Breadth · Data
Inside an Advance/Decline Count
Four published measures come out of two numbers, and the two numbers contain more than operating companies. This is the anatomy of a breadth reading: what is counted, what is quietly dropped, and which version to trust.
The five numbers an exchange publishes
A session’s breadth data is short enough to list in full, and each field carries an assumption worth knowing.
| Field | What it counts | The catch |
|---|---|---|
| Issues traded | Every security that changed hands on the venue | Includes funds, preferreds, products and multiple share classes, not a count of companies. |
| Advances | Closed above the previous close | A one-cent gain counts exactly as much as a twenty per cent one. |
| Declines | Closed below the previous close | Same, in the other direction. Magnitude is invisible to the measure. |
| Unchanged | Closed exactly level, or did not trade | Routinely dropped from published ratios, which changes the denominator. |
| Up and down volume | Shares traded in advancing and in declining issues | A whole session’s volume is assigned to the direction of the close, by convention. |
The second and third rows contain the property that defines this whole family of measures: a breadth count is a census, not a weighting. Every issue gets one vote regardless of its size or how far it moved, which is exactly what makes it useful next to a capitalisation-weighted index, and exactly why it can never confirm one.
Two measures, one set of counts
The figure below computes the two most-quoted breadth measures from a single synthetic series of sessions. Both are derived from the same advances and declines; the difference is that one accumulates and one does not, and that alone makes them disagree about what is happening.
What each of the four measures is for
The advance/decline line. A running total of net advances. Read its direction and its divergences against a price index; ignore its level entirely, since that depends only on when someone started adding up. Its strength is that a narrowing advance shows up here before it shows up in an index.
The advance/decline ratio. Advances divided by declines, session by session. No memory, so it responds immediately and says nothing about trend. Its published thresholds are the most era-specific numbers in this field, because the ratio depends on how many issues are on the list and what kind of issues they are.
Up and down volume. The same subtraction performed on share volume rather than on issue counts, which changes the question from "how many took part" to "how much stock moved in each direction". It is the closest breadth gets to weighting, and it inherits one convention worth remembering: the entire session’s volume in an issue is assigned to the direction of its close, so a stock that spent the day falling and closed a cent up contributes all of its volume to the up side.
The McClellan oscillator. The difference between two exponential averages of net advances, which converts the series into something with a zero line and a range. It is the most processed of the four and the most sensitive to the issue-count problem, since it is linear in the counts, the comparability page computes how much.
The unchanged column, and why it matters
On the synthetic series above, an average of 193 issues per session, about 6.0 per cent, closed unchanged. Real exchanges report figures of the same order, and the proportion rises among thinly traded issues, because an issue that did not trade at all is unchanged by convention.
Dropping that column has two consequences. Any ratio computed as advances over declines is describing only the issues that moved, so its denominator is not the exchange. And a session where trading was thin can look decisive: if a third of the list did not move at all, a modest number of advances against a small number of declines produces a ratio that reads as overwhelming. Publish the unchanged count alongside, or at least know what it was.
Common stock only, wherever you can get it
The single most valuable adjustment to any breadth series is restricting it to operating companies. It removes two distinct distortions at once.
The rate-sensitive block. Preferred shares and closed-end bond funds move together on interest-rate news, so on a quiet equity day with a bond-market move, a breadth count can read as one-sided while the equity market did almost nothing. The NYSE page covers why this venue is affected most.
Double counting. A company with two share classes contributes two votes that move identically, which quietly weights the census towards companies with complicated capital structures, the opposite of what a one-issue-one-vote measure is supposed to achieve.
Neither is fatal, and neither is visible in the number. That is the argument for stating which version of a breadth series you are quoting, every time: two people reading correct figures from different definitions will disagree about the session and never find out why.
Frequently asked questions
What exactly does an advance/decline count include?
Every security that traded on the exchange and closed above or below its previous close, which is a wider population than most readers assume. Alongside operating companies it contains closed-end funds, preferred shares, exchange-traded products, warrants, units, and multiple share classes of the same company. Those are all issues, they are all counted, and several of them move for reasons that have nothing to do with the equity market.
What happens to unchanged issues?
They are counted separately and then usually ignored, which is a small but real distortion. On a typical session a few per cent of issues close exactly where they closed the day before, and the proportion rises sharply among thinly traded ones, a stock that did not trade at all is unchanged by convention. Since advances plus declines plus unchanged equals issues traded, dropping the unchanged column means the denominator you are dividing by is not the number of issues on the exchange.
Which issues distort the count most?
The interest-rate-sensitive block: preferred shares and closed-end bond funds, which move together on rate news and can produce a decisive-looking breadth reading on a day when equities did very little. Multiple share classes are the other one, one company counted twice or three times, moving identically. A common-stock-only count removes both problems and is the version worth using wherever your data provider offers it.
How does one set of counts become four measures?
By choosing what to do with the same two numbers. Subtract them and accumulate, and you have the advance/decline line. Divide them, and you have the advance/decline ratio. Do the subtraction on volume rather than on issue counts, and you have up/down volume. Smooth the net figure twice and take the difference of the averages, and you have the McClellan oscillator. All four are the same session data, and they can point different ways on the same day because they answer different questions.
Why is the level of an advance/decline line meaningless?
Because it depends entirely on when the accumulation started. A running total of net advances beginning in 1990 and one beginning in 2010 will sit at completely different levels while having identical shape, so the number itself carries no information, only its direction, and its divergences against a price index. This is the same property that makes a cumulative volume line’s level uninformative, and it is the most common misreading of both.
Should the count be adjusted for the number of issues?
For any comparison spanning more than a few years, yes. The number of issues on an exchange has changed enormously over the decades, and a raw count grows with it, so an extreme reading from an earlier era is on a different scale from a modern one. Dividing net advances by issues traded turns the count into a proportion and makes the comparison legitimate. The McClellan history page works through the size of that distortion.
Does the advance/decline line lead the market?
It has diverged from price ahead of most major declines, and it has also diverged for many months without one. Both statements come from the same record. The defensible version of the claim is that the line describes how an advance is being made — whether many issues or a few are carrying it — and that a narrowing advance is more fragile than a broad one. Turning that into a dated forecast requires evidence the series does not provide.
Where does the data come from?
The exchanges publish daily advance, decline and unchanged counts, and financial data providers redistribute them; historical series are generally licensed rather than free. If you are collecting your own, record the issues traded and the exact definition alongside each session, without those two fields the archive cannot be ratio-adjusted later, and an unadjusted count is a number on a moving scale.