August 14, 2021

Prediction market Kalshi has a question called “Will 2021 be the hottest year on record?” (See also the Metaculus version of this question.) As I am writing this, a Yes share can be bought for 19 cents, and a No share for 83 cents. But it is already obvious that this question will resolve No. If we look at the data source Kalshi will use to resolve the question (No_smoothing column) and the table of monthly data from NASA (scroll to bottom), we can see that the months for which we have data were simply too cool. Yes, that’s despite the fact that this year was the hottest June in North America, and now NOAA is saying July was the hottest month of all time. For a Yes resolution, the J-D column in the monthly data table has to end up 103 or greater. This number is just the average of the monthly numbers. So given the numbers we have so far, the remaining months need to be, on average, greater than 134. (That’s the index, not the actual temperature.) Numbers this high have never occured before in the August through December columns. We did get that high a couple of times in 2016 though. So a Yes resolution would require all of the five remaining months to be extraordinarily hot. This seems so unlikely that I feel comfortable giving it less than 1% chance of happening. And NOAA agrees.

So why is Kalshi overconfident? If they were just sticking with what Metaculus says, I’d understand. But a Yes share hit 27 cents earlier this month, which is ridiculously high, and this week it has risen into the double digits yet again.

I think this is probably because there isn’t enough money to be made. The question won’t resolve until February, and for such a modest return why would someone choose to lock up their money for that long? I expect Yes shares to gradually converge to 1 cent as the close date approaches.