The prediction market company Kalshi recently inked a deal to become a partner with CNN and then CNBC. The goal is to share real-time prediction data with these media outlets, like the probability that the Democratic nominee will win the presidential election or that the Fed will cut interest rates.
There is reason for caution, however, about the value of prediction market data. That’s the conclusion of new research on the 2024 election prediction markets by my Vanderbilt colleagues Josh Clinton and TzuFeng Huang. They write:
We analyze more than 2,500 political prediction markets traded across the Iowa Electronic Markets, Kalshi, PredictIt, and Polymarket during the final five weeks of the 2024 U.S. presidential campaign involving more than two billion dollars in transactions to assess whether prices accurately and efficiently aggregate political information. While 93% of PredictIt markets correctly predicted outcomes better than chance, accuracy fell to 78% on Kalshi and 67% on Polymarket. Even the most accurate markets showed little evidence of efficiency: prices for identical contracts diverged across exchanges, daily price changes were weakly correlated or negatively autocorrelated, and arbitrage opportunities peaked in the final two weeks before Election Day. Together, these findings challenge the view that prediction markets necessarily efficiently and accurately aggregate information about political outcomes.
Let’s break that down. One thing you want from prediction markets is accuracy, especially given concerns about the accuracy of other potentially predictive tools, such as polling. For example, an event that has better than 50-50 odds, based on the price of a contract in the prediction market, should happen more frequently than events with less than 50-50 odds.
Clinton and Huang examine the 2024 election markets on these four platforms and ascertain how often those markets predicted the outcome better than chance. It’s important here to account for different kinds of markets. One type of market, the one we’re most familiar with, involves bets on who will win an election.
But other markets exist too, more so on Kalshi and especially Polymarket. These involve bets on what Clinton and Huang call “niche or low-information events that are more akin to speculation or entertainment.” This would be events like a candidate’s specific winning margin or whether the candidate will say a particular word in a speech.
The results show that those niche markets are the least accurate. And once Clinton and Huang account for the types of markets on each platform, Kalshi and Polymarket are not significantly different in their accuracy, although PredictIt is more accurate than those two. (See also political scientist Andy Hall on this point.) Interestingly, markets with more trading activity are not more accurate, controlling for the types of events that people are betting on.
The second thing that you want from prediction markets is efficiency. The markets should respond to information in a rational way: “If prices reflect a forward-looking expectation based on an aggregation of all available information – which includes the prices of similar contracts on other exchanges! – then the prices of all markets involving similar contracts should covary in meaningful and expected ways.”
Clinton and Huang study the markets for the Harris-Trump match-up in 2024. They find a lot of inefficiency:
- Across the 4 different platforms, the presidential election market prices were different and did not necessarily move together over time.
- The prices on the different platforms were correlated, but the correlations between markets were imperfect. Traders on the different platforms were not reacting in the same way to the information.
- The day-to-day changes in prices were much less correlated across the platforms. Again, traders were not reacting in the same way to information or even to the other markets.
- There were arbitrage opportunities. The prices of a Harris contract and a Trump contract should sum to $1 – meaning that the probabilities of a Harris win and a Trump win sum to 1. But on 62 of the 65 days before the election, that wasn’t true. Indeed, there were even more arbitrage opportunities closer to the election, which is the opposite of what you’d expect.
In a separate analysis, Clinton and Huang show that even though the national presidential market was relatively accurate, it was actually more inefficient than other types of markets, including the niche ones. The national presidential markets on these platforms “frequently exhibit short-term reversals rather than a smooth convergence toward the eventual outcome. This pattern can arise if the attention, liquidity, and speculative trading these markets attract cause new information to be rapidly – but often overreactively – priced in, leading to oscillations as traders correct for prior excesses or engage in profit-taking.”
In other words, yikes.
Now, this paper was written before the announcement about Kalshi’s partnerships with CNN and CNBC. The incentives created by these partnerships raise other concerns.
Andy Hall gives this hypothetical scenario:
It’s October, 2028. JD Vance and Mark Cuban are locked in a virtual tie for the presidential election. Suddenly, Vance’s price starts surging on prediction markets. CNN, which offers breathless, round-the-clock coverage of prediction-market prices through its partnership with Kalshi, documents the spike in great detail.
Meanwhile, no one knows why the price started spiking in the first place. Democrats insist the market is “rigged.” They point to evidence that a large batch of suspicious trades moved the market without any new polls or other apparent new reason to start favoring Vance.
The New York Times runs a story claiming that traders backed by the Saudi Arabian sovereign wealth fund are placing large bets on the election markets in order to drive favorable Vance coverage on CNN. Republicans say the price is justified, point out that there’s no evidence price spikes influence election outcomes anyway, and accuse Democrats of trying to stifle free speech and censor real information about the election. No one can immediately figure out what’s true.
Hall runs through previous attempts to manipulate markets. He notes that manipulation is not easy, nor is it clear that any manipulations will affect voters enough to decide the outcome of an election. But if markets are low in liquidity – i.e., there isn’t much betting activity or money being wagered – they get easier to manipulate.
Hall has a number of policy suggestions for the media, prediction market companies, and the government itself. For the media, one of the most important is to take account of liquidity.
The question is, however, whether media outlets can be trusted to do so, or whether the promise of a new “update” to a political race – no matter how chimerical – is too hard to resist.
Stay up to date on all things politics and political science. Bookmark our landing page and sign up for Good Authority’s weekly newsletter by entering your email address in the box below.
