The single biggest trap in almost any home-made stock backtest — including, partly, this site's own.
Survivorship bias happens when a backtest's universe only includes companies that still exist today, silently leaving out the ones that went bankrupt or were delisted along the way. Because failures vanish from the dataset instead of counting as losses, returns look better than what any real investor could actually have earned.
Reconstructing exactly which ~3,000 companies were the largest U.S. stocks at each past date (including ones that no longer exist) needs a point-in-time index-membership history. The official source for that (Russell 3000 fund holdings) blocks automated downloads, and there's no free equivalent — this is documented in detail on this site's own backtest page.
Don't take the absolute return number at face value. A comparison against a benchmark that shares the exact same blind spot (this site uses an equal-weight average of the same survivorship-biased universe) is more trustworthy than either number alone, because the bias mostly cancels out of the difference between them.
This site's own backtest carries this exact limitation and says so, prominently, on the results page — a backtest that doesn't mention this issue isn't necessarily better, it may just not have checked.
What is CAGR, and how should you read one from a backtest? What does "equal-weight" mean in a backtest?
Not investment advice. ← All investing terms · Screen stocks yourself →