The Lab

How we tested

• Bitcoin (and Ether as a check), hourly, June 2022 to October 2026: 52 months that include the sour 2022–23 market.
• Every book holds the same net long: 50,000 dollars of the coin at the start, so they all gain and lose the same coins on the price.
  • (a) 50,000 of the coin, nothing else.
  • (b) 100,000 of the coin and a 50,000 short, done both ways: with a dollar-settled (linear) perpetual, and with a coin-settled (inverse) one.
  • (c) 150,000 and a 100,000 inverse short.
  • (d) 250,000 and a 200,000 inverse short.
  • Neutral: coins and a short of the same size: the "break".
• Real costs: each settlement's actual funding, 0.1% to buy the coins, 0.02% on each perpetual order. Checked once a day, and resized when the hedge is more than 5% off.

The maths, step by step

1. Funding, in a year. Funding is paid every 8 hours. 0.01% each time is 0.01% × 3 × 365 = 10.95% a year, paid to the short when it is positive, and paid by it when it is negative.

2. The book is plain coins plus a neutral hedge. In (c), 150,000 of coins and a 100,000 short:

• 50,000 of the coins are unhedged: the same as (a);
• the other 100,000 of coins and the 100,000 short cancel on price. They only collect funding.

So (c) = (a) + a neutral 100,000 book. The extra money earns the funding, nothing more.

3. Why the inverse short can't be liquidated by a rise. An inverse short of Q dollars, entered at price E, holding C coins as collateral, is worth C − Q/E + Q/P coins at price P.

• If the coins are worth at least the short at entry (C × E ≥ Q), that stays above zero at every price.
• Example: 1 coin at 100 against a 100-dollar short. At 200 the account holds 1 − 1 + 0.5 = 0.5 coins, still worth 100 dollars. At 1,000 it holds 0.1 coins, still worth 100 dollars.
• That is why a fully covered inverse short is called a synthetic dollar.

4. Return on the extra money. Take the book's profit minus (a)'s profit, and divide by the extra money it needed. For (c), the extra is 100,000.

What we found

1. The extra money earns the neutral book's return, and only that. For every tilted book with an inverse short, the extra money earned exactly what the neutral book earned:

• In 2022–23 the short side often had to pay, and the year earned almost nothing.
• Over 52 months, the short paid away 7.6% of its size in negative settlements on Bitcoin, and 9.3% on Ether. Those are already counted in the table.

2. What the extra buys, and what it costs.

• It buys the funding on the hedged part: about 5–6% a year here, with single years from almost 0% to 12%.
• It costs:
  • the extra money, tied up: three times as much for (c), five times for (d);
  • fees, about 0.1% of the extra to set up;
  • the exchange: for the short to be safe, the coins have to sit there as its collateral;
  • stretches of negative funding. The worst saw the funding collected fall 2.0% of the hedge's size over 42 days from November 2022 on Bitcoin, and 3.2% over 123 days on Ether.
• The swings don't shrink. As a share of the bigger pot, (c)'s worst drop looks smaller than (a)'s (32% against 54% on Bitcoin). In dollars the two swing the same, because they hold the same net coins.

3. "Why not just 50,000 of spot?"

• If the extra 100,000 would otherwise sit idle, hedging it adds the carry, about 5–6% a year in these four years.
• If it has a better use, the tilted book is no better than 50,000 of spot plus that use.

4. Dollar-settled or coin-settled short.

• The linear short hedges a fixed number of coins. As Bitcoin rose, its dollar size grew, and so did its funding: +11.0% a year on the extra, against +5.8%. That is more carry in a rising market, but also more to pay when funding turns negative, and it needs more margin.
• The inverse short is a fixed dollar amount. With the coins as its collateral it never needed resizing at all.

5. Liquidation, said plainly.

• The inverse short, with the book's coins in the same account: a rise can't liquidate it while the coins are worth at least the short at entry. That held throughout: 2.0 times for (b), 1.5 for (c), 1.25 for (d). If the coins are moved out of that account, it can be liquidated.
• A linear short with its own separate margin: Bitcoin rose 295% above the short's entry. Only margin of about 3.3 times the short's size or more would have survived.
• A linear short in a combined account, with the coins as collateral: it never came close (the margin was at least 235 times the minimum).

6. Taking a break: the neutral book.

• The neutral inverse book needed no resizing over the 52 months. The hedge stayed exact.
• What it needs instead are funding alerts. On Bitcoin the last 3 days' funding summed below zero in 51 separate spells (Ether 64), and the last 7 days' in 23 (24). That is roughly one 3-day warning a month, and one 7-day warning every two months.

What it's called, and how professionals run it

The name. Buying the coin and shorting its future is the cash-and-carry or basis trade. With perpetual futures it is often called funding-rate arbitrage. A coin holding hedged with an equal inverse short is a synthetic dollar. Leaning it with a partial hedge has no single standard name; here we call it carry with a directional tilt.

What is documented:

• The size and swings of the carry. The Bank for International Settlements' working paper Crypto carry (Schmeling, Schrimpf and Todorov, 2023, revised 2025) describes the long-spot, short-futures trade. It found the carry averaged above 10% a year and at times exceeded 40%, varying a great deal over time. It linked the carry to demand from smaller, trend-chasing buyers of leverage, and to limits on arbitrage capital from regulation and margin.
• Why perpetuals pay. Fundamentals of Perpetual Futures (He, Manela, Ross and von Wachter) found perpetual prices strayed from no-arbitrage values far more than in currency markets, and that the gaps shrank over time as markets matured. The trade's best years may be behind it.
• A 1x short of your own coins has no liquidation price. Deribit's guide to hedging the dollar value of a coin holding says that shorting 1x or less of your balance leaves the account's dollar value fixed, with no liquidation price. It warns that funding can cost you on longer holds.
• The risks and rules of a large operator. Ethena runs this trade at scale, and its documentation names:
  • the risks: funding, liquidation, custody, exchange failure, backing assets and margin collateral;
  • a reserve fund for negative funding, with the backing moved into stablecoins earning about the Treasury rate when funding is low;
  • off-exchange settlement, so the collateral never sits with the exchange, and only unsettled profit and loss between daily settlements is exposed to a failure;
  • spreading hedges across several exchanges, cutting exposure when a venue looks riskier, and re-hedging elsewhere if one fails.
  • Its own history check: funding negative on about 16% of days for Bitcoin, with the longest negative streak 13 days.

What is our inference, not documented practice:

• treating the tilt as a separate, deliberate bet on price;
• the daily check with a 5% band (it's our own bots' rule);
• the alert levels (3-day and 7-day funding below zero);
• that a pause is "hedge everything and watch funding".

Update, 21 October 2026: the paper soak's first score

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

Since 9 October, two paper rows on exchange demo accounts have run the same basket: Bitcoin, Ether and Solana held, and shorted on coin-settled futures to earn funding.

• Neutral: fully hedged.
• Tilted: hedged more or less, by our market reading of each coin's trend.

Did the tilt do better? From 9 October 09:08 to 21 October 05:39 UTC, 11.9 days:

Each row's own value against its start, with nothing added or taken out. Other bots in the same accounts are not counted.

• Most of the neutral row's loss was a deliberate test. On 16 October, from 07:19 to 09:44 UTC, it held about 7x in coins with most of its hedges missing. That was a test of margin in one account. Leaving those hours out, it was down 0.41%.
• Without the test, the tilt was ahead by about 0.75%, or 0.08% a day on average over 10 comparable days.
  • That is well inside luck: t = 0.73, where about 2 is needed.
  • Telling a gap that small from luck would take many months.
• The tilt leaned short (hedges 1.1 to 1.4 times its coins) in a fortnight when Ether and Solana fell a little. That is one direction in one fortnight. This page's own test, over 52 months, is the longer answer.
• The neutral row also borrows a little (1.13x on average against the tilt's 1x), so the two aren't exact twins.
• Verdict: not yet. The tilt is ahead, by too little and for too few days to say anything. We'll score it again on 26 October.

Update, 26 October 2026: the paper soak's final score

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

The soak ended on 21 October, when every paper row was stopped. Scored to just before that, from 9 October 09:08 to 21 October 16:32 UTC, 12.3 days:

Each row's own value against its start, with nothing added or taken out. Other bots in the same accounts are not counted.

• The extra 11 hours changed almost nothing. The figures are within 0.04% of the first score's.
• As recorded, the tilt was ahead on most days, by 0.31% a day on average over 13 days: t = 1.54, where about 2 is needed. Most of that gap is the neutral row's stress test on 16 October, not the tilt.
• Without the test, the gap was 0.07% a day over 11 comparable days: t = 0.68. That is well inside luck.
• Verdict: a lead, not a result. The tilt finished ahead, by too little and over too few days to tell from chance. The soak is over, so this is its last score. This page's 52-month test stays the longer answer.

Credits

• Maik Schmeling, Andreas Schrimpf and Karamfil Todorov, "Crypto Carry", BIS Working Paper 1087, Bank for International Settlements, 2023 (revised 2025).
• Songrun He, Asaf Manela, Omri Ross and Victor von Wachter, "Fundamentals of Perpetual Futures", working paper, 2022.
• Deribit Insights, "Hedging USD value by shorting 1x", an exchange guide (we didn't check its date).
• Ethena's public documentation: its pages on risks, funding risk and exchange-failure risk.
• 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.
• The names cash-and-carry, basis trade, funding-rate arbitrage and synthetic dollar are standard market terms, not ours.
• The tilt as a separate bet, the 5% band and the alert levels are our own, as the page says.
• Data: Bybit's public hourly BTCUSDT and ETHUSDT prices and funding history.
• PerpGremlin and Claude, The same long, held two ways: plain coins, or coins plus a hedge that earns funding, TerminalGremlin, 2026: the tests, the code and the words.