The Historical Return You're Backtesting Against Only Includes the Funds That Didn't Die. That's Survivorship Bias
6 min read
Loading...
6 min read
You backtest a strategy against the average mutual fund return over the last ten years and it looks solid. But that average was built almost entirely from funds that were still around to be measured. The ones that performed badly, shrank, and quietly closed or merged into another fund along the way never make it into the number you are looking at. That gap has a name: survivorship bias, and it is large enough to turn a mediocre track record into an impressive one on paper.
Survivorship bias happens whenever a historical dataset only counts the things that lasted long enough to still be there when someone measured them. In investing, this shows up in two common places.
The first is fund databases. A ten-year "average return" for a category of mutual funds is usually calculated using only the funds that still exist today. Funds that got shut down or absorbed into another fund because of years of poor performance simply drop out of the dataset, taking their bad years with them.
The second is stock indices. An index like the S&P 500 only lists the 500 companies that currently qualify. Companies that were removed because they went bankrupt, got delisted, or shrank out of the index are gone from the list, even though they were part of the market's real history.
Either way, the effect is the same direction: a track record built only from survivors looks better than the full, honest history that included the ones that did not make it.
Economist Burton Malkiel measured this directly in his 1995 study "Returns from Investing in Equity Mutual Funds 1971 to 1991," published in The Journal of Finance. Using Lipper data for 1982 to 1991, he compared two groups: funds that survived the entire ten-year period, and every fund that existed in the dataset at any point during those years, including the ones that later closed.
The funds that survived the whole period averaged 17.09% a year. Every fund that actually existed during those years, survivors and non-survivors combined, averaged only 15.69% a year. The gap, about 1.4 percentage points a year, is not a difference in strategy or skill. It is purely the effect of only counting the funds that were still around at the end.
$10,000 at 17.09%/year for 10 years ≈ $48,400 (survivors only) vs $10,000 at 15.69%/year for 10 years ≈ $42,900 (all funds)
That is roughly 13% more ending wealth, generated entirely by which funds got counted, before a single real investor made a single real decision.
This is not a 1990s problem that better recordkeeping has since fixed. S&P Dow Jones Indices publishes a running count of it every year in its SPIVA U.S. Scorecard. The Year-End 2024 edition, released in March 2025, found that over the 20 years ending December 2024, nearly 64% of U.S. domestic stock funds were shut down or folded into another fund. Almost two out of every three funds that existed 20 years ago are gone from a simple, unadjusted "average fund return" calculation today.
This is a different measurement from the fund-persistence data covered in NetWort's earlier piece on why last year's top fund rarely repeats: persistence asks whether a fund stays near the top of its category, while survivorship asks whether the fund is still in the dataset at all to be ranked in the first place. A fund can vanish from a database for the same reason it would have ranked poorly if it had stayed.
Researchers Stephen Brown, William Goetzmann, Roger Ibbotson and Stephen Ross made a related point back in 1992, in a paper in The Review of Financial Studies called "Survivorship Bias in Performance Studies." Using simulations, they showed that some of what looks like genuine fund manager skill, a manager who seems to keep beating the market year after year, can actually be a side effect of a database that already dropped the managers who did not.
Survivorship bias is not only a mutual fund issue. Standard & Poor's own 50th-anniversary review of the S&P 500 in March 2007, reported at the time by PLANSPONSOR, found that only 86 of the original 500 companies from 1957 were still in the index 50 years later, about 17% of the list.
To be fair to the number: staying in the index and going out of business are not the same thing. Most of those 414 departed names left because they were acquired, split apart, or simply grew slower than the rest of the market, not because they collapsed. But the point still holds for anyone backtesting a strategy against "the S&P 500's history": today's index membership list is not the same list an investor would have seen at any earlier point, and a strategy tested only against today's survivors never has to face the companies that fell out along the way.
If you build a stock-picking rule and test it against today's list of well-known, currently-traded companies, you have quietly built survivorship bias into the test itself. Every company in that list already proved it could survive to the present day. A company that failed the same rule ten years ago and later went bankrupt is not in your test data to drag the average down, because it is not around to be included.
The honest fix is a "point-in-time" universe: testing a rule against the full list of stocks or funds that actually existed at the start of the test period, survivors and non-survivors both. That kind of dataset is expensive to build and most free, easy-to-access historical data does not include it, which is exactly why this bias is so common even among people who know it exists.
NetWort's Portfolio Health card computes your IRR, volatility, Sharpe ratio, max drawdown and beta from the real, dated price history of the specific assets you actually hold, not from a curated "best of" list assembled after the fact. That is a meaningfully different starting point from a generic "average fund return" headline, because it is your actual history, not a survivors-only sample.
One honest caveat in the other direction: NetWort's own tracked ticker list, including well-known names like the SPDR S&P 500 ETF Trust, is a curated set of currently active, currently traded symbols. Any backtest you run using only assets you can look up on NetWort or any similar tool today carries the same survivorship caveat this article describes. It is a reason to treat a backtested return as a helpful estimate, never as a guarantee.