We re-measured all 165 chart patterns across 8,346 symbols
We followed the 30 trading days after each signal's closing price. Split by market and size tier and placed beside a control group, the same pattern gave different answers in different markets.
Textbooks tell you a chart pattern means price should rise. What they rarely show is how often that actually happened. So we counted.
We re-scanned the full bar history of 8,346 symbols across Korea, the US and crypto — symbol by symbol, bar by bar — counting only the signals that were genuinely visible on that day, then followed the 30 trading days from the close of the signal candle.
How to read these numbers
A single pattern's win rate on its own says nothing: in a period when the whole market rose, every pattern looks good. So every table here places a control group on the same row — what you would have got entering on any day in the same market over the same period.
Our primary yardstick is the percentile: where this signal ranked among all bars of that symbol in that year, scaled 0 to 1, where 0.5 is any day. Win rates and returns rise whenever that symbol had a good year; the percentile divides that regime out and leaves only the power to pick the date.
Is it above 0.5? That is the first question, and most patterns are decided right there.
What stands out in the Korean market
| Pattern | Percentile | Rose by then | Reward vs risk | Sample |
|---|---|---|---|---|
| No-supply decline1 | 0.714 | 68%+23.6%p | 2.12 | 803 |
| Pipe Bottom2 | 0.643 | 71%+26.9%p | 3.65 | 1,869 |
| Descending Scallop3 | 0.586 | 50%+6.1%p | 1.16 | 23,624 |
| Bullish Harami4 | 0.584 | 53%+8.8%p | 1.33 | 25,046 |
| Piercing Line5 | 0.581 | 54%+9.6%p | 1.38 | 4,088 |
| baseline | 0.500 | 44% | 1.11 | 721,482 |
Only patterns whose 95% interval sits above 0.5 with 50+ symbols, ranked by percentile. Past record settled through 2026-07-17; not a promise about what comes next.
That chart matters more than any single number: a lucky result does not climb this monotonically.
Change the market and the answer changes
| Pattern | Percentile | Rose by then | Reward vs risk | Sample |
|---|---|---|---|---|
| No-supply decline1 | 0.623 | 52%+1.1%p | 1.29 | 1,566 |
| Measured Move Down2 | 0.596 | 52%+1.7%p | 1.23 | 4,453 |
| Three Falling Peaks3 | 0.587 | 51%+0.3%p | 1.17 | 23,325 |
| Descending Scallop4 | 0.577 | 50%−0.4%p | 1.13 | 38,841 |
| Bearish MA Alignment5 | 0.567 | 52%+0.9%p | 1.15 | 133,094 |
| baseline | 0.500 | 51% | 1.07 | 1,541,481 |
Only patterns whose 95% interval sits above 0.5 with 50+ symbols, ranked by percentile. Past record settled through 2026-07-17; not a promise about what comes next.
This is the most practical finding of the study. Pipe Bottom scores 0.643 in Korea but 0.524 in the US — and among US small caps it drops to 0.486, *below* the null. Same shape, different market, different meaning.
Size tiers split it too
| Pattern | Percentile | Rose by then | Reward vs risk | Sample |
|---|---|---|---|---|
| Measured Move Down1 | 0.634 | 62%+6.8%p | 1.45 | 1,223 |
| Three Falling Peaks2 | 0.606 | 58%+2.6%p | 1.31 | 6,199 |
| Descending Scallop3 | 0.589 | 56%+0.5%p | 1.32 | 12,524 |
| Bearish MA alignment4 | 0.585 | 57%+2.3%p | 1.23 | 5,182 |
| baseline | 0.500 | 55% | 1.20 | 478,889 |
Only patterns whose 95% interval sits above 0.5 with 50+ symbols, ranked by percentile. Past record settled through 2026-07-17; not a promise about what comes next.
| Pattern | Percentile | Rose by then | Reward vs risk | Sample |
|---|---|---|---|---|
| No-supply decline1 | 0.617 | 49%+2.5%p | 1.31 | 804 |
| Three Falling Peaks2 | 0.577 | 46%−0.2%p | 1.10 | 9,491 |
| Measured Move Down3 | 0.575 | 45%−0.6%p | 1.12 | 1,688 |
| Descending Scallop4 | 0.568 | 44%−1.5%p | 1.04 | 12,533 |
| baseline | 0.500 | 46% | 0.92 | 551,020 |
Only patterns whose 95% interval sits above 0.5 with 50+ symbols, ranked by percentile. Past record settled through 2026-07-17; not a promise about what comes next.
Crypto is a different story
| Pattern | Percentile | Rose by then | Reward vs risk | Sample |
|---|---|---|---|---|
| Measured Move Down1 | 0.560 | 42%+7.4%p | 1.23 | 312 |
| Bear Pennant2 | 0.557 | 41%+6.4%p | 1.30 | 1,058 |
| Bullish Reclaim3 | 0.552 | 42%+7.7%p | 1.52 | 350 |
| Three Falling Peaks4 | 0.544 | 36%+1.9%p | 0.98 | 1,411 |
| baseline | 0.500 | 35% | 0.95 | 75,149 |
Only patterns whose 95% interval sits above 0.5 with 50+ symbols, ranked by percentile. Past record settled through 2026-07-17; not a promise about what comes next.
Read these carefully: the control group's median is around −8%. Over this sample window, entering on any day in crypto left you well down after 30 days. So the top crypto patterns are not "places that rose" but "places that fell less" — which is why the leaders still show negative medians.
Frequent is not the same as meaningful
The most frequently firing signal is the doji, with over 460,000 observations — and a percentile of 0.506, essentially indistinguishable from any day. Frequent signals usually say nothing.
Limits — please read these with the numbers
- Only firings whose 30 trading days have fully settled are included, so
roughly the last six weeks of signals are not here yet.
- No stops, fees or slippage.
- The 95% intervals come from a symbol-cluster bootstrap — firings from the
same symbol are not independent, and counting them as if they were makes the interval falsely narrow.
- Size tiers use the median trading value of that year, not a market-cap
snapshot. That keeps 2015 events from being classified by today's size and avoids the survivorship bias of delisted companies vanishing from the list.
The table does not stand still
The figures in this article are not baked into the text; they are read from the table at request time. The batch rebuilds that table weekly, so if you reopen this article next week the numbers may have shifted. Rankings *should* change when the market changes, and the writing should follow.