Seven years of funding: the carry that paid 30.61% now pays 2.92%

The number everyone quotes is five years old

The perpetual funding carry trade is simple: buy spot, short the perpetual, collect the funding payment every eight hours. It is delta-neutral, it is easy to explain, and it is the first strategy almost every crypto quant builds.

It is also routinely described with numbers from the era when it worked best. I wanted the actual history, so I pulled it — every realised settlement, not a daily average — and the decline is steeper than the folklore suggests.

The data is free, and that matters

Binance publishes GET /fapi/v1/fundingRate: the realised funding rate at every 8-hour settlement, back to contract inception, with no API key and no subscription. Seven point one years for BTCUSDT.

This is worth stating because the obvious commercial sources are worse for the question. Aggregators carry long history but only at daily resolution that far back — and a daily OHLC of a funding rate is an aggregate. It cannot reproduce per-settlement accounting, negative-run lengths, or the distributional detail below. The venue’s own endpoint is strictly better data for this, at no cost.

Annualised carry, by calendar year

Long spot / short perp, on notional, before costs. Percentage of settlements where funding was negative — where the book pays rather than earns — in brackets.

year      BTC             ETH             SOL             DOGE
2019     7.48 (18.3%)    8.91 ( 4.8%)      —               —
2020    17.19 (14.3%)   27.41 ( 2.6%)  -12.52 (25.3%)    9.39 (15.5%)
2021    30.61 ( 7.3%)   37.54 ( 4.1%)   28.59 ( 4.7%)   38.51 ( 8.5%)
2022     4.17 (22.1%)    0.79 (34.2%)  -35.56 (45.8%)    3.45 (25.8%)
2023     7.87 (10.1%)    8.26 ( 9.1%)    1.30 (28.1%)   10.00 (10.0%)
2024    11.92 ( 8.4%)   12.96 ( 4.2%)   13.62 (10.3%)   14.00 (10.0%)
2025     5.13 (12.9%)    4.93 (16.2%)    0.35 (39.5%)    4.34 (24.8%)
2026     2.92 (26.2%)    1.97 (30.0%)   -1.04 (47.9%)    3.08 (31.5%)

The eight-year BTC mean is 10.91%/yr. 2026 is paying 2.92%. In 2021 the same trade paid 30.61% — more than ten times what it pays now. Fitting a line through the annual levels gives −1.72 percentage points per year, about −16% relative annually.

That direction is not my discovery. Published work documents the deviation from theoretical perpetual pricing compressing at roughly 11% a year, attributed to arbitrage capital arriving and competing the spread away. My contribution here is the per-settlement series showing it at venue level, from data anyone can re-pull in two minutes.

The part I did not expect: the dead zone is disappearing

Binance computes funding as a premium index plus a clamped interest term. The premium uses max(0, ·) on both sides, so while the index sits inside the perpetual’s spread the premium is exactly zero and funding collapses to the bare interest component — 0.01% per 8 hours, 10.95%/yr. Practitioners call this the dead zone, and it used to be where the market lived:

share of settlements printing EXACTLY 0.01%

year      BTC      ETH      SOL     DOGE
2019    66.9%    78.8%       —        —
2020    50.5%    56.8%    63.1%    63.7%
2021    43.2%    44.8%    52.1%    43.1%
2022    31.1%    26.0%    31.1%    44.2%
2023    38.7%    37.3%    41.3%    55.2%
2024    42.1%    39.8%    36.7%    43.6%
2025    17.2%    19.4%    19.5%    25.4%
2026     6.8%     5.3%    13.9%    22.7%

Two-thirds of 2019 settlements printed the bare interest term. In 2026 it is one in fifteen. The market has moved out of the dead band — and increasingly onto the negative side of it. BTC funding was negative in 7.3% of settlements in 2021 and 26.2% in 2026. SOL spent 47.9% of 2026 paying.

This changes what the strategy is. When half of all settlements print a venue-set constant, funding carry is close to a floating-rate coupon. When only 6.8% do, it is a directional sentiment trade wearing a delta-neutral costume.

What this means for the trade today

I measured the current window separately — 417 days, 88 venue-token pairs, reconstructed order books, a sealed out-of-sample split. BTC perpetual carry came out at 3.87%/yr on notional. Against the eight-year mean of 10.91%, that is roughly a third.

Two consequences follow, and they point in opposite directions.

The cost arithmetic has inverted. A spot-perp round trip at published retail fees is about 30 bps. Against 2021’s 30.61%/yr that is a rounding error, and you could rotate between tokens freely. Against 2026’s 2.92% it is roughly a month of carry per round trip, and rotation becomes the dominant cost. The same strategy, unchanged, has moved from robust to fee-bound without anyone editing a line of it.

But the strategy demonstrably worked. This is not a piece arguing funding carry was always a mirage. At 30%/yr on BTC, with negative funding in 7% of periods, it was a genuinely good trade and the people who ran it then were right. The honest question was never does this work — it is does it still work, at your size, after seven years of capital arriving.

What this does not say

Funding is not the whole return. A held position also marks the basis to market, and in richer eras that term was larger than the coupon — published work attributes 8.64%/yr of BTC return to price convergence against 5.06%/yr from funding payments. This series is the funding leg only.

One venue. Binance USD-M. Funding parameters are venue-specific; the interest term, the clamp and the impact notional all differ elsewhere, and the dead-zone share differs with them.

No costs, no capacity, no execution. This is the carry that was on offer, not what a book would have captured. Those require order books, and this endpoint has none.

A trend is not a forecast. 2022 fell to 4.17% and 2024 recovered to 11.92%. The line slopes down across eight points; it does not tell you about the ninth.

The one-line version

Perpetual funding carry on BTC has gone from 30.61%/yr in 2021 to 2.92%/yr in 2026, the share of settlements printing the bare interest term has fallen from 66.9% to 6.8%, and funding is now negative more than a quarter of the time. The trade was real. If you are evaluating it on numbers from the era that made it famous, you are evaluating a different trade.