Survivorship bias distorts investment comparisons when closed or merged funds disappear from the sample, leaving an incomplete performance record.
Survivorship bias is the distortion created when an investment study includes only funds or securities that still exist and omits those that disappeared. If the missing investments performed worse, the surviving group makes the original opportunity set look more successful than it was.
The problem is not that a surviving fund’s own return is necessarily wrong. It is that the group being measured has changed after the outcomes became known. A list of funds available today cannot, by itself, show what someone choosing among funds years ago could have experienced.
Imagine a database search that selects “active funds” and then ranks their ten-year returns. That search answers a question about funds that remain active. It does not necessarily answer how all funds available at the start of those ten years performed.
A liquidated fund may return remaining assets to shareholders. A merged fund may transfer investors into another fund. If the original record is removed, its returns before the event can disappear from the comparison.
S&P Dow Jones Indices’ SPIVA methodology addresses this issue by considering the opportunity set at the start of a study, including funds subsequently merged or liquidated. It also distinguishes equal-weighted from asset-weighted results. Those are separate choices: weighting the surviving funds by size does not restore funds missing from the data.
Suppose five hypothetical funds are available at the start of a year. Each receives $10,000. Their total returns include distributions and fund expenses. Ignore investor-level charges and taxes.
Funds D and E liquidate at year-end, after earning the returns shown. Their ending values are paid to investors. This timing keeps the example from needing an assumption about how early liquidation proceeds would be reinvested.
| Fund | Full-year total return | Year-end value or liquidation proceeds | Included in a survivors-only list? |
|---|---|---|---|
| A | 12% | $11,200 | Yes |
| B | 8% | $10,800 | Yes |
| C | 4% | $10,400 | Yes |
| D | -10% | $9,000 | No |
| E | -24% | $7,600 | No |
The survivors-only average is:
The original five-fund average is:
An investor who put equal amounts into all five funds would finish with $49,000 from $50,000, a 2% loss. Reporting only the survivors changes the average from -2% to 8%, an upward distortion of 10 percentage points.
Neither D nor E lost everything. Assigning both a -100% return merely because they closed would create a different error.
These numbers illustrate the mechanism, not an estimate of bias in any actual market. The result uses equal starting investments over one common year; a multi-year portfolio with cash flows needs a more detailed calculation.
A useful review starts with the question being asked. A study of funds available on a particular start date needs that date’s fund list. A strategy that adds new funds over time needs explicit entry rules instead.
| Check | Why it matters |
|---|---|
| Membership as of each relevant date | Today’s fund list may omit historical choices or include funds not yet launched. |
| Closure and merger records | The original fund’s identity and returns should not disappear when its name or structure changes. |
| Returns through the exit event | Missing final months can hide losses or gains immediately before liquidation or merger. |
| Treatment of proceeds and successor funds | Following an investor’s wealth after an event is different from describing only the original fund’s operating history. |
| Share-class and weighting rules | Counting several classes as independent portfolios can overrepresent one strategy; different classes can also have different expenses. |
Do not splice a successor fund’s earlier history onto a predecessor as though they were one unchanged portfolio. If the analysis follows investors into the successor after a merger, label that continuation and explain the assumption.
Data quality remains important even in a database designed to retain defunct funds. In their study of historical mutual fund databases, Elton, Gruber, and Blake examined omissions near fund exits and discrepancies in event dates. That historical study is a reason to inspect data and methodology, not a claim that today’s versions contain the same errors.
Survivorship bias concerns who remains in the sample. Selecting only favorable calendar years instead concerns which periods are displayed. Both can make results misleading, but fixing one does not fix the other.
There can also be overlap with look-ahead bias. A stock Backtesting exercise that applies today’s index membership to the distant past uses information that was not available then. Historical membership dates and delisted-security returns help avoid giving the strategy hindsight. Research on look-ahead benchmark bias examines this distinction.
A correct calculation on an incorrectly selected sample is still an unreliable comparison.
Changing from an arithmetic to a geometric Mean Return does not recover omitted observations. Nor does adding a dummy variable for closure supply missing returns.
If historical data cannot be recovered, an analyst can disclose the missing coverage and test explicit assumptions about the absent results. Those are sensitivity estimates, not a fully observed track record. A simulation based only on survivors can reproduce the same selection problem rather than solve it.
The direction and size of bias depend on who is missing and why. The upward effect in the example comes from removing below-average funds. There is no universal percentage adjustment to subtract from every published return, and survival alone is not evidence of investment skill or safety.
This article is educational, not a recommendation to buy or avoid a particular fund or strategy.