Issue 2 · Tuesday, October 27, 2026 · 6-minute read
Disclosure: I don't trade the 3 Bar Play, and I have no connection to Jared Wesley or Live Traders. No one paid for or reviewed this test.
Good morning. Last week, we saw how little 50 trades can prove. This week, one of the most popular day-trading setups goes through a real test: 22 stocks, 764 trades, and a final exam it had never seen.
WEEKLY QUIZ Beginner
Last week's answer: C, it's too early to tell
A coin-flipping trader with no edge wins 6 or more of 10 trades about 38% of the time. Ten trades can't separate skill from luck.
The 3 Bar Play risks 1R to make 2R. What win rate does it need to break even, before costs?
Read today's research, then vote. The answer arrives next Tuesday.
LUCK OR EDGE? · THE 3 BAR PLAY · PART 1 OF 2
We tested the 3 Bar Play on 22 stocks
The 3 Bar Play, popularized by Jared Wesley of Live Traders: a big, fast move, one quiet bar near the high, then another breakout. When buyers hold their gains, the idea goes, the move isn't finished.
We tested the rules a computer can follow. The judgment calls a live trader adds can't be backtested, and I'll say so where it matters.
The setup, as we tested it
The long setup on 5-minute bars. Shorts are the mirror image: a big down bar, a pause near its low, and a break below.
1 Bar 1 ignites. Its range is at least twice the average of the previous 20 bars. Its body fills at least 70% of the range, it closes in the top quarter, its volume is at least 1.5 times the average, and it makes a new 20-bar high. It can't come right after two other wide up bars.
2 Bar 2 rests. Its range is no more than half of bar 1's, it stays in the top half of bar 1, and its high is within 10% of bar 1's high.
3 Bar 3 triggers the trade. Buy 1¢ above bar 2's high, only while bar 3 is forming. Stop 1¢ below bar 2's low. Target: twice the risk.
4 The trade ends at the stop, the target or 3:55 PM, whichever comes first. No new trades after 3:30 PM, and one trade at a time per stock.
What we assumed
5-minute bars, tracked on 1-minute data, regular hours only. 22 large US stocks and ETFs: AAPL, AMD, AMZN, AVGO, BA, BAC, GOOGL, INTC, JPM, LLY, META, MRVL, MSFT, MU, NFLX, NVDA, PLTR, QQQ, SPY, TSLA, UNH and XOM.
Costs: $0.99 per order, in and out, plus 1¢ of slippage per share on entries, stops and end-of-day exits.
Results are in R, where 1R is the amount risked, $100 here: +2R is $200, and $1.98 of commissions is 0.02R.
How we kept ourselves honest
Before a single trade, I wrote down the rules, data, costs and pass mark in a dated document. That's pre-registration: it stops you from moving the goalposts.
2023 to 2025 was used only to check the code. January to September 2026 stayed sealed, then was run once. To pass: at least 100 trades, an average trade above zero after costs, and a p-value below 0.05 against random entries. One change after the fact: commissions at $0.99 per order instead of the registered $0.005 per share. The verdict is the same either way.
The results
| Checking the code | Sealed test |
|---|---|---|
Trades | 596 | 168 |
Win rate | 37% | 36% |
Average per trade, after costs | +0.04R | 0.00R |
Likely range of the average | −0.08R to +0.16R | −0.21R to +0.22R |
Total | +23.7R | −0.04R |
Worst drawdown | −28.6R | −30.5R |
Average cost per trade | 0.06R | 0.05R |
Cumulative result after costs, in R, for all 764 trades on the 22 stocks, in time order. Shaded: the sealed 2026 test. Notice the spike in 2026? That's Part 2.
After costs, the strategy needed to win 35.7% of its trades to break even. In 2026, it won exactly that: 60 of 168. It landed on zero.
The luck test: did the pattern beat random entries?
Every real trade got a random twin: same stock, direction, time of day, target and costs, but on a random other day, with a stop sized to that day's volatility.
Across 4,000 sets of twins, random entries averaged −0.03R per trade, and 37% of the random sets did as well as the real 3 Bar Play or better. That share is the p-value: 0.37. To pass, it had to be below 0.05.
Does the chart size matter?
We also locked in 2-, 3- and 10-minute versions. Extra tests are extra chances to get lucky, so each needed a p-value below 0.017 (0.05 ÷ 3).
Average result per trade after costs, with its likely range (95% bootstrap). 2026 p-values against random entries: 0.25, 0.08, 0.37 and 0.17.
None passed, and the averages jumped around between periods, which is what noise looks like.
Reading the result honestly
Luck isn't proof of no edge. 168 trades can miss a small edge. If the true edge were +0.10R per trade, this test would have caught it only about 1 time in 3. An edge of +0.30R, it would have caught about 9 times in 10.
Costs decide the sign. With an average trade of zero, a little more or less in costs decides between a small win and a small loss. At $50 of risk, the 2026 average drops to −0.02R.
The backtest wasn't an edge either. The +0.04R from 2023 to 2025 had a p-value of 0.057, a near miss. That's what a sealed test is for.
We tested the rules, not a trader. Picking stocks with news or skipping setups that look wrong wasn't tested, nor were small caps or pre-market trading.
APPLY IT THIS WEEK
Turn your setup into rules a computer could follow. Write down exactly what makes a bar "big" or a pause "tight". If you can't, you can't test it.
Ask what random entries would do. Same stop, target and time of day, on random days. If they often match your setup, your p-value is high, and the setup isn't what's making the money.
Know your break-even win rate, after costs. Judge your win rate against that line, not against 50%.
TERM OF THE WEEK
Break-even win rate
The win rate at which a strategy makes exactly nothing. If the average win is W and the average loss is L, it's L ÷ (W + L). Costs push it up. For the 3 Bar Play in 2026, after costs, it was about 35.7%.
NEXT WEEK
The 3 Bar Play, Part 2
Our no-edge 3 Bar Play was up nearly 30R by May 2026, then gave it all back. Next week: why short runs fool traders, and why one stock is never enough.
Your turn
Hit reply: have you traded the 3 Bar Play, or a setup like it? What did you see?
LuckOrEdge is educational content only, not investment advice. This tests the mechanical rules above, not any person's full trading method. Interactive Brokers 1-minute data, $0.99 per order, 1¢ per share of slippage, $100 of risk per trade. Past performance does not guarantee future results.