What Does a Volume Spike Mean? It Sizes the Next Move, Not Its Direction

Scope: measured on the ETFs listed (SPY, QQQ, IWM, DIA, GLD, SLV). Findings apply to the instruments measured and do not automatically transfer to other markets.

After a top-quartile volume hour, SPY's next hour ranged 0.429% against 0.225% after a bottom-quartile hour — 1.9× wider. Across six ETFs that multiple runs 1.4× to 2.8×. But the next hour continued in the same direction only 47–56% of the time in every one of the twenty-four ETF-by-quartile buckets, counting only hours where neither closed flat. High volume forecasts movement; it is silent on the sign. If you have ever bought the direction of a volume surge and been whipsawed, that asymmetry is the explanation: the surge said the next hour would be big, not which way.

Does high volume predict the next hour's range?#

Yes, and the effect is orderly. Rank each SPY hour's volume against other hours at the same clock time, then measure the hour that follows. The quietest quartile is followed by a median range of 0.225%; the busiest by 0.429%. That is 1.9× more travel available in the next sixty minutes, on roughly 480 hours per bucket.

SPY, 2025-02-10 to 2026-08-28, one-minute bars aggregated to complete regular-session hours 10:00–14:59 ET, so every hour measured has a following hour inside the same session. Each hour's volume is ranked into quartiles within the same hour of the day, so a 14:00 hour is compared only with other 14:00 hours — this removes the time-of-day pattern in volume rather than letting it masquerade as a volume effect. Next-hour range = (next hour's high − low) ÷ next hour's open, reported as the median. Continuation = the next hour closed in the same direction as this hour, counted only when neither hour closed flat (476–488 directional pairs per cell).

Note the shape. Q1 to Q3 climb gently — 0.225% to 0.260%, a difference too small to plan around. Almost the whole effect lives in the jump to Q4. It is not that busier hours are gradually followed by wider ones; it is that the genuinely heavy hour is followed by something distinctly wider.

Does high volume predict price direction?#

No. In the same SPY hours, the next hour closed in the same direction 50.2%, 49.6%, 54.0% and 52.7% of the time across the four quartiles. Widen it to all six ETFs and every one of the twenty-four cells lands between 47.3% and 55.7% — on 476–488 directional pairs each. No quartile in the set escapes the coin-flip band.

The highest continuation rate in the whole table is not in a busy bucket at all: it is IWM's quietest quartile at 55.7%. The lowest is GLD's busiest at 47.3% — a hair on the reversal side. Those two extremes sit in opposite volume buckets, which is what a table of noise looks like. Read the range column and the direction column together and the asymmetry is the finding: one column moves by a factor of 1.4 to 2.8 depending on the ETF, the other refuses to leave 50%. On 476–488 pairs a true coin flip lands between roughly 45.5% and 54.5% nineteen times in twenty; across twenty-four cells one reading just outside that band is expected, and 55.7% is that one.

Why does volume answer one question and not the other?#

Because the volume bar counts shares, not which side was in a hurry. Every share has a buyer and a seller, and the bar does not record who crossed the spread; that information exists, as signed order flow, but it is not in the bar and this note did not measure it. A heavy hour means more people showed up and more disagreement got priced — the bar alone does not say they showed up on one side. Volume clusters because attention clusters: information arrives, and it is traded on in both directions.

So what a spike leaves behind, in the six ETFs measured here (single stocks were not tested), is a more crowded, more contested market, and crowded contested markets swing wider. That is exactly the volatility half of the result. The direction half is empty for the same reason it is empty in the tape itself — the crowd that arrived was not one-sided, so its arrival carries no sign.

This note did not test breakouts, meaning volume conditional on price clearing a level; it tested volume by itself. What it can say about the familiar experience of buying a volume surge and getting whipsawed is narrower: the wider hour that follows a heavy one makes a wrong directional call more expensive, not less, precisely because the volatility half is real.

Does the pattern hold outside SPY?#

In all six ETFs, without exception on the ratio. The busiest quartile is followed by a wider median range than the quietest in every one, from 1.4× in DIA to 2.8× in SLV. Both the equity funds and the metals behave the same way, so the effect is not a quirk of one instrument.

Same window, same construction, 481–488 hours per cell. Q4 ÷ Q1 median next-hour range: SPY 1.9×, QQQ 1.8×, IWM 1.9×, DIA 1.4×, GLD 1.6×, SLV 2.8×. The middle two quartiles are not always ordered — DIA's Q2 median (0.231%) sits slightly below its Q1 (0.249%), and QQQ's Q2 (0.305%) barely clears its Q1 (0.299%). The separation that survives everywhere is Q4 against the rest.

SLV is the loudest version of the effect and also the one where it matters least in relative terms: a quiet silver hour is already followed by 0.407%, wider than a busy SPY hour. The multiple tells you how much a spike changes things within an instrument, not how the instruments compare with each other.

When do you actually know an hour was a spike?#

Only after it is over. The quartile is computed on the completed hour's volume, so the earliest moment this reading exists is the top of the next hour — the same hour whose range is being described. Nothing here is available while the spike is forming, and nothing here is a trade.

What it is, is a base rate for the size of what comes next. If the last hour was a heavy one for that time of day, the next hour has historically been roughly twice as wide as it would have been after a quiet one. That is a stop-distance and position-size input. It is not an entry, because the direction column says the next hour is a coin flip regardless. Placing an hour in a quartile also needs that clock hour's own history over a long window, eighteen months here, not the bar's height against the last few.

What this does not say#

A wider range is not profit. 0.429% is travel available in both directions to somebody with perfect timing. It says nothing about which way price went, or what survived spread and slippage.

It does not say the direction is exactly 50%. It says that on 476–488 pairs per cell, no quartile in any of the six separated itself from a coin flip. A real effect smaller than this sample can resolve would still be too small to pay for acting on.

Only the next hour was measured. Five minutes, a session, or a day could behave differently; nothing here extends past the one-hour lookahead.

The quartile split and the hour grid are this note's definitions, not standard ones. Other thresholds, other session windows, and the excluded opening half-hour and closing minute would move the exact figures.

This is descriptive. One eighteen-month window, six instruments, regular session only — what happened, not a forecast.

Related reading#

• How much does SPY move in a day? — the daily-scale version of the same range question, and why one number cannot answer it.
• Best time of day to buy ETFs — the time-of-day pattern this note deliberately ranks away.
• Do SPY and QQQ move together? — why a six-instrument table is closer to two or three independent readings.
• Before you believe a reversal story — another indicator that describes size and gets read as direction.
• Does price snap back to VWAP? — a second volume-derived reading, tested the same way.