How to read Bitcoin funding rates without fooling yourself

Negative BTC funding preceded stronger seven-day returns in every calendar year tested. Extremely positive funding did not, and its carrying cost erased the pooled price gain.

The direct answer

Funding is a holding cost before it is a trading signal. Positive funding means longs normally pay shorts at the next settlement. Negative funding means shorts normally pay longs. The sign tells you which side pays under the contract's rules, not which side is smart or what price must do next.

In 7,388 Binance BTC funding settlements with a complete seven-day outcome, negative starting funding was followed by a +2.52% average price return. All other starting states averaged +0.76%. The 1.75 percentage-point difference has a Newey-West standard error of 0.67 points, or t = 2.60.

Extremely positive funding did not behave as bullish momentum. Starting rates above 200 parts per million were followed by only +0.31% in price, and the funding actually paid during the following week reduced the average net return to -0.56% for a long before fees and slippage.

Negative funding beat its own calendar-year baseline in all seven years in the sample. That repeated direction is notable. It still does not prove that the funding rate caused the return or that the next negative print is a buy signal.

What funding actually does

A perpetual future has no expiry date. Without a settlement at expiration, the contract needs another mechanism to keep its price near the underlying spot market. Funding provides that mechanism through recurring transfers between long and short position holders.

The simplified payment is:

funding payment = position notional x settled funding rate

When the settled rate is positive, a long normally pays and a short receives. When it is negative, the direction reverses. The exchange facilitates the transfer, but funding is not the same thing as a trading fee.

The quoted rate applies to one settlement. It is not a daily or annual rate. On Binance, settlements have usually occurred every eight hours, but the actual schedule can change and the current estimate can move before settlement. The rate that settles is the one that creates the payment.

At the common 0.01% rate, three settlements per day would cost a long about 0.03% daily. One hundred parts per million, abbreviated 100 ppm, is 0.01%.

What a funding chart is showing you

A funding line usually plots the current or historical rate through time. A positive line means the long side is paying under the contract formula. A negative line means the short side is paying.

The magnitude needs a baseline. A positive rate can be the contract's ordinary state rather than evidence of exceptional crowding. In this sample, exactly 100 ppm was the largest band, containing 33.7% of eligible settlements. Calling every positive print "hot" would treat a routine value as a warning.

The shape matters too:

• A brief spike can create one expensive settlement and disappear.
• A moderate rate that persists can cost more over a long hold than one short spike.
• A negative print can reflect temporary perp weakness without telling you who initiated the underlying positions.
• Rates on different venues are not directly comparable until settlement intervals and contract rules are aligned.

Funding should therefore be read as a path of realized and potential payments, not as a red-or-green sentiment gauge.

Here is the highest funding rate in the whole sample, picked by that rule rather than by eye, with price and the rate at each settlement around it.

More generally, extreme rates fade. The chart averages the funding rate at each settlement over the week after three starting states.

Why the same rate can accompany different markets

Positive funding often appears when the perpetual trades rich relative to its reference price. That can happen while price trends upward, while leverage builds inside a range, or while arbitrageurs hold the other side. The funding rate does not separate those cases.

Negative funding can appear during a sharp selloff, a quiet bearish market or a temporary venue-specific discount. It does not prove that shorts are crowded or that longs have already been flushed.

Market regime is a major confounder. Very positive rates tend to occur in active markets where baseline returns can differ from quiet periods. A useful comparison asks whether the return after a funding state differs from other observations in the same kind of market, not only whether the pooled return was positive.

A practical way to read funding

The study below measures seven-day BTC outcomes on Binance. It does not validate this discretionary framework, identify a timing rule or show that funding causes returns. The steps separate the things the rate can tell you from the things that require other evidence.

1. Translate the quote into your holding cost. Use the position notional, the settled rate and the actual settlement schedule. The relevant question is what the position will pay or receive over the time you expect to hold it.

2. Compare the rate with its own baseline. Check the same contract and venue. Positive versus negative is less informative than routine versus unusual.

3. Check whether the state persists. One estimate can change before it settles. A sequence of realized payments is different from one temporary spike.

4. Put open interest beside it. Rising OI says outstanding contracts expanded; falling OI says they contracted. It does not identify the aggressor, but it tells you whether the funding state formed while leverage grew or shrank.

5. Compare spot and perpetual price action. If the perp is rich while spot also attracts aggressive demand, that is a different market from a perp premium unsupported by spot. Funding alone combines them.

6. Separate a forecast from a cost decision. You do not need to predict a reversal to decide that a trade is too expensive to hold. Carry can invalidate an otherwise acceptable position even when the directional thesis remains possible.

How we measured the historical result

We used BTCUSDT perpetual funding and hourly BTCUSDT perpetual prices from 2020-01-01 through 2026-10-15.

The unit of analysis is a scheduled funding settlement hour, not every intraperiod funding update or every raw timestamp variant. The source sometimes records the same scheduled settlement twice with millisecond-level timestamp jitter, such as 16:00:00.000 and 16:00:00.008. We normalize next_funding_time to its clock hour, then take the last quoted rate before that settlement. This collapses 8,479 distinct raw timestamp values into 7,410 scheduled settlement hours.

We measured price from the settlement hour to exactly 168 clock hours later. We also summed one normalized funding settlement per clock hour during those seven days.

Two settlement hours lack the starting price and another 20 lack the exact seven-day endpoint, because the last week before the sample ends has no full outcome yet. After excluding them, the result is 7,388 settlement observations. Their unconditional average seven-day price return was +1.010%, and price was higher 53.8% of the time.

The outcome windows overlap. Statistical comparisons below therefore use Newey-West standard errors rather than treating every settlement as independent.

What happens to Bitcoin after funding turns extreme?

The funding-cost column is what was actually paid over the following week. It is not the starting rate multiplied by an assumed 21 settlements. That is why a row selected for negative starting funding can still have a small positive cost over the full week: subsequent rates can turn positive.

The table separates two questions that are often blended together. Negative starting funding was associated with stronger subsequent price returns. Very positive starting funding came with a large enough subsequent carrying cost to erase the pooled price gain for a long.

The result by calendar year

Negative funding beat the same year's average in seven years out of seven. That is more informative than a pooled comparison because it does not rely on negative funding occurring only in one favorable regime.

Extremely positive funding underperformed the annual baseline in each of the four years in which it occurred. A regression with calendar-year fixed effects estimates an adjusted difference of -2.30 percentage points for the above-200 ppm band, with a Newey-West standard error of 1.29 points and t = -1.79. The direction is consistent, but the uncertainty is too large to call the adjusted effect settled.

The pooled and within-year comparisons agree here. Both put the hot-funding band below its relevant baseline, so the yearly breakdown is not overturning what the pooled table shows.

Common mistakes

• Reading funding as sentiment. The rate comes from the contract's pricing mechanism. It does not identify what every holder believes.
• Treating every positive rate as crowded. Routine positive funding can sit near the contract's baseline.
• Annualizing the current estimate without checking persistence. The rate can change before the next settlement and across later settlements.
• Ignoring the actual holding period. Carry that is trivial for an intraday trade can dominate a longer position.
• Comparing venues without normalizing. Settlement schedules and contract rules can differ.
• Claiming negative funding proves a flush. The state was associated with stronger returns here, but the rate does not identify who exited or why.
• Shorting a high rate automatically. Very positive funding can remain expensive while price continues higher.

What this study does not establish

• This is BTC on Binance, not the full perpetual market.
• Seven days is one horizon, and point-to-point returns do not show the path.
• The annual rows share overlapping seven-day outcomes and are descriptive.
• Seven same-direction calendar slices are not seven independent experiments.
• Funding is associated with market state. The study does not identify a causal mechanism.
• Net return subtracts funding but not trading fees, slippage or liquidation risk.

---

*Written by Dom at QuantumFlow. Published 5 August 2026, updated 16 October

1. Figures use Binance BTCUSDT perpetual settlement-level funding and hourly prices from 2020-01-01 to 2026-10-15.*