Rodc wrote:Looking at the full history of data, P/E10 shows about the same thing: it predicts (more technically correlates) with returns best out around 20 years, much smaller correlation at 10 years, very little correlation for less than 10 years. (note really this is only a handful of independent points so could be noise driven). So this very slow mean reversion (if it exists at all, could be noise) is nothing new to the past 20 years: it is standard market behavior.
When PE10 is high, a mean reversion is expected. Prices will crash, but when? Maybe not next year, but probably in the next 10 years. This is seen in regressions using Shiller's data. PE10 does not predict next year's return. PE10 does a much better job predicting 10 year returns.
So, use PE10 to decide 10 year investments: either 10 year bonds or stocks. Hold for 10 years then repeat. Evaluate the portfolio using 10 year returns.
You should never look at investments in isolation. Stocks may be high, and at the same time bonds are even worse. You have to operate at the portfolio level, not the asset class in isolation level.
P/E10 can be high because E10 is low or because P is high.
Please look carefully at the data GenghisKhan provided and analyze it.
Right now P/E10 is as high as it is in part because E was low 8-10 year ago. And bonds are also very highly priced. Not at all clear how this turns out.
Also, even if a high P/E10 predicted a crash (in general it does not, rather it correlates with lower long term total return), you can't really expect to make short term moves (ie new asset allocations every few years as valuations shift) based on 20-year correlations (where P/E10 is most strongly correlated with returns). One could do as you suggest, though the 10-year correlations are not nearly as strong as the 20-year correlations. But I'm going to reevaluate my financial life, how is my job doing, how are my investments doing, refine my goals, assess changes in needs and desires more often than every 10 years and that will drive my investing more than P/E10.
As for regressions, there are only about three times when P/E10 was say above 25 or so (eyeballing the graphs from shiller's website) and only three periods in single digits. Not to mention the strongest correlation is out 20-years and we only have 7 independent 20-year periods, assuming data from the 1800s is even useful today. Not a meaningful amount of data to regress.
Valuations matter. But if you want to use them you better look at and compare the valuations across asset classes, you have to understand the metrics and not look at them as simple single numbers, you should understand their limitations and understand the amount of noise in the system.
I don't think simplistic use of P/E10 in isolation it going to be terribly useful. IMHO.
We live a world with knowledge of the future markets has less than one significant figure. And people will still and always demand answers to three significant digits.