| Scheme | turn/bar | gross SR(OOS) | net1 SR | net2 SR | net5 SR |
|---|---|---|---|---|---|
| raw / top20 | 1.15 | 1.61 | 0.33 | -0.94 | -4.73 |
| resid / top20 | 0.36 | -0.80 | -1.31 | -1.82 | -3.33 |
| volconf / top20 | 0.30 | -0.36 | -1.02 | -1.67 | -3.62 |
| raw / hysteresis | 0.93 | 0.51 | -0.39 | -1.29 | -3.97 |
| resid / hysteresis | 0.31 | -1.40 | -1.60 | -1.80 | -2.40 |
Turnover is the lever: hysteresis cuts it; net at 2 bps is the real bar. 1y single-regime — directional only.
Verdict: REJECT — nothing improves it
- Beta-hedging (residual 1h momentum) makes gross worse (-0.80 vs +1.61): at the hourly horizon, market beta is the signal; stripping it kills it.
- Volume-confirmation hurts (-0.36).
- Hysteresis barely cuts turnover (0.93 vs 1.15) and slashes the gross edge (0.51): the raw hard top-20 ranking is doing the work, not persistence.
- The best scheme (raw/top20) is still net-negative at 2 bps/side (-0.94) and barely positive at 1 bp (+0.33).
So the hourly signal cannot be engineered past the cost wall with the data available. Raw 1h momentum is the only variant with any gross edge, and it is fragile (IS/OOS disagreement, only beats shuffle at p=0.044). Recommendation: do not trade hourly; keep the validated monthly conviction bot.