Archive / Crypto & DeFi / E005
Ranking perpetuals by funding rate collects less than costs
Ranking perpetual futures by funding rate collected 0.75 to 6.20 bp of funding per trade against 9.57 to 16.69 bp of costs; net Sharpe was negative in all 12 cells (-0.73 to -11.59).
A common belief says that perpetual futures with the highest funding rates are crowded with leveraged longs, so shorting them and buying the lowest-funded ones should earn the funding payment plus a price edge. We ranked 24 perpetuals by their last funding rate at every 8-hour settlement, shorted the top group, bought the bottom group and held for 8 or 24 hours. Funding received was 0.75 to 6.20 bp per trade, price moves were noisy around zero (-3.23 to +5.31 bp) and costs were 9.57 to 16.69 bp. Net Sharpe was negative in all 12 cells, from -0.73 to -11.59.
Perpetual futures with the highest funding rates are crowded longs: short them and buy the lowest-funded ones to collect the funding.
Why people believe it
Perpetual futures have no expiry, so the exchange keeps their price near spot with a funding payment every 8 hours: when the perpetual trades above spot, longs pay shorts. People argue that a very high rate marks crowded leveraged longs, that the short collects the payment, and that the crowded side is more likely to be unwound than to keep pushing the price against the short.
Why it would persist The argument, as stated before testing
People argue that leveraged speculation is a standing demand, and that hedging it ties up capital and margin, so not enough capital would close the gap.
At each funding settlement (00:00, 08:00 and 16:00 UTC) we looked at the last funding rate of each of 24 perpetual futures and sorted them. We shorted the top 20% (or top 33%), bought the bottom 20% (or 33%), weighted the coins equally in each group and held the position for 8 or 24 hours. The result counts the price change, the funding actually paid or received over the hold, and 7 bp per side of trading cost. The history was split into three consecutive windows of 60%, 20% and 20%.
Not recorded as a numeric threshold. The verdict was read from the net Sharpe in each window.
Not recorded as a rule. Read after the fact: net Sharpe at or below zero in the windows. All 12 cells were negative.
7 bp per side (5 bp fee + 2 bp slippage), charged on the turnover of weights between consecutive entries.
None. The test reports price and funding components separately for each window.
In every one of the 12 cells costs exceeded funding plus price pnl, so net Sharpe was negative (-0.73 to -11.59).
- The summary stats (n 428, net mean -9.54 bp) are the holdout cell of the 24-hour, 20% variant, one of 12 cells.
- Funding received per trade was 0.75 to 6.20 bp across the 12 cells. The 24-hour holds collected 1.82 to 6.20 bp and the 8-hour holds 0.75 to 2.30 bp.
- The price component was noisy around zero, from -3.23 to +5.31 bp per trade, and changed sign between windows in 3 of the 4 variants (for example the 8h quintile: -0.20, -3.23 and +0.15 bp; the 24h tercile stayed positive: +0.64, +0.91 and +5.31 bp, but that is below the costs of 9.57 to 16.69 bp per trade, so it was still a net loss).
- Costs were 9.57 to 16.69 bp per trade and rose over time: for the 24-hour quintile 11.36 bp in the selection window, 12.61 bp in validation and 16.69 bp in the holdout.
- Net per trade (price + funding - cost, computed from the table) was -4.50 to -13.29 bp. The best cell, the 24-hour quintile in the selection window, lost 4.50 bp over 1,284 trades.
- Net Sharpe ranged from -0.73 (24h quintile, selection) to -11.59 (8h tercile, holdout). The 8-hour holds were -3.48 to -11.59 and the 24-hour holds -0.73 to -3.09.
- Reference, not a test: the average funding rate since 2021 was about 10 to 14% a year on the major coins, 10.7% for Bitcoin. The 90-settlement rolling Bitcoin rate fell to 3 to 6% a year in 2025-2026 and was negative around 2026-05.
- Delta-neutral carry (short the perpetual, hold the spot coin) was not run. Its entry costs of 7 bp for the perpetual and 12 bp for the spot leg come to 19 bp one way, which is about 12 to 23 days of funding at 6% to 3% a year (computed from the rates above).
Funding is small next to costs
Funding receivedPrice pnlCost paidNet
Per-trade components of the 24-hour, top-and-bottom-20% variant in each window. Funding received (6.20, 1.84, 3.11 bp) never reaches the cost paid (11.36, 12.61, 16.69 bp), and price pnl adds under 5 bp, so net is negative everywhere.
Chart data (bp)
| Funding received | Price pnl | Cost paid | Net | |
|---|---|---|---|---|
| Selection (n 1,284) | 6.2 | 0.66 | 11.36 | -4.5 |
| Validation (n 428) | 1.84 | -0.4 | 12.61 | -11.16 |
| Holdout (n 428) | 3.11 | 4.04 | 16.69 | -9.54 |
Net Sharpe is negative in every cell
SelectionValidationHoldout
Annualised net Sharpe ratio of each variant in the selection, validation and holdout windows. All 12 bars are below zero, from -0.73 to -11.59; the 24-hour bars are closer to zero than the 8-hour bars.
Chart data (net Sharpe)
| Selection | Validation | Holdout | |
|---|---|---|---|
| 8h hold, 20% groups | -3.48 | -8.88 | -10.04 |
| 8h hold, 33% groups | -4.53 | -8.61 | -11.59 |
| 24h hold, 20% groups | -0.73 | -2.68 | -3 |
| 24h hold, 33% groups | -1.03 | -2.85 | -3.09 |
Why it failed Interpretation · not established
The test does not isolate a cause beyond the arithmetic. Funding is paid on notional three times a day, and an average rate of 10.7% a year is about 2.9 bp a day (computed), so even the top and bottom groups, which collected 1.82 to 6.20 bp over 24 hours, receive 0.6 to 2.1 days of average carry per trade (1.82 / 2.9 and 6.20 / 2.9, computed). The price leg did not offset that: it was positive in some windows and negative in others, with no stable sign. Costs are charged on turnover, so the higher holdout costs may mean the groups changed more between entries; this was not checked.
Lesson
A scheduled payment such as funding is worth collecting only if its size per trade exceeds the cost of entering and leaving. Here funding of at most 6.20 bp per trade was below costs of at least 9.57 bp, and the price leg showed no stable sign.
Scope of this result
Covers a cross-sectional ranking on the last settled funding rate across 24 liquid perpetuals, with 8-hour and 24-hour holds, taker costs of 7 bp per side and equal weights. It does not cover longer holds, maker execution, single-coin funding timing, other group sizes or venues, or hedged spot-and-perpetual carry, which was not run.
A rejected experiment applies to the tested conditions and does not prove an idea can never work. bp = basis point, 0.01%.
1. Test design
Pre-registeredDataUniverseSignalExecutionCostsControlsSample splitPass criterionKill criterionPower2. Result
ObservationsInstrumentsNet mean95% intervalMedianWorstFinding 1Finding 2Finding 3Finding 4Finding 5Finding 6Finding 7Finding 83. Interpretation
Why it failedLessonScope4. Reproducibility
StepsGrade