Volume · Trade size

Volume by Trade Size, Reconsidered

The method assumed a large participant leaves a large mark on the tape. Execution algorithms exist precisely to prevent that, so the inference the method was built on no longer runs, and a narrower one still does.

What the method assumed

Trade-size analysis classifies each print by the number of shares in it and tracks the classes separately. The reasoning was straightforward and, for its time, sound: a pension fund buying a large position had to transact in large pieces, and an individual buying a hundred shares left a hundred-share print. Two populations, two distinguishable footprints.

Read that way, the small-block series was retail activity and the block series was institutional, and a divergence between them was informative, a claim that recurs throughout the older literature on trade-size analysis, in several forms and with varying degrees of care.

One order, four ways of executing itA bar chart of the number of prints left on the tape by a single 500,000-share order executed four ways: as one block, sliced into five-thousand-share pieces, into five-hundred-share pieces, and into hundred-share pieces. The same order produces between one and five thousand marks depending only on the execution method.prints left on the tapeOne block1 printThe pre-slicing assumption: a large participant leaves one large mark on thetape.Sliced into 5,000s100 printsA modest algorithm. Every print is a size a retail order could plausibly be.Sliced into 500s1,000 printsOrdinary practice for a liquid instrument, and none of the prints looksinstitutional.Sliced into 100s5,000 printsThe same order, now indistinguishable from five thousand small participants.One order, four ways of executing itA bar chart of the number of prints left on the tape by a single 500,000-share order executed four ways: as one block, sliced into five-thousand-share pieces, into five-hundred-share pieces, and into hundred-share pieces. The same order produces between one and five thousand marks depending only on the execution method.prints left on the tapeOne block1 printThe pre-slicing assumption: a large participant leaves onelarge mark on the tape.Sliced into 5,000s100 printsA modest algorithm. Every print is a size a retail ordercould plausibly be.Sliced into 500s1,000 printsOrdinary practice for a liquid instrument, and none of theprints looks institutional.Sliced into 100s5,000 printsThe same order, now indistinguishable from five thousandsmall participants.
Fig. 1: arithmetic on one stated orderOne order of 500,000 shares, four execution methods, and nothing else different: the same participant, the same intention, the same quantity. As one block it leaves a single unmistakable institutional mark. Sliced into hundred-share pieces it leaves 5,000 prints, every one of which looks exactly like a small private trade. Execution algorithms exist to produce the bottom row rather than the top one, because a visible block invites the market to move against the rest of the position. The trade-size distribution the method reads is therefore a measure of execution practice at least as much as of who is trading.

What that leaves

The inference that has gone is the one from small prints to small participants, and with it the whole participant-separation claim. It cannot be repaired by adjusting the buckets, because the information it needed is deliberately not in the tape any more.

Two narrower readings survive intact, and both are worth having.

A genuine block still means what it says. A single large negotiated transaction, reported as one print, is direct evidence that someone transacted a large quantity at one price. What changed is the reverse inference: an absence of blocks now implies nothing at all.

A change in the distribution is a fact about the session. If the size profile of a security’s trading shifts sharply, something happened, but the most likely somethings are a change in execution practice, a large order being worked, or an index event, rather than a change in the mix of participants. It is a prompt to look, not a conclusion.

Why the change happened, and why it is permanent

Announcing a large order is expensive. A visible block tells everyone else that a large buyer or seller is present, and the remainder of the position is then executed against a market that has moved. Slicing is the response, and it is not a fashion. It follows directly from the cost of information leakage, so there is no reason to expect the old footprints to return.

The market mechanics page sets out the other half of the same story: a large share of retail flow is now executed by wholesalers who take the other side themselves, so the counterparty to a small print is frequently an intermediary rather than an individual. Both changes point the same way. The tape carries less information about who traded than it once did.

What to use instead

Questions this method used to answer, and what answers them now
QuestionThe old answerWhat works now
Was this session unusually busy?Block volume against its recent level.Relative volume against the instrument’s own baseline, no assumption about participants required.
Did a large transaction occur?A block print.Still a block print. This is the reading that survived unchanged.
Are institutions accumulating?Rising block volume against flat small-lot volume.Nothing on a chart answers this. Holdings disclosures answer it, with a delay.
Was the volume mechanical?Not addressed.The calendar, a rebalance, an expiry or a results date explains more sessions than any trade-size reading.

The third row is the one to be firm about. "Institutions are accumulating" is the claim trade-size analysis was mostly used to support, and it is now unsupportable from tape data, which is a better outcome than a method that keeps producing the answer after the evidence for it has gone.

The one figure worth watching, and how to read it

If a size-based series is kept at all, the useful form is the share of a session's volume in each bucket rather than the raw volume in each, for the reason every measure on this site is a ratio. Raw block volume rises with the market's overall activity, so a rising block series can mean nothing more than a busy week.

Read as shares, a sharp change is worth a look, and the list of likely causes is short and mostly mundane. A large order being worked will lift the small-print count for days. An index event concentrates an enormous quantity into the closing auction, which some data sets classify as one print and others do not. A change of execution venue or broker shifts the distribution with nothing happening in the market at all. Only when those are excluded does a distribution change say anything about participation.

That ordering (ratio rather than level, mechanical explanations before behavioural ones) is the same discipline the volume baseline page applies to total volume. It is unglamorous, it costs a minute, and it accounts for most of the sessions that would otherwise be filed as evidence of something.

Why the section stays

Because the measure is still computed, still shipped by charting packages, and still cited, so a reference that omitted it would leave a reader with the older literature and no correction. The histogram page sets out how it is built and what its buckets mean; this page states what has changed underneath it.

That is the pattern this reference follows wherever a measure has been overtaken by market structure, the breadth thresholds from an exchange half the size, the TICK levels from before automated quoting. In each case the arithmetic still works and the interpretation was calibrated on a market that no longer exists, and saying so is more useful than either repeating the old claim or deleting the page.

Frequently asked questions

What is volume by trade size?

A family of measures that classify each trade by how many shares it involved — small, medium or block — and then track the volume in each class separately. The purpose was to separate participants: the assumption was that large institutions traded in blocks while individuals traded in small lots, so the two streams could be watched independently.

Does that assumption still hold?

No, and this is the central fact about the method. Institutional orders are now routinely split into many small executions by algorithms, precisely so that they do not announce themselves. The figure on this page computes the consequence: one five-hundred-thousand-share order sliced into hundred-share prints leaves five thousand marks on the tape, none of which is distinguishable from a small participant’s trade.

So the method is dead?

The participant-identification claim is, and something narrower survives. The distribution of trade sizes is still a real property of a session, and a genuine block print, negotiated and reported as one trade, still tells you a single large transaction occurred. What can no longer be inferred is the reverse: an absence of blocks does not mean an absence of large participants, and small prints do not mean small participants.

What is a block trade?

A single large transaction, conventionally ten thousand shares or more, usually arranged away from the continuous market and then reported. Those still happen and they still mean what they always did, someone transacted a large quantity at one price. The change is in what their absence implies, which is now nothing.

Why did the market change this way?

Because announcing a large order is expensive. A visible block invites the market to move against the remainder of the position, so execution algorithms split orders across time and venues to reduce the price impact. The trade-size distribution people were reading was a side effect of a technology limitation, and the limitation went away.

Is there a modern equivalent?

Partly. Reported venue and condition codes distinguish some categories of execution, and the off-exchange share of a security’s volume is published, the short-volume study on this site uses those files. What none of it recovers is the clean participant split the older method claimed, because the information that split relied on is deliberately no longer in the tape.

What does the histogram version show?

The distribution of volume across size buckets for a period, which is a legitimate description of how the trading was structured and is worth looking at when it changes sharply. The indicator page covers its construction and its thresholds. The caution to carry from here is that a shift in the distribution is at least as likely to reflect a change in execution practice as a change in who is trading.

What should be used instead?

For the question "was this session unusual", relative volume against the instrument’s own baseline, which needs no assumption about participants at all. For "did a large transaction occur", genuine block prints, which still mean what they say. And for "was the volume real or mechanical", the calendar, a rebalance or an expiry explains more sessions than any trade-size reading will.