Distribution Days: Reading Institutional Selling Before Price Breaks
The most dangerous market is the one that looks healthy. Indexes at highs, trend intact — while underneath, the biggest holders are quietly handing their shares to the crowd. Distribution days are how you count that happening.
The definition
A distribution day is a session where a major index falls at least 0.2% on higher volume than the previous day. The falling price says sellers won the day; the rising volume says the selling was big — institutional-sized. One is noise. A cluster is a message. The classic O'Neil rule of thumb: four or more distribution days within about 25 sessions means institutions are distributing — selling into strength — and the rally is on borrowed time.
Why it fires before price breaks
Large funds cannot exit a position in a day without crushing the price, so they sell gradually, into rallies, while the index still looks fine. The price damage comes later; the volume signature is visible immediately. That is what makes distribution days the closest thing to a leading indicator in a toolkit otherwise built from lagging averages: they detect the behavior that precedes the decline.
Why these exact thresholds
Each piece of the definition earns its place. The 0.2% decline floor keeps flat, drifting sessions out of the count — a day that closed a few cents lower on heavy volume is churn, not distribution. The higher-volume requirement is the heart of it: price falling on lighter volume than yesterday means sellers were timid; falling on heavier volume means someone large needed out badly enough to accept worse prices. And the 25-session window matches how institutions actually operate — a fund trimming a major position works the order over roughly a month, so that's the span over which the footprints cluster before the decline they foreshadow.
The count also heals, in two ways, and the second is the one people forget. The first is age: each new session pushes the oldest day out of the window, so a distribution day expires after about five weeks whatever price does. The second is price. O'Neil's rule drops a distribution day as soon as the index closes 5% above that day's close, because a rally of that size means the selling has already been absorbed. A day the market has eaten is not supply hanging over the tape any more, and leaving it on the tally charges the market twice for the same shares.
That second rule is not a technicality. On 26 September 2026 our own gauge showed QQQ carrying 5 distribution days while the index sat 0.6% from its 52-week high — a contradiction a reader spotted immediately. It was real: four of the five had been cleared by rallies of 5.4% to 5.8%, and the honest count was 1. The gauge was counting selling the market had digested weeks earlier. SPY genuinely carried 8, because it had not risen 5% from any of them. Both numbers now apply the price expiry.
A live worked example: August 2026
In late August 2026, SPY sat above both its 50-day and 200-day moving averages with both rising — by trend measures, a perfect tape. Yet SPY was carrying 5 distribution days in its last 25 sessions, and QQQ 6. And the damage was already measurable where it matters: DataQuant's own breakout win rate had fallen to roughly 37% in the July window, from 64% in May–June. The index held up. The tape under it did not — and breakout traders trade the tape, not the index.
Why DataQuant scores it separately — and watches two benchmarks
This is exactly why the Market Health gauge scores distribution days as their own component (0–25 points, minus five per distribution day) instead of folding them into trend. A trend-only gauge would have read "all clear" through July while breakouts failed. The gauge also counts distribution on both SPY and QQQ and takes the worse of the two — growth stocks, where breakouts live, often come under distribution first, and in August 2026 QQQ was indeed the weaker tape.
How to use the count
- 0–2 days: normal give-and-take. No signal.
- 3 days: attention. Tighten stops on extended positions.
- 4–6 days: institutions are selling into strength. Size down, demand only the best setups, and expect breakouts to fail more often — whatever the index chart looks like.
Distribution days age out, and they also clear on a 5% rally, so a cluster that never becomes a correction simply disappears from the count.
What our own data says about them
We joined the count to 98,854 graded breakouts from 1985 to date, applying both expiries. The uncomfortable result: the distribution count on its own says very little about how a breakout goes. Across the bands the profit factor runs 1.98, 1.72, 1.75, 1.97, 1.67 — no direction, just noise. An earlier version of this study appeared to show heavy distribution preceding better breakouts, and that turned out to be an artifact of the missing 5% expiry rather than a real inversion.
What does separate is not the count at all but where the index sits relative to its own recent high:
- More than 5% below its 25-session high: profit factor 1.52. The market is genuinely broken and breakouts fail with it.
- 2 to 5% below: 2.23 — the best band by a wide margin, and it holds in every decade (2.94, 1.85, 2.45, 2.18).
- Half a percent to 2% below: 1.87.
- At the high: 1.70, the weakest. Breaking out into a market that has just run is the hardest case.
So the honest reading is narrower than the classic rule. A pile of distribution days is worth knowing about as a description of the tape, and O'Neil's threshold is a reasonable prompt to size down. But if you want the one market-context number that has actually separated breakout outcomes over forty years, it is the index's distance from its own recent high, and the sweet spot is a market that has pulled back a little without breaking.
Frequently asked questions
What exactly counts as a distribution day?
A session where the index closes down at least 0.2% on higher volume than the prior session, counted over a rolling 25-session window. Four or more is the classic warning threshold.
Can the market look strong while under distribution?
Yes — that is the whole point. In August 2026 SPY was above both its key moving averages with 5 distribution days on the count, and breakout win rates had already halved. Institutions sell into strength precisely because price still looks fine.
Why watch QQQ as well as SPY?
Growth stocks — where most breakout setups live — often come under distribution before the broad market. DataQuant's gauge counts both and takes the worse reading.
Do distribution days predict a crash?
No. They flag elevated risk, not a timetable. Clusters precede corrections more often than chance, but some clusters simply expire as the window rolls forward. Use the count for position sizing, not prophecy.