Stats › Long run returns › Survivorship bias
Survivorship bias in market data
There's a good objection to every long run stock return figure you've ever been shown. All of that data comes from markets that are still here. The ones that closed stopped producing numbers, and the people who owned them don't appear in anybody's dataset.
This page asks how big that problem actually is. The answer surprised me the first time I worked through it. The concern is entirely real, the historical cases are as bad as they sound, and the measured effect on the numbers is small. Somewhere between about 10 and 70 basis points a year, depending on which index over which period. Worth knowing, and not the thing that should worry you most.
The bigger issue lives in the same data and gets far less attention. Over 1921 to 1996 the United States was the best performing market out of thirty nine, by a wide margin, and the US is the market whose century long record everybody quotes as the record for equities. That gap is several times larger than any survivorship estimate here. The page also separates three different things people all call survivorship, because conflating them causes most of the confusion in this area. Sources and dates are underneath, and the two main papers disagree in a way that's worth seeing for yourself.
The concern is real and smaller than it sounds. Over 1921 to 1996 the US delivered the highest real return of 39 markets studied at 4.32% against a median of 0.80% for the rest, and 11 of those 39 markets suffered permanent breaks. Investors in Russia after 1917 and China after 1949 experienced a compound return of minus 100%. But the measured effect on the index is modest: 70 basis points on one estimate, 29 basis points on another from the same paper, and about 10 basis points on the long run dataset's own calculation. The bigger problem is not the closed markets. It is that the US was the best performing market of the century and is what everyone quotes.
Members only
The rest of this page is for members
Below this point there are 9 sections, 1 chart and 4 named sources, roughly 2050 words of it. Every figure carries the source it came from and the date the data is from.
You can keep browsing every statistic in the library for free. The intro and the summary are always open.