How to Read the Funding Rate
Everyone quotes it as a sentiment gauge. It is much better than that at one job and much worse at the other, and the difference is worth real money.
By Pavel Penev, MScFounder, TradeWize · 10+ years trading the marketsThe short answer
The funding rate is the payment perpetual-futures holders make to each other every 8 hours. Read it as two facts and you will be right: it says which side is crowded, and it says exactly what that side is paying. Read it as a third — a signal that the crowd is about to be punished — and 7,666 settled rates on the biggest bitcoin perpetual disagree with you. Sorting the next week's move by the funding rate before it barely separates the outcomes at all.
Spend a week around crypto traders and you will hear the funding rate quoted like a thermometer. Funding is at yearly highs, so the top is in. Funding just went negative, so that was the bottom. It is one of the few numbers that everybody watches and almost nobody tests.
So we tested it. Every settled funding payment on the biggest bitcoin perpetual — BTCUSDT, on Binance USDⓈ-M futures — since September 2019, 7,666 of them, joined to what price did next. The answer is not that the rate is useless. It is that the two things it does well are not the thing it is famous for, and the famous one does not show up in the data.
If you have not met the instrument, our perpetual-futures explainer covers the mechanism: no expiry date, so a payment every eight hours does the job that settlement does on a normal futures contract. This article starts where that one stops. The formula is on your screen and the number is blinking. What is it actually worth?
The first thing it tells you: which way the crowd is leaning
The rate is positive when the perpetual is trading above spot and negative when it is trading below. That is a direct read on positioning, because a perp only trades above spot when more money wants to be long than short at the current price. Nobody is guessing here — the payment is the crowd's own money moving from one side to the other.
Which raises the obvious question: how often is each side crowded? On bitcoin, over seven years, longs paid in 85.7% of all intervals. On ether it is 86.3%. Being long is not one state of the market; it is the default state of the market, and the funding rate is what that default costs.
| Contract | Since | Longs paid | Sat exactly at the base rate | Average rate per 8h |
|---|---|---|---|---|
| Bitcoin | 2019-09-10 | 85.7% | 35.4% | 0.0106% |
| Ether | 2019-11-27 | 86.3% | 34.6% | 0.0126% |
| Solana | 2020-09-14 | 71.3% | 34.6% | 0.0002% |
Settled rates on Binance USDⓈ-M futures, read on 2026-09-08. The fourth column is the formula's fallback — when the perp sits on the index, funding drops to a fixed 0.01% interest component, and that is roughly a third of the record. Solana is the interesting row: its perp spends nearly 29% of its life below spot, so on average the shorts there have paid.
That is the first genuinely useful read, and it is a comparative one. A 0.01% rate is not "low", it is the floor — the number the formula falls back to when nothing is happening. Judge every reading against that floor rather than against zero, and against the same contract's own history rather than against another coin's.
| Rate per 8h | What it means | What it is good for |
|---|---|---|
| below 0% | More money is short than long, so the perp trades below spot and shorts pay longs. | Rare and usually brief. It says the crowd has given up on the upside, and it pays you to disagree. |
| 0.00% – 0.01% | Longs pay shorts, but only the base interest rate the formula falls back to. | The default state of a quiet market. Two thirds of the record sits here, and it tells you nothing about direction. |
| 0.01% – 0.03% | Longs are paying a premium over the base rate to hold the position. | A trend has people's attention. Ordinary during any rally, and not a warning of anything. |
| 0.03% – 0.05% | Longs are paying three to five times the base rate, every eight hours. | The position now costs real money to hold. That is the fact — what happens next is a separate question the record does not answer. |
| above 0.05% | Longs are paying five times the base rate or more, on the way to the venue's cap. | The most expensive place to be right. It is also where the cost, not the direction, is what usually beats you. |
The second thing: the price of standing still
Here is the part that gets skipped, and it is the half that costs people money. A funding rate is a price. It is the rent on a position, charged three times a day, and it does not care whether you are right.
Convert it into the only unit that matters: how far price has to move in your favour before you have made anything. At the quiet-market floor of 0.01%, a long needs 0.03% a day just to break even — 0.9% over a month. At a crowded 0.04%, it is 0.12% a day and 3.6% a month.
This treadmill runs at the same speed at 2× and at 100×
Funding is charged on your position, and profit is earned on your position, so the two scale together and cancel. Leverage decides how fast the bill eats your deposit — that is week 80's arithmetic — but it does not change the move you need to cover it. The break-even move is one of the very few numbers in leveraged trading that you can quote without stating a leverage first.
Each line is the cumulative move a long needs, at that funding rate, before the trade has made a single cent. The steepest line is what a genuinely one-sided market charges. A view that needs six weeks to play out is being charged rent for six weeks.
Now stack that up over years, which is where it stops being a rounding error. Sum every settled payment on the bitcoin perp by calendar year and there is no year in the column pointing the other way. In 2021, holding the long side cost 30.6% of the position over the year. In 2022, price fell 65% — and longs still paid 4.2% for the privilege.
The left column is the market everybody watches. The right column is the toll, and it has no negative entries: in 4 of these 8 years price finished lower and the crowd paid to be there anyway. Over the last 365 days, longs paid 3.3% while price fell 30%.
That last pair of numbers is the whole argument in miniature. The funding rate does not know what price is doing. It is a bill for having an opinion, and it arrives in the losing years too.
The third thing, which is not there
So much for what the rate charges. What about what it predicts — the reason it gets quoted on every timeline?
The test is not complicated. Signal: the mean settled funding rate over the previous three days (nine 8-hour intervals). Outcome: the change in the 8-hour close over the following seven days (21 intervals). Every interval in the sample is used as an observation, so the windows overlap. Sort the outcomes by the signal, and if a high rate really does mark exhaustion, the buckets should fan out.
Grouped by the funding rate over the previous three days, 7,636 overlapping observations. The pale bar is the mean move over the following week; the solid bar is what was left after paying that week's funding. The dashed line is the sample's own average week.
They do not fan out. The most crowded band — where longs are paying five times the base rate or more, the reading that gets screenshotted with a skull emoji — was followed by a mean week of +0.96%, against +0.92% for every week in the sample. Price was higher seven days later 51% of the time, against 53% across the whole record. There is no crash hiding in that bucket.
Run the same sort with the windows cut so they do not overlap — 7,636 rolling observations become a few hundred independent weeks — and the ordering rearranges itself completely. That is not a result that needs interpreting. Two views of the same data that cannot agree on which bucket did best is what "no signal" looks like from the inside.
The honest limits of this test
One venue and one contract. Funding differs between exchanges, and the same reading on a small altcoin perp means something else entirely. Overlapping windows. Consecutive observations share most of their outcome, so the sample is far less independent than the count suggests — the non-overlapping check below is the honest sample size. In-sample and gross. No fees, no slippage, no attempt to trade any of it. A negative finding on one venue's flagship contract is not proof that nothing is there — it is a reason to stop treating the number as an oracle when it is unarguable as a price.
There is a second reason it makes a poor timer even when it is screaming: it does not stay screaming. The 21 stretches where bitcoin's three-day funding held above 0.05% lasted a median of 2.7 days — but the longest ran 22 days, over which a short taken on the signal would have paid the crowd for three weeks while waiting to be right.
Set a funding rate and the panel answers the three questions it can answer — and then the one everybody actually asks it, using 7,666 settled rates from 2019-09-10.
Settled 8-hour rates for BTCUSDT on Binance USDⓈ-M futures, 2019-09-10 to 2026-09-08, read on 2026-09-08. Forward returns use overlapping windows and ignore fees, so treat the last panel as a shape rather than a number to trade. Now do the thing the widget was built for: drag the slider from one end to the other. The cost panels move by a factor of 38. The panel underneath them barely moves at all — and longs still paid 3.3% over the last year while price fell 30%. 3 payments a day, every day, whichever way it goes.
Learn it by doing
Reading about it is one thing — it clicks when you do it. Learn it hands-on with free, interactive lessons on TradeWize.
Try the free lesson →The case the warning-light story has to explain
On 10 October 2025, roughly $19bn of leveraged crypto positions were force-closed inside 24 hours across about 1.6 million accounts, and perpetual open interest fell 43%. If ever there was a build-up of crowded, over-levered longs for a sentiment gauge to catch, it was that week.
| Date | Mean funding per 8h | Close |
|---|---|---|
| 6 Oct | +0.0077% | $124,628 |
| 7 Oct | +0.0083% | $121,286 |
| 8 Oct | +0.0060% | $123,238 |
| 9 Oct | +0.0027% | $121,579 |
| 10 Oct | +0.0019% | $112,715 |
| 11 Oct | +0.0068% | $110,579 |
| 12 Oct | −0.0037% | $114,894 |
| 13 Oct | −0.0012% | $115,112 |
| 14 Oct | −0.0026% | $112,984 |
| 15 Oct | −0.0004% | $110,699 |
| 16 Oct | −0.0027% | $108,129 |
Daily mean of the settled 8-hour rates, BTCUSDT. The base interest rate the formula falls back to is 0.01%.
Read the run-in. In the three days before the crash the rate never got above 0.0083% — below the ordinary base rate, and falling as the day approached. The gauge was not merely quiet, it was cooling. Then, once the damage was done and price sat 13% below where the week started, funding went negative and stayed there for days: the crowd flipped short at the bottom of the move, which is the one thing sentiment reliably does.
The lesson is not that funding lied. It is that funding measures what positions are being paid for right now, and leverage build-up is a stock, not a flow. Open interest was the number carrying that week's risk. Funding was reporting, accurately, that nobody was paying much of a premium to be long — which was true, and which had nothing to say about how much borrowed money was already in the building.
Where the rate does bite: net of what you paid
There is one column in the study that behaves, and it is the one nobody looks at. Subtract the funding a long actually paid over the week from that week's price move, and four of the five bands stay positive. The fifth does not.
What the crowd watches
The most crowded band's mean week: +0.96%
- Barely different from any other week in the sample.
- Price higher seven days later 51% of the time.
- No fan-out, no ordering, no edge to trade.
What the crowd pays
The same week, after funding: −0.31%
- The only band in the study that is negative on net.
- A quiet-market week nets +0.45% by comparison.
- The cost did what the direction never did.
So the crowded reading is real information, and it is a cost warning rather than a direction warning. It does not say the market is about to fall. It says that on the average week from here, the rent has been enough to turn a market that went slightly up into a position that went slightly down. That is a reason to size smaller or shorten the hold — not a reason to short.
And when it goes negative
Negative funding is the mirror image: the perp is trading below spot, more money is short than long, and holding a long pays you. It is genuinely rare on bitcoin — 14.3% of intervals — and rarer still in stretches. There have been 505 negative runs in seven years and the median one lasted a single 8-hour payment. The longest ever ran 24 payments, 8 days.
In the study it is the best-looking bucket: a mean following week of +2.80% against +0.92% for the sample, with price higher a week later 61% of the time. Treat that gently. It is the same overlapping-window arithmetic as everything else on this page, it leans on a handful of capitulation episodes that all resolved the same way, and the Solana contract does not reproduce it: its perp is below spot 29% of the time, the widest sample of shorts-are-paying anywhere, and across 1,942 of those observations the median following week was −0.41% with price higher only 48% of the time. What is solid is the smaller claim: when funding is negative you are being paid to hold the unpopular side, and being paid to wait is a materially different trade from paying to wait.
Who is on the other side of all this
Every payment a crowded long makes goes to somebody short the perp. Much of that somebody is not bearish at all. The hedged version — buy the spot asset, short the same size of perpetual, collect funding, carry no view on price — is one of the largest trades in crypto, and it is the reason the rate behaves like an interest rate rather than a mood.
Which means it competes with cash. Over the last 365 days, bitcoin funding paid that hedged position 3.3% on the notional. The 3-month Treasury bill paid 3.75%. On those numbers the trade is not free money — it is slightly worse than a bill, before the exchange risk of holding both legs on a venue. In 2021 it paid 30.6%, and money came from everywhere to collect it.
That is the tether under the whole number. When funding runs far above the risk-free rate, hedged sellers arrive until it does not; when it sags below, they leave. It is why the rate spends a third of its life pinned to the formula's floor and why it mean-reverts as fast as it does — and it is one more reason to read a spike as a temporary price rather than as a prophecy.
How to actually read it
- Compare it to the base rate, not to zero. 0.01% per interval is the formula's floor and roughly a third of all readings. Anything above it is a premium somebody chose to pay.
- Convert it into a daily break-even move before you do anything else. "0.04% funding" means nothing to your gut; "0.12% a day just to stand still, 3.6% a month" means everything.
- Check how long it has been there. A single hot payment is noise; a rate that has held above the floor for a week is a real cost that has already been charged several times.
- Compare across venues and contracts. The same reading means different things on a deep bitcoin perp and on a thin altcoin one, and a gap between two exchanges is usually about who is on each venue rather than about the market.
- Use it to size and to time your HOLD, not to pick a direction. High funding is a reason to hold a shorter horizon or a smaller position, because the treadmill is running faster.
- If it is paying you, notice that the crowd can still be right. Collecting funding on a short while price grinds up is a slow way to lose money with a nice-looking cash flow.
As this was written, bitcoin funding settled at 0.0082% with the 30-day average at 0.0067% — about 7.4% a year on the position, a little under the Treasury bill. Nothing to see, which is the reading two thirds of the record gives, and the one worth getting comfortable with.
The one-line version
The funding rate is a price tag, not a forecast. It tells you which side is crowded (longs, 85.7% of the time) and exactly what that side is paying (11.6% a year on average, and longs paid in every one of the last 8 calendar years). It does not tell you where price goes next — and in the one place the data does bite, it is the cost that turns the average crowded week negative, not the direction.
What does a high funding rate mean?
It means the perpetual is trading above spot because more money wants to be long, and those longs are paying shorts every 8 hours to keep the position. It is a reliable read on crowding and on cost. It is not a reliable read on direction: over 7,666 settled bitcoin rates, the most crowded band was followed by a mean week of +0.96%, against +0.92% for the sample as a whole.
Is a negative funding rate bullish?
Negative funding means shorts are paying longs, so the crowd has flipped to the downside — historically the best-performing bucket in this study (+2.80% the following week). Treat it as weak evidence: it is rare (14.3% of bitcoin intervals), it usually lasts one payment, and other contracts do not reproduce it. The solid part is that you are being paid to hold the unpopular side rather than paying to.
How much does funding actually cost?
Bitcoin's average settled rate since 2019-09-10 is 0.0106% per 8 hours, which is 11.6% a year on the position. On a $10,000 position that is about $1,159 a year, and it is paid out of your margin — so at 10× leverage it is ten times that share of your deposit.
Does leverage change the funding cost?
It changes what the cost does to your deposit, not the move you need to cover it. Funding is charged on the position and profit is earned on the position, so the break-even move is identical at 2× and at 100×. What leverage multiplies is the share of your margin the bill represents.
Can the funding rate predict a crash?
Not in this data. In the three days before 10 October 2025 — about $19bn liquidated in 24 hours, the largest such event in crypto — bitcoin funding never rose above 0.0083%, below its own base rate, and was falling. Leverage build-up shows up in open interest, not in the payment being made for it right now.
Why is funding positive most of the time?
Two reasons. Retail demand for leveraged long exposure is persistent, so the perp usually trades a little above spot; and the formula falls back to a fixed 0.01% interest component when the perp sits on the index, which is a positive floor rather than zero. On bitcoin that produces 85.7% positive intervals over seven years.
What is the funding-rate carry trade?
Buy the asset on the spot market, short the same size of perpetual, and collect funding with no exposure to price. It competes directly with cash: over the last year it paid 3.3% on bitcoin against 3.75% on a 3-month Treasury bill, though in 2021 it paid 30.6%. That competition is what pulls extreme rates back to the floor.
How often is the funding rate paid?
On the major venues, 3 times a day at 8-hour intervals, and only if you hold the position at the settlement moment. Some venues use hourly intervals on some contracts. The payment goes to the traders on the other side, not to the exchange.
Funding tells you the cost. The chart tells you the trade
Upload a screenshot or type a ticker and the chart analyzer reads the levels, the trend and the structure off the real candles — including where the crowd's stops are sitting. It never tells you what to buy. It tells you what the chart is doing, so a decision about cost is a decision you make with the picture in front of you. Free, no account, one read a week.