“When is funding highest” is a reasonable thing to ask. Group 7,372 BTC settlements from six and a half years by calendar month and you do get a ranking: February on top, averaging 2.0511% per February; June at the bottom, 0.4494%.
The problem is that February’s standard deviation is 2.7710%, larger than the mean itself.
A number with a mean of 2.05 and a standard deviation of 2.77 cannot support any decision.
The ranking first
BTC funding accumulated in the same calendar month, across years:
| Month | Years | Mean | Std dev | Worst year | Best year |
|---|---|---|---|---|---|
| Jan | 7 | 1.6187% | 1.9483 | 0.4576% | 5.9101% |
| Feb | 7 | 2.0511% | 2.7710 | −0.0634% | 7.2437% |
| Mar | 7 | 0.9892% | 1.5291 | −0.5319% | 3.1864% |
| Apr | 7 | 0.8773% | 1.6333 | −0.1772% | 4.5099% |
| May | 7 | 0.7924% | 0.7010 | 0.2313% | 2.2980% |
| Jun | 7 | 0.4494% | 0.2756 | 0.2031% | 0.8504% |
| Jul | 7 | 0.6927% | 0.6325 | −0.1741% | 1.9572% |
| Aug | 7 | 0.8074% | 0.7963 | 0.1231% | 2.4315% |
| Sep | 7 | 0.4657% | 0.3514 | 0.0974% | 1.1392% |
| Oct | 6 | 0.7411% | 0.6430 | 0.3074% | 2.0257% |
| Nov | 6 | 1.1121% | 0.8809 | −0.1737% | 2.1411% |
| Dec | 6 | 1.1293% | 0.6845 | 0.4236% | 2.1214% |
81 complete months in total. By mean, the top three are February, January and December; the bottom three are June, September and July. It looks like a story: hot at the turn of the year, cold in summer.
You can only tell that story with the standard deviation column covered up.
Drop 2021 and the ranking changes
February’s 7.2437% came from 2021. So did January’s 5.9101%. That was the only year in six in which BTC funding cleared 30% for the full year.
Remove 2021 entirely, re-average the remaining six years, and the board reshuffles:
| Month | Rank with 2021 | Rank without 2021 | Mean without |
|---|---|---|---|
| Apr | 6th | 12th | 0.2718% |
| Jul | 10th | 5th | 0.8371% |
| Jan | 2nd | 4th | 0.9035% |
| Mar | 5th | 7th | 0.6231% |
| Dec | 3rd | 2nd | 1.1724% |
April goes from mid-table straight to last; July climbs from third-from-bottom to fifth. A “pattern” that inverts when you remove one year from the sample is not a pattern.
The one genuinely stable month is June: lowest 0.2031%, highest 0.8504% across seven years, standard deviation 0.2756 — the smallest of the twelve and the only month whose standard deviation is clearly below its mean. But it is stable on the “always small” side, which is not information you can trade.
The negative-share column is more solid
Same data, different lens: what share of settlements in each calendar month were negative.
- Highest: April 26.19%, March 22.89%, September 17.13%
- Lowest: December 4.48%, January 5.53%, July 8.76%
This holds up better than the mean, because it counts occurrences and is not dragged around by single-period extremes. Even so, a big chunk of April’s 26.19% comes from one month in 2026 — that April had 64.44% of its settlements negative.
While we are on 2026: February −0.0634%, March −0.0925%, April −0.1772%, three consecutive months negative on a monthly basis. Then July through September came back at 0.5656% / 0.6183% / 0.3973%. If a single year can swing like that internally, a cross-year calendar pattern has even less to stand on.
What about day of week, or time of day
If the month does not work, try a finer grain.
Grouped by UTC weekday, the BTC per-settlement mean runs from 0.010003% on Friday to 0.011764% on Tuesday — a total spread under 0.0018 percentage points. Negative share ranges 13.20% to 15.38%. There is no weekend effect.
Grouped by the three daily settlement times:
| Settlement (UTC) | Mean per period | Negative share |
|---|---|---|
| 00:00 | 0.010147% | 15.46% |
| 08:00 | 0.010785% | 13.35% |
| 16:00 | 0.011259% | 13.19% |
The 16:00 slot really is the highest, and it beats 00:00 by 0.0011 percentage points. Suppose over a year you somehow collected only the 16:00 period, every single day (you cannot, because the position has to stay on) — the extra would be something like 0.4 percentage points, which does not cover one round trip of fees and slippage.
What this table is actually good for
Not timing. It answers a different question: can one month’s reading stand for anything.
It cannot. The same month of August was 2.4315% in 2020 and 0.1231% in 2024, a factor of nearly twenty. Extrapolating a year from any single month produces an error large enough to make the exercise pointless — which is the same reason the holding-period piece keeps insisting that one year and five years are not the same thing.
The same logic applies one level down. Waiting for a higher rate before entering is something we tested against 790 settlements, and it does not hold. The monthly layer agrees.
So what does set the level? Not the calendar — leverage demand. Why funding is usually positive covers where the money comes from. Briefly: February tops the board because longs were unusually crowded in February 2021, not because of anything to do with February.
Data and conventions
From the per-settlement records in our FundingRateTicks table for BTCUSDT: 7,372 settlements, 2020-01-01 to 2026-09-23 (UTC). Monthly totals are summed period by period, not compounded; the standard deviation is the sample standard deviation of the same calendar month across different years.
October to December have only six years of sample because 2026 has not reached them. The September 2026 cell holds only 67 settlements and is incomplete. Both quietly bias the corresponding monthly means a little. Not by much, but it should be said.
Year-by-year performance is whatever /performance/ says. A historical monthly distribution implies nothing about any future month.