If you plot returns versus P/E10, the curve fit is lousy, but the spread (or scatter) is almost ideal for calculating a standard deviation.

If you plot returns versus the percentage earnings yield 100E10/P, the fit is almost ideal.

This thread is being chronicled as another classic hocusode, but I thought I'd put a JWR mention here.

100E10P? Is that useful in regression analysis? 100 years earnings average? I can't imagine that being relevant.