Strict OOS windows (book was built on 2020-2026). Sleeves with unseen data: equity + Treasury only. Crypto (cache 2024+) and PEAD (earnings ~2021+) are IN-SAMPLE ONLY and reported as context.
OOS results (sqrt-252 annualization; CAGR annualized)
| Window | Sleeve | Sharpe | CAGR | maxDD | corrEq |
|---|---|---|---|---|---|
| 2005-2016 OOS | Equity | 0.63 | +12.16% | -54.50% | 1.00 |
| 2005-2016 OOS | Treasury (tactical) | -0.91 | -1.03% | -10.61% | -0.06 |
| 2005-2016 OOS | BOOK(eq+TSY tactical) | 0.59 | +6.01% | -30.41% | 1.00 |
| 2005-2016 OOS | Treasury (passive) | 0.80 | +4.60% | -7.69% | -0.38 |
| 2005-2016 OOS | BOOK(eq+TSY passive) | 0.90 | +9.16% | -28.43% | 1.00 |
| BOOK(tactical) vs Equity (boot p) | |||||
| BOOK(passive) vs Equity (boot p) |
| 2017-2020 OOS | Equity | 0.92 | +21.38% | -38.50% | 1.00 | | 2017-2020 OOS | Treasury (tactical) | -0.62 | -0.53% | -2.19% | 0.04 | | 2017-2020 OOS | BOOK(eq+TSY tactical) | 0.90 | +10.71% | -21.10% | 1.00 | | 2017-2020 OOS | Treasury (passive) | 1.04 | +5.23% | -6.99% | -0.22 | | 2017-2020 OOS | BOOK(eq+TSY passive) | 1.16 | +13.98% | -20.77% | 1.00 | | | BOOK(tactical) vs Equity (boot p) | | | | | 0.969 | | | BOOK(passive) vs Equity (boot p) | | | | | 0.908 |
Reading
- Equity leg is robust OOS: Sharpe 0.63 (2005-2016) and 0.92 (2017-2020), consistent with its in-sample ~0.99. The momentum factor survives unseen data.
- The tactical Treasury sleeve is a DRAG OOS (Sharpe -0.9 / -0.6): the 126d-TSMOM 'long-when-trending' rule whipsaws and underperforms passive bonds (passive TSY = +4.2% CAGR, Sharpe 0.76 over 2005-2016). So the combined book does NOT beat equity alone with the tactical sleeve (boot p ~0.97-0.99, not significant).
- A correctly-specified Treasury diversifier WOULD have helped: the passive-TSY book beats the equity leg (higher Sharpe, much lower maxDD, corrEq ~0.0) -- the diversification math is sound; the live TSY sleeve's signal needs to be passive/long-biased, not tactical-flat.
- Crypto + PEAD could NOT be OOS-tested (crypto cache starts 2024, PEAD earnings ~2021). Their in-sample contributions (combined_book2.md, 2020-2026): PEAD Sharpe 1.11 / corr 0.03; Crypto Sharpe 1.29 / corr 0.03. Those edges are UNVERIFIED out-of-sample and remain the book's biggest open risk.
- Caveat: edge_hunt9b (a factor study) touched 2005-2026 equity data, so the 12-1 momentum FACTOR is not perfectly unseen; the EQUITY+TSY COMBINATION and its net-cost treatment are new and unseen here.