Gravestone Doji: 25 Events, 44 Percent
A gravestone doji opens and closes at the bottom of its range, leaving a long upper wick and effectively no body. It is read as a rally that was completely unwound, and on this site's history it occurs 25 times, which is too few occurrences to distinguish a pattern from chance.
How it works
A gravestone doji opens and closes at essentially the same price, at the bottom of the bar’s range. The entire bar is upper wick. On a chart it looks like an upside-down T, which is where the name comes from.
It is the dragonfly doji inverted, and everything structural about that page applies here with the signs reversed: the same near-zero body threshold, the same sensitivity to tick size, the same dependence on where you cut the bars.
The story is a session that rallied and gave every bit of it back. Buyers pushed, sellers absorbed the push, and by the close price was exactly where it started. Told at the top of an advance it sounds like exhaustion.
The number, and why the number cannot settle it
On this site’s shared 576-bar history, 25 gravestone doji appear — 4.3 per hundred bars, almost exactly matching the 26 dragonflies in the same series.
Of those 25, the next bar closed higher 44% of the time.
Against a base rate of 51% across all 575 transitions in the same data, that is seven points below baseline — the largest deviation of any candle pattern counted on this history, and in the direction the pattern claims.
And it means nothing, because 25 is 25. Eleven of the 25 were followed by a higher close. Had two more gone the other way, the figure would be 52% and this page would be reporting no effect at all. A result that flips on two observations is not a result.
That is the honest position, and it is worth stating plainly because the temptation runs the other way. This is the one pattern on this site whose measured number points where the textbook says it should, which makes it exactly the number most likely to be quoted and least likely to survive.
The wider point transfers to every source you will read. A pattern occurring 4 times per hundred bars gives roughly 25 events per 600 bars, or about 100 in a decade of daily data. That is the sample size available to anybody studying it, including the people publishing confident percentages.
In practice: how the shape gets manufactured
A gap up that fades through the session produces a textbook gravestone, and there was no rally to unwind. The high was set at the open by the reopening price. The “sellers absorbed the push” story describes an event that did not occur.
Aggregation destroys it. A daily gravestone is, on a weekly chart, part of a bar with an ordinary body. Nothing about the market differs between those two views; only the bar boundaries do.
Volume is the one genuinely separate input. A full round trip on heavy participation means real size traded on both legs. The same shape on almost no volume means very little traded at all, which is itself the reason the body is missing.
The wick records where price went, not who was waiting there. By the close, whatever orders sat at the high are gone from the order book — filled, cancelled or expired. The idea that “sellers are up there” is an inference about the future from a fact about the past.
Each one traded costs 2% of a typical bar’s range in round-trip costs on this history, before any borrow cost if the trade is a short. Twenty-five attempts is twenty-five fees against a difference the sample cannot confirm.
What a gravestone doji is not
It is not a top. It is a bar shape that appears at some tops, and in a great many other places.
It is not a shooting star. A shooting star has a small but real body. This has effectively none, and the boundary between them is a threshold somebody chose.
It is not a level. The high of the wick is a price price visited once. Resistance requires price to have been turned away there repeatedly.
And it is not evidence of selling pressure. It is evidence that price rose and returned. Whether that was sellers arriving or buyers leaving is a distinction the candle cannot make and the volume histogram can partially address.
When it fails
In a range the ceiling produces them repeatedly. Price reaches the top, turns, closes near its low: the definition is satisfied by the ordinary behaviour of a range. The pattern appears most often precisely where a reversal signal tells you nothing you did not already know.
The second failure is confirmation from the one flattering statistic. The 44% figure on this page points the way the textbook says. It is on 25 events, it is unstable to two observations, and treating it as support for the pattern is exactly the reasoning error the count was computed to expose.
A third is the missing context clause. The pattern is defined as appearing after an advance. In the middle of a decline the same shape is a different thing entirely, and screeners flagging shape alone do not check.
A fourth is the tick-size artefact. On instruments where the minimum increment is large relative to movement, open and close land on the same tick frequently, and the pattern’s frequency reflects the instrument rather than the participants.
And a fifth is shorting into strength on one bar. Whatever the shape says, a single candle against an established advance is one bar of evidence against many, and the stop sits above a high that price just demonstrated it can reach.
The original data
25 gravestone doji in 576 bars, followed by a higher close 44% of the time, against a 51% base rate
across all 575 transitions in the same series. The counts, thresholds and base rate are in
research/series-measurements.json, produced by site/measure_series.py.
This page publishes a number that flatters the pattern and then declines to use it, which is the useful part. Eleven higher closes out of 25 is a result that two different bars would erase, and it comes from a synthetic series with no participants in it at all. If a figure this fragile is the strongest result in the whole set, then the honest summary of candle patterns is that nobody — this site included — has the sample size to make the claims that get made about them.
Related
Dragonfly doji is the mirror image and carries the doji frequency split. Candlestick patterns is the parent page for the naming system. And reversals treats the event this pattern claims to mark on its own terms.
I have shorted gravestone doji and I have watched them mark nothing at all, and for a long time I remembered the first kind and not the second. Counting them was uncomfortable in a useful way - it put a number on how thin the evidence was for something I had been treating as obvious.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.