Survivorship bias

Also seen as: Survivor bias

Why does survivorship bias make results look better than reality?

Survivorship bias inflates results because failure is the one outcome that removes a name from the list. Every company that fell to zero, was taken over at a discount, or was delisted for non-compliance leaves the surviving set, while every company that merely did well stays in it. The test therefore samples from a population that was defined by having done well enough to still exist — a condition that could not be known in advance. Two things follow: average return is overstated, and the depth and frequency of large losses are understated, because the events that produce the very worst losses have been filtered out.

The same logic applies to funds and strategies, not just single stocks: a study of the funds available today has quietly excluded every fund that closed after poor performance.

Worked example: an S&P backtest on today's index members

Consider a ten-year backtest that buys "the S&P 500" but sources its instrument list from the index as it stands today. Every name in that list has, by definition, survived a decade and still met the index's inclusion criteria at the end of it. The companies that were in the index at the start and were removed after collapsing — the ones that would have produced the worst positions in the run — are simply absent. The equity curve rises more smoothly than the real index did, the maximum drawdown is shallower than the real index's, and the strategy looks as though it beat the market when in fact it was handed a list of winners. Running the same rules on the index's membership as it stood on each historical date produces a materially lower result.

How does Fincanva's data handle delisted companies?

Fincanva's market data includes delisted instruments and point-in-time index membership, so names that later failed or were removed from an index remain available to a backtest rather than disappearing from history. Each instrument is used only across the span in which it actually existed — a company that listed in 2011 and delisted in 2018 is present for those years and absent on either side — which is what makes a run measurable against the market as it stood on each historical date. The widest universe — "All" — therefore still contains instruments that have since delisted, and an instrument that no longer trades can still appear in a historical run.

Data that retains its failures does not by itself make a given test unbiased: the universe and the period you choose still decide which names the test can hold. A hand-typed list of instruments you know today, for example, reintroduces the bias regardless of what the underlying data contains. See the nine biases Fincanva helps you avoid for the product behavior in context.

How is survivorship bias different from selection bias?

Survivorship bias and selection bias both produce an unrepresentative sample, but for different reasons. Survivorship bias is imposed by the data: the failures were already missing before you made a single choice, so it happens even to a careful researcher who picks names at random from the list in front of them. Selection bias is introduced by the researcher: the sample is narrowed by a hindsight-informed choice. Cherry-picking bias is a third, later step — choosing which of your existing results to report. A single test can carry all three at once.

Backtests show what would have happened — not what will. Fincanva provides no financial advice — see Is this financial advice?.