Prediction Markets and Polls: What the Research Shows

One more interesting point to add to the recent discussion about prediction markets and polls. Following my first post on the subject, Temple University political scientist Christoper Wlezien sent along the following message:

Josh, there is a lot of interest in prediction markets on the presidential race and, as you point out in your TMC piece, this year’s Intrade market seems to be mimicking 538’s probability forecast of the presidential election. We [JT: Wlezien and Robert Erikson] have two papers (here and here) on prediction markets and the polls and they show that as Election Day approaches, market prices basically follow the polls. This is true even when the polls are wrong, as in 1948. At this point in the election cycle, markets appear to add nothing beyond what the polls show.

Now Chris wrote this before the discussion on The Monkey Cage yesterday regarding the subsequent divergence between 538 and Intrade (which continues today). The social scientist in me says that the Wlezien and Erikson research suggests the emerging divergence – especially if continues for the next two weeks – represents a puzzle. However, if one wanted to be a little more provocative in the spirit of Justin Wolfer’s recent tweet and Atlantic interview, then the previous patterns of convergence identified by Wlezien and Erikson would at the very least not be inconsistent with the view that something out of the ordinary is going on now.


Update: Upon reading the above post, Chris wrote back to me with two additional points, both of which caution against interpreting the current deviation between Intrade and 538 as out of the ordinary. First, he notes that his comments had to do with the convergence of markets with polls, not the convergence of markets with forecasting models. So on the basis of the current state of the national polls, the Intrade numbers don’t look like they are deviating much at all. Second, Chris also notes that in their POQ article, they found that winner take all elections tended to have a “long-shot bias”, so that might also explain why the markets could price a Romney victory at a higher likelihood than a forecasting model like 538.

Comments are closed.