Fund databases assembled from today's fund lists exclude everything that closed or merged, and studies measuring the survivorship bias in mutual fund performance have estimated the difference between survivor-only and all-fund databases at roughly 0.5 to 1.5 percentage points of annual return — with equity funds toward the higher end — figures published in the academic literature since Elton, Gruber and Blake's work in the mid-1990s and confirmed across subsequent data eras. UZU NEWS publishes information, not investment advice, and measures the bias rather than selecting funds.
"The average fund returned X" is a sentence with a hidden qualifier: the average surviving fund. Closed funds vanish from most easily available lists, and because closure correlates with poor performance, the survivors' average is inflated by construction. The bias is the cleanest demonstration in finance of why sample construction is part of the measurement — and it is checkable whenever the data includes the dead.
Where does the bias come from?
Mechanically, from list assembly. Free and cheap data sources enumerate funds that exist now. Funds liquidate — typically after sustained poor flows and performance — or merge into better-performing siblings, and once gone, they drop out of current-fund lists. A study window that begins today and reaches backward using current membership therefore conditions on survival: every fund in the sample survived to the end. The direction of the resulting error is not subtle, because closure is not random: the funds that disappear are disproportionately the losers, so their absence raises the mean. Attrition rates themselves are documented: mutual fund studies have measured annual attrition in the several-percent range, concentrated among small and poor performers.
How large is the effect?
Measured repeatedly across eras. The original Elton-Gruber-Blake estimates put survivorship bias in the vicinity of 0.8 to 0.85 percentage points annually on their samples; later studies across the 2000s and 2010s estimated ranges of roughly 0.5 to 1.5 points depending on universe, period and fund type, with the canonical consequence — the share of funds beating their benchmark dropping materially when dead funds are included — appearing wherever the test is run. Two adjacent biases compound it: backfill bias, where funds enter a database only after they start reporting, bringing their good early history with them, inflating hedge fund databases especially; and instant-history bias, its close cousin. A performance statistic computed on a survivor-only database can also misstate risk, since the dead funds' drawdowns vanish along with their returns.
How do you check for it?
Three diagnostics, in ascending order of effort. First, ask the data vendor the two questions that matter: does the database include dead funds, and what is the documented attrition rate in the period studied. Second, reconstruct: if returns for closed funds are available, recompute the statistic with and without them — the difference is the bias, measured directly rather than assumed. Third, bound it: when dead-fund returns are unavailable, apply the literature's estimated range as an explicit correction band — reporting "survivor-only mean of X, plausibly X minus 0.5 to 1.5 points" — so the number carries its uncertainty instead of hiding it. Any published fund-performance claim that names neither its universe nor its handling of dead funds has not yet stated its own sample.
Where does it distort conclusions?
Beyond averages, three documented distortions. Performance persistence: funds that did badly tend to die, so the surviving sample overstates the continuity of good performance, weakening claims that past winners repeat. Manager skill estimates: skill inferred from survivors' histories inherits the same conditioning. Strategy backtests: rules selecting from current fund lists — momentum on fund rankings, factor tilts via funds — test on survivors and inherit the inflation. The cure is identical everywhere: point-in-time universes, the same vintage discipline this site applies to prices and fundamentals, applied to fund membership.
| Bias | Mechanism | Typical measured size | Cure |
|---|---|---|---|
| Survivorship | Dead funds absent from current lists | ~0.5-1.5 pp per year | Include dead funds; point-in-time membership |
| Backfill | Funds enter with prior history filled in | Notable in hedge fund data | Entry-date-only returns |
| Instant history | Early partial reporting retro-favored | Smaller sibling of backfill | Same: respect entry dates |
Does index investing change the picture?
Partly, in an instructive direction. Index funds rarely die for performance — they close for commercial reasons — so survivorship in index-fund data is smaller and less performance-correlated than in active funds. But the bias does not vanish: niche index products close, merge and change expense ratios, and studies of index-fund returns that condition on today's product list inherit a milder version of the same conditioning. The general law holds across product types: any statistic computed over what exists today is a statistic about the survivors, and the dead carry the difference.
What should a reader demand of any fund statistic?
Universe, window and dead-fund handling, stated together. The primary literature is accessible: the Elton-Gruber-Blake study and its successors are published in the major finance journals, and the regulatory record — fund registrations and closures — is public through the Securities and Exchange Commission's EDGAR system, with investor materials explaining standardized performance reporting at investor.gov. Averages over survivors answer a question nobody asked; the correction costs a paragraph and restores the number's meaning.
For more context, read What does out-of-sample mean in a forecasting paper?.
For more context, read alternative data evaluation.
For more context, read What regime-detection models can and cannot do.
