Depends what you mean by "efficient"

A piece in the new Economist says we shouldn't abandon every aspect of the efficient-markets hypothesis:

[I]t is important not to throw out all the insights of efficient-market enthusiasts. Although it is theoretically possible to make money by outperforming the markets, it is extremely difficult in practice.

The piece is good and worth reading. But the central question -- hold onto EMH or let it go the way of the dinosaurs -- sorta seems like it depends on how you define the theorem. My sense, based mostly on listening to Robert Shiller lecture about it, is that EMH comes, like oatmeal in Goldilocks, in three versions:


The first and weakest version just says that you can't predict the future value of an asset on the basis of past movements. If it goes up today it will not necessarily go up (or down) tomorrow. The middle (semi-strong) version says that asset prices reflect all available public information. If you're reading a story in this morning's Wall Street Journal about some new development at, I dunno, Wal-Mart, you shouldn't think "I must go buy Wal-Mart stock," because the share price of Wal-Mart is supposed to incorporate this information already. The strong version, on the other hand, says that all information -- public and private -- is incorporated into the asset's price.

The first version seems like it's about right. The second version seems like it's in need of some additional modesty. But it's the third version that really seems utterly crazy.