Survivorship bias

A selection effect that operates without anybody selecting. Companies are delisted, funds are wound up, instruments are discontinued and currency arrangements are abandoned, and the lists available today contain what is left. A test run over today's list has run over the survivors of the period it covers, and it inherited that selection silently rather than choosing it.

It reaches an index the same way. Membership is revised as companies enter and leave, so a long run of index values is a run over a changing set of constituents in which the departures are disproportionately the failures. It also reaches any record of results compiled from what people published, since accounts that stopped publishing are missing from the sample for reasons connected to how they performed.

The correction is a dataset that retains what was removed, together with the date and the reason it was removed, and such datasets are harder to obtain and are not what most retail tools supply. Where one is not available the honest statement is that the result describes survivors, which is a limit on what the number can support rather than a reason to discard it. It is one of three routes by which a backtest produces a figure better than anything the same rule would have delivered at the time.