A Violent 15-Minute Candle Barely Predicts the Close

We pulled nearly four years of 15-minute S&P 500 bars and measured what follows a sharp move.

A single 15-minute candle that leaps 30 or 50 points feels like information. Something just happened, and the instinct is that it should keep happening, or snap straight back. So we tested it the plain way: take every 15-minute bar in the S&P 500 since mid-2022 where the index displaced a set amount in that one bar, then measure where it went from the end of that bar to the 4 PM close. The answer is that the close barely becomes more predictable. There is a real bias, but it is small, and most of what looks like an edge is just the market's ordinary drift higher.

We measured it in percent rather than points

A 50-point move is not the same event across time. At an index near 7,400 it is about 0.67%. Back when the S&P was near 3,900 the same 50 points was 1.3%, almost twice the move. To compare fairly across the whole sample we converted each point threshold into the percentage it represents at recent levels, then applied that percentage to every bar in history. The four cutoffs line up like this.

Today's point moves expressed as a percentage of the index, then applied uniformly across the sample. A percentage threshold is scale-free, so it treats 2022 and 2026 the same.

Move in one 15-min bar	As a percent
50 points	0.67%
40 points	0.54%
30 points	0.40%
20 points	0.27%

Up-moves drift on, down-moves stabilize

Here are the core results.

After an UP bar of the given size: forward return to the 4 PM close. Samples shown as triggers (distinct days), since one day can produce several qualifying bars.

Up move in one bar	Samples (days)	Avg to close	Median	Closed higher
0.67% (~50 pt)	54 (35)	+0.74%	+0.32%	65%
0.54% (~40 pt)	100 (70)	+0.35%	+0.21%	60%
0.40% (~30 pt)	229 (151)	+0.28%	+0.17%	57%
0.27% (~20 pt)	732 (360)	+0.12%	+0.09%	55%

After a DOWN bar of the given size: forward return to the 4 PM close. "Closed higher" above 50% means the index tended to recover off the spike low.

Down move in one bar	Samples (days)	Avg to close	Median	Closed higher
0.67% (~50 pt)	51 (33)	+0.21%	+0.01%	51%
0.54% (~40 pt)	127 (88)	+0.19%	+0.09%	54%
0.40% (~30 pt)	305 (191)	+0.13%	+0.10%	55%
0.27% (~20 pt)	821 (401)	+0.04%	+0.04%	53%

Two patterns appear consistently across every threshold:

• After an up-bar, the index tends to grind higher into the close, with stronger follow-through after larger bars. A 0.67% up-move closed higher 65% of the time and averaged an additional 0.74% by the bell.
• After a down-bar, the outcome is closer to a coin flip that leans slightly positive. The index continued lower less than half the time and more often flattened or recovered modestly. Sharp drops in this sample did not tend to produce further downside, instead they stabilized.

The drift is tiny next to the day's range

Before anyone builds a system on this, look at the size of the edge. The biggest average in the whole study is 0.74%, after the rarest event (a 0.67% up-bar, which happened on just 35 days in nearly four years).

At the 20-point cutoff, where the sample swells past a thousand bars, the average forward move is 0.08% and the median 0.07%. That is a rounding error on a typical S&P session.

How much of this is just the market going up

The positive results in both tables have a quieter explanation: the S&P spent most of the 2022 to 2026 period grinding higher, so random afternoon moves carried a positive bias regardless of what happened in the prior 15 minutes. That structural drift accounts for much of the follow-through after up-bars and is also why down-bars still closed higher rather than flat.

The cleaner, regime-independent finding is the difference between the two: an up-bar was consistently followed by a stronger close than a down-bar of the same size, across every threshold.