Takuri line candlestick: definition, TA-Lib test and results

Takuri is a Japanese word usually glossed as "trying to gauge the depth of the water". Picture someone wading out until their foot finds the bottom, then stepping back. On a chart it describes a session that plunges, touches a level where buyers are waiting, and returns to finish at the top of the day.

The term is used loosely. Some writers apply it to the Hammer candlestick pattern: meaning, rules and reliability in general. Others reserve it for an extreme version: a doji, or nearly so, with no upper shadow and a lower shadow far longer than anything around it. TA-Lib, the library Finabase uses to detect patterns, takes the second meaning but with a definition that turns out to be much looser than "far longer".

This page explains the shape, reads TA-Lib's CDLTAKURI rule line by line, shows why it is practically the same detector as the Dragonfly doji: meaning, TA-Lib rules and a 380-case test, walks through a Nike takuri that worked and an Oracle takuri that did not, and reports what 376 real cases did next.

What the textbook takuri looks like

The textbook picture has four parts:

1. A prior decline, so there is selling pressure to be tested.
2. A body that is a line or close to it, at the top of the candle.
3. No upper shadow, or a negligible one: the session never got above its open.
4. A lower shadow that dwarfs everything nearby, often described as at least three times the body and long compared with recent candles' full ranges.

It is the most emphatic of the "rejection from below" candles. A hammer allows a real body and a shorter tail; a dragonfly doji allows a much shorter tail; the takuri is the version where the probe downward was deep and the recovery complete.

The exact rules Finabase uses

Finabase marks takuris with TA-Lib's CDLTAKURI on daily bars:

The third line is the one that matters. TA-Lib's "very long shadow" setting (ShadowVeryLong) has an averaging period of zero, so it is measured against the candle itself, its real body, with a factor of 2.0. The tail must be more than twice the body.

For an ordinary candle that would be a real filter. For a candle that has already passed the doji test, it is not. If the body is 10 cents, a 21-cent tail qualifies. If the open equals the close exactly, any tail at all qualifies.

Compare that with the Dragonfly doji: meaning, TA-Lib rules and a 380-case test, which uses the same first two tests and asks for a lower shadow longer than 10% of the average recent range. In theory each can fire where the other doesn't. In practice the two almost always agree:

The textbook's other requirement, a tail that is long compared with other candles, is not tested at all. In our data only 14 of the 376 takuris had a lower shadow at least as long as the average full range of the previous ten sessions, while 198 had a tail shorter than half of it. Most Finabase takuri markers are therefore modest dragonflies, not the deep plunge the name suggests.

What the deep probe tells you

When the tail genuinely is long, the candle compresses a dramatic session into one shape. Sellers pushed price far below the open, possibly through a level many traders were watching. Stops were triggered, and the selling probably accelerated for a while. Then the bids that met the decline were large enough to reverse all of it by the close.

That says two things. First, the market found demand at a definite level, the low of the tail, which becomes a natural reference point. Second, the sellers who drove price down that far are now holding positions that are losing money at the close, a common source of fuel for a further move up if the next session holds. Neither point guarantees anything. They describe why traders watch the candle, not what it will do.

Confirmation

Practitioners treat a takuri like a hammer: a candidate, not a conclusion.

• The next session closes above the takuri's high. Since the high and the close are almost the same, this is a low bar.
• The low holds. Many traders treat any close below the tail's low as the end of the idea.
• Supporting context: a decline of several percent into a level that already mattered, rising volume on the recovery, or a momentum reading such as the Relative Strength Index (RSI): how to read and use it leaving oversold territory.

A takuri that worked and one that failed

Nike, December 30, 2025. NKE had fallen 9.7% in ten sessions. On December 30 it opened at $61.30, which was also the high of the day, so the upper shadow was zero. It dipped to $60.64 and closed at $61.19. The lower shadow was 55 cents, five times the 11-cent body, which is a convincing takuri by TA-Lib's standard, although the whole candle spanned less than half the recent average range. The next session closed at $63.71, 4.1% higher and well above the high: confirmed. Over the next ten sessions the stock never traded below $61.70, let alone near the takuri's low, and ten sessions after the pattern it was 7.2% higher.

Oracle, June 12, 2026. ORCL had fallen 18.4% in ten sessions. The June 12 candle opened at $185.01, dipped to $179.00 and closed at $184.13, with a high only 57 cents above the open. The $5.13 tail was almost six times the 88-cent body. The next session closed at $192.64, 4.6% higher and above the high, the same textbook confirmation the Nike case had. Then the decline resumed. Five sessions after the takuri ORCL was 4.9% lower; ten sessions after it, 19.8% lower, after trading as low as $145.20. A trader using the takuri's low as a stop would have been out near $179, with a loss of about 7% from the confirmation close instead of 25%.

The two cases are a fair summary of the pattern: the same shape, the same confirmation, opposite outcomes. What separated them was what the market did next, not anything visible in the candle.

What the data says

We ran CDLTAKURI on daily bars from October 30, 2024 to October 29, 2026 for 60 large US stocks and ETFs. It fired 376 times across all 60 symbols; 367 cases had ten sessions of data afterwards. Rows marked with an asterisk are our own cuts of the same data.

The average ten-session return after a takuri was +1.00%, against +0.66% for any day. 49.0% of the measured cases followed a ten-session decline.

How to read it:

• The takuri numbers are the dragonfly's numbers. With a 99% overlap that is unavoidable. Both showed a few points more upside than an average day.
• The prior trend did not matter. Takuris after a decline did no better than the group as a whole. The shape, not the reversal setting, carried whatever edge there was.
• Confirmation made it worse. From the confirming close, 51.8% were higher five sessions later, below the baseline. The next session's gain absorbed much of the pattern's average effect.
• The textbook-length tails look striking and prove nothing. Twelve of the 14 long tails were followed by gains, but 14 cases in two years is too small a sample to conclude anything.

Takuri versus related candles

107 of the 376 takuris were also flagged as hammers. The hammer is the broader pattern in spirit, allowing a visible body, but TA-Lib's version adds a location test (the body must sit near the previous candle's low) that most takuris did not meet.

An educational outline

This is a study framework, not a recommendation to trade.

1. Insist on a real tail. Compare the lower shadow with the average range of the last ten sessions. TA-Lib will not do this for you.
2. Look for a decline into a level. A prior low, a round number or a long moving average gives the tail something to have tested.
3. Use the tail's low as the line in the sand. A close below it ends the idea.
4. Size from the distance to that low. Deep tails mean wide stops and smaller positions.

How Finabase shows takuri lines

• Chart markers. The Takuri study marks each CDLTAKURI detection. If you also add Dragonfly Doji, you will find the two sets of marks on almost exactly the same candles.
• Scanner. TA-Lib returns +100 for every takuri, and the candlestick scanner files detections by sign, so takuris appear only in the Bullish list (S&P 500 takuri lines). The scanner does not check for a prior decline.
• Signals-tab gauge. Each of the 61 patterns votes from its most recent detection within the last eight bars. The takuri always votes Buy, which it did on 9.5% of stock-days in our sample. Because a takuri is also a dragonfly, a doji and a long-legged doji, one such candle usually supplies four Buy votes to the candlestick gauge at once.

Frequently asked questions

Is a takuri the same as a hammer?

They are close relatives and the names are sometimes used interchangeably. In TA-Lib they are different tests: the takuri needs a doji body and no upper shadow, while the hammer allows a small body and requires the body to sit near the prior candle's low. Only 107 of 376 takuris in our sample were also hammers.

How is a takuri different from a dragonfly doji?

In the textbook, the takuri's lower shadow is exceptionally long while a dragonfly's can be modest. In TA-Lib the takuri asks for a tail more than twice the body and the dragonfly for a tail more than 10% of the recent average range; in our data every takuri was also a dragonfly.

Does a takuri need confirmation?

Most practitioners say yes. In our sample, measuring from a confirming close lowered the five-session hit rate from 56.7% to 51.8%, so confirmation did not improve results; it mainly protects you from cases where the low breaks immediately.

Why does Finabase show a takuri as bullish even in an uptrend?

TA-Lib returns +100 for every takuri and does not test the trend. Finabase sorts patterns by that sign, so the bullish label appears regardless of context.

Summary

The takuri line is meant to capture a deep probe below the open that was completely recovered, a doji with a long tail and no upper shadow. TA-Lib defines the tail as longer than twice the doji's own body, which in practice makes CDLTAKURI a near copy of the dragonfly doji: all 376 of its detections in our test were dragonflies, and only 14 had the long tail the name implies. The group did a few points better than an average day, regardless of trend, and confirmation did not help. On Finabase, treat the marker as a dragonfly marker and check the tail's length yourself.