Three black crows are three consecutive bearish candles with specified opening, closing and shadow relationships. Recognizing them requires more than counting red bodies. The pattern name does not reveal who traded, guarantee another decline or provide a complete entry rule. Define the candle geometry and preceding context before studying the outcome.

What the three black crows look like

TradingView describes three long bearish bodies, openings inside preceding bodies and closes at or near their lows. The terms long and near need numerical definitions for a test. For candles 2 and 3, compare the opening with the preceding candle in the trio; a condition involving the candle before the trio depends on the selected implementation. Declining closes do not require declining Highs, and three declining closes alone do not ensure that every candle is bearish.

Record the preceding context independently of the later price move. A prior advance is a contextual filter; it does not prove that the formation marks a top. Basic body and shadow geometry is introduced in candlestick patterns worth knowing first.

An explicit recognition rule

The following is a custom mathematical illustration, not a historical performance result. Require three bearish candles with strictly declining closes; O2 strictly inside candle 1’s body and O3 strictly inside candle 2’s body; each body at least twice a fixed mean of the ten bodies ending before candle 1; and each lower shadow no more than 10% of its own H − L range. Also require the four closes immediately before the trio to rise successively. Equality at an opening boundary fails this version.

For fictional EUR/USD, assume one pip = 0.0001. Each of the ten earlier bodies is 12 pips, so the fixed long-body minimum is 24 pips. The last four preceding closes are 1.0960, 1.0980, 1.1000 and 1.1020. These chosen thresholds make the example checkable; they are not optimized parameters.

CandleOpenHighLowCloseBody / range / lower shadow
11.10201.10251.09881.099030 / 37 / 2 pips
21.10101.10151.09781.098030 / 37 / 2 pips
31.10001.10051.09681.097030 / 37 / 2 pips

Every body exceeds 24 pips, and each lower shadow is 2/37 × 100% ≈ 5.41% of its range. The openings satisfy 1.0990 < 1.1010 < 1.1020 and 1.0980 < 1.1000 < 1.1010. The sum of the bodies is 90 pips, but O1 to C3 changes by only −50 pips: two openings are 20 pips above their preceding closes. Overlapping candle ranges do not mean that consecutive Close and Open are identical.

Change only O3 to 1.0980: candle 3 remains bearish, but its body is now 10 pips and its opening equals C2. It fails both the 24-pip body minimum and the strict-inside condition. Allowing that equality would be a different version, which must be fixed before examining results. The formation is known only after candle 3 closes.

The mirror: three white soldiers

The bullish counterpart, three white soldiers, uses rising bullish candles with an opening/body and upper-shadow specification. A conceptual mirror does not make two software detectors exact algebraic inverses, and neither name establishes identical trade performance. Their respective definitions and preceding contexts must be checked separately.

Why a detector can disagree with the description

A very large final candle can leave a short entry far from invalidation or nearby support, reducing the planned reward relative to risk. That is a reason to reassess the trade, but its size alone does not establish that a bounce is more likely than continuation or identify short covering.

TA-Lib 0.6.4 CDL3BLACKCROWS has its own conditions, including a preceding non-bearish candle and its High relative to the first close. It checks declining closes and short lower shadows, but does not separately require long bodies or a mature preceding uptrend. Its default very-short-shadow threshold uses 10% of the preceding ten candle ranges, rather than 10% of each current range. The custom table above is therefore not a TA-Lib compatibility test.

A synthetic code-condition example makes the distinction concrete. Place twelve candles (100,120, 90,110), followed by (98,103, 96,101), before this trio of O/H/L/C: (100,100.5, 97.9, 98), (99, 99.5, 96.9, 97), (98, 98.5, 95.9, 96). Each body is only 2 units and each lower shadow 0.1. The first body is smaller than the earlier body mean (9 × 10 + 3)/10 = 9.3. Nevertheless, the source conditions give −100 because they do not include our long-body filter. Changing only the final Low to 92 makes that lower shadow 4 units and gives 0. These are hypothetical classification checks, not trade outcomes.

The last High need not be the whole pattern's peak

In this separate model, every body is 30 units and lower shadow 2; closes fall and opens 90,80 lie inside previous bodies. But candle 2 has High 140, far above the final High 85. Our rule limits lower shadows without capping upper shadows.

Highest High in the middle candleThree falling candles with bodies 30 and lower shadows 2. Highs 105,140,85 make the second highest. Lines show 140 and 85 on the shared 48–145 scale.Whole-pattern High =14014085123High 3 =85; distance to peak =55
All geometry fits one scale. Candle 2's long upper wick remains visible although its body passes the same condition as the others.

A predefined “above the last candle” reference uses 85; “above the entire formation's maximum” uses 140. These levels differ by 55 units. Do not switch them after seeing results. The diagram specifies no suitable stop and proves neither variant better; it tests what the recorded plan actually describes.

Assumptions: entirely fictitious prices in hypothetical units. OHLC:(100,105,68,70),(90,140,58,60),(80,85,48,50). Previously fixed mean body 12, long-body threshold 24, rising context before candle 1. We test the article's custom variant, not TA-Lib's flag.

Separating recognition from a trading decision

An immediate entry after recognition, a precisely defined retracement order and no trade are different decisions. A retracement may never occur; exclude neither those cases nor failed fills from a comparison. A higher short-entry reference changes planned distances, but does not prove better net results. Any level rule should state how to draw support and resistance before later prices are visible.

Specify whether the stop reference means the first High, the last High or the maximum of all three. They need not be equal, as the diagram demonstrates. A target, size and cost model also need fixed rules; a larger reward-to-risk distance alone is not a win probability. Adding an oscillator such as RSI defines a new filter to test, rather than independent proof of strength. An ordinary stop does not guarantee its trigger price.

Do not assume an entry at the final close merely because the pattern uses that close. The default TradingView strategy emulator normally distinguishes calculation after a bar closes from execution on the next available tick; other settings can change the simulation. Record calculation timing, order submission, Bid/Ask conventions and fill assumptions. Candle geometry, order availability and realized profit are separate measurements.

A daily and an hourly occurrence use different aggregation windows; neither interval is automatically more reliable. Likewise, an engulfing pattern and an evening star have separate geometric conditions. A longer final body alone identifies neither exhaustion nor institutional short covering.

What to do tomorrow

  1. Fix one detector version. Record body and shadow thresholds, equality handling and the preceding-context rule.
  2. Check complete OHLC. Compare body sums with the net price change and locate the maximum of all three Highs, rather than assuming it belongs to the final candle.
  3. Separate recognition from execution. State the entry, expiry, exit and cost rules before testing. Include missed retracements and rejected or unfilled orders.
  4. Use the full selected sample. Record every qualifying occurrence and assess a separate evaluation period. A week of demo practice or a few selected charts does not establish an edge.

To organise your candlestick knowledge within a broader course, a good starting point is the technical analysis section in our library.