David M Rothschild on Posted on

As we leave the realm of contested primary elections, let me take a quick look at how market-based predictions did in 2016. First, market-based predictions had full coverage, this is in sharp contrast to the poll-based forecasts that covered about 50-55 of the 80 or so contests. Second, market-based prediction started earlier. This is crucial as the one big point of interest is how the prediction move with the result of the contests the week or two before. Third, they were accurate in a binary sense. I will defer to Bloomberg, that showed that market-based forecasts were not just accurate when there were polls, but when polls did not exist. Specifically we had the binary winner right in 70 of 81 contests through today.

Markets were incredibly well calibrated. I took all predictions from one, three, and seven days before all primaries. I rounded the prediction to the nearest 10. Then, I charted how many came true. When market-based predictions said there was an X% probability of something happening, it happened X% of the time!

ScreenHunter_1244 May. 09 16.40

More to follow in the coming week and months.