The win-rate trap

The short answer

A closer profit target wins more often and did not earn more per trade. GEXfreak back-tested four call-wall and put-wall rules on nine stocks, 2018–2024, each level rebuilt at the 5-minute mark; nobody placed the trades. The win rate was highest at a 0.5R target in four of four rules. The average R was highest there in none.

At 0.5R the win rate fell 3.9 to 11.4 points short of break-even in three rules. The fourth cleared it by 1.7 points, averaging +0.028R (95% range, the band the true average probably sits inside: −0.061R to +0.117R, which includes zero).

None of the 28 combinations of rule and target beat its naive trade by more than luck can explain across all 28. That is not detected, not proof of no effect: each range on the gap reaches at least ±0.07R either side of it. The levels are rebuilt, not a record of what a member saw.

The four rules and nine stocks

The nine stocks are AAPL, BAC, DVN, INTC, MSFT, NKE, NVDA, PFE and TSLA; IBM, VZ and GPRO are not rebuilt. Each rule reads a call wall or a put wall (the strike with the most call or put option gamma); the figure’s notes give each rule in full. All four are long trades entered at the open of the 15:45 ET bar, with a time exit at the close of the fifth session at the latest. Only the profit target moves: the rule’s own level target, 0.5, 1, 1.5, 2 or 3 times the stop distance above the fill, or none.

The arithmetic

The example throughout is the call-wall breakout, which enters when the prior close was at or below the call wall in the rebuilt map and price is above it by up to one implied move. Average R = win rate × average win − loss rate × average loss. At the 0.5R target: 67.3% × 0.459R − 32.7% × 1.135R = −0.062R. To break even it needed to win 71.2%; it won 67.3%.

A target half as far as the stop breaks even at 1 ÷ 1.5 = 66.7% before costs. A win paid 0.459R, not 0.5R, and a loss cost 1.135R, not 1R: modeled costs (about 0.045R a trade here) and stops gapped through.

All four rules at half a risk unit

The same entries and stops, with the target moved to 0.5R. Every win rate was highest there (58.1% to 73.1%, from 36.2% to 41.0% with no target). Three averages were below zero, but only the put-wall hold’s 95% range sits wholly below zero (−0.136R to −0.005R); the other three include zero. The exception, the put-wall reclaim after a close below, gained +0.028R over 305 trades.

Costs came to 0.029R to 0.045R a trade; adding them back, the three rules short of break-even still average −0.148R to −0.018R at 0.5R.

What the closer target gave up

With no target, all four rules averaged 0.08R to 0.25R more than at 0.5R (0.20R to 0.25R for the three scored rules), and that gap has no range of its own. Most of it is the stocks’ own rise: for the call-wall breakout, a naive trade in the same stocks with no signal rose by 0.23R over the same move, 92% of the rule’s 0.25R. For the three scored rules the naive trade gained 0.92 to 1.24 times as much as the rule (0.92 to 2.78 with the thin put-wall undercut and reclaim).

We checked 28 combinations of rule and target, each against its naive trade, with a many-tries correction (Benjamini and Hochberg, 10% false-discovery rate). None passed. The nearest miss was the put-wall reclaim after a close below at 0.5R: +0.083R over its naive trade (−0.009R to +0.176R), adjusted p 0.99. With 28 tries, about 1.05 gaps this size (p 0.0376) are expected by luck alone.

Not detected is not no effect. A combination’s 95% range on its gap over the naive trade reaches ±0.07R to ±0.27R either side for the three scored rules, so a smaller real gap could be there and not show.

Sources: Y. Benjamini and Y. Hochberg, Journal of the Royal Statistical Society: Series B 57(1): 289–300 (Oxford Academic): Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing.

By clock, price period and quarterly month

One decision clock only, 15:45 ET (the 15:40 rebuild, usable at 15:41); no other clock was run. The table splits the same trades by option-price period (see Limits) and by quarterly against other months. In all 16 splits the win rate was highest at 0.5R, and the average R was highest there in none; the average R was higher with no target than at 0.5R in all but one, the call-wall breakout in quarterly months (−0.258R against −0.030R). Quarterly months are where the accuracy check found the most rebuild error. No range or test was run on a split; one row is thin (under 20 trades at a setting).

What changed from the earlier version

The earlier version tested eight rules on twelve stocks with each level taken from a prior-close book, read once a session and held all day. This one uses the wall in force in the rebuild at 15:45 and keeps the four rules that read only what a Prism chart draws. The two levels are different constructs, not a correction: on the 6,601 sessions both cover, the earlier wall was the same strike as the wall in force at 15:45 for 60.8% of walls, and both walls on 36.5% of sessions, so a rule fires on different sessions.

Re-running the earlier rule on the earlier levels over the same sessions (parity) gives the same pattern: the win rate was highest at 0.5R in four rules and the average R in none, and 0 of 28 combinations passed. Rebuilt minus parity has a range that includes zero for every rule at its own target, but the trades differ (call-wall breakout: 312 against 400), so that does not show the level made no difference.

Score your own trades at other targets

Six steps in a spreadsheet, on one rule’s trades, each with a stop written before entry. Columns: A entry, B stop, C risk =A2-B2 (1R), D the highest price after your entry and before the stop was hit or five sessions passed, E stop touched before the five sessions ended? Y or N, F your R if you did not hit the stop (=(exit-A2)/C2), L costs in R.

• 1. Getting D. Read it off a 1-minute or 5-minute chart from your entry bar on. A daily bar includes prices from before you entered, and one bar can hold both the stop and the target: count the stop first.
• 2. Put the targets to test in G1:K1 (0.5, 1, 2, 3 and 99 for no target). In G2 enter =IF($D2>=$A2+G$1*$C2,G$1,IF($E2="Y",-1,$F2))-$L2 and fill right and down. Each cell is what that trade would have returned at that target: +t if price reached the target first, −1 if the stop came first, otherwise your own R, less costs. Average R is =AVERAGE(G2:G101).
• 3. A real trade: the first AAPL call-wall breakout in the rebuilt grid that ran up and was then stopped out short of +1R. On 3 April 2019 it entered at 195.19 with a stop at 193.59. Its highest one-minute high before the stop was 196.37, +0.737R; it was stopped out on 4 April 2019. Row: A 195.19, B 193.59, C 1.602 (the ledger’s risk), D 196.37, E Y. The sheet returns +0.5 at the 0.5R target (196.37 is at least 195.19 + 0.5 × 1.602 = 195.99) and −1 at 1R (196.79), 1.5R, 2R, 3R, no target and the rule’s own target (201.51). The study’s engine agrees at all seven settings, after costs (+0.445R at 0.5R, −1.055R at the others).
• 4. The break-even win rate before costs is 1 ÷ (1 + t): 67% at 0.5R, 50% at 1R, 33% at 2R, 25% at 3R; costs raise each. Put your cost in R in L2: =round_trip_cost_per_share/C2 (the study’s came to 0.029R to 0.045R a trade).
• 5. Compare average R across targets, not win rates. Your 95% range on a column is =1.96*STDEV(G2:G101)/SQRT(COUNT(G2:G101)). For scale, at 100 trades this grid’s ranges for the three scored rules were about ±0.15R to ±0.17R at 0.5R and ±0.39R to ±0.53R with no target.
• 6. Read the answer. A win rate is noisy too: at 312 trades a 67% win rate carries about ±5 points, and at 30 trades about ±17. A few hundredths of an R between two targets is inside those ranges: no evidence yet. Write the target down before you log trades, and read the comparison once, after 100 or more.

Limits

These levels are rebuilt at each five-minute mark; Prism by GEXfreak keeps no five-minute history of its levels. A tier-0 rebuild uses that morning’s open interest and the prior close’s option prices, re-evaluated at that mark’s market price. Only SPY has a true intraday map (from August 2025), used just to check accuracy. Within $1 of it, tier-0 matched the call wall 77.5% of the time, the put wall 83.3%, the daily flip 68.9%, the weekly flip 73.7% and the monthly flip 66.9%. QQQ and stock accuracy is unmeasured.

That check ran on SPY sessions from October 2025 to August 2026, not on these stock sessions. By hour, the marks either side of this page’s 15:45 decision matched the call wall on 74% (15:00) and 57% (16:00) of sessions and the put wall on 76% and 61%, so 15:45 is a late-day clock. A result here is about the rebuilt level, which can differ from the one Prism by GEXfreak drew live.

Through 28 March 2023 an option’s price is a stored last trade, or a modeled value where none traded (the fallback-premium era); after that date prices are native. BAC (2020–2022), PFE (2021–2022) and DVN (2022) appear only in the fallback era, so the native era rests on six stocks.

• Signal days run from 2 May 2018 to 23 December 2024, so every hold closes by 31 December 2024; nothing dated 2025 or later was read. GEXfreak chose the stocks, all still listed when the study was built.
• Nobody placed these trades. Targets fill on a touch, a same-bar tie goes to the stop, and a stop can be gapped through, so a loss can exceed 1R.
• The grid is one counted look and re-scores the same 2018–2024 sessions as the earlier version, so it is not independent evidence.