- RTH 6 bars/day, 4410 bars, 509 names
- Split (fixed): IS < 2025-10-13, OOS >= (thin)
| Mech | Book | Seg | Ret | Sharpe | MaxDD |
|---|---|---|---|---|---|
| R1 1h reversal | gross | IS | +12.3% | 0.48 | -17.9% |
| R1 1h reversal | gross | OOS | -33.9% | -1.17 | -49.0% |
| R1 1h reversal | net | IS | -61.3% | -3.05 | -71.1% |
| R1 1h reversal | net | OOS | -98.8% | -13.36 | -98.8% |
| R2 3h reversal | gross | IS | +6.3% | 0.29 | -21.3% |
| R2 3h reversal | gross | OOS | +72.0% | 1.77 | -13.8% |
| R2 3h reversal | net | IS | -43.2% | -1.78 | -57.6% |
| R2 3h reversal | net | OOS | -84.3% | -5.37 | -85.1% |
| M1 1h momentum | gross | IS | +40.6% | 1.26 | -17.9% |
| M1 1h momentum | gross | OOS | +59.2% | 1.61 | -11.3% |
| M1 1h momentum | net | IS | -51.5% | -2.36 | -63.8% |
| M1 1h momentum | net | OOS | -97.2% | -10.87 | -97.2% |
Expectation: gross may exist; hourly turnover x 10bps kills net. No live change.
Verdict: REJECT (cost-killed, as predicted)
Gross hourly signals look real in this 1-year sample (1h momentum gross OOS Sharpe 1.61; 3h reversal gross OOS 1.77), but net Sharpe is -5 to -13 for every mechanism. Turnover is 0.68-1.15 per bar = ~1000-1700 round-trips per year; at 10 bps/side that is 20-34%/yr of cost at minimum, which no gross edge can cover. The IS/OOS gross numbers also disagree (3h reversal: IS 0.29 vs OOS 1.77), so even the gross is unreliable on a single-regime year.
This closes the hourly question: hourly rebalancing/signalling is not viable for this book. The validated monthly cadence (Sharpe 1.68) stands.