How many trades until you know?

The short answer

To just clear zero, a log needs about 3.84 × (swing ÷ average)² trades, and about 7.84 × (swing ÷ average)² for four chances in five that its 95% range clears zero, if the average really is that size. Halving the average quadruples the count. The rule needs no data; only the example does.

The example is a back-tested call-wall breakout, rebuilt at the 5-minute mark. Its 312 trades averaged +0.11R with a swing of 1.8R (both rounded for the arithmetic), and their 95% range, −0.12R to +0.35R, includes zero. At those numbers 1,029 trades would just clear zero and 2,100 give four chances in five: about 191 years in one stock and 22 across nine. The earlier version of this page said 582 and 1,188. A small average cannot be proven by trading it.

What measured ranges look like

The rule goes long at 15:45 when price has pushed above the call wall (the strike with the most call-option gamma) by up to one implied move, with a stop half an implied move below the wall. On nine stocks (45 of 84 possible stock-years) it made 312 trades, averaging +0.114R. The study’s 95% range, from resampling its trades, runs −0.120R to +0.348R, a half-width (half the distance from the low end to the high end) of ±0.234R; the plain formula, 1.96 × 1.850R ÷ √312, gives ±0.205R, so the study’s range is about 14% wider.

Every range includes zero: all trades, both price eras, quarterly and other months, and each of the six stocks with 20 or more trades. Cutting only widens the doubt, from ±0.234R (all trades) to ±0.743R (INTC, 29 trades). A half-width times √trades runs from 3.3 to 4.5, median 3.6, so a half-width is about 3.6 ÷ √trades; against the plain formula the ranges ran 3% narrower to 21% wider. The chart plots these 11: one rule’s trades cut different ways, not 11 studies.

Trades needed for an average you care about

Turn the rule around. An average of A needs (1.96 × swing ÷ A)² trades to just clear zero, and about half of such samples would still cross it; for four chances in five, use (2.8 × swing ÷ A)². The table uses a swing of 1.8R, with the average to two places and the swing to one (1.8497R counts as 1.8R, +0.1135R as +0.11R). At the study’s own slightly wider range, +0.11R needs about 1,410 and 2,890 trades.

The bigger averages in the table are thresholds, not findings: no cut with 20 or more trades averaged above +0.339R (NVDA), and its range includes zero. At 30 trades the half-width is ±0.64R, at 100 ±0.35R. Two real stretches of 40 trades show it: the first (May 2, 2018 to Aug 6, 2019) averaged −0.165R (±0.573R), the next +0.294R (±0.544R), against +0.114R over all 312; neither can be told from zero or from the other. Cut into seven stretches of 40, the same trades averaged from −0.440R to +0.747R.

Sources: NIST/SEMATECH e-Handbook of Statistical Methods: Confidence Limits for the Mean; NIST/SEMATECH e-Handbook of Statistical Methods: Sample sizes required.

Years, not trades

Divide by how often the rule trades. In one stock (5.37 trades a year), 1,029 trades takes about 191 years. Pooling nine stocks (46.8 a year) cuts that to about 22, but same-day trades in different stocks move together, so the true wait is longer. The pool is nine of the earlier twelve stocks, so its pace is not like for like with the earlier version’s. Pooling also averages across stocks: among the six with 20 or more trades the average ran from −0.379R (INTC, 29 trades) to +0.339R (NVDA, 71), each range includes zero, so this page does not read that order as a difference. NVDA, MSFT and PFE supplied 124% of the rule’s total R; the other six together lost 8.5R.

Slower rules are rarer still: in a separate 2021–2024 test of 480 rule versions, not rebuilt here, a typical version that traded at all made 19 trades in four years held up to 5 sessions and 13 held up to 21.

Rates and gaps use the same arithmetic

A hit rate has a 95% range of about ±100 ÷ √trades percentage points for a rate near half: ±18 at 30 trades, ±10 at 100. The rebuilt rule’s gap over its naive trade ran −0.273R to +0.198R (±0.235R) at 312 trades, about as wide as the range on its own average.

Run it on your own trades

Six spreadsheet steps on one written rule: put each closed trade’s result in column B (rows 2 to 41). Steps 1 to 3 repeat “Score your own trades” in “Read the setup scorecard in Context”; the trades-needed arithmetic starts at step 4. Step 6 gives 40 real trades to test your formulas on.

• 1. Result in R = profit ÷ risk, after commissions and slippage. Risk is the distance from entry to stop times shares (for an option bought outright, the premium paid; for a spread, the most you can lose). Leave out any trade with no stop written before entry.
• 2. =COUNT(B2:B41) gives your trades, =AVERAGE(B2:B41) your average and =STDEV(B2:B41) your swing.
• 3. Your range’s half-width is =1.96*STDEV(B2:B41)/SQRT(COUNT(B2:B41)). If your average, ignoring its sign, is smaller, your range includes zero and the log cannot yet tell the rule from luck. The rebuilt rule: +0.114R against ±0.205R (the study’s wider ±0.234R), so its range includes zero.
• 4. Pick the smallest average you would care about, A. Trades to just clear zero: =ROUNDUP((1.96*swing/A)^2,0); for four chances in five, 2.8 in place of 1.96; swing is the cell holding your STDEV. At the rebuilt rule’s +0.11R and 1.8R they give 1,029 and 2,100; at the first stretch’s own swing of 1.850R, 1,087 and 2,218. Under 30 trades your STDEV is shaky: use 1.8R until you have 30 of your own.
• 5. Divide by your trades a year: 260 at five a week, 52 at one a week, about 1,300 at five a day.
• 6. Check the sheet on real numbers: the R of the rebuilt rule’s first 40 back-tested trades (May 2, 2018 to Aug 6, 2019), in order, for B2:B41. Trades 1 to 10: 1.840, 1.025, 2.389, -1.413, -1.016, -1.048, 1.119, 1.941, 3.190, -6.675. Trades 11 to 20: -1.028, 0.397, -1.601, -1.025, -1.116, 0.964, -1.013, -1.023, 1.383, -0.323. Trades 21 to 30: -1.047, 1.607, -1.012, 2.068, -1.055, -1.027, -1.025, 2.879, -1.027, -1.354. Trades 31 to 40: -1.132, 0.358, -1.071, 4.234, -1.076, 0.254, -1.035, -1.054, -1.034, -1.025. You should get 40 trades, an average of −0.165R, a swing of 1.850R and a half-width of ±0.573R, so a range of −0.738R to +0.408R: it includes zero.

If the answer is years

Trading the rule cannot settle it in a useful time. What can change the answer:

• Paper trades count exactly as real ones, so a log can be built without risking money, but they do not shorten the wait.
• More stocks shorten the wait, but same-day trades move together, so nine stocks are worth fewer than nine independent samples.
• A bigger average needs fewer trades, but the earlier, smaller count is not a better rule: the earlier rule on the same nine stocks averaged +0.172R and needed 341, yet the rebuilt average minus that one, −0.058R (−0.247R to +0.138R), includes zero.

What changed from the earlier version

The earlier version used the same rule on twelve stocks, every session, with the wall and gamma regime read from a book built at the prior close and held all day. This version reads the wall and regime in force in a rebuild at 15:45 for nine stocks and 45 of 84 possible stock-years. These are different constructs, not a correction: on 6,601 covered stock sessions that passed the earlier rule’s filters, the earlier wall was within one strike of the rebuilt wall on 61.6% (both walls pooled).

The middle column runs the earlier rule on the earlier levels over the same nine stocks and sessions, so the move from the left column splits into a smaller universe and a new level. Rebuilt minus that column: −0.058R, range −0.247R to +0.138R, which includes zero, on trades that share only 140 of 312. The averages are not told apart, yet the counts differ threefold, because a small average moves a count a lot. In neither version did a rule pass the study’s screen (none of eight, none of three).

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.

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, DVN and PFE sit in the fallback era only, so the native era rests on six stocks.

The one decision clock is 15:45, so the average, swing and pace describe entries then only; the earlier version also read 09:35 rules, which are not rebuilt. 15:45 is late in the day, when the SPY check matched the call wall on 74% of 15:00 marks and 57% of 16:00 marks.

• No rule qualified: the study scored three, none passed its screen, and its comparisons did not detect a gap over a naive trade (−0.037R, −0.273R to +0.198R) or over a matched line that is not a wall (−0.051R, −0.263R to +0.157R). The +0.11R probably overstates any true average; it is here because the earlier version used the same rule.
• The study re-scores the same 2018 to 2024 sessions and prices as the earlier version, so it is not independent evidence; the family of wall rules has had 21 counted looks, none added here. Nothing after 31 December 2024 was used.
• The nine stocks are a fixed set; IBM, VZ and GPRO could not be rebuilt. One stock’s pace is the median of the nine, each counted only while covered; pooled, 6.4 of the nine were covered on an average session. R is measured against each trade’s own stop distance, so a gapped stop can lose more than 1R, and costs are the study’s modeled commission, slippage and dividends. The arithmetic assumes independent trades and a steady swing. A back-test on rebuilt levels is not a signal or a promise of a future return.