Untitled Document

The Long-Term P/E

The traditional way of calculating the P/E has remained the same for at least 100 years - take the current mid-price of the share and divide it by last year's earnings per share. This gives you an easily-calculated, simple-to-understand statistic that is next to useless in predicting returns.

Why not take more years of earnings into account? This will average out the earnings over years when the market is both good and bad. It will also reduce the effect of managers massaging the earnings figures: they might for example recognise revenue this year rather than next, but for investors viewing earnings over several years, such efforts are self-defeating. Overall you should get a more reliable statistic.

This all sounds eminently sensible, and in fact it was suggested over 60 years ago by the founders of value investment Ben Graham and David Dodd in Security Analysis (1934, p.452). They said you should take into account at least five years, and preferably seven to ten years, of past earnings to gauge a company's true earnings power. Since then, nothing: there has been no academic research into whether taking more years of earnings into account makes a company's P/E more reliable.

My 2006 paper The Long-Term P/E (with Professor Chris Brooks) filled this gap. The results are straightforward: taking into account eight years of past earnings gives you a P/E statistic that is twice as powerful in predicting returns as the traditional P/E. So instead of a 6% gap in average annual returns between low and high P/E portfolios, you get a 12% gap. This alone suddenly makes the P/E much more worthwhile.

This isn't all. Matters can be improved further by decomposing the P/E.