Research · Capital flows

Where the money is, and what it cannot tell you

Every visitor arrives with one question: what will it be worth. Published answers collapse two separate bets — how much money arrives, and where the cycle is when you look — into a single number and call the second a multiplier. This page keeps them apart, measures what history says about each, tests the model the way this site tests everything, and publishes the dates by which the assumptions will have been proved or broken.

Computed from the daily snapshot for 10 September 2026

Realised cap is $1.07T today at an MVRV of 1.44, the 35.6th percentile of every day since 2013. It grew 2.71× over the last three years, against a median of 2.67× for windows that started in the 2020–23 cycle and 16.1× for the first cycle measured.

Reconciliation on 2026-09-11: market cap ÷ supply = $76,675.77 against a published price of $76,675.77 (0.0%); realised cap × MVRV = $1.54T against a market cap of $1.54T. Realised cap is derived as market cap ÷ MVRV.

What this method would have said, and what happened

Every monthly origin since 2016: the forecast made, then marked when its three years were up

What this section is. Anyone can build a model that fits the past. The question is what it would have said at the time, before the answer was known. So: stand at a date in the past, use only the data that existed then, forecast the price three years ahead, and wait for that date to arrive. Then compare. Do it at every month since 2016 and you have 91 honest tests of the method this page uses.

Why the table stops in 2023. A three-year forecast can only be marked once three years have passed. The newest origin that can be scored is therefore exactly three years old — its target date is today. Origins after it are still running; their forecasts exist and their answers do not. That is not missing data, it is what waiting for the answer looks like.

The model itself: realised cap at the origin, times how much realised cap grew over comparable earlier windows, times where MVRV sat at the end of those windows. A calibrated model puts the actual price inside its own interquartile range about half the time.

Origins tested91monthly since 2016
Actual inside the IQR28%calibrated: about 50%
Typical missforecast too highby a factor of 2.3 on average
Last 18 origins0 of 18inside the IQR

The model put the actual price inside its own interquartile range 28% of the time, against roughly half for a calibrated model, and its misses run in one direction: it forecast too high by a factor of about 2.3 on average. The reason is the deceleration in the next panel — each cycle’s capital growth has been lower than the cycle the model trained on. That is what this history can support, and every number below should be read with it in front. The pooled figure flatters the recent record: of the last 18 origins in the table, 0 landed inside the range.

Standing on each date and forecasting three years out. The scored rows run to 2023-09-06, whose forecast came due on 2026-09-06. The 36 rows beneath are forecasts already made whose answers arrive between 2026-10-06 and 2029-08-21.
Standing on
the date the forecast was made
Forecasting
three years ahead
It said, low
25th percentile
It said, middle
median forecast
It said, high
75th percentile
What happened
the actual price
Right?
inside its own range
2022-042025-04$240k$328k$601k$84kno
2022-052025-05$229k$312k$561k$104kno
2022-062025-06$220k$301k$516k$106kno
2022-072025-07$201k$274k$457k$117kno
2022-082025-08$192k$260k$425k$119kno
2022-092025-09$177k$249k$403k$114kno
2022-102025-10$162k$240k$379k$114kno
2022-112025-11$153k$231k$351k$105kno
2022-122025-12$142k$217k$303k$93kno
2023-012026-01$138k$211k$298k$91kno
2023-022026-02$136k$205k$300k$69kno
2023-032026-03$132k$199k$299k$69kno
2023-042026-04$130k$195k$296k$71kno
2023-052026-05$129k$195k$280k$80kno
2023-062026-06$109k$192k$269k$63kno
2023-072026-07$107k$188k$262k$63kno
2023-082026-08$106k$182k$246k$64kno
2023-092026-09$102k$171k$237k$80kno
Still running — 36 forecasts made since the last one that could be scored, shown with the date each comes due
2023-102026-10$99k$158k$232kstill running — due 2026-10-06
2023-112026-11$100k$154k$231kstill running — due 2026-11-05
2023-122026-12$102k$153k$232kstill running — due 2026-12-05
2024-012027-01$106k$155k$237kstill running — due 2027-01-04
2024-022027-02$105k$158k$240kstill running — due 2027-02-03
2024-032027-03$110k$167k$251kstill running — due 2027-03-04
2024-042027-04$125k$188k$282kstill running — due 2027-04-03
2024-052027-05$126k$188k$284kstill running — due 2027-05-03
2024-062027-06$130k$171k$290kstill running — due 2027-06-02
2024-072027-07$130k$167k$289kstill running — due 2027-07-02
2024-082027-08$130k$169k$285kstill running — due 2027-08-01
2024-082027-08$126k$167k$274kstill running — due 2027-08-31
2024-092027-09$124k$168k$278kstill running — due 2027-09-30
2024-102027-10$125k$169k$288kstill running — due 2027-10-30
2024-112027-11$137k$183k$321kstill running — due 2027-11-29
2024-122027-12$152k$197k$354kstill running — due 2027-12-29
2025-012028-01$159k$197k$371kstill running — due 2028-01-28
2025-022028-02$161k$197k$354kstill running — due 2028-02-27
2025-032028-03$161k$195k$318kstill running — due 2028-03-29
2025-042028-04$161k$195k$309kstill running — due 2028-04-28
2025-052028-05$168k$203k$308kstill running — due 2028-05-28
2025-062028-06$174k$210k$271kstill running — due 2028-06-27
2025-072028-07$185k$224k$279kstill running — due 2028-07-27
2025-082028-08$190k$230k$284kstill running — due 2028-08-26
2025-092028-09$195k$236k$290kstill running — due 2028-09-25
2025-102028-10$202k$244k$298kstill running — due 2028-10-25
2025-112028-11$205k$247k$300kstill running — due 2028-11-24
2025-122028-12$204k$247k$298kstill running — due 2028-12-24
2026-012029-01$204k$247k$294kstill running — due 2029-01-23
2026-022029-02$197k$238k$282kstill running — due 2029-02-22
2026-032029-03$195k$234k$274kstill running — due 2029-03-24
2026-042029-04$194k$230k$268kstill running — due 2029-04-23
2026-052029-05$194k$227k$264kstill running — due 2029-05-23
2026-062029-06$190k$220k$256kstill running — due 2029-06-22
2026-072029-07$187k$217k$253kstill running — due 2029-07-22
2026-082029-08$184k$213k$252kstill running — due 2029-08-21

Bet one: how much money arrives

Realised-cap growth over three-year windows, by the cycle the window started in
Growth multiple of realised cap, three years forward
Windows startingp25Medianp75
2013-167.5×16.1×21.0×
2017-194.0×4.5×6.0×
2020-231.8×2.7×3.0×
All windows2.7×4.1×11.3×

Realised-cap growth has fallen every cycle. The all-windows figure is dominated by years when bitcoin was a hundredth of its present size and is not a forward base rate. Growth over three-year windows and MVRV at their end correlate at -0.29: big inflows arrive at high prices, lift the cost basis, and when the cycle turns MVRV compresses onto it. The combination every large published target needs — high inflow and high endpoint MVRV — is the one history produces least.

Bet two: where the cycle is when you look

MVRV on every day since 2013
p251.29
Median1.70
p752.18
Today1.4435.6th percentile

This is the larger of the two levers. Across the interquartile range alone, the same realised cap is worth 1.7× more at the top than at the bottom. Any published range that holds it near a single value is hiding its main uncertainty, and none says which value it chose.

The surface

Price at the horizon for each pair of assumptions. Not a forecast: a map of what each assumption implies.

Read across a row: what more money buys at a fixed point in the cycle. Read down a column: what the cycle does to a fixed amount of money. The column span is about 4.1×; the row span is about 3.0×. Where the cycle is when you look matters at least as much as how much money arrived.

Price on 2029-09-11 at supply 20.46M. Rows: MVRV at the horizon. Columns: capital arriving by then.
MVRV at horizon+$0.50T+$1.00T+$1.30T+$2.00T+$3.00T+$5.30T
1.0 (trough)$77k$101k$116k$150k$199k$311k
1.3$100k$131k$150k$195k$258k$405k
1.7 (median)$130k$172k$197k$255k$338k$529k
2.2$169k$222k$255k$330k$437k$685k
3.0$230k$303k$347k$450k$597k$934k

River (Sam Baker), 2 Sep 2026 publishes $250k to $840k over 3-5 years on advisor adoption 20-40% of portfolios at 2-4%, $1.3-5.3T inflow, x3 multiplier. On this surface that requires MVRV at the horizon of 2.16 at its low inflow and 2.70 at its high — the upper quarter of MVRV’s history, assumed rather than stated.

Who holds the coins

Primary sources only: SEC filings and published on-chain addresses
Tracked holders as a share of supply
CategoryBTCShare of supplySource
US spot ETFs1,167,5185.81%SEC 10-Q/10-K XBRL, us-gaap:CryptoAssetNumberOfUnits
Listed treasury companies846,0004.21%SEC XBRL where filed; see fetch/treasuries.py
Sovereigns, on-chain00.00%published reserve addresses, read on chain
Tracked total2,013,51810.03%of 20.08M

The sovereigns row reads zero because the pipeline has not yet resolved a primary source for it; the file says which. These are the buyers who were not in the 2013–2023 windows the model trained on. If their share crosses the tripwire below, the historical base rates stop applying and no backtest can say what replaces them. That is the strongest argument against every number on this page, and it is measurable.

Tripwires

Dated, checkable, and reported against live data every morning
What would prove each assumption wrong, and when we will know
Assumption testedConditionReads nowCheck byStatus
Inflow above the recent cycle
capital arriving faster than any window of the 2020-23 cycle; River's high case in play
realised cap above $1.6T$1.07T2027-03-12not yet
Inflow below the low case
even River's low case is not arriving
realised cap below $1.3T on the check date$1.07T2027-03-12pending until check date
Buyer mix has changed
structurally new holders at scale; the historical deceleration may not apply
tracked ETF + treasury + sovereign share above 12% of supply10.03%2027-09-11not yet
Endpoint timing assumption
every published range that held MVRV near constant was wrong about timing
MVRV outside 1.3-2.2 at the horizon1.4412029-09-11pending until horizon

What this does not show

Realised cap is derived, not read: Coin Metrics publishes no direct series on the tier this site uses, so it is market cap divided by MVRV. The reconciliation line at the top shows the identity holding to the cent, and a test asserts it on every run.

Four cycles. The growth distributions above are drawn from thousands of overlapping windows and a handful of independent regimes. The walk-forward is the honest measure of what that history can support, and it says the model overshoots.

The holder table lists only what the pipeline reads from a filing or an address. Treasury companies that disclose only in prose, and governments that hold coin without a published address, are absent by design. The residual is everyone else, and it is most of the supply.

Why no number

The surface has thirty cells and every published target sits in one of them. Reading a cell without its row and column is reading a forecast without its assumptions, which is what every published target asks you to do.

The precision on this page is not in a prior. Realised cap is observable daily. Six months from now the inflow tripwire will have been checked and the surface will have lost half its columns to evidence rather than assumption. That is the only kind of precision available here, and it arrives on a schedule.

What bitcoin did with no model attached at all is on the base rates page, and what the site said on each past day is in the forward record.

Recomputed daily by fetch/flows.py and published as flows.json. Method on the methods page.