One-line conclusion: neither the long-only nor the short-only version shows a stable positive edge out of sample. Data: HTX 4-hour perpetual candles from Dec 2020 to Oct 2026 (about 5.9 years), parameters frozen (EMA20/50, EMA200 filter, 2×ATR stop), 0.6% round-trip costs plus actual funding.
| Long-only | Short-only (control) | |
|---|---|---|
| Trades / win rate | 64 / 25.0% | 77 / 26.0% |
| Total return after costs and funding | +97.2% (12.3%/yr) | −12.1% (−2.2%/yr) |
| Max drawdown | −58.5% | −68.1% |
| Last third of the sample | −11.2% (22 trades) | +45.5% (26 trades) |
| Holding ETH, same period | +329.8% |
- The long-only gain is concentrated: the first 149 days (Dec 2020 to May 2021) contributed +247.9%; the regimes after that compound to about −43%.
- Funding matters: the long side paid funding equal to 50.3% of notional. Without funding it would have been +165.3%.
- Widening the stop to 3×ATR is not supported (+25% full sample, −14% in the last third). This led the team to publicly withdraw that idea (see Corrections).
- With realistic sizing (0.5% risk per trade, ≤10% notional): long-only +9.8% total, −6.7% max drawdown.
- Caveats the researcher listed: perpetual rather than spot prices, simplified handling of funding timing, no extra slippage on stops, regime labels from a lagging SMA200, and a time-split test rather than a truly blind one.