---
title: "Method: Market breadth"
url: https://tipsheet.markets/methods/breadth/
description: "How tipsheet.markets computes market breadth: sources, start dates, checks and caveats."
updated: 2026-11-05
publisher: tipsheet (https://tipsheet.markets/)
cite: "tipsheet, “Method: Market breadth”, 5 November 2026, https://tipsheet.markets/methods/breadth/"
---

[Methods](https://tipsheet.markets/methods/)

# Market breadth

**Code:** `pipeline/tipsheet/compute/breadth.py`, `publish/breadth.py`.

**Bundles:** `markets/breadth/*`.

## Stock breadth

- **Universe:** liquid NSE main-board stocks on each date. A stock qualifies when both are true:
  
  - Its median traded value over the last 60 sessions is at least ₹1 crore a day, in constant rupees. Values are deflated by CPI to the latest month, so the bar doesn’t fall in real terms over 30 years.
  - It traded on at least 45 of those 60 sessions.
  
  ETFs and SME stocks are excluded. Delisted stocks count while they traded, so there is no survivorship bias.
- **Bundles:** `markets/breadth/stocks_daily_2y` (the last two years), `markets/breadth/stocks_daily_full` (every session from the start rule below, 6 Oct 1999; 6,731 sessions to 29 Oct 2026; from 30 Oct 2026, for the front page’s ranking of a day’s breadth against its own history) and `markets/breadth/stocks_weekly_full` (the same rows, each week’s last session). All three come from one table, so they agree cell for cell on the dates they share (checked on 30 Oct 2026: no differing cell against either).
- **Prices** come from the adjusted panel (see `prices.md`).
- **Measures:**
  
  - share of stocks above their 20-, 50-, 100- and 200-day moving averages
  - share with the 50-day average above the 200-day average
  - 52-week highs and lows, and the net of the two
  - advances, declines and the advance/decline line
  - the McClellan oscillator and summation index
  - share of stocks 20% or more below their 52-week high
  - median drawdown from the 52-week high
  - median drawdown from the all-time high: the highest adjusted close since listing or since the price panel’s first date (3 November 1994). Same population as the 52-week measure (at least 252 bars of history), so the two compare like for like. Peaks before November 1994 are censored, so the early years of the series (2003–05) understate the fall for stocks that peaked in 1992 or earlier. Added 22 October 2026 at a reader’s request.
  - stocks at new all-time highs and lows, weekly (`breadth_ath_weekly.parquet`, bundle `markets/breadth/ath_atl_weekly_full`): distinct stocks whose adjusted high (low) beat every earlier adjusted high (low) on any day of the week, as a share of distinct liquid stocks in the week with at least 252 sessions of history. “All time” has the same November 1994 floor. This base dipped to about 70 stocks in 2001, so the series starts once it stays at 200 or more: 4 July 2003. Recent listings have shorter histories, so their records come more easily; the 252-session rule removes only the trivial first-year cases.
  - one-month cross-sectional dispersion
  - one-year participation (added 2026-10-14): the median stock’s one-year price change, the share of stocks higher than a year ago, and the share that beat the Nifty 500 over the year. A stock counts once it has 250 bars spanning no more than 275 sessions. Stock prices are adjusted for corporate actions but carry no dividends, so they are compared with the Nifty 500 **price** index (niftyindices history), 250 sessions back, not the total-return index. The stock set is the liquid universe on the day, looking back, so nothing in it depends on the future.
  - the middle half (25th to 75th percentile) of stocks’ one-year price change, for the band around the median (added 2026-10-22)
- **Start date: 1999-10-06.** This is the first date the 60-session median universe reaches 200 stocks. Before that, NSE-only breadth rests on too few liquid stocks, because NSE was young and BSE carried most trading.

## Breadth by company size

`breadth_size_weekly.parquet`, bundle `markets/breadth/size_weekly`. Weekly (each week’s last session) from January 2018, when AMFI’s half-yearly lists begin.

- **Groups:** AMFI market-value rank in force on the day: largest 100, next 150 (101–250), next 250 (251–500) and the rest. Each list is used from 15 days after its period end and never carried backward, so a company sits in the group it was in then. Ranking by today’s value would pick the stocks that have already risen, and they are more often above their averages.
- **Matching:** ISIN first, then NSE symbol; 0–2 liquid stocks a week go unmatched and are left out.
- **Universe and measure:** the same liquid universe and 200-bar rule as the headline share above the 200-day average. Also published: stocks per group, share that rose that day, and the median 21-session return.
- The page’s today table reads the latest week of this series, so the table and chart always agree.

## Index breadth

- **Universe:** price-return NSE equity indices, one series per NSE index code. Bond, money-market and hybrid indices are excluded (by the Data bank’s return variant, and by name for the few it leaves unlabelled), as are currency, leveraged, inverse, futures, arbitrage and VIX indices, and the Nifty 50 dividend-points series (dividends paid in the financial year, reset to zero each April, not a price). The equity Nifty100 Liquid 15 and Nifty Midcap Liquid 15 are counted. That leaves about 124 indices today.
- **Measure:** share of indices above each moving average.
- **Start date:** published from the first date with at least 30 indices (2005-04).
- **Changes, 4 and 5 November 2026:** the count now uses price-return equity indices only, each NSE index counted once under its code. Some bond and hybrid indices (Bharat Bond, G-sec maturity buckets, the 1-day rate and debt indices) had slipped past the old name filter, and an index NSE served under two names was counted twice. On 5 November two more fixes followed: the Nifty 50 dividend-points series, which is not a price, was counted and is now out; the two equity Liquid 15 indices had been dropped by a name filter meant for liquid (money-market) indices and are now in. The share is about 1.7 points lower on average since 2016. On 28 October it was 14.5% above the 200-day (18 of 124 indices), not 19.8%.

## Member breadth for each index (`indices/breadth/<slug>`, 30 October 2026)

For each index page, each week: the share of the index’s members whose adjusted close is above their own 200-session average, on the week’s last session (`compute/index_breadth.py`). `members_counted` is how many members had 200 sessions of prices and traded that day; `members` is the list’s size. Prices are the same adjusted panel as the market-wide figure. The list decides the universe, so SME stocks count for the SME index.

**Members are point in time.** NSE publishes only today’s members for most indices, and a history built from today’s list counts the stocks that survived into it. That tilts the past towards the stocks that did well enough to stay in. So no history is built from today’s members, even labelled. Two sources give dated members:

- **Nifty 50 and Nifty Next 50, since 2008:** NSE’s month-end weightage files. A file lists the members on its month end and stands until the next file, so a change in the middle of a month shows from the next month end (a few weeks at most, around the March and September rebalances). Symbols are matched to the symbol the stock traded under on or before the file’s date, so renamed stocks keep their history.
- **Every index, since 12 Oct 2026:** the constituents lists, fetched each morning. Each new version of a list stands from the IST day it was first seen until the next version. These snapshots are the point-in-time membership table, and every index’s history grows from them, week by week. Before an index’s first list it has no rows.

Each bundle names the list in force for its latest row (`meta.members_list`: source, first and last seen, members) and where its history starts (`history_from`). A member with no price, such as a demerged company before it lists (NSE’s files then print 51 or 54 members), is in `members` but not counted. An index that re-weights another’s members (the equal-weight variants) has that index’s breadth. Test: `pipeline/tests/test_index_breadth.py`, which checks that each week counts the members it had.

## Equal weight vs cap weight

The gap between each equal-weight total-return index and its cap-weighted twin, over the last 63 and 250 sessions, in percentage points, for three pairs (names matched without spaces, so NSE’s renamings join up):

- Nifty 50: equal weight from 1995, cap weight from 1999; 1-year gap from 2000
- Nifty 100: both from 2003; 1-year gap from 2003-12
- Nifty 500: equal weight from 2005; 1-year gap from 2006

Also published: the ratio of the two levels (equal weight over cap weight), which the site rebases to 100 at the start of whatever window is shown. A rising ratio is a stretch when the average stock beat the index.

A positive gap means the typical stock beat the big ones. It is a rough gauge of breadth: return-weighted rather than a count, and it also carries the equal-weight index’s tilt towards the smaller names in the list, its quarterly rebalancing (selling winners, buying losers) and a different sector mix (the cap-weighted indices are heavier in financials). The site reads it beside the stock counts.

## Validation

I compared against BSE’s official daily advance/decline and 52-week high/low counts (via India Data Hub), 2015–2026, 2,810 days:

- **Net advances:** correlation 0.959, and 0.95–0.98 in every single year.
- **Net new highs:** correlation 0.909.

The two exchanges list different stocks, so perfect agreement isn’t expected.

## Known limits

- NSE only. BSE-only companies are absent.
- The liquidity floor is a judgement call. It is set once, here, and not tuned to results.
- The weekly full-history bundle is about 144 KB. Split it per chart when the site is built.

## Participation level versus change (2026-10-05)

Reuse the existing daily liquid point-in-time stock universe and moving-average eligibility. For 20- and 200-session participation, subtract the reading exactly 21 and 63 market sessions earlier. Changes are percentage points, without filled missing inputs. Net new 52-week highs use those same lags, with changes in counts. History uses the existing 60-session median universe threshold.

Four descriptive states use 200-session participation: at least 50% is a majority above the average, below 50% a minority; a strictly positive 21-session change is rising participation, otherwise flat or falling. Missing inputs leave the state blank. Thresholds are not optimised forecasts. Stock entry, exit and changing eligibility affect these changes; this is not a fixed-stock cohort. Existing equal-weight/cap-weight comparisons remain alongside them.

The accompanying Nifty 500 total-return changes use the same two session endpoints, without forward filling. A positive index return alongside a negative participation change describes narrowing support for the index gain; it is not treated as a tested bearish signal.

## Trading activity and liquidity proxies (21 October 2026)

The `/markets/#trading` section uses a **separate, broader universe** from price breadth. Source: `compute/trading_liquidity.py`, published by `publish/trading_liquidity.py` through the regular breadth step. Outputs: `trading_liquidity_daily.parquet`, `markets/breadth/trading_*` bundles.

- NSE main-board EQ/BE/BZ equities, ETFs and SME excluded, **no turnover floor**. All observed historical securities contribute, including subsequently delisted ones. One canonical stock/date comes from the adjusted price panel.
- A security stays on a monitoring roster for 60 market sessions after its last observed bar. Absent bars remain **missing**, not zero turnover or inferred suspensions. A historical listing/suspension master is unavailable here; delistings and prolonged gaps cannot be distinguished. The roster expires after 60 absent sessions. New listings enter when first observed.
- **Turnover direction:** each day’s nominal rupee traded value in stocks with a positive, negative or unchanged adjusted price return, divided by traded value with a valid direction. Require the previous price on the immediately preceding market session. Report direction coverage. A declining-stock trade is not an identified seller-initiated trade.
- **Activity participation:** mean traded value over the latest 20 market sessions divided by the preceding, non-overlapping 60-session mean. Require at least 15 and 45 valid observations respectively and a positive baseline. Count stocks above one, equally weighted among eligible stocks. Missing sessions are not zero-filled. Report eligible/roster coverage and median activity ratio.
- **Concentration:** the 10 and 50 highest-turnover stocks’ share of the day’s total observed traded value, ranked afresh each day. Share quantities are not added across companies.
- **Size:** use the latest AMFI half-year list assumed available 15 calendar days after its period end; match ISIN first, then contemporaneous symbol. No later list is carried backward. Unclassified stocks remain a separate group and in the market denominator. Actual publication timestamps have not all been retained: the lag is an assumption, not a verified release-date history.
- **Amihud-style proxy:** for each stock, mean of absolute daily *percentage* adjusted price returns divided by ₹ crore traded value over 20 market sessions, with at least 15 valid consecutive-session returns and strictly positive turnover. Then take the cross-sectional median. Units: percentage price movement per ₹ crore. News, volatility, nominal rupee values and a changing universe affect the measure; it is neither causal impact nor execution cost.

Full history starts alongside established breadth on 11 September 2003. Weekly bundles keep each week’s last actual observation, **not weekly turnover sums**; the daily two-year bundle retains each observation. Size history is charted from January 2018. The historical-fall table selects the ten worst Nifty 500 daily total returns within the available activity history, excluding 20 observations either side of already selected dates. These are descriptive stress episodes, not an out-of-sample forecasting test. No composite depth score is constructed.

Primary references: [NSE impact cost](https://www.nseindia.com/static/products-services/indices-impact-cost) and [Amihud (2002)](https://www.sciencedirect.com/science/article/pii/S1386418101000246). Actual available bids, spreads, circuit queues and order-book replenishment are not in the bhavcopy. High volume can coexist with very poor execution liquidity.

The breadth activity readings describe observed data and statistical proxies; the MTF selling-time calculations on the linked page are explicitly labelled **simulations**. Adjusted returns inherit the price pipeline’s retrospectively inferred corporate-action corrections; this is descriptive history rather than a point-in-time predictive test.

The daily declining-stock turnover chart draws the observed daily share as a pale coral line and its trailing 20-session average as a darker solid red line. Both are identified in the legend and reading note.

This note is the repository file `docs/methods/breadth.md`, rendered as-is.

---

Historical research, not investment advice. This page in HTML: https://tipsheet.markets/methods/breadth/. Every published dataset: https://tipsheet.markets/data/v1/manifest.json. A map of the site: https://tipsheet.markets/llms.txt
