Marubozu, and how often candle signals fail

One more shape, then the uncomfortable part: how often these patterns actually lead where they point. The answer is "not much more than half the time on their own", and understanding why is more valuable than any pattern.

Marubozu

A marubozu is a candle with no wicks, or nearly none: the open is the low and the close is the high (bullish), or the open is the high and the close is the low (bearish). This course allows wicks up to 5% of the range on each side.

   Bullish        Bearish

   +---+          +---+
   |   |          |###|
   |   |          |###|
   |   |          |###|
   |   |          |###|
   +---+          +---+

Who is trapped: everyone on the other side, all period long. In a bullish marubozu, price never dipped below the open; every seller during the period sold at a price that was immediately exceeded. There was no rejection, no pullback, no moment where sellers were winning. It is the most one-sided candle there is.

The read depends, as always, on location:

• Out of a range or base, with the marubozu closing above a level: a breakout with conviction. Sellers at the level were overrun without a fight.
• As the fifth or sixth large candle in a row after a long trend: possibly a climax. The last buyers are piling in with no one left to sell to them. The next candle often has a long wick.
• Mid-trend, of average size: continuation. Not remarkable.

A marubozu is honest about what happened in its period and silent about what happens next. Its size relative to average range is the signal strength: a marubozu with a range 2.5× the average is a different animal from one at 0.8×.

Key idea: A marubozu is maximum one-sidedness for its period. At the start of a move that is conviction; at the end of a long move it is often the last buyer or seller arriving.

How often patterns work

Here is what independent tests of candlestick patterns tend to find, across many markets and decades of data. The numbers below are rough magnitudes from the published literature, not precise facts about the chart in front of you, and results vary with the definition used, the market and the era:

Read that table honestly. A pattern that resolves as expected 55% of the time is a real edge only if it comes with a reward-to-risk that makes 55% profitable, and the same pattern, traded with a 1:1 target and a stop that gets hit by noise, loses money after costs. Many of the "60%" numbers you see quoted also measure whether price simply closed higher one candle later, which is a much easier target than reaching a profit target before a stop.

This is not an argument that candles are useless. It is an argument that the shape by itself is close to a coin flip, and the rest of the course is about what moves it away from a coin flip.

Why the raw rate is low

Three reasons, all of which you can fix:

1. Most patterns occur in the wrong place. An engulfing candle in the middle of a range traps a small group at a price nobody else cares about. When you count every engulfing candle on a chart, you are mostly counting these. Filtering to patterns that form at a level (Module 3) removes most of the noise.
2. Most patterns are against the larger trend. A hammer in a downtrend is a bounce in a downtrend. Counting bounces as "reversal signals" and then measuring whether the trend reversed produces a terrible hit rate. Filtering to patterns that form with the trend (Module 4) removes another large chunk.
3. The measurement is wrong for trading. "Closed higher next candle" is not what you trade. You trade a stop at the pattern's extreme and a target at the next level. Measuring that (Module 6) gives a different and more honest number.

When you apply the first two filters, the number of patterns on a chart drops by 80-90%, and what remains behaves noticeably better. It still fails a lot. Anyone who tells you a candlestick setup wins 80% of the time is either measuring something trivial or not measuring.

What a failed signal looks like

A failed pattern is itself information, often better information than the pattern was. A hammer at support that is followed by a close below the hammer's low means the buyers who "defended" the level have been run over and are now trapped themselves. That is a stronger bearish signal than the hammer was a bullish one, because it has two trapped groups (the wick shorts who have now been rescued and the hammer buyers who are now losing) and one has just been proven wrong in public.

Module 5 builds a whole setup out of this idea (the failed breakout).

Putting the six together

Every entry in the last column is "weak" or "none". That is not pessimism; it is the honest starting point. The next two modules add the two things that make any of them worth trading: a level and a trend.

Try it: Pick one pattern (engulfing is easiest) and one daily chart. Find every instance in the last 200 candles using the strict definition. For each, record whether the close five candles later was in the expected direction. Compute the percentage. Then repeat counting only the instances that formed within one average range of a swing high or low. Compare the two percentages. You have just done your first filter test.

Recap

• Marubozu: no wicks (≤5% of range each side). Maximum one-sidedness; conviction at the start of a move, often climax at the end.
• In isolation, candle patterns resolve as expected roughly 50-65% of the time; many quoted figures measure a trivial target.
• The raw rate is low because most patterns occur in the wrong place, against the trend, or are measured with the wrong target.
• Filtering by level and trend removes most instances and improves what remains; it does not make anything certain.
• A failed pattern is often stronger information than the pattern itself, because it adds a second trapped group.