Survivorship Bias

intermediate7 min read

Testing only on companies that survived ignores all the ones that died. The graveyard matters.

Survivorship bias is testing your strategy only on the companies that survived to today — silently ignoring all the ones that went bankrupt, got delisted, or were acquired. Because failures vanish from most datasets, your test sees a rosier history than ever existed.

The famous wartime image makes it click: engineers wanted to armour returning planes where the bullet holes were — until a statistician pointed out those were the planes that survived; the armour belonged where the holes weren’t, because planes hit there never came back. Stock backtests make the identical error: a dataset of “today’s Nifty 500” for the last 20 years only contains companies good enough to still exist — every Kingfisher, every fraud, every bankruptcy has been quietly deleted. So your strategy is tested on a universe pre-filtered for success, inflating returns and hiding catastrophic losses that really happened. The graveyard matters: you must test on a point-in-time universe that includes the companies that later died, exactly as you’d have faced them in real time.

The bias — only-survivors datasets delete the bankrupt/delisted, so the past looks far safer and more profitable than it was.
The plane analogy — judging from survivors alone points you to exactly the wrong conclusion.
The damage — inflated returns, hidden drawdowns; a strategy that “avoids losers” may just be testing on losers that were pre-removed.
The fix — use a survivorship-bias-free, point-in-time universe that includes companies as they existed (and died) historically.

ExampleBacktesting “buy and hold the current Nifty 50” over 20 years looks fantastic — but those are the 50 winners that made it to today. The real 20-years-ago investor also held names that later collapsed or were ejected. Including the dead companies turns a glowing result into a realistic, far more sobering one.

Key takeawaySurvivorship bias tests only on companies that survived, deleting the bankruptcies and delistings — like armouring planes by their survivors’ bullet holes. It inflates returns and hides real losses. Fix it with a survivorship-bias-free, point-in-time universe that includes the companies that later died.

FAQs

How big is the survivorship effect, really?

It can be large — studies suggest survivorship can overstate returns by a meaningful margin per year, and it especially flatters strategies in small-caps or distressed names (where failures are common). Always check whether your data vendor provides a point-in-time, survivorship-free universe; if it only has currently-listed names, your results are biased upward.