That measure is the cyclically adjusted price/earnings ratio, or caper, a measure compiled by Yale University's Robert Shiller. The logic for using this is compelling. Conventional PE ratios tend to be counter-cyclical. They can be very high in recessions when earnings are depressed, and low in booms when earnings are unsustainably high. But it would be silly to buy stocks at the peak of a boom and sell in a recession. The caper solves this problem by dividing current share prices not by recent earnings, but by average earnings over the past 10 years.
This reveals a problem. The caper is now 23.8, which is 44 per cent higher than the average since data began in 1881.
It seems the US market is overvalued.
Not necessarily. This 23.8 is actually a little below the post-1990 average, of 25.3, suggesting the market is slightly undervalued.
Whether you believe the S&P 500 is expensive or not therefore depends on which sample you think is most relevant - the post-1881 one, or the post-1990 one.
There is a case for preferring the longer sample. The post-1990 sample gives greater weight to the tech bubble of the late 1990s and the complacency in 2006-08 about the impending crisis. It thus oversamples periods of overpricing. This is clearly misleading. The fact that shares are now 'cheap' compared with an average that is inflated by the tech bubble is no reason to buy. The post-1881 sample avoids this bias.
However, there are two reasons to favour the message of the post-1990 sample. One is that the caper is now biased upwards because earnings collapsed in the 2008-09 recession. But it's not obvious why anyone would want to sell because earnings were low four years ago. Using the post-1990 sample avoids this counterintuitive message.
Secondly, the link between valuations and subsequent returns is much stronger in post-1990 data than it is in the longer sample. For example, since 1881 the correlation between the caper and subsequent three-yearly real changes in the S&P 500 has been -0.26. But since 1990 it has been -0.51; much the same is true for one- and five-year changes.
This gives us a dilemma. If you want to argue that the market is expensive now, you have to use the post-1881 sample. But this also tells us that valuations don't matter much; that -0.26 correlation means there's a decent chance (albeit less than 50:50) that an expensive market can get more expensive. Conversely, if you want to contend that valuations matter a lot, you must use the post-1990 data - but this tells us the market isn't expensive now.
You might think all this is horribly inconclusive. That's my point. Whether the market is overvalued or not is a matter of judgment, not fact. Valuation is nothing like a precise guide to future returns, and talk of value can sometimes give us a merely illusory sense of certainty which causes us to underestimate the extent to which the market is genuinely predictable.
MORE FROM CHRIS DILLOW...
Read more of Chris's comment pieces.
Chris blogs at http://stumblingandmumbling.typepad.com
