Archive / Crypto & DeFi / E770
Fading losing traders fails its own pre-registered test
Following skilled traders was inconclusive and fading persistent losers passed in one week (+92.4 bp) but failed out of time: +0.1 bp unhedged, and the registered both-sides condition failed (shorts -17.3 bp).
Traders on public leaderboards and copy-trading platforms are often treated as either worth following or, if they keep losing, worth betting against. We tested both on crypto perpetuals. Copying six skilled accounts earned +8.9 bp per trade, with an interval from -22 to +87 bp. Fading persistent losers passed in one week at +92.4 bp, a falling week for alt coins, so market exposure fits better than skill; over seven earlier weeks it earned +0.1 bp. Hedged against large alt coins it earned +28.7 bp, but losers' shorts lost -17.3 bp, so the registered test failed. Fading 28 copy-trading leaders earned +4.2 bp, interval -74 to +38 bp.
Some traders are skilled enough to follow, and persistent losers are predictable enough to bet against.
Why people believe it
Public rankings of traders by profit make individual performance visible, and copy-trading platforms let followers mirror a lead trader's entries. People argue that skill persists, so following the best earns, and that the losers keep losing, so taking the other side of their trades earns.
Why it would persist The argument, as stated before testing
Each account is small and its trades are spread across many coins, so a systematic fade or copy is limited by capacity and by the effort of rebuilding every account's history, which keeps large players away.
We sorted accounts into skilled, persistent losers and others using their trades from the previous three weeks. We then either copied the skilled accounts or took the opposite side of the losers' new trades one to three minutes after they opened, and held until the trader closed, however long that took. The first week looked good. We then re-ran the same rule on seven earlier weeks, with the return measured against an index of large alt coins, a hedge added after the first week because that week was a falling week for alt coins (only a Bitcoin hedge had been registered for it). The same fade was also run on the lead traders of a copy-trading platform, with positions capped at 24 hours.
Copy test: mean above 0 with the account-clustered 95% interval above 0, median account above 0, above both control groups, positive at the 5-minute delay, at least 10 skilled accounts. First fade test: at least 15 loser accounts, raw and Bitcoin-adjusted means above 0 with intervals above 0, median account above 0, 5-minute mean above 0, other accounts below losers. Out-of-time fade: pooled alt-adjusted mean at least +20 bp with the account-week interval above 0, at least 5 of 7 weeks positive, losers' longs and shorts both above 0, other accounts below losers. Copy-trading leaders: at least 600 entries, alt-adjusted mean at least +10 bp with leader- and day-clustered intervals above 0, raw mean above 0, both weeks positive.
Any pass condition not met.
11 bp round trip on the on-chain venue (two 4.5 bp taker fees plus two 1 bp slippage); 10 bp round trip for the copy-trading leaders. Return is on notional, unlevered.
Other accounts (neither skilled nor loser) as a negative control; split by the trader's side (losers' longs we short, losers' shorts we long); returns adjusted for the market by subtracting our direction times the equal-weight return of 25 large alt coins (20 for the copy-trading test) over the same candles; the first fade test was registered with a Bitcoin adjustment; its alt-index adjustment was added afterwards.
The seven-week out-of-time fade fails the pre-registered requirement that both sides be positive (losers' shorts -17.3 bp). The copy test's interval includes 0 with a negative median account, and the copy-trading fade's interval includes 0.
- Copying skilled accounts: 6 accounts and 134 trades in the test week earned +8.9 bp per trade at a 1-minute delay (median trade +46 bp, median account -2 bp, account-clustered interval -22 to +87 bp) and +18.6 bp at 5 minutes (interval -10 to +89 bp). Those accounts' own gross return was +106 bp of notional.
- Copying 5 accounts classed as persistent losers (43 trades) lost -114 bp per trade, interval -233 to -49 bp.
- First fade test, new accounts: 16 loser accounts and 196 trades earned +92.4 bp per trade (median +44.8 bp, interval +50 to +178 bp), Bitcoin-adjusted +76.8 bp, and it passed as registered. Other accounts earned +29.1 bp.
- The week was a falling week for alt coins. Losers' longs (we short) earned +200 bp and their shorts (we long) +8 bp. A post hoc check against an alt index, not part of the registration, gave +36.2 bp, with 9 of 16 accounts above 0.
- Out of time, 1,900 accounts and 90,757 episodes over seven weekly folds: the loser fade across 177 account-weeks and 4,332 trades earned +0.1 bp unhedged and +28.7 bp alt-adjusted (median +1.7 bp, interval +2.3 to +55.3 bp). Five of seven weeks were positive: -23, +8, -4, +10, +127, +70 and +48 bp; the first four weeks average -2.2 bp (simple mean of the weekly figures).
- Sign convention: a positive number means the fader made money after costs. Split by side, losers' shorts, where the fade goes long, lost -17.3 bp alt-adjusted, so the registered both-sides condition failed. The lower interval bound of +2.3 bp is barely above 0, and the alt index carries a common negative drift, which overstates any single-side hedged figure.
- Other accounts' alt-adjusted fade was -5.2 bp. Fading skilled accounts earned -15.5 bp (interval -25 to -7 bp); the -15.5 bp is after an 11 bp cost, so copying them would have grossed about +4.5 bp and netted about -6.5 bp after the 11 bp cost.
- Fading copy-trading leaders: 5,171 priced entries from 28 leaders in 83 coins (median hold 2,576 minutes before the 24-hour cap) earned -34.1 bp raw and +4.2 bp alt-adjusted (median +2.7 bp, worst -4,235 bp). The leader-clustered interval is -74 to +38 bp and the day-clustered interval -31 to +31 bp; the two weeks gave -4.8 and +17.9 bp.
- Only current lead traders were listed (245 at the snapshot, 58 eligible), so those who stopped leading during the window are missing, which biases against the fade.
Hedged and unhedged fades, with 95% intervals
Net mean per trade of each fade with its 95% interval, hedged against an alt index. The out-of-time losers' hedged mean (+28.7 bp) is above the +20 bp line, but its lower bound is +2.3 bp, the unhedged mean is +0.1 bp, and the test failed on the short side (-17.3 bp). The other rows have intervals that include 0 or are negative.
Chart data (bp)
| Mean | 95% low | 95% high | |
|---|---|---|---|
| Losers faded, 7 weeks, alt-hedged (4,332 trades) | 28.7 | 2.3 | 55.3 |
| Losers faded, 7 weeks, unhedged (interval approx.) | 0.1 | -42 | 38 |
| Skilled accounts faded, 7 weeks, alt-hedged | -15.5 | -25 | -7 |
| Leaders faded, alt-hedged (leader interval) | 4.2 | -74 | 38 |
| Leaders faded, alt-hedged (day interval) | 4.2 | -31 | 31 |
Five positive weeks, the last three carry the mean
Hedged fade of persistent losers per weekly fold, in order from late August (week 1) to early October (week 7). Five of seven weeks are positive, but weeks 5 to 7 (+127, +70, +48) carry it: the other four weeks average -2.2 bp (simple mean of the weekly figures).
Chart data (bp)
| Losers, hedged | |
|---|---|
| Week 1 | -23 |
| Week 2 | 8 |
| Week 3 | -4 |
| Week 4 | 10 |
| Week 5 | 127 |
| Week 6 | 70 |
| Week 7 | 48 |
The first-week pass shrinks with every check
Net mean per trade of the loser fade. The one-week +92.4 bp falls to +36.2 bp against an alt index and to +0.1 bp unhedged over seven weeks. The hedged seven-week +28.7 bp fails the both-sides test (shorts -17.3 bp), and the alt index carries a negative drift that flatters hedged single-side figures. The alt-index check of the one-week result was added after the result.
Chart data (bp)
| Fade of persistent losers | |
|---|---|
| 1 week, unhedged | 92.4 |
| 1 week, vs Bitcoin | 76.8 |
| 1 week, vs alt index (post hoc) | 36.2 |
| 7 weeks, unhedged | 0.1 |
| 7 weeks, vs alt index | 28.7 |
| 7 weeks, losers' shorts, vs alt index | -17.3 |
Why it failed Interpretation · not established
The first week was a falling week for alt coins, and a fade of mostly long traders is then effectively short alt coins. Hedging against an alt index shrank the result after the fact, and out of time the unhedged return was close to zero. The registered condition that both sides earn failed, which is what a market-exposure explanation predicts, though this test does not prove it. The hedge itself carries a common negative drift, so hedged one-sided numbers are overstated.
Lesson
A strategy tested in one regime can pass on exposure alone, and the check that exposes it is cheap: compare the unhedged and hedged returns, split by side, and rerun on earlier periods. A signal that is positive for longs and negative for shorts points to the market more than to the trader. Small groups selected on past results also overstate persistence: the 5 accounts that lost -114 bp per trade for one week did not predict a stable fade.
Scope of this result
Covers copying or fading accounts of one on-chain perpetuals venue (1,900 accounts selected by size and activity, any holding time, 1 to 5 minute delay) and the lead traders of one copy-trading programme (28 leaders with 5,171 priced entries in two weeks, 3-minute delay, positions capped at 24 hours). Per-account detail is not published. Prices for the out-of-time test come from a different venue than the one where the accounts trade.
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 8Finding 93. Interpretation
Why it failedLessonScope4. Reproducibility
StepsGrade