It’s election season, which means I obsessively and pointlessly check PredictWise more than once a day. It aggregates all the election betting markets (sort of like the Kayak of prediction markets). I actually have a shortcut on my iPhone home screen for the web page. Seriously. It’s a disease.
I also enjoy the periodic blog posts by the creator, David Rothschild. For example:
The state-by-state breakdown of the presidential election in 2016 varies a lot from the 2012 election; in 2012 Republican Mitt Romney needed to sweep Florida (FL), Ohio (OH), and Virginia (VA) to win the election, in 2016 Republican Donald Trump needs to win 3 of 4 from FL, OH, VA, and Pennsylvania (PA). The chart tracks the probability of victory on PredictWise since we started state-by-state predictions on February 27, 2016.
1) The assumption here is that if Trump wins FL, OH, and PA he will win enough other states to win or if Clinton wins PA and VA she will win enough other states to win. The candidates still need to win those states, but they are very, very likely to be swept along on the wave should the candidate win the bigger states.
2) Putting North Carolina (50%), Arizona (27%), Missouri (21%), and Georgia (12%) in-play is really costly to the Trump campaign. It needs to play defensive in states it could traditionally spend less time and money. But, it does not alter the reality that Trump still needs to win 3 of the 4 big states.
Up through today the winner of the general election and the winner of each state has used some of the same data, but they are not formally tied together; this has worked fine, but it is starting to get a little concerning. The probability that Clinton wins the general election should not really fall below the probability that she wins PA, the current “swing” state. Clinton is up 43.5% to 39.1% in the Pollster average (solid lead for the candidate in a state that has gone to her party every year since 1992), she is trading at $0.69 on PredictIt, fundamental prediction is down just above 50%. All of this combines in my model to 78%. So, what should you make of Clinton 75% to win the election and 78% to win PA?
1) Not a huge discrepancy, so not a big deal.
2) The national markets may be undervalued for Clinton and the state-by-state data may indicate a slightly higher valuation. I do not suspect 2012 style manipulation. I suspect that the $850 cap per person/contract on PredictIt is giving its very anti-establishment trader-base too much say versus the “good money”. It is something I will be following very closely in the next few weeks.