A wave-based forensic backtest of an ex-hedge-fund trader’s “real market edge” framework — One-Time-Framing (OTF) regime detection and Dalton’s value-area rotation rule — across Bybit crypto perps and Dukascopy FX/metals/indices.
After 18 waves of forensic audit, two findings are Strong: (1) a regime-conditioning effect — value-area fades complete more often on non-trending days, replicated on 8/8 tested assets (crypto, FX, metals, indices); and (2) a fee-and-funding-viable expectancy edge from resizing a reclaim-and-hold rule’s stop against the daily value area, replicated on 6/6 crypto perps (+0.24R BTC, +0.49R ETH full history).
The edge’s yearly expectancy correlates with realized volatility (ρ≈0.76 annual, ≈0.45–0.48 monthly) — low-vol years underperform. A rolling-vol filter/sizing rule improves held-out mean R by +0.03R to +0.11R (permutation p<0.001), though it is a mean shift, not a risk-adjusted one.
Measured execution reality: quoted spread is negligible, but market impact is 1.7–20× the quoted spread depending on order size ($1k–$100k). ETH’s edge survives every tier; BTC’s known 2023 weakness gets modestly worse.
The original framework detects a dominant market regime (trend vs range) in real time. The crypto reversion effect (H1) — short-term mean reversion conditioned on regime — is confirmed on 7/8 tested perps, with TRX the only exception (candidate explanation identified, mechanism search effectively exhausted).
Dalton’s rule: price re-entering a value area it just left and being accepted back inside tends to complete a fade. The Strong §3.2 finding: these fades complete more often on non-trending days — 8/8 assets.
The Strong §3.10 edge: take the reclaim-and-hold setup and size the stop against the previous day’s value-area width instead of a fixed stop. Fee-and-funding-viable on 6/6 crypto perps over full history.
Yearly expectancy strongly tracks realized volatility (low-vol years underperform, high-vol outperform, on every symbol tested). At trade level the relationship vanishes (R²≈0.005) — it is a regime-scale phenomenon, not a per-trade signal.
| Component | Rule | Evidence |
|---|---|---|
| Regime detection | OTF framework: trend vs range classification | Basis for conditioning, not a standalone signal |
| §3.2 fade | Value-area re-entry + acceptance → fade to opposite edge | Completes more often on non-trending days — 8/8 assets |
| §3.10 reclaim | Reclaim-and-hold, stop = daily value-area width, ~10% stop | 6/6 crypto perps; +0.24R BTC / +0.49R ETH fee-only |
| Vol filter (Wave 17) | Skip trades below rolling-vol percentile (P10–P50) | Held-out +0.03R→+0.11R, permutation p<0.001; pooled threshold beats per-symbol (Wave 18) |
| Sizing | Vol-weighted capital tilt (optional) | +0.548R vs +0.508R unweighted, no participation loss |
| Execution | Quoted spread negligible; market impact 1.7–20× spread | ETH survives $100k orders; BTC second-order drag |
| Finding | Markets | Status |
|---|---|---|
| §3.2 VA-fade regime conditioning | 8/8 — crypto, FX, metals, indices | Strong |
| §3.10 reclaim-and-hold edge | 6/6 crypto perps (BTC, ETH, LINK, STX, XMR, ALT) | Strong |
| H1 OTF crypto reversion | 7/8 crypto perps confirmed | Strong |
| H2 timeframe-scale flip | Pooled crypto held out at 18–36mo; Gold leg infeasible | Exploratory |
| H5 combination check | BTC/ETH confirmed null | Null |
| Measure | Result | Note |
|---|---|---|
| §3.10 expectancy (fee-only, full history) | +0.24R BTC / +0.49R ETH | 2020/2021–2026, 10% stop, daily-VA risk unit |
| Vol filter improvement (held-out) | +0.03R → +0.11R | Monotonic with aggressiveness; p=0.0006 (P30), p=0.0033 (P50) |
| Monthly vol-expectancy correlation | ρ≈0.45–0.48 | R²≈0.22–0.24, p<10⁻¹⁸ across 349 symbol-months |
| Trade-level vol-expectancy | R²≈0.005 | Regime-scale, not trade-scale |
| Market impact vs quoted spread | 1.7–20× | $1k→$100k order tiers; ETH survives all |
| BTC 2023 dip | not anomalous | p=0.317 block-bootstrap; driven by low-vol regime, not a fixable bug |
STRATEGY_AUDIT.md rewritten each wave (currently v16); WAVE*_FINDINGS.md is the append-only evidence trail. No result is trusted beyond N=1–2 until checked across the full tested set.
Bybit klines (last-traded price), funding and OI for 7 perps; Dukascopy for FX/metals/indices; Bybit L2 order-book archive for execution-cost measurement.
Block bootstrap and circular-permutation tests (e.g. BTC 2023 dip: p=0.317), held-out absolute-date cutoffs, permutation tests on filter rules (20,000 draws).
Wave 15 measured quoted spread (negligible); Wave 18 built a book-walking simulator to measure market impact at $1k/$10k/$100k tiers.
# read the current state of belief first
cat STRATEGY_AUDIT.md # v16 — supersedes everything else
# reproduce a specific wave's analysis (e.g. H1 / H5 / H6)
cd backtest && pip install -r requirements.txt
python scripts/run_h1.py && python scripts/run_h5.py && python scripts/run_h6.py
python scripts/regime_classifier.py && python scripts/volume_profile.py
# public dashboard
open docs/index.html # GitHub Pages: docs/