Breadth · Rate of change

McClellan Oscillator

Two exponential averages of the daily net of advances over declines, subtracted from one another. It is MACD applied to breadth, and it answers a different question from the advance/decline line built on the same numbers.

The calculation

  1. The daily net: advancing issues minus declining issues, one number per session, per exchange. This is the only input.
  2. Two exponential averages of that net: 19 periods and 39 periods, which correspond to the 10 and 5 per cent smoothing constants the McClellans specified.
  3. Subtract: oscillator = EMA(19) of net − EMA(39) of net. The result oscillates around zero.

Anyone who has read the MACD page will recognise the construction exactly: a fast average minus a slow one. The difference is the input. MACD is the difference of two averages of price; this is the difference of two averages of participation, and it inherits every property of that shape, including the fact that the two averages settle into rising at a similar rate during a long trend, which flattens the oscillator while breadth is still healthy.

An index narrowing, falling, and the oscillator at capitulationThe upper panel shows an index rising for about twenty-one sessions to a high, falling sharply for ten and then recovering. The lower panel shows the McClellan oscillator, which drifts down towards zero through the later part of the advance even while the index makes new highs, falls deeply negative during the decline, reaches its most negative reading near the low, and turns up sharply as the index recovers.INDEXMCC OSC3.37capitulationAn index narrowing, falling, and the oscillator at capitulationThe upper panel shows an index rising for about twenty-one sessions to a high, falling sharply for ten and then recovering. The lower panel shows the McClellan oscillator, which drifts down towards zero through the later part of the advance even while the index makes new highs, falls deeply negative during the decline, reaches its most negative reading near the low, and turns up sharply as the index recovers.INDEXMCC OSC3.37capitulation
Fig. 1: schematicComputed at build time from the daily net in this page's source, with 9/19 in place of 19/39 so the shape fits forty sessions. Two readings are worth taking. Through the last third of the advance the oscillator drifts towards zero while the index still makes new highs. The daily net is shrinking, which is the narrowing every breadth measure describes. Then the extreme: -8.11 on session 29, the most negative reading of the whole series, arriving at the low rather than at the top. That asymmetry is the most durable thing in this data.

Three views of one number

The daily net of advances over declines supports three separate measures, and mistaking one for another is the most common error in this section. They are not independent evidence; they share their only input.

What each does with the daily net
MeasureOperationAnswers
Advance/decline lineRunning total, no windowWhich way participation has been going, over months. Arbitrary level; slope is everything.
McClellan oscillatorFast EMA − slow EMAHow fast participation is changing now. Oscillates around zero.
McClellan Summation IndexRunning total of the oscillatorThe long view restored, with the daily noise already smoothed out of it.

The summation index is worth a moment, because it is the reason the oscillator is not usually read alone. Taking the difference of two averages discards the long horizon; accumulating the result puts it back, with the smoothing already applied. On this page’s series the summation index ends at -87.7, a number whose level means nothing whatever, exactly as on the A/D line, and whose slope is the point.

The list size problem

The oscillator scales with the number of issues on the exchange. A net of 300 on a list of 1,500 is a very different market from a net of 300 on a list of 3,500, and the American exchanges have changed size substantially over the decades, through waves of listings, mergers, delistings and the growth of funds and preferred issues that are not operating companies.

This is why the thresholds that circulate (±100 significant, ±150 rare) should be treated as folklore until checked against the data in front of you. The ratio-adjustedversion divides the daily net by the number of issues traded before the averaging, which removes the dependence and is what the McClellans themselves moved to for long histories. If a reading is being compared with one from the 1970s, that is the only version where the comparison means anything.

Where it misleads

Known failure modes
SituationWhat goes wrong
Treated as independent of the A/D lineBoth are built from the same daily net. Two views of one input are not two pieces of evidence.
Inherited thresholds±100 and ±150 depend on how many issues the exchange lists. Use the ratio-adjusted form or your own percentiles.
Long steady trendThe two averages settle into a similar rate of rise and the oscillator flattens while breadth is still fine.
Non-operating issuesFunds and preferreds move together on rate news; the fast average records it within days as a breadth event.
Zero crossings tradedFrequent, and their record depends almost entirely on whether the period trended.
Read for topsThe extremes cluster at lows. Distribution is quiet, and a rate-of-change measure cannot describe it.

What volume adds

Every measure on this page counts issues and ignores the size behind them. A session in which the net was mildly positive while almost all the volume went into the declining minority reads as a modestly good day, and it was not one.

That gap is what TRIN exists to fill: it divides the advance/decline ratio by the up/down volume ratio, so the count is normalised by the money behind it. Reading the two together is one of the few genuinely independent pairs in this section, because one of them has an input the other does not.

Frequently asked questions

How is the McClellan oscillator calculated?

Take the daily net of advancing issues minus declining issues. Compute a 19-day exponential average of that net and a 39-day one, then subtract the slow average from the fast one. Sherman and Marian McClellan specified smoothing constants of 10 and 5 per cent, which correspond to those two periods. The result oscillates around zero and is, structurally, MACD applied to breadth data rather than to price.

What does it add over the advance/decline line?

A different question from the same input. The A/D line accumulates the daily net without limit, so it describes the direction of participation over months. The oscillator measures the difference between two averages of that net, so it describes how fast participation is changing right now. One is a trend, the other a rate of change, and because they share their only input, they are not independent confirmation of each other.

What does zero mean?

That the fast and slow averages of net advances are equal; breadth is neither accelerating nor decelerating. Crossings of zero are the most quoted signal and are read as a change in the short-term direction of breadth. They are frequent, and like every crossing on this site their usefulness depends heavily on whether the market was trending or ranging, which the oscillator cannot tell you.

What readings count as extreme?

The figures usually quoted are ±100 as significant and beyond ±150 as rare, on American exchange data. Treat them as folklore until checked: they were derived when the exchange listed a different number of issues, and the oscillator scales with the size of the list. The defensible version is a percentile from the data you actually have, or the ratio-adjusted variant below, which divides the net by issues traded and is comparable across decades.

What is the ratio-adjusted version?

The same calculation with net advances divided by the total number of issues traded before the averaging, so the output does not grow simply because the exchange listed more stocks. The McClellans themselves moved to this form for long histories. If you are comparing a reading with one from the 1970s, the ratio-adjusted series is the only comparison that means anything.

What is the McClellan Summation Index?

A running total of the oscillator, read the way the A/D line is read: the level is arbitrary, the slope and the divergences carry the information. It restores the long-horizon view that taking the difference of two averages removed, which is why the pair is usually plotted together, oscillator for the rate of change, summation index for the trend.

Why do its extremes cluster at market lows?

The same asymmetry that runs through every breadth measure on this site. Forced selling is synchronised — margin calls and redemptions sell regardless of merit — so the daily net collapses across a few sessions and the fast average drops far below the slow one. Buying is deliberate and spread out, and rarely produces the mirror image. The most negative readings of a whole decline therefore usually arrive near its end rather than at its start.

Which exchange data should feed it?

Prefer common stock only where you can get it. An exchange list containing several hundred closed-end funds and preferred issues will move a large number of them together on interest-rate news, and the oscillator will report a breadth event no equity investor experienced. This distortion is larger here than on a slow measure, because the fast average reacts within days.

Is it a leading indicator?

It is faster than the A/D line and that is not the same as leading. Being the difference of two averages, it turns when the pace of breadth changes, which can happen well before or well after price does. Where it has a documented record is at capitulation extremes; as a general early-warning system it produces the same long series of premature signals as every other breadth measure read that way.

Where to go next