1926-1998, 1922-2007, 1926-2009, 1929-2008. Four periods of time. The shortest is 73 years long, and they have 70 years of overlap. And yet the average return of the US stock market over those four periods of time
varied from 9% to 11%, depending on just a few years' change in endpoints.
We shouldn't say "the historic return of the US stock market was 10%," we should say "as best we can tell, it was somewhere in the range 10%±1%."
If someone says to me "portfolio X had a 9% return over about eighty years, while portfolio Y had an 11% return over about eighty," my reaction is "I'm not impressed, because there's that
much difference in the eighty-year return of the stock market itself, with just a small difference in endpoints."
Now, making a boatload of assumptions that we know are invalid... but just to get an idea... if the CAGR of the stock market is uncertain by ±1% over eighty years, then if you cut the time period by a factor of 4, you might expect the uncertainty to increase by √4, i.e. you might expect it to be double.
In other words, if, over a twenty-year period, portfolios X and Y differ in CAGR by as much as 4%, I'm not ready to be convinced. Even that big a difference could
just be the normal noise and uncertainty of financial data.
Now, for your 3-year rolling returns, well, we ought to multiply again by √(80/3) = about 5. I say that if CAGR over 80 years is only good to ±1%, then CAGR over 3 years is only good to about ±5%. In other words, in your chart, I'm not even sure that we're really looking at leapfrogging.
We might just be looking at the ordinary variability of financial data.
I know this is heresy, because if you say "well, the difference between a CAGR of 9.88% and one of 9.45% isn't much," someone is bound to say "it's huge of you compound it out for 30 years." Well, it would be if you knew those numbers were correct in the first place (and wouldn't flip around or do handsprings if you slipped the endpoints a few years), and if you knew that the same difference in CAGR were all but guaranteed for the next thirty years. But it isn't. Heck, we can't even "predict" the past, to better than ±1%
Annual income twenty pounds, annual expenditure nineteen nineteen and six, result happiness; Annual income twenty pounds, annual expenditure twenty pounds ought and six, result misery.