Every trading book has a chapter on candlestick patterns. A Hammer after a selloff means the bulls are stepping in. A Bearish Engulfing at the top of a run means the party is over. The pictures are convincing, and the stories behind them make sense.
But do they actually predict anything? Our daily candlestick scanner has been detecting these patterns every evening, so we stored three years of prices and every pattern the scanner found, then checked what each stock did next.
The short answer: on their own, they don’t. After almost every pattern, including the famous ones, the stock went up about as often as it does on any random day. A few of the classic bullish reversals did noticeably worse than a random day.
What we tested
- Prices: 416,371 daily bars for 559 US stocks and ETFs, from October 24, 2023 to October 23, 2026.
- Patterns: every pattern TA-Lib detects, the same open-source library the scanner uses. That is 61 pattern detectors, which produced 87 distinct bullish and bearish labels and 808,622 pattern occurrences in total.
- The question: was the stock’s close higher 1, 5 and 20 trading sessions after the pattern than on the day it printed?
- The yardstick: the same question asked of every session, pattern or not. A pattern is only useful if it beats that baseline.
The baseline matters more than anything else in this post. Stocks rose over this period, so “the stock was higher five days later” was true on most days no matter what the candles looked like:
| Any stock, any day | Higher 1 session later | Higher 5 sessions later | Higher 20 sessions later |
|---|---|---|---|
| All 415,000+ sessions | 51.9% | 53.8% | 55.6% |
So if a pattern is followed by a gain 54% of the time, that sounds like an edge, but it is just the market. A bullish pattern needs to beat 53.8% at five sessions to have added anything. A bearish pattern works if the number comes in below 53.8%.
Candlestick patterns are everywhere
The first surprise: patterns are not rare. 78% of all sessions printed at least one named pattern, and many printed several. The most common labels, such as Doji, Spinning Top and Long Line Candle, each showed up tens of thousands of times.
A signal that fires on four out of every five days can’t carry much information, and the numbers bear that out.
Bullish vs. bearish: almost no difference
Grouping every bullish pattern together and every bearish pattern together:
| Occurrences | Higher 1 session later | Higher 5 sessions later | Higher 20 sessions later | |
|---|---|---|---|---|
| All bullish patterns | 497,890 | 51.5% | 53.5% | 55.2% |
| All bearish patterns | 305,011 | 52.7% | 53.9% | 55.1% |
| Days with no pattern | 89,087 | 51.9% | 54.5% | 57.1% |
| Baseline (every day) | 413,576 | 51.9% | 53.8% | 55.6% |
Read that again: stocks rose slightly more often after a bearish pattern than after a bullish one, both the next day and five days later. And the quiet days with no pattern at all did the best of the three over 20 sessions.
The famous patterns, one by one
Here are the patterns traders talk about most, with how often the stock was higher five sessions later. Remember the baseline is 53.8%. For bullish patterns higher is better; for bearish patterns lower is better.
| Pattern | Occurrences | Higher 5 sessions later | vs. any day |
|---|---|---|---|
| Bullish Three Outside Up | 7,027 | 55.6% | +1.8 |
| Bullish Dragonfly Doji | 5,586 | 55.0% | +1.2 |
| Bullish Hammer | 8,594 | 54.1% | +0.3 |
| Bullish Engulfing | 15,667 | 53.5% | −0.3 |
| Bullish Harami | 17,603 | 53.4% | −0.4 |
| Bullish Morning Star | 1,826 | 49.8% | −4.0 |
| Bullish Piercing Line | 1,261 | 49.2% | −4.6 |
| Bearish Engulfing | 17,933 | 54.1% | +0.3 |
| Bearish Shooting Star | 5,507 | 53.9% | +0.1 |
| Bearish Hanging Man | 6,949 | 53.4% | −0.4 |
| Bearish Evening Star | 2,181 | 53.4% | −0.4 |
| Bearish Dark Cloud Cover | 1,671 | 52.7% | −1.1 |
A few things stand out:
- The Hammer, Engulfing and Harami are coin flips plus the market. With 8,000 to 18,000 occurrences each, these numbers are precise to within about one percentage point, and they sit right on the baseline.
- The Morning Star and Piercing Line, two textbook bullish reversals, did worse than a random day. Stocks were higher five sessions later less than half the time. With 1,200 to 1,800 occurrences, a four-point gap is larger than chance alone would normally produce.
- No bearish pattern meaningfully predicted a decline. The best of them, Dark Cloud Cover, shaved about one point off the baseline.
- The best bullish results are small. The Dragonfly Doji and Three Outside Up beat the baseline by one to two points. When you test 87 patterns, a couple will look this good by luck, so treat these as “maybe” rather than “found an edge”.
The next-day numbers tell the same story. A Bearish Closing Marubozu (a big down day that closes at its low) was followed by an up day 55.2% of the time, against 51.9% for any day, across nearly 26,000 occurrences. Sharp down days tended to bounce, which is the opposite of what the pattern is supposed to mean.
“But you have to check the trend first”
This is the standard objection, and it’s a fair one. A Hammer is only supposed to mean something after a decline, and a Shooting Star only after a rally. TA-Lib mostly checks the shape of the candles, not the trend before them.
So we split the results by what the stock did over the 10 sessions before the pattern. First, bullish patterns that appeared after a drop of more than 5%:
| After a 10-day drop of more than 5% | Occurrences | Higher 5 sessions later |
|---|---|---|
| Any day after such a drop (baseline) | 62,899 | 54.0% |
| Bullish Hammer | 1,880 | 54.1% |
| Bullish Harami | 4,270 | 53.1% |
| Bullish Engulfing | 2,807 | 49.0% |
| Bullish Morning Star | 382 | 48.7% |
| Bullish Piercing Line | 303 | 46.2% |
Then bearish patterns that appeared after a rise of more than 5%:
| After a 10-day rise of more than 5% | Occurrences | Higher 5 sessions later |
|---|---|---|
| Any day after such a rise (baseline) | 85,961 | 53.7% |
| Bearish Engulfing | 3,929 | 54.1% |
| Bearish Shooting Star | 1,376 | 53.5% |
| Bearish Hanging Man | 1,627 | 52.9% |
| Bearish Evening Star | 756 | 56.2% |
| Bearish Dark Cloud Cover | 507 | 47.7% |
Adding the trend didn’t rescue them. Bullish reversals after a selloff did no better than any day after a selloff, and the Engulfing, Morning Star and Piercing Line did worse. Among the bearish patterns, only Dark Cloud Cover leaned the right way, on just 507 cases, where the margin of error is about four points.
There is one wrinkle worth mentioning. A Bullish Engulfing after a drop was weak five days out but better than baseline twenty days out (59.5% vs. 56.3%). Short-term weakness followed by a longer-term recovery is interesting, but it is one result among dozens, so we wouldn’t trade on it without more data.
Pattern by pattern
Each of the most-searched patterns has its own page with the full breakdown, including whether waiting for a confirming next day helped:
- Hammer and inverted hammer
- Bullish and bearish engulfing
- Doji and dragonfly doji
- Shooting star and hanging man
- Morning star and evening star
- Harami
- Piercing line
Why the per-stock numbers can fool you
Every ticker page on the scanner shows a track record panel: how that stock has done after each pattern. Small samples there can look exciting. AMD, for example, was higher five sessions after 18 of its 30 Bullish Harami patterns, which is 60%.
With only 30 cases, though, the margin of error is about ±18 points. A 60% result is entirely consistent with a stock that simply goes up 54% of the time. That is why the panels show the number of cases next to every percentage and compare it with the stock’s own baseline. Look for a large number of cases and a clear gap before reading anything into it.
Caveats
- It was mostly a rising market. The baseline of 53.8% reflects a strong three years. In a bear market every number would shift, which is why we compare each pattern with the baseline from the same period rather than with 50%.
- Survivorship. The symbol list is the scanner’s current list, so companies that were delisted during the period are not included.
- Close-to-close only. We measured closing prices with no stops, targets, commissions or options. A pattern could still be useful as a place to put a tight stop, which this test doesn’t measure.
- TA-Lib’s definitions. Other software defines some patterns differently. TA-Lib is the most widely used implementation, which is why we chose it.
So are candlestick patterns useless?
Not quite. As predictions, the data says they add little or nothing: the market’s own drift explains almost all of the “success” people attribute to them. As vocabulary, they’re still handy. “A Hammer at support” is a fast way to describe what buyers and sellers did that day, and the scanner will keep labeling them for that reason.
What we would avoid is buying or selling because a pattern printed. If you trade options, the pattern says far less about your trade than implied volatility, the strike you pick and how much you’re risking.
Quick answers
Do candlestick patterns predict stock moves? In this test of 808,622 patterns, the stock rose about as often after a pattern as on any random day: 53.5% after bullish patterns and 53.9% after bearish ones, against 53.8% for every day.
What is the most reliable candlestick pattern? None stood out by a meaningful margin. The best bullish results were Three Outside Up (55.6%) and the Dragonfly Doji (55.0%), only one to two points above the 53.8% baseline.
Does the Hammer pattern work? Stocks were higher five sessions after a Hammer 54.1% of the time across 8,594 cases, essentially the same as any day. After a drop of more than 5%, it was 54.1% against a 54.0% baseline.
Does checking the trend first make patterns more accurate? Not in this data. Bullish reversals after a selloff did no better than any day after a selloff, and several did worse.
Data: daily prices from Yahoo Finance for the scanner’s 559 symbols, October 24, 2023 to October 23, 2026, with patterns detected by TA-Lib. This post is for education only and is not financial advice.