Contrarian macro · the rebalancing test

Month-End Gravity

Here you will see whether big S&P 500 months drag bitcoin around in the last three trading days through 60/40 rebalancing — and whether the pattern existed before bitcoin ETFs, which would kill the story.

In months where the S&P 500 gained 3% or more, bitcoin's return over the last three trading days averaged +0.97% across 10 post-ETF months — against +3.20% across 24 pre-ETF months. The Welch t-statistic for that difference is -1.25. In big down-months the readings are -3.93% (n=3) after and +0.86% (n=13) before, t -2.15. The rebalancing-gravity thesis is undecided.

Updated 19 September 2026 · equities observed 2026-09-18 · bitcoin observed 2026-09-19

Big up-months, ETF era+0.97%Mean bitcoin return, last 3 trading days · n=10 months since 2024-01

Charlie's readWhen the stock market had a big winning month (up 3% or more) since bitcoin ETFs launched, bitcoin gained about 0.97% on average in the month's final three trading days. That is the 'rebalancing pull' — big funds trimming or topping up at month-end. But the average rests on just 10 months, so one unusual month can flip it.

Big up-months, control era+3.20%Same metric, 2019–2023, before any bitcoin ETF · n=24 months

Charlie's readThis is the same measurement before any bitcoin ETF existed (2019–2023): bitcoin gained about 3.20% in the final three days of big stock rallies. This row is the thesis-killer test — if the pull already existed back then, ETFs cannot be the cause of it now.

Era gap, t-statistic-1.25Welch test, post vs pre, big up-months · below |2| means indistinguishable

Charlie's readA t-statistic measures how confidently we can say two groups differ; below about 2 we cannot tell them apart. Tonight's reading is -1.25 — nowhere near 2, so the two eras look the same on this test — the rebalancing story fails its hardest test tonight.

Big down-months-3.93%ETF era, n=3 · control era +0.86% (n=13), t -2.15

Charlie's readThe mirror test: in big losing months for stocks, bitcoin lost about 3.93% at month-end in the ETF era, versus +0.86% before it (t -2.15). The two eras genuinely differ here, so this side of the story is still alive.

Current month-end setup · in progress, not in any statisticThe S&P 500 is running a -0.46% loss so far in 2026-09 (price observed 2026-09-18), with roughly 7 trading days left. It needs 2.54% more to cross the ±3% 'big month' line. If it crosses, the last three trading days are the window this page watches: post-ETF big months have averaged -3.93% for bitcoin in that window. This month is not in any statistic until it closes.

The pull, measured both eras

A big month is a ±3% S&P 500 move, close to close. The pull is bitcoin's return over the last three equity trading days of that month. If ETFs created a rebalancing gravity, the pull should be an ETF-era invention — visibly stronger after January 2024 than before it.

Mean bitcoin return in the last 3 trading days, by month type and era

92 complete months, 2019-01 to 2026-08 · equities observed 2026-09-18 · bitcoin observed 2026-09-19

Takeaway: read the post-ETF bars against the control bars, not against zero. A pull that predates the ETFs cannot be caused by them.

Every complete month: S&P 500 month return vs bitcoin month-end return

Same window · vertical lines mark the ±3% big-month threshold · red months are the ETF era

Takeaway: the scatter, not the average, shows how thin the big-month samples are — each red point beyond the lines is one observation, not a pattern.

The evidence, month by month

Every big month of the ETF era, with both numbers. This is the entire sample behind the post-2024 averages — judge its size before its sign.

Month	S&P 500 month return	Bitcoin, last 3 trading days
2024-02	+5.17%	+12.24%
2024-03	+3.10%	+1.12%
2024-04	-4.16%	-5.96%
2024-05	+4.80%	-1.18%
2024-06	+3.47%	-2.40%
2024-11	+5.73%	+4.68%
2025-03	-5.75%	-5.01%
2025-05	+6.15%	-4.58%
2025-06	+4.96%	-0.21%
2025-09	+3.53%	+4.59%
2026-03	-5.09%	-0.81%
2026-04	+10.42%	-1.37%
2026-05	+5.15%	-3.24%

Verdict · 19 September 2026undecided — the evidence is mixed: big up-months are statistically indistinguishable across eras on this test; big down-months show a stronger month-end pull in the ETF era (t -2.15). With only 10 post-ETF up-months and 3 post-ETF down-months the honest call is to keep watching, not to declare in either direction.

The contradictionThe strongest evidence against the thesis is the control era itself: before any bitcoin ETF existed, big up-months already showed a mean month-end bitcoin return of +3.20% (n=24). Whatever drives that — risk appetite, monthly flows, calendar effects — predates the ETFs. For the rebalancing story to hold, the post-2024 reading must be clearly larger; tonight the difference carries a t-statistic of -1.25, and |t| below roughly 2 means the two eras are statistically indistinguishable on this test.

How this could be wrongThis test could be wrong because 10 and 3 big ETF-era months are tiny samples, a single month can move either mean materially, and the ±3% threshold is a choice — other cutoffs give other answers. Bitcoin also trades around the clock while equities close, so the shared three-day window is an alignment choice, not a law. The window covered is 2019-01 to 2026-08; the claim should be re-run monthly and abandoned if the eras stay indistinguishable.

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: in big up-months since ETFs launched, bitcoin has averaged +0.97% in the final three trading days, versus +3.20% before ETFs — a gap that scores t -1.25, and anything below about 2 means 'cannot tell apart'. One bucket passes the page's own bar and the other does not — the evidence is genuinely mixed. This month the S&P 500 is at -0.46% so far — 2.54% short of the ±3% line, with about 7 trading days to get there. What Charlie watches next: whether this month closes beyond the ±3% line, and if it does, bitcoin's last three trading days against the averages above. What would prove the thesis wrong for good: the ETF-era and pre-ETF numbers staying statistically indistinguishable as the sample grows.

For the statistically curious

The full plumbing — exact data definitions, test 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

S&P 500 is the daily adjusted close of the index, observed 2026-09-18. Bitcoin is the daily USD close, observed 2026-09-19.
A big month moves ±3% or more close to close. The month-end window is the last three S&P 500 trading days of the month; bitcoin's return is measured over those same three trading days (bitcoin trades continuously, so the window is set by the equity calendar).
The control era is every complete month from 2019-01 to 2026-08 before 2024-01; the ETF era is every complete month after. The current month is excluded from all statistics until it closes.
Era comparisons use the Welch t-test (unequal variances). A |t| below roughly 2 means the two eras cannot be told apart on this metric. Calm-month baselines: +0.75% pre-ETF, -1.13% post-ETF.
Big-month counts are small — 10 up and 3 down months in the ETF era — so every average here is one or two months away from moving materially. Treat effect sizes as provisional.
Missing data stays missing; no month is filled in and no return is assumed zero.