The log
Claim · evidence · falsifier
Mark A. Stevens
NVIDIA director · Sequoia veteran · trustee, 3rd Millennium TrustConfirmed · Form 4 + Form 144
Did
Sold 1,848,501 shares 31 Aug – 2 Sep at $220–226 ($410.8m), then 1,022,239 more on 3 Sep ($235.6m). Combined $646m in four trading days. Filed a Form 144 on 2 Sep declaring intent to sell a further 5,000,000 shares (~$1.09bn) through Merrill Lynch. Retains roughly 26.5m shares.
Document
Form 4 filed 2 Sep and 8 Sep 2026; Form 144 filed 2 Sep 2026. Selling entity is the 3rd Millennium Trust. The 144 states the shares were acquired by private placement direct from the issuer on 6 December 2006, and discloses prior three-month sales: 4 Jun 500,000 ($109.9m); 18 Jun 885,000 ($186.0m); 31 Aug 585,000 ($128.9m); 1 Sep 63,501 ($14.0m).
Consensus read
A director sold $646m without a trading plan, at a time when insider selling is running ten to one against buying and the AI complex is visibly wobbling. The absence of a 10b5-1 plan means the sale was discretionary, which means he chose this moment. Informed insider, distribution at the top, take the hint.
Contrarian read
The missing 10b5-1 plan is evidence of less information in this trade, not more — and the Form 144 is the proof. A seller acting on private information wants a plan precisely because a plan is the affirmative defence against the accusation. Forgoing that protection at $646m, and then publishing the next $1.09bn in advance on a public form with a named broker and a fixed date, is not how an informed seller behaves. It is how someone behaves who knows the sale is uninteresting and does not expect to be second-guessed.
Evidence that discriminates
Three facts cut against the consensus read and could not be explained by it.
He pre-announced. The Form 144 telegraphs the next 5,000,000 shares. An informed seller's dominant strategy is to sell quietly into liquidity; broadcasting a billion-dollar overhang moves the price against yourself. This fact is inconsistent with the informed-seller hypothesis and fully consistent with a scheduled liquidity event.
The stock is from 2006. Acquired by private placement twenty years ago, so basis is effectively zero and the entire proceeds are gain. There is no basis decision left, and no tax reason to prefer this quarter over the next. What is left is a calendar — and calendars in trusts are set by trust and estate deadlines, not by charts.
He sold about 10% of the position. Roughly 2.87m shares against ~26.5m retained. An insider acting on a negative view does not leave 90% on the table; the retained stake is nine times the disposal and is the largest single piece of evidence about what he actually thinks.
Inference, not document: the 144's fixed sale date and the appearance of the 970 Foundation alongside the trust suggest a charitable or trust-administration trigger. The filings do not say this. Two live hypotheses remain — scheduled trust/estate administration, or simple diversification out of a position that has become absurdly dominant — and I cannot separate them from the documents available.
What would falsify meIf the 5,000,000-share Form 144 sale is followed by further large discretionary sales that take the retained stake materially below ~20m shares, or if the selling continues after a sharp price decline rather than pausing, then this is directional and the consensus read wins. A holder managing a trust calendar sells into a schedule; a holder with a view sells into weakness.
Pattern
Break in cadence, not in direction. He has been a persistent seller all year: $296m across two June dates before this. The four-day cluster is new; being a seller is not.
Stefan Larsson
Chief Executive, PVH CorpConfirmed · Form 4
Did
Bought 14,179 shares on the open market on 11 Sep 2026 at $70.53 — $1,000,001.
DocumentForm 4 filed 11 Sep 2026, trade date 11 Sep 2026.
Consensus read
Insider buying is the cleaner signal — insiders sell for a hundred reasons and buy for one. A CEO putting a million dollars of his own after-tax money into his own stock, in a month when insiders are selling ten to one, is expressing genuine confidence. Worth following.
Contrarian read
The precision destroys the signal. This trade was sized to produce a disclosure, not a position — and a purchase optimised for how it reads is, by construction, a communication rather than an investment. The rule I would take from it is the inverse of the usual one: weight an insider purchase by how inconvenient it is, not by how large. This one was maximally convenient.
Evidence that discriminates
$1,000,001 is not a share count. Someone buying because they want the stock specifies a quantity or a limit price and accepts whatever total falls out — 14,000 shares, or "up to $70." Landing on $1,000,001 requires working backwards from the dollar figure and adding a dollar so the number clears a round million in the reporting. The target was the headline, not the holding. This fact is inconsistent with an investment-first motive and has no alternative innocent explanation I can construct.
The size is immaterial to him. $1m is small against the equity compensation of an S&P-listed apparel CEO. A costly signal is only informative in proportion to the pain it causes the signaller, and this one is calibrated to cause almost none.
Where the consensus retains force: the trade is still real, still irreversible, still unhedged, and still filed under his own name. It is weak evidence, not zero evidence. My claim is about magnitude, not sign.
What would falsify meIf Larsson follows this with further open-market purchases at materially larger size, or purchases made without round-number engineering, then this was an opening tranche rather than a signal and the consensus read is right. One more round-number million would confirm my read; an awkward, unrounded, much larger buy would break it.
PatternDirection break against the tape — open-market insider buys above $1m were rare in this window.
The "insiders are selling 10 to 1" statistic
Widely circulated this month · the measurement itselfSingle-source ratio
Claim
The August 2026 insider sell-to-buy ratio was reported at roughly 10:1, described as the worst month of the year; first-half 2026 saw a record $77.6bn sold against $6.9bn bought, about 11:1.
Consensus read
The people who know their companies best are selling at a near-record pace and barely buying. That asymmetry is a warning about valuations, and the widening ratio through the year tracks the market's ascent.
Contrarian read
The ratio cannot measure what it is used to measure, because its denominator is not a decision. Insiders are paid in stock and must sell to convert compensation into money; there is no corresponding mechanism that pays them cash and requires them to buy. The numerator is compensation mechanics plus sentiment. The denominator is sentiment alone. Dividing the first by the second produces a number that rises whenever share prices rise, regardless of what anyone believes.
Evidence that discriminates
Look at what is actually in the denominator this window. The two largest open-market purchases above $1m in the 10–14 September filings were Larsson's engineered $1,000,001 and a $1,052,415 purchase by Donegal Mutual Insurance in its own affiliate, Donegal Group — a structural intra-group purchase. When a ratio's denominator is composed of one deliberate signal and one related-party transfer, the ratio is not measuring belief.
The numerator is mechanically price-elastic. The same number of vested shares sold at a higher price produces a larger dollar numerator with no change in behaviour. A "record $77.6bn sold" in a year of record index levels is close to arithmetic. This is the fact the consensus read cannot accommodate: it treats a price-scaled quantity as a sentiment measure.
The better-constructed version of this indicator exists and says something different. Practitioners who use insider data seriously count discretionary open-market purchases by number of distinct insiders, not net dollars, precisely because dollars are contaminated by compensation and price. Any ratio quoted in dollars, with no separation of 10b5-1 from discretionary, should be treated as a headline rather than a measurement.
Honest limit: I have the September ratio quoted as improving to 3.6:1 from a single source and could not independently verify either month's figure. My argument is about what the statistic can support, which holds regardless of the exact number.
What would falsify meIf the ratio were computed on discretionary non-plan purchases by distinct insider count — stripping out vesting-driven sales and 10b5-1 flow — and still showed an extreme reading against its own history, the signal would be real and my objection would be about presentation rather than substance. That version of the number would change my view.
Sergey Brin
Google co-founderConfirmed
Did
Put $102m of personal money into Building a Better California, the committee he co-founded with Eric Schmidt to defeat Proposition 40 — a one-time 5% levy on the net worth of Californian billionaires, on the ballot in November. Committee total raised: roughly $118m. Bloomberg's estimate of his exposure under the measure: about $13bn.
DocumentCommittee disclosures, reported 20 Aug 2026. The measure applies to anyone who was a California resident as of 1 January 2026.
Consensus read
A billionaire spending nine figures to kill a tax on billionaires. Read either as plutocratic self-defence or, more sympathetically, as rational insurance: $102m against a $13bn liability is a 0.8% premium, which is cheap for a chance at cancelling the whole exposure.
Contrarian read
Both framings miss what the spending reveals. This is an admission that the exit option has stopped working. For three decades the credible threat "tax us and we leave" was enough that nobody had to spend anything — mobility was the hedge, and it was free. Nine figures of disclosed political money is what it costs to hedge a tax once mobility no longer prices it. The transferable lesson runs opposite to almost all wealth-mobility content: domicile arbitrage is a depreciating asset, and strategies built on it are worth less than they are sold for.
Evidence that discriminates
The 1 January 2026 lookback. This is the fact that separates the two readings. Proposition 40 taxes past residency, so leaving before the vote does not avoid it and leaving after does not either. A measure designed this way is explicitly built to defeat mobility — and it was drafted that way because drafters learned that the mobility threat is what killed earlier attempts. The design is a response to the hedge, which is evidence the hedge had been working and is now being closed.
Restructuring is closed too. A levy assessed on net worth rather than on income or realisation cannot be deferred by not selling, cannot be timed into a better year, and cannot be routed through an entity. When both exit and restructuring are unavailable, political spending is not one option among several — it is the only remaining lever, which is why it got used at a size nobody used before.
He had not done this before. Brin has not historically been a large disclosed political donor. The behaviour is new, and it appeared at the exact moment a measure was drafted to be immune to the tools he already had.
Counter-argument I take seriously: the simplest explanation is that $13bn is just a very large number and any premium is worth paying, with no deeper meaning. But that explanation does not account for why this measure, and not the many earlier wealth-tax proposals, was the one that drew nine figures. The lookback provision is the difference.
What would falsify meIf Proposition 40 fails and subsequent wealth-tax proposals revert to prospective, exit-avoidable designs, then the lookback was an aberration rather than a learned technique, and mobility remains a live hedge. Watch whether the next state or country to attempt this copies the residency-lookback mechanism. If it does, my read strengthens considerably.
The vehicles — Hagerty Holding Corp · Control Empresarial · Silver Lake · 3rd Millennium Trust
Four of the five largest disposals, 9–11 Sep 2026Confirmed · Form 4
Did
Hagerty Holding Corp sold 10,637,500 shares of Hagerty Inc at $11.47 on 9 Sep — $122.0m, the largest disposal of the window. Control Empresarial sold $24.8m of PBF Energy on 9 Sep. Silver Lake partnerships sold roughly $70.3m of Dell on 9 Sep. Stevens sold through 3rd Millennium Trust.
Consensus read
Trusts and holding companies are how the very rich avoid tax. "Buy, borrow, die" — never sell, borrow against the position, let the basis step up at death, pay nothing. The prevalence of entity sellers in the filings is the machinery of avoidance, visible.
Contrarian read
These filings are direct evidence against the story most often told about them. Every transaction listed above is a realisation. Stevens' trust crystallised roughly $646m of gain on near-zero-basis 2006 stock. Hagerty Holding took $122m off the table. Nobody borrowed; everybody sold, and everybody will pay. The structures were not doing tax avoidance this week. Their actual function is control, continuity and centralised timing — which is a less exciting claim and a far more useful one, because it is the part that scales down.
Evidence that discriminates
The 2006 acquisition date combined with the sale. This is the single cleanest test available. If "buy, borrow, die" were operating, a twenty-year-old, essentially zero-basis position is the last asset you would ever sell, because it is the one where the step-up at death is worth most and where borrowing is cheapest relative to the embedded gain. Stevens sold it anyway, at size, in public. That choice is inconsistent with the avoidance model and consistent with a structure whose job is to execute a decision cleanly rather than to defer one forever.
The pattern is entity-as-seller, not entity-as-borrower. Four of the five largest disposals were made by an entity while the individual remained a trustee, officer or partner. What the entity supplied in each case was a single locus for the timing decision and continuity beyond the individual — not leverage.
Scope limit, stated plainly: borrowing against assets is real and is happening at scale elsewhere. My claim is narrower and better evidenced: it is not what happened in these filings, and the popular story is applied far more broadly than the behaviour supports.
What would falsify meAn 8-K or proxy disclosure showing that any of these individuals simultaneously pledged shares or drew a securities-backed facility against the retained portion of the same position. That would show the sale was one leg of a borrow-and-hold structure rather than a straightforward realisation, and would restore the consensus read.
Where consensus is right
Tested, not overturned
Two items where I looked for a contrarian angle and could not make one stand up. Both are more useful for being ordinary.
Andrew Senn selling at $126.30 into a $127.00 takeout
On 8 September the President of Growth & Innovation at Integer Holdings sold 9,408 shares at $126.30, against KKR's agreed $127.00 cash price announced 3 August. The obvious reading is that he took certainty and gave up 70 cents to stop carrying deal risk.
I tried two contrarian angles and both failed. The first — that an insider selling below the deal price signals doubt about the close — does not survive the size: 9,408 shares is not a position anyone unwinds on a view about antitrust. The second — that the spread was mispriced and he left value behind — inverts the actual economics. The last 0.55% of a merger spread is the market's price for regulatory risk, financing risk and time, and an insider is the worst-placed person to hold it, because his job, options and shares all settle on the same event. Paying 0.55% to decorrelate is correct, not impatient. Consensus wins, and the mechanic is the most directly transferable thing in this edition.
Matthew Rabinowitz trimming Natera
The founder sold $16.4m on 11 September. The tempting contrarian line is that it is bearish because Natera is simultaneously the anchor of Duquesne Family Office's book at $864.9m, 16.6% of $5.21bn — founder out, smart money in, someone is wrong.
That line does not survive scrutiny, because the two parties are not answering the same question. Druckenmiller's vehicle chose its position size as an expression of conviction. Rabinowitz never chose his; it accumulated because he founded the company. His marginal decision is about diversifying a concentration he did not select. Both can be rational simultaneously, and reading either as an opinion about price is the error. The ordinary interpretation — routine founder diversification, no signal — is the correct one unless it repeats at increasing size.
Transferable to you
Including what does not
Transfers
Put the structure in place while the asset is still cheap — the clock is the point
Medium effort · high impact
If the entity's real job is control, continuity and centralised timing rather than tax magic, then the useful question is not "how do I avoid tax like they do" but "is my structure in place before the value arrives." The billionaire version is a trust holding shares acquired in 2006. Your version is a holding company between you and your operating company, and potentially a family trust, arranged while your shares are still worth comparatively little.
The deadline matters more than the structure. Canada's lifetime capital gains exemption on qualified small business corporation shares is lost on technicalities: an asset test at the moment of sale, a 24-month holding test, and an asset test across those preceding 24 months. Excess cash and passive assets in the operating company can disqualify shares that would otherwise qualify, and purification takes two years you will not have once a buyer is at the table. A family trust capable of multiplying the exemption has to exist years before a sale, not months.
This weekOne email to your accountant: "As of today, would my shares meet the QSBC tests for the LCGE — and if not, what specifically disqualifies them?" In writing. The value is entirely in getting the answer while a 24-month clock is still cheap.
Transfers
You still control basis and location — they only control timing
Low effort · high impact
Stevens and Rabinowitz have run out of levers. You have not. Every share you acquire from here has a basis and a location you get to choose now, and for a Canadian resident the best location is registered: inside a TFSA, growth and rebalancing never create a disposition at all. Read alongside the point above — your company equity is where the timing decision will eventually be worth most and where your freedom shrinks each year.
This weekOpen CRA My Account and write down two numbers: your exact TFSA contribution room and your RRSP deduction limit today. Carry-forward makes these personal to you, not the annual limit.
Transfers
Pay the premium to decorrelate
Low effort · medium impact
Senn's trade needs no scale at all. When an exposure is correlated with the rest of your life, pay a known cost to reduce it rather than hold out for the last few percent. Your correlated exposure is unavoidable: income, company equity and professional network all depend on the same thing. If a liquidity event ever offers a certain price against a contingent one, the right question is what correlated risk the premium retires — not how much you are leaving behind.
This weekWork out what share of your total net worth — company equity included, at a sober valuation — depends on software demand and on CAD. If it is above 80%, that number is the finding.
Does not
"Buy, borrow, die"
Not what the evidence shows
Twice over. It needs a liquid public security a lender will lend against cheaply, which private shares in an Ontario CCPC are not — a lender will look at your personal covenant and your real property, not your cap table. And on this week's evidence it is overstated even for the people it is supposedly about: they sold, at size, including the twenty-year-old zero-basis stock that the theory says should never be sold. Treat the idea as a real technique used narrowly, not as the organising principle of how wealth is kept.
Does not
Spending to change the rule that taxes you
Inverts below scale
Brin's 0.8% premium only works because the exposure is $13bn and the target is a single ballot measure. Below a certain scale the arithmetic inverts completely: the cost of influencing a rule is fixed and large while your exposure is not. Included as a boundary marker, because this is the category of billionaire behaviour most often generalised into advice that cannot work.
The part that does transfer is the diagnosis, not the tactic. If domicile arbitrage is depreciating — and the 1 January 2026 lookback is evidence that drafters have learned to defeat it — then any plan of yours whose payoff depends on being able to move later is worth less than it appears. Weight structures that work where you already are.
Does not
Buying your own stock as a signal
No mechanism
Larsson's purchase works because there is a public market, a filing requirement, and an audience that reprices on disclosure. A private company has none of these. The residue that does transfer is the principle underneath my critique of it: a costly, irreversible, inconvenient commitment is credible in proportion to its inconvenience. That applies to how you signal to employees, customers and partners — and it argues against gestures engineered to look good at minimum cost, because sophisticated observers discount exactly those.
This is information, not financial, tax or legal advice, and nothing here is a recommendation to buy or sell any security. The analysis above is argued interpretation, explicitly separated from what the filings state, and is offered to be disagreed with. The Canadian structures mentioned — TFSA and RRSP room, the lifetime capital gains exemption and QSBC qualification, holdcos and family trusts — are technical, fact-specific, and change with legislation and CRA administrative position. Confirm anything tax- or structure-related with a licensed Canadian accountant, and involve a tax lawyer before creating a trust or undertaking a corporate reorganisation.