---
title: "Method: Position limits, the F&O ban list and the tests around it"
url: https://tipsheet.markets/methods/fo_crowding/
description: "How tipsheet.markets computes position limits, the f&o ban list and the tests around it: sources, start dates, checks and caveats."
updated: 2026-10-29
publisher: tipsheet (https://tipsheet.markets/)
cite: "tipsheet, “Method: Position limits, the F&O ban list and the tests around it”, 29 October 2026, https://tipsheet.markets/methods/fo_crowding/"
---

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

# Position limits, the F&O ban list and the tests around it: methods

Code: `pipeline/tipsheet/compute/fo_crowding.py`, `pipeline/tipsheet/publish/fo_crowding.py`, step `fo_crowding`. Tests: `pipeline/tests/test_fo_crowding.py`. Page: `/fo/#crowding`. Pre-registration and results: C130–C134 in `docs/research/site_calculations_spec.md`.

## What I measure

- **Utilisation.** For each F&O stock and session, open interest across exchanges divided by the stock’s market-wide position limit (MWPL). NSE sets the MWPL from the stock’s free float. The open interest is the one NSE’s ban rule used on that date: the notional open interest until 30 September 2025, and the delta-weighted “future equivalent” open interest from 1 October 2025, when NSE added that column and started computing the limit from it. NSE’s printed “limit for next day” (95% of MWPL less that open interest) checks this to within five shares on all but 18 of 756,252 rows.
- **The ban period.** Once utilisation crosses 95%, NSE prints “No Fresh Positions”: from the next session nobody may add to positions in the stock’s futures and options. The flag stays until utilisation falls below 80%. I count stocks in the ban period and above 80% each day, ban spells (consecutive flagged files) and entries per year.
- **The board.** On the latest session, F&O stocks by utilisation, the change over five sessions, ban status and length, and from NSE’s client file how many clients hold at least 3% of the stock’s MWPL and the largest one’s share.
- **Five pre-registered tests.** C130 asks whether ban entries follow unusual moves, C131 whether they lead the return, C132 whether the move into the ban reverses, C133 whether volatility falls afterwards, and C134 whether a concentrated client makes a ban more likely. The rules were written before any outcome was computed.

## Sources

- `nse_fo_mwpl`: NSE’s “Combine Open Interest across exchanges” file, `combineoi_DDMMYYYY.zip`, one per session from 1 February 2010 (earlier dates return 404). Collected in the Data bank, with history fetched once from the laptop and seeded to R2. The daily file is posted at about 19:00 IST.
- `nse_fo_mwpl_clients`: NSE’s `mwpl_cli_DDMMYYYY.xls`, one per session from 2 January 2012, posted at the same time. It lists, for each stock with any, the positions of clients holding at least 3% of the MWPL, as percentages and without names, largest first.
- Prices: Tipsheet’s split- and bonus-adjusted bhavcopy closes (`equity_prices_adjusted`; price only, no dividends), joined on date and NSE symbol and followed through renames.

## Gaps and repairs in the source

- Twelve `combineoi` files, 28 May to 13 June 2012, are corrupt at NSE: the ZIP directory and both compressed streams are broken, and a fresh download is byte-identical. Those sessions are missing. A stock’s “previous file” must therefore be at most a week older for a flag to count as an entry, so a stock first flagged after the gap is not counted as entering on 14 June.
- `combineoi` for 22 March 2014 and 1 October 2015 returns 404.
- The file for 2 November 2012 prints 5 November on every row. Its open interest equals the 2 November F&O bhavcopy (RELIANCE 17,747,000 shares), so it is kept under 2 November.
- The 30 June 2025 ZIP is unreadable; its CSV is taken from the 1 July 2025 ZIP, which carries it.
- Client files: 42 dates return 404 and two files (9 March and 20 November 2012) are corrupt at NSE, leaving 3,630 sessions. The 1 January 2013 file ends most rows with the number of clients before it (BATAINDIA has three clients, then a 3); those cells are dropped. Nineteen files misnumber their “Client N” headers, so a client’s rank is its column, checked by each row being in descending order. The 28 June 2018 file carries seven stale rows from 27 June after its alphabetical list restarts; they are dropped. Sessions without a client file are left out of C134.

## The tests, in brief

- **Universe and excess return.** U(t) is every stock in day t’s file with a close on t. Membership comes from each day’s file, so stocks later dropped from F&O or delisted stay in while they were there. Excess return is the stock’s return less the equal-weighted mean of U(t) over the same window.
- **Events.** A fresh entry is a stock flagged in t’s file, present and unflagged in its previous file, and unflagged in the 20 files before. The flag in t’s file applies from t+1, and the file appears in the evening, so forward windows start at the close of t+1.
- **Inference.** Means use a sign-flip permutation clustered by calendar month (20,000 draws). The slope and the chart’s intervals use a month-block bootstrap. C134 compares stocks at the same 5-point utilisation band and year, weighting each stratum by n₁n₀/(n₁+n₀), with a month-block bootstrap interval. The seed throughout is 20261029.

## What the results say, and what they do not

- Ban entries follow large moves in either direction, about half after rises and half after falls. The median absolute 20-session excess return is about 1.85 times the usual one, short of the pre-registered 2×.
- There is no reliable return after entry over 20 sessions, and no reversal of the move into the ban. A small positive mean over the next five sessions is a secondary, ungraded reading and should not be traded on.
- Volatility over the 20 sessions after entry is lower relative to the 20 before than for other stocks. This is partly the fading of the burst that caused the ban. It is not evidence that the ban itself calms a stock.
- Stocks with a listed client of 3% or more were a few points more likely to reach the ban within 20 sessions than stocks at the same utilisation without one. Most stocks near the limit have such a client, so the comparison group is small. The result is an association, not a claim that those clients cause bans.
- Everything here describes the past. None of it is advice to trade around the ban list.

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

---

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