---
title: "Method: IPOs"
url: https://tipsheet.markets/methods/ipo/
description: "How tipsheet.markets computes ipos: sources, start dates, checks and caveats."
updated: 2026-10-29
publisher: tipsheet (https://tipsheet.markets/)
cite: "tipsheet, “Method: IPOs”, 29 October 2026, https://tipsheet.markets/methods/ipo/"
---

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

# IPOs: methods

**Code:** `pipeline/tipsheet/compute/ipo.py`, `compute/ipo_size.py` (step `page_ipo_size`), `publish/ipo.py`. **Tests:** `pipeline/tests/test_ipo.py`, `test_ipo_size.py`. **Pre-registered spec and results:** `docs/research/ipo_spec.md` (H1–H3); `docs/research/site_calculations_spec.md` (C4, C40–C43). **Bundles:** `ipo/*`.

## What I measure

- **Listing tracker:** every mainboard and SME listing on NSE since 2016. For each one: issue price, listing-day open and close, listing gain, and the return since listing against Nifty 500 over the same dates.
- **Returns after listing:** returns at 1 month, 6 months, 1, 3 and 5 years, against three benchmarks: Nifty 500, Nifty500 Equal Weight and Smallcap 250. Reported for all listings together and by listing-year cohort.
- **Listing day:** how many listings closed day one above, at and below the issue price, by year, with the median, quartiles and mean of the listing gain, and a histogram of gains in 10-point bins. The most common bin stands in for a mode, because gains are continuous.
- **Returns from the issue price (the applicant’s view):** at 1 week, 1, 3 and 6 months, 1, 3 and 5 years, against Nifty 500 from the issue close date. Also grouped by how the first day closed. Added 2026-10-09 as a logged amendment to the spec.
- **Calendar-time portfolio:** each month, an equal-weighted portfolio of every mainboard IPO in months 1 to 36 after listing. This is the clean test of whether IPOs beat or trail the market.
- **Heat gauge:** how busy and how hot the IPO market is. It ranks two things against all earlier history:
  
  - the number of mainboard listings in the last three months
  - their median listing gain
- **Size-matched peers (C40):** each IPO from 2018 against the ten seasoned companies nearest its market value on the day before listing, in calendar time and buy-and-hold.
- **Size and promoter stake (C41, C42):** outcomes by size tercile at listing, and by the promoter’s stake in the first shareholding filing after listing (listings from 2021).
- **Who is counted (C43):** every mainboard listing that stopped trading, and how the results change if each is a total loss.
- **The IPO cycle (C4):** IPO money raised over 12 months as a share of NSE market value, against later Nifty 500 TRI returns.
- **Scorecard:** every pre-registered IPO test with its result, on the page.

## Sources

- **Issues:** NSE’s public past-issues list. My data archive fetches it daily as `nse_ipo_history`; the exchange serves it from 2016-01. It gives the symbol, issue dates, issue price and NSE’s listing date. It has no ISIN, subscription or issue size.
- **Prices:** NSE bhavcopy, through the adjusted price panel (`prices.md`). Delisted stocks are included, and prices are adjusted for splits, bonuses, rights and demergers.
- **Money raised:** SEBI’s equity capital-raising statistics via IndiaDataHub (`CMEQCR*`). Until March 2019 the FPO series is mostly public bond issues; the SEBI Bulletin’s public debt is subtracted from it and from the equity total, and totals before April 2014 are blank. Details and every other route (bonds, REITs, InvITs, offer for sale) are in `capital_raising.md`.
- **Benchmarks:** niftyindices price indices (`nifty_index_history`). Nifty 500 TRI is used as a robustness check.
- **Company size:** AMFI’s half-yearly lists of average market capitalisation (`amfi_stock_categorisation`, from July–December 2017). XBRL share counts (`nse_shareholding_xbrl`) are used only to validate.
- **Promoter stake:** NSE shareholding pattern summaries (`nse_shareholding_summary`); filings near listing exist from 2021.
- **Market value for C4:** NSE’s month-end market value of listed companies (`density_mcap_gdp.parquet`).

## Matching an issue to its listing

- **The anchor is the issue close date.** The listing is the symbol’s first NSE session after the close, within 15 sessions. Every match so far fell 2 to 11 sessions after the close; T+6 settlement applied before 2023, and T+3 since.
- **The listing must be fresh.** The ISIN on the listing row must never have traded on NSE before, under any symbol. This rules out FPOs, relistings and reused symbols.
- **BSE-first listings are excluded.** A mainboard issue whose first NSE trade is not on NSE’s listing date listed on BSE first, so its first NSE trade is not a listing day. CAMS is the example: it listed on BSE in 2020.
- **NSE’s listing date is a cross-check only.** It agrees exactly for 1,122 listings.
  
  - For 129 SME issues it holds the later migration to the main board, not the SME listing.
  - Some mainboard rows leave it blank, for example DCX Systems.
- **Exclusions:** every excluded issue is written with its reason to `.cache/derived/qa_ipo_excluded.csv`.

## Definitions

- **Listing gain** = listing-day close ÷ issue price − 1. The open gain is shown alongside.
  
  - NSE’s listing-day open is set in the special pre-open call auction for IPOs.
  - Retail and employee discounts are ignored.
- **Return from the issue price** = listing close ÷ issue price × adjusted close at the horizon ÷ adjusted listing close − 1. A later bonus or split doesn’t read as a fall. Its Nifty 500 comparison runs from the last session on or before the issue close date, when the applicant’s money is committed, to the same end date. Horizons count NSE sessions from the listing session: 5, 21, 63, 126, 250, 750 and 1,250.
- **Returns after listing** start from the listing-day close and use adjusted prices. They are price returns: dividends are left out on both sides, and the benchmarks are price indices.
- **BHAR (buy-and-hold abnormal return)** = IPO return − benchmark return over identical dates.
- **Wealth relative** = (1 + mean IPO return) ÷ (1 + mean benchmark return). Above 1 means IPOs did better on average.
- **Horizons** count NSE sessions: 21, 126, 250, 750 and 1,250. Recent listings are left out of horizons that haven’t finished.
- **A stock that stops trading** exits at its last close, and the benchmark is cut to the same dates. The `exited_early` count shows how often this happens.
- **Heat:** each row covers the three months ending that month.
  
  - Both parts, the listing count and the median listing gain, are ranked against all earlier rows. heat is the mean of the two ranks.
  - A month is hot when heat is at least 2/3.
  - Ranks start in 2017-03, after 12 months of history.
  - When testing hot markets, an issue listing in month *m* uses the row for *m*−1, so its own month never colours its label.

## Size, peers and promoter stake

- **Share counts without look-ahead.** An AMFI list gives each company’s average market value over a half-year. Divided by the stock’s mean adjusted NSE close over the same sessions, it gives an adjusted share count that does not depend on how the price moved. Market value on a date = that count × the adjusted close, so splits and bonuses cancel.
- **An IPO’s size** = the count from the list whose half-year contains the listing date × the listing-day close. Only the share count comes from the later list.
  
  - Checked against XBRL share counts filed within 45 days of listing: all 117 checkable 2025–26 listings agree within 10%, median ratio 1.000.
  - 435 of 539 mainboard listings are sized. 2016 listings and those after June 2026 have no list.
- **Peers:** from the latest AMFI list public by the listing date (15 days after its period ends), companies with at least 36 months on NSE that traded the session before; the ten nearest in log market value on that session.
  
  - Peer baskets are equal-weighted and held over exactly the IPO’s dates. A peer that stops trading exits at its last close.
  - Calendar-time test: months 2–36 after listing, IPOs minus their baskets, months with 10 or more IPOs, Newey-West t (3 lags).
- **Promoter stake:** the first filing dated 45 days before to 120 days after listing.
- **Renames that keep the ISIN** are stitched: the price panel starts a new canon at some renames (BURGERKING → RBA, 2022). Without this the stock looks as if it stopped trading. One mainboard and two SME listings were affected.

## Validation

- **Listing prices:** my listing close matched IPO Watch’s figure from NSE press releases exactly in 478 of 479 mainboard issues. The one miss is a join collision in the comparison, not a price error.
- **Against NSE’s bhavcopy MCP server** (`mcp.nseindia.in/bhavcopy/cm`), an independent route to the same exchange data: listing-day open and close agreed to the paisa for all 120 listings in a random sample of 60 mainboard and 60 SME listings from 2025–26 (checked 2026-10-09). None of the 120 had a candle before its listing date. The server lags a few sessions and lacks some older symbols, so it is a check, not a source.
- **Spot checks against well-known listings:**
  
  | Issue | Issue price | Listing open | Listing close |
  | --- | --- | --- | --- |
  | Bajaj Housing Finance | ₹70 | 150 | 165 |
  | Paytm | ₹2,150 | 1,950 | 1,560.80 |
  | Tata Technologies | ₹500 | 1,200 | 1,313 |
  | Hyundai | ₹1,960 | 1,934 | 1,819.60 |
  
  Widely quoted closes sometimes differ by a few rupees. Those are BSE closes.
- **Extreme returns checked by hand:**
  
  - Mainboard: Mazagon Dock rose 11.4× in three years.
  - SME: NPST rose 108× and KSOLVES 79×.
  
  These are real. Every bonus factor behind them matches the ex-date price gap.

## Findings (to 2026-10-14)

- **No long-run underperformance and no outperformance.** The calendar-time alpha against Nifty 500 is +4.2% a year, Newey-West t = 0.92. It isn’t significant against any benchmark.
- **Against companies of the same size, no difference either (C40):** +2.5% a year, t = 0.48, from 2018. One IPO at a time it is harsher: over three years the median IPO trailed its size-matched peers by 53 points and a third beat them, against 29 points and 40% for Nifty 500 (all IPOs from 2016).
- **The typical IPO trails the market, and a few big winners lift the average.** At 3 years the mean BHAR against Nifty 500 is +17%, the median −29%, and 40% of IPOs beat the index.
- **Listing gain says nothing about the next year:** Spearman 0.016 against Nifty 500, 0.000 against size peers.
- **Hot markets:** IPOs from hot months trailed by 5.8 points at one year, p = 0.34. Not significant.
- **Size (C41) and promoter stake (C42):** small IPOs gave applicants less (median +6% from the issue price after a year against +30% for large ones), but did no worse against their own size peers (Spearman −0.01). A smaller promoter stake went with a weaker year only faintly (Spearman +0.08, t 1.31). The eleven listings with no promoter, eight of them new-age companies listed in 2021 and 2024, had a median −19% from the issue price after a year.
- **Survivorship (C43):** 4 mainboard listings stopped trading (three mergers, one failure), none inside a three-year window. Treating them as total losses changes nothing.
- **IPO cycle (C4):** busy IPO years were followed by weaker Nifty 500 years more often than not (12-month slope negative in every non-overlapping offset), but t = −1.86 on 12.5 independent years, and nothing at 36 months (t = −0.71).
- **Heat now:** August to mid-October 2026 had 57 mainboard listings with a median listing gain of 15.2%; heat is 0.77 (hot).

## Known limits

- **NSE only, from 2016.** IPO Watch has issues from about 2000; adding them is a separate, validated step.
- **Subscription is not published.**
  
  - IPO Watch’s figures look like a snapshot taken during the last bidding day. Bajaj Housing Finance shows 35.2× there against about 63.6× final.
  - My data archive’s 06:00 fetch sees each book before the last day’s bids.
  - Since 2026-10-01 a 20:00 weekday fetch (`scripts/ipo_evening.sh`) records the end-of-day book for issues closing that day, so final subscription accumulates from now on.
- **Issue size is not used.** No source has it reliably: IPO Watch’s sizes have a units error in about two-thirds of mainboard rows. Everything is equal-weighted. Company size at listing (from AMFI) stands in where size matters, and the offer-for-sale share is known only in aggregate (SEBI, from April 2023, on the capital-raising page), not per issue.
- **What an allotted applicant earned** depends on allotment odds, which need final category-wise subscription. Only total subscription snapshots from October 2026 exist, so this is not computed.
- **Small samples in some years.** There were only 13–14 mainboard listings in 2019 and 2020, and the 5-year horizon rests on the 2016–2021 cohorts.
- **SME is descriptive only.** Thin floats and a few multibaggers dominate the means.

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

---

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