Reference
Market Analysis, Volume in Context
Volume and breadth in historical context: what the crash record actually shows about participation, and what a measurement of published data can and cannot support. Conditions rather than forecasts, and dated rather than pretending to be live.
The indicator pages on this site explain what a measure is calculated from. This section asks a different question: when those measures were at their most extreme, what was happening, and did the reading mean what the literature says it means.
That is a harder question than it looks, because the events everyone cites are the ones that resolved dramatically. The 1929 and 1987 declines, the 2000 and 2008 markets, are famous precisely because something followed, and a library of examples selected after the outcome was known will support almost any claim. Every page here is written against that pull: where a divergence or a volume reading is described as a warning, the pages also say how often the same reading led nowhere.
What the record supports
Three statements survive contact with the historical data reasonably well, and each is weaker than its popular version.
Advances narrow before they end, usually, and not usefully soon.Participation contracting while an index still rises has preceded most major declines. It has also run for many months without one. The honest form of the claim is about the character of an advance, not its remaining duration.
Forced selling is visible and indiscriminate. When holders sell because they must rather than because they want to, they sell without regard to merit, and every breadth and volume measure records it: advance/decline ratios collapse, new lows expand, TRIN spikes. This is the most reliable pattern in the whole field, and it describes lows rather than highs.
The heaviest volume in a decline tends to arrive near its end. A consequence of the previous point rather than an independent finding, and worth stating separately because it contradicts the intuition that a top should be the busiest moment. Tops are quiet; that is what makes them hard.
What it does not support
The clustering visible in the figure above is worth one further paragraph, because it is the single most useful thing the crash record has to say and it is rarely put plainly. Extreme sessions do not arrive alone. They arrive inside a few weeks of one another, in the middle of a decline that is already under way, which means the reading that identifies them is retrospective by construction: by the time a market is producing sessions like these, the thing they are supposed to warn about has already happened.
The practical implication runs the other way from how such charts are usually presented. A list of worst days is not a list of warnings; it is a description of what forced liquidation looks like once it has begun. The warning, if there is one anywhere, sits months earlier in the breadth data, quiet, unremarkable, and impossible to date.
Anything with a date in it. No measure in this reference has a demonstrated record of saying when a condition will resolve, and the studies that appear to show one almost always identify their examples after the fact, choosing the divergence that preceded a decline and passing over the dozen that preceded nothing.
Nor do the widely quoted thresholds transfer. The numbers that circulate (an advance/decline ratio below a certain level, a TRIN above two, a percentage of issues at new lows) were drawn from particular exchanges in particular decades, on lists of a particular size and composition. They are worth knowing as folklore and worth recomputing from the data you actually have before anyone acts on them.
The practical position that follows is narrower than most market writing but easier to defend. Use these measures to describe the market you are in, note the date on which a condition appeared, and check later what followed, including the many occasions when the answer is nothing at all.
The historical record
What actually happened, with the volume and breadth data that accompanied it. These are the pages where a claim can be checked against something rather than argued about.
- Stock market crashesThe major declines, what preceded each one, and what the volume looked like on the way down.
- Crash reference, dates and depthsThe same events as a reference table rather than a narrative.
- Bear marketsThe longer declines that were not crashes, and why the distinction matters, duration rather than depth.
Volume studies
Measurements rather than commentary, computed from published data and dated so they can be rerun.
- Short volume as a share of all tradingReal FINRA data: why half of all volume being short-marked says nothing about sentiment.
- Advance/decline volumeThe one breadth measure that is not a census, weighted by the stock that changed hands.
- Negative Volume IndexFosback’s index, which moves only when volume falls, and what its premise assumed about who trades when.
- Volume at market turnsWhat the sessions around major lows and highs had in common, and why the asymmetry between them is mechanical.
Method, and what a claim has to clear
Two pages about how a result is produced rather than about the market itself. Both compute their own counter-example, because the argument is easier to accept when the figure makes it.
- Monthly performance and seasonalityTwelve calendar bins produce a best and a worst month even when there is nothing there, computed on forty synthetic years to show how large the spread gets.
- Intraday trading and its arithmeticA round trip costs the same at every horizon while the move shrinks with the square root of time. The sum that comes before any indicator.
Reading a market condition
The measures that describe the state of a whole market rather than one instrument.
- Market breadthHow many issues took part, the most durable of the conditions described here.
- Volume spread analysisThe same question asked of a single bar rather than a whole list.
- Indicator libraryWhere each measure’s formula and failure modes are set out.
Frequently asked questions
What is the difference between a condition and a signal?
A condition is a description of the market as it is, participation narrowing, volume expanding, breadth diverging. A signal claims to tell you what to do about it. Everything in this section is the first kind, because the second kind requires evidence that almost none of the published claims in technical analysis actually carry. The distinction is not pedantry: a condition can be true for months and correct the whole time, whereas a signal that is early is simply wrong.
Why do crashes get a section of their own?
Because they are the only episodes where the volume and breadth data are unambiguous, which makes them the best available test of what those measures mean. Everything this site claims about participation — that a narrowing advance is fragile, that forced selling is indiscriminate and visible, that the heaviest volume in a decline tends to arrive near its end — can be checked against the record of specific dated events rather than asserted.
Does narrowing breadth predict a crash?
No, and the record is the reason to say so plainly. Narrowing participation has preceded most major declines, and it has also persisted for many months without one, sometimes resolving upward instead. Both halves of that sentence come from the same data. What breadth data supports is a statement about how an advance is being made; a forecast of when it ends is an additional claim that the data does not carry.
Is volume higher at tops or at bottoms?
At bottoms, consistently, and the asymmetry has a mundane explanation. Selling under pressure is synchronised, margin calls and redemptions force it regardless of price, so it concentrates enormous volume into a few sessions. Distribution is deliberate and spread over weeks precisely because a large seller cannot afford to be noticed. That is why the most extreme readings in every measure on this site cluster at lows, and why tops are identified by narrowing rather than by any spike.
How far back is the data reliable?
Prices and index levels go back a long way and are broadly dependable; volume and breadth counts need more care. The number of listed issues has changed substantially, the composition of the exchange lists has changed with the growth of funds and preferred issues, and the conventions for counting have not been constant. Comparisons across several decades of raw counts are not comparing like with like, which is the argument for the bounded, percentage-based forms of every breadth measure whenever a long history is involved.
Why are the studies here dated rather than live?
Because this is a static reference and an honest snapshot is better than a stale feed. Each study states the range of sessions it covers and names the public source it was built from, so the work can be repeated over any window. A page presenting a fixed number as though it updates is the kind of quiet decay that makes a reference untrustworthy.
Do you publish forecasts or market commentary?
No. A market note written years ago has no value to a reader arriving now, and a reference full of expired opinions is worse than one without any, so this section carries none. What is here is either a description of a measure or a measurement of published data, both of which remain true whatever the market does next.
What would change your mind about any of this?
For the breadth claims, a properly specified test on a long sample showing that divergences carry no information about subsequent drawdowns beyond what price alone provides. For the volume claims, evidence that the asymmetry between tops and bottoms disappears once forced selling is accounted for. Both are answerable questions, and stating them is more useful than another confident paragraph, a claim that nothing could refute is not analysis.