← all research
STUDY

"Vol-Regime Overlay on gate60 (price-only, daily)"

2026-08-25

Window: 2023-04-19 .. 2026-07-31 (1200 days, 352 gate60-long days) | features: RV10/RV60 term structure, z(RV10,90d) vol-of-vol, ATR%(14) percentile rank vs trailing 252d | all shifted +1d (decision sees bars <= D-1) | gates on long days only, gross

VariantDaysLongDaysRetSharpePFMaxDD
baseline (gate60)1200352+181.60%1.061.33-26.71%
skip rv>=0.851200111+109.07%1.341.93-16.45%
skip rv>=1.01200183+102.64%1.021.46-23.47%
skip rv>=1.151200250+169.00%1.171.44-23.37%
skip rv>=1.31200290+171.89%1.111.38-26.82%
skip rv>=1.51200314+204.35%1.161.39-26.60%
skip rv<0.71200300+163.40%1.041.35-26.71%
skip rv<0.851200241+34.69%0.451.15-28.73%
skip vov>=-0.5120090+86.98%1.312.07-11.26%
skip vov>=0.51200205+168.24%1.281.56-22.51%
skip vov>=1.01200264+169.63%1.151.42-30.62%
skip vov<-0.51200262+50.60%0.551.18-26.71%
skip atr>=0.81200217+183.29%1.251.53-18.42%
skip atr>=0.91200267+167.07%1.131.42-28.14%
skip atr<0.21200312+112.69%0.861.28-26.71%

gate60-long-day forward returns by feature quintile

FeatureQ1(low)Q2Q3Q4Q5(high)
rv+0.243%+0.923%-0.091%+0.810%-0.116%
vov+0.605%+0.864%+0.193%-0.248%+0.345%
atr+0.612%+0.607%+0.475%-0.069%+0.120%

IS/OOS split at 2025-01-01 (rule picked by IS Sharpe, frozen before OOS)

Top-5 rules on IS:

SegmentVariantLongDaysRetSharpePFMaxDD
ISbaseline256+86.70%1.131.29-26.71%
ISskip vov>=-0.560+70.85%2.022.93-8.06%
OOSbaseline96+50.83%0.991.40-18.42%
OOSskip vov>=-0.530+9.44%0.471.32-11.26%

OOS gated long-days: 30 | mean +0.3566% | bootstrap p(gate adds nothing): 0.855 | NULL-R permutation p (2000 circular shifts): 0.253

Rule neighbors (vov family), full window

VariantLongDaysRetSharpePF
skip vov>=-0.590+86.98%1.312.07
skip vov>=0.5205+168.24%1.281.56
skip vov>=1.0264+169.63%1.151.42
skip vov<-0.5262+50.60%0.551.18

Confound controls (full window)

corr(vov, prior-day ETH ret): +0.011

Control variantLongDaysRetSharpePF
control: skip if prior-dayret>3%240+226.29%
control: skip if prior-day ret<0201+6.69%0.201.07
winner AND prior-dayret<=3%74+76.40%
half size when flagged352+107.90%0.841.26

Per-year: baseline vs winner

YearBase retGated retBase LDGated LD
2023+26.35%+27.40%6611
2024+48.86%+28.32%19049
2025+48.80%+10.70%9630
2026+0.00%+0.00%00

Cross-coin: same frozen rule via vectorized twin (own SMA200/ROC60 + BTC ROC60)

CoinVariantLongDaysRetSharpePFMaxDD
BTCbaseline545+157.67%1.131.28-27.21%
BTCskip vov>=-0.5179+33.53%0.611.27-24.43%
SOLbaseline402+268.65%1.021.27-56.29%
SOLskip vov>=-0.5117+4.33%0.181.08-30.28%

Decision memo

VERDICT (2026-08-25): REJECT — family to ledger

Every adoption criterion failed:

  1. IS winner collapsed OOS: Sharpe 2.02 -> 0.47, +70.9% IS vs +9.4% OOS while baseline made +50.8% OOS — the gate skipped 66 of 96 long days and missed most of 2025's trend.
  2. Significance: bootstrap p=0.855, NULL-R permutation p=0.253 (need <0.05).
  3. Cross-coin: same frozen rule cut BTC +157.7%->+33.5% and SOL +268.7%->+4.3% — effect is not general, so it isn't an edge.
  4. Monotonicity tables are non-monotone in all three features.

The scattered full-window grid "wins" (e.g. skip rv>=1.5) are in-sample flattery; the pre-registered IS-pick protocol exposes them. Vol-regime shape (RV term structure / vol-of-vol / ATR percentile) carries no exploitable daily timing information for gate60 longs at this window. Do not re-research without new data or a different mechanism hypothesis.