What do the academics say? Using betting markets to predict election outcomes

Welcome to the latest in our occasional series highlighting interesting findings from academic research. Today it is a study into how good, or not, the betting markets are at predicting the results of elections.

A few years ago, when spread betting on election results first became popular, there was a brief period when the state of betting markets was touted as a good guide to election results. The theory was that when people decide whether or not to stake their money on a bet they are revealing what they really think, in the way that public comments often do not. Moreover, private information can be used to inform betting choices, which makes movements in betting markets one way to reveal what campaigns are really learning from their private polling, canvassing and so on.

Despite some initial very promising results in the US, the ability of betting markets to predict individual election outcomes has had a rather more mixed record since.

A trio of academics have therefore just analysed the constituency-level betting markets for the 2010 British general election (What are the Odds? Using Constituency-level Betting Markets to Forecast Seat Shares in the 2010 UK General Election by Matthew Wall, Maria Laura Sudulich and Kevin Cunningham, Journal of Elections, Public Opinion and Parties, Vol. 22 No.1, 2012).

They found that:

The performance of the seat forecasts that we have produced on the basis of the betting markets did not prove to be very accurate. In fact, the estimates performed less well than those derivable from national-level voting trends using constituency swing projections. We saw that a particular flaw of these data is their proclivity to overestimate the likelihood of long shots winning, and underestimate the chances of favourites … [But] they were rarely dramatically wrong – where favourites did not win their seats they typically came close to winning.

You can read the other posts in our What do the academics say? series here.