Survivorship bias is the error of studying only the companies or funds that still exist today. Those that failed, merged or were delisted drop out of the data, so past results look better than the investment experience really was.
A worked example
Take 100 stocks at the start of a ten-year test. By the end, 20 have been delisted after large losses. The numbers are illustrative.
| Group | Number | Average return over the period |
|---|---|---|
| Survivors | 80 | +12% |
| Delisted | 20 | -50% |
| All 100 stocks | 100 | (80 x 12 + 20 x -50) / 100 = -0.4% |
A test that uses only today's listed stocks reports +12%. The honest result for someone who held all 100 is slightly negative.
Where it appears
- Stock backtests that use today's index members or today's listed companies for past dates. See backtesting.
- Fund studies that rank only the funds still open, since poorly performing funds are often closed or merged into others.
- Data with renamed or merged companies. In India, mergers, demergers, name changes and ISIN changes can make a company seem to disappear and another seem to start, unless handled.
How to reduce it
- Use a data set that keeps delisted and merged companies with their full history.
- Rebuild the investable universe for each past date from what was listed then. See stock universe selection and point-in-time data.
- Treat the delisting return carefully: a stock that is delisted for failure should carry its loss, not be dropped with a zero.
It is one of several biases that flatter backtests. See also look-ahead bias.
Frequently asked questions
What is survivorship bias?
It is a distortion that arises when only the entities that survived to the present, such as companies or funds, are included in an analysis, leaving out those that failed.
How does survivorship bias affect backtests?
It makes historical performance look better because the worst performers are missing from the data, so a strategy appears to have avoided losses it would actually have suffered.
Is survivorship bias a problem for mutual fund data?
Yes. Funds that performed badly are more likely to be closed or merged, so data on only the surviving funds overstates the typical result.
How can survivorship bias be avoided?
By using data that includes delisted and merged securities and by rebuilding the universe for each date using only what existed then.