I am a fan of electronic markets. I find that I turn to them more and more, whether the subject is the likelihood of an FOMC rate increase in November, or where the S&P 500 will end the year, or the outcome of the U.S. presidential election. Speaking of the latter, it will be interesting to see how accurate said market is given what it currently says about the converging fortunes of Mssrs Bush and Kerry.

Mind you, economist Charles Manski at Northwestern has an interesting working paper wherein he argues fairly convincingly that electronic prediction markets don’t work the way we think they do. More specifically, he says that all-or-nothing markets — where the Bush line in the above case goes to 1 if Bush is elected, and to zero if Kerry is elected — don’t actually measure probabilities, per se.

While the explanation is somewhat mathematical, the gist is that price is the midpoint of an interval that *contains* traders’ best guess of an event’s likelihood. In other words, price is better thought of a measure of the central tendency of beliefs, not as the event probability. The upshot is that when price is above 0.5 most traders have beliefs higher than the price, and when price is below 0.5 most traders have beliefs lower than the price.

It is an interesting argument. And it fits neatly into another semi-anomaly in pari-mutuel betting, the so-called “favorite-longshot” bias. The gist: Horses with short odds (favorites) win more frequently than the odds indicate; horses with long odds (longshots) win less frequently. In other words, the prices are not really probabilities, they are more a measure of central tendency.

Of course, given how neatly split the U.S. is in the current election, with both candidates sitting at (or near) 50%, the ironic implication is that in this unusual case price in the presidential electronic markets is probability: It’s 50/50.

Related posts:

The favorites are underbet to show. Therefore you get situations where you can make more money betting a favorite to show than to place.