Recent talk about a futures exchange for terrorism has popularized the Foresight Exchange. I’m thinking of joining in the game if I can figure out how to play.

Stocks are in claims about future events. You can buy YES stocks at the market price, or you can buy NO stocks at one minus the market price. If the claim happens YES stocks pay out 1, and NO stocks pay out 0. If the claim doesn’t happen NO stocks pay out 1 and YES stocks pay out 0. A few claims have more complex payout schemes. In any case the sum of YES and NO payouts is always 1.

The idea is that the value of the stock represents the probability of the event happening. So for example, at the time of writing the probability of Bush remaining president in 2005 is at 66%

Of course probability this isn’t a perfect interpretation of what the value of the claim is. Two of the more interesting (and highly traded) ‘stocks’ are 2007-01-01 will exist, and 2015-01-01 will exist. Now we are fairly certain that these two dates will exist (and if they don’t there are bigger problems); however currently T2007 is trading at 95% and T2015 is trading at 93%. The basic meaning of this seems to be that some of the claims with due dates before 2007-01-01 are off by 5%. This is clearly a true statement as T2007 is off by 5%, but presumably there are some claims off by a larger amount, otherwise everyone would be buying T2007 at such a discount rate. This error between the actual probabilities of events occurring, and the market value of the claim seems due to the limited resources of the players. If players could get arbitrary interest free loans, the value of T2007 would immediately move to it’s true value of 100%.

But I don’t know how to play this game. I know the claim values are off by at least 5% and probably more. How do I find the mis-valued claims? I was thinking of trying to find correlated claims. It seems likely there is a relationship between the value of Dean wins in 2004 and Bush is president in 2004. So as the value of one changes so should the other. Perhaps I can make a few points between when one changes before the other. Still, it would be nice to identify the mis-valued claims directly.