Decomposing the P/E

This sounds (and in fact is) quite a complicated procedure, but the idea behind it is very simple. Take any company and ask yourself, "What is the P/E I would expect this company to have?" If the P/E the company actually has is less than you would expect, this must be a good sign.

It's well known that the overall P/E of the market goes up and down over the years. Certain sectors such as technology stocks traditionally have higher P/E's than, say, the Water sector. And finally, large companies generally have higher P/E's than smaller companies, even in the same sector. For the decomposed P/E [link to academic paper] I extracted each of these factors from the original P/E, ran a regression and weighted them according to how powerful they are in predicting returns.

Generally of course the higher the P/E the worse the future returns, but this isn't always the case. High-growth, high P/E sectors actually do perform slightly better over the years than low-growth, low P/E sectors. I take this factor out and reverse it. Up to now it has been working against us, making the P/E less powerful at predicting returns: now it is working for us.

The results are startling: using eight years of past earnings and decomposing the P/E gives a new statistic more than three times as powerful as the traditional P/E. There is a difference of over 20% between returns on portfolios of low decomposed P/E shares and high decomposed P/E. The one-tenth of shares with the lowest P/E would have returned 32% average per year since 1975 (mid-mid, annual rebalancing).

Fund managers who think that this sounds interesting should look here. Private investors who want to supercharge their returns and don't mind taking on some extra risk in a concentrated portfolio can go on to look at the naked P/E.