Prediction markets coverage towards a solid aggregation of polling as Election Day approaches. Early in the election cycle markets are all about idiosyncratic and dispersed knowledge of the campaign; polling is both thin and markets know a lot about what people will learn about candidates over the course of the election cycle. And, when news breaks, they are the first objective measures of the size of the impact. But, with the exception of early voting data, which heavily influenced the Nevada forecast, there is little relevant idiosyncratic or dispersed data that is not yet incorporated in public polling by Election Day. On Election Day, markets are estimating the probability of polling error, which is also what a smart aggregations of polls are doing.
The Michigan primary was a good example of where markets suspected something was wrong, but were anchored by the polling. FiveThirtyEight, using a reasonable aggregation of the polling, had Clinton at >99% to win the primary over Bernie Sanders. Prediction markets were hovering near 90% for Clinton. Clinton was winning by 20 percentage points in the aggregated polling. Even if the idiosyncratic and dispersed data coming into prediction markets said this was going to be tight, it is hard to for people push the market all the way to Sanders when FiveThirtyEight is at >99% and the polling average is 20 pp different. This is not a problem five weeks before the election when the markets can anticipate a shift in public opinion to come. This is is not an issue if a new scandal broke. But, barring other really pertinent information, a market is just anchored by the polling.
Markets were not missing the Trump voice; actually, the forums are dominated by Trump supporters. Trump leaning polls were cherry-picked and pushed in the forums. Every rumor of a Clinton scandal was a triumph. Yet, still, the markets leaned Clinton in Wisconsin, Michigan, and Pennsylvania for a very important reason, Clinton was consistently leading in the polls and while many people were willing to bet on Trump in these states, people are were not willing to bet more than even money on the candidate consistently losing in the polls.
I am a huge fan on markets because they are great aggregation of polls that also have a lot of insight into what will happen early in the cycle and when information breaks. And, I think they did a solid job, as always, on that front. But, my previous research has been very clear on this, they will converge to polling as Election Day approaches. That is why in this column and on my Facebook Live posts I was mainly talking about polling averages as Election Day happened. Then again, markets learn, and maybe that will not be the case in 2020.