continues to appreciate” and that a windfall from it “could offset some of the hedge fund’s public-market losses”. That cushion and the round are one exposure, because Fluidstack’s campuses are Anthropic’s buildout.

The fund led Fluidstack’s January round

The fund’s Form ADV of 11 June 2026 lists Aschenbrenner, the founder, as chief investment officer since May 2024 and Carl Shulman as director of research since October 2024, at a firm of eight employees. In June 2024 Aschenbrenner set out the thesis in an essay: “Behind the scenes, there’s a fierce scramble to secure every power contract still available for the rest of the decade, every voltage transformer that can possibly be procured”. Fluidstack signs for that capacity on a lab’s behalf.

Its Form D for the round, first sold on 10 January, shows $842.5m sold to 30 investors by May and names none of them. In July Fluidstack announced an “$830M Series A” at a “$7.5B valuation”, “Led by Situational Awareness”.

The price climbed from January’s. On 14 April Bloomberg reported, as TheNextWeb relayed, that Fluidstack was in talks to raise $1bn at an $18bn valuation. A June Form D shows $730m sold to seven investors; a Form D files no price.

An outside fund marked the rise: Capital Group’s SmallCap World Fund carried its Fluidstack preferred shares, a Level 3 holding with no market price, at $74.09 each on 31 March and at $144.11 on 30 June. On 29 October Bloomberg put the new round at $40bn, “roughly double” the valuation of a few months earlier and “more than five times” January’s.

The managers have a stake in that price: the fund’s SharonAI 13D says its general partner “is entitled to allocations based on assets under management and realized and unrealized gains”, so a higher mark counts before anything is sold. No filing says that shaped the offer, and an outside fund’s marks nearly doubled first.

Nor does any filing say whether clients would buy new Fluidstack shares or some of the fund’s own; Hedgeweek, relaying Bloomberg, reports the fund considered selling stakes in Anthropic, MatX and Fluidstack during July’s margin calls. The fund declined to comment and a Fluidstack cofounder disputed Bloomberg’s report without details. Until a Fluidstack filing shows what was sold, the $40bn rests on one outlet’s sources.

Fluidstack’s leases are Anthropic’s buildout

Fluidstack’s own filings say nothing about its leases; its landlords’ do. TeraWulf, Cipher and Hut 8, all listed, each disclosed its Fluidstack leases in an 8-K exhibit giving the term and the base term contracted revenue, the rent due before any renewal. Summed, those figures are Fluidstack’s lease book.

Barber Lake, Cipher’s site in Colorado City, Texas, is the cleanest case. Fluidstack leased it for ten years in two leases, announced on 25 September and 20 November 2025, for about $3.8bn of base term contracted revenue. Google’s backstop, its guarantee of Fluidstack’s lease payments, covers $1.73bn, under half of it. Cipher’s deck of 4 August describes that backstop as one “determined by rent payments made by Fluidstack”. At the first signing César Maklary, Fluidstack’s cofounder and president, said that “Fluidstack understands what it takes to deliver the compute this moment requires”.

On 25 September 2026 Cipher amended the lease with Fluidstack and signed “a binding commitment with a leading AI lab to lease the facility for an additional 10-year term following the conclusion of the Barber Lake Lease”. That is “a separate lease”, worth about $5.2bn.

One building holds the whole structure: Fluidstack in the middle for ten years, a lab signing for itself after, and Google’s guarantee tied to Fluidstack’s rent. Gary Wu, Fluidstack’s cofounder and chief executive, said when Anthropic announced its $50bn buildout with Fluidstack that “Fluidstack was built for this moment”; the moment both founders invoke lasts ten years, and the decade after it is the lab’s.

Exhibit 2 sums Barber Lake with the other three campuses. Counting the lab’s separate lease as Fluidstack’s would put the book at $32.2bn instead of $27.0bn.

Source: the landlords’ 8-K Exhibit 99.1, Aug to Dec 2025; author’s sum · as of 30 Oct 2026. Base terms only, Lake Mariner as signed; Abernathy’s landlord is a TeraWulf joint venture with Fluidstack, and its term may be cut to 20 or 15 years by option. River Bend is triple net, the rest modified gross.

The filings behind the rows: TeraWulf · Lake Mariner, New York, $6.7bn · TeraWulf joint venture · Abernathy, Texas, about $9.5bn · Cipher · Barber Lake, Texas (two leases), about $3.8bn · Hut 8 · River Bend, Louisiana, $7.0bn.

Three of the four campuses sit in Texas and New York, the states Anthropic named for its buildout with Fluidstack. Hut 8 reports a partnership with Anthropic and Fluidstack for at least 245 megawatts (MW), the size of its River Bend lease.

Google is tied to these halls by its chips and by its guarantee: it sells TPU8i chips to Broadcom, which leases them to Anthropic under a lease of 31 March 2026. The lease even names “the Fluidstack Property”: where Google gives credit support, Broadcom may not unreasonably refuse a site agreement there modelled on Hut 8’s. And Google backstops the rent at all four campuses, taking warrants in two landlords: about 8%, then 14%, of TeraWulf and about 5.4% of Cipher.

No listed landlord has filed a new Fluidstack campus lease in 2026. The 2026 filings naming Fluidstack show amendments, one raising two Lake Mariner leases from 162 to 168 MW, and a group led by Fluidstack agreeing to buy TeraWulf’s 50.1% of the Abernathy joint venture. No landlord filing ties Fluidstack’s other clients, Meta and Mistral among them, as TheNextWeb reported in April, or the cloud capacity it buys from IREN to a campus lease.

One new build sits outside the landlords’ filings: in April Meridian Arc HoldCo, “a joint venture owned by Next Frontier and Fluidstack”, sold $5.7bn of notes. They fund “two new data centers in Sullivan County, Indiana, which will be leased to Fluidstack”, about 430 MW “with obligations supported by Google”. No filing names the end user, and the halls sit outside the $27.0bn lease book.

Anthropic names it once

On 12 November 2025 Anthropic announced a “$50 billion investment in American computing infrastructure, building data centers with Fluidstack in Texas and New York”. Dario Amodei, its chief executive, gave the reason: “Realizing that potential requires infrastructure that can support continued development at the frontier”.

The S-1 of 26 October gives that partner one sentence. It says Anthropic has “entered into a colocation and services agreement with Fluidstack USA I Inc. (’Fluidstack’), under which Fluidstack provides data center development and operations services to us”. The next sentence is the one Fluidstack’s price turns on: “We have also executed data center leases directly and expect to do so increasingly over time”. Fluidstack builds and runs sites for Anthropic, and Anthropic means to sign more sites itself.

The wording moved across seven filings: the partner was a blank in the 1 June draft and Fluidstack by 13 July. From 9 October the sentence sat inside “Anthropic Direct”, Anthropic’s name for leasing chips itself and hosting them “in dedicated data center facilities leased and operated on our behalf”.

The S-1 prices six compute service commitments by supplier, from up to $120.5bn of compute on Nvidia chips from SpaceX.AI to the AI cloud Lambda’s about $41.3bn over terms into 2034, each larger, at its stated figure, than Fluidstack’s whole $27.0bn lease book. Its Broadcom chip leases, about $161.2bn of rent, are priced apart. Fluidstack, the partner named in the $50bn plan, gets no figure.

Anthropic already signs with Fluidstack’s landlords directly: on 6 July 2026 TeraWulf signed Anthropic for 20 years at its Justified Data site in Hawesville, Kentucky, for about $19bn, twice Fluidstack’s largest lease. Paul Prager, TeraWulf’s chief executive, said: “Our strategy is centered on owning and operating critical infrastructure assets, maintaining direct relationships with our customers, and controlling the long-term evolution of our campuses”. The release names no operator for the site.

That lease has not cut Fluidstack out so far: Fluidstack amended its leases at TeraWulf in July and at Cipher in September, and the S-1 keeps it inside Anthropic Direct. The risk is the S-1’s own “increasingly over time”.

One counter would keep the Anthropic stake a cushion: Anthropic gains when it pays Fluidstack less, so the two stakes should move apart. The S-1 points the other way: in March 2026, under an amended colocation and services agreement, Anthropic guaranteed up to $500m of a data centre provider’s power purchase obligations. The S-1 names no provider, and the only such agreement it names is Fluidstack’s.

The S-1 says rent on Anthropic’s Broadcom chip leases “may commence earlier … when the readiness of a data center facility is delayed”. So a delay at a Fluidstack site can start Anthropic’s chip rent early, while what Fluidstack owes Anthropic for a delay is not filed; if the buildout slows, both stakes fall.

The same holds for any neocloud, a company renting out AI computing capacity, whose campuses are one lab’s buildout: Nscale and Lambda, priced in the same S-1, or CoreWeave with its labs. Owning a lab and its supplier is one position, not two, when the supplier’s book is the lab’s buildout.

→ Citadel’s 2006 takeover of Amaranth’s energy book, rebuilt from the Senate record, Trade File No. 2 on Patreon.

Should a client treat the $40bn round as a bet separate from the fund’s Anthropic stake before 31 December’s filings?

Below the paid line:

• Exhibit 3: Google’s guarantee campus by campus, and the equity behind Fluidstack’s rent.

• Exhibit 4: rent per megawatt year on five rows, Fluidstack’s four campuses against Anthropic’s own lease.

• The January Series A marked at the headline price, beside the SharonAI block from the fund’s 13D/A.

• The verdict with our odds, the line that proves it wrong, and four dated checks.

Read more

In April, Citadel’s commodities head, Sebastian Barrack, told a Lausanne summit that 2022 raised margin requirements about 15-fold. His frame is volatility, which markets, he said, failed to price adequately before the US-Iran conflict.

The plainer question is how a hedge fund’s unrated trading arm can stand on the other side of producers’ hedges without the producer posting margin. If Citadel’s gas money were exchange volatility trading alone, its trading arm would need no seat in any producer’s bank line, and producers’ agreements show it holding several.

I read part of the answer in those agreements: the $5 million seat, a lender’s place in a producer’s bank line, held by Citadel Energy Marketing, Citadel’s gas trading arm. The seat would secure any swap the arm writes with that producer. It lets a producer hedge with Citadel’s unrated trading arm without posting margin, and it puts a trading house beside its banks in the collateral.

I searched EDGAR for every document that names Citadel Energy Marketing: none was filed by Citadel.

The chapters follow the gas from the producer to the seat:

• Citadel’s trading arm ranked sixth by US gas volume in 2024; it buys from producers and, since February 2026, has a contract to sell to an export plant.

• Producers’ banks make them hedge, at least 65% of expected output from producing reserves at one line, and the counterparty tests ask for a credit rating the trading arm lacks.

• A $5m seat would secure any swap the arm writes with a producer, and one producer says it would post collateral without that security.

• Before any seat, the arm lent a gas royalty company up to $100m; Citadel has since bought a gas producer’s assets.

• The paid chapter sizes the hedge book one seat sits beside, with every filing in one table.

Citadel buys America’s gas

Citadel Energy Marketing trades gas and power for Citadel, the firm Ken Griffin founded. It bought 1,885 trillion British thermal units (tBtu) of US gas in 2024 and sold 1,803. That ranks it sixth among all US gas market participants reporting on the Federal Energy Regulatory Commission’s Form 552, by Cornerstone Research’s tabulation of FERC’s data. The rank adds gas bought to gas sold, so it measures physical volume and says nothing about profit.

On a second base, the country’s use, Hedgeweek, summarising the FT in June, wrote that Citadel Energy Marketing “traded volumes equivalent to roughly 11% of total US natural gas consumption during 2025”.

Both ends of that book are filed by other companies. At the selling end, Venture Global’s Plaquemines liquefied natural gas (LNG) export plant signed a transaction confirmation with Citadel Energy Marketing on 26 February 2026. By its filing to the Department of Energy, “Plaquemines LNG will purchase natural gas on a long-term basis from Citadel”, for two years at 36.5 tBtu a year, about 2% of what the arm sold in 2024. That is a merchant gas book: gas bought, moved and sold physically.

At the buying end, CNX Resources, a large Appalachian gas producer, names a customer only above 10% of its revenue from contracts with external customers: CNX’s 10% line. It named Citadel Energy Marketing for 2021, 2023, 2024 and 2025. Its 10-K for 2025 puts Citadel third of three that year, at $206.0m. NRG Business Marketing bought $223.2m and DTE Energy Trading $208.8m.

Every credit agreement that seats Citadel Energy Marketing as a Lender is signed for it “By: Citadel Advisors LLC, its Manager“, the hedge fund’s own adviser, so the gas book sits in the fund’s commodities business, which Barrack runs.

Hedgeweek, citing Bloomberg’s sources, put that business’s profit at about $8bn in 2022, half the firm’s total. Bloomberg reported about $4bn, half that, in 2023 and again in 2024. By late December 2025, as gas profits fell, the business was on course for only a slight profit for the year, by Hedgeweek’s same account.

The seats can protect that book; they do not show that it pays.

Margin is the merchant’s problem

A producer that sells gas forward at a fixed price owes the buyer the difference when prices jump. So a spike hands the buyer a large claim on a producer whose forward sale is under water. Margin is collateral posted against a contract’s mark, and it is where volatility reaches cash. If the contract is margined, the producer must post cash at the worst moment; if it is not, the buyer carries the producer’s credit risk with nothing behind it.

Producers’ banks have a say in that hedging. A reserve-based loan is a bank line secured by a producer’s oil and gas reserves, and the Federal Reserve’s guidance on them, SR 16-17, tells banks to consider requiring “hedges for commodity price volatility”. It adds that covenants should make borrowers get “the financial institution’s approval prior to lifting any hedges upon which the institution is relying”.

Some lines turn that advice into a duty. Diversified Energy, another producer, must hedge at least 65% of anticipated production from its producing reserves over the next 12 months. The lines of TXO Partners, oil-weighted by revenue, and of PEDEVCO, another, set floors of their own. A producer under such a line must hedge while its floor binds, and where the bank relies on the hedge, it must hedge with a counterparty the line approves.

The tests are written as credit ratings: BBB or Baa2 for a hedge counterparty in Diversified’s line, BBB+ or Baa1 in CNX’s. Citadel Energy Marketing has no rating from a recognised agency, and in July 2025 TXO’s banks put that in writing.

Other lines skip the rating and name approved counterparties, and the names are merchants. PEDEVCO’s credit agreements list Citadel Energy Marketing as an approved hedge counterparty beside BP Energy, EDF Trading, NextEra Energy Resources and Cargill.

Outside the producers, the July 2023 credit agreement of Temple Generation, a gas-fired power plant, which BKV filed in May 2026, names it a “Qualified Commodity Counterparty“ beside the trading houses J. Aron, Shell, BP Energy, Mercuria and four others. Where a line names no one, an unrated merchant has one door left: a lender’s seat.

What a $5 million seat buys

CNX shows the whole mechanism in two documents. Its Fourth Amended and Restated Credit Agreement of May 2024 signs Citadel Energy Marketing “as a Lender“ with a $5m commitment, the amount a lender stands ready to fund, in a $1.4bn line. Citibank and Goldman Sachs Bank USA hold the same $5m. Three definitions turn that commitment into protection:

1. A “Specified Swap Agreement” is any swap with a counterparty that is “a Lender or an Affiliate of an entity that is … a Lender”, so any swap Citadel writes with CNX qualifies.

2. A swap there includes a “forward contract … relating to Hydrocarbons or power”, so a fixed-price forward purchase of CNX’s gas would qualify too; an index-priced physical purchase would not.

3. The “Secured Parties” include “any provider of a Specified Swap Agreement”, so anything CNX owes Citadel on those contracts is secured by the line’s collateral, ranking equally with the loans’ principal, never ahead.

CNX’s 10-K for 2025 says what that security is worth:

“None of the Company’s counterparty master agreements currently require CNX to post collateral for any of its positions. However … if CNX’s obligations with one of its counterparties cease to be secured on the same basis as similar obligations with the other lenders under the CNX Credit Facility, CNX would have to post collateral for instruments in a liability position in excess of defined thresholds.”

Security on the same basis as the banks is what spares CNX the margin; take it away and the margin comes back. Five million dollars is small change for Citadel, so the seat is easy to read as a courtesy. The definitions ask only whether the counterparty is a Lender, and CNX ties its collateral to that security: the $5 million seat buys a status, and the status does the work.

The seat moves margin rather than abolishing it. If Citadel offsets a producer’s fixed-price sale with futures, a price spike calls cash from Citadel at the exchange, and Barrack said the sector met this year’s shock with “larger capital buffers and more robust funding lines“ after the windfall of recent years.

California Resources, an oil-weighted producer, seated Citadel Energy Marketing in November 2024 for $5m, and Citadel signed again in September 2025. BP Energy, a unit of an oil major, holds the identical seat on that line: merchants used the seat before Citadel did.

TXO shows what the banks agreed to. Its Amendment No. 5, dated 31 July 2025, brought in Fifth Third, Mizuho and Citadel Energy Marketing as New Lenders. Its definition of a Potential Defaulting Lender, a lender that may default, takes in one with “a non-investment grade rating from Moody’s or S&P or another nationally recognized rating agency”.

The banks wrote one sentence for the newcomer, which the agreement calls Citadel:

“Citadel shall not, and at no time will Citadel, constitute a Potential Defaulting Lender solely because it does not have a rating from a nationally recognized rating agency.”

They wrote it “for the avoidance of doubt”: the definition catches a lender with a non-investment-grade rating, and the banks shut out any reading that would treat a missing one the same way. TXO’s line also secures “Lender Hedging Obligations”, those “under any Hedge Contract or any Gas Sales Contract” with a Lender. There the seat would also cover physical gas sales, even contracts signed before the counterparty became a Lender.

Diversified Energy shows the seat most plainly. Citadel Energy Marketing first appears in its lender schedule with the third amendment of 30 April 2026, where it holds a $5m maximum credit amount beside the $5m of Mercuria, a rival merchant.

The line’s hedge counterparty is “(a) any Secured Swap Provider or (b) any other Person” rated BBB or Baa2, and a Secured Swap Provider is “a Lender or an Affiliate of a Lender”. Without an S&P or Moody’s rating of BBB, or a guarantor that has one, the seat was the way in.

The four producers’ filings name Citadel as a Lender, and CNX’s as a customer too, but none as a hedge counterparty: the seats show which hedges would be secured, and none shows that one is live.

A loan, then gas wells

Before any seat, Citadel Energy Marketing made a loan. WhiteHawk Income, a gas royalty company, signed a credit agreement “dated as of August 3, 2023, by and among … the Issuer, as borrower … and Citadel Energy Marketing LLC“. Its S-1 describes “a senior secured term acquisition facility with a maximum amount of $100 million … at 12% per annum”.

That loan did not last. In September 2024, notes bought by two outside groups, Pacific Indemnity and three EIG funds, replaced it, and Citadel became the “Initial Swap Counterparty”. The notes’ hedge test reads “(a) Citadel Energy Marketing LLC or (b) any other Person” rated A-/A3, so the trading arm qualified by name, not by rating.

In May 2026 the notes were amended and restated, and WhiteHawk added a bank-led reserve-based loan in which Citadel Energy Marketing holds no seat. Its existing trades with Citadel stay secured, and that agreement contemplates moving them to one of the line’s lenders. Others took the lending; even the trades may follow.

Then Citadel became an owner. In March 2025 it agreed to buy the Haynesville assets of Paloma Natural Gas, a gas producer, for about $1bn by Bloomberg’s count and $1.2bn by East Daley Analytics’, a quarter to 30% of what its commodities business made in 2024.

Said, March 2025: Bloomberg reported that Citadel would not operate the assets directly. A Citadel spokesperson told it: “Investing in the growth of independent gas producers helps to ensure America’s energy independence,” and “Citadel performs an important role in that effort.”

Reported, December 2025: Paloma, renamed Apex Natural Gas, “currently operates nine rigs in northwest Louisiana“, The Center Square wrote, the state’s most active driller by the Louisiana Oil and Gas Association’s count. In May 2025 East Daley had called Citadel’s long-term strategy “a mystery” and its near-term focus to “increase Haynesville drilling”.

Filed, May 2024 to August 2026: the role Citadel performs for four independent producers is narrower: a seat in each one’s bank line that owns nothing.

→ Citadel’s four seated producers, each one’s hedge book against its bank loans from 2021 to 2025, screened from their 10-Ks, a separate note on Patreon.

How large a book of producer hedges can one $5 million seat put beside the banks, and what does that make the seat?

Below the paid line:

• The size: the hedge book one seat sits beside at CNX, from CNX’s own figures, against the seat’s cost.

• The three seats whose schedules print an amount, on one base, with the arithmetic.

• All ten companies’ filings, by role and filing date.

• The verdict with our odds, the line that proves it wrong, and three dated checks.

Read more

On 8 July 2026 GLG Partners, Man Group’s stock pickers, held a named short of 2.24% in Vistry Group. It was among the last positions Britain’s register published with a name. Man, the world’s biggest public hedge fund in a 2023 Bloomberg headline and famous for its AHL arm’s trend following computers, held more named UK shorts than anyone when the names went dark. Most were GLG’s, and most were new to the register that year.

The regulator’s new totals, one anonymous sum per company, cannot say who has covered those shorts since. A crowded UK short used to name every holder at 0.5% or more. Now a large holder beside it, the kind that matters in a squeeze, has no name.

After July, a manager’s own fund reports are the one regular place a named UK short book still shows: in bands each month, by name twice a year. GLG’s larger UK long short fund, Man Alpha Select Alternative (”Alpha Select”), publishes both. On 26 October I downloaded the short selling files of the Financial Conduct Authority (FCA), grouped every holder’s last named disclosure by firm, followed Man’s companies through the anonymous totals since, and set them against the fund’s reports.

The case runs in six chapters:

• Man held Britain’s most named shorts: 81 still open at the cutoff.

• GLG named most of it this year: 49 of its positions arrived in 2026.

• London chose to hide the names: its totals count every short from 0.2%, unnamed.

• The fund lost 6.61% in July: little of it to company size.

• The totals prove only the cuts: at Kainos, GLG covered at least 1.54 points and may hold nothing.

• Alpha Select’s schedules and size tables: how much of its bet the fund kept.

Man held Britain’s most named shorts

Until 9 July the FCA, Britain’s markets regulator, named every holder of a net short position, measured as a share of a company’s issued capital, of 0.5% or more. An open named position is a holder’s latest such disclosure in a company, notified after 31 December 2020: the population the FCA’s operational guide set for its first totals. Summed by firm, the last named file, with about 600 holders since October 2012, becomes a league.

Man held 81 of the 613 open named positions at the cutoff, in 76 companies, all last updated in 2026. GLG Partners, its discretionary arm, held 62, more than Citadel’s 47 across the whole firm. Man’s systematic managers held the rest: Numeric Investors 13, and AHL Partners, home of the trend following the label rests on, 6.

Source: FCA named file, sheet “Historic Disclosures 10.07.2026”, my grouping by firm; Citadel Securities, a market maker, excluded; Millennium’s own entities (18) and WorldQuant (6), separate firms, counted apart · as of 9 Jul 2026

Read the ranking, which leaves out a Hong Kong manager that also held 23, as a count of positions of 0.5% or more, never as money or as a measure of how bearish a firm is. The line favours concentrated stock pickers: a book spread thinly across many companies crosses 0.5% in fewer of them.

Hedgeweek, Alternatives Watch and Trustnet have ranked GLG at or near the top since 2021, and my count reproduces Trustnet’s 30 for GLG in August 2025. None that I found set a fund’s own reports against the register.

GLG named most of it this year

GLG’s count of open named positions went from 31 at the close of 2025 to 62 at the cutoff. Of the 62, 49 crossed the 0.5% line in 2026. Citadel Advisors, Citadel’s largest named entity, also doubled, from 22 to 44. Marshall Wace fell from 40 in March to 23. A crowd grew, and GLG led it.

Some of the count’s doubling may be growth: Alpha Select grew 26% in the half to £1.79bn, and a fund that grows pushes a short held at the same weight over the 0.5% line.

GLG manages the fund, with Charles Long and Nick Judge as portfolio managers, and the fund’s gross exposure was rebuilt before 2026. GLG wrote in the fund’s 2025 annual report: “Gross exposure increased during the second half of the year, especially in the third quarter, and ended at 193%.” What 2026 added was names on the register.

The fund’s monthly size table splits net exposure by the companies’ market value, in dollar bands, with index positions broken into their stocks. In April 2025 Alpha Select was net long 15.96% of its assets in the giants, companies worth more than $50bn. Below $10bn it was net short 9.10%. By June 2026 the two were 42.97% and 33.38%, with the fund close to flat overall. In pounds, its short side below $10bn grew 1.8 times between the two dates, to about £866m.

What the shorts are for

Call that shape the size tilt: long the giants, net short a third of the fund below $10bn. It is Alpha Select’s bet, not GLG’s: Man Absolute Value, GLG’s smaller UK long short fund (£320m), held the reverse on 30 June, net long 26.44% below $10bn. GLG’s two latest commentaries give a why for the shorts, though neither speaks of company size. The 2025 annual report says that “domestic stocks make up only a small part of our universe and offer exciting shorting opportunities, especially in domestic insurers and retailers.”

The report for the half year to 30 June widened the case to businesses that “need very little physical investment to grow”: “Those are precisely the businesses most exposed to artificial intelligence (AI), because much of what they do can potentially be replicated.”

On the other side, GLG wrote, companies that own physical assets, “utilities, industrial businesses and miners”, are “much harder to displace”. National Grid, Rolls-Royce, Anglo American, Glencore and Siemens were among the fund’s ten largest longs on 30 June.

Not every short is that bet. The 2025 report called one utility short “a partial hedge to the fund’s long position in National Grid”, which was still its largest long in June, at 3.45%. In June two of its ten largest shorts were again UK utilities, while its utilities sector was net long. So the size tilt is a chosen bet, and some shorts beneath it may be hedges.

London chose to hide the names

The FCA’s operational guide put it plainly: “From 13 July 2026, we will no longer disclose individual net short positions. We will instead begin to disclose anonymised ANSPs by company.”

An aggregated net short position (ANSP), what I call the total, sums every position “at or above the 0.2% threshold” in one company and publishes it without names. Funds still report to the FCA from 0.2%; what went is the public line at 0.5%, with a name. The first print, on 13 July, covered positions “held at midnight on 9 July 2026”.

The funds had asked for this. HM Treasury’s 2023 response records their case: “the current regime creates risks around copycat behaviour and short squeezes when individual positions are disclosed publicly”, and “individual position disclosures reveal proprietary trading strategies”. Jack Inglis, chief executive of AIMA, the hedge fund trade body, counted among AIMA’s wins that July “changing short selling rules to protect the confidentiality of investment managers’ positions”.

The register cannot test the first charge. It tests the second only in part: the names gave away positions, while GLG’s reasons, given industry by industry, sat in its own fund reports.

Listed companies argued the other way, the Treasury recorded: “the current individual disclosure regime helps to inform their corporate engagement with their investors”. That engagement now runs on one number per company, and on none where no short reaches 0.2%. The Treasury promised “greater transparency on the overall net short position in a company’s share than the current regime”, and on its own terms it delivered. Each total adds the band from 0.2% to 0.5% the register never printed: more of the crowd, none of the names.

Names still leak in one place: the Takeover Code’s Rule 8.3 makes anyone with a 1% interest in a company under offer disclose its shorts, but a pure short seller never discloses.

The fund lost 6.61% in July

In the three months to 30 September the FTSE 250, Britain’s mid cap index, rose 6.6% against 1% for the FTSE 100. Graeme Evans of interactive investor gave one reason: “Takeover action played a big part in the robust performance of the FTSE 250 index”. He gave another: investors “returned to heavily sold stocks” such as Hays and WPP.

Both reasons reach Man’s named list. Four of the FTSE 250’s ten best performers in the quarter, the top two among them, were on it at the cutoff: GLG’s Hays, Michael Page (PageGroup on the register) and Kainos, and Numeric’s Pinewood Technologies. Pinewood rose 67.7% in a quarter that brought a takeover. Yet Alpha Select held Hays and Michael Page long on 30 June: GLG’s named figure is its net across all its funds, so one fund can sit on the other side.

Alpha Select’s DL share class, the one its monthly reports track, lost 6.61% in July, its worst month of 2026, and 5.58% over the three months. Man’s own trading statement has its IL share class down 5.4% for the quarter, about £97m on the June assets before flows, against a 0.1% loss for the HFRX Equity Market Neutral index. Money came in, and the fund ended September at £1.79bn, where it began.

Around Alpha Select the firm had a strong quarter: Man’s assets rose 7% to $270.9bn on the same statement, so a £1.79bn fund’s bad July barely moves the firm’s numbers.

July’s movers cut both ways: two of the fund’s 30 June shorts, Vodafone and Sage, rose 18.3% and 19.3%, among the FTSE 100’s five best, and AstraZeneca, its second largest long, fell 10.4%. Its Hays and Michael Page longs rose 80.9% and 70.9%.

What company size explains

How much of the quarter is the size tilt? At June exposures, with the FTSE 250 standing in for the smaller companies, the FTSE 100 for the giants and about 3% assumed for the middle band, $10bn to $50bn, where the fund was net short 10.21%:

So the tilt explains roughly 2 of the 5.58 points for the quarter and leaves the rest to the stocks themselves. The estimate is rough: the size bands cover the fund’s whole book while the two indices are British, and its net short in UK companies was 10.35% in June.

It explains little of July: the FTSE 250 rose 4.2% that month and the FTSE 100 3.5%, a gap that costs about 0.3 points at the same exposures. Most of its estimated cost fell in August and September, when the fund made 0.50% and 0.60%, so July’s loss came from the companies it held.

GLG’s commentary covers the half to 30 June and never mentions July. It reached investors in a report the fund company’s directors approved on 26 August, after the loss. It tells them: “Investors should also expect that holding positions that differ from the consensus, in a market of this kind, will produce periods of sharp volatility in returns.”

The totals prove only the cuts

GLG was named short 1.75% of Kainos on 6 July, and at the cutoff every named position in the company added up to 2.43%. The first print put Kainos’s total at 4.46%, above the names, because it counts every short from 0.2%. The total first fell below GLG’s 1.75% on 21 August, at 1.68%, and rose back to 1.86% on 26 August. The total that replaced it on 22 October reads 0.21%, dated 22 September.

Reading a total against a fund’s last named position takes four steps:

1. The total sums every short of 0.2% or more in Kainos.

2. If GLG still holds 0.2% or more, its short sits inside that sum, so it cannot exceed the total.

3. If GLG holds less than 0.2%, it holds less than the total already.

4. Either way GLG holds 0.21% at most, so it has covered at least 1.54 points of its 1.75%.

When today’s total sits below a fund’s last named position, I call the fund capped: it holds the total at most. A fund is below the line, under 0.2%, when its company is still on the reportable list but carries no total. The same holds when the total is dated before the fund’s last disclosure, since the FCA’s Handbook dates a total by “the most recent position date included in a notification”.

Each fall below a named position is a dated proof; each rise is anonymous. The test cannot say whether GLG holds any Kainos at all. To redo it, read each last named position against the totals and the reportable list; a join on the totals file alone drops the companies with no total, which are among the strongest proofs.

GLG’s named figure nets all its funds, and where a total falls but stays above it, the totals cannot tell GLG from the crowd. That objection holds against the register, and fails, for one fund’s slice of the book, against Man’s own documents.

The totals fall as often in Citadel’s companies as in Man’s: the totals measure the crowd, the fund’s reports measure the fund.

In Alpha Select’s reports one number decides the question: its net short in companies below $10bn on 30 September, against 33.38% of the fund in June.

→ GLG’s Vistry short and the 17 other holders named beside it, costed from the FCA register and London closes and set against 44 UK crowds since 2012, a separate note on Patreon.

After a July that cost it 6.61%, how much of its bet against smaller companies did Alpha Select keep?

Below the paid line:

• The schedule table: Man’s named companies marked where GLG’s own fund was short or long.

• The size table: the fund’s exposure by company size before and after the quarter the names went dark.

• The tests: every company where the totals prove GLG cut, and the control table on what they cannot say.

• The verdict, the line that would prove it wrong, and four dated checks to watch.

Read more

Optiver’s shareholders, founder Johann Kaemingk among them, took €761 million in dividends in 2024. On 19 October 2026 a Bloomberg headline read “High Frequency Trader Optiver Seeks $900 Million Debt Package”. The Amsterdam market maker is seeking a debt package about the size of what it paid its owners in one year.

A reader of the headline might take a trader that borrows to be short of money, yet Optiver’s 2025 profit was a record and it calls its capital structure “conservative”. I read the package as Optiver borrowing to grow rather than paying its owners less, and its 2026 filed accounts will test that in the payout, the share of the year’s profit paid as dividends. That share sets how fast the equity behind Optiver’s quotes grows, and its lenders and clearing firms lean on that equity.

So why borrow at all? Optiver says in its Annual Review 2025 that it “trades on its own account, deploying its own capital across markets”. Quote, the Dutch magazine that reads its filed accounts, called the equity left after the 2024 dividend “zeer comfortabele” (very comfortable). Both describe a firm with money of its own, and neither says how the growth is paid for.

The owners take a large share of each profit, while the spending turns to computing power, a New York build out and an energy trader. Call the result pay and borrow, in which growth the owners could fund by taking less is borrowed and the owners’ share stays untouched. It breaks if the payout falls below 40% once the debt arrives.

Quote and FD, the Dutch financial daily, read the dividend from Optiver’s filed accounts in eight of the years since 2015; I set each against that year’s profit and cross checked them against its equity from 2020 to 2025. Both checks find about half of the profit going out. I also read both of its US arm’s audited balance sheets at the SEC.

The case runs in six steps, the last two behind the paywall:

• Optiver’s profit has more than quadrupled since 2019, and its spending has turned to computing power.

• At the end of 2025 its US arm’s hedged book ran on cash owed to its clearing firms.

• About half of each profit goes out, Optiver’s own equity agrees, and part of 2025’s apparent dip can be the dollar.

• Flow Traders, the listed Amsterdam market maker, stopped its dividend before taking private credit; Optiver’s record shows no such stop.

• What Optiver really keeps of its profit, and what one lean year would cover.

• Whether the owners take less once the debt arrives, and what would prove us wrong.

Source: Author’s calculation from Optiver’s filed annual accounts, years 2015 to 2024, as Quote and FD reported them · as of 31 December 2024

The payout stays at about half or more while profit rises from €279m to €1.37bn.

Each year keeps its source’s base: dividends for that year’s profit in 2015, 2018, 2019 and 2020, dividends paid during the year in 2024, and no base stated for 2016, 2017 and 2021. For 2019, the payout multiplies the €180 a share Quote reported by the 1.46 million to 1.5 million shares its figures imply. Each row is a press read of that year’s accounts, mostly published the next spring: Quote’s for 2015, 2016 and 2017, 2018, 2019, 2020 and 2024, and FD’s for 2021.

How Optiver makes its profit

Optiver quotes prices to buy and to sell, trades on those quotes and manages the inventory this leaves by hedging each position against others. What its trading income leaves after costs is the profit the owners divide, and the part they leave in adds to Optiver’s capital.

Profit was €397m in 2019 and more than tripled to €1.38bn in 2020. It fell back in 2021 and again in 2023, then rose to €1.77bn in 2025. If Optiver were borrowing because it is short of profit, its releases would show profit falling; they show a record, which Quote’s headline called “duizelingwekkende” (dizzying).

Jan Boomaars, the chief executive who signs Optiver’s annual reviews, wrote in the 2025 results release that “Our ambition is clear: to continue strengthening our position as a global liquidity provider, without compromising risk discipline.” The strengthening is where the money now goes.

The word that marks the turn in how Optiver describes its spending is “compute”, meaning computing power, which appears five times in the 2025 Annual Review and in none of the 2022, 2023 or 2024 reviews, which come no closer than “computer science” and, in 2022, “big data computing”. The build itself is older: on 7 May 2024 AMD quoted Alexander Itkin, Optiver’s US chief technology officer, on “building world-leading computational scale”, so the count dates the description; the spending began earlier.

Boomaars’s letter says “we are investing heavily in compute, data and global platforms”. The same review records “a new office in New York” and trading “in new asset classes”, U.S. credit among them, and in August 2026 Optiver agreed to buy a majority of Northpool, a Leiden power and gas trader.

Optiver prints no figure for any of these, and each is paid for before it earns. Under pay and borrow, that money comes from the profit the owners leave in or from lenders, and Optiver’s US arm already runs on one kind of lender.

Its US book runs on credit

Optiver US LLC, the group’s market making US broker dealer, has two statements online at the SEC, for 2018 and 2025, both audited by KPMG.

At the end of 2025 the arm owned $66.2bn of securities and had sold short 97% as much, most of both in options. Management writes that “market risk is substantially diminished when all financial instruments are aggregated” (note 7), so the gross figure, as in our Hudson River Trading piece, measures the size of a book largely hedged within itself and says little about its net exposure.

Note 4 shows the funding: at end-2025 the arm owed brokers, dealers and clearing organisations a net $916.1m, a line that holds $1.44bn of cash, partly offset by open trade equity (gains on positions still open). Call the cash the clearing credit, the cash the arm borrows from the firms that clear its trades, an affiliate among them (note 9). It is about 1.8 times the arm’s own equity of $803.4m.

The two year ends online differ in direction. At end-2018 the arm’s clearing firms owed it $138.6m, against $39.1m it owed them. At end-2025 they owed it $21.5m, against $916.1m. At those two dates the arm went from net creditor of its clearing firms to net debtor.

Optiver describes its trading as deploying its own capital, but at end-2025 its US book owed its clearing firms that clearing credit, so the package would add a second kind of lender, one that could term some of that credit out.

A capital squeeze is the other reading, and nothing Optiver publishes shows one. The US arm held $571m of net capital at end-2025, the liquid cushion the SEC’s Rule 15c3-1 counts, against a $250 thousand minimum. On 12 January 2026 it sent $40m of capital up to its parent, which fits no squeeze at the arm.

Optiver’s EU and UK investment firms publish their capital disclosures only “upon request”, its disclosures page says. Under the EU rules, a senior loan counts as funding, and only debt ranked below other creditors counts as capital. Bloomberg’s headline does not say how the package ranks, so it cannot be shown to add to those firms’ capital.

Half of each profit goes out

Optiver files its full annual accounts with the Dutch Chamber of Commerce, the KvK, and Quote and FD have reported the dividend from them in eight of the ten years from 2015 to 2024. For 2024 they show €761m paid in dividends against that year’s net profit, a payout of 56%.

Optiver kept paying a dividend even as capital rules tightened, FD’s headline on the 2017 results noted. Quote wrote in April 2019 that “Zoals gebruikelijk in beurshandelkringen wordt een flink gedeelte van de jaarwinst uitgekeerd aan de aandeelhouders” (as is customary among trading firms, a large part of the year’s profit is paid to shareholders).

Optiver has kept more before. Its 2013 accounts, as Quote quoted them in July 2014, said “In 2014 zullen we nieuwe kapitaalbuffers opbouwen die gebruikt kunnen worden wanneer verliezen worden geleden” (in 2014 we will build new capital buffers for use when losses are suffered), under a capital policy to be put to the Dutch central bank, and the dividend proposed out of 2015’s profit was “minder dan de helft” (less than half).

I found no report of the dividend for 2022, 2023 or 2025: for 2022 Quote saw only a shortened report, and in April 2026 it did not know the exact 2025 amounts. The KvK accounts carry each year’s dividend and were not read for this piece.

Boomaars runs the firm with Sandrine Teran, the chief financial officer, and Tadhg O’Shea, the chief risk officer, but the owners decide the payout. Quote wrote on 2 April 2026 of “de Quote 500-lieden die als aandeelhouders beslissen wat er met de winst gebeurt” (the people on Quote’s list of the 500 richest Dutch who, as shareholders, decide what happens to the profit), among them Kaemingk and a fellow shareholder, Chris Oomen.

Senior staff can also buy shares, which Optiver Holding, the group parent, must buy back from those still employed five years on, the US arm’s 2018 statement says. Many staff hold shares and share in the profit through the dividend, Quote noted in April 2023, so part of the payout reaches staff.

The equity bridge agrees

Optiver’s own releases cross check the missing years through an equity bridge, in which closing equity is opening equity plus profit, minus dividends and staff share buy backs, plus staff share issues, plus or minus currency on foreign subsidiaries. The rise in equity against profit, the share kept, mirrors the payout, blurred by share trades and currency. From 2020 to 2025 Optiver earned €7.92bn and its equity rose €3.96bn, from €1,532m at end-2019. It kept 50%, in line with Exhibit 1.

2025 alone looks different, because equity rose by only 33% of the year’s profit, which at face value says Optiver paid out two thirds. That is the pitfall: much of Optiver’s equity sits in subsidiaries outside the euro, its 2023 review says, and the dollar fell against the euro over 2025 on the ECB’s rates, so part of the gap can be currency.

The releases and Annual Reviews have described “a conservative capital structure” every year from 2019 to 2025, and the numbers it publishes agree. At a payout of about half, faster growth needs lenders or a smaller dividend, and pay and borrow takes the lenders.

Other market makers borrow too

Flow Traders took both. It measures its trading capital as net liquidity at clearing and prime brokers plus bank cash. In July 2024, a month after taking a small bank loan, its board announced a Trading Capital Expansion Plan, and its Annual Report 2025 says that “To expand the trading capital base, the Board suspended regular dividend payments until further notice”. In October 2025 a $200m private credit term loan, beside an undrawn $75m revolver, replaced the bank loan, at SOFR plus 500 basis points.

Trading capital rose by about a third in 2025, and the loan’s covenant keeps it at three times the loan or more, “on a daily basis”. The board wrote that “The use of external leverage has shown to be accretive to shareholders’ return on equity and is expected to continue to do so given the opportunities ahead.”

The bigger trading capital is on the record, much of it the loan itself; accretion is the board’s word, and the loan’s price is the bar each year’s trading must clear. Optiver’s record to 2024 shows no such stop, and it is now seeking the debt anyway.

Set against each firm’s own equity, with clearing firm credit left out, the borrowing differs in kind as well as size. Optiver’s package alone, leaving out any borrowing its KvK accounts may hold, is 14% to 15% of its end-2025 equity. Flow Traders’ two facilities came to about 27% at end-2025, 20% drawn.

Virtu Financial’s whole stock of long term borrowings was about 103% at end-2025. Jane Street’s one August issue, $14.6bn of senior secured notes raised to retire about $11.1bn of its debt, came to about 27% of the members’ equity S&P put at about $55bn for March 2026, and about 6.4% net of the debt it retires.

Set beside those, the package cuts against any reading of strain, and Optiver’s equity rose in each of the last six years.

In 2022 Moody’s wrote that Citadel Securities planned to use a term loan “to increase its trading capital”, the use Flow Traders put its own loan to. The payout decides how much of the growth the owners fund by taking less, and Flow Traders’ loan shows what the lenders’ share costs.

Once Optiver can borrow, does its 2026 payout fall, or do its owners keep taking their share of every good year?

Below the paid line:

• What Optiver kept of last year’s profit, and how much of the gap currency can explain.

• The package against a lean year’s profit, against what Optiver keeps, and against Flow Traders’ covenant.

• The verdict on pay and borrow, our odds on the next payout, and what proves us wrong.

• Four dated checks, from Northpool’s completion to the KvK filing that settles the bet.

Read more

Squarepoint’s US adviser has told the SEC since at least March 2025 that it reserves shorter-term signals for some funds. Its July 2026 brochure repeats the tests for them: higher fees, far more staff money.

Hedgeweek says most of Squarepoint’s capital comes “from partners and employees”, which suggests staff and outsiders ride together. The filing shows one master fund they do not share, 100% owned by the firm and its people. On my reading, outsiders get the slower of two Squarepoints: their funds lack the brochure’s “shorter-term signals”, which work only at limited size. So holders of Focus and Atlas, and perhaps Core, the funds outsiders buy, would trade second, at prices the brochure expects to be less favourable.

I read four of Squarepoint’s Form ADVs, from December 2022 to August 2026, and five rivals’, fund by fund; they make the case in five steps:

• Four partners own most of the firm, whose regulatory assets, a gross measure, are three times those in its December 2022 filing.

• One master fund, with 390 investors, is owned entirely by the firm and its people.

• The brochure reserves the shorter-term signals for funds with higher fees and far more staff money in dollars; Focus and Atlas carry the least, and Core is the open question.

• A number of its principals own most of STG, an options market maker that shares Squarepoint’s offices and research.

• Whether any of five rivals runs a fund its own people own on this scale, and which way staff ownership has moved since the December 2022 filing.

Bars are gross assets, which count borrowing and the funds’ holdings of one another; labels give the share its people own.

Four partners own most of it

Squarepoint began as nQuant, a team that moved from Lehman Brothers to Barclays in 2008, by the firm’s own account; it now calls itself “a privately held firm led by its four founding partners”. Its brochure names its majority owners: “Olivier Durantel, Antoine Fillet, Maxime Fortin and Grégoire Schneider”. None of the four has given an on-record interview that I could find.

On 16 December 2014 entities owned by its managers bought the business from Barclays, the brochure says. That same day, Squarepoint International Feeder Ltd, an offshore feeder of the master fund its staff now own outright, made its first sale. The US feeder of Squarepoint Core Master Fund Ltd first sold on 1 October 2015.

Its adviser, Squarepoint Ops LLC, which with its affiliates forms what its brochure calls the Squarepoint Group, runs all 28 of its accounts with discretion, so one firm decides which fund gets what it trades. Its regulatory assets under management, a gross figure that nets no borrowing, were $227bn on 31 December 2025, three times the figure in its December 2022 filing.

Growth puts more money in line for signals of limited capacity, and the brochure’s rule decides which funds get them: their capacity, it says, “has been reserved for the Higher Fee Funds”, the funds that pass those tests.

Staff own one fund outright

Form ADV gives each private fund its own row, in Schedule D, section 7.B.1, and Squarepoint’s, read fund by fund, sort into two kinds. Question 14 asks what share of a fund the adviser and its related persons own. The SEC’s instructions count among them “all of your current employees (other than employees performing only clerical, administrative, support or similar functions)”, so I call that share the insider share.

If staff and outsiders shared funds, the staff’s biggest stake would sit in a fund outsiders also own. In the August 2026 filing it sits in one they do not: Squarepoint Master Fund Ltd (the staff fund) reports $57.5bn of gross assets, 390 owners and an insider share of 100%. It is itself a fund of funds that invests in other funds Squarepoint manages, which the rows do not name, and no fund of funds owns any of it.

Squarepoint Core Master Fund Ltd (Core) is larger, at $77.3bn of gross assets, with 226 owners and an insider share of 36%.

Squarepoint Focus Master Fund Ltd (Focus) has $39.9bn of gross assets and an insider share of 14%. Squarepoint Atlas Master Fund Ltd (Atlas) has $25.6bn and 20%. On Question 14, outsiders own most of all three, so I call them the outside funds.

The rows show insider money in a fund of its own, and staff a minority, on Question 14, in every outside fund. The one quantity they leave open is staff money through the staff fund: funds of funds own between 13% and 37% of each outside fund, and the filing does not say whose those stakes are.

Citadel’s people sit with outsiders instead: on Citadel Advisors’ filing of 11 June 2026, Citadel Wellington LLC, with $30.9bn of gross assets, is 53% owned by Citadel and its people, a majority by share inside a fund outsiders also hold.

So an outside investor’s own number is the insider share of the fund he holds, and the brochure sets out what each fund gets for its money.

Higher Fee Funds get shorter-term signals

Item 6 of the brochure gives the adviser’s rule for which fund gets which signal. It defines the Higher Fee Funds by two tests. They charge “higher performance-based fees or performance allocations than other Funds, in percentage terms or in absolute dollar amounts”, and “Squarepoint staff have significantly more capital invested in those Funds” than in others. Then it says what they get:

“The Higher Fee Funds have access to shorter-term signals that are not made available to other Funds because the signals are believed to be capacity constrained, and their capacity has been reserved for the Higher Fee Funds.”

In plain terms, the short horizon signals have room for only so much money, and the firm keeps that room for the funds where its staff have the most at stake. The mechanism runs in four steps:

1. A short horizon signal predicts a price move over a short window; trade it with too much money and the fund’s own orders move the price it is chasing, which is what capacity constrained means.

2. Squarepoint gives that scarce capacity to the Higher Fee Funds, which also “have priority with respect to new strategies, methods, signals, and other opportunities”.

3. On execution, the brochure adds that “the Higher Fee Funds generally reach the market ahead of the other Funds. This is likely to result in the other Funds having their orders filled at less favorable prices.”

4. The brochure says the Higher Fee Funds “have generally experienced, and are expected to continue to experience, higher Sharpe ratios and higher gross returns than the other Funds”.

The signal rule splits the funds two ways, the Higher Fee Funds against “other Funds”. The fund rows show no fees, and the staff test is in dollars. On gross assets, the staff fund carries about $57.5bn of staff money on Question 14, and Core about $28bn. Focus and Atlas carry $5bn to $6bn each. I read the staff fund as a Higher Fee Fund, though the brochure also lists “proprietary funds” beside the tier, so it could sit outside it.

If Core is one too, which its staff money allows, its outside owners sit in the faster tier, while Focus and Atlas carry the least staff money of the four. The brochure names none, so which fund is which stays my reading. Item 6 gives the gap no number, only that one fund’s impact on another “may be material”; any cost of trading second falls on about $66bn of gross assets in Focus and Atlas, and up to $143bn with Core.

The faster tier is not free on paper. The Higher Fee Funds carry higher performance fees, in rate or in dollars, and Item 5 says certain shares restricted to staff bear other products’ expenses “on a greater than pro rata basis”. The brochure prices neither as the cost of the signals.

A further clause lets the firm copy one fund’s human traders into another, including from “other Funds in which Squarepoint Group personnel have substantial proprietary capital invested”, and the copying can run “In either direction”.

The brochure says the copy “is expected to underperform”, at times “materially”, and that the adviser “will continue to charge performance-based fees and management fees” on it. The receiving fund pays full fees for a copy the firm expects to trail.

The rule is not new: the July 2026 brochure lists no material change to Item 6 since the March 2025 one. Item 10 names another business, one that some of the principals own themselves.

Some principals own a market maker

Item 10 puts it in one sentence: “A number of the Squarepoint Group’s principals and their related persons own, directly or indirectly, a substantial majority of a separate commercial trading business, STG group”.

Its broker-dealer, STG Securities LLC, “principally operates as a market maker and liquidity provider with respect to options”. It “does not perform any services, including execution and brokerage services, for clients or the Funds”. It “may, on occasion, act as a counterparty to the Funds” on orders with a US options leg, which “will be handled by a third-party broker”.

STG Group was founded in 2022, and on 30 September 2025 its holding company bought Hull Partners LLC, now STG Securities.

FINRA’s BrokerCheck shows Kirill Gelman running it as chief executive since September 2025, with Eric Levine as head of trading, and STG Securities’ audited accounts show $7.15m of member’s equity on 31 December 2025: a startup’s balance sheet.

STG’s press statement of March 2026 presented the launch as a clean break: “STG Group was established by principals at Squarepoint and is separate from its investment management business”. Hedgeweek, citing that statement on 5 March, reported that “the new broker-dealer will operate independently from Squarepoint’s hedge fund operations and trade securities using its own capital”.

The brochure filed in July 2026 lists the ties:

• the two groups “share office space, technology, infrastructure, research, data and other resources”;

• “certain senior management of Squarepoint Group are involved in strategic, governance, operational and/or other management decisions or activities related to the Squarepoint Group and the STG Group”;

• the Squarepoint Group “expects from time to time to enter into lending arrangements … between Higher-Fee Funds and other funds managed by the Squarepoint Group and entities affiliated with the Squarepoint Group, including the STG Group”.

BrokerCheck adds that a Squarepoint services company has kept part of STG Securities’ books and records since 18 November 2025. STG’s statement and the brochure, which says the two “generally operate as separate businesses”, agree on “separate” and part at “independently”.

The brochure answers the obvious worry in one line: “The Squarepoint Group and the STG Group do not coordinate their trading activities.” Uncoordinated is not harmless. It adds that STG’s trading “may inadvertently have adverse impacts” on the funds, that “any such adverse impacts could be material”, and that “Increases in overlap and competition in business activities will exacerbate the impacts described above”.

On paper STG sits outside the adviser: its Form BD says it is not “under common control with” any securities or advisory firm, and the tie runs through “a number of” principals rather than through the adviser.

eFinancialCareers made the comparison on 5 March: STG “will have a similar relationship to the fund that Citadel Securities has to Citadel”. The lending clause, which that comparison leaves out, brings STG closer to that shape.

So STG touches an outside holder in three ways. It shares Squarepoint’s research, its own trading can move the prices and liquidity his fund meets, and on occasion it may take the other side of his fund’s options orders through that broker. The lending clause touches the funds where staff money sits, not his. STG is separate in law and close in practice.

Where will Squarepoint’s next brochure, due by the end of March 2027, leave a holder of Core: behind the staff fund, or beside it?

Below the paid line:

• The screen: the largest fund each of six firms’ people own, side by side, and where Renaissance’s Medallion sits.

• The series: each fund’s insider share across three filings, and the funds of funds’ stakes in dollars.

• The verdict, my odds on which fund gets the signals, and the line that would prove me wrong.

• What to watch: three dated checks on STG and Squarepoint’s next annual ADV.

Read more

In 2025, Jane Street Europe, the firm’s London trading company, nearly tripled its trading gains. Its return on capital went from 17.1% to 17.7%, by the measure its own accounts define, applied to the company alone, without its Indian and Korean subsidiaries.

In other words, since 2023 Jane Street Europe has earned a steady return on its capital, measured as operating profit over capital employed (total assets less current liabilities). Its 2026 accounts, due by 30 September 2027, should show it again.

That matters to anyone reading Jane Street’s London accounts: the trading company’s profit measures the capital it puts up, and London’s trading sits in a bill. Since 2023 its return has been no gauge of how London traded.

In July, Elias Ward of btw.media read the same accounts the other way. He set the 2025 falls of the London group, the company with its subsidiaries, a return on capital down from 34% to 19% among them, beside the company’s own figures. He wrote that “a lower return on capital employed shows that more capital does not automatically mean more value”. He read its independence as “a costed bet that internal control over technology and capital produces better trading outcomes than outsourced scale”.

The London group’s return did fall. The fall sits in its subsidiaries in India and Korea, whose trading gains, Korea’s included in 2025, dropped by about three quarters in a year that included an order by India’s market regulator. The company itself employed more capital in 2025 and earned a higher return on it. One question decides between the readings: does the trading company’s own return move with its trading? From 2023 to 2025 it did not.

I read the accounts Jane Street Europe and its London partnership have filed since 2022 and recomputed the company’s return from its own definition. I also matched the charge it books against the partnership’s revenue line “amounts charged to other affiliates”: they agree within 6% every year.

The case runs in six chapters:

• Since 2022 the trading company has held London’s positions, and the partnership its people.

• The trading company’s return has held near 17.5% while the charge moved with its gains.

• London’s pay nearly doubled in 2025, part of it a promise paid later.

• The largest partner, a Delaware Jane Street company, was entitled to $1.56bn of 2025’s profit.

• India carries the London group’s fall, and a $539m escrow with no provision.

• The 2026 accounts should put the return inside 13% to 22% again.

Source: Author’s calculation from Jane Street Europe Limited accounts FY2022 to FY2025, the revenue note (note 2 in FY2022 and FY2023, note 3 in FY2024, note 4 in FY2025) and the balance sheet, company column; return is operating profit over total assets less current liabilities · as of 31 Dec 2025.

Read it down: from 2023 the return stays within a point while the gains swing, and 2022, the year London’s people moved out, sits above. The swing went into the partnership’s charge, which rose from $202.7m in 2024 to $1.67bn.

London moved its people in 2022

The move began with a rewrite. Jane Street Europe’s 2022 accounts say: “During the year ended 31 December 2022, the Group’s Transfer Pricing Arrangements were revised and the Company entered into a UK Services Agreement.” Transfer pricing is what one company in a group charges another. From 1 January the company “no longer receives an allocation of employee costs”. On 1 August “all of the Company’s employee contracts were transferred to JSUP”, under the rules that carry employment contracts across unchanged when work moves to a new employer.

JSUP is Jane Street UK Partnership LLP, a limited liability partnership formed in October 2021. The accounts give no reason beyond the revision, and warn that 2022’s revenues are “not directly comparable” with earlier years.

What stayed in the trading company is the trading. Its revenues “are attributable to income from movements in the fair value of financial instruments, trading profits and losses”, plus charges to Jane Street’s US parent and from other affiliates. That is narrower than the usual answer, market making on enormous volume, and fees run the other way: commissions and fees, and interest and dividends, sit as costs inside its 2025 revenue line.

The partnership’s accounts describe its job in one line: “The principal activity of the LLP is the provision of certain operational services to its UK affiliates”. It employs London’s people, seconds them out and bears the trading company’s audit fee and future lease payments. It averaged 790 people in 2025 of Jane Street’s roughly 3,500, and its two designated members are Delaware companies: Jane Street Group, LLC and Jane Street Global Trading, LLC.

For the whole firm, Reuters reported Jane Street’s 2025 net trading revenue at $39.6bn, against roughly $12.3bn at Hudson River Trading and $12.2bn at Citadel Securities, none of it filed.

Since 2022 the trading company has held the positions and none of the people, and the partnership the people and none of the positions. The charge between them, labelled in 2022 “Amounts charged from JSUP”, is where London’s trading shows up.

London’s company earns a steady return

The trading company’s accounts name the link: it “receives/pays compensation for all trading activities performed on behalf of the Group in accordance with a trading services agreement … and the Group’s transfer pricing arrangements”. The machine runs in four steps:

1. London’s people, employed by the partnership, trade positions booked in the trading company or elsewhere in Jane Street.

2. The gains land in the company that holds the position.

3. The partnership charges that company under arrangements whose formula the accounts never state.

4. The trading company’s revenue line is left with the rest; the partnership spends the charge on its people and allocates what remains to its members.

The 2025 revenue note is the worked case, set out in Exhibit 2.

Source: Jane Street Europe Limited accounts FY2025, note 4 p29, group column except the first line, which is the company’s own; the India and Korea line is the author’s calculation, group less company · as of 31 Dec 2025.

The lines sum to the filed revenue within rounding; the second line is why the London group’s column misleads.

In the company’s own column the charge took 71.5% of its gains, and its revenue line was left with $498.1m. Its operating profit was $491.3m on capital employed of $2,770.7m, which is 17.73%. The accounts define the measure themselves: “Return on capital employed is calculated by dividing operating profit by total assets less current liabilities”.

To redo it, take Jane Street Europe’s accounts for 2022 to 2025, use the company column rather than the group’s, and apply that definition each year. Then set the charge beside the partnership’s revenue line “amounts charged to other affiliates”, as Exhibit 3 does. The trap is the London group’s column, which carries India and Korea: on it the 2025 charge looks like 67% of the gains instead of 71.5%, and the return falls with India’s gains.

Source: Jane Street Europe Limited accounts FY2022 to FY2025, company column; Jane Street UK Partnership LLP accounts FY2022 to FY2025, the revenue note (note 2 in FY2022 and FY2023, note 3 in FY2024 and FY2025), its FY2022 a 14-month period · as of 31 Dec 2025.

Read across: one bill seen from both ends, then the two inputs of Exhibit 1’s return.

The obvious rival is a cost plus fee, which would give the partnership a small, steady markup on its costs and leave the trading company’s return to swing. The partnership’s profit has instead ranged from 84% to 118% of its costs; the steady number is the trading company’s, though from 2023 its gains slipped 8% and then nearly tripled.

What the four years show instead is a return that follows the capital. From 2023 to 2025 it stayed inside a band of about two thirds of a point. In each of those years the charge moved by 97% to 125% of the change in the gains.

It is the shape left by a method that pays the capital a return and leaves the rest to the people: a split that gave the company a real share of what remains would have lifted its 2025 return, which barely moved. That is my reading of the filings, which disclose no method.

KPMG has identified a fraud risk over the transfer pricing arrangements in every audit of the trading company since 2022, at first tied to the changes in them. The 2025 report, signed by James Voyle as senior statutory auditor, words it this way:

“On this audit we do not believe there is a fraud risk related to trading revenue because of the nature of the transfer pricing arrangements used by the Company, as well as the systemised nature of trading revenue streams. Given the complex and inherently judgmental nature of transfer pricing arrangements … we have identified a fraud risk over the transfer pricing arrangements.”

It is a risk the audit tested, and each year’s opinion says the accounts “give a true and fair view”. The trading is systemised; the judgement, and London’s year, sit in the bill.

London’s pay nearly doubled

The partnership’s biggest cost is its people. Its 2025 payroll was $1.66bn, up 93.5% on 2024. eFinancialCareers’ Alex McMurray, working from the same accounts, headlined “Jane Street’s UK pay: $1.9m per head, $55m if you’re a partner“ on 16 September, and my reading matches it: salaries alone, $1.48bn over 790 people, come to about $1.88m each.

From 2023 to 2025 headcount rose by about a quarter while payroll more than tripled.

The 2025 accounts carry a sentence the 2024 set does not: “In addition, the LLP provides other long-term benefits to participants in a compensation plan. The estimated value of future payments related to these benefits are recognised as an expense within staff costs.” A new line arrives with it, $287.0m of accruals falling due after more than a year, where the 2024 balance sheet showed none.

So part of 2025’s pay is a promise booked now and paid later. The accounts do not say how much of that balance was expensed in 2025; if all of it was, payroll still rose about 60%. The filing does not say what the plan is.

Nor did 2025’s jump come from people moving in: every contract moved in 2022, and in 2025 headcount rose only about 15%.

Per head, London’s payroll came to $2.11m in 2025, against a Jane Street average for 2025 of about $2.7m that Bloomberg reported, on a different base: a reported pay pool spread over about 3,500 people. Pay rose most in the year the charge did, and what it leaves goes to the partnership’s members.

The biggest partner is in Delaware

The charge lands in the partnership as revenue, and after pay the biggest share of that revenue is one member’s. The partnership “is controlled by and is a subsidiary of JSG, an entity that was formed under the laws of the State of Delaware”, and M R Berger signed its 2025 accounts on behalf of both Delaware members.

Its $3.88bn of 2025 revenue has two payers. “Amounts charged to other affiliates” came to $1.67bn, matching the trading company’s charge to about $2.5m. “Amounts charged to the Parent” came to $2.21bn, for London’s work for the rest of Jane Street, and no filing shows how it is set.

Of that revenue, 43% went on pay and 40% to a single member, which the accounts call “a corporate Member”, entitled to $1.56bn of the $1.86bn of members’ profit. The partnership averaged eight members in 2025, “a mix of corporate entities and natural persons”. Its register has only ever listed two corporate ones, the two Delaware designated members.

On 16 September 2026 McMurray put the top member at $1.54bn and wrote that, if that went “to one of the entities”, “the remaining partners received an average of $55.6m”. Note 5 of the partnership’s 2025 accounts files the largest entitlement at $1,563,388,000, to “a corporate Member”, which leaves $292m for the other seven members. One of those seven is the second Delaware company, so the six people on the register averaged at most $48.7m each.

At the end of 2025 the trading company still owed its affiliates nearly $1.5bn, most of the year’s charge.

The people on the two sides of the bill overlap too. John Steven Mackenzie, a director of the trading company and of Jane Street Financial, a London investment firm, has been a member of the partnership since 1 August 2022, and Maanit Desai, the trading company’s other director, since 1 January 2025. The payer’s directors sit in the payee, and the biggest partner is Jane Street itself.

India carries the London group’s fall

The London group’s fall sits in two Indian subsidiaries of the trading company. JSI Investments Private Limited and JSI2 Investments Private Limited are “registered in India and regulated by the Securities and Exchange Board of India”, and the trading company owns them. Read as the London group’s column less the company’s, an approximation, the subsidiaries’ trading gains went from about $32m in 2023 to $592m in 2024, roughly two fifths of the London group’s trading gains that year.

In 2025 they fell to about $149m, including Jane Street Korea, acquired on 1 October and counted from the year’s opening balances, while the trading company’s own revenue and operating profit rose.

On 3 July 2025 India’s market regulator, known as SEBI, issued an interim order signed by Ananth Narayan G., one of its whole time members. It said “the intraday reversal of positions in the cash market was conducted by JS Group in its India entities viz. JSI Investments Private Limited”. SEBI says the matter is still at the investigation stage.

SEBI ordered the four Jane Street entities in its case, two of them outside the London group, to deposit a sum jointly and severally. The two London-owned Indian entities “have credited a sum of 48.44 billion INR (equivalent to approximately $539 million) to an escrow account with a lien marked in favour of SEBI”. The London group carries it as “Other receivables” due after more than a year, and “no provision has been recorded”. The sum is about 18% of the London group’s equity.

The trading company’s directors, Desai and Mackenzie, answer in its accounts that Jane Street “is cooperating with SEBI’s investigation and plans to challenge the findings of the Interim Order through the appropriate process”. On 21 October 2026 India’s Securities Appellate Tribunal reserved its order on Jane Street’s plea for SEBI’s documents.

India’s gains and deposit sit in the London group’s column, so none of it reaches the trading company’s operating profit.

The 2026 accounts can break it

In 2026 Bloomberg reported a record first quarter for Jane Street, with $16.1bn of trading revenue, and Reuters a $15bn hit in July, in a report CNBC carried on 14 August.

If the steady return holds, none of that reaches the trading company’s return: a strong year or a weak one moves the partnership’s charge, and the return lands near 17.5% again. The band I set is 13% to 22% of capital employed, far wider than the three years’ own range.

The 2022 transition year, at 23.7%, sits above that band, in a year the company paid a $350m dividend and still borrowed from its parent at interest. Added back, that dividend alone moves 2022 to 19.75%.

The ratio rests on capital at the year end, which Jane Street can move. The steady years also begin with the loan between the company and its parent turning interest free on 1 January 2023, and by the end of 2025 the company had lent its parent $1.42bn on those terms, about half its capital employed.

I would put 80% on the 2026 return landing inside that band, against three years in three since the revision. I am lower than three in three because the arrangements have been rewritten once in four years.

Only 2025 tested much: in 2024 the gains moved 8%, and 2026 tests the reading only if its gains move a long way from 2025’s. The evidence is an audited filing for the return, and no filing at all for the formula behind it.

The lease rules that start on 1 January 2026, “increasing both total assets and total liabilities”, can lower the ratio a little without the argument being wrong.

Flip the assumption and suppose the company is paid for its own trades. Had the charge stayed at 2024’s 24.4% share of its gains, its 2025 return would have been about 57% on unchanged capital, and still over 40% with the extra profit kept as capital. Leaving the band would be ordinary.

The reading reaches past Jane Street, though not far: McMurray’s March survey of “20 top electronic trading firms and what they pay in London“ reported that two of the twenty run partnerships, Jane Street and XTX. The rule travels as a test, never a presumption: Jane Street Financial is billed through the same partnership, yet its return on capital went from 19% in 2022 to 16% in 2023, then climbed to 23% and 31%. Where a trading group’s London company earns a steady return on its capital, read its trading year in the bill it is charged.

The argument is wrong if Jane Street Europe Limited’s 2026 accounts, due at Companies House by 30 September 2027, show its operating profit outside 13% to 22% of its capital employed, total assets less current liabilities in the company column.

Three trip wires come with it:

• An explicit statement that the transfer pricing arrangements, the trading services agreement or the UK Services Agreement were revised, amended or replaced voids the test rather than failing it.

• So does a change in capital that moves the ratio by itself: a dividend or other return of capital above about $540m, a fifth of 2025’s capital employed, would lift 2025’s return to 22%, and an injection of about $1bn would pull it to 13%.

• A charge further from the partnership’s 2026 revenue “charged to other affiliates” than 6% or $25m, whichever is larger, would mean the two sides no longer describe one bill.

If the 2026 accounts show the trading company’s return inside that band again, London’s trading year will have landed in the partnership’s bill for a fourth steady year, and the company will have been paid for its capital.

Further reading

• How HRT and Jane Street Made $22.5B in One Quarter: The Models, the 50.1% Edge, and the SEBI Order: the firm-wide quarter and the SEBI order that this piece reads from London’s side

• How Hudson River Trading Beat Goldman’s Desks, at What Risk: the rival Jane Street is ranked against here, read from its own quarter

• How Market Makers Like Citadel Securities Actually Make Money on 0DTE Options: how a market maker earns on its own positions, the gains this piece follows into a bill

Next: Jane Street’s July loss, and which of its books took it, read from its own filings.

→ The five line screen for where a London trading group’s year shows up in its accounts on Patreon, open to everyone.

Navnoor Bawa · YouTube · LinkedIn · Corrections: reply or comment.

Logo: Jane Street

On 15 July 2026, Michael Stanton signed the papers for a Morgan Stanley feeder into Turion, Steve Cohen’s AI fund. He signed for its general partner, an affiliate of iCapital, which became Morgan Stanley’s feeder fund service provider in 2019; HedgePremier is the name on those feeders for wealth clients. No client had bought on filing day.

Morgan Stanley can now offer its richest clients Point72’s AI fund, which fell 11.4% in July while Point72’s flagship fell 3.3%, Hedgeweek reported, citing Bloomberg. The Point72 those clients know is the flagship, the diversified platform the firm’s own Form ADV brochure calls its Multi-Strat Funds. The same brochure files Turion apart, among funds that “pursue a single investment strategy”.

So which fund does a buyer get? The same-ledger test reads the books of one UBS fund that owns both, turning its quarterly marks on each into returns. Its marks for the quarter to December cover October. In the week to 15 October, Goldman Sachs’s prime brokerage data showed hedge funds cutting technology at the fastest proportional pace since its records began in 2016, Hedgeweek reported. A feeder client’s money rides on which fund those marks describe.

I read Point72’s registration, every Form D filed for Turion’s feeders and the quarterly filings of two UBS funds that own Turion.

Six chapters make the case:

• Cohen built Turion to be different: one theme, few names, and a gain of about 30% by mid-November 2025.

• Outside money, sold by outsiders: two outside distributors filed this summer to offer Turion.

• Morgan Stanley’s shelf added the theme: a buyer pays two layers of fees, into a fund with a 25% gate.

• One UBS ledger marks both funds: two kinds of UBS filing turn into a return series for each fund.

• Turion’s slope against the flagship’s: settles how far each fund moves with the Nasdaq-100 and with chip stocks.

• What a Turion buyer holds: settles who sits beside him, what he pays for, and the dated test.

Cohen built Turion to be different

The brochure gives Turion a narrower job than the flagship’s. Point72 launched it in October 2024, when its onshore feeder made its first sale. The main funds pursue “long/short equity, systematic and global macro investment strategies”. Turion’s are “focused on investment in technology and technology-enabled businesses believed to be impacted by artificial intelligence”, their holdings “expected to be concentrated in a limited number of financial instruments”. So the flagship spreads risk across many books and styles, and Turion puts it into one theme and a short list of names.

Turion’s chain of control runs to Cohen: Point72 Turion GP, LLC, its general partner, runs the funds, and Turion’s onshore and offshore partnerships’ Form Ds name Steven A. Cohen as “Sole beneficial owner of the Managing Member of the Issuer’s General Partner”. Eric Sanchez, Turion’s portfolio manager according to Hedgeweek, runs the book. The brochure’s Turion section gives the fund’s premise: “Point72 believes outsized returns come from variant investment opinions resulting from deep fundamental research in sectors experiencing rapid change.”

Cohen said what he thought of the sector four months after Turion’s first sale. “We’re in the early stages of something transformational,” he told the iConnections conference in Miami on 28 January 2025, Hedgeweek reported. “Great companies are going to be expensive,” he said. A fund built around the second sentence needs room to stay long through a rally. Hedgeweek says Turion has that room: “a wider mandate than the hedge fund manager’s main fund, giving it the flexibility to hold longer-biased positions”. Hedgeweek reported on 12 November 2025 that Turion had made about 30% on the year, including about 9% in October.

The printed view of Turion is that it is a long short book. Reuters, on 16 January 2025, called it “the hedge fund that picks winners and losers in the AI supply chain”. the annual report of A&Q Long/Short, the UBS fund that owns both, says that at the end of 2025 “the Fund was invested in Investment Funds that employed Equity Hedged strategy”, Turion and the flagship among them. Both can be true while Turion moves with the market, since a book can run shorts and still be mostly long.

July put Turion in its league. The Nasdaq-100 fell 6.6% that month and the chip index, the PHLX Semiconductor index, 20.6%; Turion fell 1.7 times as far as the first and about half as far as the second.

Its loss sat beside Altimeter’s 11% and the 21.7% lost by Whale Rock’s flagship on Hedgeweek’s list of the month’s biggest losers. The platforms barely moved: Millennium lost 2.1% and Balyasny 1.5%. Point72’s flagship fell about half as far as the Nasdaq-100. In July the fund built on one theme lost with that list.

Outside money, sold by outsiders

Turion is small inside Point72, and most of the money in it comes from outside. Alex Morrell of Business Insider reported that “Point72 hit $50 billion in assets in April” and that Turion had “nearly $4 billion”, so about 8% of the firm’s capital sits in the AI fund.1

Point72’s registration shows who holds it. At the end of 2025, Turion’s master had 121 beneficial owners, and investors outside the US held 63% of it, against 49% of the flagship’s main fund, Point72 Associates, LLC.

The feeders split the inside money unevenly. The offshore partnership, Point72 Turion Offshore, L.P., had $2.17bn of gross assets and 48 owners, and 79% of it was owned from outside the US. Point72 and its related persons, among them its affiliates, Cohen and its employees, owned none of it. The onshore feeder, Point72 Turion Onshore, L.P., was the US door: 11% of it was owned from abroad, and 42% of its $610m belonged to Point72 and its people. So the firm’s own money sits on the US side.

The feeders’ filings show what came in after Cohen made his case. Each Form D reports cumulative sales since the first sale, gross of redemptions, and the amendments filed in September 2025 and September 2026 sit a year apart.

Over that year the onshore feeder sold another $160m to 36 more investors. The offshore partnership sold another $144m to 11 more. Together Point72’s three Turion feeders, the third an offshore company, report $1.29bn of Regulation D sales to 177 investors since 2024, about $7m a head.

Two new doors opened this summer, and neither is Point72’s. ShoreBridge Point72 Turion filed on 25 June 2026, to be sold by ShoreBridge Capital Securities, of Summit, New Jersey, in 13 states. HedgePremier/Point72 Turion followed on 15 July, and neither had made a sale when it filed.

So the AI fund is small beside the firm, owned more from abroad than the flagship, and now offered through doors Point72 does not run.

Morgan Stanley’s shelf added the theme

Counted on EDGAR, HedgePremier has carried 103 feeders since 2009, 15 of them since 2020; it first carried a Point72-named feeder in 2019, and two of this year’s three are Turion’s.

In a Sacra interview on 11 August 2025, an iCapital managing director explained why the shelf exists: “the funds themselves aren’t really set up to take a thousand Morgan Stanley investors or a thousand UBS investors.” By Point72’s Form ADV, Turion’s master asks a $5m minimum and its own feeders $150,000, so the limit a feeder solves is the count: it lets many brokerage clients come in as one investor.

In 2024 the shelf added new Millennium and Jain Global feeders, and in July 2026 Cohen’s theme; the filings do not say whose choice Turion was. By the Turion filing and its Ltd. twin, iCapital’s affiliate sponsors the feeder, Morgan Stanley Smith Barney sells it in all states, and Point72 manages the fund underneath. The plumbing decides who gets in, and it leaves what they own unchanged.

Who gets in is narrow: under Rule 506(b) and section 3(c)(7) of the Investment Company Act, only qualified purchasers, a test of investments held, may buy.

The fees come in layers, as FINRA, the brokers’ regulator, warned of feeder funds in a 20 January 2026 note: “Master-feeder structures typically have several levels of fees.” The Form D names the top one: “Certain private placement clients of Morgan Stanley, Smith Barney, LLC (”MSSB”) are required to pay (i) an upfront placement fee to MSSB and (ii) an annual fee to the fund (which is then paid to MSSB).” Neither HedgePremier filing prints a rate for either fee.

Underneath, Point72’s brochure says it typically charges each fund a monthly advisory fee “at rates ranging from 1.50% (per annum) to 2.85% (per annum) of the net asset value”. Performance compensation, a share of net profits at a rate the brochure does not print, comes on top. If Turion’s rate sits in that range, which a fund’s rate may exceed or undercut, a $10m stake pays $150,000 to $285,000 a year, before the performance fee and Morgan Stanley’s two.

The HedgePremier feeder’s own exit terms are not on file; UBS’s, as a holder of the onshore feeder, are. A&Q Technology Fund, the other UBS holder, owns it with quarterly redemptions on 65 days’ notice. Its report footnotes the holding: “The Investment Fund is subject to an investor level gate of 25%.”

At the end of 2025 that stake was also under a lock-up, with no exit before 31 March 2026 except for a fee. A&Q Long/Short holds the flagship’s US feeder, Point72 Capital, L.P. (the flagship feeder), on 45 days’ notice behind the same gate.

On those terms, a holder who read Turion’s July loss when Hedgeweek reported it, on 6 August, had already missed the 27 July notice date for a 30 September exit, so the earliest way out was 31 December, and then for a quarter of the stake at most. A flagship holder could still give notice for 30 September until 16 August. From January 2027 the flagship lets holders out 8.33% a quarter instead of 25%, Hedgeweek reported, citing Bloomberg.

In a holder’s units, July cost a $10m stake about $1.1m in Turion and about $0.3m in the flagship, before any feeder’s fees. A buyer of Cohen’s theme pays two layers of fees into a fund that lets holders out quarterly. What he holds comes down to each fund’s beta to the Nasdaq-100 on UBS’s books, how far its quarterly return moved per 1% move in the index.

One UBS ledger marks both funds

The same-ledger test needs two kinds of filing and one division. A&Q Long/Short Strategies Fund, a registered UBS fund of hedge funds, files an N-PORT report each quarter with the value of every holding, and its annual report prints each holding’s cost. Value per dollar of cost is the first divided by the second, and its change from one quarter to the next is the holding’s return.

On that ledger the flagship feeder, held since 2018, compounds to +17.6% for 2025, against about 17.5% in the press. For the first half of 2026 it compounds to +14.6%, against about 14.5%.

Turion sits on the same ledger, bought on 1 February 2025 at an unchanged cost of $3.33m, so its six quarters run from February 2025 to June 2026. The one pitfall is a redemption. On A&Q Technology’s books, where a 30 June redemption cut cost by 5.41%, raw value gives +24.1%; per dollar of cost both UBS funds give +31.2%, as A&Q Long/Short’s June filing shows.

Which Point72 does a Morgan Stanley feeder client own on UBS’s books, the hedged platform or a one-theme tech book, and what will the December filing decide?

Below the paid line:

• The exhibit: both funds on one UBS ledger, quarter by quarter, against the Nasdaq index and the chip index.

• The Form ADV ownership line, each fund’s return beyond the market, and the betas without the decisive quarter.

• Our odds on UBS’s December filing, and the dated line that would prove us wrong.

Read more

On 23 June Coatue’s founder, Philippe Laffont, told CNBC he owned chip equipment makers to avoid picking the winning chip. His 30 June filing put the quarter’s biggest open market buy elsewhere: 2.98 million more Micron shares, added as Micron tripled, while the equipment lines were trimmed.

Most of that filing’s chip weight came from price rather than buying. Coatue Innovative Strategies Fund (CTEK), the firm’s registered fund for qualified individual investors, hedged part of its Micron near the top. CTEK prints its shorts, and Coatue’s next 13F, due 16 November, tests whether the firm kept its Micron.

At Friday’s close the filing’s shares are worth billions less than Coatue reported, for every book that holds them.

Coatue’s June book was nearly half chips

Eleven lines did it. I call them the chip complex: TSMC, Lam Research, Micron, Applied Materials, Broadcom, Intel, Cerebras, Nvidia, ASML, Qualcomm and AMD. They were 46.4% of the June book, from 37.1% in March. A year earlier they were 17.9%. A 13F shows only the reported long book of US-listed stock, filed up to 45 days late.

Micron shows what the weight is made of: Coatue held 165,931 shares on 31 March and 3,142,269 on 30 June. Micron’s close went from $338 to $1,154 over the quarter, and a dollar column cannot tell the two effects apart.

I rebuilt the filing’s 211 rows at Nasdaq’s closes: each filed value divided by its shares gives the 30 June close to the cent on all 64 holdings with a usable close.

The book rose $19.6bn in the quarter, 67% in reported value, buying included. Tiger Global had 39.3% of its June book in the same eleven lines, the nearest of three fellow Tiger cubs. Lone Pine had 12.6% and Viking 5.1%.

A 10% move in the chip complex would move the book by $2.26bn at June’s values. This week Lam fell 17% and Applied 19%. Micron fell 15%, and any book long those names, or copied from this filing, took the same fall.

Price made the weight; Micron was the buy

Laffont runs that book: the brochure, dated 29 July 2026, gives him in Item 4 “overall responsibility for… the implementation of all portfolio transactions.” On CNBC on 23 June he gave the logic:

“All of them at the end of the day will need the same machines.”

“If I’m a supplier to the fabs, I don’t need to make an exact bet on which of the chips is going to win… That’s sort of the reasons why we own the semi caps.”

The fabs’ suppliers get paid whichever chip wins, and the 13F for 30 June, filed 14 August, agrees on the dollar column: Lam, Applied and ASML came to $7.73bn, twice the $3.63bn Micron line.

The coverage read the add as conviction. The Motley Fool headlined it on 2 September: “Billionaire Philippe Laffont Just Bought Micron Stock in Q2…” Daniel Scrivner, who blogs on managers’ holdings, called it “a strong conviction bet on memory chips as a critical AI infrastructure bottleneck.”

That holds as far as it goes, and a share count shows a count, not a motive. It leaves open how much of the weight Coatue bought, and whether the Micron line was still live when the filing was read.

The share column answers the first: March shares times the price change is the price effect, and the share change times the June close is the trading. It is the share column method we used on Tiger Global.

The chip complex’s $11.80bn rise has three parts. Price on March’s shares made $6.59bn, and Cerebras shares Coatue held before its May listing made $1.47bn. Open market buying, net of trims, made $3.74bn at June closes; bought evenly across the quarter, the same shares would have cost about $2.52bn. Outside the listing the rise is $10.33bn. Price is 64% of it on March’s shares, or $9.51bn and 92% when June’s shares weight the price change; the lower figure carries the claim on its own.

Source: Coatue 13F-HR infotables, 31 March and 30 June 2026 · as of 30 June 2026

Lam’s line rose from $2.15bn to $4.09bn on fewer shares, so its weight is all price.

The paid half settles whether to own Micron now or wait for the printed Coatue cost, and which two filings must come first.

Below the paid line:

• The June book marked to 16 October, its chip lines one by one.

• CTEK’s Micron cost, its short by 30 June, and its monthly cards since.

• The position, with its entry tied to the two November filings.

Read more

Hudson River Trading made a record $11.4bn trading in the second quarter of 2026, Bloomberg reported on 19 August. In December Bloomberg said HRT was minting billions by “embracing more risk”; to see what that risk looks like, I netted the stock its US dealer owned against the stock and futures it sold, at all eight year-ends filed electronically.

Hudson River Trading out-traded each Goldman Sachs desk in the second quarter, and its US dealer’s audits show how it carries the risk: a bigger book, hedged within the same band. From 2019 to 2025 the dealer’s book and its unhedged residual both grew about tenfold, and a loss would land first on its partners’ capital.

The piece runs in four steps: the quarter against Goldman’s desks, why volume does not pay for it, the risk Bloomberg described, and what eight audits show about it.

HRT out-traded each Goldman desk

Goldman Sachs had posted its quarter five weeks before Bloomberg’s report. Its 8-K of 14 July 2026 showed $7.4bn of equities revenue and $4.6bn from FICC, its fixed income, currencies and commodities desk. HRT’s $11.4bn is larger than either. It is also about 1.5 times the $7.5bn the two desks made from intermediation, which is buying and selling for clients, the part of a bank’s trading closest to what a market maker does.

Add the $4.5bn the desks earned from financing, lending to clients against their securities, and they made $12.0bn. HRT reached neither that nor JPMorgan’s $12.1bn of markets revenue. It did clear Morgan Stanley’s $8.8bn and the record $7.3bn at Citadel Securities, a non-bank market maker.

The figures are not the same kind of revenue. Goldman’s is client business, reported in a filing; HRT’s is a private firm’s own trading revenue, reported by Bloomberg and then by the FT from a person familiar with the results, and HRT declined to comment. Virtu Financial, a market maker that files its numbers, reported $0.86bn of trading income for the same three months.

Volume alone doesn’t pay for it

A sceptic has two explanations for the record that need no audit. The first is a volatility windfall, which predicts a jumpier market. The FT’s report, in Hedgeweek’s summary, credited “Market volatility linked to the war in Iran and dramatic swings in technology and artificial intelligence stocks”.

The market’s volatility index was calmer: the VIX averaged 18.31 in the quarter against 23.56 a year earlier, on our average of daily closes.1 An index average cannot see single stocks swinging against each other, so the windfall survives only in the FT’s narrower form, swings inside technology and AI stocks.

The second is a tide that lifted every intermediary. Jamie Dimon, JPMorgan’s chief executive, described it in his bank’s 8-K: “Markets revenue growing 35% due to elevated client activity, strong trading performance and continued demand for financing in Equities”. A tide predicts that every firm grows alike, and at about the pace of volume. Nasdaq’s 10-Q puts US share volume up 10% on a year earlier, and options volume up 27%.

Exhibit 1 · HRT grew faster than the tide and faster than Citadel Securities.

Source: Bloomberg Law and Hedgeweek for HRT; Bloomberg and Hedgeweek for Citadel Securities; the Goldman Sachs, Morgan Stanley and JPMorgan Q2 8-Ks; Virtu’s and Nasdaq’s Q2 10-Qs · as of 30 Jun 2026

Volume grew by about a tenth to a quarter, the banks by a third to just over a half, and the two non-bank market makers about three and four times. Virtu grew at the tide’s pace: its market-making measure rose 28%, against HRT’s fourfold.

Raman Kalra of Coalition Greenwich, a market research firm, put the non-bank liquidity providers’ revenue pools up 45% from 2024 to 2025, in a 3 June 2026 release. “Whether these outsized returns are the new normal or just a byproduct of a unique macroeconomic moment remains to be seen,” Kalra said. One quarter cannot settle that. It does show HRT taking more than the tide, and Bloomberg had printed an explanation the previous December.

The story says more risk

On 2 December 2025 Bloomberg ran “More Risk, Less Speed: Hudson River Mints Billions Slowing Down”. Its second paragraph begins: “The high-frequency firm is increasingly holding positions for days and even weeks, embracing more risk”. Business Insider had described the engine nine months earlier, from people familiar with the firm. HRT, it said, was “earning billions in profits from hedge-fund-style trading strategies that require greater risk and longer holding periods”.

The unit running them is Prism, which “generated more than $2 billion” in 2024, according to the same report, with “ETF arbitrage and index rebalance” its noteworthy strategies. Roughly half of profits still came from the high-frequency business HRT calls Classic.

Iain Dunning, who heads AI research at HRT, gave both halves on Bloomberg’s Odd Lots podcast on 31 October 2025. “We make money only because someone else wants to trade,” he said. “If no one was trading, we wouldn’t exist.” On the longer horizons he went further: “I think the market making service provision does break down.”

The filings back Bloomberg on holding. HRT Financial LP, the group’s US dealer, files a Form 3 each time it becomes a 10% owner of a listed company, and it filed one in 2023, two in 2024, four in 2025 and 66 in 2026 through 16 September, nearly all for micro-caps. Adam Nunes, the dealer’s president and chief executive since May 2010, signs them. One Form 4 shows Aditxt shares bought on 18 June 2026 and sold to zero on 22 June: held four days.

The dealer’s quarterly 13F, which Nunes has signed every quarter since at least 2017, grew too, to $54.8bn at 30 June 2026, 2.4 times its level a year earlier.2 The 13F shows HRT holds more; how long it holds is Bloomberg’s reporting, not the filing’s.

The second half, “more risk”, is a different claim. If risk means a net position that moves with the market, a bigger book sold short against itself carries little more of it, and a bigger book left open carries a lot. Which one HRT built is written down once a year, in the dealer’s audited balance sheet. I found no published netting of it; Bloomberg and Business Insider describe the holding, not the net.

So the two answers predict different balance sheets. If HRT’s extra risk is a growing directional bet, what is left after its hedges should grow faster than its book, and its share of the book should climb out of its old range. If the extra risk is a bigger book hedged the same way, the dollars should grow and the share should not.

The audits show the same band

Every US broker-dealer files an audited Statement of Financial Condition each year, on Form X-17A-5. The dealer’s statement for 31 December 2025 opens its notes: “HRTF provides liquidity to financial markets and trades financial instruments utilizing a proprietary electronic trading system.” It adds: “The Partnership is engaged in a single line of business of proprietary trading as a securities dealer.”

The test is one subtraction, repeated. From each year-end’s equities owned, take away the equities sold, not yet purchased, then the futures sold, and add back the futures bought, all as the statement prints them, futures at notional. Notional is the full value of the stock or index a contract stands for. Call what is left the residual: the dealer’s net stock position after its hedges.

The trap is to use the 13F instead, which lists no shorts and no futures, so its residual is always 100% of the book.

Exhibit 2 · The dealer’s 2025 residual was $2.63bn, 11.3% of the stock it owned.

• Equities owned: $23.23bn, 100%

• Less equities sold, not yet purchased: $9.86bn, 42%

• Less futures sold, notional: $11.07bn, 48%

• Plus futures bought, notional: $0.33bn, 1%

• Residual: $2.63bn, 11.3%

Source: HRT Financial LP, audited Statement of Financial Condition, fair-value table and derivatives note; assumes the futures sold offset stock, since the note does not name their underlyings · as of 31 Dec 2025

At end-2025, and at three of the four year-ends before it, short sales offset about two fifths of the stock the dealer owned, and futures sold offset most of the rest. Exhibit 3 repeats it at every year-end, and sets the residual against partners’ capital, the owners’ equity a loss reaches first.

Exhibit 3 · From 2019 the residual stayed between 3.0% and 17.3% of the stock owned, and between 14% and 88% of partners’ capital.

Source: Author’s calculation from HRT Financial LP audited Statements of Financial Condition, fair-value tables and derivatives notes; each year’s statement is in the dealer’s EDGAR filing list. Partners’ capital is from the same statements; in 2018 and 2019, before the dealer became a partnership, it is members’ capital · as of 31 Dec 2025

Two things hold across the table. The residual share topped 20% once, in 2018, on a book of less than $1bn; from 2019 the dealer sold short, or sold futures against, between 83% and 97% of what it owned. And the share moves without a trend. It fell to 6.4% in 2021, when the book first passed $18bn, and to 3.0% in 2024.

If the extra risk Bloomberg described were a growing bet, the 2025 statement would show the residual above its 2019 to 2024 range. On the stock owned, 2025’s 11.3% sits inside that range; on partners’ capital, its 62% sits inside it too, below 2023’s 88%. A band that wide cannot tell a slowly growing bet from a steady one.

The residual set a record

A hedged book still carries risk: the spread between what it owns and what it sold, and whatever the futures fail to match. That is the form of “more risk” the audits show: more of that spread risk, carried on a bigger book.

The number that cuts against our claim is the latest one. The residual rose from $0.48bn at the end of 2024 to $2.63bn a year later, 5.5 times, the largest in dollars of the eight year-ends, though only 15% above 2023’s $2.29bn. Its share of the stock owned went from 3.0% to 11.3%. In the year Bloomberg described, the bet grew in dollars, and its share went back to where it had been.

Against the partners’ capital, the record is large. The 2025 statement shows $4.22bn of it, so the residual equals 62%, second among the eight year-ends only to 2023’s 88%. Its subsequent-events note adds: “During this time, the Partners have withdrawn capital of $500.0 million.” That leaves $3.72bn before any 2026 profit, and puts the residual at 71%. A 10% fall in whatever the residual is long would cost about $0.26bn, 7% of that smaller base.

Who would bear that loss is on file. FINRA’s BrokerCheck lists the dealer, which Nunes runs, as a Delaware partnership owned through HRT companies. Jason Carroll, a co-founder Business Insider calls “a current managing partner”, holds between 50% and 75% of HRT Capital LLC, which indirectly owns 75% or more of the dealer.

No HRT document says why the dealer hedges what it carries. I infer it from the structure, with moderate confidence: when a loss lands first on the partners’ capital, a dealer has reason to keep its residual a small share of its book. Eight audits show the result that inference predicts. None of them states the motive.

What the audits cannot see

1. The futures’ underlyings and direction are not named. Netting only the stock leaves 57% to 58% of it in 2021, 2023, 2024 and 2025, so the futures carry the claim, and this is where the case against it is strongest. Two facts fit index hedging without proving it. At end-2025 the net futures sold, $10.74bn, hedged 80% of the stock-only net of $13.37bn; what they left is the residual. And the two largest lines on the June 2026 13F are the SPDR S&P 500 and Invesco QQQ index funds. The dealer also trades Treasuries, so some futures could be on rates. From 2021 the notes do not label futures bought or sold either: this piece reads the liability-side contracts as sold, as the 2018 to 2020 notes labelled them, and 2022’s fit that reading less cleanly.

Citadel Securities LLC’s audited statement for 31 December 2025 shows $34.16bn of equities and ETFs owned, 61% of it left after short sales, beside HRT’s 58%. Its derivative notionals appear only gross, $2.28tn of equity contracts with no bought or sold split, so the full netting cannot be run on it.

1. Option deltas are not disclosed. The 2025 note shows $31.6bn of options bought and $31.8bn sold, at notional,. Matched notional says the book sits on both sides. It gives no direction, and no delta is printed.

2. A year-end is a snapshot. The gap since the last audit is wide. The 13F’s shares and fund units rose from $24.26bn at 31 December 2025 to $39.92bn at 30 June 2026, 65% more, hedged by an amount no public filing shows. The next audit is the first look at that book.

3. The stock it owns has changed. The long book now reaches into micro-caps, where nearly all of 2026’s Form 3s were filed, while the hedges the evidence fits are index futures and index funds. A micro-cap hedged with an index carries size, liquidity and basis risk an index holding does not, so the residual can hold near 11% of the stock owned while the risk in each dollar rises. A constant ratio is not constant risk. For now each line is small: the largest of the stocks it crossed in the second quarter sat at $255,000 in the dealer’s 30 June 13F.

4. One entity is not the group. On 31 August 2022 Moody’s cut HRT to Ba2 from Ba1, outlook negative, and located the added risk in its “expansion into certain regions and markets outside of its core historical areas of focus”. The dealer’s statement cannot test that: it covers one US entity, and risk taken through prime brokers elsewhere need not touch it.

So the finding stops at the dealer, at its year-ends.

The next audit decides it

The argument is wrong if HRT Financial LP’s audited statement for 31 December 2026, due on the SEC’s EDGAR system by 1 March 2027, shows a residual above 20% of equities owned. That would make the dealer’s bet larger against its book than at any year-end since 2019.

I would bet, at about 75%, that the share stays under its line, because the dealer has hedged more than four fifths of its stock at seven straight year-ends while the book grew about tenfold.

The base rate alone would say 80% to 88%. The line was crossed at one of eight year-ends, on a book under $1bn, and the largest one-year rise in the eight statements, 8.3 points from 2024 to 2025, would carry the share from 2025’s 11.3% to 19.6%, 0.4 points short. My odds are lower because the futures’ underlyings are unnamed and the dealer is not the whole firm.

A market maker’s gross book shows its volume; only its net shows its bet.

HRT’s 2026 audit lands by 1 March: a net under a fifth of the stock it owns means the bigger book stayed hedged; above it, the extra risk Bloomberg found has become a bet.

Further reading

• How HRT and Jane Street Made $22.5B in One Quarter: The Models, the 50.1% Edge, and the SEBI Order: the same two firms’ trading engines, a year earlier

• How to Read a Market Maker’s 13F Options Line: Jane Street and Susquehanna’s SpaceX Book, Ranked Against Their Own: reading a dealer’s own filings without mistaking inventory for a bet

Next Monday: a named fund’s bet, read from its own filing, with the position.

→ A screen of every Form 3 HRT’s dealer has filed as a new 10% owner, each timed to its last reported trade on Patreon, open to everyone.

Navnoor Bawa · YouTube · LinkedIn · Patreon

Logo: Hudson River Trading LLC

On 13 October 2026, Paul Tudor Jones told CNBC he would be “short the whole year” in 2027. His firm’s 30 June filing leans against narrower targets: small caps, the AI chip leaders and junk bonds. Its junk bond puts grew 13 times from March, while its Nasdaq-100 puts and calls roughly cancel. In other words, Tudor’s new insurance sits on credit and the chip leaders. That matters to the firm, whose book reads narrower than its founder’s words, and to any PM holding the same exposures.

Short on air, level on the Nasdaq

Jones told CNBC’s Squawk Box that “2027 is going to be really, really challenging for the stock market” and added, “I’ll be short the whole year.” CNBC’s headline read it as a call on the stock market.

Tudor Investment Corporation, which Jones founded, filed its 13F for 30 June on 14 August. A 13F lists only the options a filer owns, as underlying shares and their value, with no strike, expiry or premium. The SEC’s instruction is plain: “You should not report put or call options that you write.”

The June filing shows puts on QQQ, the Nasdaq-100 fund, over $3.49bn of shares and calls over $3.55bn. The line is level, with puts and calls about equal. In March the puts had led.

Reading one line of the filing

The junk bond line lists puts on HYG, the iShares iBoxx $ High Yield Corporate Bond ETF, over 10,000,000 underlying shares valued at $799.7m. Divide the value by the shares and the filing marks HYG at $79.97 on 30 June.

The same arithmetic nets a whole position. A net put is the put shares, minus the call shares and the shares owned outright, priced at the filing’s quarter end mark. A negative result is a net long. It measures notional, not delta, so a put struck far below the mark is valued the same as one struck near it, though only the second pays on an ordinary fall.

Netted that way, the book’s largest position against stocks is a net put of $5.98bn on IWM, the Russell 2000 fund. That is 8.3% of the $71.9bn in Tudor’s 13F. The put has appeared in every filing since the end of 2023. From September 2024 to December 2025 it ran between about 18 and 26 million shares at each quarter end. March 2026 was a dip, and June restored it. That put is standing, so the change lies in the smaller, newer lines on chips and junk bonds.

For a PM long the S&P 500 and little else, Tudor’s filing changes nothing. If he owns AI chip names or high yield credit, Tudor held protection on exactly those at 30 June, three and a half months before Jones went on air. An allocator sees the firm’s new insurance on credit and chips, and an analyst who wants to work there sees the craft of choosing what to insure.

Several managers share one filing

Tudor’s Form ADV reports $106.8bn of regulatory assets as of 31 December 2025, including leverage. That places it between Bridgewater Associates at $150.2bn and Caxton Associates at $71.6bn on each firm’s latest Form ADV. That is a different scale from the 13F’s $71.9bn, because regulatory assets are gross holdings while the 13F values listed lines at the shares beneath them.

Not every line is Jones’s. Tudor’s Form ADV brochure says: “Mr. Jones is the sole controlling principal of Tudor and is the Co-Chairman and Chief Investment Officer of TIC,” Tudor Investment Corporation. Other managers run their own books inside the firm, all added into one 13F. So is a given line Jones’s view, another manager’s, or a hedge on something the filing cannot show?

The filing cannot say, so this piece reads every line as the firm’s choice at 30 June.

Tudor’s chips turned net put

The chip trade here means seven lines: Nvidia, Micron, TSMC, AMD and Broadcom, plus two semiconductor funds, SMH (the VanEck Semiconductor ETF) and SOXX (the iShares Semiconductor ETF). Across Tudor’s last ten filings, the seven lines were net long in seven of them, every quarter from March 2024 to September 2025.

The sign changed in December 2025, when the seven lines tipped barely to a net put, and by March 2026 the tilt was clear. Call it the chip turn. From March to June the put lines doubled in value while the count of put shares rose 16%. All but AMD and SOXX now carry more put shares than call shares and owned shares combined.

Jones gave his reason and left the trade unsaid. CNBC summarised his worry as a wave of AI equity supply, with offerings from “Anthropic and OpenAI, along with one just finished for SpaceX”. In his own words, “You’ve probably got the most negative flow picture for the U.S. stock market that we’ve had, say, since 2002.”

New AI shares need buyers, and the listed names most exposed to AI spending are the chipmakers. So puts on the chip complex express that view without selling the whole index. The June filing is consistent with that view, but it cannot say it is the view.

The number this piece computes is the size of the chip turn, Tudor’s net put notional across the seven chip lines at 30 June. Tudor’s 13F for 30 September, due by 16 November 2026, will confirm or reverse it.

The paid half settles whether to buy the same junk bond protection Tudor holds, at what size, and when to drop it.

Below the paid line:

• Exhibit 1 sets Jones’s words against Tudor’s filing, row by row

• Exhibit 2 nets the seven chip lines name by name, with the sum shown

• Exhibit 3 tracks five quarter ends of the turn, and how much of it was price

• The position is one listed junk bond option, with entry, payoff level and exit rule

Read more

Bridgewater Associates sold more than half its AI shares in the second quarter, its 14 August filing shows. Its chip, cloud and power stocks fell from 34.0% of its reported stock book in March to 20.8% in June, the deepest cut among the large funds I checked. It looks like a vote against AI, cast in the quarter its own research said compute would need rationing. Its own filings say otherwise: they show its AI holdings moving inside a band, and a cut like this one before, followed by buying. If you sold chips on this filing, you sold beside a fund that has bought back before.

The reading that followed the filing was simpler: Bridgewater dumped its AI stocks, so the world’s biggest macro fund has turned on the AI trade. Coverage of the June filing read it as a cut to big tech and chips. The cut is real. The turn is missing from the record.

A fifth, stretched to a third

Bridgewater’s 13F lists the US listed stocks and listed options it held long at a quarter end, filed up to 45 days later. It shows no shorts, futures or currencies, and nothing of when in the quarter the firm traded.

That list is one sleeve of a larger firm. Bridgewater’s Form ADV, signed on 18 August, reports $150.16bn of regulatory assets, a gross measure that counts borrowed money. The $24.38bn stock book in its June 13F is about a sixth of that. The second quarter’s AI sale, $7.09bn net, came to more than a quarter of that stock book.

I sorted the AI lines in its filings into one basket of 38: chipmakers and their suppliers, the hyperscalers (the cloud and internet giants that buy the chips) and the power producers that feed data centres. Across six quarter ends, four of the basket’s weights sit between 18.7% and 24.0% of the reported book. I call that band the AI range. March 2026, at 34.0%, is the stretch, the only quarter end at which AI passed a quarter of the book.

The basket’s 38 lines, as filed: Advanced Micro Devices, Alphabet (class A and class C), Amazon, Applied Materials, Arista Networks, ARM Holdings (ADS), ASML (NY registry shares), Astera Labs, Broadcom, Celestica, Cisco, Coherent, Constellation Energy, CoreWeave, Credo Technology, Dell Technologies, GE Vernova, Intel, KLA, Lam Research, Marvell, Meta Platforms, Micron, Microsoft, Monolithic Power Systems, Nebius, Nvidia, Oracle, Qualcomm, Sandisk, Seagate, Super Micro Computer, Talen Energy, TSMC (ADS), Vertiv, Vistra and Western Digital.

Equity lines only; the six Bridgewater filings used here carry no option lines. A line’s weight is its filed value over the filing’s total.

I read Bridgewater’s 13F as a sizing machine: it holds AI inside that band, will stretch past it by choice, and sells back into it, whatever its research says about AI. The March stretch came from buying. The June sale handed it back after a rally had made it larger still.

The quarter’s arithmetic shows how much larger. The AI stocks Bridgewater held at the end of March rose about 60% over the quarter, weighted by their March values. Had it kept its March share counts, with the rest of the book as it ended the quarter, AI would have been about 39% of the book. The rally added about five points; the sale took off eighteen points, handing back the rally and the stretch the firm had chosen in March.

For a portfolio manager, a cut from the stretch back into the band is weak evidence of a turn. For an allocator, the band across six quarters says more about Bridgewater’s view of AI stocks than any one quarter’s cut.

The research expected more spending

In January, Bridgewater estimated what AI spending would add to US growth in 2026. In February it put the four largest spenders’ capital spending at $650bn for 2026, against $410bn in 2025, and expected them to spend more than forecast. In April, the page for its first quarter chief investment officer call said compute was running short. Exhibit 1 sets those words beside the June filing.

Source: Author’s calculation from Bridgewater research of 7 January, 23 February and 22 April 2026 and its Form 13F-HR filings for 31 March and 30 June 2026, as of 30 June 2026.

The research and the filing part where the quotations end and the share counts begin. Micron, TSMC and Marvell each lost more than nine tenths of their shares in the quarter its research page described compute running short, while Nvidia lost less than a fifth.

The words are about the economy; the filing is about position size. The research says how much the spenders will spend and what that does to growth. The filing says how much of one stock book a theme should fill after a rally. A firm can believe the first and still sell to satisfy the second.

March pushed AI to the top

The quarter before built what June sold, and in it the research and the filing pointed the same way. The January and February notes expected more AI spending, and the book bought the stocks that sell into it.

Bridgewater raised its AI share count by about 29%, weighted by value, in a quarter when the basket’s prices rose only 0.5%. Its Nvidia shares rose between the end of 2025 and 31 March. It also sold a net $5.06bn of other single names, so the rest of the book shrank. AI reached 34.0%, the highest weight in Exhibit 2, because the firm bought AI and sold other stocks, not because prices lifted it.

Source: Author’s calculation from Bridgewater’s Form 13F-HR filings for 31 March 2025 to 30 June 2026, as of 30 June 2026.

March 2025, at 14.8%, is the one quarter end below the band, before the build that followed.

The second quarter’s rally would have stretched it further, and chips led that rally. The price of SOXX, a fund of chip stocks, divided by that of SPY, an S&P 500 index fund, rose from 0.505 on 31 March to 0.858 on 30 June on Nasdaq closing prices. Chips gained more than two thirds on the index in three months, and a book that held still would have let that rally decide how much AI it owned.

Micron led the June sale

Counted in shares and weighted by each line’s value, Bridgewater cut its AI holdings by 58.3% in the quarter. That was a net sale of $7.09bn at 30 June prices, more than a quarter of the stock book it reported that day.

Micron, the memory maker, was the largest single sale: Bridgewater sold a net 1,359,038 shares, the gap between the two Micron counts in Exhibit 1. A 13F values each line at the quarter’s last close, and the same rule prices the sale at Micron’s 30 June close:

1,359,038 shares × $1,154.29 ≈ $1.57bn

That one line is more than a fifth of the whole AI sale. I priced every net trade the same way, at 30 June closes, and valued lines sold out entirely at their 31 March price, because a line that is gone has no June price in the filing. On that one convention the quarter adds up.

Source: Author’s calculation from Bridgewater’s Form 13F-HR filings for 31 March and 30 June 2026, as of 30 June 2026.

The first six rows sum to the last. Inside the AI row, chipmakers and their suppliers carried $5.50bn of the sale, the hyperscalers $1.18bn and the power producers $0.40bn. About a third of the money went back into other listed holdings, the largest part into the index: SPY took $713.5m, iShares Core S&P 500 $242.0m and Vanguard S&P 500 $205.9m. The rest left the part of the book a 13F can see. The book’s dollar value still rose, from $22.40bn to $24.38bn, because prices rose.

Bridgewater cut its weight the most

I ran the same basket of 38 lines through the 13Fs of ten other large funds for the same two dates, counting equity lines only. Eight of the eleven sold AI, net, in the quarter, so selling was common. What stood out was how far each fund let its AI weight move.

Source: Author’s calculation from each fund’s Form 13F-HR filings for 31 March and 30 June 2026, as of 30 June 2026.

Bridgewater’s weight fell the most, by 13.2 points. Millennium and Two Sigma held theirs roughly flat while cutting their AI shares by more than a third, weighted by value, and each sold more AI stock in dollars than Bridgewater did. Weight is the fair base because it nets out the rally: a fund that sells about as fast as prices rise ends the quarter where it started, whatever its dollar sales say. Bridgewater sold far faster, so its weight is the one that fell.

Point72 is the cleanest case of the rule. It sold nearly a third of its AI shares, weighted by value, and finished the quarter at the AI weight it started with, to the decimal, because the rally replaced in price what it took out in shares.

Three funds took the other side. Coatue Management, AQR and Lone Pine were net buyers of the same names in the quarter, Coatue by the most. The basket Bridgewater sold down to a fifth of its book made up about two thirds of Coatue’s.

Tiger Global sold a net $2.73bn of the basket and its AI weight still rose, to 70.1%, which fits our read of Tiger’s own filing: it sold more stock than it bought while prices carried its book higher.

Nvidia shows the move twice

Nvidia’s column in Exhibit 2 is the machine’s fingerprint. Bridgewater more than doubled its Nvidia shares in the second quarter of 2025, cut them by about two thirds in the third, rebuilt them over the next two quarters and trimmed them in the latest. Its Nvidia count moved the same way as its AI weight in all five quarterly moves, and on 30 June 2026 it held almost exactly as many shares as at the end of 2025.

The third quarter of 2025 is this quarter’s twin. Bridgewater cut Nvidia by 65.3%, Alphabet’s class A shares by 52.6% and Meta by 48.3%. The AI weight fell from 24.0% in June 2025 to 18.7%, the lowest of the five quarter ends from June 2025 to June 2026. In the next quarter it was buying Nvidia again, and by March AI stood at its peak. Rebuilding that fast is what a sizing machine does after a cut.

For a reader who holds the same names, the twin is the useful part, and the weak one: one precedent is thin, and it is the only one these six filings hold.

Any reader can run the test on any fund: multiply each 13F line’s share count by the quarter end price, then set the fund’s AI weight against its own last six quarters and against its peers in the same quarter.

September’s filing can break the range

The case against the sizing reading rests on three facts. The first is the split inside the sale. Bridgewater cut its chip, supplier and power names harder in share terms than the hyperscalers, by about 63% against about 44%, weighted by value, which looks like a call against chips. In dollars, though, its chip exposure fell 32% and its hyperscaler exposure 37%, close to the same cut. That is what a rebalancer does: the stocks that ran fastest need the most shares sold to bring their dollars back down.

The second is the research after the quarter. On 28 July, a Bridgewater talk page summarised the discussion as covering “emerging risks to frontier labs’ revenue growth across regulation, adoption challenges, and competition from cheaper models”. That is cooler than April’s talk of rationing, because frontier labs are among the buyers of compute. It came after the quarter closed, so it cannot explain the June sale, but it could drive a September one.

The third is the names. On the 13F’s own quarter end prices, Micron rose 242% in the second quarter and Marvell 201%, and the filing left each with under a tenth of its March shares, while Nvidia rose 15% and kept more than four fifths of its. That much fits sizing by name. TSMC rose 41% and lost 97% of its shares. Amazon rose 14% and lost 54%. Those two are choices, and the filing does not say why.

The next filing settles it. Since 30 June the chip sector has fallen against the index: SOXX/SPY stood at 0.858 on 30 June, 0.746 on 30 September and 0.700 on 12 October. Price alone can move the AI weight, so the test has to be read in trades, the way Exhibit 3 reads the second quarter.

This reading is wrong if Bridgewater’s 13F for 30 September 2026, due by 16 November 2026, shows it sold AI shares again on net, valued at 30 September prices, and AI below 18.7% of its reported stock book, the floor of the five quarter ends before it. That outcome would mean Bridgewater is leaving AI in the part of its book a 13F shows. A net purchase says the band holds, as it did after the cut in the third quarter of 2025. A net sale that leaves AI inside the band says the trimming is not finished, and the reading stays open.

Until that filing lands, the verdict is the AI range: Bridgewater did not turn on AI in June; it sized its AI stocks back into the band, as it did in the third quarter of 2025, and the last time it did that it bought back.

Further reading

• Tiger Global Sold $3.7bn of Stock, and Its 13F Shows One Chip Bet Built Up: Intel: the same share count method, on a fund that built one chip bet

• How Elliott Is Long Synopsys With Software Hedged, From Its Own 13F: one fund’s bet read line by line from its own 13F

• Bridgewater’s Pure Alpha Returned 33% in 2025. Here’s the Exact System Behind It: the macro system behind the stock book this 13F shows

Next Thursday: a named fund’s bet, read from its own filing, with the position.

→ Bridgewater’s 2025 AI cut and its refill, name by name on Patreon, open to everyone.

Navnoor Bawa · YouTube · LinkedIn · Patreon

Cover photo: Web Summit · CC BY 2.0, adapted · via Wikimedia Commons

Man AHL’s trend fund gets paid by a bond selloff only once the selloff becomes a trend. In 2026’s first half its bond and rate futures lost $1.05m while the fund still made money. With the 10-year Treasury yield at its highest since 2002 this month, anyone holding trend funds as a hedge against a bond rout needs to know when that hedge pays. I read the program’s SEC filings and Man’s own monthly reports against Treasury’s yield curve.

The coverage credits the bond selloff. On 2 October, Hedgeweek relayed an FT report that “Computer-driven funds that seek persistent market trends have built sizeable positions against fixed income”. By that account Graham’s Tactical Trend fund was up more than 31% this year, Winton’s Diversified Macro 17.5% through late September and Aspect’s flagship about 21%, according to people familiar. The same report noted that the managers’ returns “have not been driven solely by government bond positions”. Treasury’s par curve, its daily yield for a notional bond priced at par at each exact maturity, had the 10-year at 5.31% on 5 October. On 5 October I worked through where AQR’s own trend rule turns its Treasury short off. This piece asks what the bond trade has actually paid.

Man AHL’s own records describe a different first half. The fund made 11.86% for its SEC feeder’s investors, yet its bond and rate futures lost money through June, in the filing’s dollars and in Man’s own monthly figures, while currencies, stock indices and energy made the return. In July and August, with the fund already short as yields rose, bonds and rates paid. Trend following is a rule that loses small and often and is paid by whichever markets keep moving; in bonds, the paying started only once the selloff had become a trend. Call it no trend, no pay. The turn that starts a trend costs first.

Three of Man’s funds appear below, each named once here. The SEC feeder is Man-AHL Diversified I LP, $64.4m at 30 June; its 11.86% is the return on its Class A Series 1 units after costs. The trading company is the master fund it invests in, Man-AHL Diversified Trading Company L.P., about $145m, whose accounts sit inside the feeder’s 10-Q with gains and losses by sector. The $1.05m is the trading company’s loss on interest rate futures, a line that holds both government bond and short rate futures. The flagship is Man AHL Diversified, a fund of about $313m that Man runs on the same AHL Diversified Programme and reports on monthly. I profiled the firm in December; here I take its trend machine apart.

If you hold a trend allocation as protection against falling bond prices, the first half says it pays on a selloff that has become a trend, and that it can lose at the turn that starts one.

AHL’s machine, in its own words

AHL’s 2022 annual report describes the program without decoration. It is “designed in the anticipation that most of its trades will be unprofitable; the objective of overall profitability depends on the system identifying certain major trends which occur and recognizing significant profits from participating in such trends”. The same section adds that “in trendless or stagnant markets, the AHL Diversified Program is unlikely to be profitable”.

In plain terms, AHL expects to be wrong on most trades and right, by a lot, on a few. The losses are the cost of being in position when a big move starts.

The 2026 10-Q adds the second half of the design. “All of the strategies and systems of the AHL Diversified Program are designed to target defined volatility levels rather than returns.” The program chooses how much risk to run. It doesn’t choose how much to make.

The filing describes the machine in words. The tested public version comes from researchers at AQR, and it has three parts: a signal, a size and a spread across markets.

The signal is each market’s own past

Time series momentum is the rule that a market’s own recent return decides whether you are long or short it. Stock pickers’ momentum ranks one stock against another; here each market is judged only against its own history.

A 2012 study of 58 futures markets by Moskowitz, Ooi and Pedersen found that a market’s past 12-month excess return, its return above cash, predicted its next month’s. It held “for every asset contract we examine (58 in total)”. A market that had gone up tended to keep going up for a while, and a falling one kept falling.

Why trends exist at all is the question a sceptic should ask first. The study reads its result as “consistent with sentiment theories of initial under-reaction and delayed over-reaction”. In the CFTC’s positioning data it found that “speculators profit from time series momentum at the expense of hedgers”. A 2017 AQR study adds a reason that matters for bonds: “when central banks intervene to reduce currency and interest rate volatility, they may slow down the rate at which information is incorporated into prices, thus creating trends.” A central bank that moves rates in announced steps is close to a trend generator for its own bond market. That study’s version of the rule blends 1-, 3- and 12-month signals with equal weight across 67 markets, 15 of them bond markets, and rebalances monthly.

Now apply it to bonds. A bond future’s price rises when yields fall, so a falling 10-year yield turns the signal long, and a rising yield turns it short. The rule doesn’t ask what the Fed will do next. It asks only which way the price has gone over the lookback.

That has one consequence for a selloff. The signal can only go short after yields have risen long enough to show up in the lookback. With a 1-month signal in the blend the lag is weeks, not months, but it is always there.

Positions sized to a volatility target

The second part is size, and it is where “defined volatility levels” comes from. The 2012 study sets each market’s position to 40%/σ. Here σ is the market’s ex ante volatility, a forecast of its annualized volatility made before the trade. The study builds it from an exponentially weighted average of past squared daily returns with a centre of mass of 60 days, a weighting that gives each new day about a sixtieth of the weight. AHL doesn’t publish its own estimator; these are the study’s choices.

Plug in two markets. Say one has a σ of 10%, so it gets 40/10 = 4 times capital in exposure. Say another has 20%, so it gets 40/20 = 2. The quieter market gets the bigger position, so each contributes about the same risk. The 2017 study then scales the whole book to a 10% annualized volatility target.

So a calm bond market can carry a large position. When bonds start to swing, σ rises and the position shrinks, whatever the signal says. The size is set by how noisy the market is, never by a view on yields.

Bonds carry a sixth to a third of the risk

Man’s monthly reports put bonds and rates at between a sixth and a third of the fund’s risk at each 2026 month end, which is the third part of the machine: the spread across markets. Each gives a 99% value at risk (VaR) by sector, a statistical estimate of a loss the positions should exceed with 1% probability; the report doesn’t state the holding period. The feeder’s 10-Q prints a table it also calls VaR, but by its own words it uses “exchange initial margin requirements” as the measure, so I use Man’s figures.

Source: the flagship’s monthly reports for February, March, June, July and August 2026, Class DN USD. Man reports five sectors (bonds and rates, commodities, credit, currencies, stocks); the sum adds them with no credit for diversification, and the shares are my arithmetic. Net exposure is the positions’ notional value as a share of the fund’s capital, with bond positions restated as 10-year bond equivalents.

A large part of the book, then, never the whole of it. The last column is the turn in one line. At the end of February the fund was long bonds worth 61.6% of its capital in 10-year terms; a month later it was 143.8% short.

The payoff looks like a straddle

The 2012 study found that trend following’s returns look “similar to an option straddle on the market”. A straddle is a call and a put at the same strike. Its owner pays a premium up front and makes money only on a large move, in either direction.

AHL’s line that most trades “will be unprofitable” is the premium. Each trend that fails costs a small loss, kept small by the volatility target. Each one that runs pays for many failures.

The analogy breaks in one place. A straddle’s premium is known the day you buy it. A trend follower’s isn’t, and a trend that reverses in a few sessions can cost far more than the small, steady bleed the design expects.

Long into March, short by April

The flagship was long bonds when this year’s selloff began, and the worked case starts at that turn. The 10-year par yield fell from 4.18% at the end of 2025 to 3.97% on 27 February, 21 basis points over January and February (a basis point is a hundredth of a percentage point). Falling yields mean rising bond prices, and the flagship went from 114.1% net short at the end of January to 61.6% net long at the end of February.

Then the market turned. The 10-year rose 33bp to 4.30% by 31 March. The feeder’s first quarter 10-Q reads: “Fixed income trading proved challenging, as long bond positions detracted amid rising yields.” Positions in Italian, Canadian and French rates were among the worst; shorts in German, Australian and Korean bonds offset part of it. Man put the month’s bonds and rates contribution at −1.50%.

That is the cost the program describes in advance. The signal was long because prices had risen, and prices fell before the lookback caught up.

It flipped fast. By the end of March the flagship was 143.8% net short, and in April the second quarter 10-Q says “Fixed income trading contributed to gains, led by short ten-year Japanese Government Bond positions.” Then the trend stalled. The 10-year rose 10bp in April and 5bp in May, and slipped 1bp in June, while Australian, European and Korean bonds rallied against the book’s shorts.

Meanwhile the fund still made money. The Class A Series 1 units returned 8.39% after fees in the first quarter and 3.20% in the second. Compounded, that is the half’s 11.86%.

Currencies and stocks paid the first half

Currencies, stock indices and energy paid for the first half, and the trading company’s own table shows it: the 10-Q’s notes give gains and losses by sector, realized and unrealized, and I added the two for each sector.

Source: Man-AHL Diversified I LP 10-Q, quarter ended 30 June 2026, notes to the trading company’s statements; forwards and futures combined, excluding currency translation and spot contracts. The trading company reported a total trading gain of $17.93m for the half.

Interest rate futures lost $0.72m in the first quarter and $0.33m in the second. On the trading company’s $145m that is about 0.7% of capital, and the rest of the book paid for it many times over.

The feeder’s share of the trading company’s gains was $8.07m of the $17.93m, about 45%. Fees and other expenses took $1.90m, an annualized 5.77% for the Class A Series 1 units, and interest on its cash added back $1.18m. That leaves $7.36m, about 11.4% of its average capital of $64.8m.

The flagship tells the same story at a different size. Its bonds and rates lost 2.48% gross over the half, in four losing months out of six. It is a different fund run at different risk, so I compare the direction of the two, never the size.

So through June, the bond short the coverage credits was a cost. Currencies, stock indices and energy carried the half.

The rule you can check in advance

The fund’s net bond position going into a month, set against that month’s yield move, called the bond leg’s sign in two of 2026’s three big months. A rule fitted after the outcome can explain any quarter; this one can be checked before the month starts, because Man’s report gives the fund’s net bond exposure at each month end. Going into a month, which way is the book facing, and which way do yields move?

Sources: the flagship’s monthly reports, December 2025 to August 2026, gross contribution; exposure is the prior month end’s; Treasury par curve month ends, my arithmetic.

In the five months the 10-year moved less than 25bp, the bond leg’s results net to about −0.2%. That is noise. In the three months it moved 25bp or more, the position going in called the sign twice. March, long into a 33bp selloff, cost 1.50%. July, short into a 31bp selloff, made 1.08%.

February broke the rule. The book went into a 29bp rally net short and still made 0.19%, because, in the first quarter 10-Q’s words, “Long SONIA and SOFR positions generated profits in rates trading”. SONIA and SOFR are the overnight sterling and dollar rates, and futures on them made money while the bond shorts lost. The US 10-year is my yardstick, not the book, which trades Japanese, Australian, European and Korean bonds as much as American ones.

The trading company’s longer record says the same thing in dollars, with a denominator.

Sources: the trading company’s derivative tables in the FY2022 10-K, the first quarter 2023 10-Q, the third quarter 2023 10-Q, the third quarter 2024 10-Q and the second quarter 2026 10-Q, realized plus unrealized; share of the limited partners’ capital at the period’s end ($188.0m, $182.2m, $194.9m, $172.0m, $145.0m); Treasury par curve. My arithmetic.

In 2022 the bond leg was the year. The 10-year rose from 1.52% to 3.88% and the units returned 14.63%. Interest rate futures made $24.8m of the trading company’s $29.7m of net gains on derivatives, about 84%. Sized against the yield move, the two trend periods paid roughly 3% to 6% of capital per 100bp. In the third quarter of 2024 the 10-year fell 55bp, every month, and the bond leg still lost 1.9%, because the book came into the rally short. The 10-Q says fixed income lost in July, was flat in August, and that in September “The Partnership’s transition to long fixed income over the quarter was rewarded”. That is the lag, with a price on it.

Where the machine breaks: 13 March 2023

A trend follower’s worst day is a trend that reverses at once. AHL’s own record has one. Its first quarter 2023 10-Q reads: “A decline of 61bp on 13th of March for US 2-year Treasury yields was the largest decline in over 40 years, was against the prevailing price trend, and was detrimental to a short in the instrument and indeed all other tenors of US treasuries traded by the Partnership.”

The 61bp is AHL’s market figure. Treasury’s par curve, which marks a constant 2 year maturity, shows 57bp, from 4.60% on 10 March to 4.03% on 13 March. Over three sessions from 8 March the fall was 102bp. The trading company’s interest rate futures lost $8.49m that quarter. The units lost 5.60% in the quarter and finished 2023 down 4.57%.

The rule had done what it was built to do, short a market that had been falling, and the market reversed faster than a volatility target can shrink a position. I read the mechanics this way. AHL doesn’t publish its estimator, but in the 2012 study’s version σ moves slowly by construction: each new day carries about a sixtieth of the weight, so three sessions barely move it before the damage is done. The volatility target protects against a market that gets noisier over weeks. It has no answer to a few sessions.

The 2012 study found the same shape in 2009, when trend following took “sharp losses when the crisis ends in March, April, and May”. Between the big years the units’ returns were thin: 1.09% in 2024, and 5.45% in 2025 after a 14.52% loss in its first half, per the 2025 annual report and the 2026 10-Q.

September 2026 is the test

September is the month the rule has to get right, because July to September supplied the trend the first half lacked. The 10-year rose 85bp, from 4.44% on 30 June to 5.29% on 30 September, 54bp of it in September. Man’s reports already show the bond leg collecting: bonds and rates added 1.08% in July and 0.97% in August, and the flagship went into September 157.1% net short. On my reading of the table above, September should pay.

This is wrong if Man AHL Diversified’s monthly report for September 2026, published in October 2026, shows bonds and rates with a negative contribution for the month. In 2026 that line was crossed in four of eight months, but never in a month when the 10-year moved 25bp or more in the direction the book already faced. There has been one such month, July, so the evidence is thin. The filing check follows: the feeder’s 10-Q for the quarter ended 30 September 2026, due on EDGAR by 16 November 2026, carries the trading company’s interest rate futures line for the quarter.

On the day each lands, you can check it yourself. The flagship’s report prints the month’s bonds and rates contribution. On EDGAR, search CIK 1052354 and open the 10-Q; in the notes, find the trading company’s table of gains and losses on derivatives and add the realized and unrealized lines for “Interest rates”.

A few limits sit under all of this. The 10-Q gives sector dollars only by quarter, its monthly paragraphs are the fund’s own account in a filing it is required to make, and its statements are unaudited. The flagship’s monthly figures are gross, before fees, for a larger fund than the SEC feeder. The 2017 study’s century of results, positive before costs in every decade from 1880 to 2016, is a backtest of the published rule that no fund traded for most of that span.

The open question is the turn. The flagship is short 157.1% in 10-year terms, and the next reversal will cost before the lookback catches it; how much depends on how fast it comes. Trends pay over months. Reversals collect in days.

Further reading

• AQR’s Managed Futures Fund Bet Against Treasuries. AQR’s Own Trend Rule Says When It Stops: the same trend rule run tenor by tenor on the Treasury curve, with the yields at which it flips

• Inside the $168B Systematic Fund Making Money While Others Lose (Man AHL): the firm and the research process behind this machine

• The $5.6 Billion Unwind: How Trump’s Tariff Shock Broke the Quant Machine: what a fast reversal does to systematic books, from April 2025

Next Monday: a named fund’s bet, read from its own filing, with the position.

→ Man AHL’s 2022 bond rout, rebuilt quarter by quarter from its filings on Patreon, open to everyone.

Navnoor Bawa · YouTube · LinkedIn · Patreon

Logo: Man Group plc

In Q2, Chase Coleman’s Tiger Global sold about $3.7bn more stock than it bought. Its 13F still showed a book that grew, because prices rose faster than Tiger sold. The selling sat outside chips; inside them, Tiger swapped names and let the rally make semiconductors a bigger share of the book. The one chip line it built in both quarters is Intel, worth owning beside Tiger only below today’s price. Anyone long Broadcom, Lam or TSMC had a large holder cutting into the rally.

Lam grew while Tiger cut it

A 13F’s value column is each line’s share count times the stock’s close on the quarter’s last trading day. The Form 13F instructions value every holding “at the close of trading on the last trading day of the calendar year or quarter.” So the column moves with prices whether the manager trades or not. The filing lists long US-listed positions only, at quarter end, up to 45 days later, with no shorts or swaps.

Lam Research shows how far the two can part. Tiger cut its Lam share count by 18.9% between 31 March and 30 June. The same line rose $537.5m in value, because Lam’s close doubled.

Source: Author’s calculation from Tiger Global’s 13F-HR filings for 31 March and 30 June 2026 · as of 30 June 2026.

The price move added more than Tiger took out, so a list sorted by value shows Lam among Tiger’s biggest gains in a quarter when Tiger was selling it. Run that split across all 46 lines Tiger reported in June and the 17 it closed, and the real question appears: how much did Tiger sell, and from where?

Most of the selling sat outside chips

The answer comes in four steps.

1. Priced at the June close, Tiger’s share changes and the lines it closed net to about $3.7bn of stock sold.

2. Lines outside chips and storage carry nearly all of it, led by Alphabet, whose Class A count Tiger cut by 45.4%.

3. Inside chips and storage, money moved from Broadcom, TSMC, Lam, Nvidia and Applied Materials into Intel, AMD, Seagate and Cerebras, a newly listed AI chipmaker. At June prices the two sides came out within a tenth of a billion of each other.

4. Chip prices ran, the VanEck Semiconductor ETF (SMH) rose 71% in the quarter, so chips became a larger share of a book Tiger was shrinking.

When SMH rises 71% in a quarter, a chip line’s value mostly tells you what SMH did; only the share column says what Tiger did, and only the sum of every line says how much.

For your book, that changes two readings. If you own Broadcom or TSMC, Tiger’s count fell 51.1% in Broadcom and 12.3% in TSMC. If you are weighing whether to follow Tiger into chips, the filing says it added close to nothing to them on net.

Tiger had no Intel line at the end of 2025 and opened one in the first quarter. In the second, its count rose from 1,638,700 to 4,252,690 shares. Over that quarter Intel’s close went from $44.13 to $139.63.

Tiger was also alone. Coatue and Lone Pine, two growth funds with the same kind of public book, ran through the same rally, and put through the same split, both come out as net buyers.

The paid half holds the swap table, the same split on Coatue and Lone Pine, and the Intel entry with its levels. The call settles on 16 November 2026, when Tiger’s filing for the September quarter is due; the 45-day deadline falls on a Saturday.

The question it settles is whether a 13F follower should buy Intel now, wait for a lower price, or skip it before Tiger’s November filing.

Below the paid line:

• The decomposition table: Tiger’s reported change split into price, trading and exits.

• The table of ten chip and storage lines, each flow at the June close.

• The peer table: the same split on Coatue and Lone Pine, beside Tiger.

• The position: instrument, entry, levels, sizing and the dated line that takes it off.

Read more

In January 2026 Brevan Howard made Fash Golchin lead portfolio manager of its flagship fund and gave him its largest cash allocation, Hedgeweek reported. By August the ten largest allocations in that fund held 71% of its adjusted capital.

From the outside, Brevan Howard looks like a pod shop. Its website advertises “150+ Portfolio Managers“ and sells “multi-portfolio manager strategies.” We described it that way ourselves in January, when we wrote that it ran “20+ portfolio managers with individual mandates” and read its flat 2025 partly as those books netting each other out.

But the flagship’s own disclosure says otherwise. BH Macro Limited, the listed fund that puts nearly all its money into the Brevan Howard Master Fund, reports every month how that fund’s capital is spread. In August 2026 the three largest allocations held 41% of it, and three allocations each ran more than a tenth of the money. Brevan Howard is a 150-manager firm, but its flagship is a concentrated macro fund built on about ten allocations.

For anyone who owns BH Macro, that is the position: about ten books, three of them holding two fifths of its adjusted capital, and the largest run by Golchin’s team of eight. It was not always so. In early 2025 as much as 43% of the fund’s capital sat in temporary books outside any trader’s allocation. Those books have since shrunk to 7%. Six parts of the record show how the flagship got here:

• The 2011 machine

• The flagship shrinks inside the firm

• Inside BH Macro’s monthly report

• From temporary books to allocations

• 2025, the flat year

• The December 2026 test

Source: BH Macro Limited monthly shareholder reports, as printed. The temporary books were called “strategic books” in April 2024 and “Risk Overlay Strategies” from August 2024.

How to read it. Adjusted capital is each allocation’s capital scaled by Brevan to the same severe loss risk, so a cautious book and a geared one can be compared; the weights are not published. The temporary books are tactical positions the fund opens and closes on its overall risk, outside any trader’s allocation. The report counts them in the total but not in the top three or top ten, so when they are large the printed shares understate how much of the traders’ own capital the biggest allocations run.

The 2011 machine

In April 2011 the consultant Cliffwater wrote a due diligence report on Brevan Howard for the Rhode Island Treasury, and a redacted copy is public. It is an early look from inside, with two limits: it is one allocator’s consultant writing at one date, and parts are blacked out.

Five partners had founded Brevan Howard in 2002, all from Credit Suisse First Boston: Alan Howard, Christopher Rokos, Trifon Natsis, James Vernon and Jean-Philippe Blochet. The Master Fund launched in April 2003 with $870 million. By 31 March 2011 it held $24.7bn of the firm’s $32.6bn, or 76%. In money terms, Brevan Howard was one fund.

The firm had 339 employees. Cliffwater counted “more than 85 investment professionals, including 45 traders, 14 economists, and 4 strategists”. The fund ran on those traders. “The fund is comprised of a group of approximately 45 different traders,” the report says, each with “a strict, unique written mandate that specifies the limits of the trading strategy including capital allocation, risk limits such as VAR and DV01”. VaR is value at risk. DV01 is the loss from a one basis point move in rates.

Each mandate came with a ladder of losses:

1. At a 4% loss, the trader reviews positions with the risk officer.

2. At 8%, the trader is “subject to a mandatory reduction in risk limits”.

3. At 12%, the trader “may be required to sharply reduce risk or take time off from trading”.

Cliffwater added a sentence that tells you how the ladder worked in life: “In practice, a trader’s capital will be cut sharply before they reach the 12% drawdown”. The top rung was a backstop. Capital was cut well before it.

Capital came from a committee. “The firm’s investment committee distributes capital among the individual traders,” the report says. It met monthly, reviewed every trader’s capital formally once a year, and could move capital away from a trader whenever it judged it necessary.

The risk function sat outside the star system. Aron Landy, then chief risk officer, reported to joint chief executive Nagi Kawkabani “rather than to Mr. Howard”. Even the founder was a line on the risk sheet: “Mr. Howard does not participate in these discussions and his book is specifically reviewed and discussed at the weekly meetings.”

Put those pieces together and you have the risk machine people now associate with the big platforms that split capital across many managers. Traders hold written mandates. A committee hands out capital, and losses are cut on a schedule. A risk chief who does not answer to the founder enforces it.

What differs is scale and scope. The 2011 machine had about 45 traders, and it governed one fund that held three quarters of the firm’s money. Today the firm advertises 150+ portfolio managers across many products, and Landy, the 2011 risk chief, runs it as chief executive. The machine kept growing. The flagship’s place inside it did not.

By 2011 Brevan Howard had built a pod shop’s risk machine and pointed it at a single macro fund.

The flagship shrinks inside the firm

The Master Fund grew fast before the firm grew around it. Its year end assets went from $2.65bn in 2003 to $15.3bn in 2007 and $24.7bn in 2010, per the same Cliffwater report.

Then the firm grew around it. Brevan Howard now reports “over 1000 team members” in nine hubs, from London and New York to Bengaluru and Austin. It sells “multi-portfolio manager strategies, single-portfolio manager strategies and thematic co-investment opportunities”. In October 2019 the firm said Alan Howard would step down as chief executive by the end of that year, with Landy taking over, and that “Alan will focus on his own trading activities,” Reuters reported.

The manager ran about $34bn at the end of 2025, according to BH Macro’s chair. The Master Fund’s net assets that day were $11.06bn. Divide one by the other and the flagship is roughly a third of the firm, against 76% in 2011. Firm assets barely moved over that span, from $32.6bn to about $34bn. The flagship’s share fell by more than half.

Source: Cliffwater for the Rhode Island Treasury (2011); Brevan Howard Master Fund audited statements and BH Macro chair’s statement (2025).

That correction applies to us too. Our January piece called Brevan a “$34B Master Fund” and built a scale argument on it. The $34bn is the whole firm; the Master Fund was about a third of that.

The shrinking continued in 2025. The Master Fund lost $1.59bn to net redemptions during the year, against $655m of investment gains before the fees its feeder funds charge, so it began the year near $12.0bn and ended it at $11.06bn. The interim statements published on 30 September put net assets at $11.39bn on 30 June 2026. Hedgeweek puts the Master Fund and its sister, Alpha Strategies, at roughly $11bn each. The flagship is now one of two equals.

The fund also changed how it holds its traders. In October 2023 BH Macro said the manager expected the Master Fund to redeem its stakes in three sister master funds, BH Alpha Strategies, AS Macro and FG Macro, by 2 January 2024. Direct allocations to the teams would replace them, and holdings in other Brevan funds would fall to about 17% of assets. In 2011 Cliffwater had put that share at about 7-8% of the fund’s net asset value.

The same report sorted “every portfolio manager (’PM’) trading on behalf of BHMF” into two styles. Alpha traders had “more narrowly defined trading mandates and smoother return profiles”, and the report’s table sums up their P&L as “Smoother - try to hit ‘singles’”. Macro traders were “Lumpier - try to hit ‘home runs’”. Alpha held 55.6% of the fund’s capital and Macro 44.4%, so in late 2023 the narrow mandates were the larger half.

The partnership that had employed Brevan’s London traders since 2002 changed shape as well. On 10 May 2025 its business moved into a new company, and on 31 May Companies House recorded Alan Howard, Landy and Natsis leaving it as members. The accounts say the partnership is to “remain inactive”.

The flagship went from three quarters of Brevan Howard to a third, and the business of many managers that the label describes is the firm that grew up around it.

Inside BH Macro’s monthly report

BH Macro’s report defines its own terms in a footnote. From the January 2025 report:

“Adjusted Capital applies weightings to the traders’ nominal capital in order to normalise the data to a consistent level of potential severe downside. Total Capital exceeds AUM, accounting for gearing. Core BHMF Trader Allocations excludes Risk Overlay Strategies (’non-core books’).”

The temporary books are “temporary tactical books that will open/close based on the aggregation of risk factors across the fund as well as investment opportunities at the time.” The top 3/5/10 calculation “excludes Risk Overlay Strategies from the numerator but includes them in the denominator.”

One detail limits how far the table can be pushed. The report gives the temporary books as a share of “total BHMF capital”, a different base from adjusted capital, so the two figures cannot simply be divided into each other. We read the printed top ten share as a floor on the traders’ share and go no further.

Now the case, end to end, for August 2026:

1. The ten largest allocations held 71% of adjusted capital, temporary books included in the total.

2. The temporary books held about 7% of total capital, so the floor sits close to the traders’ own figure.

3. Three allocations each held more than 10% of the fund’s capital, counted at nominal value.

4. On the printed figures, ten allocations hold at least seven tenths of the adjusted capital and three hold two fifths.

Sized roughly, a 10% loss across those three allocations would cost the fund about 4% of its adjusted capital: 41% times 10%. That is an order of magnitude, not a forecast, because adjusted capital is not net assets.

“Allocations” is the careful word. The report says traders manage capital “directly or indirectly”, and the largest allocation has a team behind it. Golchin’s appointment as lead portfolio manager came with a trading team of eight, and his own $1.4bn FG Fund was to stop running outside money at the end of June, Hedgeweek reported citing Bloomberg.

Compare that with the picture our January piece drew. Twenty books of equal size would put half the capital with the top ten, and less once the temporary books are counted in the total. The fund printed 71%. No firm that runs many managers publishes a comparable figure, so the flagship is measured here against that benchmark, its own past and its own sister fund.

For an allocator, the portfolio manager count answers a question about the firm. The question about the flagship is the top ten share, read beside the temporary books, and it sits in a free PDF every month.

In 2026, on the fund’s own reporting, the flagship’s capital sits with about ten allocations, and three of them hold about two fifths of it.

From temporary books to allocations

In April 2024 the top ten held 84% of adjusted capital and the top three 47%, with 20% of capital in what the report then called “strategic books”. After dipping to 13% in August, the temporary books rose to 34% by December 2024, when the top ten’s printed share was 50%. In January 2025 the temporary books reached 43% and the printed top ten 44%.

The table cannot say how much of that fall belongs to the traders. With nearly half the capital outside their allocations, their printed share had to fall, and the report’s two bases do not allow a clean adjustment. What moved without any arithmetic is the temporary books themselves. For 11 of the 12 months of 2025, more than a fifth of the fund was run that way, outside any named trader’s allocation.

2025 went badly early. In March, with the Master Fund down 5.4% for the year, Landy imposed stricter trading limits on some portfolio managers, Hedgeweek reported from an investor letter seen by Bloomberg. That is the 2011 machine’s tool, a cut in a trader’s limits, used at the top of the firm.

Through the middle of the year the fund looked most like the label. From March to November 2025 the printed top ten ran between 50% and 59%, and the temporary books held 27% to 28% of capital from June to November. In August, September and October 2025 no allocation held more than a tenth of the fund. Our January piece described the fund in exactly that window.

Then the books closed and the allocations grew, in this order:

1. By December 2025 the temporary books were down to 18% and the printed top ten up to 69%.

2. In January 2026 Golchin became lead portfolio manager, and his own fund was set to close.

3. From February 2026 four allocations held more than a tenth each, and four still did in April and May.

4. By July the temporary books were 7% of capital and three allocations were above a tenth.

The reports give the sequence, not the reasons. Read in order, they show capital leaving the books run on the fund’s overall risk and going to a small number of named allocations, one of them led by the trader Brevan had just promoted.

The fund’s worst month sits in that stretch. March 2026 cost BH Macro’s dollar shares 6.22%. In the 234 months of BH Macro’s published NAV table, which starts in March 2007, no other month fell by more than 4.11%. The March report puts the losses on “interest rate and equity trading”, with the equity losses from directional exposure to Asian indices, in a month when four allocations held more than a tenth each.

The flagship has moved its risk out of books run on the fund’s overall exposure and back into a few named hands.

2025, the flat year

Chris Rokos, one of the five founders in the 2011 report, left to build his own firm, and 2025 is where his record and Brevan’s part. Rokos Capital made about 21% last year, according to Hedgeweek citing Bloomberg, which put its assets at roughly $22bn in January 2026. Brevan’s flagship was flat for its shareholders. The Rokos figure is a reported estimate; Brevan’s below come from its own documents.

BH Macro’s net asset value (NAV) per share rose 0.8% in dollars and 1.4% in sterling after fees. Inside the Master Fund, before the feeder funds’ management and performance fees, the main dollar share class returned 2.91%. Hedgeweek reported an estimated 0.75% for the Master Fund and about 8% for Alpha Strategies.

The fund’s own attribution shows where the year went. Equities and commodities added about eight points to the sterling shares; FX, rates and digital assets took nearly eight back, with FX alone costing 4.87. Discount management, the buybacks, added 0.74, which is most of the 1.4% the sterling shareholder kept. Gains in one set of markets were spent covering losses in another, so on this attribution the netting reading in our January piece holds up.

BH Macro’s chair, Richard Horlick, called the year’s return “less than satisfactory whilst it is within expected bounds of return”. Among the reasons he gave for the shares’ discount was “a lack of returns from the two historically principal drivers of return, interest rate trading and FX trading”. He added that “the Board is encouraged by the changes to process which the Manager has implemented during the course of the year”. The January 2025 report had already put the year’s first losses “primarily from interest rate and FX trading”.

Shareholders stayed anyway. The sterling shares traded at a discount of 6.7% to 10.6% during 2025, and the company bought back £115m of stock, after £116m in 2024. In February 2026, votes to close the share classes failed, with 96.23% of sterling votes and 99.91% of dollar votes against closure.

Source: BH Macro Limited monthly shareholder report, August 2026 (monthly USD NAV per share), author’s calculations.

Since 2024 the fund has returned 2.8% a year at 9.0% volatility, against 12.7% at 7.6% in its first era, so it now takes more risk for far less return. The pattern has turned before: the weak 2013 to 2018 stretch was followed by 28.89% in 2020 and 21.17% in 2022.

2025 was the year equity and commodity gains were spent covering losses in the fund’s two traditional engines, while the founder who left made 21%.

The December 2026 test

One claim is on trial: that the flagship is a concentrated fund, built on about ten allocations, and not a platform of many small books.

The firm’s portfolio manager count is not a test, because the Master Fund’s reports do not disclose how many managers trade for it. Neither is the printed top ten share on its own, because it falls whenever the temporary books grow. The test has to hold the temporary books low and then look at the allocations.

2025 shows why the temporary books matter. From March to November that year the printed top ten held 50% to 59%, and for three months no allocation held more than a tenth. But the temporary books never fell below 21% of capital in that stretch, too large for the printed shares to settle the question. In August 2026 the top ten held 71% and the temporary books 7%. If the temporary books come back above 15%, the printed shares stop being a clean test until they fall back.

We are not forecasting the fund’s return. It was up 1.78% in dollars through August 2026, after the 6.22% fall in March.

The limits, stated once. Adjusted capital is Brevan’s own weighting, and the weights are not published. The temporary books are reported against total capital, a different base from adjusted capital, so we read the printed shares as they stand. The concentration figures are the manager’s own month end estimates, and the definition of the temporary books changed during 2024. An allocation can be a team. The table shows twelve month ends; the other months cited are linked where they appear. The reports do not disclose the number of portfolio managers inside the Master Fund. Drawdowns from month end data understate losses inside a month. Concentration and returns sit side by side in this piece, never as cause and effect.

A 150-manager firm can still own a flagship that rests on ten allocations, and the monthly report is where that shows.

The argument is wrong if, by 31 January 2027, BH Macro’s monthly report for December 2026 shows the temporary books at 15% of total capital or less and the ten largest allocations below 60% of adjusted capital.

Further reading

• How Millennium, Citadel & Point72 Structure Pods: Team Economics Behind $428B Multi-Manager Industry: how a true pod shop splits capital and pay across hundreds of teams, the structure this flagship turned out not to have

• D.E. Shaw, Millennium and Citadel Lock Up Cash for Years. Rokos Chose the Opposite: the redemption terms behind the same funds, the other half of how a flagship’s capital stays put

Next Thursday: a named fund’s bet, read from its own filing, with the position.

→ The screen that reads BH Macro’s monthly report for concentration, with all 26 months in a sheet on Patreon, open to everyone.

Navnoor Bawa · YouTube · LinkedIn · Patreon

Cover: Brevan Howard logo · Public domain · via Wikimedia Commons

On 27 August, AQR’s $3.2bn managed futures fund reported it was short every US Treasury future on 30 June. Anyone short duration beside the trend funds, or waiting to buy what they sell, needs to know when that selling stops. I read the filing line by line and rebuilt AQR’s published trend rule on Treasury’s daily yield curve to find out.

The claim: AQR’s published trend rule, which averages a 1-, 3- and 12-month signal and rebalances at each month end, does not need a bond rally to stop being short Treasuries. If the 10-year simply holds near 5.28%, the calendar and the bond’s carry turn its fastest signal long at a month end this autumn and its average net long by the end of December.

The stake: Trend funds built large positions against bonds in this selloff, the FT reported. A rule that stops selling with no move in price matters to every short duration book leaning on that flow.

The catalyst: The Fed meets on 27 and 28 October and on 8 and 9 December. AQR’s rule decides at each month end.

Wrong if: AQR Managed Futures Strategy Fund’s N-PORT for 31 March 2027, due on EDGAR by about 1 June 2027, reports a positive 10-year bucket in its US dollar DV01 (a 10-year book that gains when yields rise) while the published rule’s 10-year position, set at the 26 February 2027 month end on Treasury’s par curve, is net long.

In August I argued that Bessent’s doubled buybacks would not cap the long end, and said I was wrong if the 30-year held under roughly 5.10% for two weeks once the larger operations took effect on 9 September. It never did, closing at 5.63% on 2 October, so that leg stands; the 4 November refunding is the open test.

Trend funds rode the bond rout

The money on the other side of that market has done well. According to people familiar, as reported by the FT and relayed by Hedgeweek, Graham Capital’s Tactical Trend fund was up more than 31% for the year, Winton’s Diversified Macro 17.5% through late September and Aspect Capital’s flagship 21%. Not all of it came from bonds: the same report says Winton and Aspect also gained in energy. The FT said the computer driven funds had built large positions against bonds, and that the 10-year reached 5.342% on 1 October, its highest since 2002. Treasury’s par curve, which this piece uses throughout, peaked at 5.29% on 30 September.

None of the three publishes its futures positions, so I can’t check their books. I went to a large trend fund that has to show the SEC what it holds.

The usual way to frame a trend fund’s exit is a price trigger: the funds stay short until a rally through some level forces them to cover. AQR’s published rule, run on Treasury’s own data, says something stranger. Read on 2 October, its fastest signal needed a rally the 10-year has not produced in any month of the past year. Read at the October month end, when the rule actually decides, the same signal needs no rally at all.

AQR is short the Treasury curve

Form N-PORT is a registered fund’s holdings report to the SEC, made public as a quarter end snapshot weeks later. AQR’s report shows the book on 30 June and was filed on 27 August. The fund trades frequently, so the report shows direction, and size only as of that date.

The direction is plain. The fund was short 2-year, 5-year and 10-year note futures, the long bond and the ultra bond, with no long US Treasury future anywhere in the book. Those lines carried −$1,116.1m of notional, or 34.8% of net assets. It was also short all seven SOFR contract months it held.

Item B.3 of the form asks for the change in value of a fund’s portfolio for a one basis point move in rates, which desks call DV01. AQR’s US dollar buckets sum to +$825,142, positive in every bucket from one year to 30 years. On a $3.2bn fund, that is about 2.6bp of NAV for every basis point US yields rise.

The trigger resets each month

AQR’s 2017 paper on a century of trend following, by Brian Hurst, Yao Hua Ooi and Lasse Pedersen, tested one trend rule on 67 markets, 15 of them bond markets, with data back to 1880. It averages three signals in equal weight, built on 1-, 3- and 12-month windows, and it rebalances once a month. Each window is a lookback, the stretch over which the rule measures a market’s past return. Positive means long, negative means short, and each position is sized to the same volatility inside a book scaled to a 10% annual volatility target.

The return it reads is a return over cash, the definition AQR’s original trend paper tested across 58 futures and forward markets. So I take what a constant maturity 10-year earned over each window, yield plus price change, and subtract what a three month bill earned. The yield the bond earns over the bill is carry: what it pays you for sitting still.

Two things in that rule decide the exit, and neither is a fixed price. At each month end, every window’s start moves to a later month end close. And the bond earns carry: the 10-year closed at 5.28% on 2 October against 4.19% on a three month bill. A leg flips when the 10-year finishes the month below its window’s starting yield plus a small carry allowance.

Say the 10-year jumps 50bp in September and then goes nowhere. At the end of October the 1-month window starts from September’s high, so a flat October shows a small positive return over cash, just the carry, and the leg goes long. After a steady selloff, the start points the windows roll onto are higher than the ones they drop, so the trigger climbs toward the price even if the price stands still.

A trend rule’s trigger is a date with a level attached, and the date keeps moving.

The chain, one step at a time:

1. The 10-year has risen 110bp this year, to 5.28% on 2 October.

2. So at the September month end each lookback’s return over cash was negative, and the published rule went short on all three.

3. AQR’s fund was short across the curve at 30 June, at about 2.6bp of NAV per basis point.

4. In a flat market the trigger climbs to the price at the month ends, and the legs turn long one at a time, fastest first.

If you are short duration partly because the trend money is short, your thesis carries a clock that runs whether or not the 10-year moves. If you are waiting to buy duration once the selling stops, the stop can arrive in a market that has not rallied at all.

So price triggers describe half the exit, and time is the other half. One number decides how they combine: at what 10-year close does the fastest leg flip at the October month end?

What follows lets you decide whether to stay short duration beside AQR through the October Fed, or step aside before its rule does.

Below the paid line:

• The month end flip table: where each of the three lookbacks turns long on 30 October, with the arithmetic.

• The schedule a sideways market sets through December, in three month end rows, leg by leg.

• The filing line by line, with its rate risk per maturity and what the June book would have earned since.

• The position: instrument, entry, payoff and kill levels, size in risk units, and the exit date.

Read more

Elliott kept cash settled swaps on 4,042,300 Seadrill shares while it sold voting stock through September, its latest Schedule 13D amendment shows. Anyone holding SDRL trades beside a fund with 12.3% economic exposure that has sold in 2023, in 2024 and again this year. After my read of Elliott’s hedged Synopsys campaign, whose call is open until 16 November, I rebuilt its Seadrill record, which shows a trading book around a swap held since 2023.

The claim: Elliott trades Seadrill’s voting shares around a swap position it has kept between 3.9 million and 5.4 million shares since July 2023, on average buying lower than it sells, and September was its largest selling window since the spring of 2024.

The stake: 92.5% of the shares Elliott has had to itemise since 2023 sold at $44 or higher, while its unitemised buying came in a quarter when SDRL averaged under $40. A holder at the 30 September close of $44.67 is trading against the remainder of that supply.

The catalyst: Seadrill’s buyback, with $188 million of its $500 million authorisation left at 30 June, runs to 31 December 2026. Every share it retires lifts Elliott’s percentage without Elliott trading, which lengthens the selling Elliott can do before its filings stop.

Wrong if: Elliott’s next Schedule 13D amendment, filed by 31 March 2027, itemises a session sold at a weighted price below $44, or reports swaps on fewer than 3,917,981 Seadrill shares, its lowest count since July 2023, per EDGAR.

Seven of nine analysts rate Seadrill a Buy

The consensus on Seadrill is a Buy. Nine analysts rate it, seven of them Buy or Strong Buy, and their average target of $59.40 sits about a third above the price on MarketBeat’s tally, read on 30 September. Put Elliott’s 12.3% beside that and it reads like a second vote for the same upside.

The filings read differently. Elliott has sold, repeatedly, in the same price zone, and its buying has come lower down.

Nine filings show a seller’s band

Nine filings show Elliott selling Seadrill in one price band and buying lower down. It received 7,413,399 Seadrill shares on 3 April 2023, in exchange for its Aquadrill units when the two drillers merged, according to its first Schedule 13D, filed two days later. That filing said it believed the shares were undervalued and might propose changes to the board, the capital structure or a sale. It carried the stake at $163.6 million, $22.07 a share.

Eight amendments later, the tape of trades it had to itemise runs to 70 sale lots. I summed them: 4,759,651 shares for $230.5 million, a weighted $48.43. The cheapest lot went at $35.45 in May 2023 and the dearest at $55.09 in July 2024. The same filings show five purchase lots, 609,400 shares at a weighted $33.38.

Elliott’s purchases came after falls. It bought 100,000 shares at $46.94 on 5 August 2024, a session SDRL fell 5.7% to $46.96, and 509,400 more at $29.95 to $32.02 in December 2025. Between those purchases SDRL traded as low as $17.74 in April 2025, about 69% under its July 2024 high of $56.46. Elliott’s 13Fs show 3,703,746 shares at every quarter end from September 2024 to September 2025. It sat through the fall and bought after it.

The swap is the core position

The swap is the core position and the voting stock is the trading book, in a chain of five links, each with its own filing.

1. The swap is the core. It has referenced 3,917,981 to 5,444,919 shares in every filing since July 2023 that states it, and 4,042,300 now, 6.5% of the company.

2. The voting shares are the trading book. 92.5% of the itemised sales printed at $44 or higher; most of the itemised buying came near $30.

3. Seadrill’s buybacks hold the percentage up. The share count on the latest 10-Q is 17.3 million below the count after the merger. On the 2023 count, Elliott’s voting stake would read 4.56%, not 5.8%.

4. Small moves wait. Rule 13d-2 forces an amendment within two business days of a 1% change; a smaller move waits for the next amendment or the quarter end 13F.

5. The 5% line ends the filings. A holder at or under 5% owes no Schedule 13D.

A cash settled swap lets the two halves move apart. Elliott’s filing describes its swaps as giving “economic results that are comparable to the economic results of ownership” without the power to vote or dispose of the shares, and it disclaims beneficial ownership of them. So the swaps count toward the 12.3% of exposure it reports and not toward the 5.8% that keeps it a Schedule 13D filer. Elliott can shrink the stake the filing rules see while keeping most of its price exposure.

Its first filing reserved exactly that. Item 4 lets Elliott “enter into transactions that increase or hedge their economic exposure to the Common Shares without affecting their beneficial ownership”, and no amendment has changed it. The swap book was then built between the first two filings, as the voting shares started going out. Since July 2023 Elliott has cut the leg that votes by half and the leg that only pays by a quarter.

The worked case is 17 September. Elliott sold 150,000 shares that day at a weighted $47.99, in a range of $47.47 to $48.24. Nasdaq’s quotes put the day’s volume at 443,170 shares, so Elliott was 33.8% of everything that traded. SDRL closed at $48.11, up 0.6% after a 3.9% fall the session before, and by 30 September it was 7.2% lower at $44.67. One session proves nothing about demand; it shows how much of the tape this seller can take.

The filings leave one count open: how many more voting shares Elliott can sell before it passes 5% and its filings stop, and how far Seadrill’s buyback moves that line.

Whether to own SDRL beside Elliott, at what price, and after which filing turns on that number.

Below the paid line:

• A table of all nine filings: voting shares, swaps, cost and percentage at each amendment.

• The ten sessions of Elliott’s September Schedule 1, each against Nasdaq’s traded volume.

• Elliott’s runway before its filings stop, on two share counts, with the buyback’s addition.

• The position: instrument, trigger, levels, size and dated exit.

Read more

Berkshire’s Form 3 of 21 September hides how its Lennar stake was built: 65% of the Class A it has added since June was bought before any form reported a trade. Anyone reading the Forms 4 as the whole campaign is working from the smaller half. After my August look at its shrinking cash, I rebuilt the campaign from its June 13G, the Form 3, which lands less than one share past the 10% line, and both Forms 4.

The claim: The 10% filing marks the point at which Berkshire’s Lennar buying became visible, well after it began, and this quarter’s buying is four times last quarter’s.

The stake: A fund that follows Berkshire into Lennar at Monday’s $82.03 close pays more than Berkshire paid on any day in the filed tape, and less than its average since April 2025.

The catalyst: Berkshire’s Schedule 13G amendment, due 7 October 2026 under Rule 13d-2(c), covers a holding above 10% at the end of September. Its 13F for 30 September is due 16 November 2026.

Wrong if: Berkshire’s 13G amendment, due 7 October 2026, or failing that its 13F for 30 September, due 16 November 2026, reports fewer than 25,160,137 Lennar Class A shares.

The full piece rebuilds the campaign from the June 13G, the Form 3 and both Forms 4, prices the filed and unfiled cost, and names the position: Berkshire Is Betting on Lennar Up to $82 Before Its 13F Shows the Buying.

Navnoor Bawa · YouTube · LinkedIn · Patreon

Cover photo: Mark Hirschey · CC BY-SA 2.0 · cropped · via https://commons.wikimedia.org/wiki/File%3AWarren%20Buffett%20KU%20Visit.jpg

Berkshire’s Form 3 of 21 September hides how its Lennar stake was built: 65% of the Class A it has added since June was bought before any form reported a trade. Anyone reading the Forms 4 as the whole campaign is working from the smaller half. After my August look at its shrinking cash, I rebuilt the campaign from its June 13G, the Form 3, which lands less than one share past the 10% line, and both Forms 4.

The claim: The 10% filing marks the point at which Berkshire’s Lennar buying became visible, well after it began, and this quarter’s buying is four times last quarter’s.

The stake: A fund that follows Berkshire into Lennar at Monday’s $82.03 close pays more than Berkshire paid on any day in the filed tape, and less than its average since April 2025.

The catalyst: Berkshire’s Schedule 13G amendment, due 7 October 2026 under Rule 13d-2(c), covers a holding above 10% at the end of September. Its 13F for 30 September is due 16 November 2026.

Wrong if: Berkshire’s 13G amendment, due 7 October 2026, or failing that its 13F for 30 September, due 16 November 2026, reports fewer than 25,160,137 Lennar Class A shares.

The press read a value play

The consensus read is a value play. CNBC quoted CFRA Research’s Cathy Seifert calling it a “classic Berkshire value play”. It counted a 93% rise from the 13.4 million shares Berkshire held at 30 June to 25.9 million across both classes, and reported $212.4 million and $136.4 million of purchases in two filings. The arithmetic is right. It leaves open how much of the rise sits on the filed tape. Berkshire’s Form 3, filed on 21 September, reports 21,050,601 Class A shares on 17 September, the day the holding crossed 10%. Its June 13G and 13F put the June figure at 13,111,741.

Nothing between those two dates required a report. So 7,938,860 shares arrived without one, and only the 4,327,533 bought since the crossing sits on a filed tape. The unfiled shares outnumber the filed ones by 1.8 to 1. Berkshire’s cash was already falling: it dropped to $359.2 billion at 30 June, the first decline since 2022. Its own 10-Q confirms the figure.

Three filings give the gap

Three counts give the gap. Berkshire held 13,111,741 Class A shares at 30 June, the figure in both its 13G and its 13F, a long only snapshot of that date. It held 21,050,601 on 17 September, per its Form 3, and 25,378,134 after its last Form 4, for trades through 25 September. The first step adds 7,938,860 shares and the second 4,327,533, for 12,266,393 since June. Set against that whole gain, 65% arrived before the reporting clock started and 35% after it. The Forms 4 show the second number and nothing else.

The June quarter added 3,012,099 shares, from 10,099,642 at 31 March. So this quarter’s 12,266,393 is 4.1 times last quarter’s with days still to run. Four times the June addition on top of the June holding is 25,160,137 shares, the count the claim above must clear. As shares of the 210,506,003 Class A that Lennar counted on 31 May, the three holdings are 6.2%, 10.0% and 12.1%, and the 13G’s own 6.2% agrees with the first.

The middle one is not a round accident. Ten percent of that count is 21,050,600.3 shares, so the Form 3’s 21,050,601 sits less than one share past the line on the last count Lennar had published. Several of Berkshire’s earlier Form 3s look different. They show round holdings just under the line at Bank of America (9.99%) and BNY Mellon (9.998%). Southwest (10.19%) and HP (10.35%) were loose crossings.1

I read Lennar’s as a reporting cut chosen with the count in hand: the first tranche runs to the edge of Section 16 and the reporting begins there. The buying itself did not pause, since 677,749 more shares followed on the same day. Lennar’s buyback had also made the May count stale. A holder may still count against the latest published report. Rule 16a-1(a)(1) takes the ten percent test from the Section 13(d) rules. Rule 13d-1(j) lets a holder rely on the issuer’s latest report.

Section 16 starts its clock partway through

Section 16 reporting begins at the line. A holder of more than 10% of a class files a Form 3 within 10 days, and under Section 16(a) it reports each later trade before the end of the second business day. Rule 16a-2(c) exempts the purchase that carries a holder across the line. The reporting clock therefore starts only once the position is 21 million shares, and Berkshire’s starts on the line. Any holder that builds through 10% has a stretch nobody reports. What is specific here is its size, 1.8 times the filed tranche, how precisely it stops at the line, and a quarter four times the last.

Spread evenly over the 54 sessions to 16 September, the 7,938,860 shares work out at 147,000 a session.2 That is 5.6% of Lennar’s Class A volume. The filed lots from 17 to 25 September were 9.3% of the volume across those seven sessions. No filing dates the first tranche, so the comparison holds only if it was bought evenly.

One quantity is still open. The 25,378,134 Class A shares are 12.1% of that class on Lennar’s May share count, but Lennar has a second listed class with ten votes a share. Berkshire’s share of the votes is therefore smaller, and by how much decides whether it can ever matter to the board.

The paid half settles whether to own Lennar against the home construction ETF and at what ratio, with a trigger dated to the 16 November 13F.

Below the paid line:

• The table of the six sessions Berkshire bought, by class, price and volume share, and the unfiled shares’ cost.

• Which of Berkshire’s six insurers holds each lot, before and after.

• The vote table, Berkshire against Stuart Miller, and what Section 16(b) really costs a seller.

• The position: instrument, entry, levels, size and dated exit.

Read more

Elliott’s 13F for 30 June carries a $1.43bn put on the software ETF, in the quarter its partner Jesse Cohn joined the Synopsys board. Synopsys sits inside that ETF, and Elliott’s stake barely registers on the 13F. I dated each of Elliott’s sector puts against the campaigns it has made public. They cluster in the quarters the campaigns went public, and the software line points at Synopsys.

The claim: Elliott’s sector puts tend to arrive with its activist campaigns and sit on the funds that hold its targets, so its June book is long Synopsys with the software sector hedged.

The stake: Since the 30 June report date Synopsys has slipped 4.6% while the software ETF rose 17%, which leaves Elliott’s campaign stock back near the price at which its stake surfaced in March.

The catalyst: Elliott’s 13F for 30 September, due by 16 November 2026, shows whether the software hedge survived a quarter in which the pair lost 18%.

Wrong if: an 8-K reports Jesse Cohn leaving the Synopsys board, or Elliott’s November 13F shows neither a software put nor any Synopsys shares.

Coverage reads Elliott’s puts as bets or skips them

The consensus treats Elliott’s puts as bets or leaves them out. A Benzinga headline in August 2025 had Elliott betting against the S&P 500, the Nasdaq 100 and Nvidia. In July, 24/7 Wall St listed Paul Singer’s top picks as five longs and skipped the puts entirely.

In August I showed that Schonfeld’s put lines ran 60, 3 and 129 across three consecutive filings, with no rule requiring a reported book to be continuous. Elliott’s put lines persist from filing to filing, and what moves them is outside the filing: its campaigns.

Take the new software line. Elliott’s put on the iShares Expanded Tech-Software Sector ETF, IGV, was absent at 31 March and covered 15,750,000 shares at 30 June, more than six times any earlier software put it filed. In between, Synopsys signed a cooperation agreement putting Cohn on its board from 1 June. The fund’s own schedule for 30 June holds $434.0m of Synopsys. Elliott’s own 13F lists 227,500 Synopsys shares worth $101m, against a stake the Wall Street Journal reported on 22 March as worth billions. The rest is exposure the form does not carry.

A 13F put is priced at the stock

A 13F put line is the stock under the option, shares multiplied by the close at quarter end. The SEC’s Form 13F guidance sets the rest: the form reports long positions in listed US securities at quarter end, filed up to 45 days later. A manager lists only the options it holds, never the ones it writes, and for an option the columns describe the underlying stock.

Elliott’s software line shows the arithmetic. Its 15,750,000 shares at IGV’s 30 June close of $90.60 make exactly the $1,426,950,000 in the filing, the stock behind 157,500 contracts. I checked every put line in the June filing the same way, and each equals its shares times that day’s close. The strikes, the expiries and the premium are nowhere in the document.

Some of the lines are large against the funds themselves. The IGV put covers 10.6% of IGV’s $13.43bn of net assets at 30 June. Its put on IHI, the medical devices fund, covers 20.4% of that fund’s $3.02bn, beside a call on 1,400,000 IHI shares.

The form also leaves out the stake itself. Elliott’s Form ADV brochure describes its hedged strategies as ones in which a long position in one asset is associated with a short position in a different one. When the long is an activist stake held largely as economic exposure, which a 13F does not report, the filing shows the short leg and hides the long.

PepsiCo shows the order of operations. Elliott’s filing for 30 June 2025 was amended in November, when confidential treatment expired, to add puts on 4,000,000 PepsiCo shares worth $528m. Elliott disclosed its $4bn stake on 2 September, two months after a quarter end at which it already held puts on the stock.

The reason to hedge the sector is the length of a campaign. A board seat, a strategic review or a breakup takes quarters to land, and the target can fall with its sector in the meantime. Selling the stake would give up the votes the campaign needs, so a put on the fund that holds the target covers the sector’s move and leaves the stake whole.

That leaves one thing to test. When Elliott makes a campaign public, does a put appear or grow on the fund that holds the target in the same quarter?

Whether to own Synopsys against the software ETF into Elliott’s 16 November 13F, and at what size, rests on the campaign timing table.

Below the paid line:

• Sixteen Elliott campaigns dated against the quarter a put appeared on a fund holding each target, with the base rate

• The weight of each target inside its fund, from nine N-PORT schedules

• Fifteen filings of Elliott’s put book, and the one campaign sector it leaves bare

• The position: entry, payoff and kill levels, staged sizing, and the November test

Read more