The Lab

The maths, step by step

1. Funding into a yearly rate. Funding is paid three times a day on most exchanges.

• Example: 0.01% every 8 hours is 0.01% × 3 × 365 = 10.95% a year to the short.
• Real rates move, and in a sour market they turn negative: then the short pays.

2. A neutral hedge. Hold 1 coin and short 1 coin's worth of the perpetual. If the price rises 10%, the coin gains 10% and the short loses 10%, which cancels. The book earns only the funding, less fees.

3. An inverse short leans on its own. An inverse perpetual is sized in dollars but paid in the coin.

• Example: a short of 10,000 dollars' worth of contracts at a price of 100 covers 100 coins. At 125, the same contracts cover only 80.
• So the hedge shrinks as the price rises and grows as it falls, and must be reset each day.

4. Its average price is harmonic, not plain. That inverse leg's average entry has to be worked out the right way.

• Example: 100 one-dollar contracts at 100 is 1 coin, and 100 more at 125 is 0.8 coins. Together: 200 contracts for 1.8 coins, an average of 200 ÷ 1.8 = 111.11.
• The plain average, (100 + 125) ÷ 2 = 112.5, is wrong.
• That difference mattered. See finding 3.

5. The tilt leans the hedge with the trend.

• A tilt of 0.3 hedges 70% of the coins when the trend reads up, all of them in a range, and 130% when it reads down.
• 0.6 leans twice as hard.

6. How "sideways" a year was: the efficiency ratio (Perry Kaufman's) is how far the price ended up, divided by how far it travelled.

• Example: 100 → 110 → 95 → 105 travels 10 + 15 + 10 = 35 to end up 5 away: 5 ÷ 35 = 0.14.
• 0 means it wandered and came back. 1 means a straight line.

What we found

1. On one exchange, perpetuals alone have no carry.

• Over a falling year (to October 2026), what a short received: BTC +2.7% to +3.3%, ETH +2.1% to +2.3%, across the USDT, USDC and inverse perpetuals.
• The three products fund almost alike. The best pair between them on BTC made +0.47% a year after fees, and none on ETH cleared +0.1%.
• Hopping to whichever pair paid best lost 12.8% to its own fees.
• The carry that exists needs a spot leg, or a second exchange.

2. Between two exchanges there was a steady gap.

• Over the same year, one exchange (Hyperliquid) paid its shorts more than the other (Bybit): BTC +5.9% against +2.7%, ETH +6.3% against +2.3%.
• Long on Bybit and short on Hyperliquid collected +3.2% (BTC) and +4.0% (ETH), positive in all 13 months, against 0.06% in fees.
• It is small, and it carries costs not in the number:
  • two separate pots of collateral, so a squeeze can liquidate one leg while the gain sits on the other exchange;
  • the price gap between the two venues.

3. The hedged basket, counted correctly, earned its funding and little more.

• The basket: BTC, ETH and SOL at a third each, each held as spot with its own short, rebalanced once a day past a 5% drift.
• The first count was wrong. It showed ×1.52 over two years for the neutral basket. A check found two errors:
  • the inverse short's average price was taken plainly, not harmonically;
  • the hedge was counted against the spot coins alone.
  • Together they invented most of that return, and quietly made "neutral" bet against the move.
• Corrected, over two years (October 2024 to October 2026): the neutral basket made ×1.13 with a 1% worst drawdown, about 6% a year, its funding.
• Borrowing to run it at 3.4 times, at a 3.9% loan rate, gave about 8% a year with an 8.8% drawdown. Leverage bought little.
• A funding rule that lifts the hedges when funding turns negative made the book directional: ×1.41 at a 31% drawdown, against ×1.13 at 1% without it, in two rising years. That return came from holding unhedged coins, not from carry.

4. The tilt: two ways of reading the trend gave two answers.

• With the trend read the way our live bots read it (a strength indicator on 4-hour candles), every tilt lost to neutral over two years:

In brackets: the worst drawdown. Leaning slower, after 72 hours, did worse still (30%: ×1.041; 50%: ×0.975 at 18%).

• With the research model's own trend read (daily and 4-hour structure), the tilt added a little in the trending years. See finding 5, where the same model is run year by year.

5. In a year that goes nowhere, the hedge holds the price, not the funding.

• The most sideways 12 months in the history (lowest efficiency ratio across the three coins) were June 2022 to May 2023. The ratio was 0.020, with BTC ×0.85, ETH ×0.96 and SOL ×0.45.
• Next came calendar 2025, at 0.021.

• In the sour year of 2022–23 the shorts paid funding, 10.2% in all, and the neutral basket lost about that much.
• Against chance: each tilt was rerun with its trend spells in shuffled order, 20 times (a permutation test, after R. A. Fisher).
  • In the up year the real timing beat 16–17 of 20 shuffles.
  • In 2025 and the down year it was no better than chance.
  • In 2022–23 it did worse than all but one or two shuffles. The tilt cost 5 points (0.3) and 13 points (0.6) against neutral.

On our paper accounts

Our own paper bots, on exchange demo accounts, with no real money. Percentages and ratios only.

Two paper accounts have run the basket since early October 2026:

• carry: the neutral basket, with some hedge-pair grids beside it, from 8 October.
• tilt: the same basket with a tilt of 0.3 and no loan, from 9 October.

• Six and five days say almost nothing. Telling a Sharpe like the tilt's apart from luck needs about 289 more days. The "chance it is above 0" and the days needed come from Bailey and López de Prado's probabilistic Sharpe ratio (2012).
• The carry account's figure is the whole account, grids included, not the basket alone.
• Each account's record can be scored again as it grows.

Credits

• The carry trade (long the coin, short its futures, collect funding) is long-standing market practice, not ours. Maik Schmeling, Andreas Schrimpf and Karamfil Todorov, "Crypto Carry", BIS Working Paper 1087, 2023, measure it in crypto.
• Robert J. Shiller, "Measuring Asset Values for Cash Settlement in Derivative Markets: Hedonic Repeated Measures Indices and Perpetual Futures", Journal of Finance, 1993: the perpetual future, proposed. Crypto exchanges built it, starting with BitMEX in 2016.
• Holbrook Working, "The Theory of Price of Storage", American Economic Review, 1949: the price of carrying a thing to its future's date.
• Perry J. Kaufman, Smarter Trading, McGraw-Hill, 1995: the efficiency ratio.
• R. A. Fisher, The Design of Experiments, Oliver and Boyd, 1935: the permutation test.
• William F. Sharpe, "Mutual Fund Performance", Journal of Business, 1966: the Sharpe ratio.
• David H. Bailey and Marcos López de Prado, "The Sharpe Ratio Efficient Frontier", Journal of Risk, 2012: the chance a Sharpe is above 0, and the track record needed.
• The basket, the tilt and the tests are our own design.
• Data: Bybit's public prices and funding history; Hyperliquid's public funding history.
• PerpGremlin and Claude, Funding carry and the hedged basket, TerminalGremlin, 2026: the tests, the code and the words.