Why relying on prediction markets in the elections may not be a safe bet The U.S. midterm elections are almost here. Pollsters have been busy asking voters whether they are likely to vote and who they plan to vote for. This data indicates that there are several close races in the House, with Democrats holding a clear edge. Many Senate races are competitive in what have been traditional Republican strongholds like Texas and Alaska. Prediction markets have also entered the midterm prediction arena.

A question that candidates and voters may be asking is: Will pollsters or the prediction markets provide the most accurate forecasts? To assess this question requires a deep dive into both approaches.

Polling has evolved over time, yet its basic premise remains the same: sampling a group of people from which a prediction about an election outcome can be drawn. A sample of 1,000 people typically results in a plus or minus three-point margin of error. This means that in close races, like one where candidate A is polling at 49 percent, Candidate B is polling at 47 percent, and 4 percent are undecided, the margin of error dampens the value of polls as a predictive tool. There are many such races in this election cycle.

Prediction markets take an entirely different approach. People trade contracts based on a yes/no question. The balance between yes/no bets determines the price of contracts and the effective odds of winning. Prediction market contracts are priced between $0 and $1, which reflects the market’s estimate of a candidate’s probability of winning. For example, if a contract trades at 60 cents, the market assigns that candidate a 60 percent chance of victory. Note that prediction markets do not provide a margin of error for their forecast.

Looking back, prediction markets were assessed to be more predictive than polls for the 2024 presidential election. One possible explanation is that odds are set by people placing bets on who they believe will win. This contrasts with polls, where people are sampled by pollsters and asked questions about candidate preferences. This means that voters who place a bet are self-selected hoping to win money, while voters have no real choice on whether they are sampled by pollsters.

The bigger issue is whether people who engage with prediction markets are representative of the people who will vote. This then begs the question: Is money being wagered based on who bettors believe will win or who they want to win?

Given that anyone with access to a prediction market platform can place bets on the election, barring state restrictions, while only U.S. citizens are eligible to vote, there is no reason why prediction markets should serve as a reliable prediction tool. Moreover, the objective for bettors is to win money. To improve their odds, they are likely to assess all news information available, including polls. This means that polls inform prediction market odds, adding some dependency in their forecasts. Polls are also snapshots in time, while prediction markets dynamically update odds in real time as money flows to each candidate.

Americans are generally opposed to betting on elections; recent polls indicate that more Democrats than Republicans oppose election betting (69 percent versus 61 percent). Similarly, more Republicans than Democrats support election betting (16 percent versus 12 percent). Other betting demographics, like the high number of younger males, may further obfuscate the predictive value of prediction market odds. If Republican bias translates into bets being placed, this may give Republican candidates an edge in prediction markets that may not translate into the election outcomes, though arbitrage and market efficiencies will in theory correct for this.

Polling is a tried-and-true process for election prediction, while modern prediction markets are a modern phenomenon that has grown out of the digital economy. What will likely challenge both approaches in this year’s midterms is the large number of races with candidates polling well within the margin of error. Contradictory forecasts between polls and prediction markets are representative of true toss-ups.

Like every election, voter turnout matters. There is no way to determine this until after the election is complete. Factors like inclement weather on Election Day or an unexpected personal emergency can derail the best laid plans to vote.

If the demographics of people that show up to vote are not captured in polls, the polls are likely to provide misleading information. If the demographics of people placing bets in prediction markets do not represent how voters will act based on the amount of money wagered that set contract prices, they too will not offer much predictive value.

After the elections, the results will be reported and with 20-20 hindsight, both pollsters and prediction markets will claim their share of victories and explain away missteps. Yet the stakes are radically different. Future polling errors can be addressed by changing how polls are conducted. Prediction markets are a business platform where accuracy drives attention and revenue.

Of course, the true winners aren’t those who placed bets or pollsters conducting surveys, but rather, the voters who take the time to make their voices heard. They provide the data that matters most.

Sheldon H. Jacobson, Ph.D., is a professor of Computer Science at the University of Illinois at Urbana-Champaign. He applies his expertise in data-driven risk-based decision-making to evaluate and inform public policy.