Survivorship Bias: The Planes, Funds and Stocks You Never See

Abraham Wald and the planes that never came back

In the summer of 1942 a group of mathematicians and economists set up the Statistical Research Group at Columbia University to work on military problems. Its members included Milton Friedman, George Stigler and the statistician Abraham Wald. Around 1943 Wald wrote a series of eight memoranda on a deceptively simple question: how do you estimate where an aircraft is most vulnerable when the only damage you can inspect is on the planes that made it home?

His insight was that the missing planes are the data. If returning aircraft rarely show hits in a certain area, that is not evidence the area is safe. It suggests planes hit there tended not to return. Wald turned that idea into a formal method for estimating the chance of being downed by a hit to each section. The memoranda stayed obscure until the Center for Naval Analyses reprinted them in 1980.

Markets run the same filter every day. Companies fail, funds close, ticker symbols disappear, and what remains on your screen is the population that made it home. Here are five places where that bites investors.

The S&P 500 quietly replaces its losers

The S&P 500 is a list that a committee keeps updating. Companies leave when they are acquired, shrink too much or collapse, and the committee picks replacements. When Enron was dropped from the index during its collapse in 2001, Standard & Poor's gave the slot to Nvidia, then a little-known chip maker. Nvidia's 2002 proxy statement confirms it joined the index in the fiscal year that ended in January 2002.

The index itself is not the problem. While Enron was a member, S&P 500 investors owned it and took the loss; the index's published return includes that hit. In our dataset the S&P 500 price index (dividends excluded) fell 49.1% from 1,527.46 on March 24, 2000 to 776.76 on October 9, 2002, and that decline already includes Enron's collapse. The bias appears when someone takes today's member list and looks backward, as if those 500 names had been the obvious picks all along. Our article on whether you can beat the S&P 500 covers why the index is such a hard benchmark in the first place.

Mutual funds: the graveyard behind the ads

Fund companies rarely advertise the funds they shut down. A fund with a weak record is often liquidated or merged into a sibling, and its history disappears from the family's marketing. S&P Dow Jones Indices built its SPIVA scorecards to fix this: they count every fund that existed at the start of a period, including the ones that later vanished, and use returns from CRSP's survivor-bias-free fund database.

The SPIVA U.S. Mid-Year 2025 scorecard (data to June 30, 2025) shows how big the graveyard is. Of 2,373 U.S. domestic equity funds alive at the start of the 20-year window, only 37.29% still existed at the end. For large-cap funds the survival rate was 34.70%. SPIVA counts only surviving funds that beat their benchmark as winners, so a fund that closed is scored as a loser; on that basis the same report shows 91.03% of large-cap funds trailing the S&P 500 over those 20 years. The scorecard's own glossary says funds usually close because of continued poor performance. The pattern held in the shorter run too: S&P's year-end 2025 scorecard, published in March 2026, found that 79% of active large-cap funds lagged the index in calendar 2025.

If you judged active management only by funds you can still buy today, you would be grading the class after expelling most of the students who failed.

"Best stocks of all time" lists start the clock late

Lists of the greatest stocks are true and still misleading. They are built backward from the winners, so they skip the thousands of companies that looked just as promising at the time. Finance professor Hendrik Bessembinder measured this across every U.S. common stock in the CRSP database since 1926. In a 2018 study he found that most individual stocks earned less over their lifetimes than one-month Treasury bills, and that, measured in lifetime dollar wealth creation, the best-performing 4% of listed companies accounted for the entire net gain of the U.S. stock market since 1926.

Even the winners on those lists went through near-wipeouts. In our price data, Amazon's share price fell 94.4% from its December 1999 peak to its September 2001 low and did not close above the 1999 high again until October 2009. Cisco lost 89.3% from its March 2000 peak to October 2002; even with dividends reinvested it did not get back above its March 2000 peak until August 2021. The dot-com bubble killed many of their peers outright, and those peers are absent from every "greatest stocks" list.

Backtests built on today's winners

A backtest is a simulation of how a strategy would have done in the past. The classic error is to pick the test universe from a current list, say today's S&P 500 members, and run it from 2000. That universe excludes Enron, WorldCom and Lehman Brothers, which were large, popular stocks at the start of the test, and it includes companies that only grew into the index later. Nvidia, for instance, joined the S&P 500 only in 2001, as Enron's replacement. By the end of November 2001 its stock was already worth about 11 times its first-day close in January 1999, gains a current-member backtest would credit to an "S&P 500 stock". The result flatters almost any strategy.

Survival is also not the same as success. Citigroup never went bankrupt, so it sits in every current-member backtest. Yet in our dataset its stock, with dividends counted, fell 98.0% from December 2006 to March 2009 and on October 1, 2026 was still about 66% below that 2006 peak. A backtest that only checks whether a company still exists misses this kind of permanent damage.

Startups and "how I got rich" stories

The founder on the podcast and the investor with the bestselling memoir are survivors by definition. People who followed the same playbook and lost do not get book deals, so the advice you hear is filtered by its outcome. Private markets are worse because the failures are rarely priced in public. WeWork was valued at $47 billion at its peak as a private company and filed for Chapter 11 on November 6, 2023, about two years after it listed on the stock market through a SPAC.

The filter can run the other way too. Lists of famous bankruptcies assume shareholders always end up with nothing. Hertz filed in May 2020, its stock dropped from a first-quarter high of $20.85 to a low of $0.40 by June, and yet the reorganization plan confirmed in June 2021 paid creditors in full in cash and, by the company's count, gave existing shareholders more than $1 billion of value. Cherry-picked disasters distort the picture as much as cherry-picked successes.

Famous collapses that drop out of the charts

How to spot survivorship bias: a checklist

In a chart

• Ask who chose the start date. A stock chart that begins at a crash low or right after an IPO is selling a story.
• Ask about the peers. If the chart shows one company that grew 100-fold since 1997, look up what else was in the same sector in 1997 and how many of them still trade.
• Check what is being measured. Price index or total return, nominal or after inflation. The S&P 500 levels quoted in the news exclude dividends.

In a fund ad

• Look for the denominator. "Our five-star funds beat the market" says nothing about how many funds the family launched, merged or closed.
• Check the inception date and fund history. A fund that absorbed a weaker sibling may report only the stronger record.
• Compare with a survivor-free source. SPIVA scorecards count every fund that existed at the start of the period.

In a backtest

• Universe as of each date. The stock list must be the one investors could actually buy on each historical date, including companies that were later delisted.
• Count the bankruptcies. A 20-year U.S. stock backtest with zero failures in it is almost certainly broken.
• Treat delistings as real losses. If the data simply stops when a stock disappears, the loss on the final leg may be missing.

How to correct for it as an investor

1. Always ask for the full starting population. How many funds, startups or stocks began the race, and how many finished? The success rate is meaningless without the starting count.
2. Respect the skew. Bessembinder's results imply that a small, concentrated portfolio usually misses the handful of stocks that drive the market's long-run gains. His paper links this directly to why poorly diversified active strategies so often lag.
3. Judge decisions with the information available at the time. In 2000, Enron, WorldCom and Cisco were all large, widely held companies. A fair test asks what you would have done then, not what you know now.
4. Separate surviving from winning. Citigroup and Cisco survived, yet investors who bought at their peaks waited many years, or are still waiting, to break even.

Why Blindfolio keeps the failures in the game

A stock simulator that offers only today's listed companies is a survivorship-bias machine. Blindfolio's catalog of about 290 instruments deliberately includes companies that went bankrupt or were delisted, such as Enron, Lehman Brothers, WorldCom, Blockbuster and Pets.com, alongside today's giants. Every company appears under a temporary codename, so you cannot skip the losers by recognizing a name. In a random round the years are hidden as well; if you choose a start decade instead, you see the dates, but the names stay masked. Time runs week by week, you can pause at any moment, and you pick from the market as it looked then. The real names are revealed only at the end, when your portfolio is compared with the S&P 500, dividends reinvested, over the same period.