Chart Patterns
Trading Classic Chart Patterns
The only Bulkowski book that grades a pattern before you enter it. Roughly ten yes or no tests, each worth plus one or minus one, and a total above zero says this pattern should beat the median for its type. The tests are sensible. The point values behind them came from five years of data.
The one idea
The Encyclopedia of Chart Patterns tells you how an average double bottom behaves. It cannot tell you whether the double bottom on your screen right now is one of the good ones. That gap is the reason this book exists. Think of the Encyclopedia as the price guide for a model of used car and this book as the inspection checklist for the specific car in the driveway.
Bulkowski built it by slicing each pattern's sample one trait at a time. How long was the price run leading into the pattern. Is the pattern tall or short compared with others of its type. Where does the breakout price sit inside the last twelve months of range. Was volume rising or falling while the pattern formed. Was breakout day volume above average. Small cap or large cap. For each trait he compared the two sides against that pattern's own median result and stamped the better side plus one, the worse side minus one. You read your chart against ten or so tables, add the marks, and get a single number.
The claim is that the number sorts winners from losers. Broadening tops with a positive score rose 38% on average against 21% for negative scores. Ascending triangles, 50% against 21%. Rectangles, 55% against 22%. Across all thirteen bullish setups in the summary table, the positive group roughly doubles the negative group, and the pattern repeats on the bearish side. That single comparison is the whole argument of the book.
What it actually teaches
The scoring system is a fixed sequence you walk once per candidate trade, keeping a running total as you go.
FIND THE TREND START
Stand on the first day of the pattern and look backward. If prices fall away from the pattern going back in time, find the lowest low before a close 20% or more above it. If prices rise away, find the highest high before a close 20% or more below it. That point is the trend start, and several later tests hang off it. The 20% figure is Bulkowski's definition of a trend change everywhere in the book, borrowed from the conventional bull and bear market threshold.
TIME THE RUN-UP
Count calendar days from the trend start to the pattern's first day and drop it into a bucket: short term (0 to 3 months), intermediate (3 to 6 months), long (over 6 months). The score is not universal, you look it up per pattern. Broadening tops with upward breakouts pay best on a long run-up (41% average rise) and score short-term run-ups minus one. Descending triangles breaking out upward want the opposite, a short run-up, worth 49%.
LOOK FOR A SHELF IN THE PRICE PATH
A horizontal consolidation region is several weeks of sideways trading between the trend start and the pattern, sitting where price now has to travel. They average 37 to 43 days long, so they are hard to miss. On broadening tops with upward breakouts, having one in the way scores minus one (31% average rise against 35% with a clear path), and prices stall inside the region 38% of the time going up and 58% of the time going down.
PLACE THE BREAKOUT IN THE YEARLY RANGE
Take the highest high and lowest low of the twelve months before the breakout, cut that range into thirds, and see which third the breakout price falls into. Broadening tops breaking out in the top third rose 37% against 22% for the middle third. The rule underneath is momentum: do not buy a falling stock hoping it bottoms, and do not short a rising one expecting it to crack.
MEASURE HEIGHT AGAINST THE PATTERN'S MEDIAN
Highest high minus lowest low inside the pattern, divided by the breakout price. Compare that percentage with the median Bulkowski publishes for that specific pattern: 17.49% for broadening tops with upward breakouts, 18.29% for downward, 17.09% for double bottoms. Above the median is tall and scores plus one. Tall broadening tops rose 42% against 27% for short ones. Tall beats short in all thirteen bearish setups and eleven of thirteen bullish ones, which makes this the most reliable test in the book.
SCORE VOLUME TWICE
First the volume trend from the pattern's first day to its last, fitted with linear regression (the trend function in a spreadsheet). Second, breakout day volume against the stock's own recent average. Above-average breakout volume scores plus one nearly everywhere and is worth about five points of average return, 38% against 33% on broadening tops. The volume trend test is the weakest thing in the book, splitting 36% against 33%, and it still counts the same as everything else.
CALL THE RETRACE AND THE TRAP
Decide whether price will throw back to the breakout level (throwback rates run 29% to 65% depending on the pattern, pullbacks 43% to 61%) and whether the breakout is a trap. A bull trap is a rise of less than 10% followed by a drop below the pattern low, a bear trap the mirror image. Bull traps run 9% to 20% of cases by pattern, bear traps 16% to 43%. This is where the checklist stops being arithmetic and becomes your read of overhead resistance.
ADD MARKET AND SIZE, THEN TOTAL
Check whether the S&P 500 rose or fell between the pattern's first and last day, and bucket the stock by market capitalization: small up to $1 billion, mid $1 to $5 billion, large over $5 billion. Small caps score plus one in eleven of the thirteen bullish setups. Then add everything up. Above zero, the pattern is worth considering. Below zero, walk away. Exactly zero leaves the decision to you.
What it looks like Monday morning
You stop taking breakouts on sight. Before the next one you run four measurements that take about two minutes on a chart: pattern height against the published median for that pattern type, where the breakout price sits in the twelve-month range, breakout day volume against the stock's average, and market cap. Those four carry most of the spread between the good patterns and the bad ones. Two or more negatives and you pass, no matter how clean the shape looks.
The second change is that you stop shorting these patterns. The book's own horizon failure tables are brutal on the short side: for most bearish setups roughly a third of shorts are already underwater a week after the breakout and close to half are underwater at two months, and fewer than 4% of any bearish pattern falls more than 50%. If you do hold longs, the beta-adjusted trailing stop table in Chapter 4 is usable on Monday: look up the stock's beta and daily high, take the stop percentage off the high (a beta of 1.3 on a $39 high gives 13%, a stop at 33.93), recalculate on every new high, and never move it down.
| Test | What the chart showed | Score | Running total |
|---|---|---|---|
| Run-up length | 132 days from trend start, intermediate term | +1 | +1 |
| Consolidation in the path | None overhead | +1 | +2 |
| Breakout in yearly range | 22.13 breakout equalled the yearly high, top third | +1 | +3 |
| Pattern height | (22.13 - 14) / 22.13 = 37%, against a 17.49% median | +1 | +4 |
| Volume trend | Falling across the pattern | +1 | +5 |
| Breakout volume | Average, no surge | -1 | +4 |
| Partial decline | None before the breakout | +1 | +5 |
| Bull trap | No overhead resistance, none expected | +1 | +6 |
| Market during formation | S&P 500 fell from 1395.86 to 1261.32 | -1 | +5 |
| Market capitalization | Small cap | +1 | +6 |
Final score plus six, and the stock then rose 234% against a 22.81% median for this pattern, which is the result the chapter was chosen to show.
Do not consider the scoring system as a robust, mechanical trading system, but more as an investment checklist.Thomas N. Bulkowski, Trading Classic Chart Patterns
Where it fails
- The scores were fitted and tested on the same data. Bulkowski derived every plus one and minus one from a database of 500 stocks over five years plus 230 more with one to five years of extra data. He then reports the system works because positive scores beat the median 58% to 79% of the time. Those hit rates were measured on the same patterns that set the thresholds, which guarantees they look good. The book contains no out-of-sample table. The preface says he is pleased with out-of-sample tests but never shows one.
- Five years of data, and they were the strangest five years on record. The charts and trades in this edition run roughly 1996 to 2001, covering the back half of the dot-com bubble and its collapse. Two of the book's headline findings, that small caps beat large caps and that breakouts near the yearly high beat the rest, describe that specific tape extremely well. There is no test of whether either plus one survives a flat market. The 2005 second edition of the Encyclopedia later split every statistic by bull and bear market, which is exactly the control missing here.
- Two of the tests ask you to predict the future. The trap score and the throwback score both require you to judge in advance whether prices will reverse after the breakout. Asked directly whether a trap can be predicted, Bulkowski answers yes, if you are lucky. In the Adam and Eve double bottom case study he notes that if the reader disagrees with his bull trap call the difference will not change the outcome anyway. That is an admission that part of the total is opinion carrying the same weight as a measured statistic.
- Every test counts the same, so the total hides which signal fired. Tall against short broadening tops is 42% against 27%, a fifteen point spread. Rising against falling volume trend inside that same pattern is 33% against 36%, three points and well inside the noise. Both are worth exactly one point. A pattern scoring plus two tells you nothing about whether the two points came from the tests that matter or from two coin flips, and the book never publishes a weighted version.
- Several score tables run on samples too small to trust, by his own standard. Bulkowski states that anything under 30 samples may be inaccurate, then keeps scoring off tables that break that rule: 20 samples for long-term run-ups into broadening tops, 14 for the lowest third of the yearly range, 10 and 2 for six and eight touch trendlines in Chapter 1. He dropped diamonds with upward breakouts from the book entirely because he could only find 106 of them, which shows he knew the threshold mattered.
- Half the book argues for trades the same book tells you not to take. Every pattern chapter carries a downward breakout section, and the numbers there are dismal. Broadening tops breaking down lose 16% on average, only 3% fall more than 50%, and 55% are trading above the breakout price two months later. Bulkowski writes out the obvious question himself, asking why anyone would short if half of them are higher in two months, and then keeps scoring short setups for another 300 pages.
- The author moved past the scoring system within three years. The scoring system appears in this book and nowhere else in his catalogue. The 2005 Encyclopedia second edition replaced it with per-pattern checklists ranked by measured effect size, and Getting Started in Chart Patterns does not use it either. Nothing here has been restated or re-tested on the 1991 to 2004 sample that powers the current Encyclopedia. You are buying a 2002 method on 2002 data that its own author retired.
- It quietly assumes data and patience most people do not have. Scoring one pattern requires shares outstanding, the twelve-month high and low, a linear regression across the pattern's volume, the stock's average volume for the breakout comparison, and the S&P 500 close on the first and last day of the formation. On top of that, every performance number measures from the breakout price to the ultimate high, which by definition is not confirmed until price reverses 20%. That is a multi-month hold, and Bulkowski excluded any pattern whose 20% reversal had not arrived before his data ended. Day traders and swing traders get nothing usable. The prices are pre-decimalization too, quoted in sixteenths.
Who it is for
Buy it if
You already own the Encyclopedia, you trade US stock breakouts on daily bars, and you want a repeatable way to reject setups before you commit money. The trend start rule, the beta-adjusted trailing stop table, and the horizon failure rates appear nowhere else in his catalogue. Buy it as your second Bulkowski, never your first.
Skip it if
If you own none of the three, buy the Encyclopedia of Chart Patterns and stop. It covers more than 50 pattern types against the dozen here, it runs on newer data, and its For Best Performance checklists do the same filtering job this scoring system does. Getting Started in Chart Patterns is the cheaper entry point for a beginner and covers the same ground more gently. Owning all three means paying three times for one research database.
As an Amazon Associate I earn from qualifying purchases. It costs you nothing extra, and it does not change what gets recommended or what the verdict says.