One of the concerns about giving a lot of credence to the money wagered on elections — and thus how much “wisdom of the crowds” is really “wisdom” and/or “crowds” — is that it seems the people doing it aren’t exactly a representative cross-section of the American voter.
Point being, it seems highly unlikely 75-year-old farmers in Iowa and 55-year-old divorcees in California are as active on prediction markets as 23-year-old data nerds in Brooklyn. As a result, it would seem intuitive that the odds in these markets don’t reflect the true odds of how people will vote.
A new academic paper finds plenty of evidence to back this demographic inconsistency. Examining political betting markets spanning more than 100 years — well before the current prediction market mania — the researchers didn’t find a single group of bettors that represented the voting public.
But they also found that this didn’t matter.
The bettors’ politics weren’t throwing off the prices, and American capitalism — basically, having money on the line — gives the participants in these markets plenty of reason to bet with their brain, not their political leanings.
That’s the main finding from Harry Crane of Rutgers University and Koleman Strumpf of Wake Forest University, whose working paper, “Diversity, Efficiency, and Incentives in Political Prediction Markets,” examines a criticism likely to get another workout with the midterms next week.
“Every market I’ve ever looked at, the people who trade in them are totally unrepresentative of voters,” Strumpf told InGame.
The demographic differences were substantial, bordering on comical. On Kalshi in 2024, 86% of election traders were male and more than half were under 30. On PredictIt in 2016, men made up 92% of presidential traders.
Crane and Strumpf also went deep. They tracked down census records for 80 people betting on the 1900 election in New York. Every one of them was a man, 44% were bankers or brokers, and the typical bettor was older than the average for the city’s male population.
Different eras, different crowds, plenty of dudes.
The authors argue that what kept these markets useful was the heavy influence of traders who were simply trying to make money, even when other bettors were happy to put a few dollars on their preferred candidate.
Harris money
The most detailed part of the paper uses records supplied by Kalshi, covering election trading from Oct. 4-Nov. 5, 2024, and it showed people were using their brains instead of their hearts.
When Crane and Strumpf weighted traders’ home ZIP codes by trading volume, those areas voted for Donald Trump just 28% of the time. Yet some 60% of trading volume went toward Republican positions.
But among people who made only one single trade, 73% of the money landed on Kamala Harris.
The researchers offer two possible explanations for those one-time bets, and they’re both reasonable and obvious. Either they were backing their preferred candidate, or they preferred betting on the longshot at the time, as Harris was trading around 40 cents.
The data doesn’t settle which explanation mattered more, and the ZIP codes obviously can’t tell us how each individual trader voted.
A reason to be right
As the trading ramped up, researchers found, the political leanings of the traders mattered less and less.
Among people who made more than 100 trades in the last month of the 2024 election cycle — and this is 93% of the entire volume over that time — the researchers found zero difference in which side the money backed based on the politics of the traders’ ZIP codes. It was basically a pure profit play.
“Their demographics don’t seem to impact how they trade,” Strumpf said. “And I should say this is disproportionately true for heavy traders.”
Additionally, many active traders would be taking positions on both sides of a contract, trading as they went, which the authors say keeps bettors’ political preferences at bay. And, of course, if any short-term enthusiasm for a candidate pushed the price too high, traders were quick to come in on the other side.
“To me, ‘dumb’ money is actually essential for these markets to work,” Strumpf said. “They’re the chum providing the profitable trading opportunities for the ‘smart’ people.”
All told, Kalshi’s crystal ball/crowd wisdom results in 2024 were strong. In the Trump-Harris race, its prices beat Silver Bulletin, FiveThirtyEight, and The Economist, per the paper. Kalshi also beat the available poll-based models in every battleground state and across the group of 19 Senate races studied.
The 2024 House races, however, were another matter.
Kalshi opened 24 district markets six days before the election. They attracted just $236,000 combined, and FiveThirtyEight performed better in 20 of them.
The researchers cautioned that the sample is too small to say whether the problem was a lack of trading or simply a bad week. Still, the House results leave something worth asking as we get ready for the 2026 midterms: How much money needs to be floating around the markets in a congressional race for the price to be considered a true measure of the probability?
“I guess I won’t go to the mat to say markets are going to work great,” Strumpf said. “But I guess I’m very curious to see.”
