Annualising one funding payment is a guess

Every perpetual venue shows a funding rate, and most show it with an annualised figure beside it. The annualised figure is the number people act on. It is also the one number on that screen that is not a measurement.

An annualised rate is a single observed interval multiplied by a number derived from the length of a year. The multiplier assumes the observed rate repeats unchanged for every interval in that year. Funding rates do not repeat unchanged. So the annualised figure describes a rate you have seen once, not a rate you will be paid.

This is worth stating plainly because the arithmetic behind it is simple enough to audit, and almost nobody audits it.

The multiplier is a claim, not a conversion

Take the rate printed on a dashboard and divide it by the venue's funding interval. That gives one interval's payment. Multiplying by the number of intervals in a year is where the claim enters: 365 days divided by the interval, or 8,760 hours for an hourly one.

For an hourly venue, that multiplier is 8,760. If the screen shows +0.00125% per hour, the annualised figure reads +10.95%. That number is the product of one hour's rate and 8,760. It is not a forecast, a yield, a measurement, or a history. It is arithmetic applied to an assumption.

Two venues can print different annualised numbers for what is economically the same carry, purely because they settle on different intervals and each multiplies by its own constant. This is not a rounding difference. It is the same underlying payment expressed through two different assumptions about the future.

The interval is a venue setting, not a law of nature

The three venues whose documentation I read do not agree on the interval, and each says so plainly.

Hyperliquid: "The funding rate on Hyperliquid is paid every hour." The mechanism is unusual enough to be worth reading closely. Its documentation states that "The funding rate formula applies to 8 hour funding rate. However, funding is paid every hour at one eighth of the computed rate for each hour."

So on Hyperliquid the published formula is an 8-hour formula, and what lands in your account is one eighth of it, once an hour. A rate quoted per hour and a rate quoted per 8 hours are the same economic quantity divided by eight. Confusing them is an eight-fold error, not a small one.

dYdX: "We charge funding rates every hour on our platform. The funding rate is calculated at the end of each hour and is based on the average of premiums collected over the last 60 minutes." The premium behind it is also a per-minute object, because "At the end of each funding-sample period (default to 1 minute), the median FundingPremiumVote is taken as the sample for that period."

OKX: "The funding fee will be charged or paid out every 8 hours (00:00, 08:00, and 16:00 UTC) by default, unless specified otherwise (i.e. every 1, 2 or 4 hours)."

And OKX adds the sentence that undercuts annualisation more directly than any formula: "The funding fee settlement time and funding rate cap/floor may be adjusted in real-time according to market conditions."

An annualised figure multiplied out by 365 assumes the schedule is fixed. On at least one of these venues the schedule itself can change.

OKX normalises the interval, and that changes what you compare

OKX publishes the interval explicitly inside its formula. It divides by an interval factor: "N = the contract's funding settlement interval, in hours. Supported values are N ∈ {1, 2, 4, 8}. The (8 / N) divisor normalizes the per-period rate so that the daily-equivalent funding cost remains consistent across all supported settlement cycles."

The divisors are ÷ 1 for 8-hour, ÷ 2 for 4-hour, ÷ 4 for 2-hour, ÷ 8 for 1-hour.

This is a real design decision with a consequence for anyone comparing venues. If a venue normalises the per-period rate so the daily cost is the same across settlement cycles, then an hourly contract and an 8-hour contract on that venue are engineered to cost the same per day. The per-interval headline numbers differ by a factor of eight and mean the same thing.

A dashboard that annualises each venue by its own interval partly absorbs this. A dashboard that shows per-interval rates side by side does not. If you are comparing venues, compare the normalised daily or annual figure, not the per-interval one.

The two components, and one of them is not the market

Both components of these formulas are documented, and they behave differently.

The premium component is the market-reading term. Hyperliquid's documentation states: "The premium component fluctuates based on the difference between the perpetual contract's price and the underlying spot oracle price."

The interest component is set by the venue, not discovered from the order book. Hyperliquid: "interest rate component is predetermined at 0.01% every 8 hours, which is 0.00125% every hour, or 11.6% APR paid to short. This represents the difference in cost to borrow USD versus spot crypto."

OKX: "Interest rate = 0.01% (fixed across all settlement intervals)". dYdX: "the interest rate component for cross markets is 0% ", and "the default interest rate component for isolated markets is 0.125 bps per hour or 1 bps per 8 hours."

Read those side by side. Two venues charge shorts a positive fixed carry; one charges cross markets nothing. That is a venue pricing decision, and it sets the sign of your funding before the market has said anything at all.

The interest component is also why annualising a single sample is fragile on a specific mechanical level: the premium can sit inside the clamp, in which case the rate you are annualising is not a market reading but the venue's own constant. Hyperliquid's formula is Funding Rate (F) = Average Premium Index (P) + clamp (interest rate - Premium Index (P), -0.0005, 0.0005). When the clamp binds, the difference term is pinned at its bound and F collapses to the interest rate. You are annualising the venue's borrowing assumption.

What a sign flip does to the number

The sign of funding is the direction of the transfer, and it reverses. Hyperliquid's documentation is explicit about both directions: "If the contract's price is higher than the oracle price, the premium and hence the funding rate will be positive, and the long position will pay the short position. Conversely, if the contract's price is lower than the spot price, the funding rate will be negative, and the short position will pay the long position."

OKX states the same in its own terms, including that the venue takes no cut: "our platform only facilitates the exchange of funds between traders and doesn't charge any service fees under this mechanism."

That is the second important property: funding is a transfer, not a fee. Hyperliquid puts it plainly: "Funding is purely peer-to-peer and no fees are collected on the payments." Nobody is charging you to hold the position. The only path to a positive funding payment is for the other side of the contract to hold an offsetting position and lose.

Which is why the annualised figure has no floor. Take the Hyperliquid interest rate of 0.0001 per 8-hour interval as the magnitude. Held constant and positive for a year of 1,095 intervals, it is +10.95%. Held constant and negative, it is -10.95%. Now let the sign flip, with equal magnitude either way:

| Intervals positive | Net simple annualised | |---|---| | 100% | +10.95% | | 60% | +2.19% | | 50% | 0.00% | | 40% | -2.19% | | 0% | -10.95% |

The break-even is exactly 50%, because funding is zero-sum between the two sides. A headline that says +10.95% and a history where you were positive 40% of the time describe the same contract. The first is a forecast. The second is what happened.

On Hyperliquid, deriving from the published formula, the funding rate turns negative only when the premium index falls below -0.0005, since for any premium between -0.0005 and +0.0005 the clamp binds and F equals the interest rate. Negative funding there is not a drift. It requires the contract to trade meaningfully below the oracle.

The arithmetic trap, in one worked number

The clearest demonstration is OKX's own worked example, because the venue publishes the inputs.

OKX takes "a perpetual contract with a 1-hour settlement cycle (N = 1) and an average premium index of 0.10% in the latest settlement window" and arrives at "Funding rate = 0.05% / (8 /

1. = 0.00625% per 1-hour period."

One documented sample. Now annualise that single number two ways:

| Convention | Arithmetic | Result | |---|---|---| | Simple | 0.00625% × 8,760 | +54.75% | | Compounded | (1.0000625)^8760 − 1 | +72.89% |

An 18-point spread, from one sample, with no change in the rate that was observed. Nothing about the market differs between those two rows. Only the multiplier choice differs, and the dashboard picks for you without telling you.

The same effect is visible in Hyperliquid's own published figure. Its documentation converts 0.01% per 8 hours to "11.6% APR". Simple annualisation gives 10.95%. Compounding 1,095 daily intervals gives 11.57%, which rounds to the published figure.

That is worth internalising, because it means two platforms can display 11.6% and 10.95% for the identical rate on the identical contract, and the gap is arithmetic convention rather than a data disagreement. Before treating a difference between two venues' annualised numbers as a real finding, check which convention each uses.

The same trap, scaled by the clamp

The clamp bounds are larger than they look once annualised. Hyperliquid's clamp on the interest-minus-premium term is -0.0005 to +0.0005 on the 8-hour rate. Held at its bound for a year, 0.0005 × 3 × 365 is 54.75% APR. So the widest possible contribution from that term, sustained, is the same magnitude as the entire OKX worked example above.

The range of outcomes implied by the published formula is wide. The published parameters permit outcomes from negative to positive, at magnitudes reaching several dozen percent annualised, depending entirely on which side of the clamp the premium lands and how long it stays there. The annualised figure on your screen is a single point from that range, and it is usually the point you happened to look at.

Read the series, not the headline

Four things to do instead of reading the annualised number.

Get the interval first. It is a venue setting, it differs by venue, on some venues it varies by contract, and on at least one it can be changed in real time. Without it, the annualised figure cannot be checked.

Count intervals per year from that interval, and do the division yourself. Then check it against the screen. A discrepancy usually means the screen is using the other convention.

Read the rate's own components. On venues that publish them, subtract the interest component to see how much of the rate is the market. Where the premium sits inside the clamp for long stretches, the rate is the venue's constant and carries no positioning information.

Sum the actual payments over many intervals, and read the sign distribution rather than the mean. The mean of a sign-flipping series understates the risk in it, because the series that averaged +10.95% over a year may have spent the year in the red and recovered in a handful of intervals, or the reverse. What matters for a held position is the sum, and the proportion of intervals that ran against you.

Two more things the arithmetic does not cover

What the rate is applied to is not the same across venues. Hyperliquid's documentation is specific: "the funding payment at the end of the interval is position_size * oracle_price * funding_rate. In particular, the spot oracle price is used to convert the position size to notional value, not the mark price." OKX uses the mark price in its own position value definition. So a position's funding charge is computed on a different base on each venue, which changes the dollar amount even when the rate matches.

Holding through the assessment is a decision with a deadline. OKX: "You're obligated to pay or receive the funding fee if you hold open positions at the point of fee assessment. If you close your position before the funding fee assessment, you're exempt from paying or collecting the fee." It also notes that assessment "may take up to a minute", so a position opened seconds before the boundary can still be charged. On an hourly venue that window arrives 8,760 times a year. That is an execution cost with a timestamp, and it is priced into the rate only if the rate is being read as an average rather than as a boundary event.

What this does and does not license

It does not mean annualised funding figures are fraudulent. Most are computed correctly from a documented formula. It means a correctly computed figure answers a question nobody asked: what would this single rate be worth if it never changed.

It does not mean the rate is uninformative. The premium term is a genuine measure of where the contract trades against the index, and on the markets where the clamp is not binding it is a real signal about positioning. The error is in the multiplier, not the underlying data.

It does not generalise across venues without work. Three venues document three different intervals, three different carry baselines, and three different formulas. Anything you conclude has to be traced to the specific venue's specific documentation.

Sources

• Hyperliquid funding documentation: the peer-to-peer no-fee statement, the predetermined 0.01% per 8 hour interest component and its 11.6% APR, the hourly payment of one eighth of the 8-hour formula, the full clamp formula, the 5-second premium sampling, the negative funding condition, the 4%/hour cap, and the oracle-price basis for the payment. Fetched 5 October 2026.
• dYdX funding documentation: the hourly charge, the 60-minute premium average, the 1-minute funding-sample epoch and median vote, Funding Rate = (Premium Component / 8) + Interest Rate Component, the 0% cross-market and 0.125 bps per hour isolated interest components, and the 600% * (Initial Margin - Maintenance Margin) rate cap. Fetched 5 October 2026.
• OKX perpetual funding fee mechanism: the default 8-hour schedule with 1, 2 and 4-hour variants, the platform-facilitates-transfers statement, the N ∈ {1, 2, 4, 8} normalisation divisor and its per-cycle effect, the fixed 0.01% interest rate, the ±0.05% clamp, the real-time adjustment of settlement time and cap/floor, the exemption for closing before assessment, the mark-price position value, and the worked 1-hour-cycle example producing 0.00625% per hour. Page states last updated 27 August 2026. Fetched 5 October 2026.
• Binance's funding rate documentation did not load for this piece. Both the support article and the developer endpoint returned an empty 202 response. No Binance figure, interval or formula is quoted or paraphrased here.

Every arithmetic result in this piece is computed from a published rate and a published interval, and the two figures in the first table are the only conventions compared.

Cryptocurrency trading, leveraged perpetual futures, and automated algorithmic strategies carry significant risk of rapid and total financial loss. Never risk funds you cannot afford to lose completely.