Do chart patterns actually work? We measured 46,009 pattern windows
Most articles quote a success rate for flags or head and shoulders without saying how it was measured. We have the data to check: every time a textbook pattern appeared in 2,312 stocks since 2015, we looked at what the price did next.
The short answer
Patterns tilt the odds a little, not a lot. A random chart in our data ended 10 trading days later higher 53% of the time (average +0.58%). The pattern with the biggest edge over a random chart was gap down: it moved the expected way 52% of the time, against 47% for a random chart. The weakest was bear flag at 42% (random: 47%). No pattern beat a random chart by more than about 5 percentage points.
The results, pattern by pattern
"Moved as expected" means higher for bullish patterns and lower for bearish ones. A random chart moved up 53% of the time, so bearish patterns face a headwind: about 47% of random charts fell.
| Pattern | Examples | Moved as expected | Average change | Median change |
|---|---|---|---|---|
| Volatility contraction | 14,736 | 53% up | +0.64% | +0.49% |
| Parabolic run-up | 7,747 | 51% | +0.68% | -0.20% |
| Gap up | 4,887 | 50% | +1.20% | +0.00% |
| Gap down | 3,790 | 52% | +0.16% | -0.43% |
| Capitulation (sharp drop) | 2,874 | 56% | +3.53% | +2.47% |
| Accumulation (volume favours buyers) | 2,346 | 50% | +0.84% | +0.00% |
| Head and shoulders | 1,778 | 51% | -0.18% | -0.07% |
| Inverse head and shoulders | 1,772 | 56% | +0.65% | +0.93% |
| Double top | 1,681 | 50% | +0.40% | +0.07% |
| Double bottom | 1,531 | 54% | +0.70% | +0.79% |
| Accumulation (volume favours sellers) | 1,222 | 46% | +0.11% | +0.20% |
| Trading range (tunnel) | 1,123 | 52% up | +0.27% | +0.22% |
| Bull flag | 400 | 51% | +0.29% | +0.26% |
| Bear flag | 122 | 42% | +1.37% | +1.75% |
| Random chart (baseline) | 2,499,144 | 53% up | +0.58% | n/a |
Measured 10 trading days after the pattern, close to close, on 2,312 stocks since 2015. Patterns with fewer than 100 examples are left out. Past results, not a prediction.
What we learned
- No pattern is a coin you can bank on. The strongest edges are a few points, and every pattern still fails often. That is why one trade tells you almost nothing.
- Averages hide the shape. A pattern can have a high average because of a few big winners while the median is close to zero. Look at both columns.
- Context matters. The rise of the whole market, the stock's trend and the news behind it change the odds. A pattern alone is one input, not a verdict.
- Process beats outcome. A good decision can lose and a bad one can win. Judge your calls against the odds, over many tries.
How we measured it
We did not use hand-picked examples. For every stock and every day since 2015, a rule-based check looked at the last 60 daily candles and asked whether a pattern was really there (for example, a flag needs a sharp pole of at least 12% followed by a tight, shallow pause). Windows that passed were counted, and the change over the next 10 trading days was recorded. The rules are the same ones we use to choose the charts in the game; you can read them on each pattern page.
Limits: 10 days is one horizon (shorter or longer targets give different answers), results ignore trading costs, and stocks in the data are those with price history available today, which can flatter the numbers.
Test your own eye against these odds
In SwipeTrainer every answer shows the real odds of the pattern, so you learn to judge the quality of a call, not just the result.
Frequently asked questions
Do chart patterns work?
In our test they gave only a small edge. Across 2,312 stocks since 2015, most patterns ended 10 trading days later within a few percentage points of a random chart. Patterns are better seen as a way to frame the odds than as a signal.
What is the success rate of a bull flag?
Online figures vary and rarely define "success". In our data, bull flags that passed strict rules ended higher after 10 trading days 51% of the time, against 53% for a random chart.
How did you measure it?
For every window since 2015 that passed our strict textbook rules for a pattern, we recorded the close-to-close change over the next 10 trading days and compared it with the 2,499,144 ordinary windows in the same data.
Is this investment advice?
No. These are past results for education. They ignore costs, taxes and slippage, and past odds do not predict the future.