Breadth · Comparability
McClellan Oscillator Across Eras
A reading of plus ninety means one thing on a list of 1,600 issues and something else entirely on a list of 3,400. The raw oscillator scales with the size of the exchange, which makes every borrowed threshold era-specific whether it says so or not.
The question behind the request
People arriving at a page like this usually want one of two things: the current reading, or a long history to compare it against. This is a static reference, so it carries neither, and the more useful answer is that the second request contains a problem most published breadth history quietly ignores.
The oscillator is built from net advances, a count of issues. The number of issues available to be counted has changed by a factor of two or more across the period people want to compare. So the same market behaviour produces systematically larger readings in a later era, and a threshold learned from an earlier one is not conservative or aggressive; it is measuring on a different scale.
The ratio between the two extremes is not a coincidence or an artefact of the smoothing: it is the ratio of the issue counts, because the calculation is linear in its input. Multiply the number of issues by anything and the raw series is multiplied by the same amount.
What the ratio adjustment does
Divide net advances by issues traded before smoothing, and the input becomes a proportion. The two panels above then collapse into one line, because they always described the same market.
ratio-adjusted net = (advances − declines) ÷ issues traded
The McClellans recommended this form themselves, precisely because the exchange list was growing under the measure they had published. It is the only version in which a multi-decade comparison means anything, and it is the less commonly published one, which is how thresholds from the 1970s ended up being quoted against exchanges twice the size.
One consequence is worth being explicit about: the ratio-adjusted series has a different numeric range from the raw one, small decimal values, usually rescaled by a constant such as a thousand for readability. So the familiar levels do not carry over to it either. There is no way to keep the old numbers and gain comparability; the numbers were the problem.
What a long breadth history can and cannot support
| Use | Raw form | Ratio-adjusted form |
|---|---|---|
| Shape within one period, narrowing or broadening | Sound. The scaling is constant inside a period, so it cancels. | Sound, and directly comparable with other periods. |
| Comparing an extreme with one from another decade | Invalid. The two readings are on different scales. | Valid, subject to the composition caveat below. |
| Applying a published threshold | Only to the era and exchange it came from. | Recompute as a percentile of your own window, always. |
| The Summation Index level | Meaningless across eras: the scaling error accumulates. | Direction and turning points, not the level. |
The part the adjustment cannot fix
Correcting for the number of issues leaves a second problem untouched, and it is the reason to be modest about even a properly adjusted long history: the contents of the list changed as well as its length.
The share of exchange-listed issues that are not operating companies has grown substantially, closed-end funds, preferred shares, exchange-traded products and multiple share classes of the same company. Those issues move together on interest-rate news, as the NYSE page sets out in more detail. A breadth count that includes them is answering a different question from one that does not, and the proportion has drifted continuously.
So the honest position on a multi-decade breadth series is narrower than it looks. Ratio-adjust it, read its shape rather than its levels, and express any extreme as a percentile of a defined window. A statement like "this reading is in the most extreme two per cent of the last ten years" survives all of the above. "This reading is above 150" does not.
Getting the underlying data
Advance, decline and unchanged counts come from the exchanges, and the historical series are generally licensed rather than free, which is the practical reason this reference computes its figures from published free sources such as FINRA’s daily short-volume files and illustrates everything else with clearly labelled synthetic series.
If you are assembling your own history, record three things alongside each session: the issues traded, the definition used for the count, and whether the series is common-stock-only. Without those, a breadth archive cannot be ratio-adjusted later, and an unadjusted archive is a set of numbers on a moving scale. That is a mundane data-hygiene point and it is the difference between a series you can still use in ten years and one you cannot.
Frequently asked questions
Does this page publish the historical McClellan series?
No, and the reason is worth stating rather than apologising for. A static reference page cannot carry a series that changes every session without becoming quietly wrong, and the underlying breadth data is licensed from the exchanges rather than freely republishable. What this page does instead is answer the question that usually sits behind the request: whether a reading from an earlier period can be compared with one from today. The short answer is that the raw form cannot.
Why are old raw readings not comparable?
Because the oscillator is computed from net advances — a count of issues — and the number of issues on the exchange has changed enormously. A list of about 1,600 issues cannot produce the same counts as a list of twice that size, even on a session where exactly the same proportion of issues advanced. So a raw reading grows with the size of the list, and a threshold drawn from one era systematically understates extremes in a later one. Nothing about the market needs to change for the number to change.
What is the ratio-adjusted form?
The same calculation with net advances divided by the number of issues traded before the smoothing is applied, so the input is a proportion rather than a count. Two sessions with the same breadth then produce the same reading whatever the size of the list. The McClellans themselves recommended this adjustment for exactly this reason, and it is the only form in which a long history is meaningful. It is also less widely published, which is why so many quoted thresholds are era-specific without saying so.
How large is the distortion?
Proportional to the change in the number of issues, which is the useful way to think about it: double the list and you roughly double the raw readings for identical breadth. The figure on this page computes both series from one set of breadth percentages at two list sizes and reports the ratio between their extremes. It is not a subtle effect at the edges; it is a scaling of the whole series.
So where do the familiar thresholds come from?
From particular exchanges in particular decades, mostly the NYSE in the years when the oscillator became popular. The levels that circulate (a reading beyond plus or minus one hundred as significant, beyond one hundred and fifty as extreme) were reasonable descriptions of that list at that time. Applied to a larger list they are too low, and applied to a much smaller one they are too high. Recompute them as percentiles of the series you actually have.
What else changed besides the issue count?
The composition, which matters as much and is harder to correct. The proportion of exchange-listed issues that are not operating companies (closed-end funds, preferred shares, exchange-traded products, multiple share classes) has grown substantially, and those issues move together on interest-rate news. A breadth count including them is measuring something different from one that does not, so even a correctly ratio-adjusted long history is comparing lists with different contents.
Is the Summation Index affected the same way?
More so, because it accumulates. The Summation Index is a running total of oscillator readings, so any scaling error in the input compounds into the level rather than staying proportional, which makes its absolute level across decades close to meaningless and its direction and turning points the only part worth reading. Anyone comparing a Summation level from one era with another is comparing two different measurement scales.
What can a historical breadth series legitimately be used for?
Two things, provided it is ratio-adjusted. Its shape: whether participation was narrowing or broadening ahead of a given episode, which is a within-period comparison and unaffected by the scaling. And its percentiles within a defined window: a reading in the most extreme two per cent of the last ten years is a statement that carries across eras in a way that a fixed number does not. What it cannot support is a fixed threshold quoted from a book.