What Was the Most Likely Cause of This Crash?
Bitcoin fell 21.3% between 4 October and 13 November, where this note's measurements stop, and fell further over the weekend that followed. To that date the record splits the fall in two. For five weeks it fell with the Nasdaq, by about what its usual link to the Nasdaq implies. Then, over four days in which the Nasdaq ended where it began, interest-rate expectations jumped, the largest stock offering of the autumn set its price range, gold fell with Bitcoin, and borrowed positions built up during the decline were closed by force.
Revised 16 November 2026: the “After the cut” box now covers the weekend of 14 and 15 November, with the weekend's low, its liquidations and the bitcoin in Binance's disclosed wallets, and §4 adds the reported demand for Anthropic's offering, which confirms that explanation's premise but measures no seller. The window, every measurement, the verdict and the shares given to the phase-two triggers are unchanged.
Every sharp fall in Bitcoin arrives with a dozen confident explanations, and this one has more than most: a record exchange theft, an index provider deleting the largest corporate holder, a chief executive's resignation at the most valuable chip company, an election, eleven straight days of withdrawals from the largest fund, a hot inflation report, a very large stock offering said to be draining cash, an exchange said to be selling its reserves. Each is true as an event. Whether any of them moved the price is a different question, and it can be tested. This note fixes the dates of the decline from the price alone, asks first whether the move is unusual and how much of it is simply the stock market, then lays each proposed cause against what it would have to leave behind in the data: in the hour it happened, in who sold, and in what else fell at the same time. The measurements stop at the close of 13 November; the fall continued past it, and §1 reports what followed. This is an account of what happened to that point. It says nothing about where the price goes next.
What this note answers
- What happened, exactly? The dates, the two phases, and where this decline ranks among Bitcoin's others.
- How much was just the market? The share that Bitcoin's ordinary link to technology stocks accounts for, and the share it does not.
- Who sold? Funds, long-held coins, miners, corporate holders and futures traders, in bitcoin and dollars on one scale, with the lines nobody can measure named.
- What started phase two, and what made it bigger? The sequence, hour by hour.
- Which popular explanations fail? The ten most repeated and twelve further arguments, each tested.
- What is the most likely cause, how confident is that, and what would change it?
Terms used throughout, defined once. Spot ETF: a US-listed fund that holds bitcoin itself; money leaving the fund means the fund sells coins. IBIT is the largest. Beta: how much one asset has typically moved for each 1% move in another; here, Bitcoin's fund against the Nasdaq-100 fund (QQQ), measured over the year before the decline. Residual: what is left of a move after the beta share is taken out. Long-term holder: a coin that has not moved on the blockchain for about 155 days or more; the count falls when such coins are spent. Perpetual future: a futures contract with no expiry, the main way to hold bitcoin with borrowed money. Open interest: the number of futures contracts outstanding; it rises when new positions open and falls when they close. Funding rate: the fee paid every eight hours between futures buyers and sellers; positive means buyers are paying to stay in. Liquidation: an exchange closing a borrowed position by force when its collateral runs out. Basis point (bp): one hundredth of a percentage point. Log points: returns measured so that they add across periods; over small moves, one log point is about 1%.
The case that Bitcoin had its own seller
The strongest case against this note's reading, in the terms its proponents would use.
- The largest seller by far was inside Bitcoin, and it sped up. Long-term holders' supply fell by 317,552 coins in forty days, about $25 billion at each day's price, against $3.2 billion that left the US spot funds. The pace was 4,280 coins a day from the series' peak on 24 July to the price peak and 7,939 a day after it. No stock-market story explains why coins held for five months or more would be spent at nearly twice the rate.
- Bitcoin took the falls and half the bounces. On the twelve days the Nasdaq rose during the window, compounding to +16.1%, Bitcoin's fund gained 8.0%. On the seventeen days it fell, compounding to −22.5%, the fund lost 25.8%. An asset that only followed the market would have come back when the market did. After 5 November the path the Nasdaq implied recovered about five points and Bitcoin's fund fell another eight.
- The buyers of the breakout left. The spot funds took in $3.06 billion in the eleven sessions of the September rally and gave back $3.16 billion over the window. The largest fund had withdrawals on eleven consecutive sessions through 13 November. The withdrawals were from Bitcoin funds, and at the largest fund they continued on every day the Nasdaq rose in November.
- The largest corporate holder was pushed out of the indexes. MSCI published its final rule on 15 October and its deletion list on 11 November, effective at the close on 30 November. Strategy's shares fell 32% over the window against 20% for Bitcoin's fund. The company holds 850,364 coins, pays dividends on preferred shares, and has already sold this year: its third-quarter report shows 5,552 coins sold in the quarter and 6,947 in the nine months to September, about $432 million, to fund those dividends and share repurchases. It is the one holder whose distress would land on the coin itself.
- Miners sold in size. One miner disclosed selling 13,597 coins after 30 September, about $1.1 billion, and said more would follow.
- In phase two it was Bitcoin, not crypto. The other large tokens usually move about 1.3 times as far as Bitcoin. From 9 to 13 November they fell less than Bitcoin did or about the same. Whatever hit the market that week weighed more on Bitcoin than its usual relationship to the other tokens implies.
What this note tests: claims 1, 3 and 5 are measured in §3; claim 2 in §2; claim 4 in §4 and §5, against the hour of each announcement and against every filing the corporate holders made; claim 6 in §4. What it cannot test: who owned the long-held coins that moved, and why; whether any holder sold through private dealing desks, which leave no public record; and what Strategy did in the week of 9 to 15 November, which it had not yet reported when this note was cut.
What happened, exactly
The dates below were set from the price path before any explanation was examined, by three rules. The peak is the highest daily close after the 30 June low: $86,521 on 4 October. Bitcoin first closed at that level on 21 September ($86,505) and spent two weeks beneath it; the decline is measured from the last close at the high. The cut, where the measurements stop, is the last settled daily close when the analysis was run: $68,096 on 13 November. It is not the end of the decline. The break between the two phases is the day on which a two-line fit to the path changes slope most cleanly: 9 November, $75,161. Before it the price lost about 0.3% a day; after it, 2.4% a day. One day stands out from the price alone and sits inside phase one: 5 November, when Bitcoin fell 5.8%.
Exhibit 1 — The decline and the dated record beneath it
How to read it. The black line is Bitcoin's daily closing price in dollars from 14 September to 13 November 2026; the short dashed segment beyond the last pin joins the two settled closes of the weekend after the cut, drawn on a compressed time scale; they enter none of the measurements. The three gold pins mark the peak, the break and the cut; the gray band is phase one and the cream band is phase two, each with its percentage change. The rule beneath the chart is a timeline on the same date axis: each numbered circle sits on the day of a dated event, and the key below names it. Compare where the circles fall with where the line bends. The chart shows that the steepest stretch is the four days of phase two, and that most of the numbered events fall in November.
Is this unusual? Not by Bitcoin's record, at the cut. Since 2018 Bitcoin has fallen 15% or more from its 90-day high twenty-three times, between two and three times a year, and eight of those have come since US spot funds launched in January 2024. The median such decline reached 23.0% before recovering. This one stood at 21.3% at the cut: thirteenth deepest of the twenty-three, fifth of the eight.
Exhibit 2 — Where this decline ranks
How to read it. Each circle is one decline of 15% or more from Bitcoin's highest close of the previous 90 days, placed by how far below that high it reached before the price came back within 3% of it. Filled circles are declines since January 2024, when US spot funds began trading; open circles are 2018 to 2023. The gold triangle is the current decline, measured at the 13 November close. The vertical gold line is the median of all twenty-three. Compare the triangle with the line: this decline sits just short of the middle of the record.
The speed of phase two is rarer than its depth. A four-day fall of 9.4% or worse has occurred in 6.3% of all four-day spans since 2013 and in 1.9% of spans since the funds launched, where it has now happened ten separate times, roughly once every three and a half months. Measured against how much Bitcoin had been moving beforehand (about 40% a year, annualized), the four days were a 2.3 standard-deviation move and the whole decline 1.8. The word in this note's title is the one readers are using. The measured description, to the cut, is a median-sized decline that sped up.
The decline sits inside a larger one. Bitcoin's record close was $124,824 on 6 October 2025. It fell to $58,525 on 30 June 2026, rallied 48% to the October peak, and by the cut had given back 66% of that rally. At the cut it stood 45% below the record and 16% above the June low. The decline did not stop at the cut; what followed is set out in the box below, and none of this note's measurements includes it.
How much was just the market?
Bitcoin has traded with technology stocks for most of the past two years, so the first question for any fall is how much of it is that link and nothing more. The test is simple. Over the year before the window, Bitcoin's largest fund moved about 1.09% for each 1% move in the Nasdaq-100 fund. The link is loose day to day: the Nasdaq explains about a fifth of the fund's daily variation (R² 0.22). Hold that number fixed, apply it to what the Nasdaq did on each day of the window, and compare the result with what Bitcoin did.
Exhibit 3 — Bitcoin against what the Nasdaq implied
How to read it. Both lines start at 100 on the 5 October close, the first session after the weekend peak. The black line is the price of the largest Bitcoin fund (IBIT). The dashed gold line is where that fund would have been had it moved each day by exactly 1.09 times the Nasdaq-100 fund's move, the link measured over the prior year. The vertical dashed line is the break on 9 November, and the cream area is the gap between the two lines from 5 November on. Where the lines run together, the stock market accounts for Bitcoin's move; where they part, something else does. They run together for a month and part in the final week.
In phase one the two lines are hard to tell apart. The Nasdaq-100 fund fell 10.2%, semiconductor shares fell 23%, and the link implied an 11.0% fall in Bitcoin's fund against an actual 12.9%. That is 85% of the move, and the remainder is well inside ordinary daily noise. The broader market barely moved: the S&P 500 fund lost 2.6% over the whole window and bank shares rose. This was a fall in the technology and artificial-intelligence trade, and Bitcoin fell with it. Its daily correlation with the Nasdaq Composite was 0.73 inside the window against 0.46 in the year before.
The largest day makes the point most plainly. On 5 November the Nasdaq Composite fell 5.5%, its worst session since April 2025, after election results and a chipmaker's forecast raised doubts about spending on data centers. Bitcoin's two worst hours of that day were the hour of the US stock-market open and the hour after it. Its fund lost 5.1% on the day, less than the 7.1% the link implied.
Phase two is the opposite. From 9 to 13 November the Nasdaq-100 fund rose 0.1% and Bitcoin's fund fell 8.0%. The stock market explains none of the four-day move. It is not absent from every day of it: the broader Nasdaq Composite lost 0.5% over the four sessions, and on 13 November alone the Nasdaq-100 fund fell 1.2% and semiconductor shares 3.0%, which by the usual link accounts for about 1.3 of the 4.5 points Bitcoin lost that day. The Nasdaq's gains earlier in the week cancel it.
Two cautions belong beside the 85%. First, the link is not a constant. Measured over only the sixty sessions before the peak it was 0.79, not 1.09, and at that value stocks account for about 62% of phase one and 38% of the whole decline, not 53%. Second, the second point of the opposing case holds in full: Bitcoin captured 1.17 times the Nasdaq's losses on down days and 0.52 times its gains on up days. Most of that asymmetry comes from the week of 9 to 13 November, when stocks recovered and Bitcoin did not. Before 5 November the two lines never differed by more than about three and a half points. From 26 October, when the prospectus discussed in §4 became public, to 9 November the fund fell 4.3 points short of the link, all of it from 6 November on.
In log points, which add up: the Bitcoin fund lost 22.1 over the window. The Nasdaq link accounts for 11.6 (53%). Phase one's remainder is 2.1 (9%), too small to distinguish from noise. Phase two's 8.4 (38%) is entirely unexplained by stocks, and is the part the rest of this note is about.
Who sold?
Someone sold every coin that changed hands, and several of the sellers can be measured. The ledger below puts each on one scale over the same forty days.
Exhibit 4 — The seller ledger, 4 October to 13 November
How to read it. Each bar is the net change, in bitcoin, in what one group of holders held over the window: bars to the left of the center line are net sales, bars to the right are net purchases, and the label gives the coins and, where a dollar figure exists, the value at each day's price. Black bars are measured sales and gold bars measured purchases; the dashed outline is a transfer that one outlet reported and that is not established as a sale. Bar lengths share one scale, so they can be compared directly. The rows overlap: a fund's or a miner's coins that were more than about 155 days old also count in the long-term holder row, so the bars must not be added together. The two lines at the foot name what cannot be measured. The chart shows one row far longer than every other seller, and its mirror on the buying side.
Long-term holders. The count of coins unmoved for about 155 days or more fell by 317,552 over the window, about $25 billion: 274,364 in phase one and 43,188 in the four days of phase two. The coins went somewhere: the count held by short-term holders rose by 335,680.
Two facts limit what this proves about cause. The series peaked on 24 July and fell by 308,193 coins between then and 4 October while the price rose from the mid-$60,000s to $86,521 (625,745 coins since July in all), so selling of this kind and size ran straight through a rally without stopping it. And its timing inside the window is mixed. The pace was 8,607 a day from the peak to 4 November, 1,510 a day over the five days before the break, then 10,797 a day in phase two, the fastest stretch of the window. Its two large falls in phase two, 26,767 on 10 November and 34,857 on 13 November, came on the leg's first and last days, both days the price fell; but it rose by 504 on 11 November and by 17,932 on 12 November, a day the price fell 3.4%.
The figure also measures a category, not a holder. Any wallet whose coins had sat for about five months counts, whether it belongs to an early holder, a fund's custodian, a miner or an exchange. It is not a count of sales, and it errs in both directions: a coin moved between one owner's wallets counts as spent, which overstates selling, and coins aging past the threshold over the same forty days offset spending, which understates the old coins that moved.
The named groups account for little of it. If every coin that left the spot funds (42,110), every coin miners disclosed selling (15,745), the miner-wallet transfers above their usual pace in phase two (about 7,100) and a transfer of 9,261 coins from wallets linked to the US government to Coinbase Prime on 6 and 7 October, reported by one outlet and not established as a sale, had been old enough to count, that is about 74,000 coins, under a quarter of the row. About 243,000 coins, roughly three-quarters, belong to none of the groups this note can measure; funds listed outside the United States, exchange wallets and treasury companies that do not file in the United States are among those it cannot.
Two further readings add detail without names. The profit ratio on long-held coins that moved was below 1, meaning sold for less than they cost, on 8 of the 35 days to 7 November, where the vendor's series ends; over the whole period they were sold at a profit (the ratio averaged 1.14). And the wallet-size bands show no exit by large wallets: wallets holding 1,000 to 10,000 coins ended the window about 20,000 coins higher, and those holding 100 to 1,000 ended it little changed after falling about 47,000 from a high on 2 November, with the caution that exchange and fund custody wallets sit in those bands.
Spot funds. The eleven US funds had net withdrawals of $3.16 billion, about 42,110 coins, on 18 of 30 sessions. The largest fund accounts for about half and had withdrawals on eleven straight sessions from 30 October. In phase two the funds lost $864 million, four sessions of four, 80% of it from the largest fund. The symmetry with September is exact enough to notice: $3.06 billion arrived in the eleven sessions of the breakout. The quarterly reports in which institutions list their holdings (Form 13F) show positions as of 30 September; they say nothing about who sold in November.
How much price can those withdrawals account for? This house's study of fund flows and price (605 trading days) found about 0.30% of same-day price change for each 1,000 coins of net flow. Applied here, the withdrawals line up with as much as 54% of the decline. That figure is a ceiling, and it overlaps the stock-market share; the two do not add. The same study found that flows follow the price by a day more strongly than they lead it, so part of every withdrawal is a response to a fall that had already happened.
Part of the withdrawals was not selling in the ordinary sense. Hedge funds commonly hold the funds' shares against a short position in CME bitcoin futures and earn the gap between the two; when they close the trade, coins leave the funds and futures are bought at the same moment. The regulator's weekly positioning report shows leveraged funds' net short in CME bitcoin futures falling by 22,416 coins between 6 October and 3 November, against 17,511 coins that left the spot funds over the same span. Three cautions apply. The weekly match is looser than the totals: the short fell in weeks of fund inflow too, and not every short is one leg of this trade. The usual reason to close was not present until the end: the gap between CME futures and spot ran at an annualized 4.6% to 5.4% through October against a 3.9% cost of overnight funding, and fell to or below that cost only over 10 to 13 November (2.9% on 13 November). And the reports cover 17,511 of the roughly 30,050 coins that left in phase one and none of phase two; the later reports were not yet released. For the weeks covered, much of the fund withdrawal is consistent with a hedged trade closing, which is less pressure on the price than an outright sale, though not none: the redemption still places coins on the spot market while the offsetting purchase is in futures. That takes some weight off the third point of the opposing case for phase one and leaves it open for phase two.
Miners. Companies that mine bitcoin disclosed 15,745 coins of sales, about $1.3 billion. MARA reported selling 13,597 after 30 September, at about $81,039 each, in filings dated 9 November, and used $750 million to repay loans secured on bitcoin; Hut 8 sold about 2,148 in October to buy back shares. Neither gives dates within those spans. MARA's average price points to October, when Bitcoin traded near that level, and a small part could fall on 1 to 3 October, before the window. All of it is dated on or before 9 November, so none of it belongs to phase two. No miner had filed an October or November production report by the cut, so this row is a floor.
A second source reaches into phase two. Coins leaving wallets identified as miners' ran at about 660 a day through the three months before the peak and through phase one, then 4,325 on 10 November and 4,260 on 13 November, the first and last days of the fast leg and, of its four days, the two on which long-term holder supply fell most. That is about 7,100 coins above the usual pace in four days, around $500 million, against about 12,000 leaving the spot funds. A transfer out of a miner's wallet is not proof of a sale, and on the 68 earlier days in the vendor's four-year record with outflows this far above normal, Bitcoin's moves were ordinary (median −0.05% on the day and +0.2% over the next five).
Corporate holders. The companies that hold bitcoin as a treasury asset bought. Strategy reported purchases in four of the five weeks from 5 October to 8 November, 2,364 coins in all, and wrote of the fifth week that it "did not purchase or sell any bitcoin". Strive bought 13,691 over the same weeks, though its buying fell from 6,182 coins in the week to 23 October to 29 in the week to 6 November. No company of this kind disclosed a sale in the window. Strategy is not a holder that never sells: its quarterly report shows 5,552 coins sold in the third quarter, about $349 million. But its weekly reports give its holdings at each week's end, and from 14 September on those holdings rise by exactly the coins bought, which places every one of those sales before the window. The treasury companies' purchases (16,055) almost exactly offset the miners' disclosed sales.
Futures. Futures traders do not sell coins, but their positions can be measured. On OKX, one of the largest venues, open interest in bitcoin perpetual futures rose from 31,108 coins on 21 October to 37,052 on 12 November, a 19% increase while the price fell 15%. The ratio of accounts positioned for a rise to accounts positioned for a fall went from 0.98 on 5 October to about 2.9 in November, and on Binance it also tripled and on Gate more than doubled. The count of accounts is not the weight of money, though. The funding rate on OKX averaged positive on 39 of the window's 40 days, but it sat below its neutral setting on about three readings in four at OKX, Binance and Gate, and on Binance the perpetual traded below the spot price in nearly every hour. Futures were under net selling pressure while the number of accounts betting on a rise grew. On Binance even the largest accounts' positions leaned long, about two to one, so the data do not show who was on the other side. Positions were being added into the decline on both sides. The same build shows on every venue with a public history: over 9 to 12 November open interest rose on OKX, Binance, Gate and Kraken, 6.6% taken together.
What cannot be measured. Private dealing desks, funds listed outside the United States, coins held on exchanges, and any holder who sold coins younger than 155 days. The marginal seller in any given hour may be in this row.
What started phase two, and what made it bigger?
Phase two is where the stock market stops explaining anything, so it is where the other candidates have to be tested. Four dated events fall inside it.
Exhibit 5 — Phase two, hour by hour
How to read it. Two panels share one time axis, hourly from 9 to 13 November 2026 in UTC. The upper black line is Bitcoin's price. The lower gold line is open interest in bitcoin perpetual futures on OKX, in coins: it rises when traders open positions and falls when positions close. The gray vertical bands are Asia trading hours. The four numbered dashed lines are dated events, named in the key. Read each event line down through both panels: look at whether the price moved in that hour, and whether open interest was rising or falling as the price fell. The chart shows the price falling in steps while open interest climbs, until 13 November, when both fall together.
The inflation report (1). At 13:30 UTC on 10 November the October consumer price index came in at +0.4% for the month and +0.3% excluding food and energy, each a tenth above forecasts. The Federal Reserve had raised its policy rate in September, the first increase since 2023, and held it in October with two officials voting for another increase. Bitcoin fell 0.9% in the hour of the report and 2.0% over four hours. Between 9 and 12 November the two-year Treasury yield rose 30bp to 4.86%, its highest of the window, and the ten-year rose 22bp to 5.10%. On Polymarket, the probability that the Federal Reserve leaves rates unchanged in December fell from 22.5% on 9 November to about 1.5% on 13 November, most of it between the 10th and the 11th, and the probability of an increase of half a point or more rose from about 0.5% to about 10%.
Rates alone would be a thin case; yields rise on many days when Bitcoin does not fall. What makes it more than thin is gold. The largest gold fund fell 6.3% over the same four sessions, while the Nasdaq-100 fund ended the four sessions slightly higher and credit spreads narrowed. Bitcoin and gold pay no income, so both lose appeal when the return on cash and bonds rises. A fall in both at once, with stocks steady, is what a rate shock predicts. It is not what an index deletion, a miner's sales or fund withdrawals predict; none of those reaches gold. Adding gold to the stock-market yardstick lifts the explained share of phase two from nothing to 22%. In the prior year there were fourteen days on which the two-year yield rose 8bp or more; Bitcoin fell on ten of them, with a median move of −1.9%.
The limit of this reading should be stated as plainly. Over the prior year the day-to-day correlation between Bitcoin and changes in yields was about zero. Rates have mattered to Bitcoin on shock days, not as a steady relationship, and one week is one week.
The index deletion (2). MSCI's list became public on the evening of 11 November: The Block reported it by 23:09 UTC and the press release is stamped 23:32 UTC. Bitcoin rose 0.1% in the hour that contains both times and fell 0.2% over four. The following session Strategy's shares fell 6.5% against 4.5% for Bitcoin's fund: the announcement landed on the stock. That is what the mechanism implies. Funds that track the index must sell Strategy's shares by 30 November; nothing in the rule requires anyone to sell bitcoin. The coin would be affected only if the company itself sold, and its filings through 8 November show purchases.
The resignation (3). Nvidia's chief executive resigned before the US open on 12 November, after several days of reports about a board investigation, so part of the news was known earlier. Nvidia fell 13% at the open and closed down 4.3%; the Nasdaq-100 fund lost 0.3% on the day. Bitcoin fell 1.2% in the opening hour and 3.4% on the day. A fall in technology shares can explain a fall in Bitcoin. A day on which technology shares as a group barely fell cannot explain much of one.
The fourth day (4). The sharpest stretch began at about 02:00 UTC on 13 November, late evening in New York, with no dated event in that hour. The price fell 1.0% in the 02:00 hour, 1.8% in the 03:00 hour and 0.6% in the 05:00 hour, and open interest on OKX fell across those hours. By the end of the day open interest was down 9.2% and the price 4.5%. CoinGlass, as reported by The Block, counted $1.47 billion of positions liquidated across crypto in the 24 hours to 05:10 UTC, $1.31 billion of them bets on a rise.
This is the clearest finding in phase two, and it settles a question the popular account gets backwards. A liquidation cascade is often named as what started the fall. For three days the opposite was true: open interest rose 6.7% from 9 to 12 November, up on two of the three days, as the price fell 5.2%. Traders were adding positions into the decline, and on 13 November open interest fell on all four venues with a public history, 5.9% taken together, which removed almost exactly what the three days had added. Those positions were the fuel, and on the fourth day they were closed. Across the whole window, hourly changes in open interest follow price changes; they do not lead them. The build-up came first, and liquidation did not begin the leg, which was already three days old. Whether forced closing started the 13 November slide or followed its first hour cannot be read from hourly data.
The second candidate: cash for a very large stock offering. The explanation that spread fastest in the week of 9 November is that Bitcoin is doing what it did before SpaceX listed in June. In its strongest form it has three parts. Investors who want shares in a heavily subscribed offering sell what is most liquid first, and Bitcoin trades around the clock. Up to a fifth of the deal may go to retail buyers, who overlap heavily with holders of Bitcoin, so the two compete for the same money. And the pattern has happened before: Bitcoin fell about 19% in the fifty days before SpaceX's shares began trading, made its low nineteen days after the offering priced, and then recovered. This note tests the third part (Exhibits 6 and 7) and bears on the first only indirectly. It cannot test the second: no data set identifies subscribers who hold Bitcoin.
The two prospectuses fix the dates and sizes. SpaceX filed publicly on 20 May, printed its price of $135 on 3 June, priced on 11 June and raised $86.25 billion, the largest offering on record. Anthropic filed publicly on 26 October, printed a range of $70 to $78 on 10 November and expects to price on 18 November and trade on the 19th under the ticker ANT. It is selling up to $64.35 billion of shares, or $74.0 billion if the underwriters' option is used, with a further $26 billion placed privately with suppliers. The company expects retail allocations of up to 20%, about $12.9 billion at the top of the range, with final amounts set by demand, and its offering site tells applicants to be ready to pay for every share they request. Each company printed its price eight days before pricing.
Exhibit 6 — Bitcoin around two offerings, aligned on pricing day
How to read it. The horizontal axis counts calendar days from each offering's pricing day, which is zero: 11 June for SpaceX, and 18 November, as expected, for Anthropic. Each line is Bitcoin's daily close, set to 100 fifty days before that pricing day. The gray line is the SpaceX episode, which runs on for 45 days after pricing; the black line is the Anthropic run-up, ending at the gold dot on 13 November, where this note's measurements stop, with a dashed step to the latest settled close, on 15 November. The two dashed vertical lines mark when each prospectus became public (23 days before pricing in both cases, to within a day) and when each price was printed (8 days before). Compare the two lines up to the gold dot, then follow the gray line on. The chart shows two paths that fall together from the day the prospectus became public, and a SpaceX path that kept falling for nineteen days after pricing.
Anchored on those filing dates, Bitcoin's price fell from the day each prospectus became public. Its shortfall against the stock market is less tidy. Before SpaceX filed, Bitcoin was already 14.0 log points short of a rising stock market. Before Anthropic filed it was slightly ahead, and the shortfall opened between the 5 and 9 November closes, about two weeks after the filing and just before the range was printed (§2).
Exhibit 7 — The two offerings, stage by stage
| SpaceX (priced 11 June) | Anthropic (pricing expected 18 November) | |
|---|---|---|
| The 27 days before the public filing: Bitcoin; Nasdaq-100 fund | −1.9%; +7.1% | −4.2%; −4.4% |
| Same 27 days: Bitcoin fund's shortfall against the Nasdaq link | −14.0 log points | +1.2 log points (ahead) |
| Public filing to pricing: Bitcoin | −17.4% in 23 days | −14.9% in 19 days, to the cut |
| Same span: Nasdaq-100 fund; gold fund | +2.2%; −6.1% | −3.5%; −3.0% |
| Same span: two-year and ten-year Treasury yields | −8bp; −22bp | +17bp; +10bp |
| Same span: Bitcoin fund's shortfall against the Nasdaq link | −22.4 log points | −13.5 log points |
| Same span: US spot funds, net flow | −55,221 BTC | −35,330 BTC |
| Same span: stablecoin supply | −$7.3 billion | −$12.7 billion |
| Same span: long-term holder supply | +224,843 BTC | −120,897 BTC |
| Price printed to pricing: Bitcoin | −4.6% in 9 days | −9.4% in 4 days, to the cut |
| After pricing | −7.8% over 19 days to the low; then +9.2% in twenty sessions while the Nasdaq-100 fund fell 10.1% | not yet priced |
How to read it. Two columns, one per offering, and one row per stage, with the same measure applied to both. The stages are fixed by filing dates: the public prospectus (20 May and 26 October), the first printed price (3 June and 10 November) and pricing (11 June; 18 November expected, with Anthropic measured to the 13 November cut). The shortfall is how far the largest Bitcoin fund fell short of what its link to the Nasdaq implied over the span, in log points, using the link estimated on the 150 sessions before each run-up (1.63 for SpaceX, 0.88 for Anthropic). Compare across each row. The table shows a small fall in Bitcoin's price before either filing, though before SpaceX it was already well short of a rising stock market; a fall of about a sixth after the filing in both cases that the stock market does not account for; yields that fell in one case and rose in the other; and long-term holders with opposite signs.
What supports the explanation:
- The price fell from a dated event in both cases. In the 27 days before each prospectus became public Bitcoin fell 1.9% and 4.2%, the second in line with the Nasdaq. From the filing to pricing it fell 17.4% and 14.9%, with the Nasdaq-100 fund up 2.2% and down 3.5%.
- The same things were sold. Spot funds lost 55,221 and 35,330 coins over those spans, stablecoin supply fell $7.3 billion and $12.7 billion, and gold fell in both. Over the last ten sessions of each run-up, 13 of 16 assets compared moved in the same direction.
- Rates do not explain the fall after SpaceX filed. From SpaceX's filing to its pricing the two-year yield fell 8bp and the ten-year 22bp while Bitcoin fell 17.4%. Bitcoin then made its low nineteen days after pricing and rose 9.2% over twenty sessions while the Nasdaq-100 fund fell 10.1% and the ten-year real yield rose another 21bp.
- The cut fell inside the window in which buyers must find the money. Anthropic printed its range on 10 November, the first day of phase two.
- The book was heavily oversubscribed before the cut. Bloomberg reported on 11 November that the book was several times oversubscribed, and Reuters on 13 November, citing people familiar with the deal, that orders had reached about six times the shares on offer by 12 November, from about twice by the end of 10 November; and CNBC and Reuters reported institutional demand running ahead of SpaceX's at the same stage. Anthropic's filings through 15 November keep the range at $70 to $78 and the offering at 825 million shares. This confirms the premise of the explanation, a deal heavily subscribed while Bitcoin fell, but not its mechanism: institutional orders in a book this oversubscribed can be inflated in expectation of cutbacks and need not be funded before allocation, so the multiple bounds the cash that might have been set aside without measuring it.
What limits it:
- Nobody has measured subscribers selling anything. No brokerage, exchange or data provider has published a figure for either deal. The one published test of the crypto leg, by CoinDesk on 6 June, found stablecoin and exchange flows inside their normal range. Retail money-market fund assets slipped about $17 billion, or 0.5%, around SpaceX's pricing and have been flat across this decline.
- Buyers may have borrowed. Margin debt at brokerages rose $86.5 billion in June to a record and fell $84.8 billion in July. That is consistent with the SpaceX deal being financed on credit, which needs no sale of Bitcoin, though the rise began in April and the series cannot be tied to the deal.
- The selling did not stop at pricing. A simple raise-cash-then-stop account puts the low at pricing. After SpaceX, Bitcoin fell another 7.8% and the spot funds lost a further 39,386 coins before the low.
- The two run-ups differ in what else happened. Over the fifty days the Nasdaq-100 fund rose 10.1% before SpaceX and fell 7.9% before Anthropic; semiconductor shares rose 26.5% and fell 19.8%. Long-term holder supply rose before SpaceX and fell before Anthropic. Across the sixteen assets, the two fifty-day patterns are close to unrelated (a rank correlation of 0.11).
- The pattern is thin. A fifty-day fall of this size has occurred in about one in ten such windows since the spot funds launched. The two price paths track closely (a correlation of 0.81, matched by about 3% of random pairs of windows), but their day-to-day changes barely correlate (0.18). Across sixteen large US offerings since 2012, Bitcoin fell in the fifty days before seven and rose before nine; the three that raised $20 billion or more, Alibaba in 2014, SpaceX, and SK hynix in July 2026, were each preceded by a fall of 18% to 24%, but the last two share one episode.
- June had other explanations, and this time rates rose. Bitcoin's shortfall against the stock market predates SpaceX's filing, and the weeks that followed held a long run of fund withdrawals and concern about Strategy. From Anthropic's filing to the cut the two-year yield rose 17bp, and 30bp over the four days from the printing of the range, so in this episode the two leading explanations point the same way.
On this evidence the offering explanation stands level with the rate shock. It has what the rate shock lacks, a second episode in which Bitcoin fell by a similar amount over the same stage of an offering, that time with yields falling, and it lacks what the rate shock has, a price move in the hour of its event. Neither has a measured seller. The reported demand for Anthropic's offering confirms its premise and measures no seller, so it does not move this. The two are not exclusive: a market short of buyers because cash is committed elsewhere falls further on a rate shock. They make different predictions from here, which §6 sets out.
Three readings bear on who was selling in those four days. More than half of phase two (54%) occurred in Asia trading hours, and 12% in the hours the US stock market is open, which is when the funds trade. Coinbase, the main US venue, traded at a smaller discount to offshore venues than in phase one. And the other large tokens fell less than their usual multiple of Bitcoin's move, the reverse of phase one. The leg was heavier in Bitcoin than in crypto generally, and heavier outside US hours than inside them. That fits a shock to Bitcoin as a rate-sensitive asset and a forced unwind in offshore futures; it does not fit a US fund-led sale, and it leaves the spot seller in Asia hours unidentified. The same facts limit the rate reading: 23% of the leg printed in the wider US session (13:00 to 21:00 UTC, which includes the hour of the inflation report and most Treasury trading; the 12% counts only stock-market hours), and a rate shock does not by itself explain why other tokens, which also pay no income, fell less than Bitcoin.
Asia hours do not point to an Asian cause. Over the four days the yen weakened 1.0% and the Nikkei rose 4.0%, the opposite of the August 2024 unwind of yen-funded positions, and the premium Korean buyers pay over US prices rose to 1.5% from about 1.0% in phase one, so there was no sign of regional selling (Exhibit 8). On 13 November Bitcoin was already down 0.7% at 02:00 UTC while Japanese equity futures were still up and US ones flat, and the equities that followed fell less.
Which popular explanations fail?
A search of the published commentary found forty-five distinct explanations in forty-three queries. The ten that recur most are numbered below, in order of how often they appeared. Twelve further arguments and data sets follow them.
Exhibit 8 — The explanation scorecard
| The explanation | Finding | What the record shows |
|---|---|---|
| 1. A hot inflation report revived fear of rate increases | HOLDS · PHASE TWO | Fits the hour, the bond market and gold. Fails for the first five weeks, when the ten-year yield fell 40bp. Bitcoin's standing link to yields is weak and no rate-driven seller is measured. Set level with the stock-offering row below (§6) |
| 2. The technology sell-off, sharpened by the Nvidia resignation | HOLDS · PHASE ONE | 85% of phase one. Fails for phase two taken together: the Nasdaq-100 fund rose 0.1%, though its 1.2% fall on 13 November accounts for about 1.3 points of that day |
| 3. A liquidation cascade | PARTLY | It enlarged 5 and 13 November. It did not start either: open interest rose as the price fell |
| 4. Withdrawals from the spot funds | PARTLY | A measured seller, about 42,110 coins. Flows follow the price by a day; most of phase two fell outside the funds' trading hours |
| 5. A stronger dollar and higher Treasury yields | HOLDS · PHASE TWO | For yields, the same finding as 1 and the same limit: over the first five weeks they fell. The dollar index rose 1.4% in phase two to its highest since April 2025, after slipping 0.2% over the first five weeks |
| 6. Shrinking stablecoin supply | PARTLY | Real and large: supply fell $17.1 billion (5.6%) over the window, at about $0.2 billion a day until 28 October, $0.7 billion a day to 9 November and $0.9 billion a day after; its 30-day fall reached $16.2 billion, the most since June 2022. Tokens frozen after the exchange theft were reported not burned as of 5 November; on the data checked, none of the fall traces to them. It shows cash leaving, not why |
| 7. Recent buyers selling at a loss | A DESCRIPTION, NOT A CAUSE | Buyers are at a loss because the price fell. This names who sold after the fall, not what caused it |
| 8. MSCI's deletion of Strategy | NOT SUPPORTED | No move in Bitcoin at the release; the shares took it; no sale in any filing inside the window (the company's sales this year came earlier, in the second and third quarters). The "$2.8 billion of forced selling" in circulation is a 2025 estimate about the shares |
| 9. The election result and the stalled market-structure bill | NOT SUPPORTED | Polymarket's contract on the CLARITY Act being signed in 2026 stood at 3.2% the day before the vote, and Bitcoin rose 3.6% the day after. The election reached Bitcoin through technology shares on 5 November |
| 10. Selling by miners and large holders | PARTLY | One miner's 13,597 coins are real and were sold on or before 9 November. Miner wallets then moved 4,325 and 4,260 coins on 10 and 13 November against about 660 a day before; whether those were sales is not established. Claims about "whales" rest on wallet-size bands, which include exchange and fund custody wallets |
| Also tested | ||
| Cash raised for a very large stock offering, as before SpaceX in June | PARTLY | Bitcoin fell about a sixth between each public prospectus and pricing (Anthropic's measured to the 13 November cut), with the Nasdaq-100 fund up 2.2% and down 3.5%, in June with yields falling, and the same things were sold in both run-ups (Exhibits 6 and 7). No one has measured a subscriber selling, June's rise in margin debt began in April and cannot be tied to the deal, Bitcoin's shortfall predates SpaceX's filing, and across sixteen large offerings there is no general pattern. Set level with row 1 (§6) |
| An unwind led from Asia: yen-funded positions, Japanese rates, regional markets, Korean selling | NOT SUPPORTED | Over 9 to 13 November the yen weakened 1.0%, the Nikkei rose 4.0% and Korean shares 5.4%; the Korean premium rose to 1.5%; on 13 November Bitcoin fell before Asian equities did. Japan's 10-year yield rose about 12bp, mostly on a day Bitcoin rose, and its 13 November level was not available; the Hang Seng fell 3.3% |
| Hedge funds closing a trade of fund shares against futures | PARTLY | Their net short in CME futures fell 22,416 coins from 6 October to 3 November, against 17,511 leaving the funds. Reports for the last ten days were not yet released |
| Rule-based selling when trend lines broke | NOT SUPPORTED | After the 14 earlier weekly closes below the 50-week average the next day's median was −2.3% (10 of 14 down), the next five days' +1.8% and the next twenty days' −8.0%; this time +0.3% the next day and −9.2% over five. After the 18 earlier first closes below the 200-day average the next day's median was +0.6%; this time −4.5% |
| Gold fell for a reason of its own | NOT SUPPORTED | Silver fell 8.8% and gold miners 10.7% with it; a search of wire reports found no margin increase or official sale; the largest gold fund's holdings were flat. Gold's fall is in line with the dollar and real yields on a weekly fit that explains about a quarter of gold's moves |
| Year-end redemptions and fund tax-year selling | NOT SUPPORTED | October redemption notices at hedge funds were reported at 1.62% of assets against a five-year average of 2.06%; mid-November figures are not yet released |
| The leverage pattern beyond one venue | CONFIRMED | Open interest rose on all four venues with a public history over 9 to 12 November (+6.6% together) and fell on all four on 13 November (−5.9%) |
| Hedging through options ahead of the fall | NOT SUPPORTED | Put volume on Deribit was ordinary on 10 to 12 November (9,523 to 19,271 coins a day) and jumped to 57,528 only on 13 November, though more of the put volume than usual sat at strikes of $70,000 and below on 10 and 12 November. Options followed the fall |
| The Binance theft, and a belief that the exchange was insolvent or was selling its bitcoin | NOT SUPPORTED | The theft adds less than a point on its own two days. Binance's price never fell more than 5bp below OKX's, its token matched Bitcoin over the window, and the insolvency contract was back below its pre-theft level by phase two. The bitcoin in its disclosed wallets was 0.6% higher at the cut than at the peak, while $11.6 billion of stablecoins left; 18,270 of the stolen coins were bitcoin, which had to be supplied, not sold. Binance says holdings that include bitcoin funded its buyback of stolen tokens; whether any bitcoin was converted is undisclosed, and a sale on its own order book or from undisclosed wallets would not show. That part cannot be tested |
| An effort to discredit Binance and move business to US venues | NOT SUPPORTED | Binance's share of spot volume fell about ten points from the theft on and Coinbase's rose about six, but US funds lost coins and US prices stayed at a discount |
| A deliberate push to force liquidations | NOT SUPPORTED | Of four marks, two are absent: none of five sharp slides stopped at the densest liquidation level, and spot volume rose faster than futures. One is ambiguous: futures were under net selling pressure, which a short being built and ordinary hedging both produce. One is present for one slide of two: the 8 November recovery held. A push meant only to start a cascade cannot be tested, and intent cannot be read from prices |
| The four-year cycle | CANNOT BE TESTED | A calendar pattern is not a testable cause. What can be tested is whether long-held coins were sold, and §3 shows they were, starting in July and through a rally |
How to read it. Twenty-two rows. The first ten are the explanations that recurred most across published commentary between 29 October and 14 November, numbered in order of how often they appeared; the twelve below the divider are further arguments and data sets tested against the same evidence. The middle column is this note's finding: gold-outlined labels hold, or are confirmed, for the part of the decline named; gray labels hold in part or describe an effect rather than a cause; pale labels are not supported or cannot be tested. The right column gives the evidence that decides the row. Compare the labels down the column: three explanations hold, each for one phase only, one pattern is confirmed, six hold in part, ten are not supported, and none holds for the whole decline. Long-term holder selling, the largest measured seller, is measured in §3 and has no row here.
The pattern in the errors is consistent. Most of the failed explanations take an event that shared a day with the fall and treat it as the reason. Several reverse cause and effect: withdrawals, liquidations and losses among recent buyers are all things a falling price produces. And the most-shared one, the index deletion, treats a sale of the company's shares as a sale of its coins; the stronger version, that the company itself sells, is the first finding in §6 that would change this view. Two of the rows need more than a line.
The Binance theft
The theft of about $6.3 billion began, by the tracing firms' accounts, at about 20:05 UTC on 29 October, inside the second-worst hour of the window (−2.5%), and the largest discount of US prices to offshore prices came over the next three days. But the same two sessions held earnings disappointments at two large technology companies: the Nasdaq-100 fund fell 4.8%, and Bitcoin's fund fell 6.0% against 5.3% implied. Whatever the theft added is less than a point.
A market that believed the exchange insolvent would leave marks that are absent here. Bitcoin on Binance never traded more than 5bp below its price on OKX, and during the halt in withdrawals it traded at a premium of about 4bp. Binance's own token fell more than Bitcoin on the theft days (7.3% against 4.9%), but 20.8% over the window against Bitcoin's 21.1%, and less than Bitcoin in phase two. Polymarket's contract on a major exchange being declared insolvent in 2026, which names Binance among five exchanges, stood near 3% before the theft, rose to 4.45% on 1 November and was at 2.45% through 11 to 13 November.
One mark is present, and it bears on the related idea that the episode served to discredit Binance and move business to US venues. Across Binance, Coinbase and OKX together, Binance's share of spot Bitcoin volume fell from about 58% before the theft to about 47% afterwards and stayed there, and Coinbase's rose from about 22% to about 28%. The shift dates from the theft, not from the selling. Activity moved onshore; coins did not: the US funds lost about 42,110 coins and US prices stayed at a small discount to offshore ones. A fall in Bitcoin's price hits every venue alike, so it is a poor instrument against one of them.
A stronger version holds that Binance is itself the large seller, short of the reserves it claims and selling bitcoin for cash. Five records bear on it.
- What was stolen. By TRM Labs' tracing, 18,270 of the stolen coins were bitcoin, about $1.47 billion of the $6.36 billion, and none of it was frozen. Making customers whole on that part meant supplying bitcoin, not selling it.
- What Binance said. Its co-founder Yi He wrote on 30 October that the company was buying back the stolen tokens with holdings that "include a large amount of assets such as BTC and ETH". Its founder wrote on 2 November that "Binance did not sell assets." The two statements do not sit easily together, and the company has not said how much bitcoin, if any, it converted.
- The emergency fund. Half of the fund's 15,000 bitcoin left its published address on 31 October. Followed on-chain, about 4,800 of the 7,500 coins were consolidated on 3 November into a wallet that now holds about 173,000 bitcoin and that one tracking site labels a Binance cold wallet; the rest went out in separate transactions. Whether any were sold cannot be read from the transfers. The fund has not been refilled.
- The wallets. As tracked by DefiLlama, whose readings are taken at 00:00 UTC, Binance's disclosed wallets held 640,832 bitcoin on 4 October and 644,676 on 13 November, 0.6% more with the theft in between, and the balance rose by 12,920 coins between the 9 and 13 November readings. Most of that rise, 9,246 coins, came in the day before the 12 November reading, the moment of the reserve report's snapshot, and may be stolen bitcoin being replaced. Binance did lag: nine other exchanges tracked the same way (OKX, Bybit, Bitfinex, Bitget, Gate, KuCoin, HTX, Deribit and Robinhood; Coinbase and Kraken are not covered) gained 7.1% together. Between the 1 and 3 November readings, as withdrawals reopened, Binance lost 25,997 coins (4.0%), and on 9 November it was still 13,291 below its 1 November level (2.1%) while the others had gained. Dollars left in larger size: stablecoins in its wallets fell from $53.0 billion to $41.3 billion, against falls of about $1 billion each at OKX and Bybit. Wallet data cannot split that between the theft and customer withdrawals, and cash raised to pay a reimbursement would not stay in the wallets, so the fall in dollars does not rule a sale out. From the 13 November reading to early on 15 November, the last day of the window and the day after it, Binance's bitcoin fell 6,071 coins (0.9%) and its stablecoins rose $0.6 billion while the other nine gained 964 coins. That is the direction the claim predicts, at about 3,000 coins a day, the average size of its daily change since September.
- The reserve report. Binance's own report for 12 November, as carried by crypto.news, shows 688,283 bitcoin held against 687,912 owed to customers, a ratio of 100.05%. The report is the company's own and the margin is thin.
Taken together: Binance has said that its holdings, which include bitcoin, funded the buyback; whether any bitcoin was converted is not disclosed, and nothing located shows a sale at a size that matters beside the 317,552 coins long-term holders shed or the 42,110 that left the funds. The test has limits. Wallet balances combine customers' coins with the company's, and a sale by Binance to its own customers on its own order book, or from a wallet it has not disclosed, moves nothing the trackers can see. That part of the claim cannot be tested.
The argument that the fall was engineered
One explanation circulates after most sharp falls: that someone pushed the price down on purpose to force liquidations and collect the proceeds, the charge usually made under the word "manipulation". In its strongest form it runs as follows. Accounts positioned for a rise on borrowed money had tripled relative to those positioned for a fall, so a large pool of forced sellers sat just below the market. The two sharpest moves with no news behind them, the 21:00 UTC hour on Sunday 8 November and the 02:00 to 05:00 UTC hours on 13 November, both came outside US trading hours. On 8 November the densest cluster of long liquidations sat only 0.5% below the price, and a seller needs only to tip the nearest cluster; the forced selling does the rest. And $1.47 billion of positions were liquidated in the day to 13 November, money that went to whoever held the other side.
Intent cannot be read from prices, so the test is of footprints. A push designed to extract as much cash as possible has a recognizable shape: short positions are built first, so open interest rises while the funding rate turns negative; the selling leads in futures, where a small sum moves the price, and not in spot; the price is carried to the level where the most long positions are liquidated and stops near it; and once the forced selling has been bought, the price recovers and holds. A looser version, a push meant only to tip the nearest cluster and let the selling run, cannot be tested here, because it cannot be told apart from a cascade nobody started.
On those marks the record is mixed, and mostly against.
- The size of the forced sale is as the argument says, and any liquidation produces it.
- The timing is less telling than it sounds. Ranked by typical volume over the three months before, the Sunday 21:00 UTC hour sits near the middle of the week's 168 hours, and the four Friday hours from 02:00 to 05:00 UTC range from its busier third to its quieter third.
- The build-up is ambiguous. Open interest rose over 9 to 12 November, and accounts positioned for a rise outnumbered those positioned for a fall about three to one. But funding, though positive, sat below its neutral setting on most readings, and perpetuals on Binance traded below spot in nearly every hour, so futures were under net selling pressure. That is what a short position being built looks like. It is also what hedging by funds and dealers looks like, and the two cannot be separated here.
- Spot grew faster than futures. Average daily spot volume in phase two was about 80% above its level over the previous five weeks, and the ratio of futures to spot volume fell from 6.2 to 4.8, so coins were being sold.
- The recoveries differ. After the 8 November slide the price recovered 1.8% in six hours and its daily closes stayed above that low for the next three days. After the 13 November liquidation it recovered 2.5% in six hours and fell below that low within a day.
- Where each slide stopped is the sharpest mark. This house keeps a daily log of where estimated long liquidations cluster below the market. The rule was set in advance: a slide counts as aimed if its low lands at the densest cluster then below the price, or within 1% below it.
Exhibit 9 — Where the five sharp slides stopped
| The slide | Densest long-liquidation cluster below the price when it began | Low reached | Low against that cluster | What happened |
|---|---|---|---|---|
| 5 Nov, US open | $77,152 | $75,035 | −2.7% | Ran through it and stopped 0.1% above the next |
| 8 Nov, Sunday 21:00 UTC | $75,260 | $73,477 | −2.4% | Ran through it; the next was not reached |
| 10 Nov, inflation report | $71,555 | $72,643 | +1.5% | Crossed a lesser cluster; never reached it |
| 12 Nov, US open | $70,461 | $70,600 | +0.2% | Stopped just above it |
| 13 Nov, Asia hours | $70,461 | $67,555 | −4.1% | Ran through it and the one below, and 1.9% past the lowest logged |
How to read it. One row for each of the five sharpest slides between 5 and 13 November; the second and fifth are the two with no news behind them. The second column is the price level, among those still below the market when the slide began, at which the largest share of estimated long liquidations sat, taken from the most recent daily log (5 to 19 hours old). The third is the lowest price in the twelve hours after the slide began. The fourth is the gap between them: a slide that carried the price to that level and stopped would show a figure between 0 and −1%; a larger negative figure means the price ran through it, a positive one that it never got there. The table shows no row inside that band.
None of the five meets the rule, including the two with no news behind them. Nor do the clusters draw the price on ordinary days: across the 110 days the log covers, the day's low landed in that band on 8.2% of days, against about 10% for a level at the same distance given only how far the price fell on those days, and on the sixteen days with a fall of 3% or more it did so once, with a median overshoot of 1.7% among the nine that reached the cluster. Two of the five lows stopped just above a cluster; a low within half a percent of some cluster occurs on about a third of all days and on three of the sixteen days with a fall of 3% or more. The 13 November slide went through every logged level below it and kept going.
This does not show that nobody acted with that purpose, and the rule is narrower than the argument. A push meant to start a cascade, not to stop at a level, would look like 8 November and 13 November. A holder selling coins while short on a venue not measured here would leave the spot-led pattern that was seen, and the seller of spot bitcoin in Asia hours on 13 November is unidentified. The log is one reading a day of a vendor's estimate, positions grew about 4.5% on 12 November after that day's reading, and five events are few. What the record shows is that the level mark of a deliberate, cash-maximizing push is not there, that the recovery mark is there for one slide of the two, that the lean in futures is ambiguous, and that the moves can be accounted for without one. The finding would change if a later slide's low landed within 1% below the densest cluster then below the price after funding had turned negative with open interest rising and futures leading spot, or if an exchange or a regulator disclosed trading aimed at liquidations inside the window.
The verdict, and what would change it
Is it unusual? At the cut, no: a median-depth decline of a kind that has happened twenty-three times since 2018, with a four-day fall inside it that has happened ten times since 2024. It has since gone deeper (§1).
The ranking, by role.
| Role | Phase one (4 Oct–9 Nov, −13.1%) | Phase two (9–13 Nov, −9.4%) |
|---|---|---|
| Trigger | The fall in technology and AI shares (about 70%) | The rate repricing after the 10 November inflation report, and cash raised ahead of a very large stock offering, about a third each and not ranked against each other. The remaining third is spread across long-term holder selling, fund withdrawals, transfers by miners and corporate-holder stress, each at roughly 7 to 10%, with all others about 2% together |
| Seller, measured | Long-term holders 274,364 coins; spot funds about 30,050, part of it consistent with a hedged trade closing; miners 15,745 disclosed, dated 1 October to 9 November | Long-term holders 43,188; spot funds about 12,060; miner wallets about 9,750 moved, about 7,100 above the usual pace |
| Amplifier | Liquidations on 5 November (the US open) | Futures positions added over three days and liquidated on 13 November, on all four venues with a public history (about 85%) |
| Not the cause | Rates (yields fell); the election; the market-structure bill; the exchange theft beyond its own two days | The stock market (unchanged over the four sessions); the index deletion; the Nvidia resignation; miners' disclosed sales (on or before 9 November) |
On the evidence against each, three candidates sit close together for phase two. Against the rate shock: Bitcoin's weak standing link to yields; June, when Bitcoin fell after SpaceX filed with yields falling and recovered with them rising; and a leg of which only 23% printed in the wider US session, while other tokens, which also pay no income, fell less than Bitcoin. Against the offering: no measured subscriber selling; a shortfall that predates SpaceX's filing; a rise in margin debt that began in April and may show buyers borrowing; selling that ran nineteen days past SpaceX's pricing; and a leg mostly in Asia hours, when the US funds its match with June rests on do not trade. Against long-term holders: gold's fall, which that reading does not predict. It is largely untested rather than refuted.
The ranking sets the rate shock and the offering level, and both ahead of long-term holders, by judgment. The rate shock alone has a dated event inside the leg with a price move in its hour: the first fall came in the hour of the inflation report, and the repricing of the December decision came between the 10th and the 11th. The offering alone has a second episode: in June and again now, Bitcoin fell about a sixth between the public prospectus and pricing (this time measured to the cut) while the Nasdaq-100 fund rose 2.2% and fell 3.5%. Its price range was printed on the day of the inflation report, so that day's fall cannot be assigned to either, and both predict gold's fall. Long-term holders have timing by the day only: their two largest falls came on the leg's first and last days, but the series rose on 12 November as the price fell and did not lead the price over the window. Among the smaller candidates, miners were added on their wallet transfers on the leg's first and last days, which overlap the long-term holder row, and fund withdrawals were not lowered, because the positioning data that argue for a hedged trade stop on 3 November; the seventh finding below decides it. The percentages are a judgment, not a measurement, and the four smaller candidates are not ranked against one another.
The shares. The stock market accounts for 53% of the decline (38% on the lower estimate of the link), phase one's unexplained remainder for 9%, and the four days of phase two for 38%. Inside that 38% the trigger and the amplifier cannot be separated cleanly: 13 November was the largest of the four days, about 4.6 of Bitcoin's 9.9 log points on daily closes.
What the data cannot settle. Whether long-term holder selling was a cause or simply the supply that met whoever was buying; it is the largest measured group, and about three-quarters of it cannot be attributed to any group this note can measure (§3). Who sold spot bitcoin in Asia hours on 13 November. Whether miners' transfers on 10 and 13 November were sales, and what miners sold in October and November. Whether the fund withdrawals after 3 November were, like many before, a hedged trade closing. What Strategy did in the week of 9 to 15 November. Whether Binance sold bitcoin of its own to customers on its own order book, which wallet balances would not show. Whether any seller acted in order to force liquidations: intent is not observable, and a push meant only to start a cascade cannot be told apart from one nobody started. And whether the rate shock or deal-related cash raising did more of the work in phase two: the range was printed on the day of the inflation report, so the four days cannot separate them.
This house's own record. The Bitcoin Quarterly dated 22 September put a 22% probability on a new record high by year-end against 9.9% in the market, and 70% on $100,000 against 46%. It named a weekly close below the 50-week average as the condition on which that 22% would be withdrawn to the market's range, and a ten-session net outflow from the funds with every close above $80,000 as a signpost pointing the same way. Both were met during this decline: the outflow signpost on 8 October, four days after the peak, and the withdrawal condition on the 8 November weekly close ($74,968 against $75,488). The house's number was the higher of the two, and the condition set against it fired. The year-end contracts are unresolved; from the cut, the $126,200 high those contracts are settled against would require +85% in 48 days. A third signpost from July, a sustained premium on US venues over offshore ones, has not been met on any day since the September note. The June-low condition from July, a daily close below $58,525, has not been met.
What would change this view. Eight findings would move the ranking, and each is observable.
- A sale by a corporate holder. If Strategy's report for 9 to 15 November, or any treasury company's filing, shows bitcoin sold inside the window, the corporate-stress explanation moves from unsupported to the leading candidate for phase two. Fear of a forced seller and an actual sale are different findings; only the second is in the price mechanism.
- Miner reports. If a miner's report or filing shows that the transfers of 10 and 13 November were sales, or if November reports show sales for the month above the usual pace by about the 7,100 coins moved, miners move up for phase two. If not, the transfers stay unestablished.
- The rate link failing in reverse. If yields and rate expectations fall back while gold recovers and Bitcoin's gap to the Nasdaq does not close, the rate reading was a coincidence of one week and a seller specific to Bitcoin remains unidentified.
- The gap widening. If Bitcoin's shortfall against the Nasdaq-implied path, 10.5 log points at the cut, grows beyond 15 without a rate shock, the stock-market reading of phase one was the smaller part of the story.
- Long-held coins leading. Over the forty days the daily change in long-held supply moved with the same day's price (correlation 0.22) and did not lead the next day's (−0.20, the wrong sign for a lead); four days are too few to test the leg alone. If the decline extends past the cut and the daily series leads the price by a day over ten or more sessions, or an identified holder's filing dates sales ahead of the 13 November fall, long-term holder selling moves from a measured seller to the trigger.
- The offering against the rate path. Anthropic's pricing is reported for about 18 November and the Federal Reserve decides on 9 December. If Bitcoin's shortfall against the Nasdaq-implied path narrows in the weeks after the listing while the two-year yield stays at or above its 9 November level, the stock-offering explanation moves ahead of the rate shock. If the shortfall narrows only once yields fall, the rate shock stands and the June resemblance was coincidence.
- The positioning reports. If the regulator's reports for 10 and 17 November show leveraged funds' net short in CME futures still falling by about as many coins as left the funds, the fund withdrawals of the last ten days were a hedged trade closing and fall out of the ranking. If the net short rose, those withdrawals were outright selling and move up.
- Binance's bitcoin. If the bitcoin in Binance's disclosed wallets falls by more than 5%, about 32,000 coins, over any ten days while its stablecoins rise and the other nine exchanges' bitcoin does not fall, or if a reserve report, auditor or regulator shows the company's own holdings down by a comparable amount, the claim that Binance is a large seller moves from not supported to open.
The conditions that mark each cause as ended or still running. None is a price level.
| What to watch | By | If it happens | If it does not |
|---|---|---|---|
| Strategy's weekly report for 9–15 November | 18 Nov 2026 | A sale: the corporate-stress explanation is confirmed for phase two. No sale: it stays unsupported | No report by the date: the company has changed its reporting, itself a finding |
| MSCI's changes take effect at the 30 November close; the filings for that week | 8 Dec 2026 | No bitcoin sale disclosed: the index event passed through the shares alone | A sale disclosed: finding 1 above applies |
| US spot funds: five consecutive sessions of net inflow | 31 Dec 2026 | That seller has stopped | The withdrawals that began 30 October are still running |
| Long-term holder supply: a rise over any 30-day span | 15 Jan 2027 | The largest measured flow has ended | Distribution that began 24 July is still running |
| The Federal Reserve's decision on 9 December, and the two-year yield the next day against 4.56%, its 9 November level | 10 Dec 2026 | Yield back below 4.56%: the rate shock has reversed | A half-point increase or a yield above 4.86%: it has extended |
| OKX perpetual open interest above 37,052 coins with funding positive | 31 Dec 2026 | The amplifier is rebuilt | Leverage stays lower than before the liquidation |
| Bitcoin's shortfall against the Nasdaq-implied path, from 5 October | 31 Dec 2026 | Narrower than 5 log points: phase two was a passing shock. Wider than 15: finding 4 applies | Between 5 and 15: unresolved |
| Miners' October and November production reports and filings | 15 Dec 2026 | The 10 and 13 November transfers shown to be sales, or November sales above the usual pace by about 7,100 coins: finding 2 applies | Ordinary sales: the transfers stay unestablished and miners stay where they are in the ranking |
| Bitcoin's shortfall against the Nasdaq-implied path over the twenty sessions after Anthropic's first trade, read beside the two-year yield | 18 Dec 2026 | Narrower by 5 log points or more with the yield at or above 4.56%: the stock-offering explanation leads. Narrower only with the yield below 4.56%: the rate shock stands | Not narrower: neither is confirmed, and finding 4 is the one to watch. No listing by the date: the test lapses |
| Leveraged funds' net short in CME bitcoin futures in the reports for 10 and 17 November, against coins leaving the spot funds | 24 Nov 2026 | Net short down by about as much as the funds lost: a hedged trade closing. Net short up: outright selling | Reports delayed: unresolved |
Sources and Method
The window. Bitcoin daily closes are the CoinMetrics reference rate (the UTC day's close). The peak, break and end were fixed by rule from that series before any other data was opened, and were not moved; prices after the 13 November close are reported separately as after the cut (the settled daily closes of 14 and 15 November and the weekend's hourly low from Coinbase, with open interest from OKX, implied volatility from Deribit, stablecoin supply and the bitcoin in Binance's disclosed wallets from DefiLlama for those days, the weekend's liquidations from CoinGlass's tally and the Binance founder's statement both as reported by The Block on 15 November) and enter none of the measurements; the two closes are the CoinMetrics reference rate, and long-term holder and miner readings, where shown, are from bitcoin-data.com. Fund and equity prices are Robinhood regular-session closes; the 13 November figures are the session's closing bars, pulled over the weekend before the official closing field had posted.
History. A decline is counted when the close is 15% or more below the highest close of the trailing 90 days, and ends when the close is back within 3% of that trailing high. Frequencies of four-day and forty-day falls use every overlapping span since 2013 and since 11 January 2024.
The market share. The Bitcoin fund (IBIT) regressed on the Nasdaq-100 fund (QQQ) by daily log returns from 4 November 2025 to 2 October 2026 (slope 1.09, R² 0.22); the slope is held fixed inside the window and the intercept dropped. Alternatives reported: the S&P 500 fund, semiconductors, small companies, a three-factor version with banks and gold, and the sixty-session slope at the peak (0.79).
Funds. Daily net flows by fund from Farside Investors, in dollars, converted to coins at each day's close. The price-per-flow figure is from this house's flow study (605 trading days, an aggregate series from bitcoin-data.com, which differs from Farside by more than $50 million on two days of the window).
On-chain. Long-term and short-term holder supply and coin days destroyed from bitcoin-data.com (BGeometrics), one vendor for each series. Spent-output profit ratios, the miner position index and wallet-size bands were released by the vendor late on the build day and are used only where the text says so; the profit ratios end on 7 November.
Corporate holders and miners. Every current report, quarterly report and press exhibit filed on EDGAR between 15 September and 14 November 2026 by twenty bitcoin treasury companies and miners, 196 filings. Strategy's and Strive's purchases are from their weekly current reports, and Strategy's third-quarter sales from its quarterly report (note 3); MARA's sale from its quarterly report (note 18) and shareholder letter of 9 November; Hut 8's from its quarterly report. Metaplanet does not file with the SEC and is not covered.
Futures and venues. OKX public data for perpetual open interest, funding and the long/short account ratio; Coinbase and OKX hourly prices for sessions and the US premium, adjusted for the dollar price of USDT; Deribit for implied volatility. One venue's open interest stands in for the market. Hourly liquidation history is not public; the totals quoted are CoinGlass figures as reported by The Block (13 November) and other outlets.
Rates and expectations. Treasury yields, real yields, credit spreads, reserve balances and the overnight rate from FRED; the bond market was closed on 11 November. December rate probabilities and the market-structure bill's odds from Polymarket daily prices.
Dated events. The Bureau of Labor Statistics release (10 November); the Federal Reserve's statement (28 October); MSCI's November 2026 index review list and the timestamp of its release; Nvidia's current report (12 November); Strategy's current reports; CNBC, CoinDesk, The Block and Yahoo Finance for the 29 October, 5 November and 13 November sessions. The exchange theft's size is as reported by those outlets. The transfer from US-government-linked wallets rests on one outlet and is shown as unestablished.
The two later tests. Binance's Bitcoin price from the exchange's public market-data service against OKX, hourly; its token's daily closes; Polymarket's daily prices for the exchange-insolvency contract; Binance's announcement of 12 November on reimbursement; daily balances of bitcoin and stablecoins in the disclosed wallets of Binance and nine other exchanges from DefiLlama; the composition of the theft from TRM Labs; the emergency fund's transfers from the Bitcoin blockchain, starting at the address Binance published; the executives' statements as quoted by crypto.news, CoinDesk and PANews; the 12 November reserve figures as carried by crypto.news. The two offerings' dates, prices, sizes and retail allocations are from the companies' registration statements and prospectuses on EDGAR (SpaceX: filed 20 May, price printed 3 June, priced 11 June; Anthropic: filed 26 October, amended 2 and 10 November) and Anthropic's offering site. Daily closes for the sixteen assets compared are from this house's Bitcoin series, Robinhood and, for eight funds and stocks, Massive; Treasury yields from FRED; the Nasdaq Composite for the sixteen earlier offerings from FRED; that list of offerings from Renaissance Capital and press records of each deal. Retail money-market fund assets are the Investment Company Institute's weekly series; margin debt is FINRA's monthly series. Spot-fund flows in Exhibit 7 are bitcoin-data.com's daily series in coins; stablecoin supply and long-term holder supply are the series used in §3. The link to the Nasdaq in Exhibit 7 is estimated on the 150 sessions before each fifty-day span.
The liquidation-level test. This house's daily log of the three largest clusters of estimated long and short liquidations around the price, read once a day from CoinGlass's liquidation heatmap for Binance's bitcoin perpetual since late July 2026 (110 readings with a full following day; taken at varying hours, mostly at about 07:30 UTC in November), with placebo levels; lows from Coinbase hourly bars. The rule (the low at, or within 1% below, the densest long cluster still below the price when the slide began) was fixed before the results were read and then corrected on review to count only clusters still below the price when each slide began, which changed two rows and not the count of none in five; the comparison across all 110 days uses the densest cluster at each morning's reading.
The further tests. Leveraged funds' positions from the Commodity Futures Trading Commission's weekly report on CME bitcoin and micro bitcoin futures (latest 3 November); CME futures session prices, obtained through the Massive market-data service, against Coinbase for the gap between futures and spot. Open interest, funding and account ratios from Binance's public data files, Gate and Kraken beside OKX (Binance's November funding is rebuilt from its hourly premium index, which tracked October's actual funding with a correlation of 0.88); Bybit refused the request and Hyperliquid and Deribit do not serve open-interest history. Option trades from Deribit's public trade history; the largest gold fund's daily holdings from its sponsor. Stablecoin supply from DefiLlama. The dollar index rebuilt from European Central Bank daily fixings, because the usual price service refused requests; spot gold and silver and the fund closing prices from Alpha Vantage. Japanese yields from the Ministry of Finance; Asian indices, currencies and equity futures from Yahoo Finance; Korean prices from Upbit. Miner wallet outflows, long-held coins' profit ratio and wallet-size bands from bitcoin-data.com, released by the vendor late on the build day (the profit ratio ends 7 November). Hedge-fund redemption notices as reported by HedgeCo from an SS&C GlobeOp release. Past breaks of the 50-week and 200-day averages from the CoinMetrics daily series since 2011.
The commentary search. Forty-three search queries, forty-five distinct published explanations, each recorded with its author, date and outlet; the ten numbered rows of Exhibit 8 are those that recurred most across outlets.
Limits. The stock-market link is an estimate and moves over time. The long-term holder series is a net count with no identities. Rows of the seller ledger overlap. The rate reading of phase two rests on four sessions, and the stock-offering reading on one earlier case. Nothing in this note is a statement about future prices.
Disclosures
Information only. TON618 Capital. This report is for information purposes only. Nothing here is an offer to sell or a solicitation of an offer to buy any security, fund interest, or digital asset, and nothing here is personalized investment advice or a recommendation regarding any instrument.
Publisher's exclusion. All research is published solely as general, impersonal information of regular circulation. It is not tailored to the objectives or circumstances of any individual and is not issued in connection with compensation from any client. The Fund has no clients and distributes all research free of charge. On that basis it publishes in reliance on the publisher's exclusion from the definition of "investment adviser" under the Investment Advisers Act of 1940 (§202(a)(11)(D); cf. Lowe v. SEC, 472 U.S. 181 (1985)).
Registration & conflicts. TON618 Capital is not registered as an investment adviser or broker-dealer in any capacity. The Fund is a Bitcoin fund and may hold or transact in the securities or digital assets it discusses. This note discusses bitcoin, the spot bitcoin funds (among them IBIT and FBTC), Strategy (MSTR), Strive (ASST), MARA Holdings (MARA), Hut 8 (HUT), Nvidia (NVDA), the QQQ, SPY, SMH and GLD funds, and prediction-market contracts on Polymarket; it makes no recommendation on any of them and states no view on the future price of Bitcoin. The Fund receives no compensation from any party in connection with its research.
Use of AI. Artificial intelligence is used in the creation of this research. All methodology and data integrity are reviewed and approved before publication by TON618 Capital's Chief Investment Officer, Keyth Beck; errors may nonetheless occur, and readers should verify independently.
CFA. This report was prepared to align with CFA Institute analytical standards (methodology only). CFA® and Chartered Financial Analyst® are registered trademarks owned by CFA Institute. That reference describes the analytical framework applied; it does not imply the report was prepared, reviewed, or authored by a CFA charterholder, and the report is not issued, reviewed, endorsed, certified, or approved by — nor affiliated with — CFA Institute.
Risk & feedback. Past performance is not indicative of future results. Digital assets and equities are volatile and may result in total loss of capital. Corrections and feedback are welcome — please direct them to CIO Keyth Beck at keyth@ton618capital.com.