# P/E forecasting, from Steve Ellison

January 3, 2011 |

You can get as-reported earnings for the S&P 500 from 1988 on at Standard & Poor's website.

Using 12-month trailing earnings for each year's September quarter (the last that would be known by Dec. 31) to calculate an E/P ratio for the S&P 500 as of Dec. 31, I get a somewhat positive correlation of trailing E/P to year-ahead returns with t=1.10, R sq=0.057, p=0.29, and N=22.

## Larry Williams writes:

My model for the DOW suggests a 12.25% growth for the year, slightly above the long term average growth.

For the S&P, I get 10.6 % barely above the long term average.

## Bruno Ombreux writes:

There are two things I don't like in P/E or E/P studies.

1) Your regression is in the form:

-1 + P(t)/P(t-1) = f[P(t-n)/E(t-n)] + e we have the same variable on both sides, and even if it is lagged I am not sure standard regression is OK to handle this type of formula. Just to give you an idea, multiplying both sides by P(t-1), it is actually P(t) = P(t-1) + P(t-1)* f[P(t-n)/E(t-n)] + P(t-1)*e

This is certainly amenable to study, but not with the standard regression toolbox.

2) Price is more volatile than earnings. There is a subtle bias introduced by the fact that over the estimation sample, high P/E will be naturally followed by lower P/E, and vice-versa. This is a bit like regression to the mean but more subtle. This can lead to spurious mean-reversion.

The issue is not really the dependent variable. It is using the Shiller variable with its serial correlation. One way to use the Shiller variable would be to take every tenth month. That might work but you would have one tenth the data. You still might have the Holbrook Working flaw because of the averaging. The averaging also leads to the Slutsky-Yule effect which creates spurious sinusoidal artifacts in the adjusted variable when no such sinusoidal effect is actually present in the original data..

`SELECT * FROM wp_comments WHERE comment_post_ID = '5710' AND comment_approved = '1' ORDER BY comment_date`