Evening star candlestick: how to spot a three-day top

An evening star marks the moment a rally runs out of road. A tall green candle carries the advance to a new high, the next session opens even higher but produces only a small body, and the third session is a decisive red candle that gives back a large part of the first day's gain. Three days, three moods: confidence, hesitation, retreat.

In the Japanese vocabulary popularized in the West by Steve Nison, the evening star is Venus at dusk, the last bright point before darkness. It is the top-of-the-market twin of the Morning star candlestick pattern: rules, examples and test results, built from the same three ingredients turned upside down.

Below you will find the anatomy, TA-Lib's exact conditions, one configuration detail that keeps this pattern out of Finabase's own signals, two real cases from 2025, and a measurement of 154 evening stars on large US stocks and ETFs. Spoiler: they were roughly a coin flip.

Anatomy of an evening star

Look for these three sessions at the end of an advance:

1. Day one: a long green body. Buyers push the close far above the open. Nothing yet looks wrong; if anything, the rally seems to be accelerating.
2. Day two: the star. The session opens above day one's close and its body stays above day one's body, but it is small. It can be green or red. Price went up at the open and then could not make progress.
3. Day three: a long red body. Sellers drive the close deep into day one's green body. Many traditional sources want the close below the midpoint of that body; the deeper it goes, the more of the rally's last push has been erased.

Two variations have their own pages. If the star is a doji, it is an Evening doji star pattern: a doji at the top of a rally. If the doji is separated from both neighbors by gaps that even the shadows don't bridge, it is the bearish Abandoned baby candlestick: the rare double-gap reversal.

TA-Lib's test, condition by condition

Finabase's pattern detection uses TA-Lib. For CDLEVENINGSTAR, the function returns −100 on the third candle when every line of this table is true. The averages are TA-Lib's default candle settings, each taken over the 10 candles before the one being judged.

Compared with the textbook version, three simplifications stand out:

• The rally is assumed, not measured. TA-Lib's own source notes that the pattern matters in an uptrend "while this function does not consider the trend". In our sample 58.6% of detections followed a 10-session rise; the other 41.4% did not.
• The third day needs only a 30% retracement. Nison's preference is a close beyond the first body's midpoint, a stricter 50%.
• No gap is needed after the star. A gap down into day three makes the formation more convincing for most authors, but TA-Lib ignores it.

The penetration setting that hides it

That last condition contains a tuning knob, optInPenetration, which TA-Lib reads as a fraction of the first candle's body. Finabase's candlestick configuration sets it to 10 for this pattern, a value copied from the legacy Stock Master app and preserved so both platforms compute identical results. The Morning star candlestick pattern: rules, examples and test results page walks through the arithmetic; in brief, 10 asks day three to close ten first-body lengths below day one's close, roughly a 50% one-day collapse for a typical setup.

In 29,760 daily bars that never happened, so Finabase's markers, scanner and gauge will not show an evening star. The real examples and statistics on this page re-run the same function at TA-Lib's default of 0.3, which is also what the chart link at the top of the page does.

Reading the three days

What does each candle say about the people trading?

Day one belongs to the bulls. Late buyers chase the move, early buyers sit on profits, and the close near the high encourages more of the same.

Day two opens higher still, which should be good news. Instead the session goes sideways. The buyers who drove the gap cannot extend it, and sellers are willing to meet them at that level. The small body is a stalemate at the top of a rally, and stalemates after a strong run are where reversals often begin.

Day three is the verdict. Price falls through the star and well into day one's range. Everyone who bought during days one and two is now losing money. Some will sell to limit the loss, which is how a single bad day can feed on itself.

None of this guarantees a downtrend. A rally can pause for three days and resume, and in a rising market that is the more common outcome, as the data below suggests.

What counts as confirmation

Most traders want proof that day three was not a one-off:

• A close below the third candle's low within the next session or two. This is the confirmation used in our test.
• A gap down on the following day that is not filled.
• Heavy volume on the red third candle compared with recent sessions.
• A level that matters: the star forms at prior resistance or a round-number high.
• Momentum disagreeing with price: for example, the Relative Strength Index (RSI): how to read and use it making a lower high while price made a higher one (a bearish divergence).

Evening stars on real charts

GE Aerospace, March 26, 2025 (worked). GE had climbed 6.3% in ten sessions. On March 24 it rose from $206.92 to $210.23. On March 25 it opened above that close and finished at $212.13, a modest body above the first. On March 26 it closed at $207.37, well into the March 24 body. The next session closed at $205.88, below the pattern's low, confirming it. Within five sessions the stock was 3.7% lower and after ten it was down 9.7%, as the broad market slid in early April 2025.

AMD, July 18, 2025 (failed). AMD had rallied 13.8% in ten sessions when it printed a textbook-looking evening star: a strong green day on July 16, a small red star above it on July 17, and a red close at $156.99 on July 18. It never confirmed. The next session closed flat, above the pattern's low, and five sessions later the stock was 6.0% higher; after ten, 9.4% higher. Strong uptrends routinely shrug off three-day pauses, and an unconfirmed star in a powerful advance is weak evidence.

What the data says

The study covers 60 large US stocks and ETFs, two years of daily bars (October 30, 2024 to October 29, 2026), 29,760 day-samples. For a bearish pattern, a "hit" is a lower close five sessions later. The baseline is the share of all days that were followed by a lower close.

The hit rate was almost exactly the baseline, and the average stock rose after an evening star, by 0.50% over five sessions and 0.70% over ten. Confirmation nudged the hit rate to an even 50% but left the average return positive. During a two-year stretch in which large US stocks mostly climbed, a bearish three-candle pattern was not enough to lean against the tide.

Patterns that resemble the evening star

An educational outline for trading it

Here is how traders commonly frame an evening star. It is an outline for learning, not a recommendation, and every number is illustrative.

1. Is there a rally worth reversing? A steady advance into a known resistance level gives the pattern meaning. A star in the middle of a range gives it little.
2. How deep did day three close? Past the midpoint of day one's body is the traditional standard.
3. Confirm or not. Some act at the third close; others wait for a close below the third candle's low.
4. Where is the idea wrong? Above the star's high. A close up there means buyers absorbed the selling.
5. Size from that distance, and know the exit before you enter.

How Finabase shows the evening star

• On charts. Finabase's candlestick list includes Evening Star (CDLEVENINGSTAR), but at its penetration of 10 the study marks nothing. The link at the top of this page passes a penetration of 0.3, which makes the chart flag evening stars at TA-Lib's default.
• The label trap. The chart marker labeled "Evening Star" belongs to a different study, TA-Lib's CDLGAPSIDESIDEWHITE (side-by-side white lines). The legacy app copied the wrong name onto it, and Finabase keeps the label for compatibility. If you see that marker, you are looking at a gap followed by two similar white candles, not at this pattern.
• Scanner. The candlestick scanner places detections on the Bullish or Bearish list by the sign of TA-Lib's output. The evening star always returns −100, so it belongs to the Bearish list, which stays empty for this pattern at Finabase's setting.
• Signals tab. On a symbol's Signals tab (AAPL, for example), the Candlestick gauge counts each pattern's latest detection in the last eight bars as a Buy or Sell vote. With nothing detected, the evening star always votes Neutral.

Frequently asked questions

Is the evening star a reliable sell signal?

Not on its own, according to our data. At TA-Lib's default, 47.4% of evening stars were followed by a lower close five sessions later, practically the 46.7% of any day, and the average move was up.

Does the star have to gap above the first candle?

Its body does. TA-Lib requires the whole star body to sit above the first candle's body, though the shadows may overlap. The gap into the third candle is optional in TA-Lib.

Which is stronger, an evening star or an evening doji star?

Traditional sources rank the doji version higher because a doji signals more complete indecision. In our sample the Evening doji star pattern: a doji at the top of a rally had a lower hit rate, 44.0% across 50 cases, so the ranking did not hold here.

Why does my Finabase chart show "Evening Star" on a pattern that looks different?

Because that marker is the side-by-side white lines detector carrying a legacy label. The real evening star detector, with Finabase's penetration of 10, does not fire.

Summary

The evening star records a rally losing its footing over three sessions: a strong up day, a stalled day above it, and a strong down day that cuts into the first. TA-Lib codifies it without checking the rally, and Finabase's inherited penetration setting means its own tools never flag it. At TA-Lib's default, 154 evening stars on large US stocks did no better than chance in a rising market. If you use it, use it as one input among several, at a level that matters, with confirmation and a defined exit.