Funding

A perpetual contract never expires, which removes the mechanism that normally forces a futures price back to spot. Funding replaces it: a recurring payment between the two sides of the market, sized so that whichever side is crowded pays the other to be there. Funding is paid between traders. The venue takes no share of it — every unit paid by one side is received by the other, and settlement is rejected if those two totals do not match exactly.

​

Direction

The effect is a continuous drag on the crowded side. A long position held through a sustained premium bleeds cost even while the price is flat — which is the point. Without it, leverage on the popular side accumulates until it unwinds violently instead of gradually.

​

What you pay

$$\text{funding payment} = \text{position size} \times \text{mark price} \times \text{funding rate}$$ Position value uses the mark price at settlement, not the index and not the last trade. Mark price is the harder number to push, and using it means the size of a funding payment cannot be moved by someone printing a trade in the seconds before the hour.

​

Building the rate

Three layers of averaging sit between the order book and the rate you pay. Each one closes a different attack.

​

The premium index

The premium measures how far the contract trades from the index — but not by comparing mid prices, which would be trivially pushed by a single quote on each side. Instead it uses depth-weighted prices: the average price you would actually get if you spent a defined amount of capital sweeping the book. That amount, the impact notional, is derived from the market’s maximum leverage, so a market where large positions are permitted is probed to a correspondingly greater depth. $$P = \frac{\max(0,\; \text{DW bid} - \text{index}) \;-\; \max(0,\; \text{index} - \text{DW ask})}{\text{index}}$$ The $\max(0,\cdot)$ on each side means the premium is zero whenever the index sits inside the depth-weighted spread. Only a book that is genuinely displaced in one direction registers at all. To move this, you cannot post one order — you have to hold real depth away from the index, which anyone can trade against.

​

Sampling and averaging

Premium is computed every block, then reduced to one value every five seconds by taking the median of that window’s blocks. Medians rather than means, again: a single anomalous block cannot carry the sample. The rate for a settlement period is the arithmetic mean of those five-second samples across the whole period — 720 of them in an hour. Sampling has validity conditions. A block’s premium is skipped when the book is crossed, when either side has three levels or fewer, or when the index price is more than ten seconds stale. Those are broken-market conditions, and feeding them into a funding rate would let a malfunction become a charge. Funding is skipped only when fewer than 20% of the expected samples arrived — for an hourly market, fewer than 144 of the 720. Short of that the period is charged on whatever samples were collected. The threshold is deliberately low: it exists to catch a period that was barely measured at all, not to excuse one that was measured imperfectly.

​

The interest component and clamps

$$F_{8h} = \bar{P} + \operatorname{clamp}\!\left(I - \bar{P},\; -0.05\%,\; +0.05\%\right)$$ The rate is quoted on an eight-hour basis, and a settlement charges the fraction of it that the market’s interval represents: $$F = F_{8h} \times \frac{\text{interval seconds}}{28{,}800}$$ $$F_{\text{final}} = \operatorname{clamp}\!\left(F,\; F_{\min},\; F_{\max}\right)$$ The interval is a per-market setting, and only six values are permitted: 30 minutes, 1, 2, 4, 8, or 24 hours. An hourly market therefore charges an eighth of the eight-hour rate each settlement; an eight-hour market charges all of it. This is the step most often got wrong when estimating carry. A quoted rate of 0.01% on an hourly market is 0.00125% per settlement, not 0.01% — and over a day it is 0.03%, not 0.24%. The interest component reflects the cost of carry between the two currencies of the pair. It is a per-market constant, published as default_funding_rate_ppm on the eight-hour basis. It is the figure to reach for when estimating what it costs simply to hold a position in a flat market: with the premium at zero, the rate settles at the interest constant, scaled by the interval. Read it from the contract specification for your market rather than assuming a value. The clamp around it bounds how much it can pull the rate away from what the premium actually measured — so in a strongly trending market the premium dominates, and in a flat market the interest term sets a small baseline. Finally the result is clamped to per-market limits, configured at listing:

​

Settlement

Funding settles on the first block of each period — typically hourly, on the hour. The period length is configurable per market. Settlement is atomic: it applies to every position in the market in one execution step, or it does not apply at all. It runs as a Clearinghouse stage inside block execution, not as an external sweep, which is why there is no window during which some accounts have been charged and others have not. After each settlement the totals are checked. If what longs paid does not equal what shorts received, the transaction is rolled back and an alert is raised rather than leaving an imbalanced ledger.

​

What this means in practice

• Check the rate before holding through a settlement. The current rate and the countdown to the next settlement are shown on the trading screen and available through the API.
• A high rate is information. Persistently expensive funding means the market is crowded on your side, which is also the condition in which forced unwinds happen.
• Funding is a cost, not a fee. It goes to other traders, and it is separate from trading fees, which go to the protocol.

​

Reading the live values

Every figure on this page is on-chain per-market configuration, not a constant of the protocol. The parameters that drive it: They are published in each contract specification and readable from the chain. An integration that hard-codes them will eventually compute a number the network is not applying — see Developers.

​

Where to go next