Calculation of weekly P/E ratio

The P/E-ratio is defined as $$\frac{\text{Market value per share}}{\text{Earnings per share (EPS)}}$$ I have weekly observations of stock prices, but what measure should I use for EPS? Should it for example be the EPS of the previous 12 months? What is the typical approach?

References to journal papers computing weekly P/E-ratios are very welcome.

I strongly recommend reading an undergraduate finance textbook like Investments by Bodie, Kane, and Marcus. Your methodology may be limited by your data. For example, using forward P/E requires next fiscal year's EPS estimates. NTM (next twelve months) requires quarterly EPS estimates. If you do not have estimates, the best method is TTM (trailing twelve months). For longer time horizons, you may want to look at CAPE (cyclically adjusted P/E).

• Thank you for your answer. I have EPS for each company on a daily and quarterly basis from 2004-2013 (both years included). Would you recommend the TTM then and if so, how would you calculate the P/E ratio for a given week? – Sunv Sep 9 '14 at 17:59

I think you are confusing the goal with the means. The calculation of the PE is not the goal, the true goal is assessing whether a particular stock is an interesting investment opportunity (cheap) under an investment thesis (set of hypotheses).

Therefore, there is an infinite number of ways to calculate PE ratios, as a results of a set of different assumptions as well as a result of an infinite adjustments deemed necessary by the analyst (removing the contribution of one-off items, for example). This also explains why the definition of PE ratio, as per textbooks, is not really "precise" as you may otherwise expect coming from a more technical background.

It is hard to recommend a PE calculation methodology without knowing your investment thesis.

EDIT

I believe that even under a panel data model framework the choice of PE should be motivate and interpreted through a fundamental approach. Let me make a couple of examples:

• Trailing 12M PE and Last FY: backward looking metrics tend to favor an "as-it-is" approach, ie assuming company bottom line might not change. Does the market "revalues" current bottom lines? What is the pattern there?
• Forward looking PE metrics: How far does the market look? 1-qr, 1-yr and 2-yr growth assumptions might offer different performances, although probably correlated as intuition suggests. @ProbablyPattern suggests the use of CAPE, it is a great hint.
• Sector rotation: different PE metrics might offer different performance in different industries and sectors.

Bottom Line:

1. Different PEs might try to capture different market dynamics. You might even try to combine a few metrics - For example backward looking PEs and EPS growth expectations
2. I guess it is important to understand what the choice of PE implies
• Thank you for your comment. Actually, I'm not operating with an investment hypothesis, but trying to predict stock returns with a number of individual firm variables in a panel data model framework. It is my understanding that the P/E ratio is typically used in portfolio based empirical analysis of stock prediction. I'm considering applying it to my panel data analysis. What are your thoughts based on this new information. Once again thank you for your time. – Sunv Sep 10 '14 at 11:46
• Instead of extending the problem, you were supposed to give an answer here. – emcor Sep 10 '14 at 19:44
• I so disagree with you @emcor. The question was explicitly inquiring about the "typical approach". And, voilà, here is my 50 cents on a "typical approach", based on my experience in the hedge fund industry. Besides, it seems like the op found the answer somehow useful. I see no need of moderation here. – pincopallino Sep 11 '14 at 6:45
• @pincopallino Thank you for your continued interest in my question. From your recent edit, I understand there are many ways of calculating the P/E-ratio depending on what your hypothesis is. Just to be sure, by fundamental approach you mean what exactly? In my empirical analysis I include firm specific characteristics (and no sorting into portfolios). How would you go about calculating a P/E-ratio on a weekly basis with weekly closing prices and weekly EPS observations? – Sunv Sep 11 '14 at 9:24
• @emcor Do you have any alternative suggestions? – Sunv Sep 11 '14 at 9:25

As your model is to predict stock returns via P/E, I suggest you try out all possible P/E's in a backtest and select the one with best forecast ability(lowest forecast error).

• Thank you for the answer. Your answer brings me back to the original reason for asking the question. How would you calculate a P/E-ratio on a weekly basis with weekly observations of closing prices and weekly EPS observations? – Sunv Sep 11 '14 at 9:33
• @Sunv If you have time series of forward or trailing EPS, you can use them directly and see the prediction quality. If the time series does not change weekly, its no problem you just see it as constant until the next change. – emcor Sep 11 '14 at 9:53
• I see. The EPS I have is calculated quarterly by adding the most recent fiscal year end EPS and all subsequent quarterly EPS and subtracting quarterly EPS prior to the most recent four quarters. So this would actually work, wouldn't you say? – Sunv Sep 11 '14 at 10:00
• @Sunv Yes you can do it. You can technically choose any time period (though it only makes sense for when the data actually changes in between). Whether you use quarterly or 10-year EPS for your P/E series is both feasible, but the prediction power would be higher for quarterly EPS I guess, so you can just try it out. – emcor Sep 11 '14 at 11:00