Reference
Market Breadth: How Many Took Part
An index tells you what the largest constituents did. Breadth tells you how many issues took part, and the gap between those two statements is the most durable warning in market analysis.
Every measure in this section counts issues instead of weighting them. That single choice is what makes them a complement to an index rather than another version of it: in a capitalisation-weighted average, a few very large companies can carry the number while the majority of the list declines, and no amount of study of the index itself will reveal it.
The three groups below are the same data at three speeds. Advance/decline figures react daily and are the noisiest. The 52-week extremes are slow, because a stock has to travel a long way to set one, and are correspondingly steadier. Volume-weighted measures put back the one thing counting discards, the size behind each participant, and are the natural bridge to the rest of this site.
Three ways breadth relates to the index
Whatever the measure, the reading is comparative, breadth on its own means very little, and almost every use of it falls into one of three cases.
Confirmation
The index rises and participation expands with it. This is the ordinary state of a healthy advance and it is unremarkable, which is precisely why it is worth naming: most of the time breadth tells you nothing you did not already know from the price, and an indicator that is usually silent is being used correctly when it is silent.
Divergence
The index makes a higher high while participation shrinks. Fewer issues are carrying each successive push, which means the average has become dependent on its heaviest members. This is the reading breadth exists for and the reason the data survived from the 1920s into every modern internals screen. It is a condition rather than a signal: divergences have run for many months, and some have resolved upward.
Capitulation
The mirror image, and it behaves differently. Where tops are built slowly by narrowing, lows tend to arrive suddenly, an advance/decline ratio under 0.2, net new highs deeply negative, selling that has stopped discriminating between what is worth holding and what is not. Indiscriminate selling is characteristic of forced liquidation, which is why the most extreme readings in a whole decline usually appear at its end rather than at its start.
The same two numbers, four ways
Almost everything in this section begins with one pair of counts: how many issues closed higher, and how many closed lower. What separates the measures is what is done with that pair afterwards, and the differences are worth holding in mind because a threshold quoted for one is meaningless applied to another.
| Measure | Operation | Reads as |
|---|---|---|
| A/D ratio | Advances ÷ declines | One session. Resets daily; unbounded above, floored at zero. |
| A/D line | Running total of advances − declines | A trend. Arbitrary origin, so only the slope means anything. |
| McClellan oscillator | Two exponential averages of the daily net, subtracted | The rate of change of breadth. Oscillates around zero. |
| TRIN | The A/D ratio divided by the up/down volume ratio | Whether the volume matched the count. Above one, declines carried more volume than their number implied. |
Because they share their only input, they are not independent confirmation of each other. Two of them agreeing is one observation seen twice. The genuinely independent readings in this section come from the 52-week extremes, which count a different thing, and from the volume-weighted measures, which add an input the counts do not contain.
Reading breadth honestly
Three habits do most of the work, and none of them requires a subscription or a faster feed.
Date the divergence. Write down the session on which participation stopped expanding. It converts an impression into a fact that can be checked later, including against the many occasions when nothing followed, and it is the only way to notice that you have been reading the same warning for four months.
Compare like with like. An exchange-wide count against a large-capitalisation index is a comparison of two different lists of companies, and the mismatch alone will manufacture apparent divergences. If the breadth data covers a list, compare it with an index built from that list.
Distrust thresholds you did not derive. The numbers quoted in the literature (a ratio under 0.2, a McClellan reading beyond ±100) were drawn from a particular exchange with a particular number of listed issues, decades ago in most cases. Recompute the percentile from the data you actually have; the exercise takes an afternoon and it replaces inherited folklore with something you can defend.
Advance/decline data
The daily count of what rose against what fell, read three ways: as a session reading, as a running total, and as a smoothed oscillator.
- Advance/decline ratioAdvances divided by declines, one session at a time, and why the exchange’s own composition distorts it.
- Advance/decline lineThe running total. Its level carries no information at all; its slope against the index does.
- McClellan oscillatorThe daily net through two exponential averages.
- Advancing and declining issuesThe five fields an exchange publishes, which issues distort them, and how two numbers become four measures.
52-week extremes
A slower measure: how many issues are at the edges of their own year-long range. Harder to distort than a daily count, and more informative in the case people overlook.
- New highs and new lowsThe counts, the high-low index, and why both being large at once is the interesting reading.
- High-low range chartsThe extremes plotted as a range rather than as two counts.
- Hindenburg OmenThe signal built on both counts being high, sound observation, weak record.
Volume-weighted breadth
Counting issues ignores the size behind them. These measures put the volume back in.
- TRIN (Arms Index)Breadth divided by the volume behind it, in one number.
- Up volume and down volumeThe volume traded in advancing issues against declining ones, the one breadth measure that is not a census.
Frequently asked questions
What does breadth tell me that the index does not?
How many things moved, rather than how much the largest things moved. A capitalisation-weighted index can rise on the strength of a handful of its biggest members while most of the list falls; the index cannot show you that, and breadth measures exist for no other reason. When the two disagree, the disagreement is the information.
Are breadth divergences reliable?
Reliable as a description, unreliable as a trigger. Narrowing participation has preceded most major declines, and it has also persisted for many months without one, sometimes resolving upward instead. Treat a divergence as a change in the character of an advance, which is what it is, rather than as a countdown.
Why do the readings behave differently at tops and bottoms?
Because selling is more synchronised than buying. Fear moves the whole list at once, so the extreme readings — an advance/decline ratio under 0.2, a collapse in net new highs — cluster at lows rather than at highs. Tops are built slowly by narrowing, bottoms arrive suddenly by capitulation, and the same indicator therefore looks quite different at each end.
Which exchange should the data come from?
Prefer a common-stock-only count where you can get it. The NYSE list contains a large number of closed-end funds, preferred shares and other rate-sensitive vehicles that are not operating companies and move together on bond news, which pushes every breadth reading around for reasons unrelated to equity participation. It is also why historical thresholds drawn from earlier decades describe a differently composed exchange.
Can an equal-weighted index be used instead of breadth data?
It is a good substitute and not the same thing. An equal-weighted version of an index gives every member the same influence, so comparing it with the capitalisation-weighted version shows whether the largest members are doing the work, the same question breadth asks, answered in returns rather than in counts. What it cannot show is how the participation was distributed: a small number of large moves among the smaller members produces a rising equal-weighted index without any broadening of the list.
How many issues does a market need before breadth is meaningful?
Enough that a count is a statistic rather than an anecdote, a few hundred at least for a market-level reading, and at minimum several dozen for a sector. Below that the numbers move on individual company news, and a ratio built from twelve advances and eight declines carries no information about anything. This is the main reason sector-level breadth is more useful on large sectors than on small ones, and it is also why exchange-level data has survived while most narrower variants have not.
Do these measures work outside the American market?
The arithmetic is universal and the calibration is not. Every threshold quoted in the literature was drawn from American exchange data, on lists of a particular size and composition, and none of it transfers unexamined to a market with two hundred listed companies or one dominated by a handful of very large issues. The concepts hold everywhere; the numbers have to be rebuilt from the market you are looking at.
Is breadth useful for timing an entry?
Rarely, and it is the wrong question to bring to this section. Breadth describes the condition of a market (broad or narrow, participating or narrowing) over weeks and months, and the extreme readings that come closest to being actionable appear at capitulation lows, which is exactly when acting is hardest. Used as a filter it earns its place: it can tell you that an advance you were about to trust is being carried by fewer and fewer names. Used as a trigger it produces a long series of early exits.