McClellan oscillator

Definition

The McClellan oscillator (Sherman and Marian McClellan, 1969) is the difference between a fast and a slow EMA of daily net advances (advancers − decliners) — MACD machinery applied to breadth instead of price. Positive readings mean breadth momentum is improving; extremes mark unusually one-sided participation. The summation index is its running total, a slower cumulative gauge.

How it works / structure

• Formula: oscillator = EMA(19) − EMA(39) of daily net advances (the McClellans’ conventions); ratio-adjusted versions divide net advances by total issues first — mandatory for cross-era comparability as listing counts changed; summation index = cumulative sum of the oscillator.
• Parameters (engine-executable): universe (pinned — NYSE composite vs S&P 500), ratio adjustment (pinned on), the two EMA windows, and the signal conventions — zero-line crosses, extreme thresholds (e.g. ±70 ratio-adjusted, treated as tunable), divergences (mechanically pinned), summation-index levels.
• Inheritance: every property of its parents — EMA lag and window sensitivity (indicator-ema), breadth’s universe- contamination and structural-drift issues (indicator-breadth-advance-decline).

When it applies

Breadth-momentum tracking within the market-regime toolkit (lens-market); washout/thrust extreme detection with the small-sample caveats of all named breadth events; divergence monitoring as fragility flags, not timing.

Risk profile & failure modes

• Double inheritance: EMA-family whipsaw plus breadth-data contamination compound; the oscillator is two transformations away from anything tradable.
• Extreme-threshold folklore: ±70/±100 boundaries are the originators’ conventions from mid-century NYSE data; the ratio adjustment helps comparability, the thresholds remain uncalibrated lore.
• Small-sample event signals: oscillator-based “breadth thrust” definitions have single-digit historical instance counts — unreliable by construction (lens-quantitative).

Evidence & limits

The McClellans’ exposition (1970) is the construction source. No peer-reviewed validation of oscillator-specific rules exists; it inherits the mixed breadth-timing evidence discussed in indicator-breadth-advance-decline and the technical-survey verdicts (Park-Irwin 2007). Descriptive use (participation momentum measurement) is sound arithmetic; every timing convention is a labeled hypothesis.

Falsifiable-thesis examples

Illustrations only, not signals:

• “Ratio-adjusted oscillator readings below −70 are followed by positive 10-session index returns on average across this replay decade” — falsified by the tally.
• “This mechanically-defined oscillator divergence precedes an index lower-high within 2 months” — falsified at the mark.

Cross-references

• Parent data: indicator-breadth-advance-decline; parent machinery: indicator-ema, indicator-macd
• Consuming framework: lens-market, regime-volatility
• Method caveats: lens-quantitative (small samples)