#58 — Indicator screen → momentum×efficiency → regime-gated daily mean-reversion

Session 079, 2026-05-28/29

Verdict: negative · ⛔ closed negative — the regime-gated survivor was retracted at the premise (Coda)

An honest spread-deducted screen of new indicators found only daily-horizon mean-reversion positive; a wiggle/vol regime gate produced the project's high-water mark, +13.0p/trade daily-clustered, t=2.25, p=0.025 — later retracted at the premise: the Coda's field-scan re-run (2026-07-17, Markov gate) re-measured the underlying daily-MR signal at +1.1p gross, negative net of the daily rollover financing cost.

Hypothesis — what it tests

Whether any new indicator (TRIX, Vortex, Fisher, momentum×efficiency) carries a spread-net directional edge at any timeframe or bar type — and, once only daily mean-reversion survived, whether a volatility/choppiness regime gate can turn that unstable effect into a stable, statistically defensible edge.

Description

The user asked to explain and test TRIX and Vortex and brainstorm new indicators. A deliberately honest test harness — spread-deducted average P&L per trade at a bounded fixed hold, avoiding the TP-only loss-deferral trap that fooled the momentum book — showed TRIX, Vortex and Fisher have no spread-net edge at any timeframe (M15–D1) or bar type; the only positive was daily-horizon mean-reversion (RSI2 contrarian, +4.73 p/trade net, IC +0.137, 10/12 pairs). Deepening it (z-score vs SMA10, ±2σ, 3-day hold) exposed regime dependence — walk-forward folds went +6.8/−1.8/+8.6/+23.0 — while the user's momentum(pips/min)×efficiency idea died in all three roles (continuation entry, fade entry, exit overlay). The redemption was using efficiency as a REGIME gate: trade the daily MR only when the market is both calm (low realized-vol percentile) and choppy (low intraday efficiency), which made all four walk-forward folds positive. Hardening with daily-clustered statistics and block bootstrap gave the honest number: +13.0p/trade, t=2.25, p=0.025 — the only signal of the session to clear overlap-robust significance, all-folds-positive WF, bounded margin and an economic rationale, but marginal (5%, not 1%) and thin (~63 trades/yr), so it earned a paper-test spec rather than live deployment.

Key result

TRIX/Vortex/Fisher: −1.6 to −12 p/trade at every TF/bar type; daily MR (RSI2): +4.73 p/trade net, IC +0.137, 10/12 pairs; ungated WF folds +6.8/−1.8/+8.6/+23.0; G6 wiggle gate (eff<0.5 & vol<0.6): all 4 folds positive, sealed OOS +31.9p t=3.43 62% WR 8/12 pairs; honest daily-clustered +13.0p, t=2.25, bootstrap p=0.025; ~63 trades/yr, max 9 concurrent; any stop halves the edge (worst MAE −592p) | Coda re-run 2026-07-17 (Markov gate, not this weighted-sum config): underlying daily-MR signal +1.1p gross — negative net of daily rollover cost; the flag for a paper test was closed at the premise.

Indicators

TRIXVortexFisher TransformRSI(2)z-score vs SMA(10)Kaufman Efficiency Ratiorealized-volatility percentilelag-1 return autocorrelation (ac60)momentum (pips/min)path efficiency (|net|/path)

Algorithms

grid search4-fold walk-forwardblock bootstrapdaily-clustered t-statisticsevent-bar (volume/tick) samplingregime conditioning

Code

Interactive version (search, filters, figures): the experiment explorer.

← #57 Finite-margin re-validation of the live momentum book #59 Oracle optimal-trade ceiling (GBP_JPY S5) →

This page is one experiment from the audited record behind the book The Edge That Wasn't — get the book · about the project.