Failed Breakout Fade
a pre-registered null — the popular fade that isn’t“Most breakouts fail” is one of the most repeated lines in Brooks price action, and it spawns a popular trade: when a stock pokes to a new multi-week high but closes back inside the range, the breakout buyers are trapped — so fade it short, and ride the reversal as they cover. This study asks the blunt question: does that fade actually pay on the next session? We pre-registered the hypothesis, then ran it over 1,433 failed breakouts and 2,901 held breakouts on 15 liquid US large-caps and ETFs (Jan 2022 – May 2026). The answer is a clean no. We publish it because a null is a result: it tells you where the edge isn’t.
The fade resolves in the reversal direction exactly as often as a coin. No edge.
Expectancy of the fade in ATR units is indistinguishable from zero — slightly negative, if anything.
At the 20-day daily scale, breakouts holdabout 2-to-1. “Most fail” is a lower-timeframe truth, not a daily one.
Does fading the failure beat a coin flip?
The failed-breakout reversal rate’s 95% confidence interval contains 50% — the fade is statistically a coin flip. 15 tickers · 2022–2026 · n=5,000 bootstrap.
The “trapped buyer” signature that wasn’t
If trapped breakout buyers really drive a reversal, then on an otherwise identical new-extreme day, the session that FAILS (closes back inside) should be meaningfully more bearish the next day than the one that HOLDS. Measured in ATR-normalized next-day return, the gap is real in sign but tiny — and its confidence interval crosses zero in both directions. The split carries no significant information.
| New 20-day extreme | Failed (closed inside) | Held (closed through) | Held − Failed | CI95 of diff |
|---|---|---|---|---|
| New high → next-day return | +0.042 ATR | +0.054 ATR | +0.012 ATR | [-0.050, +0.076] |
| New low → next-day return | +0.034 ATR | +0.064 ATR | +0.031 ATR | [-0.051, +0.108] |
Both confidence intervals straddle zero — you cannot reject “fail and hold behave the same next day.” The trapped-buyer story may be real intrabar, but it does not survive to the next daily close.
Per-ticker detail
Sorted by reversal hit rate. The values scatter from 41.1% (GOOG) to 58.6% (AMD) on both sides of 50% with no consistency — exactly what you would expect from noise. Only AMD has a hit rate whose interval barely clears 50%, and a single name out of 15 is what chance produces.
| Ticker | N failed | N held | Reversal hit % | Fade return (ATR) |
|---|---|---|---|---|
| AMD | 99 | 200 | 58.6% | +0.206 ATR |
| AVGO | 62 | 177 | 54.8% | -0.056 ATR |
| TSLA | 81 | 180 | 53.1% | -0.008 ATR |
| MSFT | 95 | 193 | 52.6% | +0.080 ATR |
| IWM | 98 | 181 | 52.0% | -0.029 ATR |
| XLK | 107 | 204 | 51.4% | +0.028 ATR |
| AMZN | 90 | 174 | 51.1% | +0.043 ATR |
| AAPL | 98 | 185 | 51.0% | -0.090 ATR |
| META | 87 | 162 | 50.6% | +0.068 ATR |
| SPY | 134 | 229 | 50.0% | +0.013 ATR |
| QQQ | 106 | 231 | 50.0% | -0.081 ATR |
| XLF | 97 | 208 | 47.4% | -0.066 ATR |
| NVDA | 87 | 187 | 46.0% | -0.135 ATR |
| JPM | 102 | 209 | 42.2% | -0.087 ATR |
| GOOG | 90 | 181 | 41.1% | -0.138 ATR |
What this means for trading
- The failure is not the signal — the confirmation is.Brooks never says “short the instant a breakout closes back inside.” He says wait for the reversal to confirm: a strong reversal bar, a second entry, a lower high. This study tests the blind, mechanical version of the fade and finds nothing, which is fully consistent with the Brooks framing. The edge, if there is one, lives in the confirmation — and a daily close-to-close test can’t see it.
- Don’t pre-position against a new extreme.Selling short simply because price poked above a 20-day high and closed back inside has zero next-day expectancy here. If you trade failed breakouts, your edge has to come from the entry trigger and the stop, not from the pattern’s mere existence.
- “Most breakouts fail” is timeframe-dependent. At this 20-day daily horizon, breakouts actually HOLD about 2-to-1 (2,901 vs 1,433). The Brooks maxim is an intrabar / intraday observation about how many attempts fail, not a daily close-through statistic. Mapping a lower-timeframe truth onto a higher-timeframe trade is where the myth gets made.
- A null protects your capital.The most expensive trades are the ones that feel obvious and aren’t. “Trapped buyers must cover” is a compelling story, and it is exactly the kind of intuition that needs a number on it before you risk money. The number here says: coin flip.
Pre-registered verdicts
Three pre-registered verdicts, three failures — and that is the value. A clean null, locked before the data was read, is far more trustworthy than an edge discovered by tuning the rule after the fact. The fade-the-failure trade does not work at the daily close-to-close horizon. Where it might work — on the confirmation trigger — is the companion Prior-Day Extremesgallery’s territory.
Honest caveats
- This tests a blind, mechanical, close-to-closefade. It does NOT test a discretionary entry on a reversal signal bar, a second-entry trigger, or an intraday stop. A null on the mechanical version says nothing about a skilled trader’s confirmed entry — it only rules out the naive version.
- “Failed” here is defined at the daily close (new 20-day extreme intraday, close back inside the prior bar’s range). A stricter rejection filter — a long tail, a close in the bottom third of the day’s range — would isolate fewer, cleaner events, but changing the rule after seeing this result would be exactly the post-hoc tuning the pre-registration exists to prevent. That belongs in a new, separately pre-registered study.
- Horizon is one session. The reversal could play out over 2–5 days even if day +1 is a coin flip. A multi-day holding-period sweep is a natural, honest extension.
- Universe is the same 15 mega/large-cap US names as the prior studies — all highly liquid, all in a broadly bullish 2022–2026 window (note the 52.9% unconditional next-day up-rate). Failed breakouts in choppier or bearish regimes, or in small-caps, may behave differently.
Methodology
1,433 failed-breakout events + 2,901 held-breakout events across 16,170 sessions on 15 names (2022-01-03 – 2026-05-30) via Massive/Polygon daily bars. New extreme: today’s high > max of the prior 20 highs (mirror for lows), the 20-bar window excluding today. Failed: close falls back below the prior bar’s high (above the prior low). Held: close clears it. Outcome: next-day close-to-close return normalized by Wilder ATR(14) computed strictly through today. Reversal return signs the next-day move in the fade’s favor. Bootstrap CI (n=5,000, seed=17). Engine: scripts/ml/backtest_failed_breakout.py. Run 2026-06-21.
Related: Prior-Day Extremes · Outside Day Reversion · Inside Day Expansion