Pre-registered parameters from the original study (validation/carry_basis_overlay.py:DEFAULT_LOOK_H=168, DEFAULT_CARRY_Z=1.5, DEFAULT_BASIS_Z=2.0). Four rolling-origin cuts; the OOS verdict at each cut uses the same frozen params.
Cutoffs (cut date / IS span / OOS span):
- 2024-10-01 / IS 2024-04-01..2024-10-01 / OOS 2024-10-01..2025-04-01
- 2025-01-01 / IS 2024-07-01..2025-01-01 / OOS 2025-01-01..2025-07-01
- 2025-04-01 / IS 2024-10-01..2025-04-01 / OOS 2025-04-01..2025-10-01
- 2025-07-01 / IS 2025-01-01..2025-07-01 / OOS 2025-07-01..2026-01-01
Adoptable: overlay Sharpe lift >= 0.10 AND bootstrap p < 0.05 in at least 3/4 OOS cuts.
Per-coin results
ETH
data: 2024-06-04 .. 2026-08-31 (819d)
| Cutoff | OOS days | OOS long | bucket% | base Sharpe | overlay Sharpe | lift | p | OOS pass? |
|---|---|---|---|---|---|---|---|---|
| 2024-10-01 | 182 | 72 | 1.6% | +0.66 | +0.75 | +0.08 | 0.189 | N |
| 2025-01-01 | 181 | 22 | 1.1% | -0.15 | -0.06 | +0.10 | 0.364 | N |
| 2025-04-01 | 183 | 67 | 2.7% | +2.48 | +2.76 | +0.27 | 0.008 | Y |
| 2025-07-01 | 184 | 74 | 3.8% | +2.04 | +2.62 | +0.58 | 0.009 | Y |
ETH verdict: REJECT (2/4 cuts pass) (2/4 OOS cuts pass; 0/4 cuts had NEGATIVE overlay lift).
BTC
data: 2024-06-04 .. 2026-08-31 (819d)
| Cutoff | OOS days | OOS long | bucket% | base Sharpe | overlay Sharpe | lift | p | OOS pass? |
|---|---|---|---|---|---|---|---|---|
| 2024-10-01 | 182 | 114 | 4.4% | +2.80 | +2.53 | -0.27 | 0.964 | N |
| 2025-01-01 | 181 | 102 | 4.4% | +1.86 | +1.63 | -0.24 | 0.790 | N |
| 2025-04-01 | 183 | 134 | 4.9% | +1.26 | +1.47 | +0.21 | 0.021 | Y |
| 2025-07-01 | 184 | 82 | 3.3% | +0.97 | +1.14 | +0.16 | 0.086 | N |
BTC verdict: REJECT (1/4 cuts pass) (1/4 OOS cuts pass; 2/4 cuts had NEGATIVE overlay lift).
Combined verdict
- Per-coin: ETH REJECT (2/4 cuts pass), BTC REJECT (1/4 cuts pass)
- Total OOS cuts passing across all coins: 3/8 (pooled; per-coin is the source of truth since each sleeve is independent)
- Overall: REJECT (0/2 coins clear the 3/4-cuts threshold; see per-coin rows for details)
Reading
- The original study used one OOS cut at 2025-01-01. That one cut showed Sharpe 0.99 -> 1.30 (p=0.008). With 4 rolling-origin cuts, we ask whether that lift is consistent across different regimes.
- An overlay that lifts OOS Sharpe by >= 0.10 with p<0.05 in only 1-2/4 cuts is consistent with the original being a lucky window. Adoptable here requires >= 3/4 cuts per coin.
- NEGATIVE lift on a cut means the overlay HURT Sharpe on that window. Even one negative lift is a yellow flag; multiple negative lifts suggest the overlay is regime-dependent and not safe to deploy live without a gate.
- The 'bucket%' column is the fraction of long days where the overlay fired a non-passthrough bucket. If it's <5%, the overlay is INERT (the original study's failure mode #1).
- Pre-registered parameters are NOT retuned between cuts. Re-tuning would re-introduce the look-ahead bug class.
Time: 5s.