Survivorship bias explained, and why it flatters backtests

Survivorship bias is the error of looking only at the things that lasted and ignoring the ones that did not. In investing, it means judging a strategy, a market or a group of funds using only the companies or funds that still exist today. The ones that failed, merged or closed are missing, so the past looks better than it was.

It is one of the most common reasons a backtest or a performance table looks more impressive than the reality behind it.

How it gets into a backtest

The easiest way to build a stock universe is to take today's list of companies and download their price histories. But today's list is made of survivors. A company that went bust in 2008 or was taken over in 2015 is not on it, so a test that starts in 2005 never sees it fail.

Index membership causes the same problem. Testing "the S&P 500" with today's members means testing companies that grew enough to be in the index now. Many of them were smaller, riskier names twenty years ago, and the ones that fell out of the index along the way are missing from the test.

How it shows up in fund tables

Funds that perform badly are often closed or merged into other funds, and their records can disappear from the tables people look at. A list of funds with ten-year records is, by construction, a list of funds that survived ten years. The average of that list is higher than the average of every fund that was available ten years ago.

Why it matters for a strategy

A rule that buys after large falls looks very good if every stock it bought eventually recovered. In a survivor-only universe, they did, because the ones that never recovered are not in the data. The same rule, run on the full set of companies that existed at the time, would have bought some that went to zero.

So survivorship bias does not just inflate returns a little. It can turn a rule that loses money into one that appears to work, especially rules that buy weakness.

How to reduce it

Use a point-in-time universe: the list of companies as it actually stood on each date, including those later delisted. Use price data that includes delisted companies. And be most sceptical of results from rules that buy fallen stocks, because they are the most exposed.

When the data cannot do that, say so. A clear statement that a test is subject to survivorship bias lets a reader discount the result appropriately.

Our own limits

Our ticker pages test strategies on tickers that are listed and trading today, so they are subject to survivorship bias, and our pages say so. That is one reason we read our results cautiously, and one reason the comparison we show is always against buying and holding the same ticker over the same dates: both sides of that comparison share the same survivor, so the bias does not favour one side over the other in the way it would favour a rule tested against cash.

Even with that shared starting point, most of the strategies we test do not beat holding. Across the 38 published tickers, 2,565 of 21,014 backtests beat holding the same ticker over the same dates, which is 12.2%.

Figures are from the 38 StrategyLens published ticker cards, updated 29 October 2026. They are hypothetical backtests with no trading costs, before tax, and they will change as the cards are refreshed.

Questions to ask of any result

Does the universe include companies that no longer exist? Was index membership taken as of each date, or as of today? Does the rule buy weakness? And is the comparison affected by the same bias, or only the strategy? If the answers are not given, assume the result is flattered.

Other biases that travel with it

Survivorship bias rarely arrives alone. Look-ahead bias uses information that was not available at the time, such as revised company figures or today's sector labels. Selection bias picks the tickers to test after seeing which ones did well. Reporting bias publishes only the tests that worked. Each makes the past look kinder, and they often stack: a survivor-only universe, chosen after the fact, with only the best results shown.

The common thread is that the person running the test knows how the story ended. Good test design tries to put the tester back in the position of someone who did not know, using only the information and the universe that existed on each date.

An everyday version

The same error shows up outside backtests. Stories of investors who held a single share through a huge rise are told because the share rose; the people who held shares that collapsed do not write articles about it. Lists of successful companies that "started in a garage" leave out the far larger number of garage start-ups that failed. Whenever a sample has been filtered by success, its lessons about how to succeed are distorted.

A useful habit is to ask, of any impressive group or record: who is missing from this list, and what happened to them?

This is research, not financial advice. Past results, real or simulated, do not tell you what will happen next.

What is survivorship bias in investing?

Judging past results using only the companies or funds that still exist, which leaves out the failures and makes the past look better than it was.

Which strategies are most affected?

Rules that buy after large falls, because in a survivor-only universe every stock they bought appears to have recovered.

How do you avoid survivorship bias in a backtest?

Use a point-in-time list of companies, including those later delisted, and price data that covers them. Where that is not possible, state the limitation.