Polls vs punters: an explanation?
Nearly a month ago, I noticed that betting markets were giving long odds (3.5 to 1) against a Labor win in the (presumably) forthcoming election. That would be a good bet if you thought Labor had a better than 22 per cent (1/(1+3.5)) chance of winning. Given that the polls were pretty much tied, I thought those were good odds.
Since then, the polls have moved steadily in Labor’s favor to the point where their lead is just about statistically significant in a meta-analysis (add lots of independent samples and the margin of error declines). At the same time, the government has barely had a good news day. Their one big hit, the kerfuffle about 10-year projections of tobacco tax revenue (a bipartisan policy) blew up in their faces a few days later when Turnbull and Morrison couldn’t/wouldn’t state the cost of their company tax plan. It seems that they had the $50 billion number ready, but had hit on the clever plan of having the Treasury announce it today, just before Parliament is dissolved so that Labor couldn’t … I’m not sure what (cue underpants gnomes). As a result of all this, the pundits, who dismissed the idea of a Labor win as implausible until very recently, are now coming around to the idea
Yet despite all this, the odds are barely unchanged at 3.3 to 1. That’s good for anyone who gives Labor a 23 per cent chance. There are a few possible explanations of this
(a) The idea that betting markets are highly rational aggregators of information is wrong
(b) Those betting in these markets have inside information or else insights unavailable to the rest of us.
(c) The markets don’t really exist in any substantial form and are just a publicity stunt for the bookmakers. That’s the argument of this 2013 article by Michael West, whom I’ve usually found to be sensible and reliable.