SDS Quant Research

Survivorship bias in backtests

2026-11-21 · Learn

Survivorship bias is what happens when a test includes only the things that survived. In backtesting, it usually means using today's list of index members to test a strategy over the past.

How it creeps in

Suppose you want to test a strategy on the S&P 500 from 2000. The easy way is to download the current list of 500 companies and their price history. But that list is made of companies that did well enough to be in the index today. The firms that were removed, taken over cheaply or went bankrupt along the way are missing.

Your test universe has been quietly filtered by the future. A strategy that buys falling stocks, for example, will look much safer than it was, because the stocks that kept falling are not in the data.

Why it matters

The bias is not random noise. It pushes results in one direction, upward, and it is largest for strategies that buy weakness, hold small or volatile companies, or run for many years.

How to avoid it

  1. Use point-in-time membership. On each simulated date, the tradable universe should be the companies that were actually in the index on that date.
  2. Include delisted securities. Their prices, including the final collapse, must be in the data.
  3. Handle corporate actions correctly. Splits, mergers and dividends change what a holding was worth.
  4. Check your data against a published benchmark. If your data's returns for a broad index differ sharply from published figures, something is missing.

What we do

Our methodology uses historical index membership, so each simulated date is limited to what could have been chosen then. It is one of three checks, alongside look-ahead tests and a sealed final test.

This article is educational and is not investment advice.