What do IEM prices tell us?

The Keynes analogy is the correct one. Participants on the Iowa Electronic Markets (IEM) essentially are predicting the behaviour of others (the beauty contest), based on proxy information. The question you seem to be raising is, what information are they using? And I would add another question: what does the market price actually tell us? So, for the first question. I think we can both assume that the market players are using sources of information that we would expect: poll numbers of all kinds (tracking likely, registered, right/wrong track, approval) and subjective analysis, like observing debate performances. The so-called 'students' of elections would be expected to consider historical trends in presidential races as well. Such sources include Yale prof Ray Fair's economic model (which shows Bush dominating in November) and data on historical swing voter movements. I tried to do some statistical analysis of what you identify as a 'lag' factor between IEM prices and released polls, but since I couldn't find enough poll data from before October 1, I couldn't do anything meaningful. At first glance, the numbers seem to bear out the claim that market players are being reactive to the polls rather than proactive, but to resolve that, we have to demonstrate a correlation between the lagged market price and polling data. As for the second question, it's not entirely clear what the market prices actually mean. There's a paper (on PDF) by Northwestern's Charles Manski on this very subject. It's pretty, well, economics-y, but I'll draw out the main question Manski seeks to answer, and briefly describe his conclusion. He writes:

What is the logical basis for interpreting the price of an all-or-nothing futures contract as a market probability that the event will occur? Researchers engaged in empirical study of prediction markets have been uncomfortably vague. Forsythe et al. (1992) and Berg, Nelson, and Rietz (2003) refer to Hayek (1945), who argued broadly that market prices aggregate information. Leigh, Wolfers, and Zitzewitz (2003) write (p. 2): “Markets aggregate opinions and, by requiring a trader to ‘put your money where your mouth is,’ theylessen the cheap-talk problem and create incentives for individuals to reveal their true beliefs.” These and other recent papers on prediction markets provide no formal analysis showing how such markets aggregate information or opinions.

In other words, the reason the IEM seem so seductive as a predictor is because they tell us that people are willing to put they money where their mouths are, and that as a result their opinions have some greater validity. But the price on the market is not a straight-up consensus that either candidate has a certain probability of gaining a majority in the popular vote (ie, if Bush is selling at .70, 70 percent). It is, Manski argues, just an average of market beliefs weighed by the amount of money each trader is prepared to bet. In other words, there might be only one guy who thinks the right/wrong track is the most important predictor of the eventual popular vote winner, and he bets heavily on that outcome. This does not mean that most people believe that Bush, say, will win. Unlike an efficient market, there is no consensus about what should make up the price of the contract: some people use different kinds of information (compare this to, say, the stock market, where market prices are determined by some pretty commonly available information, such as corporate earnings, etc). This conclusion seems to make sense, and gets at your point about the beauty contest -- everybody uses different methods of predicting who other people will vote for. In the IEM case, though, the price does not necessarily reflect the beliefs of all market players. I don't think there is any systematic way of determining the volumes bought and sold by individual traders, so we'll just leave it at that. Also, MacDuff, consider that even 'students' of past elections have no idea how this one will end up. We may believe that without winning Ohio, Bush cannot win the electoral college vote, because it has never been done before -- but there are several scenarios in which it can happen. Likewise, we may say that the Ray Fair model has never before failed in picking the president, but this year, it looks like he may be wrong if current opinion poll trends hold. The truth is, this is a highly unusual presidential election cycle, for reasons that are obvious -- tax cuts during a time of war, the spectre of an invisible terrorist enemy, job losses not seen since the Depression, and other boons and anxieties. I can't predict with any certainty which side will win, and anybody who says they can -- whether their money is riding on it or not -- cannot be any more certain. There is another thing to consider: what the IEM actually trades are contracts that either candidate will win the popular vote. In other words, the candidate could win the popular vote and yet lose the election. I tentatively predicted last April that Bush would win the popular vote and lose the electoral vote. At this point, I'd have to crunch the numbers to see if that's even possible, let alone likely. But it would be an interesting, and somewhat ironic, conclusion to the race.