An 81% win rate that pays $0.00

This is the strongest probability edge our desk has found in five years of NQ data, and the money test that killed it. If you only ever read one drop, read this one.

The dial

At 10:20 ET, look at how much of its opening move NQ has given back: retrace = (high − price) / (high − low), everything measured through 10:20. Then ask: does the session make a new high after 10:20?

Base rate first — the denominator nobody quotes: with no condition at all, NQ extends its high after 10:20 on 63.8% of days (n=1,403, 2020-12-07 → 2026-07-27). 64% is the coin flip here, not 50%.

Monotonic, every bucket split-stable across both halves of five years, p<0.0001. Honest caveat in the same breath: the ends are partly mechanical — price sitting near its high doesn’t have far to go. Expectation-setting, not discovery.

The primary stat

Stack two conditions: high set in the first 10 minutes AND 70%+ retraced by 10:20.

Result: a new high follows on only 29.9% of days (n=284) — 33.9 points below base rate, p<0.0001, both halves agree (30.3% / 29.5%). Roughly 70% of the time, the high is already in. Same days, the low side: a new session low follows on 84.9% (base 58.1%).

That is slide one of every guru deck ever made. Now the slide they never show you.

The money test

We converted it to trades three ways. Entry at the 10:20 close, short. Real stop, 1.7-point round-trip cost, stop assumed to fill first on same-bar touches. n=284.

That middle row is the headline. Targeting the morning low — the “obvious” trade — wins 81% of the time and produces −0.6 points per trade. An 81% win rate, exactly zero expectancy.

An 81% win rate produced −0.6 points per trade.

*The third row looks closest to alive, and we’re not promoting it either: it’s the best of five exit rules tested on the same 284 sessions, so its p-value is flattered by the search. Uncorrected p=0.057 across five tries is effectively p>0.25. It sits on the watchlist pending out-of-sample data — that’s what a watchlist is for.

Why a 34-point edge pays nothing

The stop distance scales with the exact volatility that creates the setup. The days that give back 70% of a big opening range are violent days, so the stop is wide (median risk on this cell: 130 points). Win often, win small; lose rarely, lose big. The hit rate is real. The payoff structure eats all of it.

Hit rate is not edge. Expectancy is edge. Any stat quoted to you without an EV test after costs is decoration.

What the desk does with a dead trade

We use this cell every day — to not do things. After an early high and a deep give-back, chasing longs is fighting a 70/30 number, so: size down, stop chasing. A stat can fail as a trade and still pay as a filter. Almost everything real the data gives you is a reason to do less.