David Pennock writes:
http://PredictWiseQ.com is our (beta) prediction contest which aims to estimate not just the marginal probabilities of election outcomes this November, but millions of correlations among outcomes as well, like the chance Obama will win both Ohio and Florida, or the chance Romney will win if the September jobs numbers are negative. It’s a working example of a combinatorial prediction market design we published this summer in the conference ACM EC’12.
And here’s Pennock’s blog, which supplies more background.
It’s hard to see how I could win this game with Nate Silver as my counterparty.
Ken:
I can’t be sure, but when I worked with Nate a few years ago, I found that the correlations of his forecasts did not make complete sense. I recall that the correlation between the forecast vote in any two states was close to a constant and was not generally higher for states that are closer to each other. That might make sense, or maybe not; the point is that it’s not really so important for Nate’s forecast to be just right on these sorts of compound events. If Nate put in the effort he could get a more reasonable error structure but it’s probably not worth his effort.
Perhaps my real message is that there’d be no reason for Nate to want to play this sort of game with his forecasts. Or, maybe I should say that Nate could very well want to play this game but not with the goal of winning money but rather with the goal of using the forecasting market to reveal subtle problems with his forecasts, thus paying a small amount of error to crowdsource his model checking.