Survivorship bias
What is Survivorship bias?
Survivorship bias is drawing conclusions from investments, funds, managers, or companies that remain observable while omitting those that failed or disappeared.
Backtest datasets should retain delisted securities, closed funds, failed managers, mergers, and historical constituents with point-in-time identifiers. Current membership cannot be projected backward. Coverage reports should quantify missing failures and stale instruments, since sophisticated calculations on a survivor-only universe still produce misleading precision.
Vendor changes and identifier mapping should be versioned because a missing security can otherwise vanish silently when a database migrates or a corporate entity is reorganized. Coverage differences require explicit reconciliation and transparent disclosure.
How it enters data
Databases can remove liquidated funds, delisted securities, bankrupt companies, closed accounts, failed forecasts, and departed managers. Historical constituents can be replaced by today's winners. Voluntary reporting can retain successful products and lose poor ones. The resulting sample looks stronger and sometimes less risky than the actual opportunity set investors faced.
Portfolio consequences
Average fund return, manager persistence, factor performance, business quality, and backtested strategies can all be overstated. Failure rates, drawdowns, fees, and liquidity stress are understated. Choosing today's successful companies and testing them in the past gives the strategy information unavailable then, even if every historical price used is accurate.
Example
A backtest buys the current members of a stock index ten years ago. It excludes firms removed after decline or bankruptcy and includes companies added after strong growth. The result measures hindsight-selected survivors, not an implementable index strategy. Point-in-time constituents, delisting returns, and corporate actions are required.
Data controls
Use point-in-time universes, dead-fund files, delisted prices, merger proceeds, bankruptcies, and historical identifiers. Record fund incubation and backfill, reporting start and stop dates, and manager changes. Compare complete cohorts by launch vintage. Missing outcomes should be quantified and stress-tested rather than assumed equal to surviving observations.
Practical interpretation
Ask who or what is absent, why reporting ended, and whether inclusion depended on later success. Reconstruct the investable set and costs at each date. Survivorship correction does not eliminate selection, look-ahead, or publication bias, so multiple checks remain necessary. Treat extraordinary survivor stories as possible outcomes, not base-rate forecasts.
Sources and further reading
- Behavioral Patterns of U.S. Investors, U.S. Securities and Exchange Commission