Is your Indian equity backtest lying to you?

Two bugs live in the data, not your strategy — and they quietly inflate almost every Indian equity backtest:

• Survivorship bias — delisted / merged companies are dropped from nearly every Indian data source, so your backtest never holds the names that actually blew up (DHFL, Jet Airways, Reliance Capital, IL&FSENGG…).
• Look-ahead bias — retail screeners show restated / latest fundamentals, not what was knowable on the date. Your strategy silently "sees the future."

This is a free, real-data sample of a point-in-time, survivorship-bias-free NSE/BSE dataset built to kill both. Real tickers, real dates — poke at it yourself.

What's in here (/sample_data, CC-BY 4.0)

file	size	what it proves
survivorship_universe.csv	3,838 names	every name that traded 2010–2026, including 1,209 dead/delisted with first- and last-traded dates. The names other sources silently drop.
prices_sample.csv	99 names	daily CA-adjusted close + total-return close (tr_close) + delivery % across 2010–2026. Split/bonus crashes are corrected and demergers handled (total-return continuous across the spin-off) — the unadjusted-split bug that other samples ship is gone. Rename chains (e.g. Bajaj Auto Finance→Bajaj Finance) are spliced into one continuous series. Rights / capital-reduction events (whose total-return can't be cleanly reconstructed without subscription terms) are excluded from this sample and adjusted in the full dataset. deliv_pct is blank (NaN) before 2021-06-07 — NSE did not publish deliverable-% before then, so it's marked unavailable rather than a false 0.
fundamentals_sample.csv	17 names	as-reported quarterly revenue / PAT / EPS stamped with announce_date — the date the number became public, not a restated figure. Median 33-day lag from period-end.

(Money columns in ₹ crore; eps_basic in ₹/share; announce_lag_days = announce_date − period_end. In survivorship_universe.csv, status is complete for all 3,838 names; company_name is present for ~62% and listing_date for ~61% — the rest are delisted / BSE-only names where a clean public name isn't available.)

Poke it yourself (30 seconds)

1. Open survivorship_universe.csv, filter status == inactive → DHFL, Jet Airways, RCOM, IL&FSENGG, Andhra/Dena/Vijaya Bank. Your screener doesn't have these — that's survivorship bias, and it's why your backtest looks better than reality.
2. Open fundamentals_sample.csv and read DHFL down the page — the collapse is right there in the as-reported numbers: 2019-03 PAT −₹2,223 cr, 2020-03 −₹7,507 cr, 2020-12 −₹13,095 cr — each stamped with the date it was actually announced.
3. Every announce_date falls after its period_end (median 33 days). That gap is exactly what restated screeners erase — and exactly how look-ahead bias sneaks into a backtest.

Why it matters — run the 60-second demo

pit_sample_in_use.ipynb runs directly on the real sample CSVs: it filters the survivorship-free universe to the 1,209 delisted names, reads DHFL's collapse straight from the as-reported fundamentals, and plots a delisted stock to zero — every number traceable to a file above. Published research puts survivor-only overstatement around +4.9 pp/yr on the NIFTY Smallcap 250 — the gap is real, not theatrical. The tell that a backtest has these bugs: a Sharpe that looks too good and a suspiciously shallow drawdown.

(A separate PIT_survivorship_demo.ipynb illustrates the four-way strategy comparison with a clearly-labelled, research-calibrated simulation — np.random, not the sample data. It is an illustration only; the linked notebook above is the real-data one.)

The full dataset

This is a taste. The full set: ~3,800 names of CA-adjusted total-return prices (2010–present, delisted included), as-reported fundamentals across ~1,400 names with full announce-date vintages, and point-in-time universe membership — reproducible from source, maintained, and delivered as files into your own environment (not behind an API).

Want the bias measured on your tickers, or the full panel? Open an issue or reach out — pick 15–20 names and a date range, and diff it against whatever you use today.

Honest scope

Income-statement fundamentals today (full balance-sheet / cash-flow on the roadmap). EOD, not intraday. This sample is a bounded slice, not the maintained universe.

---

Data is derived from public exchange filings (NSE/BSE bhavcopy, corporate-action announcements, and as-filed XBRL results) and provided for research and illustration.

Built by Finance-broski. Free under CC-BY 4.0 — use it, fork it, build on it; just keep the attribution.