# Funding rate

Also written funding, funding rates.

*https://cryptomarkets.tools/glossary/funding-rate · next to Backtesting & Research Libraries*

**Definition:** A periodic payment between the two sides of a perpetual swap that holds its price near the index. Longs pay shorts when the rate is positive, shorts pay longs when it is negative, and the venue takes no cut. It is quoted per settlement interval, not per year, and the venue picks the interval — eight-hourly is the common convention and several settle hourly, so two quoted rates can be the same word and a different quantity.

Every perpetual swap has one. It is the mechanism that makes the instrument work, and it is the
number most often quoted without the one piece of information that makes it mean anything.

## How it works

A perpetual never expires, so nothing forces its price to converge on spot. Funding is what does
the forcing. At each settlement instant the venue computes a rate from a premium component — how
far the contract has traded above or below its index — plus a fixed interest component, then moves
cash between the two sides of every open position. Positive rate, longs pay shorts. Negative rate,
shorts pay longs.

**It is not a fee.** Binance's documentation states that it charges nothing on funding payments and
that they are transferred directly between traders holding opposing positions. OKX says the
platform only facilitates the exchange of funds between traders and does not charge a service fee
under the mechanism. The distinction has a practical edge: funding is not in the fee schedule, is
not discounted by a volume tier, and does not appear in any maker-taker calculation. A cost model
that gets fees exactly right can still be missing this cash flow completely.

**It is charged on notional at an instant, not accrued across the period.** OKX computes the
payment as position value times the funding rate, and states that a position closed before the
assessment is exempt from paying or collecting. Position value is itself venue arithmetic:
contracts times contract size times multiplier times mark price for a USDT-margined contract, the
same expression divided by mark price for a crypto-margined one. Hyperliquid converts position size
at the spot oracle price rather than the mark price, and says so in as many words.

## The interval is half the number

A funding rate is a rate per settlement, and venues do not settle on the same clock.

- **OKX** settles at 00:00, 08:00 and 16:00 UTC by default, and documents contracts on one-, two-
  and four-hour schedules as well.
- **Binance** uses the same eight-hour default, with an interest component of 0.03% daily, which is
  0.01% per interval. Since 2 May 2025 settlement switches to hourly automatically when the rate
  reaches its cap or floor — so the interval is not even fixed per contract.
- **Hyperliquid** pays every hour, at one eighth of the computed eight-hour rate, capped at 4% per
  hour.
- **dYdX** runs a funding tick every hour by default and computes the rate as the premium component
  divided by eight plus an interest component, which is 0% on cross markets and 0.125 basis points
  per hour on isolated ones.

Two of those four divide an eight-hour figure to get an hourly one, which means a number labelled
"the funding rate" may be the rate that will be paid this interval or the eight-hour rate it was
derived from. Annualising without the interval is not approximate, it is wrong by the ratio of the
two intervals: 0.01% per eight hours is 1,095 payments a year and 10.95% simple, while the same
0.01% per hour is 8,760 payments and 87.6%.

Compounding is the second silent convention. Hyperliquid's documentation gives its fixed interest
component as 0.01% every eight hours and calls that 11.6% APR paid to shorts — the compounded
figure. The simple one is 10.95%. Both are correct descriptions of the same rate, and they differ
by more than the spread on a lot of carry trades.

## Why it matters here

A backtest of a perpetuals strategy that charges no funding is not a pessimistic backtest. It is a
measurement of a different instrument — the price leg alone, with one of the position's two cash
flows deleted. [What a crypto backtest silently assumes](https://cryptomarkets.tools/guides/what-a-crypto-backtest-assumes)
works through what each engine does about it, venue schedule by venue schedule; the summary for
this page is a count.

All eleven cards in [backtesting and research libraries](https://cryptomarkets.tools/categories/backtesting-libraries) mention
funding, which is not the same as charging it. Four record a model.
[NautilusTrader](https://cryptomarkets.tools/tools/nautilus-trader) settles it at the venue's own boundaries, from funding
records you supply. [Freqtrade](https://cryptomarkets.tools/tools/freqtrade) downloads mark and funding-rate candles itself in
futures mode, which makes it the only one of the four that fetches the data.
[LEAN](https://cryptomarkets.tools/tools/lean) applies rate times position notional on one eight-hour schedule to every crypto
venue it supports, and charges nothing at all when no rate feed is loaded.
[QuantConnect Cloud](https://cryptomarkets.tools/tools/quantconnect-cloud) is hosted LEAN with those rate datasets already
attached. The other seven — backtesting.py, Barter, hftbacktest, [Jesse](https://cryptomarkets.tools/tools/jesse),
Lumibot, [VectorBT](https://cryptomarkets.tools/tools/vectorbt) and VectorBT PRO — charge nothing, and in several of them the
word does not appear in the package.

On the data side the word is everywhere: 65 of the 92 cards here mention funding, across all six
categories. Three things are worth checking on any of them before the series is used as a number
rather than as a shape.

**Which interval the series is in**, and whether the vendor normalised. A provider that folds an
hourly venue and an eight-hourly venue into one column without saying so has produced a chart, not
a dataset. [Coinalyze](https://cryptomarkets.tools/tools/coinalyze) is explicit about the distinction its listing keeps,
carrying current, predicted and historical funding as separate endpoints.

**Predicted or realised.** Several cards here sell both. The predicted rate is the venue's running
estimate for the interval that has not settled yet, and it moves until it does; charging it as if
it were the settled rate makes a backtest slightly clairvoyant.

**Whose clock.** A funding timestamp is the venue's settlement instant, and an engine with a
hard-coded eight-hour schedule will line a dYdX or Hyperliquid series up against boundaries that
venue does not have. The rate is right and the payment lands in the wrong place, which is the kind
of error that survives every sanity check on the data itself.

The other half of the perpetuals pair is [open interest](https://cryptomarkets.tools/glossary/open-interest), which has the
same problem in a different form: a number whose units the vendor chose.

## Where you will meet this

- [NautilusTrader](https://cryptomarkets.tools/tools/nautilus-trader.md)
- [Freqtrade](https://cryptomarkets.tools/tools/freqtrade.md)
- [LEAN](https://cryptomarkets.tools/tools/lean.md)
- [QuantConnect Cloud](https://cryptomarkets.tools/tools/quantconnect-cloud.md)
- [Jesse](https://cryptomarkets.tools/tools/jesse.md)
- [VectorBT](https://cryptomarkets.tools/tools/vectorbt.md)
- [Coinalyze](https://cryptomarkets.tools/tools/coinalyze.md)
- [3Commas](https://cryptomarkets.tools/tools/3commas.md)
- [AGGR](https://cryptomarkets.tools/tools/aggr.md)
- [Amberdata Market Data API](https://cryptomarkets.tools/tools/amberdata.md)
- [Artemis](https://cryptomarkets.tools/tools/artemis.md)
- [Backtesting.py](https://cryptomarkets.tools/tools/backtesting-py.md)

## FAQ

### Is the funding rate a fee paid to the exchange?

No. Both Binance and OKX state in their own documentation that the payment moves directly between traders holding opposing positions and that the venue charges nothing for it. It is not in the fee schedule, it is not reduced by a VIP tier, and a cost model that gets maker and taker fees exactly right can still be missing funding entirely.

### How do I annualise a funding rate?

Multiply by the number of settlements in a year, which means knowing the interval first. A rate of 0.01% settled every eight hours is 1,095 payments a year, or 10.95% simple. The same 0.01% settled hourly is 8,760 payments, or 87.6%. A rate with no interval attached cannot be annualised, compared across venues or charged in a backtest.

### Which crypto backtesting engines charge funding?

Four of the eleven cards in this catalogue's backtesting category record a funding model — NautilusTrader, Freqtrade, LEAN and QuantConnect Cloud — and only Freqtrade fetches the rate history for you. The other seven charge nothing at all, in several cases because the word does not appear in the package.

### Does a position pay funding if it is closed before settlement?

Not on the venues cited here. OKX states that a position closed before the funding assessment is exempt from paying or collecting it. Funding is charged on the notional open at an instant, not accrued across the interval, which is why a strategy that never holds through a settlement boundary can ignore it and one that holds for days cannot.

## Sources

1. [Perpetual futures funding fee mechanism](https://www.okx.com/en-us/help/iv-introduction-to-perpetual-swap-funding-fee) — OKX, 2026-08-27
2. [Introduction to Binance Futures Funding Rates](https://www.binance.com/en/support/faq/introduction-to-binance-futures-funding-rates-360033525031) — Binance, 2026-03-06
3. [Funding](https://hyperliquid.gitbook.io/hyperliquid-docs/trading/funding) — Hyperliquid, read 2026-09-21
4. [Funding](https://docs.dydx.xyz/concepts/trading/funding) — dYdX, read 2026-09-21

*Last updated 2026-09-21. A reference page, corrected in place — not a dated post.*