No. of Recommendations: 16
"Since 1872, the S&P 500’s annual starting PE (using trailing 12-month earnings) and forward 1-year returns have an R-squared of 0.01—essentially no causality! R-squared using three- and five-year returns is 0.03 and 0.02. That means just 3% and 2% of US stocks’ forward three- and five-year returns, respectively, even possibly stem from PEs," wrote Fisher. "Randomness. Believing the reverse is just dumbness."
R-squared is a measure of the degree to which one thing determines another.
Though I don't have any reason to doubt those numbers, I might suggest that R-squared isn't really the right metric to get much insight. Nor trailing one year P/E. Current P/E is indeed a pretty weak predictor of anything. Price to peak-to-date E is much better, and CAPE better still.
As an alternative view, quintiles of CAPE, 1960-2018, and one year forward US stock returns for dates starting in that valuation quintile:
Most expensive 1/5 of the time: 3.1%
Next 1/5: 7.6%
Next 1/5: 7.8%
Next 1/5: 13.1%
Cheapest 1/5 of the time: 18.7%
A one year forward return is itself a bit of a torture test, as valuation is famously a pretty terrible predictor of stock market prices for time frames under around two years. But that table seems to me pretty darned far from "no causality" or "dumbness" if the thinking is about the general notion of valuation levels mattering to stock returns.
Jim
Data from a study by Columbia Threadneedle Investments done with data to end 2019. I think that for their study they used 7-year CAPE rather than the usual 10-year, but it doesn't make much difference.
