Survivorship bias happens when a backtest only sees the winners that lasted. Build a stock list from what trades today and every company that went bankrupt along the way is gone from your history. The losers didn't stop existing, you just stopped looking at them... and your results look better because of it.
What is survivorship bias?
Testing on a group chosen by who survived until today, instead of the group that existed back then. A strategy that bought stocks in 2015 could have bought ones that later went to zero, but a list built in 2025 doesn't contain them. The test skips the worst trades it would have taken, so the average looks better than any live version of it could be.
Where does it show up?
In more places than stock lists:
| Where | What survived | What's missing |
|---|---|---|
| Stock lists | companies still listed | bankruptcies and delistings |
| Crypto | coins still trading | coins that went to zero or got delisted |
| Index members | today's members | companies that dropped out |
| Strategies you copy | traders still posting results | traders who blew up and went quiet |
| Your own tests | ideas that looked good early | ideas you dropped along the way |
The last two are about you, not your data. You hear from the traders who are still around, and you keep testing the ideas that worked first, which is where data snooping starts.
How do you avoid it?
Test on the list as it was on each date, not the list as it is today. That means a universe that includes delisted names, and a rule that picks from that list on the day, rather than a hand picked group you already know did well. If you can't get that data, be suspicious of any "top 20 today" version, and prefer markets where the instruments don't disappear: major forex pairs and index futures carry very little of this bias, because the same EURUSD and the same S&P contract existed through the entire test.
Why it's easy to miss
Nothing in the backtest looks wrong. The curve is smooth, the trades are real prices, every number checks out. What's wrong is the list of things the test was allowed to buy, and that list was written with knowledge of who survived, which makes it similar to look ahead bias. It's one of the reasons a backtest should be read as the best case, not the forecast.
Frequently asked questions
Related terms
- Expectancy — What one average trade is worth to you, after the wins and losses cancel out.
- Overfitting — When a strategy is tuned so tightly to the past that it only works on the past.
- Out of Sample — The data your strategy never saw while you built it, kept back to test if the edge is real.
- Walk Forward — Out of sample testing done over and over, rolling through your whole history instead of one split.