Survivorship bias
Also written survivor bias
What happens when a dataset contains only the securities that still exist. Companies that were acquired, went bankrupt, delisted or left an index drop out of the history along with their returns, so the remaining record describes the winners of a period rather than the period. It flatters every figure computed from it, and it is a property of the data rather than a mistake in the arithmetic.
How it works
Take today's list of index members, pull each one's price history back ten years, and you have built a dataset in which nothing ever failed. Every company in it was still listed, still solvent and still a member on the day you took the list. The companies that were acquired, delisted, went to zero or simply fell out of the index are missing, and so are their returns.
Two separate mechanisms produce the effect, and a dataset can suffer from one and not the other.
Delisted securities absent from the price history. A provider that keeps only currently traded symbols has no record of what happened to the rest. This is why academic work leans on databases built to keep them — CRSP's US stock databases exist substantially to carry the securities that stopped trading, with their delisting information attached.
Index membership taken as today's list. Membership changes constantly, each change is announced individually, and there is no public authoritative archive of who was in on a given date. S&P Dow Jones Indices publishes each change as its own press release with an announcement date and an effective date — Coinbase joining the S&P 500 in May 2025, Honeywell's aerospace business in June 2026 — and reconstructing a decade from those is a manual archaeology project. The membership history itself is licensed data sold by the index provider.
The result is not a small distortion. The names that left did so disproportionately after falling, and the names that joined did so after rising, so a series built from today's members inherits both halves of the selection.
What honest coverage looks like
A dataset that handles this keeps the dead securities with a permanent identifier, so a symbol that was reassigned to a different company does not merge two histories; carries the final prints and the delisting event rather than ending the series quietly; and can answer "who was a member on this date" rather than only "who is a member".
Vendors built for it say so and price accordingly. Norgate Data and CSI Data sell delisted history and historical index membership as a product; AlgoSeek and FirstRate Data carry delisted symbols in their historical files; QuantRocket bundles datasets chosen for this property. A free API that returns a clean ten-year series for a ticker list is almost always giving you the survivors, and will not say so.
The cheap way out is forward-looking. ETF issuers publish full daily holdings for their index funds as a dated file — the SPDR S&P 500 trust's daily holdings file is one — so archiving one every day builds a membership series you own outright. It costs nothing but two years of patience, and it does nothing for the past.
Why it matters here
Anything that measures a rule over history — a screen replayed over past years, a portfolio reconstruction, a comparison of one period with another — inherits this from its data before any code is involved, and no amount of care in the code corrects for records that are not there. It is the first thing to establish about a dataset and the last thing most datasets document.
Two questions do the work: does the history keep securities that stopped trading, and where does the membership list for a past date come from? The sibling entry point-in-time covers the same problem for figures that were later revised rather than for securities that later vanished, and historical index constituents has what the membership history actually costs.
Where you will meet this
The cards where this changes a decision, then the rest that use the word.
Sources
- Coinbase Global Set to Join S&P 500 — S&P Dow Jones Indices,
- Honeywell Aerospace Set to Join S&P 500 & S&P 100; Others to Join S&P MidCap 400 and S&P SmallCap 600 — S&P Dow Jones Indices,
- SPDR S&P 500 ETF Trust daily holdings file — State Street Investment Management,
- CRSP US Stock Databases — Morningstar Indexes (CRSP), read
FAQ
What should "survivorship-bias-free" mean when a vendor says it?
That a security which stopped trading keeps its permanent identifier and its full price history in the dataset, including its final prints, and that a membership list for a past date returns the names that were members then. Ask which of the two the vendor means, because the phrase is often used for only the first.
Can I avoid paying for this?
Going forward, yes. ETF issuers publish full daily holdings for their index funds as a dated file. Archive one every day and in two years you own a point-in-time membership series nobody can revoke. For the past, the membership history is licensed data and there is no authoritative free copy.
Updated