图表形态：证据说明了什么，又没说明什么

Chart patterns are technical analysis at its most storied: the head and shoulders, the double top, the flag, the triangle. Every trading book carries the same gallery, drawn with the same clean lines, and the lines always lead where the caption says. This sheet keeps the gallery but swaps the folklore for base rates: what each pattern strictly is, where its target number comes from, what published research actually supports — and why a failed pattern deserves as much of your attention as a completed one.

Why pattern lore survives: hindsight is a flattering editor

Every pattern illustration you have ever seen was chosen because it worked. The author searched completed charts for a clean head and shoulders that resolved exactly as the theory says, cropped it, and labelled the parts. The hundreds of look-alike formations that broke the “wrong” way were not photographed. This is selection after the outcome — the textbook gallery is a highlight reel presented as a base rate.

Two features of human perception keep the reel convincing. First, the eye finds shapes in randomness: charts generated from pure coin flips reliably produce formations that experienced traders will happily name and trade. Second, memory keeps score unevenly — the double top that nailed a major high is retold for years, while the five that failed the same month dissolve. Folklore compounds the way anecdotes do, not the way evidence does.

None of this proves that patterns are empty. It proves something narrower and more useful: the evidence for them cannot come from looking at galleries. It has to come from strict definitions, written down before the outcome, and counted. That is the standard this sheet applies — and the standard you should demand from anyone making a pattern claim at you.

The big four families, defined strictly

The classic patterns sort into four families by what they claim about the move that came before them. The family matters more than the individual name, because each family has its own precondition — and a pattern quoted without its precondition is folklore by definition.

Strictness is the whole game, so take one example apart. A double top is two highs within a few pips of each other, separated by a meaningful trough — and, the part routinely skipped, a close below that trough. Until the trough breaks, “double top” is a guess about the future wearing the name of a pattern. The same applies across the gallery: the head and shoulders needs its neckline broken, the flag needs its break in the original direction. Patterns are only readable after completion, which means real entries come later, and at worse prices, than the textbook drawings imply.

The preconditions do equal work. A “reversal pattern” in the middle of a sideways range has no trend to reverse — whatever the silhouette looks like, the claim behind it is unavailable. Naming the family forces you to check the precondition first, which quietly disqualifies most of what the eye wants to label.

Measured moves: where the target number comes from

Most pattern lore ships with a target: project the height of the pattern from the point where price breaks out, and that distance is the “measured move”. It is worth seeing exactly how mechanical this is.

Now the money, both ways. A trader sells one standard lot at the neckline break, 1.0850. If price reaches the measured target, the trade earns 100 pips — $1,000 at $10 per pip. If the break fails and price travels 100 pips the other way instead, the same arithmetic charges $1,000. The projection is symmetric arithmetic, not a promise, and nothing in the drawing tilts which outcome arrives.

Where does the convention come from? The early twentieth-century charting tradition — it was chosen because it is drawable with a ruler, not because it was derived from data. Counted studies disagree on how often measured moves complete, and the honest range is wide: partial moves are common, full completions are far from guaranteed, and the rates shift with market and period. The defensible use is comparative, not predictive: if the projected move is 100 pips and the honest stop on the failure side is 80, the trade needs a completion rate that no published count supports. The target's best job is to disqualify trades, not to promise destinations.

What pattern studies actually show — and what they can't

The evidence on chart patterns comes in two genres. The academic genre detects patterns algorithmically — smoothing the price series and scanning for defined shapes across thousands of instruments — and then compares what returns follow. Its headline finding, replicated in various forms: some patterns coincide with return distributions that differ measurably from random. That is genuinely interesting. It is also far short of a trading edge — the differences are small, they vary by market and decade, and they shrink or vanish once realistic spreads and slippage are charged against them.

The practitioner genre counts by hand: thousands of patterns catalogued from historical charts, with per-pattern “success rates” that get quoted endlessly — often in the 60% to 80% range. Those counts represent real effort, and they are where most pattern statistics in circulation come from. They also carry methodological limits that the quotations always drop:

• Definition drift: loosen the rules and any chart contains any pattern; tighten them and the sample shrinks toward anecdote.
• Hindsight selection: detecting patterns on completed charts uses information no live trader had at the right edge.
• Generous success criteria: “price moved in the predicted direction at all” is a far lower bar than “the measured move completed before the stop was hit”.
• No costs: spreads, swaps, and slippage are absent from most counts — and costs are exactly what decide marginal edges.
• Regime dependence: a rate measured on 1990s stock charts transfers poorly to this morning's EUR/USD.

There is also a quieter, mathematical reason to distrust precise rates: strict patterns are rare. A head and shoulders that genuinely meets its definition might appear a handful of times per year on a daily chart, so even an honest personal count takes years to reach a sample worth trusting — and by the time it has, the market that produced the early instances has changed underneath it.

Put the two genres together and the honest summary is one sentence: patterns are weak evidence, sometimes, in context — never instructions. When someone quotes you a precise success rate for a pattern, the questions that matter are mechanical: defined how, counted by whom, measured against what bar, net of which costs. Claims that survive those four questions are rare, and none of the survivors resembles the posters.

The failure case is information

A strictly defined pattern has one virtue that needs no folklore: it tells you where other people's orders are. A textbook double top that breaks its trough attracts sellers at a known price; their stops sit just above the break, at another known price. If the move fails and price climbs back through the pattern, those stops are bought back — forced buying, stacked at a predictable level, fuelling the very move that hurt them. This is why failed patterns so often travel fast: the fuel was loaded by the people trading the textbook.

Seen this way, a pattern stops being a prophecy and becomes a map of commitments. Completion and failure are both tradeable hypotheses, each with a definable trigger and a definable invalidation, and some experienced traders prefer the failure side precisely because the trapped side's exits are identifiable. The symmetry from the measured-move example holds here too: the same 100 pips pays $1,000 to whichever side read the evidence correctly and charges $1,000 to the other. Neither side owns the odds — what changes is only the quality of the evidence each is acting on.

One mechanical caution belongs here rather than in a footnote: breaks are where execution gets rough. Spreads widen at the moments patterns resolve, fast moves fill stops beyond their level, and a position held across a weekend can reopen past both stop and target. The sheet on slippage, gaps, and fills covers the machinery; the planning rule is simply that pattern trades must budget for worse fills than the drawing assumes.

A checklist that forces definition before entry

Everything above condenses into a discipline you can apply to any pattern claim, including your own. The checklist's purpose is to move every definition to before the outcome — which is the single difference between evidence and folklore:

1. Name the pattern and its family while it is still forming — in writing, with the precondition checked.
2. State the completion condition as a price event: what must close where.
3. State the failure condition the same way, and place the stop there — not at a round number of pips.
4. Project the measured move and compare it with the stop distance, in pips and in money at your size.
5. Size the position from the stop distance, never from the target.
6. Log the outcome against the written definition, and judge the pattern only on a sample of twenty or more.

Run that discipline for a season and you will have something no book provides: your own base rates, on your pairs, your timeframe, your costs. Most pattern claims do not survive being written down in advance. The few that do are the only ones worth carrying into the next sheet's territory — timeframes, and then the full machinery of testing a system.