"FWIW, I used to do that a lot, looking at the correlation between past returns and initial starting value and building nice fits. Ultimately I decided that there was a different way to come up with a reasonable expectation for forward returns that was both more reliable AND vastly simpler to do."

Thank you, Jim. I also use the method you describe. It works quite well, especially for forecasting returns over longer time periods than 2-3 years.

"This overshoot effect might happen again, but more to the point, it might not. A far more conservative and reasonable assumption would be that the end point is merely a normal valuation level, not super cheap."

With Berkshire repurchasing shares when they're cheap, the overshoot on the cheap side might be more mooted in the future. However, as you know, since Dec '99 the overshoot has been substantial, with P/B ranging from 1.0 to 2.0.