New Bybit perp listings are supported by market makers who are contractually obliged to provide liquidity — commonly structured as a 7-day TWAP over a contract period often quoted at ~90 days. When the obligation ends, MMs withdraw liquidity, which should cause a drop in price, volume and open interest. The original hypothesis was to short new listings after the first seven days and hold for 90 days to harvest that predictable decline.
We tested this rigorously: 349 new USDT perps listed in the last 365 days, daily OHLCV + volume + open-interest, 321 symbols with ≥14 days of history, separated into 162 crypto-native vs 109 stock-ETF perps (AAPL, TSLA, NVDA, QQQ…).
| Cohort | Short d7→14 | Short d7→30 | Interpretation |
|---|---|---|---|
| Stock-ETF (109) | –0.90% (45% WR) | –1.83% (37% WR) | Trend up, OI +10–50×, no edge |
| Crypto-native (162) | +4.13% (59% WR) | +7.87% (70% WR) | Edge present but fat-tailed |
| Horizon | Avg Return | Median | Win Rate | Profit Factor |
|---|---|---|---|---|
| Short d7→14 (n=321) | +2.42% | +0.87% | 54.5% | 1.46× |
| Short d7→30 (n=282) | +4.71% | +4.65% | 58.9% | 1.46× |
| Short d7→90 (n=154) | +28.13% | +40.03% | 81.8% | 4.19× |
Why it breaks down under risk management:
Every single trade (100%) sees an adverse move before paying off. Median MAE 15–20%; many exceed 50–100%.
| Stop-loss | Surviving trades | Expectancy |
|---|---|---|
| None | 100% | +16.0% (d7→30) |
| 5% | 5% | –3.7% |
| 10% | 18% | –1.4% |
| 15% | 31% | +1.5% |
| 20% | 43% | +2.8% |
| 30% | 59% | +6.5% |
Launch day is extremely volatile: average intraday adverse excursion –12%, favorable +19%. Day 0–4 routinely swing 10–20% against the open.
Critical insight: a strong launch (day-0 return > +10%) predicts continuation, not reversal — these symbols average +51% over the next 7 days. The pump is a fade only after it exhausts — which is why condition-based entries (waiting for breakdown) beat time-based entries.
Execution: Enter short at next day's open · Stop = entry + 3.0 × ATR(14), trailing down · Max hold 30 days · 1 trade per symbol.
After Bybit taker fees (0.055%/side) + estimated slippage (≈0.21% round-trip): expectancy +17.20%, profit factor 2.70×.
Grid-search over ATR stop multiplier (1.5–5×), pump threshold (5–40%), hold (15–90d), trades/symbol (1–3):
| Config | N | WR | AvgW | AvgL | Exp | PF |
|---|---|---|---|---|---|---|
| ATR3 · Pump30 · Hold30 (best) | 30 | 73.3% | +37.4% | –37.7% | +17.4% | 2.73 |
| ATR3 · Pump35 · Hold30 | 27 | 74.1% | +39.2% | –35.2% | +19.9% | 3.18 |
| ATR4 · Pump30 · Hold30 | 30 | 73.3% | +37.4% | –49.6% | +14.2% | 2.08 |
| ATR3 · Pump30 · Hold45 | 30 | 70.0% | +37.8% | –47.5% | +12.2% | 1.85 |
| ATR4 · Pump30 · Hold30 (no ATR contraction) | 32 | 78.1% | +39.2% | –58.0% | +17.9% | 2.41 |
Filters that hurt: volume-spike-at-entry (too rare, high stop-out), distribution-day only (21 trades, negative expectancy), EMA20+EMA50 combo (over-filters). Wider stops (ATR4+) raise MAE exposure without better risk-adjusted return.
| Split | Train symbols | Train exp | Test symbols | Test trades | Test exp |
|---|---|---|---|---|---|
| 60 / 40 | 97 | +17.6% | 65 | 3 | +16.0% |
| 70 / 30 | 113 | +17.4% | 49 | 0 | — |
| 80 / 20 | 129 | +17.4% | 33 | 0 | — |
Train expectancy is extremely stable across splits. Out-of-sample counts are small because the signal is rare — a real constraint of this universe.
| Statistic | Value |
|---|---|
| Median expectancy | +17.68% |
| 5th percentile | +5.23% |
| 95th percentile | +28.94% |
| P(expectancy > 0) | 98.9% |
| P(expectancy > 5%) | 95.3% |
| P(expectancy > 10%) | 84.5% |
The edge is statistically significant — not a lucky cluster. But sample is small (30 trades): treat estimates as lower-confidence until 50+ live/paper signals.
| # | Symbol | Return | MAE | Held (d) | Stop hit |
|---|
Top-10 winners avg +53%; the 5 losers avg –57% and are the only trades that hit the stop. BSBUSDT survived a +148% adverse excursion to finish +22.6% — exactly the fat tail the 3×ATR stop is designed to tolerate.
We also tested going long on the pump and flipping to short on distribution (long when price > VWAP & EMA50 after a strong launch; exit long & short on a distribution day: down-day + volume ≥1.5× SMA20 + close in lower third of range).
| Leg | N | WR | AvgW | AvgL | Exp | PF |
|---|---|---|---|---|---|---|
| Long pump | 147 | 33.3% | +65.8% | –21.2% | +7.8% | 1.55 |
| Short after distribution | 138 | 50.7% | +39.7% | –48.8% | –3.9% | 0.84 |
| Combined | 285 | 41.8% | +50.5% | –32.5% | +2.1% | 1.11 |
Verdict: The long pump leg works (huge wins, +66% avg) but with a 33% win rate, and the short-after-distribution leg is negative — distribution-day signals are too rare and enter too late. The pure short-on-breakdown (optimal strategy) is strictly better.