Maybe I’m the idiot, but I think last week’s fevered discussion about polls and models vs. prediction markets in the wake of the midterms is the most idiotic fevered discussion between normally smart people I’ve ever seen or heard.
Now I will grant you: I may not be seeing the forest for the trees, but when you have one side claiming the polls and the models were the clear winners, and another side saying the prediction markets were the clear winners, I’m pretty certain that means both sides are missing a bigger truth.
And I’m pretty sure the bigger truth is this: Without polls and models, people betting into election results on prediction markets would be flying blind. Also: With prediction markets, pollsters and modelers get to see what people are willing to do with their information when being wrong costs them money.
I mean, this seems pretty simple. The polls inform the models, the models inform the trades.
To me, it’s like how you can eat peanut butter, jelly, and bread in any combination you want, but it’s not a peanut butter and jelly sandwich unless you have all three working together.
Does that metaphor work? Maybe? How about we bring in Slats Grobnik to help me out here.
Slats what?
OK, for those who aren’t steeped in newspaper columnist lore, Slats Grobnik was a fictional character created by Mike Royko of the Chicago Daily News and later the Chicago Sun-Times. Royko is pretty much the best newspaper columnist ever, and he used “Grobnik” as a sounding board.
Well, I have my own “fictional” character, except this one is a real person: Captain Jack Andrews. Basically, whenever something is happening in the wide world of gambling that makes my eyebrows raise, I reach out to Captain Jack to help guide me through.
This case was no different.
Andrews started with the obvious distinction: money.
The people answering a poll don’t lose anything if their answers turn out to be wrong, but people betting into prediction markets quite obviously can lose money.
So, yes, that’s a great reason to pay more attention to the prediction markets. But at the risk of turning this whole discussion into an ouroboros, the people risking their money betting on elections are … looking at polls, evaluating them, looking at models, evaluating them, maybe even creating their own models. They might even have information the public models don’t.
Andrews brought up the Polymarket whale from the 2024 presidential election, whose approach included asking people who they were voting for and who they thought their neighbor was going to vote for. We went back and forth about how that might reveal more, or at least how it did in 2024.
But even there, I kept coming back to the same thing: That was polling, modeling, and then placing the trade.
Somewhere in all of this, I tried the peanut butter and jelly sandwich argument on him.
“Yeah, I see that,” he said.
Good. Lunch was settled.
The question of who gets to take a victory lap was proving more difficult.
“How can you say if something happening was 72%, versus someone else saying something happening was 87%, and the thing happened,” Andrews asked. “Who was right?”
In other words, if two forecasts both favor the eventual winner, the result doesn’t tell me how impressed I should be with either one.
More PB&J, and one NFL game
Back to my peanut butter and jelly sandwich thing again. (Sorry.) No single ingredient there is the DNA for the whole thing.
Meaning, a poll asking who you intend to vote for is asking a different question from a market asking who will win. A candidate getting 55% in a poll is not the same thing as that candidate having a 55% chance of winning. And 87% of the people in my little Twitter poll below picking a football team to win doesn’t give that team an 87% chance of winning, either. (By the way, at the close of this poll, the Jaguars were at 76% to win on Kalshi. They eventually won. Was my poll “more” right?)
Captain Jack, meanwhile, was considerably less interested in declaring himself an election expert than seemingly everyone else on the internet.
“It’s out of my depth because it’s just not something I’ve ever put a lot of thought into,” he said.
He did, however, recognize the people announcing afterward that the midterm results had been obvious.
“They didn’t seem to make that known beforehand, right?” he said.
Which is a great point. I witnessed this in a handful of races that I was holding (thank you, Alex Keeney of Eventful), specifically Gina Hinojosa to win the governorship of Texas and J.D. Ford in Indiana’s Fifth District.
Early on Election Day, Hinojosa was trading at about 35 cents, Ford was at 25 cents. They both won.
Heading into the election, Hinojosa was trailing Gov. Greg Abbott by 2.3 points in the polls, according to Real Clear Polling, and Ford was trailing by 3 points.
So who got it “right”? I’ll wait while you run the numbers.
To be fair, the prediction markets were quicker to get to 99 cents (i.e., call the races) than the Associated Press and others were willing to call it, but to that I say “duh.” Betting markets don’t decide who wins and loses. Voters do. And of course the news outlets are going to wait until the math is solid before declaring anything. (Lessons, learned.)
Anyway, I’m happy to hear an argument that a particular model added more value, that this or that poll nailed it, that these particular traders are truth machines. Show me who said what and when they said it and then show me how we’re going to measure all of it when mixed in with the results. That sounds like a useful, if mentally painful, discussion.
In the meantime, all I think I know is this: Polls inform models, models inform trades, strawberry jam over grape jelly for the sandwich.
