Will Bitcoin Make a New High by Year-End?

In July the house named three signposts that would mark the turn from a bitcoin floor to a bitcoin trend. Two fired — flows on 16 July, the 50-day average on 21 August — weeks before Monday's billion-dollar ETF day, and the third has not. The market still pays one chance in ten for a new all-time high by year-end, the options market agrees to the decimal, and the house's own cycle tools put it nearer one in five, with the house's own reference class as the reason to doubt them. This first issue of the Bitcoin Quarterly scores the July signposts, reads the state of play, and arms three new paths to 31 March 2027 with the tripwires that would end them.

TON618 Capital Research · As of 22 September 2026 · Download PDF

On Monday 21 September the US spot bitcoin ETFs took in $998.96 million, their largest dollar day since 6 October 2025 and, at roughly 11,530 bitcoin, their largest intake in coin terms since November 2024. The price closed at $86,505, up 13% in eleven sessions and 48% from the 30 June low; it reclaimed its 200-day average on 19 August after 289 sessions beneath it, and the week ending 20 September closed above the 50-week average for the first time in 44 weeks. The argument that followed is the one every bitcoin turn produces: proponents of the price read the flows as the arrival of the price-insensitive buyer, and skeptics read a 31% drawdown from a high eleven months old, a Federal Reserve that raised rates six days earlier, and a prediction market that gives a new high by year-end one chance in ten. This issue takes that argument in three parts: the July signposts, scored on their own thresholds; the state of play — who is buying, who is selling, what the market pays for the rungs above the price and where the house's instruments land; and three new paths to 31 March 2027, each with the condition that confirms it and the tripwire that ends it.

About this series. This is the first issue of a quarterly note: the house's read of the state of the bitcoin market, on a fixed plan that does not change from issue to issue — score the tripwires the previous issue armed, read the current state with the same instruments, and arm the next set. It builds on work the house has already published rather than starting over: the July floor note, Is the Bottom Forming?, whose signposts it scores; Structural, or Just Winter? (July 2026), on whether the marginal buyer had left for good; the ETF-flow elasticity study behind the finding that fund flows follow the price; What Actually Drives Bitcoin, on which macro and on-chain series lead; and the live engines — the regime composite, the drawdown monitor, the Markov forecaster and the moving-average cascade — that supply the numbers in Part Two. None is required reading, and each is cited where it is used. Issue two follows the 31 December resolution of the year-end rungs; issue three scores the March paths.

Terms used throughout, defined once. A signpost is a public, checkable condition named in advance — the July note named three, each with a threshold, and this note grades them against that threshold and nothing else. The 50-day and 200-day averages are the mean of the last 50 and 200 daily closes, and the 50-week average the mean of the last 50 weekly closes (weeks ending Sunday); the price is "above" one when the close exceeds it. Spot ETF net flow is the bitcoin created or redeemed by the US spot bitcoin exchange-traded funds in a session, here in bitcoin rather than dollars so that days in different price regimes compare. A touch contract pays if the price trades at or above a level at any moment before a date — not where it finishes; the Polymarket contracts in this note resolve on any one-minute high on Binance's BTC/USDT pair. An options-implied probability is the chance a strike is reached that is consistent with what options at that strike cost; implied volatility is the annualised dispersion those prices assume. The regime chain is the house's forward model: it sorts each day of bitcoin's history into a rising, flat or falling state by the trailing 20-day return, measures how often each state follows each other, and replays those odds forward. A reference class is the set of past episodes that share the present set-up, used to ask what happened next. A long-term holder is a coin unmoved for 155 days or more.

The July note, Is the Bottom Forming?, found that the transfer of coins from weak hands to strong hands was real and verifiable, and that on the evidence it times the floor, not the turn. It named three signposts that would mark the turn, each with a threshold, and one falsifier. This section grades all four against their thresholds, on the same data series the July note used, and against nothing else. The order below is the July note's.

What the grade means, on the July note's own terms. The note said that any one signpost alone "has flickered before in this bear," that two together "would mean something structural — that the price-insensitive seller of the past nine months (fund-share redemption) has been replaced by a price-insensitive buyer," and that this demand-side event is what accumulation alone cannot supply. The second signpost completed on 21 August. From that date the July note's frame says the market changed from floor-forming to something structural, and the flow and price series since have behaved accordingly: 34,866 bitcoin of net creations over the last 30 sessions on the house feed before Monday, and the price above every average it publishes. What the frame does not say is that a structural change is a new high. The three signposts were confirmation conditions for a turn; the distance from a turn to the old high is the subject of Part Two.

What the July scenarios said, and what happened. The July note's six-month frame, to early January 2027, gave 30% to a bull case in which the signposts confirm and the price reaches about $85,000, 45% to a base case in which the floor holds and the averages cap rallies at $66,000, and 25% to a bear case in which the June low breaks and the price finds the high $40,000s. The bull case's level printed on 21 September, fifteen weeks before its horizon; the base case's cap, the 50-day and 200-day averages, was cleared in August and both now sit below the price; the bear case's trigger never fired. Read as a scorecard, the July frame under-called the size of the move and correctly called its shape: floor first, then the demand-side event, then the trend. That is the reason Part Three replaces the frame rather than extending it — its horizon and its levels are spent.

One correction to the record, in the house's own accounting. The two signposts fired on 16 July and 21 August. Neither was scored publicly when it fired; both are scored here, five and nine weeks late respectively. The delay changes nothing about the grade — the thresholds were fixed in July and the series are archived — but a signpost that is published as a public test should be scored on the day it clears, and the house has now armed the third signpost and the year-end rungs with dated checks so that the next grade is not late.

Three things describe the market on 22 September: who is buying and who is selling, what the market pays for the rungs above the price, and where the house's own tools land. In that order.

Who is buying, who is selling

The buyer is the fund complex and the seller is the holder cohort that built the floor. Monday's $998.96 million of net creations was the fund complex's largest dollar day since 6 October 2025 and, at roughly 11,530 bitcoin, its largest intake in coin terms since November 2024; the house feed shows 34,866 bitcoin of net creations over the 30 sessions before it. On the other side, long-term-holder supply peaked on 24 July at 16.86 million bitcoin and has released 364,000 since, 94,000 of them in the last 30 days, into the rising price. In the July note's frame this is the ownership transfer of June running in reverse: conviction holders who absorbed panic supply at $58,000–$65,000 handing it to price-insensitive fund creations from $64,000 up to $87,000. That is what a turn looks like — and, as the case for one in ten notes, what a distribution top looks like, which 100 days cannot separate. The house's three trend and risk instruments read the same tape the same way. The moving-average cascade is golden on both anchors: the daily 50-over-200 cross printed on 8 September, and the weekly has been golden for 141 bars, since early 2024. The drawdown monitor reads 35.7 — "moderate," inside the 26.5-to-41.4 range it has held since May, its hottest input sentiment at 80 of 100. The Markov forecaster has the price at the start of a rising state, with 46% odds of still being in it twenty sessions on against 26% for a falling one. None of the three is a probability for 31 December; all three are consistent with the state the regime chain replays below.

What the market pays: two markets that agree

The contract at the centre of the argument is Polymarket's Bitcoin all-time high by December 31, 2026? It pays Yes if any one-minute candle on Binance's BTC/USDT pair between 16 December 2025 and 11:59 pm Eastern on 31 December 2026 prints a high above every prior one-minute high — in practice, above the $126,199.63 of 6 October 2025. On 22 September it traded at 9.9% with $1.82 million of volume, and it is the liquid one: its sibling on $100,000 by the same date printed 46% on about $20,000 of volume, and the $100,000-by-March-2027 contract 59.5% on $3,000. The market's view is therefore precise on the high and thin on the rungs beneath it. The same crowd prices a dip to $50,000 by year-end at 6% and to $45,000 at 4%; a $95,000 print in the remaining days of September at 10.65%; a new high by 30 September at 0.75%.

The second market is Deribit, where December options on bitcoin carry 118,429 bitcoin of open interest against a volatility index of 37.7. The December future trades at $87,616, a 5.0% annualised carry over spot. Implied volatility at the $95,000 strike is 38.5%, at $100,000 38.8%, at $125,000 43.7% — a smile that slopes gently up toward the high, so the market pays a little more for the tail than a flat surface would. Read as terminal probabilities on the 25 December expiry, the call spreads say 30% for a finish above $95,000, 21% above $100,000, 2.5% above the old high. But the Polymarket contracts pay on a touch, not a finish, and a touch is roughly twice as likely as a finish for a level the price would have to cross: on the standard barrier arithmetic, extended six days to 31 December and using each strike's own implied volatility and the future's carry, the December options imply a 63% chance of touching $95,000, 46% for $100,000 and 9.0% for the high.

The two markets agree. On $100,000 they print 46% and 46%; on the high 9.9% and 9.0%. They are priced by different participants — Polymarket's contract by a retail crowd in a binary token, Deribit's surface by dealers and funds hedging in bitcoin — under different mechanics, and neither can arbitrage the other cleanly. The agreement is the strongest fact in this note about the market's number: it is not a quirk of one venue. Any house number above it is a claim that both crowds are wrong in the same direction.

What 9.9% requires. A gain of 45.9% in 100 days. At the options market's 43.7% implied volatility for the high strike, that is a move of about 1.7 standard deviations over the period — reachable, and priced as such. At bitcoin's realised volatility over the last 30, 90 and 365 sessions (42.7%, 39.7%, 45.5%), a driftless replay gives the touch 5.7% to 9.3%. The market's number is, to the decimal, what the asset's recent volatility with no directional view produces. A house number above it must therefore come from a directional view: that the current state has drift.

Where the house lands: four lenses, one method

The house committed to its method before it saw the numbers. The published probability is the regime chain replayed from the current state at this month's realised volatility; the other lenses are checks, reported in full, and the reference class is reported as the reason to doubt the published number rather than as a way to shade it. The four lenses are the options market (above), a realised-volatility baseline (above), the regime chain and the reference class.

The regime chain. Each day since 1 January 2014 is sorted by its trailing 20-day return into a rising state (above +5%), a falling state (below −5%) or a flat one, using only returns through the prior day. Rising days are followed by rising days 89% of the time, falling by falling 87%, flat by flat 76%; the current state is rising: the 21 September close took the trailing 20-day return to +11.8%, so the label applies from the next session. Replaying 50,000 100-day paths from that state, drawing each day's return from its state's own history, gives the high 36%, $100,000 73% and $95,000 82%. The rising state's history has an annualised dispersion of 66% against 42.7% realised this month, and that dispersion — 2017 and 2021 resampled — is doing part of the work. Rescaling each state's returns to this month's dispersion, keeping each state's own centre, gives 22% for the high, 70% for $100,000 and 82% for $95,000. That is the house's number. Starting the sample in 2020 instead of 2014 gives 20%; using the options market's 38.5% in place of realised volatility gives 19%. The range across those choices is 19–22% for the high and 68–70% for $100,000; the $95,000 rung sits at 80–82% under every variant.

The reference class. Seven prior weeks whose close crossed above the 50-week average after at least eight weekly closes below it while still at least 20% under the running high — the shape of the week ending 20 September 2026, which followed 44 weeks below.

Four of seven gained at least the 9.8% that $95,000 requires; four of seven at least the 15.6% that $100,000 requires; three of seven at least the 45.9% that the high requires. None made a new high inside 100 days, because the three that gained 46% or more started from 71–79% under their highs. The median best gain is +33%, which is why the rungs at $95,000 and $100,000 read as ordinary in this class and the high does not. The one episode near today's depth, March 2022, is the class's worst on both counts: a 1.5% best gain and a 59% worst point. Seven is not a sample from which to estimate a probability; it is a list from which to read what has and has not happened. What has not happened is the event the house prices at one in five.

Reconciling the two. The chain says the rising state has drift and the market's driftless replay omits it; the reference class says the drift after a reclaim from near this depth has, once, been zero. The house's published number is the chain's, because that was the method declared, and because the chain's own sensitivities are narrow. The honest description of one in five is: a directional model's number, above two agreeing markets, with a base rate of zero in the seven cases that most resemble today. The reader who weights the reference class over the model will land at the market's number, and the note does not think that reader is wrong. The house's claim is narrower: the market's 9.9% is a driftless number, and the state the price is in has not, historically, been driftless.

What drives the house's read, and what dissents

The chain says the current state has drift. Whether the house's other instruments agree is the composite's question. The house's composite regime score blends 29 live signals, each weighted by its own measured track record and by how many others move with it, into one reading between −1 and +1. On 22 September it reads +0.117, a lean toward rising, with 60% of the weight on that side and a margin of three: remove any three of the same-sign leaves and the sign flips. It is a lean, not a call.

The drivers. Long-term-holder supply is the largest single contribution, and it needs reading with care: the leaf scores the series against its own history, where the July build still reads as unusually high, while the raw series has turned, as the buyer-and-seller reading above showed. The growth-and-inflation nowcast sits in the quadrant of rising growth with falling inflation, the composite's second-largest rising contribution. The five valuation leaves with verified edge — RHODL, reserve risk, MVRV, the 200-week deviation and NUPL — all read below their own long-run centres rather than stretched. ETF flow itself contributes modestly because the scored transform is the five-day z-score, and Monday's print had not yet entered the house feed.

The largest dissents are macro, and they are the July note's ceiling. Twelve of the 29 leaves vote down. The five largest — the funds rate, the dollar, miner selling, Tether issuance and global money supply — carry roughly three-quarters of the falling weight, and three of the five are macro; the other seven (the ten-year, M2 convergence, net liquidity, Puell, Fear & Greed, address growth and weekly RSI) are each small. The Federal Reserve's 16 September increase to 3.75–4.00% was unanimous, and the median projection for year-end, 4.1%, implies another before December. The July note's base case — an on-chain floor under a macro ceiling — assumed exactly this configuration; what has changed is that the price walked up through the averages while the ceiling stayed. The composite does not resolve the disagreement. It reports it.

Part One retired the July scenarios; Part Two ended on a disagreement the composite reports and does not resolve. This section turns both into a forecast that can be scored, and what it arms is scored in the next two issues: the year-end rungs in issue two, the six-month paths in issue three. The levels are fixed today and are observable: a bull path in which the 31 March 2027 close is at or above $100,000; a base path in which it sits between today's 200-day average, $70,582, and $100,000; a bear path in which it is at or below the 200-day. Three probabilities are shown for each, from three lenses that disagree, and the house's published number is the average of the first two: the options market's risk-neutral bucket from the March 2027 chain (34,102 bitcoin of open interest, future $88,650), and the regime chain at this month's volatility run 190 days. The chain alone is not published at six months: compounding a rising state's drift for 190 days is the model's known weakness, and its bull bucket — 67% — is the number the reference class was built to check. The third lens is that class: where the seven prior reclaims of the 50-week average stood 190 days later, in the same buckets.

What the three lenses are saying. The options market is a driftless number with a fat left tail: it puts 23% on a close below the 200-day, which the chain puts at 6.5% and the reference class saw once in seven. The chain is a momentum number: it puts two-thirds on a close above $100,000, which the market puts at 28% and the reference class — five of seven — supports more than it supports anything else in this note. The same caveat applies in reverse: all five began 55–79% under their highs, and the one that began near today's depth, March 2022, is the one in the bear bucket. The house's average is therefore a bull path at 47%, a base path at 38% and a bear path at 15%. Read against the year-end rungs: the house gives a new high by 31 December one chance in five; by 31 March 2027 the options market's own touch probability for the old high rises to 19%, and the chain's to 51%. The house does not publish a six-month number for the high; the bull path's confirming observable, a weekly close at or above $100,000, is the test that would make one worth publishing.

The record this creates. Six dated rows. Three are scored on 31 December 2026 on the Binance one-minute high: a new high — house 22%, market 9.9%; $100,000 — house 70%, market 46%; $95,000 — house 82%, options-implied 63%. Three are scored on the 31 March 2027 close: bull 47%, base 38%, bear 15%, against the options market's 28 / 50 / 23. Both sets were published on 22 September and neither can be revised.

Tripwires (in-note; none is a publishing hold). The three path confirmations and the three path enders in the Part Three table are armed as dated checks · the Deribit quarterly expiry of 25 September, with 185,907 bitcoin of open interest · the 28 October FOMC, the meeting at which the projections' further increase would come · the 30 October and 25 December Deribit expiries · the 31 December resolution of the Polymarket contract, on which the house will publish its score against the market's — a Brier score on each of the three rungs for both.

Observed (primary and vendor series, dated). Daily closes, highs, the 50- and 200-day averages and the 50-week average (ISO weeks, Sunday close) from the house daily series (CoinMetrics reference rate, 2010 to 21 September 2026; the ta-cascade daily candles from 2016 for volatility); US spot ETF net flow in bitcoin from the BGeometrics feed behind the Signal Ledger's flow leaf, as of 18 September, with Monday's figure from two press compilations of the issuers' daily creations; long-term-holder supply from the same vendor's 155-day cohort series; the Coinbase-versus-OKX five-day premium from the Signal Ledger; Polymarket contract prices, volumes and resolution terms from its public API on 22 September, archived; the Deribit December and March option book, index, future and volatility index (DVOL, hourly, archived as data/deribit_dvol_2026-09-22.json) from its public API on 22 September; the FOMC statement of 16 September 2026.

Inferred (computed from the observed, with the code archived). The signpost grades and their dates, from the archived series and the July thresholds; the options-implied terminal and touch probabilities, the realised-volatility baselines, the regime chain and its sensitivities, the reference-class table, and the six-month scenario buckets under all three lenses (data/house_probability.py); the composite score and its leaf table (the house composite, run 22 September, archived); the tape exhibit's series (data/tape_series.json).

Assumed (model choices, stated). The scenario buckets are fixed at today's 200-day average and the $100,000 round number, chosen for observability, not for fit; the house scenario probability is a stated equal average of the market's risk-neutral bucket and the chain's, a rule adopted because the reference class supports neither lens alone at six months. The barrier arithmetic treats the price as lognormal with the strike's own implied volatility and the December future's carry, held flat for the six days between the Deribit expiry and 31 December; a touch on a one-minute high is slightly more likely than the daily-close basis the chain and the reference class use, so both understate by a small margin the event the contract pays on. The regime chain's states are defined by the trailing 20-day return at ±5%, the house forecaster's own parameters; returns are resampled independently within a state, which ignores dependence inside a state and is the model's known limitation. Each state's historical daily centre since 2014 is retained in the chain as a description of the sample and archived with it (data/house_probability.json); it is not a forecast of any return and none is stated here. The 2014 start is the post-Mt.Gox era; the full-history variant, in which the rising state's dispersion is 100% annualised, is shown in the archive and used for nothing.

CoinMetrics community reference rate, BTC PriceUSD, via the house cache (btc-regime-forecast/prices.csv, 2010-07-18 to 2026-09-21); ta-cascade daily OHLCV (~/ta-cascade/data/candles_1d.csv, 2016-02-07 to 2026-09-21). BGeometrics / bitcoin-data.com v1/etf-flow-btc (US spot ETF net flow in BTC, 2024-01-11 to 2026-09-18) and v1/long-term-hodler-supply-btc (to 2026-09-22). KuCoin News, "U.S. Bitcoin Spot ETFs Record $998.96M Net Inflow on September 21" (issuer breakdown: IBIT $381.37M, ARKB $289.12M, FBTC $238.84M); CryptoSlate, "Bitcoin ETFs just absorbed 11,500 BTC in their biggest buying day in nearly two years," 22 September 2026. Polymarket gamma API: bitcoin-all-time-high-by-december-31-2026 (9.9%, $1.82m volume), will-bitcoin-hit-100k-by-december-31-2026 (46%), will-bitcoin-hit-100k-by-march-31-2027 (59.5%), the September monthly and dip contracts, pulled 22 September 2026 and archived in data/. Deribit public API: get_book_summary_by_currency (BTC options), get_index_price, get_volatility_index_data, 22 September 2026, archived. Binance BTC/USDT one-minute high of 6 October 2025, $126,199.63 (TradingView; Binance.US quotes $126,198.07). Federal Reserve Board, FOMC statement and Summary of Economic Projections, 16 September 2026. TON618 Capital, Is the Bottom Forming? The On-Chain Floor vs. the Macro Ceiling, 7 July 2026, for the signpost thresholds, the falsifier and the scenario table; Structural, or Just Winter?, 19 July 2026, and the ETF flow elasticity study for the flow-follows-price finding. House engines: the BTC regime composite (v3, 22 September), the drawdown monitor (10:10 UTC, 22 September), the Markov regime forecaster (22 September), the ta-cascade state file (11:45 UTC, 22 September). Method for the probabilities and the reference class: data/house_probability.py, seeded, with its output data/house_probability.json.