Appearance
Stress-Exit Pullback — VIX-latched exit on H1 (Phase 2f)
The strongest single result of the Phase 2f strategy search: keep H1's entry untouched and latch the exit to the vol state at entry. Cycles entered while the VIX sits in its top trailing-252-day quintile get a wider, longer-leashed exit (prior-10-day-high target, 15-bar time stop); all other cycles are byte-identical to H1 (prior-M-high, 10-bar stop). Code: pullback_stress_exit.py, registered as pullback_stress_exit.
Pre-registration. Hypothesis, grid (H1's exact 6-point grid — zero new fitted parameters), and every fixed prior (stress =
vix_pct_rank_1y ≥ 0.8at entry, K = 10, stress stop = 15 bars, latch fixed for the cycle's life) were written in the module docstring before the first real-data run, then audited: an independent agent re-applied the patch and reproduced every reported number to four decimals, and cleared the diff for look-ahead.
Why the exit, not the entry
Three Phase 2c overlays proved that vetoing entries on vol/positioning stress destroys value — the vetoed entries are exactly the bounces reversion pays best on. And Phase 2f's i_vol_target_exit showed a static entry-anchored ATR target fails the other way: flush-inflated ATR makes the target procyclically demanding. This candidate threads both autopsies: vol information changes how long the bounce is given to develop, while the target stays H1's path-adaptive prior-high (which decays as lower highs roll in). ~84% of days are non-stress and behave exactly like H1.
Results (walk-forward OOS, 1 tick/side slippage)
Micros 2020-12 → 2026-03 (13 splits); deep e-mini parents 2012-01 → 2026-04 (35 splits, $1M account for leverage parity, gates composed from bars+VIX only):
| Metric | MES | H1 MES | MNQ | H1 MNQ | ES parent | H1 ES | NQ parent | H1 NQ |
|---|---|---|---|---|---|---|---|---|
| OOS Sharpe | 1.06 | 0.84 | 0.76 | 0.86 | 0.77 | 0.74 | 0.60 | 0.58 |
| Sharpe 90% CI | [0.57, 1.69] | [0.26, 1.31] | [0.42, 1.14] | [0.25, 0.97] | ||||
| Max drawdown | −2.4% | −2.8% | −4.9% | −4.9% | −4.1% | −4.9% | ||
| Total return | +14.0% | +10.5% | +19.1% | +20.5% | +20.1% | +27.7% | ||
| Trades / exposure | 70 / 32% | 71 / 30% | 64 / 28% | 66 / 27% | 158 / 26% | 170 / 29% | ||
| Sharpe at 2 ticks / next-open | 1.05 / 1.02 | 0.75 / 0.72 | 0.76 / 0.77 | 0.60 / 0.60 | ||||
| corr to H1's OOS returns | 0.95 | 0.92 | 0.84 | 0.82 |
Reading
- 3 of 4 spans improve, and risk never worsens. MES +0.22 Sharpe with a better drawdown; both 14-year parents marginally above H1 with CIs excluding zero. This is the only structural edit in eight improvement candidates that added value anywhere, and it validates the campaign's central mechanism lesson: vol information belongs in the exit, not as an entry veto.
- MNQ degrades (−0.10), and the autopsy exonerates the exits themselves: on stress days the delta is ≈ 0; the loss is knock-on — a 15-bar stress hold occupies the book longer, delaying re-entries, and the shifted later cycles lost, concentrated in 2024's choppy Nasdaq retests (parameter selection was split-for-split identical to H1's, so selection drift is ruled out).
- Why this is not adopted as champion. The MES/MNQ split is one realized path of a noisy mechanism — there is no economic reason stress-exits work on the S&P and fail on the Nasdaq. It is also the best of ~10 improvement attempts, so its +0.22 carries family-wise selection risk. Fleet-wide adoption is a portfolio wash (equal-weight both-micro Sharpe 0.926 vs H1's 0.924).
Verdict — registered finalist; forward-validate before adoption
Recorded in the ledger as borderline / QUALIFIED. The recommended path (Phase 2f judge): run pullback_stress_exitbeside H1 in any paper-trading phase on MES, and adopt only if the live delta is consistent with the backtest's +0.22. Do not replace H1's exit fleet-wide on this evidence. The open follow-up (a new pre-registration, not a tweak): whether the knock-on re-entry blocking — not the exit definition — is the true bottleneck, which points at portfolio-level capacity rather than exit design.
Paper trading — signal-only shadow on MES (Phase 2g)
Per the verdict above, the stress-exit runs in the paper stack as a signal-only shadow on MES beside H1: it logs intended entries/exits and virtual P&L (no capital, role="shadow" in the ledger), so its live delta vs H1 accumulates at ~zero marginal cost. Pre-registered adoption bar: fleet-wide adoption of the VIX-latched exit is not discussed until the live delta is consistent with the backtest's +0.22 (MES 1.06 vs 0.84) — and never on the strength of MNQ, where it degrades. Because the two rules diverge only on stress-latched entries (a minority of ~13 trades/yr), expect years to a verdict; the value is that it rides along for free. Details: paper-stack note.