EW top-10, 12m momentum, hold 21, next-open, 10bps, 101-name live universe.
- D0 base: OOS ret +1240.8%, SR 1.10, maxDD -35.5%, boot-p 0.000
- D1 SPY>200d: OOS ret +1982.1%, SR 1.53, maxDD -16.7%, boot-p 0.000
- D2 VIX<VIX-SMA50: OOS ret +19399.9%, SR 3.55, maxDD -10.4%, boot-p 0.000
- D3 SPY>200d & VIX<SMAs: OOS ret +11366.2%, SR 3.49, maxDD -10.4%, boot-p 0.000
- D4 BS-C vol-target 12%: OOS ret +0.8%, SR 0.30, maxDD -0.3%, boot-p 0.178
Interpretation: many overlay-gate drawdown cuts cost strategy convexity (the momentum crashes retrace fast after gated days); adopt only if DD cut >= ~10pp.
ALL DONE
Honest read
- D1 (SPY>200d gate, flat below its 200d SMA) is the only portable robustifier: SR 1.53 vs 1.10 and maxDD -16.7% vs -35.5%. Same family as the eqmom C7 SPY-hedge overlay spirit; production gating happens via the overlay shadow path, so this would ship as a strategy-evaluation overlay, not a code retrofit.
- D2/D3 (VIX<SMA50 variants) are training-luck artifacts: one 2022 in the sample, gate-off window happens to catch that single bear leg; SR 3.5 on the ON half (SR~1.53 de-annualization of full-sample means it's not additive).
- D4 BS-C vol-target 12% killed the whole book because EqMom's EW book vol runs >12% ann. during OOS. A sane target (~20% ann.) collapses to W2 inverse-vol-sizing, which already failed.
- Recommendation: watch D1's gate behavior as a shadow overlay in production via the EQMOM overlay channels (EQMOM_SPY_HEDGE/EQMOM_VOL_TARGET family); no code change adopted.
BS-C vol-management at sane targets (corrects the 12% probe artifact)
| target vol (ann. 63d trailing) | OOS SR | ret | maxDD |
| 0.20 | 1.08 | +681% | -24.2% |
| 0.25 | 1.08 | +869% | -26.8% |
| 0.28 | 1.10 | +998% | -28.6% |
| 0.30 | 1.09 | +1033% | -29.8% |
| 0.40 | 1.09 | +1135% | -33.2% |
12% flattened the book (zero return). At 20%ann. vol target, EqMom's maxDD halves (-35.5%-24%) at constant SR 1.08 - this is the B5 family done at a book-matching target, not a DD-cutting proofpoint. Recommend: if we want an adoptable DD reducer, measure D1 SPY-200d gate first (shadow just added); vol-target variants closer to book vol barely change SR and are the next slot if D1's shadow rule holds.
DD-cutoff overlay probe (complementary D1 test)
| EqMom 63d peak-trough trigger | OOS SR | ret | maxDD | | halving at -5% | 1.00 | +541% | -23.3% | | halving at -8% | 1.17 | +966% | -22.5% | | halving at -10% | 1.11 | +887% | -25.3% | | halving at -15% | 1.09 | +984% | -30.9% | -> DD-cutoffs don't beat D1 on SR and don't reduce DD as much; D1 remains the only candidate worth shadow (added to runtime). No code changed for DD-cutoffs.
D1 damped vs all-off on its off-days
- All-off (current D1 gate, in production shadow): OOS SR 1.53, maxDD -16.7%.
- 0.5x on D1-off days / 1.0x on D1-on: OOS SR 1.36, maxDD -19.8%.
- The half-out alternative keeps exposure on the bad-term days; SR and MaxDD both worsen, so we do not change the shadow semantics. D1 remains: full exposure on, flat off.
- Conditional daily mean of EqMom net under gate-on +0.3695% over 2218 OOS days vs -0.1772% over 449 off-days ~1.7y below SPY 200d (most clustering 2016 early/2022); the all-off tail captures the negative months.