I think it’s still too early to perform a full postmortem on the election because some margins still aren’t known, but my current hypothesis is that the presidential markets had uniquely poor calibration because Donald Trump convinced many people that polls didn’t matter, and those people were responsible for a large part of the money put on him (as supposed to experienced, dispassionate gamblers).
The main evidence for this (this one is just about irrationality of the market) is the way the market has shifted, which some other people like gwern have pointed out as well. I think the most damning part here is the amount of time it took to bounce back. Although this is speculation, I strongly suspect that, if some of the good news for Biden had come out before the Florida results, then the market would have looked different at the the point where both were known.[1] A second piece of evidence is the size of the shift, which I believe should probably not have crossed 50% for Biden (but in fact, it went down to 20.7% at the most extreme point, and bounced around 30 for a while).
I think a third piece of evidence is the market right now. In just a couple of minutes before I posted this, I’ve seen Trump go from 6% to 9%+ and back. Claiming that Trump has more than 5% at this point seems like an extremely hard case to make. Reference forecasting yields only a single instance of that happening (year 2000), which would put it at <2%, and the obvious way to update away from that seems to be to decrease the probability because 2000 had much closer margins. But if Trump has rallied first-time betters, they might think the probability is above 10%.
There is also Scott Adams, who has the habit of saying a lot of smart-sounding words to argue for something extremely improbable. If you trust him, I think you should consider a 6ct buy for Trump an amazing deal at the moment.
I would be very interested in knowing what percentage of the money on Trump comes from people who use prediction markets for the first time. I would also be interested in knowing how many people have brought (yes, no) pairs in different prediction markets to exploit gaps, because my theory predicts that PredictIt probably has worse calibration. (In fact, I believe it consistently had Trump a bit higher, but the reason why the difference was small may just be because smart gamblers took safe money by buying NO on predictIt and YES on harder-to-use markets whenever the margin grew too large).
To be clear, my claim here is bad news came out for Biden, then a lot of good news came out for him, probably enough to put him at 80%, and then it took at least a few more hours for the market to go from roughly 1⁄3 to 2⁄3 for Biden. It’s tedious to provide evidence of this because there’s no easy way to produce a chart of good news on election night, but that was my experience following the news in real time. I’ve made a post in another forum expressing confusion over the market shortly before it shifted back into Biden’s favor.
I think it’s still too early to perform a full postmortem on the election because some margins still aren’t known, but my current hypothesis is that the presidential markets had uniquely poor calibration because Donald Trump convinced many people that polls didn’t matter, and those people were responsible for a large part of the money put on him (as supposed to experienced, dispassionate gamblers).
The main evidence for this (this one is just about irrationality of the market) is the way the market has shifted, which some other people like gwern have pointed out as well. I think the most damning part here is the amount of time it took to bounce back. Although this is speculation, I strongly suspect that, if some of the good news for Biden had come out before the Florida results, then the market would have looked different at the the point where both were known.[1] A second piece of evidence is the size of the shift, which I believe should probably not have crossed 50% for Biden (but in fact, it went down to 20.7% at the most extreme point, and bounced around 30 for a while).
I think a third piece of evidence is the market right now. In just a couple of minutes before I posted this, I’ve seen Trump go from 6% to 9%+ and back. Claiming that Trump has more than 5% at this point seems like an extremely hard case to make. Reference forecasting yields only a single instance of that happening (year 2000), which would put it at <2%, and the obvious way to update away from that seems to be to decrease the probability because 2000 had much closer margins. But if Trump has rallied first-time betters, they might think the probability is above 10%.
There is also Scott Adams, who has the habit of saying a lot of smart-sounding words to argue for something extremely improbable. If you trust him, I think you should consider a 6ct buy for Trump an amazing deal at the moment.
I would be very interested in knowing what percentage of the money on Trump comes from people who use prediction markets for the first time. I would also be interested in knowing how many people have brought (yes, no) pairs in different prediction markets to exploit gaps, because my theory predicts that PredictIt probably has worse calibration. (In fact, I believe it consistently had Trump a bit higher, but the reason why the difference was small may just be because smart gamblers took safe money by buying NO on predictIt and YES on harder-to-use markets whenever the margin grew too large).
To be clear, my claim here is bad news came out for Biden, then a lot of good news came out for him, probably enough to put him at 80%, and then it took at least a few more hours for the market to go from roughly 1⁄3 to 2⁄3 for Biden. It’s tedious to provide evidence of this because there’s no easy way to produce a chart of good news on election night, but that was my experience following the news in real time. I’ve made a post in another forum expressing confusion over the market shortly before it shifted back into Biden’s favor.