Wednesday 7 October 2009

iPredict's forecasting performance: awesome [updated]

Matt has been doing great work over at iPredict, New Zealand's event stock market. His latest update: iPredict's rather good forecasting accuracy since it's opened. On binary contracts (events that either happen or don't), folks far too often in the iPredict forums, or even the most bombastic prediction markets blogger, say that the market has failed if its price close to closing is far away from the final price. Not so. If a stock is trading at $0.90 for a $1.00 contract, if that kind of contract doesn't close at zero 10% of the time, the market is seriously wrong. So we want prices to follow a 45 degree line. Here's the picture.

A bit of underpricing around the 30-60% range, but on the whole, things are looking very good. Matt goes through a bunch of other ways of judging performance. Go read the whole thing, then start trading! At least if you're a Kiwi.

Update: Chris Masse is right: the Olympics markets are not the kinds of events for which we would expect good prediction market performance. I just have fun poking noses. I have the underlying data for iPredict from Matt that will let me see whether we're closer to the 45 degree line for the kinds of contracts that are more like the Olympics one as compared to the kinds of contracts that are more likely, a priori, to aggregate wisdom of crowds. Will check as soon as I'm unburied from the mountain of grading.

No comments:

Post a Comment