Encyclopedia of Chart Patterns — Bulkowski: COMPLETE

Before Bulkowski there were two kinds of pattern literature: hand-drawn classics like Edwards & Magee, full of beautiful examples but without a single outcome frequency, and trivial cookbooks without depth. Bulkowski's revolution was to treat every formation as a statistical experiment. He collected an archive of thousands of measured formations distributed over several dozen types, identified each individual case according to rules set up in advance, and then noted what happened after the breakout — on average, at best, at worst, and how often it failed.

A pattern that looks like a head-and-shoulders and behaves like a head-and-shoulders is not enough; it must be measurably defined for the result to be replicable.

The method's backbone is the division by market state. Bulkowski divides the entire material into primary bull and bear markets before he counts anything at all, because the same formation shows different expected value depending on the prevailing climate. From there he builds his performance tables: average rise or decline from breakout to the best and worst point respectively under his measurement horizon, the share of cases that give an adverse move beyond certain thresholds, how often the course of the price is thrown back to the breakout level, how long the moves take, and a weighted performance rank that sorts the formations against each other.

The tables are the book's actual content — the prose is only the explanation of the numbers.

Three definitions you should lock in right away. A breakout is as a rule a closing price through a formation's defined level — convergence line, neckline, or resistance — not an intraday intrusion that closes back. The measurement point is the breakout moment, not the formation's start, which makes the statistics tradeable: the numbers describe what happens if you actually buy the signal.

And the failure rate is defined as the share of patterns where the course of the price moves less than a fixed threshold — on the order of 5 to 10 percent depending on edition and formation — in the expected direction after the breakout. Everything in the rest of the course hangs on these three load-bearing concepts.

The book's four financial ratios do the following. The average rise or average decline is measured from the breakout to the best or worst point under the measurement horizon — on the order of 20 to 40 percent for most reversal bottoms, with large variations between formations and editions. The failure rate is the share of cases where the course of the price does not even reach a minimal threshold move in the expected direction; here the formations span from low single digits up to over twenty percent — a fearsomely large difference in expected value.

The throwback rate tells how often the course of the price returns to the breakout zone before the continuation, and the rank sorts the formations against each other within his material.

An average rise is a historical fact, not a physical law — and Bulkowski is himself the first to point that out. The numbers shift between editions because samples and periods shift: several averages fell when he moved from the first edition's material to the second's, and he openly reports the number of observations per formation so the reader can judge the uncertainty.

Some formations rest on hundreds of cases; others on considerably fewer, and their table rows are thus noisier. The rank is moreover relative: being number one in one sample does not mean being number one in the next decade or on the next market.

Three pitfalls worth engraving. First, averages are pulled up by tails — a few monstrous moves can lift an average most traders never get to experience; the median often lies lower. Second, the numbers are about raw price moves before transaction costs, slippage, and tax, which eats into the gross margin. And finally there are sampling problems and effects of multiple testing of a valuation multiple of formations: when dozens of formations are tested on the same history, some will look good by pure chance.

Bulkowski's discipline — report sample, count, and definitions — makes his tables better than most, but not immune to this.