In principle a prediction market should generate more accurate predictions than a simple poll.  For example, in an election, the outcome of a poll should be known to the traders and incorporated into their trades.  But in practice, the advantage of prediction markets is small, or so suggests a new study.

In a new study, Daniel Reeves, Duncan Watts, Dave Pennock and I compare the performance of prediction markets to conventional means of forecasting, namely polls and statistical models. Examining thousands of sporting and movie events, we find that the relative advantage of prediction markets is remarkably small. For example, the Las Vegas market for professional football is only 3% more accurate in predicting final game scores than a simple, three parameter statistical model, and the market is only 1% better than a poll of football enthusiasts.

More here. My view is that there is no theoretical reason for interpreting a market price in a prediction market as a probability. That is, if Coakley is trading at 75 cents there is no reason to interpret that as a 75 percent probability that Coakley will win.  At best there is an ordinal relationship:  a higher price means a higher probability.  Likewise with a poll.

Studies like these should instead be measuring the conditional probability of an event given the price observed in the market.  Because, even an “innacurate” prediction can be a very good one.  To take an extreme case the market might always underprice the probability of a Coakley win by 25%.  But then by dividing the market price by .75 we can infer the right probability with perfect accuracy.