How do perpetual futures work?
A perpetual future (a “perp”) is a futures contract with no expiry date. You can hold it as long as you like, with leverage, long or short. A normal future is pulled towards the spot price because it settles at a fixed date. A perp never settles, so exchanges use a different tool to keep its price close to spot: funding.
Every few hours, the two sides of every perp swap a small payment. When the perp trades above the spot index (more people want to be long), the funding rate is positive and longs pay shorts. When it trades below, the rate is negative and shorts pay longs. The exchange takes nothing; it passes the money from one side to the other. The payment is the rate times the size of your position: at a 0.01% rate, a $10,000 position pays or receives $1.
On Binance, Bybit and OKX funding is paid every 8 hours (some markets every 4 or 1). On Hyperliquid and Lighter it is paid every hour. That one small number, paid over and over, is often the largest cost of holding a perp. We added it up over almost seven years.
Longs pay most of the time
Of 7,488 funding payments on Binance's BTCUSDT perpetual from 1 Jan 2020 to 31 Oct 2026, 86% had a positive rate: the long side paid. For ETHUSDT it was 86.2%. One reason is built into the formula: when the perp trades close to spot, the rate settles at a default of 0.01% per 8 hours (the interest part), and 33% of all BTC payments were at exactly that rate. Crypto traders also tend to be long more often than short.
That default rate alone is about 0.03% a day, or 10.95% a year. The average over the whole period was higher: the rates add up to 11.6% a year for BTC and 13.7% for ETH. The largest single payment was 0.3% (12 Feb 2020) and the most negative -0.3% (13 Mar 2020, the March 2020 crash).
The cost of holding for a week, a month, a year
| Held for | Median | Middle 80% | Best / worst | Windows where the long paid |
|---|---|---|---|---|
| 1 week | 0.13% · $13.20 | 0.01% to 0.56% | -1% / 2.75% | 91% |
| 1 month (30 days) | 0.55% · $54.50 | 0.11% to 2.43% | -1.31% / 7.71% | 93% |
| 1 year | 8.22% · $822 | 4.08% to 30.16% | 3.21% / 33.56% | 100% |
A typical week cost a $10,000 long $13.20. A typical month cost $54.50, but one month in ten cost more than $243 and the worst 30 days cost $771. Over a full year the median was $822, or 8.2% of the position, and in every single one-year window in the data the long paid: the cheapest year still cost 3.2%, the most expensive 33.6%.
Compare that with the cost of getting in. A taker fee of 0.05% to open and 0.05% to close is 0.10% in total, $10 on $10,000. The median month of funding cost about 5.5 times that; the median year about 82 times. For a position held more than a few days, funding is the main cost, and fee comparisons miss it. The same windows for ETH: median week 0.14%, month 0.56%, year 8.3%.
Year by year
Average funding rate each month, turned into a yearly percentage. Positive = longs pay. Binance BTCUSDT and ETHUSDT (8-hourly), 1 Jan 2020 to 31 Oct 2026; Hyperliquid BTC (hourly) from 12 May 2023.
Show the numbers
| Month | Binance BTC | Binance ETH |
|---|---|---|
| 2020-01 | 21.4% | 24.7% |
| 2020-02 | 57.2% | 77.3% |
| 2020-03 | -6.3% | 11.8% |
| 2020-04 | 0.6% | 9.8% |
| 2020-05 | 11.4% | 15.3% |
| 2020-06 | 9.1% | 12% |
| 2020-07 | 23% | 30.8% |
| 2020-08 | 28.6% | 47.4% |
| 2020-09 | 8.1% | 11% |
| 2020-10 | 6.3% | 15.5% |
| 2020-11 | 23.6% | 41.5% |
| 2020-12 | 25% | 34% |
| 2021-01 | 69.6% | 97.7% |
| 2021-02 | 94.4% | 111.1% |
| 2021-03 | 37.5% | 51.3% |
| 2021-04 | 54.9% | 66% |
| 2021-05 | 27.1% | 26.6% |
| 2021-06 | 2.9% | 6.1% |
| 2021-07 | -2% | 6.2% |
| 2021-08 | 13.8% | 17.9% |
| 2021-09 | 13.9% | 19% |
| 2021-10 | 23.9% | 26.6% |
| 2021-11 | 26.1% | 18.2% |
| 2021-12 | 10.8% | 9.6% |
| 2022-01 | 5.9% | 3.4% |
| 2022-02 | 3.1% | 2.9% |
| 2022-03 | 6.1% | 7.7% |
| 2022-04 | 3% | 0.4% |
| 2022-05 | 5.5% | 3% |
| 2022-06 | 3.2% | -0.5% |
| 2022-07 | 7.3% | 4.3% |
| 2022-08 | 1.9% | 0.5% |
| 2022-09 | 4.4% | -18.9% |
| 2022-10 | 6.3% | 3.7% |
| 2022-11 | -2.1% | -2.9% |
| 2022-12 | 5% | 5.2% |
| 2023-01 | 9.4% | 11.6% |
| 2023-02 | 9.1% | 8.9% |
| 2023-03 | 6.2% | 6.7% |
| 2023-04 | 5.5% | 6% |
| 2023-05 | 6% | 5% |
| 2023-06 | 7.3% | 5.4% |
| 2023-07 | 6.6% | 6.8% |
| 2023-08 | 6.3% | 4.2% |
| 2023-09 | 1.2% | 2.1% |
| 2023-10 | 4.4% | 6.7% |
| 2023-11 | 12.4% | 13.9% |
| 2023-12 | 19.9% | 21.6% |
| 2024-01 | 13.2% | 13.6% |
| 2024-02 | 16.5% | 16.6% |
| 2024-03 | 37% | 37.9% |
| 2024-04 | 10.2% | 10.2% |
| 2024-05 | 7.5% | 9.7% |
| 2024-06 | 10.3% | 11% |
| 2024-07 | 7% | 8.1% |
| 2024-08 | 1.4% | 1.7% |
| 2024-09 | 1.9% | 4.6% |
| 2024-10 | 8% | 9.5% |
| 2024-11 | 15.8% | 18.5% |
| 2024-12 | 14.2% | 14.1% |
| 2025-01 | 8.5% | 9.5% |
| 2025-02 | 5.1% | 3.1% |
| 2025-03 | 2.1% | 2.4% |
| 2025-04 | 2.6% | 1.9% |
| 2025-05 | 5.1% | 6.1% |
| 2025-06 | 2.9% | 4.8% |
| 2025-07 | 8.6% | 8.3% |
| 2025-08 | 7.1% | 6.3% |
| 2025-09 | 5.3% | 3.4% |
| 2025-10 | 3.6% | 3.5% |
| 2025-11 | 5.4% | 5.2% |
| 2025-12 | 5% | 4.2% |
| 2026-01 | 5.4% | 4.8% |
| 2026-02 | -0.8% | -4% |
| 2026-03 | -1.1% | -1.1% |
| 2026-04 | -2.2% | -1.6% |
| 2026-05 | 2.7% | 3.4% |
| 2026-06 | 2.5% | 0.6% |
| 2026-07 | 6.7% | 4.1% |
| 2026-08 | 7.3% | 5.7% |
| 2026-09 | 5.5% | 4.9% |
| 2026-10 | 4.8% | 3.3% |
| Year | Binance BTC | Binance ETH | Hyperliquid BTC | Hyperliquid ETH |
|---|---|---|---|---|
| 2020 | 17.2% | 27.5% | – | – |
| 2021 | 30.6% | 37.5% | – | – |
| 2022 | 4.2% | 0.8% | – | – |
| 2023 | 7.9% | 8.3% | 8.7% | 12.9% |
| 2024 | 12% | 13% | 24.2% | 22.1% |
| 2025 | 5.1% | 4.9% | 10.6% | 8.5% |
| 2026 | 2.6% | 1.7% | 4.8% | 5.1% |
Funding follows the mood of the market. In the bull market of 2021, longs paid 30.6% of their size over the year on BTC and 37.5% on ETH, with monthly rates above 50% a year at the peaks. In 2022, a bear year, BTC funding added up to only 4.2% and ETH to 0.8%: rates spent long stretches near zero or below. The years since have been in between, and 2026 has been one of the cheapest years to hold a long. Hyperliquid (hourly, from mid-2023) charged more than Binance in each full year: 24.2% against 12% for BTC in 2024. A recent check against OKX gave a similar picture to Binance: from 6 Aug 2026 to 10 Nov 2026, OKX BTC funding averaged 6.2% a year.
What this means if you hold a perp
- Count funding in the cost of a trade. For a long held a month, the typical funding bill was several times the trading fee.
- The sign matters. Shorts received the same amounts longs paid: a short held through 2021 was paid 30.6% of its size, before any price move.
- The rate changes every period. Today's rate is a guide for the next few hours, not for next month. The figures above show how wide the range is.
- Different exchanges charge different funding on the same coin at the same time. The live funding board shows today's rates side by side, and the calculator adds funding for your hold time.
- Spot has no funding. If you plan to hold for months without leverage, buying the coin avoids it (but pays spot fees and spread instead).
This is history, not a forecast. Rates in the next year could be higher or lower than any year shown here, and none of this is advice to trade.
Method
Binance BTCUSDT and ETHUSDT funding comes from Binance's public data archive (data.binance.vision), monthly files from January 2020 to October 2026: every 8-hour funding payment, 7,488 rows per coin with no gaps. The archive has no files for 2019 (Binance's USDT perpetual launched in September 2019), and Binance's live API refuses requests from our research server's location, so the study starts on 1 Jan 2020.
Hyperliquid BTC and ETH funding comes from its public fundingHistory API, one payment per hour from 12 May 2023 (its first record) to 10 Nov 2026. OKX's public API only returns about three months, which we show as a recent check (6 Aug 2026 to 10 Nov 2026). Bybit's funding history API refused our server's location, so Bybit is not in this study.
The cost of holding is the sum of the funding rates paid while the position is open, as a share of the position size: a long of fixed size pays the sum when it is positive and receives it when negative; a short is the mirror image. For the week, month and year figures we started a position at every funding time in the data and added up the next 7, 30 or 365 days. That gives thousands of overlapping windows; we report the median, the 10th and 90th percentiles and the extremes. This is the funding only: trading fees, slippage and price changes are separate.
Funding rates on these exchanges are calculated from the gap between the perp price and the index price plus a small fixed interest part, with caps. We did not recalculate them; we used the rates the exchanges paid.
The numbers behind every chart and table are in data.json (free to reuse with a credit to CostDepth). This is fee and cost data, not financial advice.