Patterns at Levels, Not in a Vacuum

Six shapes, two families

Reversal shapes all describe the same physics: one side took a price and the other side took it back before the close. A hammer has a small body at the top of its range and a long lower wick — sellers reached down, buyers absorbed everything and closed near the high. A shooting star is the mirror at the other end, which is why it matters near highs rather than lows. A bullish engulfing candle has a body that fully covers the previous session’s opposite body, so the prior session’s move was erased in one period; a bearish engulfing is its twin at the top. A morning star is a three-bar sentence: a strong down candle, a small indecision candle, then a strong up candle that closes well into the first one’s body. Read them as sentences rather than spells — each says “this side tried, that side answered, and the close is evidence.” Continuation shapes describe agreement instead. Three white soldiers are three consecutive strong up closes, each opening inside the prior body — buying that keeps arriving rather than exhausting. A marubozu has almost no wicks at all, which means one side was never seriously contested: open at the low, close at the high, no argument in between, and at a breakout level that is a strong statement about who was present. A flag — a sharp move followed by a tight, drifting consolidation of a few sessions — is the same idea on a larger scale, and it is what a breakout often looks like before it resolves. The honest part is the arithmetic of reliability. Large-sample studies put most candle patterns near 55–65% “behaves as expected”, which is meaningfully better than a coin and nowhere near a certainty. Two consequences follow. First, a pattern that fires mid-range, against the higher-timeframe trend, on thin volume is inside the noise of that 55–65% band and should be ignored; the same pattern at a level that has been tested repeatedly, in line with the frame above, with volume expansion, is the version studies find any edge in. Second, a single pattern is one observation: at a 60% base rate, seeing three fail in a row is routine, so judgement has to be in samples rather than in the most recent outcome. That is why the journal in TR12 records the setup and the location, not just the result. What each shape measures, in the terms of TR1 — Hammer / shooting star: Long wick on one side, small body: rejection at a price ← · Bullish / bearish engulfing: Body covers the prior opposite body: one session erased ← · Morning / evening star: Three-bar sequence: exhaustion, indecision, reversal ← · Marubozu: Body ≈ range: no contest, one-sided conviction The engulfing definition is worth being strict about: the body must cover the prior body, not merely the prior range. A candle that wicks over the prior high while closing lower is not an engulfing candle, and calling it one is how a discretionary read becomes a story.

Location is the edge, and it is measurable

A location qualifies a pattern in three ways that a learner can check rather than feel. Repetition: a level that has turned price twice or three times carries orders from participants who remember it, while a level touched once is one transaction. Time: a level tested eight months ago is weaker than one tested last month, because the orders that were there have had time to change. Frame: a hammer that appears at the bottom of a decline on the daily chart, when the weekly is still above a rising trend line, is a pullback in an uptrend — a different trade from the same candle in a downtrend with no support anywhere near. Repetition, recency and frame are all observable before the trade, which is what makes them usable discipline rather than commentary. Volume is the fourth qualifier because it is the only one that measures participation at the moment the pattern fires. The same hammer on 2.4 times average volume says thousands of participants agreed at that price; on 0.6 times average it says the session was thin and the wick was one order away from noise. The threshold matters less than the habit of checking: if the pattern cannot show that participation arrived, treat it as an observation and wait for the next close. And the next close is the last qualifier, because a pattern is a claim the following session has to confirm — a hammer followed by a close below its low is a failed statement, not a signal that “needs time”. • Repetition: how many times the level has turned price, ideally on a chart older than this month. • Recency: orders at a level decay, so an old level is a weaker claim. • Frame: the same candle is a pullback in an uptrend and a hope in a downtrend. • Volume: participation at the moment of the pattern, measured against the average. • Confirmation: the next close, which either agrees with the statement or refutes it. Pattern names are a shared vocabulary, not evidence. Two traders calling the same candle a “hammer” have agreed about a shape and nothing about whether it matters, so the label should never appear in a journal entry without the location beside it: “hammer at the 57.50 level, third test, 2.4× volume” is a setup. “Hammer” is a word.

How to test a pattern honestly

The base rates attached to patterns in published work are close enough to coin flips that the interesting question is not whether a shape works but whether a shape works **more often than the same market in the same conditions without it**. That is a different claim, and testing it requires a benchmark. The honest procedure has four steps, and the third is the one almost everyone skips. First, define the pattern mechanically enough that a computer could find it — a close below the prior low, a body larger than the average of the last twenty bars — because a definition that requires your judgement cannot be tested. Second, collect every instance over a long enough sample, including the ones that looked ugly, since selective collection is how a pattern acquires an imaginary hit rate. Third, record what the instrument did over the same number of bars **after every bar in the sample**, whether or not the pattern was present. Fourth, compare the two: the pattern’s edge is the difference between them, not its hit rate on its own. The benchmark step is where most pattern claims die, because a rising market produces a rising hit rate for almost any long signal. If the unconditional probability of a gain over the next five bars is 54%, a pattern with a 56% hit rate is barely distinguishable from owning the instrument, and the small difference may not survive the spread you pay to act on it. This is also why patterns behave so differently across regimes: a continuation pattern will look excellent in a trending market and worthless in a range, and the base rate quoted from one sample will not transfer. Two further cautions keep the exercise honest. The first is **multiple comparison**: if you test twenty shapes and report the best, the reported number reflects the search as much as the shape. The second is cost — a pattern that produces a small statistical lift can still be unprofitable once the spread and the slippage of entering on the next bar are deducted. A pattern does not need to be a secret to be useful, but it does need a lift larger than the friction of trading it, and for most shapes that is a higher bar than the folklore suggests. • Define the shape mechanically, so the sample cannot be curated by your judgement. • Compare against the unconditional outcome over the same number of bars — the lift is the edge. • Test in more than one regime; a base rate from one sample does not transfer. • Deduct the spread and the entry slippage before believing a small lift.

Every shape carries its own invalidation, and the target is a distribution

A pattern is a claim, and a claim has a price at which it is false — which the shape itself supplies, so nothing has to be invented. A hammer is refuted when its low trades; an engulfing candle is refuted at the low of the engulfing bar for the bullish version and the high for the bearish one; a morning star is refuted below the low of the small inner bar, because that bar is the moment the two sides were balanced and losing it means the buyers never really took control. Using the pattern’s own extreme is the tightest honest stop available: a percentage stop is a guess placed on the chart, whereas the pattern’s extreme is a level the market itself defined during the session that made the claim. It is also, deliberately, a stop inside the ordinary range of the instrument — which is exactly why the pattern has to be at a level that matters, since a structural stop with no structure behind it is a tight stop with no thesis. The second number the shape supplies is a **measured move**. A flag has a height — the sharp move that preceded it — and the folklore projects that height from the breakout point; a double bottom projects the depth of the trough to the neckline. Treat the projection as a modal expectation rather than a target, and its value becomes pre-trade arithmetic instead of a forecast. A flag four percent tall above a breakout at 100 gives 104 as the reference case, and against a stop at the flag’s low — say 98 — the trade is now a six-point reward for a two-point risk, which is a statement that can be compared with the base rate before any money is committed. The measured move is not what the market owes you; it is the number that lets you decide whether the setup’s payoff is worth its 55–65% hit rate at all. When the projection is smaller than the stop distance, the pattern is telling you the trade is badly shaped regardless of how clean the shape looks. The third idea is the one that turns pattern reading from a prediction into a probability exercise: **the failed break is itself a setup**. A breakout that reverses back inside the range within a bar or two has trapped everyone who acted on it, and the position of the trapped traders is the fuel for the move in the other direction. In the same large-sample data the shapes come from, the failure rates of the classic continuation patterns sit high enough — commonly a third to a half — that the reversal of a failed breakout is not a consolation prize but a strategy in its own right. The practical consequence is that the entry trigger and the invalidation price are the same geometry seen from opposite sides: the level that would make the bull case true is the level that makes the bear case true, so a trader who has written both down is never without a trade, only without a direction. • Let the shape set the stop: hammer at its low, engulfing at its body’s low, star below the inner bar. • Read the measured move as a reference case for reward-to-risk, not as a target the market owes you. • If the projected reward is smaller than the refutation distance, the shape is decorative. • A failed breakout is a setup: the trapped positions are the fuel for the reverse move. • Entry trigger and invalidation are the same level, read in opposite directions. This is why the journal entry from the previous read asks for the level and the trigger rather than the pattern name: “bullish engulfing at the third test of 57.50, refuted below 56.80, measured move to 60” is a trade with a stated way to be wrong. “Bullish engulfing” is a shape with no price in it, and a shape with no price cannot be sized, stopped or reviewed.

Sources

• Pattern base rates over large samplesBulkowski, “Encyclopedia of Candlestick Charts”
• Reversal shapes and their context rulesNison, “Japanese Candlestick Charting Techniques”
• Why location dominates the shapeSchwager, “Technical Analysis” (interviews on pattern context)