Archive / Crypto & DeFi / E826
The house side of leveraged trading showed no reliable return
Over 182 days, the liquidity-provider token of a large on-chain perpetuals exchange, hedged against its asset basket, returned +0.34 bp per day with an interval from -1.3 to +2.1 bp; the median day lost 0.59 bp.
Some on-chain perpetuals exchanges let holders of a liquidity-provider token act as the counterparty to leveraged traders, and pay them the trading fees. The belief is that retail traders lose on average, so the house should earn a steady return once the price risk of its asset basket is hedged. Over 182 days, the hedged token returned +0.34 bp per day (a basis point, bp, is one hundredth of a percent). That is not distinguishable from zero: the interval runs from -1.3 to +2.1 bp, and the median day lost 0.59 bp.
The liquidity-provider token of a perpetuals exchange earns the house edge over leveraged retail traders, and hedging the basket leaves that edge.
Why people believe it
People argue that leveraged retail traders lose more than they win, and that fees on opening, closing and borrowing come on top. The token holders are the other side of every trade, so they receive the traders' losses and the fees. The pool's value is a basket of SOL, ETH, BTC and stablecoins plus fees plus trader losses minus trader gains. Strip out the basket's price moves with short positions and, in this view, the fees and the traders' net losses remain.
Why it would persist The argument, as stated before testing
Taking the other side of retail leverage needs capital and tolerance for lumpy losses when traders win in a trending market. The fee income is paid for carrying that risk, so in this view it would not be competed away.
We took the daily price of the liquidity-provider token and removed the price moves of its asset basket by subtracting the return of short positions in SOL, ETH and BTC perpetuals (futures contracts with no expiry) sized to the basket's published target weights. We added the funding those shorts would receive (the periodic payment between long and short holders of a perpetual) and charged 2 bp per day for rebalancing the hedge. The first question was whether what is left is positive. The second was whether the day after a day of large market moves differs from other days. Both rules and both rejection criteria were fixed before any token return was computed.
Mean hedged daily return above 0 with a 95% interval (the plausible range, from a bootstrap that resamples calendar weeks) whose lower bound is above 0, median above 0, and positive in each calendar half-year. Second question: the difference between after-big-move days and other days has an interval above 0.
First question: the interval's lower bound is at or below 0. Second question: the difference interval includes 0.
2 bp per day for hedge rebalancing, short funding received added. Pool deposit and withdrawal fees, and any slippage when entering or leaving the token, are not modelled.
Residual sensitivity of the hedged return to each of the SOL, ETH and BTC returns, to check the hedge removed the basket exposure.
The interval's lower bound is -1.3 bp, at or below zero, so the first question fails; the difference after big-move days has an interval from -17 to +26 bp, which includes zero, so the second fails.
- The hedged daily return averaged +0.34 bp (about 1.3% a year) after 2 bp per day of hedge cost and with funding received. That does not count as a carry: the 95% interval runs from -1.3 to +2.1 bp and the median day lost 0.59 bp.
- The worst day lost 65 bp and the best day gained 90 bp, both far larger than the average.
- The two calendar half-years disagree in sign: -0.37 bp per day from April to June (about 2 months) and +0.77 bp per day from July to October 2026.
- Measured against each of the 3 hedge legs (SOL, ETH, BTC), the residual sensitivity of the hedged return was about zero, so the hedge removed the basket's price exposure; what was left cannot be split between fees, trader results and noise.
- Only 7 days followed a day of large market moves, too few to read anything into. The hedged return on those days was +3.7 bp against +0.2 bp on the other days, with a difference interval from -17 to +26 bp.
The hedged return cannot be told from zero
Mean hedged daily return over 182 days with its 95% interval (bootstrap over calendar weeks), and the two calendar half-years. The pass rule needed the interval's lower bound above 0; it is -1.3 bp. The half-years are point means only.
Chart data (bp/day)
| Mean | 95% low | 95% high | |
|---|---|---|---|
| Whole history | 0.34 | -1.3 | 2.1 |
| April-June (about 2 months) | -0.37 | Not reported | Not reported |
| July-October | 0.77 | Not reported | Not reported |
A small mean beside large single days
Hedged daily result in bp: the mean is +0.34, the median is below zero, and the best and worst days are 90 bp and -65 bp. The mean is positive but a median day lost money.
Chart data (bp)
| Hedged daily result | |
|---|---|
| Worst day | -65 |
| Median day | -0.59 |
| Mean | 0.34 |
| Best day | 90 |
Seven days after a big move prove nothing
Mean hedged daily return on the 7 days that followed a big market move against all other days. The difference interval runs from -17 to +26 bp, so the +3.7 bp bar is not distinguishable from the other bar.
Chart data (bp/day)
| Mean hedged return | |
|---|---|
| Day after a big move (7 days) | 3.7 |
| Other days | 0.2 |
Why it failed Interpretation · not established
The token's headline yield appears to be mostly the basket's own return. Once the basket is hedged away, what is left may be fees roughly offset by what traders won from the pool over these 182 days, leaving nothing distinguishable from zero. This is an interpretation: the test shows the net, not how it splits between fees, trader results and noise.
Lesson
Being the counterparty to a crowd that is believed to lose is only a carry if the crowd's losses plus fees exceed what you pay out when they win, and a hedged return has to be read against the hedge's own noise. Here a mean of +0.34 bp per day sits inside an interval of -1.3 to +2.1 bp. A short history says nothing about a carry this small, and a positive mean with a negative median is not evidence of one.
Scope of this result
Covers one liquidity-provider token, one exchange, fixed target weights, and 182 days from April to October 2026.
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 53. Interpretation
Why it failedLessonScope4. Reproducibility
StepsGrade