Stock Market Crashes, Volume, Breadth and the Historical Record

It warns of instability, not of a date. What recurs across these episodes is not a single spike but a change of character in the weeks beforehand: rallies made on lighter volume than the declines that interrupt them, and breadth narrowing while the index still rises. That combination has preceded major breaks often enough to be worth watching and has also appeared many times without one following. Anyone claiming a volume pattern that predicts crashes is selling the survivors of that record.

An enormous session — several times the recent average — where price falls hard and closes near its low, followed within days by another heavy session that fails to make a new low. The first is forced selling; the second is the test that finds no more of it. The pattern is legible after the fact and treacherous in real time, because the same first session occurs in the middle of declines that then continue much further.

Speed. The S&P 500 fell roughly 34 per cent in 33 days, the fastest decline of that magnitude on record, then recovered its level within about five months. The volume signature compressed accordingly: what took months to develop in 2008 happened inside three weeks. It is the clearest demonstration that these patterns describe a sequence of behaviour, not a fixed timetable.

Because it is the largest number available, and a headline needs one. The 89 per cent figure is real, but the market that produced it had margin rules that permitted leverage now prohibited, no circuit breakers, no obligation to halt, far less disclosure and a much narrower list of listed companies. Most of the mechanisms that made that decline possible were legislated away in direct response to it. The figure is accurate; the analogy is the part that needs care.

No, and the two are close to independent, which the depth and speed charts on this page show for the episodes themselves. Heavy volume says a great many holders needed to transact, which is a statement about the present rather than about the eventual extent. The deepest decline on this list developed over nearly three years with long quiet stretches inside it; the fastest was among the shallowest.

Use it to recognise a sequence, not to forecast one. Knowing that advances on thinning volume, narrowing breadth and heavy down sessions have preceded severe declines is worth something: it is a reason to reduce exposure and to check what participation is doing rather than to assume an advance is sound. Knowing how often that same configuration has resolved into nothing is the other half, and a page that gave you only the first half would be selling something.

Yes, and it is the clearest structural difference between the modern episodes and 1929 or 1987. Market-wide breakers introduced after 1987 halt trading at defined percentage declines, which interrupts the feedback loop in which falling prices force selling that pushes prices lower. They do not prevent a decline, March 2020 triggered them repeatedly and the market still fell about a third, but they change its shape, spreading what might have been one uninterrupted session across several. Any comparison of intraday behaviour across that boundary is comparing two different market designs.

The same sequence at very different speeds, which is the useful comparison. 2008 developed over months: repeated heavy down sessions separated by rallies on lighter volume, a pattern that repeated through the autumn and gave many opportunities to observe it. 2020 compressed the identical structure into about three weeks, the distribution phase, the acceleration and the exhaustion all occurred inside a month. Anyone whose rules were calibrated to the 2008 tempo was still waiting for confirmation when the 2020 low had passed.

No absolute one, and this is where volume analysis is most often oversold. What recurs is a relationship rather than a number: the heaviest session of the whole decline tends to arrive at or very near its end, and the informative event is the session after it, a second heavy session that fails to make a new low. Neither can be specified in advance as a share count, because what counts as heavy depends entirely on the instrument and the era.

Because the sequence is not about the cause. Margin debt, an oil shock, a valuation regime, a mortgage market and a public-health shutdown have no shared mechanism, but each ended with a large number of holders needing to reduce exposure through a market that could not absorb them simultaneously. That constraint is structural rather than causal, and it is why the volume signature is similar across episodes that share nothing else.

Weaker ones, and the asymmetry is real. Advances are typically made on volume that expands modestly and unevenly; declines expand it sharply and uniformly, because fear is more synchronised than optimism and because forced selling has no price sensitivity. This is why almost every volume reading in this reference has more to say about weakness than about strength, and why the extreme readings in breadth data cluster at lows rather than at highs.

There is no reliable answer, and any source that gives one is describing a small sample. In the episodes above the character change (advances on lighter volume, heavy sessions being down sessions, breadth narrowing) was visible for anywhere between a few weeks and well over a year beforehand. That range is the whole practical problem: the pattern identifies a condition, and the condition can persist far longer than a position can.

Very little, which is worth stating because "distribution on heavy volume" is a common phrase. Heavy volume at a high can be aggressive buying that continues, or supply being absorbed by the last buyers. What distinguishes them is not the volume on that day but what the following sessions do with it: an advance that cannot extend after its heaviest session is a different market from one that can. The single session is a question, not an answer.

In their specifics, no, and the historical record on this is unambiguous. Several of them were widely discussed in general terms for years beforehand (leverage in the late 1920s, valuations in 1999, mortgage credit from 2006) without that discussion helping anyone time the outcome. Some commentators were right about the mechanism and years early, which in practice is indistinguishable from being wrong. Recognition of a condition is achievable; timing is not.

Increasingly little, and 1987 is the clearest case. The decline that day was driven less by any reassessment of value than by portfolio-insurance programmes selling futures mechanically as prices fell, into a market whose ability to absorb them had disappeared. The 2020 episode had a comparable component. Describing these as changes of opinion misses what actually happened, which was a market structure failing to clear, and volume is the only series that shows it directly.

No, and conflating them makes the history look more dramatic than it was. A crash is a sudden violent decline over days or weeks; a bear market is a decline of twenty per cent or more, which can take years and often contains no crash at all. 1987 was a crash without a recession or a prolonged bear market. 2000–02 was a long bear market that included no single crash day comparable to October 1987.