Research status recorded: 2026-08-15
39.9%
break-even 29.2% at its realized average win/loss
re-measured (top 10 coins · 1H · 2026-11-20): 52.25%
Percentage of trades that were profitable
1.61
re-measured (top 10 coins · 1H · 2026-11-20): 1.02
Gross profit / gross loss ratio
38.2%
Largest peak-to-trough decline
as published (50 coins · 4H · 2026-05-04)
Marks next to a metric: ~ = borderline (PF between break-even and 1.5; win rate ≤5 pts above break-even, or 50–55% if none shown) · ! = below the bar, or the strategy's research status (dated on its page) is below the bar — that dated record outranks this number. No mark = clears it and its research status is not below the bar.
The dated profile on this card reports bull/bear windows and half-year walk-forward losses. Read those measurements separately; an aggregate PF does not remove losses in a smaller window.
Card re-measured: 2026-11-20 · 978 trades analyzed
Overview
Keltner Channels plot an ATR-scaled envelope around an exponential moving average. This preset waits for a squeeze — the Bollinger Bands compressing inside the Keltner channel, i.e. price volatility (σ) running low relative to the ATR envelope — and then enters SHORT when price breaks down through the lower band — trading the directional expansion, not a fade.
Original 2026-05-04 sweep (1,680 combinations × 50 coins, IS/OOS split) selected SL 3% / TP 10% and reported an OOS profit factor (1.91) above IS (1.76). At the time we read that as evidence the edge was structural. A fresh out-of-sample re-test on 2026-06-28 (product engine, dedicated regime-robustness + 5-axis adversarial kill) retracted that conclusion: once the result is decomposed against the market regime, the SHORT performance is bear-beta — profit from a falling market rather than a regime-independent squeeze edge. The backtest is reproducible; we no longer claim a confirmed live directional edge.
How It Works
- Channel setup — 20 EMA center, ATR × 1.5 for upper / lower bands
- Squeeze detection — Bollinger Bands (20, 2σ) inside the Keltner channel (compressed volatility)
- Trigger — close breaks below the lower Keltner band (downward expansion out of squeeze)
- Entry — SHORT on the next bar’s open after the trigger
- Exit — TP 10% / SL 3% / 12-bar (2-day on 4H) timeout
- Risk — 1:3.3 R:R (3% risk, 10% reward), accepting 60% loss rate
Why It Works (Thesis)
Squeezes precede the largest directional moves. When the ATR-based channel tightens (volatility compression), potential energy builds. The lower-band break signals sellers have won the compression battle. With a 3% SL, false breaks are cheap; with a 10% TP, the real expansions are captured in full.
The OOS > IS result (1.91 vs 1.76) originally looked notable. But the 2026-06-28 re-test shows it is not robust: the May 2025–May 2026 OOS window was itself a falling market, so a SHORT book made money from direction, not from the squeeze filter. Adjusted for that bear-beta, the standalone squeeze edge does not survive — treat this thesis as unconfirmed.
Results (2-year backtest, IS/OOS split, measured 2026-05-04)
| Metric | IS (May24–May25) | OOS (May25–May26) | Combined |
|---|---|---|---|
| Total trades | 447 | 476 | 923 |
| Win rate | ~42% | ~38% | 39.9% |
| Profit factor | 1.76 | 1.91 | 1.61 |
| Coins profitable | 39/50 | 38/50 | — |
Caveats
- Failed fresh out-of-sample regime-robustness (2026-06-28). The earlier “OOS improves / structural edge” claim was retracted — the SHORT result is bear-beta, not a confirmed standalone edge.
- 2026-06-28 measurement: the out-of-sample regime check did not support the earlier cross-regime claim.
- Taken off the curated roster on 2026-08-15.
- Record scope: the measurements above describe their historical settings and windows. Machine-readable research fields remain in the strategy record; they are not a permanent judgment about other settings or windows.
- The backtest figures above (PF 1.61, 39.9% win rate, measured 2026-05-04) still reproduce. A strategy’s fate and the validity of its backtest numbers are different questions.
- Low win rate (40%) is inherent to the 1:3.3 R:R structure. The equity curve has frequent small losses punctuated by larger wins.
- These historical backtest measurements do not establish live execution performance or a current directional edge.
- 4H timeframe means about one signal per coin every 40 days on average (923 trades / 50 coins / 2 years) — far lower frequency than 1H variants.