In today’-s soapbox:
Yet it turns out that in 2002, IEM markets indicated that Republicans would lose the house. Which we now know is wrong. They were pretty consistent in their 2004 presidential prediction, however. I recall that Surowiecki discussed the fact that the predictive markets for congressional elections tended to be less accurate than the presidential ones–-because fewer people got involved in the betting.
Wrong. Volumes are OK. If I remember well, the polls were wrong in 2002, so don’-t look any further for the cause of failure of those IEM prediction markets.
Paper That Documents This 2002 IEM Debacle: Iowa Electronic Markets – (PDF) – by Paul Gomme – 2003-04-15
Until September, Republican control of the House was seen as a 50–50 proposition, while their control of the Senate received a probability of around 20 percent. In October, the likelihood of a Republican-controlled House fluctuated between 65 percent and 90 percent while the likelihood of a Republican Senate fluctuated around 40 percent. It was not until election day results came in that market participants locked in on the eventual outcome: Republican control of both the House and Senate. Of course, this outcome was generally a surprise: Neither pollsters nor political commentators called the Republican win in the Senate.
I Disagree With This Paragraph Of The Same Paper:
Better than Gallup? The IEM political markets have a couple of advantages over their closest “competitor,” the public opinion poll. One advantage is that data from the IEM are available virtually instantaneously and almost continuously. Results from polls are typically several days old when they are reported and are taken at discrete intervals. Consequently, data from the IEM are more amenable to studying events like the untimely death of a Senate candidate. A further advantage of the IEM is that contracts can be written based on intrinsically interesting events, such as who controls the House or Senate. Poll results require more massaging to answer such questions.
My Take: Give the polls a break, and quit saying that our predictive market-generated technology is a “-competitor”- to the polls. Without polls, the political prediction markets could return to the locker room. Plus, prediction markets aggregate probably [see comment] more than just the polls.
Hint: A public roasting of David Perry over his impersonating of George Gallup is in the tube and will be published soon on this blog.
Overstatement and understatement in turn…
I bet there are many contracts at TS and WSX for which there is no good polling data. I bet those will be fairly predictive. Less so than those for which corresponding polls exist, but hard to tell whether that is due to polling data informing the markets or the same races attracting both polls and thicker markets. Someone should do some analysis.
Probably? Including private polling.
Point #1. I don’t agree. I think that when the polls are wrong, the political prediction markets are always wrong. The 2004 Alaska prediction market was inaccurate because the polls were wrong, am I correct?
Point #2. You’re right. I will edit my post and suppress the word “probably”.
I don’t know about Alaska, other than that it is the only US Senate seat TS got wrong in 2004. If polls also got it wrong does that imply causality?
Another thing to analyze would be how strongly markets react to new polling data over many instances.
But my conjecture was that there are many races with no or poor (stale or conducted in a biased fashion) polling data. Are markets worthless for those races? That’s what I took “without polls…” to mean.