Prediction Markets Are Mirrors, Not Crystal Balls
What recent primary upsets tell us about how to trade political prediction markets:
The favorite lost in Iowa, Georgia, South Carolina, and New York over the last month. The prediction markets (Kalshi and Polymarket) called only one of those upsets ahead of time, missed two, and only saw the third coming once the votes forced their hand. That scorecard isn’t a knock on prediction markets, it’s a good lesson on what they’re actually good for.
The story going around about the last months primaries is that Trump’s grip on Republican primaries is loosening, and the Democratic establishment’s hold on its own races is too.
Trump-backed candidates lost in Iowa, Georgia, and South Carolina, while in New York three insurgents backed by Mayor Zohran Mamdani beat the party’s preferred picks. But Polymarket and Kalshi didn’t necessarily see this coming.
Understanding why Prediction Markets, which are often cited as being more accurate than polls, had the wrong candidate favored is important for political traders.
Start with the thing people forget. A prediction market isn’t a window into the future. It’s a running count of what its traders believe right now. In a low-turnout primary, that belief mostly comes down to two inputs: who has the bigger name, and who has the endorsement. The market reflects those two things faithfully. What it struggles to do is see past them.
That distinction is the whole story of the last month. In most of these races, the market didn’t see the upset coming. It had the eventual loser favored, often heavily, right up until the votes came in. And where it got the call right, it was for one specific reason that tells you exactly when to trust these markets and when to fade them.
The Races the Market Missed
Iowa is the cleanest example of the market reading an endorsement instead of an electorate. Trump endorsed Representative Randy Feenstra and called him “MAGA all the way,” so the markets did the obvious thing with a sitting congressman holding a presidential endorsement and made him a heavy favorite. Kalshi had Feenstra around 70 to 75 percent late in the race, with farmer Zach Lahn down near 30. Lahn won. The market wasn’t reading Iowa voters; it was reading the endorsement.
New York’s 13th District ran the same way. Adriano Espaillat, a five-term incumbent and chair of the Congressional Hispanic Caucus, was priced around 67 to 69 percent to keep his seat. He lost to Darializa Avila Chevalier, a 32-year-old organizer, by about three points. A Data for Progress poll before the election had the challenger ahead. The poll was measuring voters. The market was repricing the incumbent’s résumé.
In both cases there was no mechanism to force the price off the endorsement before election night. The wrong number simply rode all the way to the finish.
South Carolina: Watch the Machinery Work
South Carolina is the most useful race of the bunch, because you can watch the endorsement do its work and then watch it fail in real time.
Trump backed Lieutenant Governor Pamela Evette late in the campaign. On the morning of the June 9 primary, Polymarket gave her an 80 percent chance of winning the nomination, with state attorney general Alan Wilson sitting at 12. That 80 percent was the endorsement, full stop.
Then Evette actually had to get votes, and she led the first round by just 29 to 26, nowhere near the majority she needed to avoid a runoff. The price she’d been carrying suddenly looked unearned.
Over the next two weeks it came apart. Nancy Mace, who finished near the back of the primary, endorsed Wilson. New polling put him ahead. The market flipped hard, moving Wilson from the 70s up to about 98 percent by the runoff, and he won going away.
The market got there in the end, but notice when. It only corrected after the first round of real voting contradicted the endorsement out loud. The runoff was a public checkpoint that the endorsement couldn’t survive, and that checkpoint is the only reason the price ever moved.
Georgia: the One the Market Actually Nailed
Then there’s Georgia, which matters most because the market got it right from the very start.
Trump endorsed Lieutenant Governor Burt Jones. Governor Brian Kemp backed him too. By the logic that ran the Iowa and New York markets, Jones was the easy trade, pile on the endorsements and price him as the favorite. But he lost.
For months, Polymarket had businessman Rick Jackson ahead of Jones, pricing him in the 50s and 60s even as the endorsements stacked up on the other side. Jackson, who put more than $100 million of his own money into the race, won the June 16 runoff by about 53 to 47.
Georgia is the tell. Its traders didn’t take the shortcut. They didn’t just price the endorsement, they looked at a self-funder with a hundred million dollars and a real business record and made their own call. That’s the entire difference between the races the market nailed and the ones it blew. When traders priced the candidate, they were early. When they priced the endorsement, they were wrong until the votes said otherwise.
The Fair Rebuttal, and Why It Doesn’t Hold
The honest objection here is that I’m leaning on the misses and underselling how fast these markets correct. South Carolina is the rebuttal, the price did flip to the right answer in two weeks, well before the runoff finished. And the markets were flatly correct in plenty of races.
In New York’s 10th District, Kalshi had Brad Lander near 98 percent and he won by two-to-one. In the 7th, Claire Valdez sat near 78 and won by twenty points. These markets are not always a step behind.
But look at what the South Carolina correction actually required. It happened because the primary came in two rounds, and the first round was a public checkpoint the endorsement couldn’t survive. Iowa and New York had no runoff, so there was no moment to force the price off the endorsement before the votes were counted. The market didn’t wise up on its own, the votes made it.
And the races the market called cleanly from the start, the Landers and Valdezes, were never close. Telling you the obvious favorite will win isn’t worth much, because the voter registration numbers already tell you that. The whole pitch for prediction markets is that they earn their keep in the tight races, where real money is supposed to sharpen the guess past what polls and pundits manage. There were four genuinely contested races so far this cycle. The market read the candidate instead of the endorsement in exactly one of them, Georgia, and that’s the one it got right ahead of time.
What This Means if You Trade It
The takeaway isn’t that prediction markets are dumb, and it isn’t that they beat the polls. Both are too simple. These markets are very good at gathering up the consensus and not good at seeing past it. That makes them great for pricing a sure thing and shaky in exactly the races where you’d want the help, the close ones, where the consensus is about to crack.
So here’s the practical filter. When a market has a candidate at 95 percent in a lopsided race, believe it; the consensus and the outcome are the same thing. When a market has a candidate heavily favored mostly because of an endorsement in a contested race, treat that price with suspicion. The endorsement is already in the number. The question is whether anything will force the market to look past it before election day.
Two things can crack a stale endorsement price. The first is a structural checkpoint, like a runoff or an early-voting release, that publicly contradicts the consensus and forces a repricing. The second is traders doing the actual homework, pricing the candidate’s money, organization, and fit instead of just the endorsement. Georgia had the second. South Carolina got the first. Iowa and New York had neither, so the wrong price rode all the way to election night.
The edge, then, isn’t “trust the market” or “fade the market.” It’s knowing which kind of race you’re looking at. In a blowout, the market is a mirror showing you the obvious. In a tight race priced on an endorsement, it’s a mirror showing you the consensus, and a mirror is only as smart as whoever is standing in front of it.