Comments on: Shiller P/E Ratios and 10Yr Annualized Real Returns

By: the social scientist

the social scientist — Sat, 12 Jun 2010 07:41:26 +0000

The number of points in the scatter, and the cluster at a PE greater than 35, suggest to me that the unit of observation is months, or perhaps quarters. This creates an auto-correlation problem in your regression. What does this scatter, and equation, look like if your unit of observation is calendar year? you will still have an auto-correlation problem, but it only invalidates the statistical significance test. You can probably control for the auto-correlation with an auto-regression multivariate model.

By: the social scientist

the social scientist — Sat, 12 Jun 2010 02:41:26 +0000

By: dave.d

dave.d — Fri, 11 Jun 2010 04:16:48 +0000

He's building on Shiller's work which uses PE10. Stocks that went bust are tracked however that is done in the S&P 500, or whatever Shiller's data set uses for periods before the S&P existed.

By: Biscosc

Biscosc — Thu, 10 Jun 2010 18:15:38 +0000

Great work as always! How about doing the same analysis for a 5 year average P/E? Leuthold uses something like that rather than the 10 year he used to use. I'm not sure why he switched, maybe 5 year has better predictive power?

By: OJFrant

OJFrant — Thu, 10 Jun 2010 12:44:58 +0000

where are the stocks which went bust?
and how do you account for stocks which get taken over during the ten year period?