Market breadth

The index went up. Did the market?

Those are different questions, and a capitalisation-weighted index cannot answer the second. Breadth is the family of measures that counts participants instead of watching price — and when the two disagree, the disagreement is the signal.

Why an index is not the market#

In a cap-weighted index, the largest constituents dominate the level. When a handful of very large companies rise while most others fall, the index still goes up. Nothing is wrong with the index — it is doing exactly what it was designed to do — but "the market rose" is not a fair description of that day.

Breadth measures ask a different question: how many things went up, not how much weight went up.

import { calculateNetAdvances } from "fintech-algorithms/market-breadth-and-internals/advance-decline-breadth/net-advances";

Start with counting, then fix the counting#

Net advances — advances minus declines — is the simplest measure and the input to most others.

It has one flaw that took the field decades to fix: the number of listed issues changes. A net-advance reading of +800 meant something different in 1970 than it does today, because the universe has grown. Raw counts are not comparable across eras.

The fix is to make it scale-free:

• Advance/decline ratio — advances divided by declines
• Normalized A/D line — net advances divided by issues traded, then accumulated

Everything downstream inherits this problem, which is why the McClellan family exists in both traditional and ratio-adjusted forms.

The cumulative line, and what it is for#

Running net advances into a cumulative total gives the advance/decline line. Its level is arbitrary — it depends entirely on where you started the accumulation — so the level carries no information at all.

The shape does. Specifically, divergence: when the index makes a new high and the A/D line does not, fewer and fewer names are participating in the advance. That has preceded most significant tops, and it is visible weeks before price confirms it.

One caution the reference page enforces: a gap in a cumulative series propagates forward forever. Miss three sessions and every subsequent value is wrong by that amount. The routine returns the sessions it could not find rather than silently bridging them, and it applies a knowledge cutoff so revisions arriving later do not quietly rebuild history into a line nobody traded.

The McClellan family#

The McClellan oscillator is the difference between a 19-day and a 39-day exponential average of net advances. Fast, noisy, and useful for spotting short-term extremes.

The summation index accumulates the oscillator, turning a fast signal into a slow regime gauge. The oscillator says what is happening this week; the summation index says what has been happening for months.

Both come in traditional and ratio-adjusted forms, and the choice matters: only the ratio-adjusted versions are comparable across decades. Historical threshold values quoted for the traditional oscillator were derived on a much smaller universe of listed issues.

There are volume-weighted variants too, and when the issue-based and volume-based readings disagree, that disagreement is itself informative — many small advancers versus a few heavily-traded ones.

Participation: how many are actually in an uptrend?#

Advance/decline measures one session. These measure position over a long lookback:

• New highs and new lows — how many issues reached 52-week extremes. A rally where fewer and fewer names make new highs is thinning.
• Percent above the 200-day average — a market with 80% of names above their long-term average is a different market from one with 35%, even at the same index level.

The 20-, 50- and 200-day variants read different horizons: short-term turnover, intermediate trend health, and long-term participation.

A detail worth noting in the reference: these return equal_count separately. Securities sitting exactly on their moving average are neither above nor below, and folding them either way biases the reading.

Pressure and exhaustion#

Some breadth measures capture force rather than direction.

TRIN (the Arms index) divides the advance/decline issue ratio by the advance/decline volume ratio. Above 1 means declining issues are absorbing proportionally more volume than their numbers suggest. It is a comparison of two ratios, so inverting either one flips the reading — which is why the reference page exposes both components.

The Zweig breadth thrust is the rare one: the advance ratio moving from below 0.40 to above 0.615 within ten sessions. It fires a handful of times in a generation, and has historically marked the start of major advances. That rarity is the point — and also why it cannot be validated on a short sample.

Upside/downside volume ratio at extremes — a 9-to-1 up day — is a recognised initiation signal for the same reason: it is rare.

Concentration: the most topical question in markets#

"Seven stocks are holding up the entire market." How would you prove it?

import { calculate } from "fintech-algorithms/market-breadth-and-internals/concentration-and-diffusion/top-n-index-contribution";

Top-N contribution answers it directly: how much of the index's move came from its largest N members. It is a claim that is usually asserted and rarely measured.

Two companions make it rigorous:

• Herfindahl concentration — the sum of squared weights. Squaring is what makes it a concentration measure: it is dominated by the largest holdings in a way a simple count never is. Note it is reported both as a fraction and in points, which differ by a factor of 10,000 and are frequently confused.
• Effective number of constituents — the reciprocal of HHI. An index of 500 names with an effective number of 60 is, for risk purposes, a 60-stock portfolio. The gap between the actual count and the effective count is the entire message.

Diffusion asks the same question across groups rather than constituents: what share of sectors, or of factors, is improving? A move concentrated in one sector and a move spread across all eleven look identical at index level.

Reading breadth honestly#

Three cautions, all of which the reference pages encode.

Ready flags, not zeros. Rows carry a ready flag, and a row that is not ready is excluded rather than treated as zero. A missing count and a count of zero mean opposite things about participation.

The universe is the argument. Two systems disagreeing on breadth almost always disagree about which securities were in the universe, not about the arithmetic. That is why every reference page here returns the universe id and revision alongside the number.

Point-in-time or nothing. Breadth data is revised routinely. A cumulative line silently rebuilt from revised inputs is not the line anyone traded, so the cumulative topics take a knowledge cutoff and ignore anything that arrived after it.

A dashboard, in order#

If you are assembling a breadth view, this sequence builds up:

1. Net advances and the A/D ratio — the daily reading
2. The cumulative A/D line — the shape, watched for divergence
3. McClellan oscillator and RASI — fast signal and slow regime
4. Percent above 50- and 200-day — participation across horizons
5. TRIN and upside/downside volume — pressure
6. Top-N contribution and effective N — how narrow is this?

Steps 1–3 tell you what breadth is doing. Steps 4–6 tell you whether to believe the index.

Where this connects#

Breadth is computed over a universe, and the universe has to be reconstructed point-in-time — which is corporate actions territory. It is computed from prices that need adjustment, which is the same domain. And the concentration measures are the same arithmetic an index provider uses when capping weights.

The reference pages state each contract exactly.

The reference pages for this domain#

Market Breadth and Internals — 28 algorithms across 5 families, each with a parameter contract and an executed example.