Funding rate arbitrage where the receiving side rarely flips · Data 21:33:16 JST

The two pages start from different assumptions, so the rows differ, and which one is better is not something this site can decide.

How long you can keep holding is set by your own circumstances.

• Front page: only pairs on the assumption that you open and close within two days; it states there, reason by reason, how many were left out — nothing left once costs were taken out, the direction disagreed, the gain buried in the swings.
• This page: combinations whose gap between funding rates has stayed in the same direction for a long time, on the assumption that you hold for days to weeks. One round of buying and selling is repaid by holding on, so it is read by the realised annual rate and the days to cost recovery.

By the realised annual rate over the last 30 days, highest first. It is the same calculation over the same 30 days as the realised annual rate on the pair detail page, so the two show the same figure.

The realised annual rate is what has actually been paid and received over the last 30 days (over the whole observed period for a pair seen for less than that), restated as one year, with no interest earning interest (compounding). As the note at the foot of the screen says, a high annual rate does not mean safety.

• Only this annual rate enters the ranking — stability, days observed, giveback and cost are not part of it.
• The instantaneous annual rate on the detail page is a hypothetical calculation of this very moment lasting a year, so it is a different figure, and the two can even carry opposite signs.

A — means that period's figure is not produced, and the three numbers are read by how the short and long periods disagree.

Stability is the share of that period in which the same exchange stayed on the higher-funding side (that is, the side you sell in this direction); 1.00 means it never flipped. High at 15, 30 and 90 days alike is sturdy; high at 15 days only means it only turned that way recently, a passing state; high at 90 days but low at 15 means it is breaking down right now. Faint type means there were few records in that period, and putting the cursor on a cell tells you which meaning a — has, or the exact days observed.

• In the 15-day column: rows in the list above have at least 15 days observed, so a — always means there is no record of funding passing during those 15 days.
• In the 30-day and 90-day columns: it can additionally mean the days observed are too short to evaluate that period. Days observed is printed rounded, so a row reading 30 days that is really 29.7 days shows —.

It shows how much of the accumulated funding profit was given back along the way, and rows above 0.25 are not listed here.

It is the largest give-back divided by the final profit and loss — a number for how close the curve is to a straight line.

• 0.02 — it piled up in almost a straight line.
• 0.50 — half of what had been accumulated was given back at one point.
• A — means the figure is not produced: either the records needed for the calculation are still few, or the row involves an exchange that sets a separate rate for the buying side and the selling side.

No — "awaiting evaluation" means no judgement of good or bad has been made.

A combination that could be seen on both exchanges for less than 15 days does not reach even the shortest stability period (15 days), so the yardstick for measuring whether it is sturdy does not exist. It is in a separate box only because there is nothing yet to judge it on, and once the days observed pass 15 it moves up to the list above.

• What lines up in this box: up to 20 rows, in order of days observed, longest first.
• Plus, for each exchange that appears nowhere in the table above, its single row with the longest days observed.
• The number left out is always stated inside the box.

There are several reasons; the main ones are that it fell out on this site's own conditions, or through your own settings or the ranking.

Other reasons are possible too. Separately, on a day when data ingestion is behind, a band saying evaluation is on hold appears at the top of the screen.

• This site's conditions (1): across the whole period observed, the share in which the same exchange stayed on the higher-funding side dropped below 70% — the receiving side flipped.
• This site's conditions (2)(3): the open interest on the thinner of the two exchanges fell below $5.0M (down to $2.0M is allowed only for the eight assets BTC, ETH, SOL, BNB, XRP, DOGE, HYPE and ASTER); or the asset is no longer carried on one of the two exchanges, or the selling side and the buying side are no longer available in a form that can be held at the same time.
• This site's conditions (4)(5): the difference you would receive became buried in the day-to-day swings, as the residual of two large flows paid and received; or giveback went above 0.25.
• Your side (6)(7): it was cut by the 24h volume or open interest floor you entered yourself — the number excluded is then stated near the top of the screen; or, since the table holds only the top 100 rows, it slipped below that.

No. It restates what has been observed over a short stretch as if it lasted a year; it is not a forecast of what you would receive.

Three things are outside it. It has no compounding — it is the average of what has actually passed between the two sides, multiplied out, not interest earning interest. It has no cost taken out: opening and closing are paid separately, and how many days that takes to repay is the cost recovery column. And it is not a record of one year — the days observed column says how long the pair has actually been watched, and the rate is set afresh every few hours, so the direction itself can turn (rows where it has are marked).

• A high annual rate does not mean safety — the note at the foot of the screen says the same.
• Whether to take any of them is not something this site says (it is not investment advice).

What is left is the funding difference less the cost of opening and closing, and the break-even is the days to cost recovery on that row.

The cost column adds three things, at both exchanges, for opening and for closing: the exchange's own fee, the gap between the bid and the ask, and the worse price a $10,000 order gets as it fills at once. The size it was estimated at is stated at the top of the screen. The smaller your own size, the heavier the fee share, so a small position does worse than the screen says. Funding, meanwhile, arrives a little at each settlement, so the row only turns positive once enough settlements have passed.

• The fee assumed is each exchange's published standard rate for an order that fills straight away; a discounted tier pays less.
• Where the orders standing on one of the exchanges could not be read, the cost is shown with a ≈ in front — that exchange's share is estimated from open interest and volume, the same estimate the pair page uses.
• Not included: moving funds between the two exchanges, and tax.

There is no way here to tell in advance. What this page can show is how much of the time the same side has stayed on top so far, and it drops a row once that breaks.

So the list is itself the sieve: the conditions below are applied to every row, and what is left is what has held so far — not what will hold. How the three stability columns are read is answered separately above.

• Across the whole period observed, the share in which the same exchange stayed on the higher side must be 70% or more.
• A pair seen on both exchanges for less than 15 days goes to the awaiting-evaluation box instead of this list.
• A row whose giveback went above 0.25 is not listed.
• None of the three is a promise about what comes next.

Not necessarily — volume is what was traded over the past day, not what is standing in the market right now.

The cost of getting in and out is measured from those standing orders, not from volume: the price a $10,000 order gets as it fills at once. So a row can carry a large 24h volume and still show a heavy cost. The volume floor you can type in yourself is applied to the thinner of the two exchanges, and the number of rows it removed is stated near the top of the screen.

• A large volume is not evidence that the orders standing there are thick.
• The 24h volume column shows whichever of the two exchanges traded less over the last 24 hours; a — means the exchange does not publish it, not that the figure is small.
• Where the orders standing on each side are thin, part of the price slip is an approximation and the cost carries a *.

Each row is one funding-rate arbitrage candidate: sell on the exchange paying more, buy on the one paying less, and hold that spread.

The list is sorted by what the funding actually paid out over the period observed, not by what it might pay next. A row appearing here does not mean it is worth taking — the columns to its right show what it costs to open and close, how many days it takes to repay that, and how steady the difference has been.

Yes — that is what it is. It is a screener over funding-rate pairs: you set the floors and the list narrows to the rows that clear them.

The controls above the table are the screen: minimum open interest, minimum 24h volume, minimum days observed, and which exchanges to include. Each one is applied to the thinner of the two sides, because a pair can only be traded to the size the thinner side allows. The count of rows removed by each filter is shown, so you can see what a screen is costing you rather than only what it leaves.

Yes. Every row links through to the two exchanges' own history, and the ranking itself is built from the past, not from this instant.

The realised annual rate is the average of what has actually been settled, and the stability figures are the share of the last 15, 30 and 90 days in which the gap kept the same direction. Both are records of the past. Clicking an exchange name opens that one market's own history — funding, open interest and volume over time — which is where you look to see whether today is ordinary for that pair or unusual.

A dated contract settles on a named day. What it pays is the gap between its own price and the spot price today, and that gap closes once — it does not keep arriving the way funding does.

Because it closes once, the total a dated contract can pay is the gap as it stands today, in basis points, and no longer. Turning that into a yearly rate divides by the days left, so the same gap reads larger the closer the end date is: 15 basis points over 20 days reads 2.7% a year, and the same 15 basis points over one day reads 55%. The cost of opening and closing is paid once and does not shrink with the days left, so on this list a dated row is only carried when the gap itself is larger than that cost. Measured on 2026-09-05 across the 18 dated contracts collected, 8 of them showed between 2.4% and 3.5% a year while their gap was smaller than one round trip.