Contrarian macro · the echo test

The Halving Is a Phase-Shifted M2 Echo

Here you will see whether bitcoin's four-year halving cycle is better explained by US money-supply growth arriving on a delay than by the halving clock itself — and where that leaves the current cycle.

Over 164 months from 2013-01 to 2026-08, bitcoin's 12-month log return lines up best with US M2 growth shifted back 3 months (correlation +0.19, versus +0.18 unshifted). Raced head to head on the same months, the halving clock alone explains +19% of the variance and lagged M2 alone explains +4%. By that measure the cycle reads more like a real halving rhythm.

Updated 19 September 2026 · M2 observed for 2026-07-01 · bitcoin price observed 2026-09-18

Best M2 lag, scanned 0–18 months3 monthsCorrelation +0.19 at that shift vs +0.18 unshifted · 2013-01 to 2026-08

Charlie's readCorrelation measures how often two things move together, from −1 (always opposite) to +1 (always together). The best match appears when US money growth is shifted back 3 months — meaning today's money growth tends to show up in bitcoin's yearly return about 3 months later. With no shift at all the two barely notice each other (+0.18), so the relationship only exists on the delay.

Halving clock explains+19%Share of variance in 12-month bitcoin log returns, 164 months

Charlie's readThis is the share of bitcoin's year-to-year moves that the halving calendar alone can explain: +19%. That is a modest share — most of what bitcoin does year to year has nothing to do with the halving countdown everyone watches.

Lagged M2 explains+4%Same months, same target — input is M2 YoY at the best scanned lag

Charlie's readSame question, but the only input is US money growth heard 3 months late: it explains +4% of bitcoin's yearly moves. The clock still beats money tonight — the halving rhythm is not just an echo, at least not on this measure. (Fair warning: the lag was chosen on this same data, so treat this number as a best case.)

Cycle position · 2026-09-182.4 years past halvingLast halving 2024-04-20 · next estimated April 2028 (estimated)

Charlie's readWe are 2.4 years into a cycle that has historically run about four. That puts today in the middle stretch — historically where the big swings of past cycles happened. The next halving is estimated around April 2028 (estimated).

Money first, price later

US M2 grows in pulses. Bitcoin's celebrated four-year rhythm may be those pulses arriving late. The chart puts the bitcoin price (log scale) against M2 year-over-year growth; the lag scan then asks exactly how late the echo arrives.

Bitcoin price (log) vs US M2 year-over-year growth

Price observed 2026-09-18 · M2 observed for 2026-07-01 (monthly series, published with a delay) · halvings marked

Takeaway: the money pulses and the price waves rhyme, but only once M2 is allowed to speak first.

Lag scan: correlation of 12-month bitcoin return with M2 YoY shifted 0–18 months

164 monthly observations, 2013-01 to 2026-08

The race is deliberately unfair to the money story: M2's lag was chosen on the same data it is then scored on, which flatters it. Even so, the unshifted correlation of +0.18 rises to +0.19 at a 3-month shift, and the halving clock — the variable everyone watches — explains +19% on its own. If the four-year cycle were purely a supply event, the clock should win this race comfortably. It still does.

The race: halving clock vs lagged money

Two one-variable models on identical months. Model A knows only how many days have passed since the last halving. Model B knows only M2 growth at its best scanned lag. The winner is the better explanation of the cycle.

Model A · days since halving+19%Variance explained in 12-month log returns over 2013-01 to 2026-08

Model B · M2 YoY, best lag — the halving clock wins+4%Same target, same months · lag chosen in-sample, so read as an upper bound

Current implied position · 19 September 2026M2 growth observed for 2026-07 implies a 12-month bitcoin log return near +62% under the lagged model; the actual reading for the latest complete month (2026-08; the still-open current month is excluded) is -32% — price is running behind the money echo. The halving clock sits 2.4 years past the 2024-04-20 halving. If the echo framing is right, the variable to watch for the next leg is the M2 print, not the countdown to April 2028 (estimated).

The four echoes, overlaid

Every cycle drawn from its halving date, log return against days elapsed. The overlay shows how similar the arcs are — the question is whether a supply schedule alone should produce that, or whether an external metronome is conducting.

Log return since each halving, days elapsed

Halvings 2012-11-28, 2016-07-09, 2020-05-11, 2024-04-20 · price observed 2026-09-18

Each curve starts at a halving and tracks the log return from that date. The current cycle is 2.4 years in. In the echo framing, similar shapes across cycles are exactly what a shared external driver would produce; a true supply shock would not need M2 to rhyme with it.

The contradictionThe awkward fact for the halving story: an unshifted correlation of +0.18 means money growth and bitcoin returns barely acknowledge each other in real time — the relationship only appears after a 3-month delay that was itself chosen by this scan. And the awkward fact for the echo story: with four halvings and 164 months, both models are fitting a handful of waves. The halving clock wins tonight's race at +19% against +4%, but that margin is not a proof.

How this could be wrongThis could be wrong because M2 is revised, published with a delay (latest observation month: 2026-07), and measured for one currency area while bitcoin trades globally; because the best lag is fitted in-sample and will decay; and because 164 monthly observations contain barely three full cycles. If the 3-month lag keeps drifting from cycle to cycle, the echo is a coincidence of timing, not a mechanism, and the page should say so next run.

The bottom line

Everything above, compressed into plain English: what the numbers say right now, what Charlie watches next, and what would prove the whole thesis wrong.

Charlie's read · 19 September 2026As of 19 September 2026: raced head to head over 164 months, lagged US money growth explains +4% of bitcoin's yearly moves against +19% for the halving clock — so the cycle reads more like a genuine halving rhythm than a money echo. Right now price is running cooler than the money echo implies (model implies +62%, latest complete month printed -32%). What Charlie watches next: the monthly M2 print, not the halving countdown — under the echo framing, money growth observed today is the forecast for about 3 months from now. What would prove the thesis wrong: the best lag wandering from cycle to cycle, or the clock model winning this race outright — either means the echo was a coincidence of timing.

For the statistically curious

The full plumbing — exact data definitions, model choices, and honest limitations. Nothing below changes a number above; it only explains how the numbers were built.

Methodology, data sources and limitations

How the numbers are built

Bitcoin is the daily USD reference price (Coin Metrics community reference rate), observed 2026-09-18. M2 is the US monthly money-stock series, latest observation month 2026-07-01.
The target variable is the 12-month log return of month-end price, 2013-01 to 2026-08 — 164 months. The still-open current month is excluded: its last print is a mid-month price, not a month-end close.
Model A regresses that return on years elapsed since the most recent halving. Model B regresses it on M2 YoY shifted by the best of 19 scanned lags (0–18 months). R² is reported for both on identical month sets.
The best lag is selected on the same data it is scored on; Model B's R² is therefore an in-sample upper bound, and the page labels it as such rather than hiding the advantage.
M2 covers one currency area and is revised after publication; bitcoin trades globally, 24/7. A matched calendar is a real limitation, not a footnote.
Missing data stays missing. No month is filled in, and no correlation is reported on fewer than 24 paired observations.