Delta-neutral funding, net of costs

What the board finds

Three questions, answered with numbers

1. 01

   Does it pay after all costs?

   It pays when the funding forecast for your hold is larger than the cost to open and close both legs.

   How the board computes it

   • The hold. You pick 7, 30, 90 or 365 days, from the open to the close. A short hold pays the same costs with less funding, so fewer positions pay over 7 days than over 365 days.
   • The forecast. The funding of each position is a forecast for the hold: a weighted average of the funding that the venues settled. A day counts half as much for each half-life back, so recent days count more.
   • The costs. Each leg pays the fee, the spread or the measured impact at your size, and slippage, once to open and once to close. The base case fills as taker on both legs, with 3 bp of slippage on each side.
   • The net. The funding of the hold less the costs. The board leads with the APR on capital: the net on all the capital, the margin and the reserve too, scaled to a year without compounding. Under it are the dollars of the hold and the percent on capital.
   • Left out and marked. A pair with less than 90 days of settled funding is left out. A pair that the rates now pay the other way is marked, ranks last and does not count as profitable.

   The funding forecast of each hold

   Hold	Funding forecast
   7 days	Weighted average of the last 30 days, half-life 7 days
   30 days	Weighted average of the last 90 days, half-life 14 days
   90 days	Weighted average of the last 365 days, half-life 60 days
   365 days	Weighted average of the last 365 days, half-life 60 days

2. 02

   At what size?

   Up to the largest size whose net stays above 0, measured in the order books.

   How the board computes it

   • The books. The proxy reads the order book of each market about every 10 minutes. It measures the cost of a market order at six sizes, from $1,000 to $1M.
   • The scan. The board prices 28 sizes of each leg, from $1,000 to $1M, against the funding of your hold. “Best size” has the largest net in dollars. “Profitable up to” is the largest size with a net above 0.
   • The end of the books. The scan stops at the last size that both books filled, or at the largest margin tier of a perp. A + after the size means that the net still grows there: the books were not read deeper.
   • Past the books. A size that you type past the depth that was read gets an estimated cost, and the pair ranks last. An estimate alone never makes a pair profitable.
   • The age of a read. Each pair shows the age of each book read. A read older than 15 minutes missed a round and is marked old. “Use this size” sets the size of each leg to the best size.
3. 03

   When to close?

   Close both legs together when the first of four rules applies.

   How the board computes it

   • Not before the break-even. The days of funding that pay the costs to get in and out. A close before them loses money.
   • The forecast line. Close if the forecast falls below the cost to close both legs at today's book and your size, spread over the hold as a rate a year. Below it, the funding of the hold earns less than the close costs.
   • The flip rule. Close if the funding pays the other way for a number of hours in a row: 8 hours on a 7-day hold, 48 hours on a 30-day hold. For 90 and 365 days, the forecast line alone decides. The lengths come from a backtest on the settled funding of 55 perps and 81 perp-perp spreads, 2023-10 to 2026-10. The table gives the share of positions that still earned the cost to exit over the next hold.
   • The exit gap. The gap between the mid prices of the two legs at entry. A close at that gap or lower gives back nothing that the entry locked in. It is the best time to close, not a reason to close on its own.

   The flip rule of each hold

   Hold	Close after	After the flip, earned the exit	All held hours
   7 days	8 hours the other way in a row	27%	65%
   30 days	48 hours the other way in a row	43%	81%
   90 days	No number: the forecast line decides	After 72 hours: 54%	83%
   365 days	No number: the forecast line decides	After 72 hours: 73%	93%

What the numbers do not know

Forecasts fall short

A backtest on the settled funding of 55 perps on Binance, Bybit, OKX and Hyperliquid, from 2023-10 to 2026-10, checked each forecast against what was paid next. A point is one percentage point of APR.

Over all the perps, the forecast of a 7- or 30-day hold was close to what was paid. The forecast of a longer hold was too high. The 5 best forecasts of each day paid less than forecast, more so for a long hold. A perp-perp spread in the direction of its forecast also paid less than forecast.

The board states these figures and does not take them off the net: they come from one test, and the fall is true of the best picks, not of each position. The 365-day hold has only one year of start days, so its figures are the weakest.

The backtest of each funding forecast

Hold	All perps: the forecast was	5 best forecasts each day	Perp-perp spreads
7 days	0.01 points too high	Paid 13% less	Paid 0.96 points less
30 days	0.07 points too low	Paid 20% less	Paid 0.93 points less
90 days	1.09 points too high	Paid 31% less	Paid 1.19 points less
365 days	3.98 points too high	Paid 49% less	Paid 2.36 points less

Hindsight

The board picks the direction of a perp-perp pair from its funding history. So today's direction over the same past weeks is hindsight: it looks better than a trader could have done.

The funding by week of a perp-perp pair first shows what the forecast's direction paid. Each week takes the direction that the forecast gave from the days before it, and a week that turns the direction pays a round trip at today's costs. Where the board shows today's direction, it says “in hindsight”.

Assumed margin tiers

With isolated margin, the default, each perp leg takes margin at 3x, or the most that the venue's tier allows at the size, and a reserve of 10% of the notional. The board reads the margin tiers of Bybit, OKX, Hyperliquid, KuCoin and MEXC.

Binance publishes its tiers only to a signed request, and Bitfinex's margin grows by size steps that its public config does not hold. For their perps, the leverage and the liquidation distance are assumptions, and each capital figure that rests on them says “assumed”. Check them on the venue before you open.

Fees, transfers and taxes

The fees are each venue's base tier, typed in by hand from its fee page. The fee table is dated 2026-10-28. A higher volume tier, or maker orders, cost less.

Not in the numbers: transfers, withdrawal fees, borrow costs and taxes. The funds also sit on exchanges, with their risk.

Funding is not locked

Each venue sets a new rate each interval, so the past is not a promise. Most venues pay at settlements every 1, 4 or 8 hours, so a hold can get one payment more or less on each leg. The price gap between the two legs can also move before you close.

Questions, answered

What is a delta-neutral funding position?

A long and a short of the same coin, of equal size, so that a price move gains on one leg what it loses on the other. In a spot-perp pair, the position earns the funding that the short perp receives. In a perp-perp pair, it earns the spread between the funding rates of two venues.

Is funding rate arbitrage risk-free?

No. Each venue sets a new rate each interval, so the funding is not locked. The price gap between the legs can move, a leg on margin can be liquidated after a large move against it, and the fees and margin rules are assumptions. The board lists the risks of each position.

Which costs does the net take off?

On each leg: the fee, the spread or the measured impact at your size, and slippage, once to open and once to close. The percent on capital also counts the margin and the reserve, which earn nothing. Transfers, withdrawal fees, borrow costs and taxes are not in the numbers.

How good is the funding forecast?

In a backtest on 55 perps from 2023-10 to 2026-10, the forecast of a 7- or 30-day hold was close to what was paid. Longer holds were forecast too high, and the 5 best forecasts of each day paid 13% to 49% less than forecast. The board states this and does not take it off the net.

Why does a high funding rate not make a position profitable?

The costs to open and close come first. A short hold pays them with less funding, and a large size pays more impact in the books. A pair that the rates now pay the other way, or whose size goes past the books that were read, ranks last and does not count as profitable.

Can I read the data from my own code?

Yes. The web proxy serves the top of book and the funding of each market, the hourly funding, a year of settled funding, and the hourly gap between two markets, as JSON over HTTPS. The API docs list each route and field.

Web proxy API

Go further

• Live · FundingFunding boardThe positions now, at your size and hold, with the risks of each one.→
• Integrate · APIWeb proxy APIThe funding, the books and their history as JSON, for your own code.→
• Data · OracleThe oracleSigned prices from independent nodes, checked by quorum. The web proxy behind the boards is a client of these nodes.→
• Board · ArbitrageCross-exchange arbitragePrice gaps between exchanges after taker fees, with the size and the gap at which to stop.→