If you've spent any time around trading charts, you've met the candle patterns: the hammer, the doji, the dreaded head-and-shoulders. The promise is always the same — learn to read these shapes and the chart will tell you what happens next.
That's a testable claim. So we tested it, the same way we test everything before it gets near the bot: against every 4-hour candle Bitcoin has printed since 2015.
What we did
We wrote a precise, math-only definition for 20 patterns — engulfing candles, hammers, dojis, shooting stars, bull and bear flags, failed breakouts, and more — locked the definitions before running anything, then found every occurrence across 23,600 four-hour candles (and the daily chart too). For each one we measured what price did 1 to 30 days later, against what Bitcoin does anyway over the same horizon. A bullish pattern only counts if it beats the market's normal upward drift.
Method, briefly
Textbook definitions, pre-registered, no tuning after seeing results. Overlapping signals de-duplicated. Testing many patterns guarantees lucky-looking ones, so significance uses false-discovery control (q = 0.10); survivors then had
to agree with an untouched 18-month holdout. Edges shown net of a
0.28% round trip.
What we found
| Pattern | Events | Edge vs. drift | Verdict |
|---|---|---|---|
| Bullish engulfing | 1,005 | −0.3pp | NOISE |
| Hammer (after a dip) | 185 | +0.4pp | NOISE |
| Shooting star | 99 | +0.9pp | DIED ON HOLDOUT |
| Bull flag | 62 | +2.8pp | TOO RARE TO PROVE |
| Failed breakdown (“the spring”) | 298 | −2.3pp | REAL — BUT BACKWARDS |
| Volatility squeeze in a downtrend | 294 | +6.4pp | REAL |
| Two inside bars after a down-move | 207 | +5.0pp | REAL |
| …13 more | — | — | NOISE |
Note what's missing from the winners' column: every famous pattern. We measured over a thousand bullish engulfing candles — the most celebrated shape in the book — and they predicted nothing. The classic "spring" (a failed breakdown that's supposed to be bullish) turned out to be real but backwards: price did worse than normal after it. The only genuinely verified signals were the least dramatic things on the chart: volatility quietly drying up while price sits below its recent average (after which Bitcoin's usual upward drift simply fails to show up for weeks), and two quiet "inside" bars after a down-move — each candle nesting inside the previous one, typically a sign of more weakness ahead. Quiet beat dramatic, twice.
The honest conclusion
We'd have happily added any of these to the bot. Even the one real pattern got a full audition — and made the strategy worse, because the bot's trend rules already avoid the dead zone it flags. So the bot stays boring, the drama candles stay decorative, and the negative result gets published. That's the whole point.