Key takeaways
- Of the 2,376 US domestic equity funds available to buy at the start of the 20 years to end-2024, only 36.41% were still running at the end.
- Measured on US data from 1962 to 1995, survivorship bias in mean annual fund return grew from 0.07% over one-year samples to 0.94% over 20-year samples.
- Attrition isn't random. Of fourth-quartile US equity funds, 24.89% merged or liquidated within five years, against 6.81% of first-quartile funds.
- Only 41.40% of the 157 UK large- and mid-cap equity funds trading at the start of 2015 were still trading at the end of 2024.
- The best objection cuts the other way: one structural estimate puts a reverse survivorship bias of 76 basis points a year on measured manager alpha.
The fund record you can read belongs to the funds that made it
Look up how active funds have done over 20 years and you're reading a table assembled from the funds that lasted 20 years. Most of them didn't. S&P Dow Jones Indices builds its scorecard from the full opportunity set instead: every fund a person could actually have bought on day one. For the 20 years to 31 December 2024, that set held 2,376 US domestic equity funds. At the end, 36.41% of them were still running.
The other 63.59% were merged into other funds or wound up. Their records didn't move to a footnote. In most published performance tables they simply aren't there, and that absence has a size you can measure.
Mark Carhart, Jennifer Carpenter, Anthony Lynch and David Musto did the measuring. They assembled every known US diversified equity fund from January 1962 to December 1995, then asked what a survivor-only sample says compared with the complete one. Over 20-year windows, the survivor-only sample showed a mean annual group-adjusted return of 1.02%. The complete sample showed 0.08%. The gap is 0.94 percentage points a year.
Group-adjusted return is a fund's return minus the average return of every other fund with the same declared objective. It's a peer measure, so something near zero is what the average fund ought to produce by construction. Cut the dead funds out and that near-zero turns into a full point of annual outperformance that nobody earned.
S&P states the principle it works from directly: "Many funds might be liquidated or merged during a period of study. However, for someone making an investment decision at the beginning of the period, these funds are part of the opportunity set." That is the whole argument in two sentences. The relevant question isn't how the survivors did. It's how the choice you faced did.
Survivorship bias grows with the length of the record, not down
The instinct is that a longer track record is a safer one. For this particular problem it's the other way round. The chart plots the same authors' estimate of survivorship bias against the length of the sample it's measured over, and the line rises the whole way.
Over one-year windows the bias is 0.07% a year, which is nothing. At five years it's 0.37%. At ten years, 0.66%. At 15, 0.85%. At 20, 0.94%. Their full 34-year sample puts it at 1.06%. As they wrote, "For samples of more than 15 years, the hypothesis that survivor bias is 1% per year is not rejected."
The mechanism is the survival rule. A fund doesn't close after one bad quarter. It closes after a run of bad years, which means survival depends on cumulative performance rather than the last print. Each extra year you add to a sample is another year in which the worst cumulative records get cut away, so the surviving group keeps getting more selected the further back you look.
SPIVA's own survivorship column traces the same curve in raw counts. Of US domestic equity funds, 83.76% survived the five years to end-2024. Over ten years the figure was 63.92%, over 15 years 49.49%, and over 20 years 36.41%. Large-cap funds fared worse over the longest window, at 32.95%.
The funds that disappear are the funds that were losing
None of this would matter much if funds closed at random. A random cull removes good and bad records in equal measure and leaves the average where it was. Closure isn't random, and there's a clean way to see it.
The S&P U.S. Persistence Scorecard ranks domestic equity funds into quartiles on their December 2014 to December 2019 record, then follows each quartile for the next five years. Of the fourth-quartile funds, 24.89% were merged or liquidated by December 2024. Of the first-quartile funds, 6.81% were. Among large-cap funds alone the split was 21.43% against 7.14%. Sorted into halves rather than quartiles, 20.34% of the bottom half went, against 9.37% of the top half.
Carhart and his co-authors found the same asymmetry in returns rather than counts. "By these measures, nonsurviving funds underperform survivors by about 4% per year," they reported, and liquidated funds were the worst of the lot. Measured month by month, their survivor-only performance estimate ran 0.08% above the complete-sample estimate.
The Financial Conduct Authority reached the same place from the supervisory side. Its 2017 Asset Management Market Study found little evidence of persistent outperformance, and noted that "worse performing funds were more likely to be closed or merged into better performing funds". That's the machinery of survivorship bias described by a regulator: the losing record doesn't get published for longer, it gets absorbed.
What 0.94 points a year is worth over a 20-year table
An annual figure of 0.94% reads as small, and compounding is what makes it not small. Run it over the 20 years the estimate is measured across and it comes to 20.6% of extra cumulative return, none of which any investor received. That's the difference between a category average that looks like a modest edge over peers and one that looks like the near-zero it has to be, given that peer-relative returns across a whole group sum to about nothing before you count what the funds that closed were doing.
It's also why survivorship bias is a bigger problem for the long-horizon numbers people quote in arguments than for the one-year ones they ignore. The 0.07% bias in a single year is invisible next to a fund's tracking error.
UK funds don't last as long as US funds do
The attrition is faster on this side of the Atlantic. SPIVA Europe counts 157 UK large- and mid-cap equity funds trading at the start of the ten years to end-2024, of which 41.40% survived. The broader UK Equity category started with 356 funds and kept 46.35%. The comparable US large-cap figure over the same ten years was 65.14%. The two scorecards draw on different fund databases, so treat that as a wide gap rather than a precise one.
A UK equity holding picked in 2015 was chosen from a menu of 157 funds, and fewer than half of that menu is still trading. S&P put the running rate plainly: "A cross-category average of 5% of funds in our sample were liquidated or merged in 2024 (see Report 2 and Report 7); a similar rate, if compounded, would result in a survivorship rate of only 60% over 10 years." In 10 of the 31 categories it covers, more than half the funds available at the start never lasted long enough to post a ten-year number at all.
That's what makes a ten-year league table of UK funds a harder document to read than it looks. Roughly half the field isn't in it, and the half that's missing is tilted towards the weaker records. It's also why performance chasing is measured against the full universe when it's measured properly, and why the SPIVA scorecard reports its survivorship column next to its underperformance rates rather than in an appendix.
The strongest objection: for manager skill the bias runs the other way
There's a serious argument that survivorship bias has been pointed at the wrong quantity, and it deserves its best statement rather than a strawman. Juhani Linnainmaa's Reverse Survivorship Bias, drafted in February 2010, makes it. Funds often close after a stretch of bad luck rather than bad management. If a fund's poor run came partly from negative idiosyncratic shocks, then the alpha you estimate from its truncated history understates the alpha it actually had.
Linnainmaa estimated a structural model on CRSP fund data running to September 2009, and put the gap between true and observed mean alpha at 76 basis points a year under a market model. Under a three-factor specification the gap was 85 basis points, and under a four-factor specification 83 basis points. In other words, adding the dead funds back in, which is exactly what a survivorship-bias-free database does, pushes the measured average manager alpha too low by roughly a point a year.
So which is it? Both, because the two effects are answering different questions. Linnainmaa is explicit about the boundary of his own result: the reverse bias "affects the measurement of fund managers' true alphas but does not bias the estimates of the returns available to mutual fund investors". If you're trying to rank human skill, survivor-corrected data is harsh. If you're asking what the average pound in the average fund actually earned, the correction is the number you want, and his own conclusion is that most managers still ran negative net alpha.
What survivor-bias-free data still doesn't fix
The standard remedy is a database that keeps the dead. CRSP's Survivor-Bias-Free US Mutual Fund Database exists for that reason and starts in December 1961. It was built out of Mark Carhart's 1995 dissertation, which was titled "Survivor Bias and Persistence in Mutual Fund Performance", so the fix and the finding come from the same work.
Keeping the dead doesn't clear every selection problem. CRSP's own user guide flags one it hasn't solved: "A selection bias favoring the historical data files of the best past performing private funds that became public does exist." Funds incubated privately can carry their prior record into the public database when they list, and the ones that didn't work out never list at all. That biases the record upward at the point of entry rather than at the point of exit.
The other limits are worth stating plainly. The 0.94 point estimate comes from US diversified equity funds between 1962 and 1995, an industry with a different growth rate, fee structure and closure culture from the one you're buying into now. SPIVA's survivorship columns are counts of funds, not of money, so a category in which only 36.41% of funds survived may have kept far more than 36.41% of its money. The FCA's own analysts flagged the same weakness in their charts, warning they were "likely to be subject to survivorship bias because they condition on funds which existed for the full assessment period". And none of these figures says anything about which individual fund lasts.
What the numbers do support is narrower and firmer. Where a performance table draws only on funds that exist today, its average is high by an amount that grows with the horizon, and a serious source publishes the attrition alongside the returns. That's one of the first things worth checking when reading a fund factsheet, because a factsheet is by definition a document about a survivor.
What would change the conclusion
Three things would move it, and each is observable rather than hypothetical.
The first is the closure rule itself. The whole effect rests on funds dying for performance reasons. If closures came to be driven mainly by consolidation of fund ranges or by platform economics, attrition would decouple from returns and the bias would shrink towards zero even at long horizons. The 24.89% against 6.81% split is the number to watch, and it's republished annually.
The second is the pace. S&P's 5% cross-category liquidation rate for 2024 is what compounds into a 60% ten-year survivorship rate. A sustained fall in that rate would make ten-year tables less selected than they are today, and a rise would make them worse.
The third is the FCA's observation that not all persistently poor funds get closed and that it can take a long time for those that do. If closure got faster, measured survivorship bias would rise, because more weak records would leave the sample. Faster tidying of the fund range and a cleaner-looking industry average are the same event seen from two directions, which is the part that's easy to miss.