Please read carefully paragraph 8 and especially Table 3 at this site.
These are rolling 40-year periods. There were, in fact, 89 of them during the period discussed since 1871.
-'Copernicus'
my user name on FR is rimini, but I, 'Copernicus,' am the author of this set of "Social Security and Stocks" articles. Thanks for your vital interest.
P.S. The expansions of the financial--and all--markets are without limit. What appear to be limitations and restrictions are transient episodes of contraction, soon followed by expansions. See Market Timing.
I'm aware that they are rolling. They are not, however, independent. There are at most 3(!) independent samples you can choose. Not a good basis on which to predict the next 40 years, IMO.
And the number 89 is arbitrary. The more willing you are to draw improper conclusions from dependent samples, the more periods you can have. Simply allow finer divisions of the start and end dates, i.e. choose start dates of "March 1, 1900", "March 2, 1900", etc. You could get thousands and thousands of samples from an hourly division of start end dates.
By doing this you can prove that in the past, that the timing of entry is irrelevant, provided your investment horizon is long enough. However the length of that horizon is a non-trivial fraction of the record, and prevents valid conclusions about the future.