Morning Star and Evening Star: Do They Reverse? 240 Backtests

A morning star is three candles. The first falls, the second has a small body, the third rises and closes past the midpoint of the first body. The evening star inverts every part of that. In the Sakata tradition both belong to the santen or three-river family.

The name comes from the middle candle floating free of the two around it, the way a single star sits in the dawn sky. That picture requires a gap on each side of it.

So that was counted first. Across three timeframes and two years of USD/JPY, 4,558 signals fired and 0 of them had a gap on both sides.

Measured against the colour sequence it is built on, a down bar then anything then an up bar, the pattern came out ahead in one cell of six.

20252024dashed: the colour sequence on its own, that year

What the morning star and evening star actually test

For a morning star, Formiq's backtester requires three things at once:

1. The colour frame: the first bar falls and the third rises. The middle bar's colour is not checked.
2. A small middle: the middle body is under half the first body and under half the third body.
3. The recovery: the third close passes the midpoint of the first body.

The evening star inverts all of it.

No gap is tested. The condition reads open and close positions only; whether the middle bar is separated from the first is never asked. This article counts the gaps, then removes and moves the other clauses one at a time.

Is there really a gap?

Two definitions were counted. A body gap puts the middle body entirely below the first bar's close (for a morning star) and the third body entirely above the middle body. A full gap requires the same of the wicks, which is what leaves visible white space on a chart.

The full gap is zero in all six cells, 0 out of 4,558.

The reason is the trading day. Currency does not close overnight and reopen in the morning; it runs continuously from Monday to Friday, so each bar opens at almost exactly the previous close.

Some difference exists on 93 to 94% of bars, but a full pip of separation appears on 0.85 to 3.42%. The rest is under a pip. The largest break on all three charts is the same bar in each year: 204.05 pips on 5 October 2025 and 94.80 pips on 27 October 2024, both at the Sunday reopen.

Waiting for a star that is genuinely separated therefore leaves a handful of candidates a year, all at the weekly reopen. The rest of this article measures the gapless shape, which is the one that actually gets traded.

Settings and how to add a threshold

Nothing is drawn. Three candles that are already on the chart make the pattern, so no line and no lower pane appears.

Neither condition has numbers

Both thresholds are fixed in the code: the middle body at half, the recovery at the midpoint. This article moves them from outside instead.

The published trading rule

Enter at the close of the third bar, long on a morning star and short on an evening star. Each one only produces a single side, so a rule using just the morning star never sells. Switching both on lets the morning star close shorts and the evening star close longs.

How this was measured

Against the colour sequence alone

The pattern adds two clauses on top of "a down bar, then anything, then an up bar". Here is what that base does by itself.

Only the 15-minute chart in 2025 came out ahead, by 0.61 points. The other five are negative, and the 4-hour chart in 2024 is 11.10 points down.

A colour sequence landing near 50% is unremarkable: a down bar followed two bars later by an up bar happens more than 3,000 times a year on the hourly chart. What the table shows is that adding two clauses, which cuts the count to about a seventh, produced a lower number than the count it started from.

How far does the third bar have to recover?

The shipped rule puts the third close past the midpoint of the first body. Cutting that threshold at six levels, with the small-middle requirement left in place:

Hourly chart. Within a year the whole ladder spans 1.53 points in 2025 and 0.67 in 2024. The same threshold moves 4.12 to 5.56 points when the year changes. The shipped 50% level reads 50.00% and 45.60%.

Signals fall from 570 to 267 across the ladder. Cutting more of them did not raise the rate.

Does the middle bar's size or colour matter?

Removing the other clause, "the middle body is under half of each outer body":

The recovery-only column tracks the colour sequence almost exactly. The six differences run from -1.88 to +0.26 points, and on the hourly chart they are 0.25 and 0.26. That clause removes about 40% of the signals and leaves the rate where it was.

Keeping only the small-middle clause is slightly above the sequence on the 15-minute and hourly charts in 2025 and below it everywhere else. Combining the two, which is the shipped condition, is not better than either alone.

The middle bar's colour was counted too. It ran opposite to the third bar in 41.41 to 46.06% of signals.

The opposite-coloured middle leads on the 15-minute and hourly charts in all four cells and trails on the 4-hour in both. Four of six, but not consistent across timeframes.

Trading all 240 runs

Closing ten bars after entry. A single star produces one side only, so there is no opposite signal to close against.

Hourly chart. Two of the ten conditions cleared both years, both of them the morning star with a filter, and the ATR version took 33 and 31 trades.

The 15-minute chart cleared three (the morning star alone, with the body filter, and with the 50-MA direction) and the 4-hour cleared two (the morning star and the evening star, each with the body filter). No condition cleared both years on all three charts.

Subtracting the market's own drift

The morning star only buys and the evening star only sells. Buying repeatedly in a year that rose pays regardless of the pattern, so what an unconditional ten-bar hold would have returned is subtracted from the per-trade result.

USD/JPY rose through 2024, so on the 4-hour chart holding any ten bars returned 9.49 pips on average. The evening star's 2024 result carries that headwind in full.

The hourly evening star lost 9.58 pips a trade in 2024. Removing the 2.53 pips of drift still leaves -7.05 pips.

After the drift, four of the 30 timeframe-condition pairs kept a positive per-trade result in both years: the morning star with the body filter on the 15-minute and 4-hour charts, the morning star with the ATR filter on the hourly, and the evening star with the body filter on the 4-hour. All four take between 16 and 80 trades a year.

Last year's winner, held through this year

The 15-minute and hourly picks fell below the following year's median. The 4-hour pick stayed first, on 17 trades in 2025 and 15 in 2024.

Stops, targets and cost

A 60 / 30 exit lifts the evening star to a 61.67% win rate and still loses 1,038.4 pips in 2024. The entries are identical; the ratio of stop to target sets the win rate.

Spread paid matched trade count times 0.3 pips everywhere. The hourly morning star still returns -19.3 pips at zero spread, so the loss comes from the entry rather than from paying for it too often.

Related articles

• Candlestick pattern win rates: these two among 17, over the same two years and the same kind of control
• Is the engulfing candle profitable?: a two-bar reversal measured against the colour flips that did not engulf
• Double tops and double bottoms tested: a larger reversal shape over the same two years
• Do fair value gaps really fill?: the gap counted here, promoted into a trading condition
• Do Indicator Settings Transfer Between Pairs? 3,024 Tests: this setting won the 4-hour chart on USD/JPY in 2023 and is one of the few that also made money on EUR/USD, GBP/USD and EUR/JPY

Notes

The data is USD/JPY from 1 January 2024 to 31 December 2025. Fills are at the close, the spread is fixed at 0.3 pips, and the lot is 0.1. Detection uses the same code path as the backtest and reads bars from before the test window. Because the horizons run out to 20 bars, that many bars at the end of each year are excluded. The body gap is measured on opens and closes; the full gap on highs and lows. The drift subtraction averages the ten-bar close-to-close move over every bar of the window, then removes it from long-only results and adds it to short-only ones.