Survivorship bias arises when analysis retains only objects still present at observation time and excludes those that disappeared. Trading examples include delisted shares, closed funds, vanished tokens and instruments whose histories a provider no longer retains.
Today’s list is not yesterday’s
A test starting in 2015 but using today’s index members introduces future membership information. Later successful entrants appear prematurely, while failed or excluded companies may disappear entirely. A defensible universe requires membership and availability on each historical date.
Hypothetical example
Ten instruments begin with equal weights. Eight gain 10%; two lose 100%. The original set’s mean simple return is:
(8 × 10% − 2 × 100%) / 10 = −12%.
Removing the two disappeared instruments produces a reported +10%. The example assumes buy-and-hold, no reweighting and no costs. It illustrates selection bias, not a typical market return.
Missing data can have other causes
Licensing, technical loss or identifier changes can also remove histories. Investigate these separately rather than calling every omission survivorship bias.
Do not add assets that were genuinely unavailable at decision time. The criterion is historical availability, not collecting absolutely every asset.
Audit the selection rule
Check historical membership, listing dates, delistings and final payouts; ticker changes, mergers and splits; and the treatment of incomplete histories. Define exclusions before examining results and assess whether missingness is associated with bad outcomes.
Choosing only favourable years also distorts the evaluated population. Bailey and colleagues discuss selection across many tested variants. See the broader hypothesis-testing protocol.