Statistical Bias5 分で読めます

The Reality of Survivorship Bias in Funds

When we see a number like 'average fund return over the past 20 years,' we believe it's the record of every fund that survived those 20 years. But if that statistic left out all the funds that 'went under and disappeared,' would that average be the truth?

Vanished Funds Don't Stay in the Statistics

Funds aren't forever. When performance is poor or assets don't gather, the management company liquidates (closes) the fund or merges it into another fund. The problem is that funds that disappear this way are also erased from the 'past performance statistics.'

If you gather only the funds 'alive today' and average their past performance, funds that closed midway due to poor results were never counted in the first place. What remains are only the passable survivors. So the average looks better than reality. This is fund survivorship bias.

It's the same principle as a class average coming out high after the poor-performing students have transferred away. That average is not the ability of the whole class, but only the ability of 'the remaining students.'

Survivorship bias is a type of the broader concept of 'selection bias.' Because the sample (surviving funds) fails to represent the population (all funds that existed in the past), the conclusion is distorted.

How Many Funds Disappear?

Far more than you'd think. U.S. data show that over a long period of about 20 years, more than half of equity funds were liquidated or vanished through merger. One tally reported that, as of the end of 2024, about 64% of U.S. equity funds over the past 20 years had disappeared, while another category (small-cap funds) showed a 20-year survival rate of around 50%.

On an annual basis, an average of about 100 U.S. equity funds are liquidated or merged each year, roughly 5% of the funds that existed at the start of that year. In 2009, when the market was severely shaken, 275 disappeared in a single year.

In other words, the 'list of surviving funds' we see today is a little over half of the funds that stood at the starting line. The report cards of the other half were quietly dropped from the statistics.

The exact disappearance rate is reported differently, roughly in the 50%–64% range, depending on the period, fund classification, and data source. Here it's safer to understand it as the scale of 'around half disappear' rather than as one specific number.

So How Much Are Returns Inflated?

Academic studies estimate that survivorship bias inflates the average return by roughly a fraction of a percentage point up to 1–2 percentage points per year. The numbers differ by source because the periods and methods studied differ.

Notably, Elton, Gruber & Blake (1996) estimated the bias at about 1.4 percentage points per year in mutual fund data, and Brown, Goetzmann & Ross (1995) put it at around 1–2% per year. The work of Carhart, Carpenter, Lynch & Musto showed more precisely that in short 1-year samples the bias is a small 0.17 percentage points per year, but in long samples of 15 years or more it grows to about 1 percentage point per year. That is, the longer the period, the more vanished funds accumulate and the larger the bias.

1 percentage point per year may look small, but compounded over 20–30 years it creates a difference of tens of percent in the final amount. This is why it's hard to take at face value the ad line that 'even the average fund earned this much.'

Estimates of the inflation size range from 0.17 percentage points per year (short term) to 1–2 percentage points (long term), depending on the source. This very disagreement shows that 'the size of the bias varies by period and sample.'

Honest Comparison Counts the Vanished Funds Too

A representative resource that handles this problem honestly is S&P's SPIVA Scorecard. When comparing active funds against an index, SPIVA includes funds that vanished through liquidation or merger during the period. Because it doesn't count only the surviving funds, survivorship bias is corrected.

The result is sobering. According to SPIVA data as of the end of 2024, over 15 years about 89.5% of U.S. large-cap active funds failed to beat the S&P 500 index. Had they excluded the vanished funds and counted only the surviving top performers, this number would have looked far better.

This is also why The Return of Almost Everything does not boast a particular fund's 'surviving record,' but instead tries to show the long-term performance of indices and real assets exactly as it was — together with maximum drawdown and loss periods. Knowing what was left out of the statistics is the first step in reading numbers.

Active-fund underperformance is not this site trying to condemn active investing, but a teaching example of how carefully the statistic 'average fund return' must be read.

よくある質問

Q. Are survivorship bias and 'fund survivorship bias' different?

The basic principle is the same. Survivorship bias means 'the error of judging by looking only at what survived,' and fund survivorship bias applies that principle to fund statistics. It refers to the phenomenon where funds that vanished through liquidation or merger drop out of the past-performance statistics, so that the average from the remaining funds looks better than reality.

Q. How can I tell whether data is survivorship-bias-corrected?

Check the data's methodology description. If it states 'funds that disappeared during the period were included' or 'a survivor-bias-free database was used,' it's corrected. Conversely, an average based only on funds currently alive — like 'funds currently on sale' — likely still has survivorship bias remaining.

Q. So should I not trust past fund return statistics?

You don't need to distrust them unconditionally, but you must also ask 'what was left out.' Especially the longer the averaging period, the more funds have vanished and the larger the bias. The 'average fund return' in an ad may be weighted toward surviving funds, so it's safer to build the habit of comparing it against data that counts the vanished funds too (e.g., SPIVA).