Does the shooting star pattern work? 6,672 tested
The shooting star is taught as a bearish reversal. We found 6,672 of them in 758,032 sessions of S&P 500 history and measured what happened next.
Mean five-day return after the pattern was +0.36%, against +0.31% for the sample as a whole. Two-sided p = 0.065.
What that means
A shooting star is meant to mark exhaustion at the top of a move. The measured drift here is very slightly upward, which is the exact opposite of the claim. The effect is not big enough to trade either way.
Shooting star: what it looks like on a chart
Two real ones, drawn from the same sample as the table above. One is an instance the textbook called correctly and one is an instance it called wrong.
Neither is a showpiece. Within each group the instance shown is the one whose five-day return sits at the median, so these are the ordinary members rather than the dramatic ones, and the screen that picked them looks only at whether the chart is legible. Compare them at the ringed bar. That is the whole of what a 1.1 point difference in the odds looks like when you draw it.
The ringed bar is where the pattern completes and the call would be made. Everything right of the dashed line was hidden at that moment. Prices are normalised to 100 at the ringed bar, the way the game normalises its own charts.
The ringed bar is where the pattern completes and the call would be made. Everything right of the dashed line was hidden at that moment. Prices are normalised to 100 at the ringed bar, the way the game normalises its own charts.
How the pattern was defined
Long upper shadow at least twice the body, almost no lower shadow, closing above the 20-day average. The definition is mechanical, which matters: a pattern you can only recognise after the fact is not a pattern you can trade. Anything requiring judgement to spot would score differently for every reader, and could not be tested at all.
On this sample the shooting star did not separate from an ordinary bar. That is not proof it never works. It means that if there is an effect here, it is smaller than 6,672 instances can detect, which is a strong constraint on how much weight the pattern deserves on its own.
The half a backtest cannot answer
A backtest says what the market did. It cannot say what a reader does when the pattern appears, and that is where most of the damage happens. Read the Tape deals the same five blind charts to everyone, so we can see both at once: across 165 settled charts the textbook lean was right on 49.7% of them, and the crowd followed it on 77.0%.
How this was measured
Every pattern is scored the same way. Take 153 S&P 500 names, 2005-01-01 to 2025-01-01, which is 758,032 daily sessions. Mark every bar where the pattern completes, using only information available at that bar. Measure the close 5 trading sessions later, the same horizon the game itself uses. Compare against the base rate across all bars in the same sample, which is 54.7% higher after five sessions.
The comparison is a two-sided test of proportions and the threshold is the conventional 5%. "No measurable effect" means the pattern failed that bar, not that the effect is exactly zero. The code is one file and the definitions are written out in full on each page, so you can disagree with a threshold and see what it changes.
What this does not show
These are unconditional five-day results on large-cap US stocks over one twenty-year window. They say nothing about intraday trading, about other markets, or about a pattern used in combination with something else. A trader who uses the hammer as one input among several is not tested here. Neither is any pattern at a horizon other than five sessions, and a signal can be useless over a week and useful over a quarter.
These results are published under CC BY 4.0. Quote the figures or rebuild the table anywhere you like, with credit to Read the Tape and a link back to this page.
Call five blind charts yourself →