MTF trend gate on DONCHIAN_20 — does a higher timeframe help as a bias?

2026-06-24 · status: inconclusive · 3.1s

Hypothesis: A causal daily/weekly trend gate (only trades in the direction of the higher TF) improves risk-adjusted DONCHIAN_20 performance if the filtered-out counter-trend trades are systematically worse — and that BROADLY across several trend definitions (EMA50/EMA200/MACD/weekly), not just one.

Verdict: SIGNAL REAL, FORM OPEN — the per-trade edge is broad (4/4 gates, median +21.4bps): counter-trend DONCHIAN trades ARE worse. But the hard gate hurts (EV thrown away) and the mean-neutral tilt lifts the Sharpe in only 1/12 configs (best ema200 k=0.3: ΔSharpe +0.09). No clear deployable uplift in-sample; before taking it seriously, cross-asset + DOF deflation. The multi-timeframe idea has a real core, but it is no free lunch.

Key metrics

Setup

DONCHIAN_20 live config, re-sim from 2021-01-01. 339 trades (Long 160 / Short 179). Gate = only trades whose direction matches the causal (1d-lagged) daily trend. Baseline = all trades.

Baseline (all trades): ret 2463% · maxDD 16.9% · Sharpe 5.15 · Calmar 146.09 · n 339

1) Hard gate (ON/OFF) — causal, 1d-lagged

r_keep/r_cut = mean per-trade r (bps) of the kept vs filtered-out trades; edge_bps = difference (positive = the filtered-out counter-trend trades are worse).

2) Mean-neutral sizing tilt (keep EV, reweight toward trend quality)

Instead of killing trades: aligned trades size↑, counter size↓, mean_mult=1 (pure timing, no leverage). This is the right form when the counter trades are still positive.