How to forecast elections
I was reminded to go back to the Iowa election markets (run by the University of Iowa's business school) by a post at Chicago Boyz. The idea is in principle a simple one. For example, they are currently offering an auction on who will control the House and the Senate. They list the possible outcomes: (1) Republican House and Republican Senate; (2) Republican House and Not-Republican Senate; (3) Not-Republican House and Republican Senate; (4) Not-Republican House and Not-Republican Senate. (The not-Republican is a catch-all to cover the in-principle possibility that no party controls one house or the other). The bidding is electronic, but the idea is roughly this. The auction house sells a set of slips, one for each of the four outcomes, for $1. After election day, the slip calling the actual outcome is worth $1; the other slips are worth $0. So you can buy a set, do nothing, and break even (boring). The slips, however, do not need to be kept together; you can sell one or more and keep some. So if you think option (1) will happen, sell the other three. The market's website reports transaction prices for the four outcomes. Since having a slip for each outcome is clearly worth $1, the combined prices for the four slips are driven by arbitrage to equal $1. However, nothing requires an individual slip to have particular value, so the prices fluctuate depending on participants' assessments of the likely outcomes.
Here is why it works well (Hal Varian of UC-Berkeley explained this better than me). Suppose I am keen to have outcome (1) take place, but I haven't a clue what will actually happen. I might well buy slips for outcome (4), which is basically the Democrats control both houses. This is essentially insurance. If the Republicans win, I'm happy. If the Democrats do well, at least I made some money. The problem for the auction is that by buying option (1), I am helping push up its price, even though I haven't a clue what will really happen. So why does the auction predict well (not perfectly: it predicted a more comfortable victory for Bush, but it usually beats the other polls)? Suppose there are lots of people like me buying up option (1), pushing up its price, but in fact anyone with good information knows the Democrats are going to clean up this year [note: this is a hypothetical; don't email me telling me the Democrats are really going to lose]. Because option (1) is being pushed up in price, the other options are falling in price (remember that the prices will sum to $1). If piles of loyal Republicans push down the price of option (4), and an insider with good information knows it is really valuable, the insider buys up option (4) because it is a good way to make money. Option (4) isn't worth much, but it is very likely worth a dollar. Even though there are lots of Republicans in the market in this example, they don't determine the price. The person with good information determines the price because it pays for him to enter, arbitrage, and make money.
As I write this, the combined price of the Republican House slips is 95 cents (versus 5 cents for the alternative), and the combined price for the Republican Senate slips is 52 cents (versus 48 cents for the alternative). The site gives you current prices, and historical price series. For political junkies, it is both fun and useful.