Milkman Trades — Compression Drift Continuation

ES & NQ futures, 3-minute chart — 18 years (2008–2026) + NQ holdout + 2026 forward test
System v1.0 — Frozen 2026-07-10

← Master Sheet milkmantrades.com
01 — The Pattern at a Glance
Squeeze drifts one way → expansion fires the same way → go with it
first 5 squeeze bars set the range new highs keep printing: breach ≥0.5×ATR up, low side holds <0.25 = DRIFT UP expansion bar fires UP = aligned → TRADE enter next 3m open, long stop = entry − 1×(3m ATR14) hold to stop or session close — nothing else SKIP THESE drift up, expansion fires DOWN (anti-drift) suspect print — no trade either way flat squeeze break: continued in testing but FAILED the NQ holdout as a system — skip
02 — Setup (all five must be true)
The signal checklist
  • Squeeze: Saty compression ON for ≥8 consecutive 3m bars (RTH 09:30–15:59 ET; a single 1-bar flicker off doesn't reset the count).
  • Mark the first 5 squeeze bars' high/low. That range is the reference.
  • Drift: later squeeze bars breach one side by ≥0.5×ATR while the other side holds within 0.25×ATR. Breach side = drift direction. (ATR = 3m ATR14, always.)
  • Expansion fires in the drift direction: first bar with compression OFF closes past the midpoint of the last 5 squeeze bars' range, on the drift side.
  • Volatility gate: 3m ATR14 ≥2.0 pts (ES) / ≥2.6 pts (NQ). Below the gate the edge measured smaller than round-trip friction — skip.
03 — Execution (nothing discretionary)
Entry, stop, exit
  • Entry: market at the next 3m bar's open after the expansion bar closes. Everything is known at that close — no anticipation, no chasing mid-bar.
  • Stop: 1× the 3m ATR14 (value at the expansion bar), intrabar. NOT the daily ATR: median stop ≈ 3.5 ES pts (~$173/contract), IQR 2.6–5.1, p90 ≈7.
  • Exit: the stop, or market-on-close of the same session. No target, no trail, no partials — measured worse every way tested; the P&L lives in the runners held to the bell.
  • Sizing: contracts = risk budget ÷ (ATR × $/pt). Risk scales 1:1 with ATR, so size down automatically on hot days.
  • One trade per signal; new signals while in a trade are independent — take them only if your risk budget allows the overlap.
04 — What to Expect (per instrument)
This is a low-win-rate runner system
~21%
win rate — 4 of 5 trades stop out at −1 ATR
+$45–49
net per trade after costs (1 ES / 1 NQ contract)
~135/yr
signals per instrument (ATR-gated, 2008–2026 avg)
  • Expect losing streaks and losing months at any size. Forward 2026 ran −$13.0k (Feb) to +$31.5k (Jun) on 1 ES + 1 NQ.
  • Shorts measured stronger than longs on both instruments (ES +1.28 vs +0.56 pts/trade) — take both sides.
  • Start on MES/MNQ. 18-yr max drawdown was 1,522 NQ pts — about $30k on a full contract, $3k on a micro.
05 — Why We Trust It (evidence chain)
Three gates, in order, no re-fitting between them
GateSampleNNet/Tradet (day-clust)
Discovery (in-sample)ES 2008–262,558+0.91 pts+2.93
Pre-registered holdoutNQ 2008–264,450+2.46 pts+2.85 — PASS
Forward (unseen data)ES+NQ Feb–Jul 2026273+$83+0.66 (in line)
Holdout specs & pass lines were frozen in writing before NQ was touched. Forward window is too short for significance by construction (an 18-yr t=2.85 edge projects t≈0.45 over 5.5 months) — it checks "did it break," and it didn't. Same rules replicate directionally on 1m and 10m charts; net of friction the 3m is strongest (t 1.46 / 3.04 / 1.07).
06 — Measured & Rejected (don't add these back)
Every one of these was tested and made it worse
  • Fading the drift / expecting mean reversion: the original hypothesis. Across 10,648 episodes the drift direction kept winning after expansion — stronger drift, stronger continuation (top quartile +0.113 ATR by bar 3, t=4.3).
  • Trading anti-drift expansions (either direction): weakest group, not statistically separable as a fade. Treat as "stand aside."
  • Flat-squeeze breakouts as a system: best per-event stats in-sample, then failed the pre-registered NQ holdout (t=1.42, shorts ≈0). Chart context only.
  • 10m confluence filter (same-direction = trade / opposite = skip / squeeze = half size): would have roughly halved total P&L on both instruments; buckets not separable (t 0.5–1.9). Run it unfiltered.
  • Profit targets, trailing stops, fixed-bar exits: all reduced net vs stop-else-close. The skew IS the strategy.
★ Bottom Line
The whole system in three lines
milkmantrades.com · Compression/PO by Saty Mahajan · spec: analyst/po_comp_drift_STRATEGY.md ES/NQ futures · costs incl. (0.31 / 0.405 pts RT) · System v1.0 frozen 2026-07-10