On This Page
Key Takeaways
- Survivorship bias comes from studying only the names still around today. Keep every name that existed at the start, with its actual outcome, when you test the past.
- The direction and size of the difference depend on which entities are omitted, their outcomes and the strategy’s weights. Acquisitions end some listings at a gain.
- Liang’s hedge-fund database study found a return difference exceeding 2% per year in its sample. Measure the gap on your own dataset and strategy before you trust a backtest.
- Backtests and screens are especially exposed when they start from a current index or a present-day symbol list, which already excludes everything that dropped out.
- The fix is a point-in-time, survivorship-bias-free dataset that carries delisted names and closed funds with the returns they actually posted before disappearing.
What is survivorship bias?
Survivorship bias is the distortion that arises when a sample is selected by which entities still exist at a later date, while the analysis concerns an earlier opportunity set. A backtest built from today’s index members, or a fund study that drops closed funds, measures only the survivors, so its average reports the survivors’ record instead of the return the full starting universe delivered.
Closures, mergers, acquisitions and delistings end with different outcomes, so keep each exited name with its actual returns, distributions and exit proceeds.
Formula
Survivor-selection difference = average result of retained entities − average result of the defined full cohort, using a common horizon and weighting.- The surviving sample contains only entities that still exist at the measurement date.
- Count every name in the starting cohort: its return through exit, plus whatever its exit proceeds earn through the common year end.
- A positive difference means the survivors flatter the record by that many points. Your own portfolio return also depends on its holdings, weights, timing and costs.
Survivorship bias example: a 100-stock cohort
Start with 100 equally weighted stocks over one year. By year end, 88 remain listed and 12 have exited. Exit proceeds sit in zero-return cash through the common year end.
- The 88 retained names have a weighted return sum of 970 percentage points, giving an average of 11.0%.
- The 12 exited names have a weighted return sum of −480 percentage points, giving an average of −40%.
- All 100 names average (970 − 480) / 100 = 4.9%.
- Retained-cohort difference: 11.0% − 4.9% = 6.1 percentage points.
The survivors average 11.0%; the full starting cohort averages 4.9%. A backtest built on the 88 survivors overstates the cohort’s return by 6.1 points, all of it from leaving out the 12 exits. Run your own holdings through the same test when your strategy trades only part of the cohort.
The chart follows a fixed starting cohort of 100 stocks over one year. Eighty-eight names remain listed and twelve exit, and the exit proceeds sit in zero-return cash through year end.
The retained names’ return sum is 970 percentage points, giving 11.0% across 88 names. The twelve exited names sum to −480 points and average −40%.
All 100 names average 4.9%. Studying only the survivors overstates the cohort by 6.1 percentage points.
An equal-weight buy-and-hold of all 100 names earns the 4.9% average before trading costs. A momentum screen with changing holdings needs its own portfolio calculation.
How to build a survivorship-bias-free backtest
Put every name on the same horizon. Carry each exited holding’s proceeds to the common endpoint under one rule, instead of averaging a full-year return for survivors with a shorter lifetime return for exits.
A current index list also contains entrants that were absent earlier. A point-in-time universe tracks eligibility, entries and exits at each historical decision date. Use the same eligibility and contract definitions in research and execution, and evaluate the strategy with the names and contracts available at each decision.
Drawdown and other path measures come from the portfolio’s day-by-day sequence, so rebuild that sequence with the exited names before you read the worst loss.
How to measure survivorship bias
Subtract the full cohort’s metric from the retained cohort’s metric using matched weighting and horizon. The direction depends on the omitted outcomes, because an acquisition or merger exits some names at a gain.
Liang’s hedge-fund database study found a difference exceeding 2% per year in its sample, with variation across investment styles. Treat that 2% as evidence the gap is large enough to matter, and measure your own dataset’s gap directly.
Reverse survivorship bias and manager skill
Linnainmaa’s reverse-survivorship research looks at funds that exit after poor returns driven partly by negative idiosyncratic shocks, the bad-luck part of a fund’s return. In that model, the measured alpha understates the managers’ skill.
Keep that skill question separate from the full cohort’s realized return. To fill a missing history, use the actual records or a modeled estimate built from them; a flat haircut leaves the size of the error unknown.
Limitations of survivorship bias corrections
- Measure the bias on your own data. The example’s survivors overstate the cohort by 6.1 points and Liang found more than 2% a year in hedge-fund databases, but your dataset, omitted names and weights set your gap.
- Use a dataset that carries delisted names with their actual returns and exit proceeds. Where records are missing, a flat haircut leaves the error unknown, so size the strategy for a wider range of outcomes.
- A present-day membership list used for historical decisions carries look-ahead bias as well, so rebuild the universe at each date to fix both.
- When acquisitions and mergers end listings at a gain, the bias runs the other way. Check what the dropouts earned before you mark a backtest down.
- A full-cohort return reports what investors earned. Judging a manager’s skill takes a separate test, because bad luck closes some good funds.
Bottom line
Survivorship bias distorts history by studying only the entities that lasted. Test strategies on a point-in-time universe that keeps every delisted stock and closed fund with its actual exit outcome, and measure the effect on your own dataset rather than applying a generic adjustment.
Clear answers
Frequently Asked Questions
Sources and Further Reading
- CFA Institute: Backtesting and Simulation
- CFA Institute: Reverse Survivorship Bias (Digest Summary)
- Bing Liang, Hedge Funds: The Living and the Dead, Journal of Financial and Quantitative Analysis
How entries are sourced, calculated and updated: editorial methodology.