Every second market commentary carries the same sentence: nobody picks stocks anymore, everybody buys ETFs, and once a company sits inside an index it gets a permanent mechanical buyer — whatever the balance sheet says. In Bitcoin the sequel goes like this: since the spot ETFs launched, institutional inflows drive the price.

The question nobody asks: what would the world look like if that were not true?

It can be checked in two places. The first sits in equities and has been documented for decades. The second we computed ourselves, pre-registered, and in both directions — because measuring only one of them is exactly why this subject looks the way it does.

The floor: what index inclusion delivers

Before a claim can prove anything, it needs the one moment where it would have to show. If index membership brings a mechanical buyer, then the day of inclusion is worth something, and it has to be worth more as index money grows, not less. This is not a technicality. It is the most direct prediction the claim makes, and it is measurable to the day.

Robin Greenwood and Marco Sammon measured it across forty years of S&P 500 additions and deletions.

And the demand shock grew over that period rather than shrinking. Net buying by index trackers around an addition was close to zero percent of shares outstanding in the early 1990s. Today it runs above six percent.

Six percent of the float changes hands in a single session and the price does not move. That is the base rate for everything told about the mechanical buyer: at its most visible moment, it has vanished.

The explanation is unglamorous, which is why it convinces. The other side of the trade did not disappear, it turned professional. Trading desks moved staff and computing power onto index rebalancing, large passive managers run their own liquidity teams for reconstitution days, and volume now concentrates on exactly those dates. On top of that, a growing share of additions are no longer new entrants but promotions out of the S&P MidCap — the buyer already owned the stock.

Passive is bigger than the statistics say

That does not make the starting observation wrong. Alex Chinco and Marco Sammon derived the passive ownership share not from assets under management but from trading volume on reconstitution days: how much money would have to track an index to explain the volume spike observed when a stock switches in or out? Their answer for 2021 — index funds held 16 % of the US stock market, while total passive ownership stood at 33.5 %. The gap is pension funds replicating an index in-house and active managers who effectively hug their benchmark.

What does work: the ongoing flow

Here is where the popular version goes wrong. It conflates two different things: the one-off rebalancing shock at inclusion, and the ongoing inflow into a position already held. The first is gone. The second is well documented.

Hannah Unterberg identifies it through a calendar mechanism rather than a correlation: 401(k) contributions arrive at the start of the month, mechanically, and do not respond to fund performance. Over the first three trading days of the month, low-active-share funds earn 2.6 basis points more when passive flows are elevated — high-active-share funds earn 2.1 basis points less. Active fund flows show no comparable pattern.

The second finding in that work weighs more. Price pressure from active fund flows reverses within a few quarters. Price pressure from passive flows still sits at roughly half its initial magnitude three years out.

So membership does not pay, the stream does. And it pays more at the top than at the bottom: Hao Jiang, Dimitri Vayanos and Lu Zheng show that passive inflows disproportionately lift the largest firms, and that the aggregate rises even when the flows come entirely from investors switching out of active into passive. No new money is required. "For every buyer there is a seller" is true as accounting and irrelevant as economics, as long as the two sides differ in price sensitivity.

Three findings that constrain any strong version

A mechanism that works is not yet a mechanism that rules. Three numbers limit the reach. Vanguard puts index funds at a little over one percent of trading activity and active funds at two percent. In 2022 the S&P 500 fell 25.4 % from its 3 January high to its October low while the passive sector did not shrink across three losing quarters — a bid that permits that is not a floor. And in 2017 the best performer in the index rose 133.7 % while the worst fell 50.3 % at near-identical weights; a buyer that does not discriminate does not produce that spread.

So between "works" and "rules" sits a number. Valentin Haddad, Paul Huebner and Erik Loualiche measure that the remaining active traders offset two thirds of a behavioural shift. The rise of passive over twenty years made demand for individual stocks eleven percent more inelastic. Eleven percent, not one hundred.

Crypto: we measured both directions

Bitcoin sits differently, and the difference is larger than the shared word ETF suggests.

An index fund decouples two things: the decision and the purchase. Whoever buys an MSCI World ETF is deciding about a portfolio, not about Apple. The Apple purchase is a derived obligation of the fund and happens at any price. That decoupling is the inelasticity Gabaix and Koijen measure in the aggregate and Unterberg isolates through payroll deduction. It is not a property of ETFs. It is a property of the separation between whoever decides and whoever buys.

A Bitcoin spot ETF decouples nothing. Whoever buys IBIT is deciding about Bitcoin. Decision and transaction are the same act. The flow therefore inherits the buyer's full price sensitivity — it is not mechanical demand, it is discretionary demand in a wrapper.

On top of that comes what is missing anyway: no additions, no deletions, no cross-section, no weight drift. And no automation. No autopilot routes into a Bitcoin ETF every month because an HR department set it up that way. You see it in the tape: after roughly $2.44 billion of net inflows in April 2026, about $1.26 billion flowed back out across six consecutive trading days.

So we measured not one direction but both. Same method, same window, both pre-registered with a written abort rule and a written prior. First the question everyone asks: do flows move the price? Then the one almost nobody asks: do flows follow the price?

The two tests are structurally the same thing — one day of offset, once forward and once backward. Window from 1 June to 28 August 2026, 62 daily observations, 31 inflow and 31 outflow days.

Forward, nothing holds. Split by sign it briefly looked alive, and it inverted our own written prior: after inflows the next day continued, after outflows nothing happened. We had written down the opposite.

Backward, everything holds. A three percent down day is associated with roughly $190 million of outflow the following session, against a daily flow standard deviation of $250 million. Put in signs, which is how anyone actually watches this:

44 hits out of 62, or 71 %, against a base rate of exactly 50 % because the window holds as many inflow as outflow days.

Three checks that could have rescued the forward result

A result drawn from 62 days can mislead in several ways. We applied the same three checks to both arms — checking asymmetrically would be precisely the error we accuse others of.

First half against second half. Forward, the sign flips: −7.9 in the first half against +44.9 in the second. Backward it holds: +51.2 against +55.1.

Without the outlier days. Excluding the three days that moved more than five percent, the forward result falls from +21.6 to −1.9 basis points. The backward result stays at +57.4 with t=3.47. Without the entire melt-up week of 17–24 August it still reads +54.1 with t=3.85.

Dose response. A real relationship scales with size. Forward, it does not: the largest inflow tercile produces a worse next-day median than the smallest. Backward it runs cleanly — strongly negative prior days carry a median outflow of $155 million, middling ones plus $56 million, strongly positive ones plus $99 million.

The entire forward finding was the melt-up of 19–21 August. The backward finding survives every single check the other one died on, and each month computed separately carries the same sign.

Why this direction is cleanly measurable at all

The arrow of time comes free with the market structure here. Bitcoin trades around the clock, the ETF only between 13:30 and 20:00 UTC, and the daily close falls at midnight UTC. Yesterday's move is therefore finished thirteen and a half hours before the opening bell and cannot possibly have been caused by that session's flow.

Almost nothing in this field hands you a causal arrow for free. Here it does, which is why the unpopular direction is the only one about which anything solid can be said.

That also gives the much-quoted same-day number a second reading. Our own same-day estimate is 67 basis points per $100 million, close to the 53 basis points circulating as a multiplier for this market. That figure has been read as a demand impulse. It reads at least as well as a receipt: the price moves, and the day's net flow records afterwards what that did to the people holding it. We cannot separate the two readings on this data. We can say that of the two lagged tests, only the backward one holds.

What would overturn this finding

The explanation above was formed after the result, not before it. That makes it a prediction rather than a confirmation — and it is testable, because the structure is changing right now.

It would look different if Bitcoin were not bought as its own product but as a component of a mixed vehicle: a broadly diversified portfolio of equities, bonds, gold and a small Bitcoin sleeve. Then exactly the decoupling that carries the equity market appears, and it brings three things that do not exist today.

First, price sensitivity disappears. The buyer decides about the portfolio, not about Bitcoin. The Bitcoin purchase becomes a derived obligation of the vehicle.

Second, the sign of the rebalancing flips. In a portfolio with a target weight, the Bitcoin sleeve rises above target when Bitcoin rises — and the vehicle sells. When Bitcoin falls, it buys. That is not merely non-pro-cyclical, it is counter-cyclical. The opposite of what we measured.

Third, an inclusion date appears. When a model portfolio series adds a Bitcoin sleeve, there is a one-off mechanical buying need with a date on it — the crypto counterpart to index inclusion. And what happens to that is in the first half of this piece: the effect shrinks the more anticipated it becomes.

This is no longer a thought experiment. BlackRock added IBIT to two model portfolios in February 2025 at a one to two percent weight — the Target Allocation variants that permit alternatives, out of a model portfolio universe of roughly $150 billion. In June 2026 the BlackRock Investment Institute formally recommended the same range for multi-asset portfolios.

The larger lever sits in retirement savings. On 30 March 2026 the US Department of Labor proposed a rule creating a safe harbour for fiduciaries who add alternative assets including crypto to 401(k) menus. It follows an executive order of 7 August 2025, after the 2022 caution guidance was withdrawn in May 2025. Lawyers point out that employees will hardly find standalone crypto funds on the menu, but limited sleeves inside target-date funds. That is precisely the route that creates the decoupling.

On magnitude, with a provenance warning: depending on the definition, US 401(k) plans are cited at roughly ten to twelve trillion dollars. One percent of that would be a multiple of what today's spot ETF market moves in a day. The timeline is not short, though — broad availability is considered realistic from 2027 to 2029 at the earliest.

Our result therefore has a shelf life, and it can be named. We are not measuring a property of Bitcoin. We are measuring a property of the vehicle through which Bitcoin currently gets bought. If the vehicle changes, the backward coefficient has to shrink. If it does not, the explanation was wrong.

The strongest counter-argument

Two objections come to mind, and both deserve an answer.

First: 62 observations across three months is nothing. You can show anything with that. The window is narrow, and it is declared — daily resolution is not obtainable further back from our source. But the objection hits both arms equally hard, and the two arms reach opposite conclusions on identical data. A window that is too short generates noise, and noise does not produce findings that survive six subsample checks with the same sign. It produces exactly what the forward arm produced.

Second: perhaps the flow data is attributed to the wrong day. Then your backward result is just the same-day correlation relabelled. Good objection, and we tested it internally rather than merely noting it. If the attribution were shifted, the prior-day return would have to be the strongest arm. It is not: the same-day arm leads with t=7.47 against t=5.07. That does not rule out mixing through settlement paths, but it makes a simple one-day mislabelling unlikely.

What to do with this

The ETF figure that scrolls across the ticker in the evening is mostly a receipt for the day, not its engine. Whoever reads it as a demand impulse and extrapolates to tomorrow is reading the causal arrow backwards — and the test that would support that does not hold on our data.

This explicitly does not mean flows never move the price. Both directions can run at once. What is established is the weaker and cleaner statement: on this window, only the backward direction is detectable.

And it closes the loop with the equity half. There, the mechanical buyer vanished at its most visible moment because the other side turned professional. Here, it was never mechanical to begin with. Both times, the automatic bid is the story people tell, and not the one sitting in the data.

The open question is the one that should be answered before any multiplier is quoted: at which horizon — fixed in advance, not found in hindsight — has an ETF flow ever predicted a price move it was not already behind? And the second one hanging off it: at what share of mixed vehicles in Bitcoin ownership does the relationship we measured turn over?

FAQ

What does "index inclusion effect" mean? The abnormal price move on the day a stock joins an index. It measures what the index funds' mechanical buying need is worth, and it has fallen from 7.4 % in the 1990s to 0.3 % over the past decade.

Why is the same-day relationship worthless? Because within a single day flows respond to prices and prices respond to flows. Without a lag or an instrument, no direction can be read out of a same-day correlation, however large it is.

Does this mean ETF flows never move the Bitcoin price? No. It means that on this window only the reverse direction is detectable. Both directions can run at once; only one of them is established.

Why only 62 observations? Because the data source delivers daily resolution for roughly 90 days only. Longer requests return the same number of rows across a wider span, so roughly 15-day sums rather than daily flows. The window is a property of the data, not a choice.

Does this hold for Ethereum and Solana ETFs? Untested. Transferring it without a separate run would be exactly the error this piece describes.

What if Bitcoin becomes part of a mixed ETF? Then presumably it no longer holds, and that is the heart of the matter. A mixed vehicle separates the decision from the purchase and rebalances counter-cyclically. The single-asset ETF does neither. The relationship we measured is a property of the vehicle, not of Bitcoin, and it should shrink as the share of mixed vehicles grows.

How do I know you did not compute until something fit? Both questions were fixed in writing before the first data pull, including the abort threshold, the correction for multiple testing and a written prior. That prior fell — we had written down that outflows should hit harder, and they turned out to be the empty arm.

---

Not investment advice, not a recommendation, not a forecast. Historical patterns are no promise for the future.

Study the Past — Improve your Future 🥋