Experiment on election prediction markets August 5, 2009Posted by Sverre in : Methods in political science, Political behavior , trackback
I’ve recently become involved (as a participant) in an interesting experiment performed by PhD student Sveinung Arnesen at the University of Bergen in which we are asked to predict the election result through a market model, buying and selling fictive “shares” in the outcome based on our own evaluations. This is based on prior experiments like Iowa Electronic Markets experiments by the University of Iowa in connection with American Presidential Elections, and the work of Robin Hanson.
Participants have been recruited through the political party organizations (at least I was), and appear to only have the option of buying or selling “stock” in our own party and/or government coalition. I assume part of the reason why we are restricted to our own party is the need for keeping the results secret to avoid incentives for strategic attempts at driving up the predicted value.
Unlike the Iowa experiments, this experiment is not conducted with “real money”. Each participant is given a number of virtual credits, and as far as I know there is no winning prize. This does of course give weak incentives towards predicting results accurately. If the result wasn’t being kept secret until election day, we might even have a serious adverse incentive to manipulate the predicted result to give our party “wind in the sails” towards the election. Such secrecy seems to be maintained, though, also to the extent of not letting us know the predictions made on other parties.
It seems highly likely to me, though, that the weak incentives for “winning” will cause party member participants to systematically overvalue their own party’s election result, especially when unable to also “invest” in other parties. The only alternative to holding “stock” in your own party is not holding stock…. i.e. having inert credits.
Of course it would possibly be likely that such a bias is so systematically distributed that it can be predicted and controlled for trhough discounting by some factor. Still I can’t help but think that there would be a chance for better predictions if we were either allowed to “invest” in all parties or had actual economic incentives to “win”.
In any case it’s interesting to see new methods of prediction getting tested. The Iowa results look rather good.