Research note
Minervini's Holy Grail, Checked
Mark Minervini's 'holy grail' is a simple recipe: average winners of 15%, losers held to 5%, win about half the time, and turn the edge over as often as possible. The compounding math checks out. But his headline ratio overstates what you actually keep — and when we tested the closest codable version of his system, it produced 68 trades and not a single winner.
Mark Minervini has a simple recipe for what he calls the holy grail of trading: win about half your trades, make about 15% on the winners, lose about 5% on the losers, and turn that edge over as often as you can. He laid it out in an eight-minute video — one hand-drawn sketch, no charts, no tickers, no backtest — and the clip has been circulating again since he shared it this week.
We checked it. Three findings: the compounding math is right; the headline ratio overstates what you actually keep; and when we tested the closest version of his system we could write down as rules, it produced 68 trades and not a single winner.
What the video actually says
The video is an eight-minute explainer, about a year old. The whole thing is one sketch: a buy point on a line, upside running to infinity, downside running to zero, and a wavy line showing the price wandering above and below the buy point after entry. There is no price chart, no named stock, no entry rule, and no performance record in it.
His numbers, as he states them: the average winner makes 15%, the average loser is held to 5%, and across his career about 45% to 50% of his trades have been profitable. He prefers a three-to-one reward-to-risk ratio and will accept two-to-one. The argument is that trading is not about predicting peaks or buying lows — that is, in his telling, close to impossible and unnecessary. It is about finding the gain you can capture repeatedly, keeping the loss small enough not to choke off the winners, and compounding the result by doing it again and again.
The compounding math holds
The middle of the video is a compounding comparison: one trade that doubles your money, against several smaller trades compounded one after another. The arithmetic checks out.
| Path | Result |
|---|---|
| 1 trade of +100% | 100% |
| 4 trades of +20%, compounded | 107.4% |
| 8 trades of +10%, compounded | 114.4% |
Four 20% gains compound to 107.4% (1.20 to the fourth power, minus one). Eight 10% gains compound to 114.4% (1.10 to the eighth power, minus one). The intuition behind it is sound: a 10% move is far easier to find than a 100% move, and you get many more chances at it in a year. Smaller, repeatable gains really can beat hunting for one huge one.
One small error on the slide itself: its first row prints "1 x 75%" next to "100%", while the narration says one trade of 100%. A single 75% trade returns 75%, not 100% — the printed row is wrong and the spoken words are right. The other two rows, 107% and 114%, are correct as printed.
The ratio is gross, not net
This is the part that needs correcting. Minervini adjusts his reward-to-risk ratio for his win rate with a simple formula: the percentage gain times the percentage of winning trades, divided by the percentage loss times the percentage of losing trades. At 40% wins, that gives (15 × 40) / (5 × 60) = 2 to 1, and he presents it as making $2 for every $1 risked.
That 2 to 1 is gross dollars won divided by gross dollars lost. It is not what you keep. What you keep is net expectancy per trade: the win rate times the average win, minus the loss rate times the average loss. The risk on each trade is 5%, so net profit per $1 risked is the expectancy divided by 5.
| Win rate | Gross ratio | Net expectancy per trade | Net per $1 risked |
|---|---|---|---|
| 50% | 3.00 to 1 | +5.00% | $1.00 |
| 45% | 2.45 to 1 | +4.00% | $0.80 |
| 40% | 2.00 to 1 | +3.00% | $0.60 |
At 40% wins you keep $0.60 per $1 risked, not $2. At 50% wins you keep $1.00, not $3. The headline ratio overstates what you keep by about three times — 3.0× at a 50% win rate, 3.3× at 40%.
To be fair, the edge is still real at his own numbers: net expectancy is positive at every win rate in his stated 45–50% range. A trader who genuinely averages +15% winners, −5% losers, and wins half the time has a working edge. The correction is about size, not existence — the ratio as presented is a gross figure wearing net clothing.
The premise, tested
The video's premise needs a system that actually delivers average winners of +15% with losses contained at −5%. So we tested the closest version of his system that can be written down as rules. The test was pre-registered — the spec was frozen before any run — and it is the same test we ran when his setup first circulated: his screen plus his stop, with a plain breakout entry standing in for the pattern entry, which cannot be coded (more on that below).
The rules: S&P 500 stocks, daily bars, January 2016 through November 2026. The screen requires a stock within 25% of its 52-week high, priced at least $30, with a rising 200-day average, the 50-day above the 200-day, at least double its 52-week low, and a new 252-day closing high in the last six months. Entry is a close above the trailing 20-day high while passing the screen. The exit is his stop only: twice the 20-day average true range, capped at 10%, fixed, with no profit target and no time exit. One percent of equity risked per trade, positions capped at 10% of equity, 2 basis points of friction per side, starting from $1 million. The bar to pass was a net Sharpe ratio of at least 0.80 with at least 30 closed trades.
| Metric | Result | Bar |
|---|---|---|
| Closed trades | 68 | 30 |
| Win rate | 0.00% | — |
| Expectancy | −1.0073R | — |
| Total return | +185.71% | — |
| Net Sharpe | 0.4966 | 0.80 |
| Maximum drawdown | −42.27% | — |
Verdict: FAIL. The trade count clears the bar; the Sharpe does not.
The 0% win rate is not a data error. It is a mathematical consequence of the spec: the stop below the entry price is the only exit, so every closed trade exits below its entry and loses about its 1% risk by construction. All 68 trades exited at the stop price, with risk multiples between −1.0153 and −1.0038 — the small excess over −1.00 is friction. There is no way for a closed trade to win.
And the +185.71% total return is entirely unrealized. The 68 closed trades lost $432,396 between them. The gain sits in four positions still open at the end of the data, carrying $2,289,449 of unrealized profit — most of it one name, AMD, entered in October 2019 and never once touching its stop. The system's return describes a four-name buy-and-hold tail, not a repeatable trade edge.
The comparisons make the point sharper. The same entries with no stop at all returned +410.39% at a Sharpe of 0.73 — the stop cost about 225 percentage points. An equal-weight buy-and-hold of the same universe returned +493.14%, and the S&P 500 itself returned +334.95%. The out-of-sample half, 2022 through 2026, repeats the pattern: 31 trades, 0% wins, Sharpe 0.4952. Two honesty notes: the universe is the current S&P 500 list, so survivorship bias flatters every number including the benchmarks; and a cash rule dominated the trade count — 8,444 signals fired but only 72 entries were taken, because the 10% position cap means about ten slots, and a few survivors held theirs for years.
One more result matters, because it locates the failure precisely. The same screen with an exit that can win does clear the bar: a trailing stop at three times the average true range produced 799 trades, a 35.54% win rate, and a Sharpe of 0.8843; a 60-day time exit produced 744 trades, a 33.60% win rate, and a Sharpe of 0.8310. Both were added to our Terminal as voteless presets. It is the exit rule — not the screen — that fails in the stop-only version.
What this does and doesn't prove
The most important caveat is what we did not test. Minervini's actual entry is the Volatility Contraction Pattern: a series of shrinking pullbacks on drying volume, bought on the breakout. That pattern is visual and discretionary, and when we operationalized it, it fired 4 times in 16 years of S&P 500 data and 17 times on the S&P 400 and 600 combined — too few to test. It is closed as untestable, not proven worthless.
So this post tests the codified screen-and-stop version, not his discretionary trading. His discretionary entry could differ from the breakout entry we substituted, and we cannot test what we cannot write down as rules. Nothing here claims he is wrong as a trader.
What the evidence does show is narrower, and it is enough: the compounding arithmetic is right, the headline ratio overstates net expectancy by about three times, and the codified system does not deliver the +15% winners and −5% losers the premise needs. The holy grail, as stated, is a true statement about arithmetic attached to a winner profile that the written-down system never produced.
FAQ
Does Minervini's compounding math check out?
Yes. Four +20% trades compounded return 107.4%, and eight +10% trades compounded return 114.4%, both beating a single +100% trade. Smaller, repeatable gains really can beat hunting for one huge move.
Is a 3:1 reward-to-risk ratio as good as it sounds?
Only as a gross figure. At a 50% win rate with +15% winners and −5% losers, gross wins are three times gross losses — but net expectancy is +5% per trade, or $1.00 kept per $1 risked, not $3. At 40% wins it is $0.60 per $1 risked, not $2.
Did you test Minervini's actual strategy?
No — we tested the closest version that can be written down as rules: his screen, a breakout entry, and his stop. His real entry, the Volatility Contraction Pattern, is discretionary and fired too rarely to test when operationalized. The stop-only version failed with 68 trades and a 0% win rate; the same screen with a trailing stop or a time exit passed.
Why did every trade lose?
Because the stop was the only exit. A fixed stop below the entry price means every closed trade exits below where it entered, losing about its 1% risk. The system's entire gain was unrealized profit in four positions that never touched their stops.
Sources
- Mark Minervini, "Mark Minervini Explains the Holy Grail in Trading" (YouTube, about a year old) — youtube.com
- The post sharing it: x.com
- The backtest is our own pre-registered test: spec frozen November 9, 2026, run November 9–10, 2026, on S&P 500 daily data from January 2016 through November 2026. The follow-up exit variants were tested under the same frozen spec and independently reviewed.
Educational only, not financial advice. This post checks arithmetic and tests a written-down version of a published system. It is not a judgement on any trader's discretionary results, and past performance — ours or anyone else's — does not guarantee future results.