You see a funding rate — say 0.01% — and the first instinct is to ask whether that is high or low.
There is a precise answer, but it comes in two steps. Step one is where it ranks across the market. Step two is how many hours its interval is. Skip step two and step one is off by a factor of two.
Bucketing the 608,656 settlements recorded across 450 Binance USD-M contracts between 1 January and 24 September 2026, 03:00 UTC by settlement interval, and sorting each bucket:
| Percentile | 8-hour (462,521 periods) | 4-hour (141,354 periods) | 1-hour (4,774 periods) |
|---|---|---|---|
| 50th | 0.005000% | 0.005000% | −0.047244% |
| 75th | 0.005000% | 0.005000% | −0.016582% |
| 90th | 0.010000% | 0.005000% | −0.000341% |
| 99th | 0.045119% | 0.051617% | 0.045742% |
| Arithmetic mean | −0.006791% | −0.003142% | −0.112109% |
Read the first column. If you are looking at an eight-hour contract:
- 0.005% is the median — and also the 75th percentile. More than half of all settlements are exactly that number.
- 0.01% is already the 90th percentile. It looks like an ordinary figure; it is higher than nine settlements in ten.
- To reach the top 1% you need 0.045119%.
Why the median and the 75th percentile are the same number
This is not a statistical quirk. The funding formula contains a fixed interest term, and when the premium index sits near zero — when the perp and the spot are trading at much the same price — the rate lands on that default. Most of the time, that is exactly the state of the market.
So normal in this market is not a distribution, it is a constant. The formula is taken apart in how is the funding rate calculated.
It also explains something people find odd: stare at funding all day and the number is always one of the same few. Because it is.
Step two: the same reading, twice the annualised figure
Convert that table to annualised terms (period rate x periods per day x 365) and the identical rows separate:
| Percentile | 8-hour | 4-hour | 1-hour |
|---|---|---|---|
| 50th | 5.475% | 10.950% | −413.853% |
| 75th | 5.475% | 10.950% | −145.254% |
| 90th | 10.950% | 10.950% | −2.987% |
| 99th | 49.405% | 113.041% | 400.696% |
The same 0.005% is 5.475% a year on an eight-hour contract and 10.950% on a four-hour one. The contract is not richer than its neighbour; it just collects twice a day.
Which is why comparing raw readings is meaningless. 325 of today’s 450 contracts run on a four-hour cadence, and plenty of market pages show a bare percentage with no interval attached. Exchanges do not even handle it consistently: moving from eight hours to four, OKX halves each period so the annualised rate is unchanged, while Binance does not, so the annualised rate simply doubles — and both announcements say only that the settlement frequency has been adjusted. That story is in eight hours to four.
The one-hour column needs reading on its own
The median of that third column is −0.047244%, or −413.853% annualised. That is not an arithmetic error; those contracts genuinely spend most of their time paying the other way.
The reason matters: when Binance temporarily switches a contract to hourly funding, it is managing a basis that has already gone out of control. So when a scanner shows you a very large annualised number, check first that it is not a contract from the third column — that column really can hit 400% annualised at the 99th percentile, and its median is −413%. What that same cohort did on a cumulative basis this year is in summing 2026 funding contract by contract.
One more thing worth noting: the arithmetic mean of all three buckets is negative, while the medians of the eight-hour and four-hour buckets are positive. The mean is dragged away by a few extreme negatives, so on this particular question the median is the more informative statistic.
Check it against today’s actual prints
The settlement at 00:00 UTC on 24 September 2026, for the four majors (all on eight-hour cadence):
| Contract | This period | Position in the 8-hour bucket | Annualised |
|---|---|---|---|
| ETHUSDT | 0.005263% | just past the median | 5.76% |
| SOLUSDT | 0.001129% | below the median | 1.24% |
| BTCUSDT | 0.000132% | far below the median | 0.14% |
| BNBUSDT | 0.000000% | exactly zero | 0.00% |
That is what the majors actually look like today: one of the four barely past the median, and BTC’s period annualising to under 0.15%. Single-period readings swing a lot, which is why you should not annualise one of them — the shorter the sample, the more the annualised figure reads like a fortune cookie. BTC’s 799 periods year to date sum to 2.1419%, and that is the comparable number.
A lookup table you can use directly
For a reading on an eight-hour contract:
| Reading | Position | Annualised | How to read it |
|---|---|---|---|
| 0% or below | below roughly 40% of periods | 0% or below | you are paying, not collecting |
| 0.005% | median | 5.475% | the market’s default state |
| 0.01% | 90th percentile | 10.950% | already above nine periods in ten |
| 0.045% | 99th percentile | 49.3% | rare, and rarely on a major |
| 0.30% or more | 66 of 608,656 periods | — | happened 66 times market-wide this year |
That last row is worth remembering on its own: it is roughly the size of one round trip in fees, meaning only 66 periods all year across the whole market would have let you get in, collect one period, and get out without a loss. The arithmetic is in holding only across the settlement.
Two notes on using this
One: these are cross-sectional percentiles, not time-series ones. Every percentile above pools all periods from all 450 contracts, and the overwhelming majority of those periods come from small contracts. If you only trade BTC, its own historical distribution is a different table.
Two: high is not the same as good. Contracts in the 99th percentile column usually have thin spot books and wide spreads, and a high-funding state delivers only about 45% of itself on average (measured in the week after a funding scanner crowns a leader). Use the percentiles as coordinates, not as a target.
Our own year-by-year record, worst year included, is on performance; what a funding rate is and where the money comes from is in funding rate basics.
Data: per-settlement funding rates from our own collection database (sourced from Binance’s public USD-M endpoints), window 2026-01-01 00:00 UTC to 2026-09-24 03:00 UTC, 608,656 periods across 450 symbols — 462,521 in the eight-hour bucket (450 symbols), 141,354 in the four-hour bucket (325), 4,774 in the one-hour bucket (95), plus 7 periods on a two-hour interval not shown. Percentiles are the empirical distribution of single-period rates, with each period of each contract counted as one observation; annualised = period rate x (24 / interval hours) x 365, summed not compounded, excluding fees and spread. Single-period annualisation is for cross-sectional comparison only and should not be read as a sustainable yield. Contract tickers are shown to document the data, not as recommendations. Past data does not indicate future returns.