● Learn
Is funding rate arbitrage still profitable in 2026?
Funding rate arbitrage, or carry, means holding spot and shorting the same amount of the perpetual future. Price moves cancel out, and you collect the funding that long traders pay every eight hours. The question is how much that funding really comes to after costs. Here is what 2,190 real settlements per symbol on BTC and ETH show.
What it earned, year by year
Annualised funding, before leverage:
| Year | BTC | ETH |
|---|
| 2024 | +8.18% | +9.40% |
|---|
| 2025 | +5.13% | +4.93% |
|---|
| 2026 (to June) | +1.00% | +0.35% |
|---|
Across the two years the net figure, after both legs’ fees and slippage, is +4.88% on BTC and +4.91% on ETH before leverage. At 3.4x reference leverage with a 25% reserve held back for tail moves, that is about 12.5% a year on total equity.
The yearly table is the honest answer to the question: funding is not a fixed rate. It is paid by traders who want to be long, so it is richest when the market is keen and thinnest when it is quiet, as it has been in 2026.
The decision that matters most: don’t churn
The single biggest difference in the data is not the year. It is how often the position is opened and closed. Over the same two years:
- Held, never traded: +9.8%
- Opened and closed 14 times a year: +0.6%
- Opened and closed 52 times a year: −25.2%
Each round trip costs about 34 basis points of the position, which is roughly 25 days of funding to earn back. The break-even is 14.9 round trips a year, at every leverage. Trying to time the funding, closing when it dips and reopening when it recovers, spends the yield on fees. Our funding carry break-even calculator runs this same arithmetic for your own rate and costs.
Three checks before you run a carry bot
- Count its round trips. A bot that trades in and out to “optimise” funding is paying for it every time.
- Ask how its funding data was joined. Matching settlements to prices by exact timestamp quietly drops about 40% of them (39.4% in our data), and not at random: the quarters where it dropped the most were the ones paying the most. A backtest built that way is measuring a different yield from the one you would earn.
- Look at the yearly numbers, not only the average. An average hides the quiet years.
CARRY is built on exactly these figures: it holds, it does not time the funding, and its verify command regenerates every number above from the raw settlements.