Four years ago, I grabbed some numbers off Intrade in the days leading up to the U.S. presidential election, ran some smoothing functions on them, and discovered a strong correlation with the vote in each of the fifty states (R-squared value of 0.77, and every state 'called' perfectly hours, even days before the actual voting).
This year several things were different, including no incumbent candidate, larger margin -- and expected margin -- of victory, new dimensions around polling and voting blocks that seemed hard to predict, tumultuous external events, etc.). Nonetheless, the state-by-state markets on Intrade coughed up predictions that were almost as good as 2004. Using my extremely unscientific sampling method, the state-by-state markets produced an R-squared value of 0.72 for data taken around Monday Noon).
In 2004, the markets were (as I recall) run as mirror images of one another. If you were going long on Bush, you were, by definition shorting Kerry, buying (at least in theory) from someone who'd lost confidence in or perhaps wanted to take profits and reduce losses in the Kerry contract at the market price (even if, perhaps, said trader still hadn't lost confidence in the candidate himself -- an entirely separate question we're not even going to approach).
This year, by contrast, Intrade ran separate markets for each state (plus DC) in: McCain (i.e., to win a particular state), Obama, and 'Other', or 153 markets in all. That's a lot on which to maintain liquidity, but on the swing states (e.g., Missouri, Indiana, North Carolina, Virginia, etc.) the markets appeared to be quite deep indeed.
Now, here''s the interesting part...
As the chart below depicts, the McCain and Obama markets as of Monday Noon (12:45PM EST, actually) -- roughly 18 hours before the polls opened -- were both trading at what ended up being an R-squared value of 0.72* (the square of the correlation coefficient, shaking out poor correlations from good ones) relative to the near-final results I took off CNN yesterday (Thursday). *Actually, they differed by two one-thousandths -- hardly significant.
As of Tuesday (Election Day) morning however, the correlation on the Obama market had dropped significantly (to 0.59). It dropped further through Noon on Tuesday then bottomed out. By contrast, the correlation of the McCain markets to the final results (again, state-by-state, not overall) maintained a relatively high correlation throughout the day.
Did Obama market participants lose interest? Were some of them trading on a non-rational basis for other (perhaps partisan, utilitarian or sentimental) reasons? I'm at a loss to explain the difference, since nothing of any note happened to the 'Other' markets (all except Wyoming). They traded -- if they trade at all -- with paper-think liquidity at values of $0.10 on a contract that paid $100.00. (The contract for a Wyoming third party victory was trading at $1.00)