The Lab

Where the idea comes from

• The carry trade holds a coin and sells the same amount short on futures, and collects the funding. Its origins and maths are on our theory page "Funding carry: getting paid to hold a hedged coin".
• Borrowing to make it bigger multiplies the funding, and adds a loan to pay for. When funding falls below the loan's rate, the extra size costs more than it earns.
• Not trading unless it pays is old, settled theory. Michael Magill and George Constantinides (1976) showed that, once trading costs money, the best plan leaves a "no-trade zone": small departures from the ideal are left alone. Mark Davis and Andrew Norman (1990) worked out that zone exactly. Our rule (d) below is a simple version of the same idea.
• The stand-down rule tested here is our own software's. The question came from it.

How we tested

• Funding: Bybit's public 8-hourly funding on its coin-margined (inverse) futures, Bitcoin, Ether and Solana, from January 2022. Solana's market didn't settle from November 2022 to April 2024, so those dates are left out.
• The book: coins bought with the money plus a loan, and an equal short. Held this way, the price cancels, so only funding, the loan's interest and fees move the money. Built at 3, 5 and 10 times the money.
• The loan rate: a flat 3%, 4%, 6% or 8% a year. There is no free history of the real rate, so its swings are not tested.
• Fees: 0.1% on the coins and 0.055% on the short, each time the size changes.
• "The funding" is what our software reads: the last 7 days of funding as a yearly figure.
• The rules:
  • (a) today's: to 1x the moment funding is under the loan's rate;
  • (b) patience: the same, but only once it has been under for 1, 3, 7 or 14 days in a row;
  • (c) steps: cut by a quarter each day while under, not all at once;
  • (d) cost-aware: cut only if the gap, over the next 7, 14 or 30 days, would cost more than the fees of cutting and rebuilding;
  • (e) never cut;
  • (f) never borrow.
  • Every rule builds back by 5% a day, and only while funding is above the loan's rate.
• No peeking: (b)'s days and (d)'s horizon were picked on 2022 to 2024 alone. Both picked 14 days. Then every rule was scored on January 2025 to October 2026, held back.
• The bar, set before the run: a rule beats today's only if, in the held-back years, it earns more on at least 2 of the 3 coins at every leverage, with a worst fall no deeper. Judged at each loan rate.

The maths, step by step

1. The funding as a yearly figure. 0.01% paid every 8 hours for a week is 21 payments: 0.01% × 21 × 365 ÷ 7 = 10.95% a year.

2. What the loan does. At 3 times the money, the short collects 3 × 10.95% = 32.85%, and the loan of 2 times the money costs 2 × 3% = 6%. The book makes 26.85% a year before fees.

3. What a cut costs. Cutting the loan and building it back trades it twice, on both sides: 2 × (0.1% + 0.055%) = 0.31% of the loan. At 10 times the money the loan is 9 times it, so one round costs 9 × 0.31% = 2.79% of the money.

4. The cost-aware test. Funding at 0 against an 8% loan: over 7 days the gap costs 8% × 7 ÷ 365 = 0.153% of the loan. That's less than 0.31%, so (d) waits. Over 30 days it's 0.658%, more than the fees, so (d) cuts.

5. The way back. At 5% a day, 1x back to 10x takes ln 10 ÷ ln 1.05 = 47.2, so 48 days.

What we found

1. No rule passed the bar. At every loan rate, none of (b) to (e) beat today's rule in the held-back years. Never borrowing, (f), did pass it: in 2025 and 2026 it beat today's rule on Ether and Solana at every leverage.

2. The worry is real in good years. At 10x and a 4% loan:

• In 2023, today's rule on Ether at 10x paid 14.5% of the money in fees and was cut back on 70% of the days the funding beat the loan. It made +6.0% a year. Never cutting made +41.9%.
• In 2022 it was the other way round. Never cutting lost 34% to 59% a year at 10x. Today's rule lost 4% to 11%.

3. The held-back years, coin by coin, at 10x and a 4% loan:

• The cost-aware rule did better on Bitcoin and Ether at a 4% loan in 2023, 2024 and the held-back years, and at a 3% loan in the held-back years too. It cut 2 times where today's rule cut 24 to 31.
• But it fell further on Solana, whose funding was often poor, and lost at 6% and 8% loans. That is what failed it.
• Today's rule is expensive and safe. In the held-back years, at a 4% loan, it paid 6% to 16% of the money in fees and was cut back on 42% to 51% of the days the funding beat the loan. But its deepest fall over the three coins was the shallowest of all the rules that borrow, at every leverage.

What it means

• There is a real trade-off, and no free fix. Acting at once is costly when funding dips and recovers. It protects when funding stays low, as in 2022 and on Solana.
• Waiting, or weighing the fees, earned more on Bitcoin and Ether, but took deeper falls where funding stayed poor. By the bar set in advance, neither is better than today's rule.
• In the latest years, the loan barely paid at all. At a 4% loan in 2025 and 2026, never borrowing beat every rule that borrows on Solana, and was within half a point of the best on Ether.
• The loan rate decides it. At 3% or 4% the patient rules looked best on the big coins. At 6% and 8% they lost.

Update, 16 October 2026: two corrections

Added after this page was published. The text above is left as it was.

• "Cut from 10 times its money" described the setting, not the book. The book was set to borrow up to 10 times its money, but some of its orders were refused. When it was cut back to 1 time, it held about 2.9 times its money in each of Bitcoin and Ether (5.7 times in all), each with a short of about 3.2 times. Our page "The carry trade in one account: does it survive the crashes?" tells that day in full.
• Solana's missing dates were misstated under Limits. Its funding is missing from January to 24 March 2022, and from 10 November 2022 to April 2024: about four and a half months of 2022, all of 2023 and the start of 2024. The test ran as described under "How we tested".