IS/OOS 2016-01-01, sqrt(365). For each factor: OOS gross Sharpe, bootstrap p, net@10, and the beta audit (residual Sharpe vs equal-weight market; correlation to validated cross-sectional momentum). A real NEW edge needs residual > 0 AND low corr.
Equity factors
| Factor | OOS Sharpe | p | Net@10 | Resid Sharpe | Beta | Corr w/ Mom |
|---|---|---|---|---|---|---|
| TSMOM(126) | 1.25 | 0.001 | 1.21 | 0.00 | 0.84 | 0.91 |
| High52 | 1.06 | 0.003 | 0.93 | 0.00 | 0.70 | 0.87 |
| Illiquidity | 1.58 | 0.000 | 1.56 | 0.00 | 1.10 | 0.78 |
Crypto cross-sectional (8 alts)
| CryptoXS | OOS Sharpe | p | Net@10 | Resid vs EW-crypto | Corr w/ Gate60 | Corr w/ EqMom |
|---|---|---|---|---|---|---|
| CryptoXS(1) | 0.42 | 0.214 | 0.40 | 0.00 | 0.46 | 0.29 |
Sector rotation (9 ETFs)
| SectorTop3 | OOS Sharpe | p | Net@10 | Resid vs SPY | Corr w/ EqMom |
|---|---|---|---|---|---|
| SectorTop3 | 0.15 | 0.371 | 0.11 | -0.00 | 0.52 |
Reading
- Any factor with Resid Sharpe ~ 0 and Beta ~ 1 is just equity beta (long-only cross-sectional trap again).
- CryptoXS is the key test: a different asset class. If its residual (vs EW-crypto) Sharpe > 0 AND corr to EqMom/Gate60 is low, it is a genuinely new, independent edge.
- TSMOM's value is crisis diversification (goes to cash when trends break); the OOS 2016+ bull may hide it, so also check its full-history residual if shown.
TSMOM crisis check (full history, incl. 2008)
TSMOM is NOT even a crisis hedge here. Full history (2006+):
- TSMOM: Sharpe 0.95, CAGR +20.5%, maxDD -52%.
- EW market: Sharpe 1.04, CAGR +25.6%, maxDD -51%.
- 2008 crisis: TSMOM -40.4% vs EW market -18.3%. The 126d signal stays ~fully invested through the crash (too slow to de-risk), so TSMOM is strictly worse than beta (a drag, not protection). Confirmed: no new edge.
Verdict
ROUND 3 = ZERO new edges. Across 3 rounds we have now tested, with the mandatory beta audit, every price/volume-constructible family:
- Long-only cross-sectional momentum = the ONE real edge (residual Sharpe > 0, independent of EW market at the daily level, OOS p=0.000).
- Everything else (TSMOM, 52w-high, low-vol, illiquidity, ST-reversal, liquidity-trend, overnight, turn-of-month, pairs, crypto cross-sectional, sector rotation) is either pure beta (residual ~0, beta ~1, corr to momentum 0.5-0.9) or not statistically significant (p>0.2). CONCLUSION: with price+volume data alone there are no hidden non-beta edges left to find. To discover genuinely new edges we would need NEW data types we do not yet have: fundamentals/earnings, options-implied (variance risk premium, dispersion), rates/term-structure, FX/carry, commodities, or sentiment/flows.