Sixty-Four ES Brackets, Zero Winners, and the Expensive Part Is the Middle

We opened one long at a random minute of each ES cash session for twelve and a half years, then scored the same 3,093 entries under 64 different stop-and-target combinations. Every one of the 64 loses money. The best averages −$13.12 a trade, the worst −$40.61, and not a single cell beats the null of holding to the close.

That is the headline, but it is not the useful part. The useful part is the distribution underneath it, which tells you how much room an ES trade needs before the question of a stop is even worth asking.

Why the entries are random

This is a measurement of cost, not a strategy test.

A bracket does two things at once. It caps a loss, and it also decides which of the trade’s future paths you are allowed to be in. Attach it to a signal and those two effects are tangled: a result could come from the entry, from the geometry, or from an interaction between them. So we removed the entry.

One long is opened at the open of a random minute in each cash session. Nothing about that entry carries information. The bracket is the only thing that changes from cell to cell, which means whatever separates the cells is the geometry and nothing else. The expected value of a random entry is roughly zero before costs, so what the grid measures is what the bracket subtracts from zero.

That is the point. This article does not tell you which setup to trade. It tells you what every setup is fighting before its edge shows up.

The rule

• One long per cash session, at the open of a randomly chosen minute, 3,093 sessions.
• A stop at N ticks below the entry, a target at M ticks above it.
• N and M each drawn from 2, 4, 8, 12, 16, 24, 32 and 40 ticks. Every combination, 64 cells.
• Position held at most 120 minutes, then closed at the market.
• If one minute bar contains both the stop and the target, the stop is taken.
• $29.50 per round trip out of every trade in every cell.

Identical entries in all 64 cells. Only the two distances move.

How far a random ES long goes the wrong way

Before any of the grid, this is the part a reader can use tomorrow. For each entry we measured the worst excursion against the position and the best excursion in its favour over the following hour.

Over the next hour	Against the position	In its favour
Median	16.0 ticks	17.0 ticks
75th percentile	37.0 ticks	36.0 ticks
90th percentile	71.0 ticks	64.0 ticks
99th percentile	193.1 ticks	—

Two things fall straight out of that table.

A four-tick stop is not a stop. 86.3% of entries traded at least 4 ticks against the entry price at some point in the next hour. Six entries in seven. A stop at that distance is not filtering bad trades from good ones, because it fires on nearly all of them; it is an exit that triggers on ordinary noise and charges you the round trip for the privilege. At 16 ticks the share is still 50.9%, so even a stop at the median adverse excursion is a coin flip on its own terms.

The asymmetry people hope for is not there. The adverse and favourable columns in the table above are almost the same shape at every percentile. Price travels about as far one way as the other over an hour, which is what a near-martingale looks like when you measure it. So there is no stop distance at which the trade is structurally more likely to reach the target first. You cannot choose your way to a favourable bracket. You can only choose how the same distribution gets converted into fills.

The chart below is those two distributions on the same axis, clipped at 58 ticks, with the median and the 90th percentile of the adverse side marked.

The grid

Sixty-four cells, same entries, same costs. 0 of 64 are positive. Best −$13.12, worst −$40.61, median −$23.79.

The best cell is stop 40 / target 24, at −$13.12 a trade with t = −2.27. It is the least-bad cell in a grid of losers, and worth naming only because of where it sits, at one end of a U.

Here are four of the eight stop rows, across all eight targets, in average P&L per trade.

Stop ↓ / Target →	2	4	8	12	16	24	32	40
2 ticks	−$35.28	−$30.11	−$24.17	−$20.69	−$18.24	−$16.18	−$14.26	−$15.40
8 ticks	−$40.23	−$34.20	−$29.11	−$26.32	−$25.45	−$23.76	−$21.69	−$22.85
16 ticks	−$39.93	−$34.00	−$29.45	−$25.64	−$23.81	−$21.57	−$20.29	−$20.79
40 ticks	−$37.19	−$32.11	−$27.19	−$19.55	−$15.76	−$13.12	−$14.83	−$13.92

The full 64-cell heatmap is below, with every cell’s average printed in it. The scale is sequential because there is nothing positive to diverge from: dark is worst, pale is least bad.

The shape is the finding, and the shape is a U. The expensive part of the grid is the middle. The 2-tick stop row averages −$21.79 a trade and the 40-tick row −$21.71, eight cents apart at the two ends. Every one of the six rows between them is worse, and the 12-tick row is the worst of the eight at −$28.22. The best cells split the same way: stop 40 / target 24 leads at −$13.12, and the third-best cell in the whole grid is stop 2 / target 32 at −$14.26.

The two ends get there by opposite routes. A 2-tick stop fires on nearly everything, since 86.3% of entries trade at least 4 ticks against the entry, but each firing costs 2 ticks plus the round trip and no more. A 40-tick stop sits out past the 75th percentile of adverse excursion, so it rarely fires and the outcome is decided by where price went rather than by the noise on the way. The middle rows pay both bills. A 12-tick stop is wide enough that every hit costs 12 ticks, and still inside the 16-tick median adverse excursion, so it keeps getting hit. Neither end is managing risk better than the other. They are cheaper ways of being wrong than the distances in between.

The null

The baseline is the same entries with no bracket at all: buy, hold to the session close, pay the same commission once.

Hold to the close
Trades	3,093
Win rate	51.1%
Average	−$8.36
Total	−$25,843.50
Profit factor	0.975
Sharpe	−0.12
t	−0.40
p	0.6857

Cells beating it: 0 of 64. The best bracket in the grid is worse than doing nothing.

The honest qualifier belongs next to that sentence. Holding to the close is itself slightly negative, at −$8.36 a trade. So the comparison is between a small loss and a larger one, not between a loss and a profit. At t = −0.40 and p = 0.6857 the hold-to-close result is statistically indistinguishable from zero, which is exactly what a random entry should produce: the $8.36 is the round trip and the drift working against each other, and nothing else.

Nothing in this article makes money. That is the correct outcome. If a random entry had produced a profitable bracket, the finding would have been a bug in the fill logic.

What this is not saying

This is not an argument against stops.

A stop does a job the grid cannot score: it bounds the worst trade. The worst single outcome in the stop-40 rows is −$529.50, and in the hold-to-close baseline it is −$7,754.50. If you are sized such that one bad session can end the account, that difference is the only number on the page that matters, and no expectancy calculation overrides it.

What the grid says is narrower and harder to argue with. A stop cannot create an edge. Once the entry carries no information, every tick of the bracket is a cost decision, not a risk-management one — it changes which part of the same distribution you get paid on, and it charges you for the change.

It also follows that this grid is not a lookup table. A setup with a real edge changes the entry distribution, and a different distribution has a different best cell. The right way to read the numbers here is as the surface a strategy has to climb out of: this is what the bracket costs when the entry adds nothing, so anything a strategy earns has to clear that first.

Where this sits in the series

This is the machinery underneath two earlier results.

The three-setups article found win rates of 77.0%, 57.7% and 72.9% on setups that all lost money. The wickless candle study was the extreme version: 91.1% of trades reached their target and 16.5% of them made money.

Both looked like anomalies at the time. They are not. Read down a column of the grid above, holding the target fixed and widening the stop, and the win rate marches while the money stays negative throughout. In the 4-tick target column it climbs from 32.5% with a 2-tick stop to 87.7% with a 40-tick one, and that last cell loses $32.11 a trade.

The 2-tick target column is the same point at its limit. All eight of its cells win 0.0% of the time, because a 2-tick target grosses $25.00 against $29.50 of cost, so a trade can reach its target and still book a loss. Pick the bracket and you have picked the win rate. The two are the same choice, and the P&L is settled somewhere else entirely.

What we changed

Nothing in the book. No sleeve was rebracketed on the strength of a random-entry grid, and it would be a mistake to rebracket one: our sleeves have entry information, and this test deliberately has none.

What changed is a check. Any new sleeve whose stop sits inside the 16-tick median adverse excursion now has to justify it explicitly, because we know what that distance does to a trade with no edge. It is not a veto. It is a question with a number attached.

Our ES tick archive, minute bars and trade prints back to 2014, is in the historical data packages.

Methodology: ES cash-session minute bars from our own tick archive, 3 January 2014 to 9 September 2026, 3,093 sessions. One long per session at the open of a randomly chosen minute; excursions measured over the following 60 minutes; bracketed positions held at most 120 minutes and then closed at the market. Stops and targets from the set 2, 4, 8, 12, 16, 24, 32, 40 ticks, all 64 combinations on identical entries. When a single bar contains both the stop and the target, the stop is taken. One contract, $4.50 commission plus two ticks of slippage per round trip, $29.50 on ES; one ES tick is $12.50. Sharpe is annualised on the trade rate.