Prediction markets can be useful forecasting tools, but they won’t replace political polls. Although political polls failed to fully capture Donald Trump’s chances of winning the 2016 presidential elections, they are still startlingly accurate.
G. Elliot Morris points out early in his book Strength in Numbers that “these days, national polls miss the winning party’s vote share an average of just one or two points in most elections. Even in 2020, they were only off by two and a half [points]. Few other indicators have such a record of accuracy in measuring public opinion.”
Prediction markets may seem like promising alternatives to political polls. Although prediction markets have demonstrate predictive power, they have certain shortcomings.
For example, sportsbook odds don’t impact the chances of a bet winning. If a moneyline moves in favor of one team, the team isn’t more likely to win. In politics, public perception of a candidate’s polling can impact a voter’s decision to vote for them in an election, which can skew prediction market predictions.
Instead of seeking alternatives to prediction markets, individuals should improve their reading of political polls. Polls can’t perfectly tell the future, as the 2016 presidential election and some 2020 state results demonstrated. However, they have key advantages over prediction markets that suit polls for political predictions.
Why The United States Doesn’t Have Political Prediction Markets
Prediction markets haven’t taken off in part because of American online gambling restrictions. The United States passed the Unlawful Internet Gambling Enforcement in 2006. This was done shortly after the United Kingdom passed its 2005 Gambling Act. The UK act regulated rather than banned online gambling companies. Would-be prediction markets would have had to acquire expensive gambling licenses and wouldn’t be able to offer markets to all 50 states. With varied state gambling regulations, private-run prediction market companies were untenable as gambling models.
Kalshi, a prediction market company, launched by securitzing ‘yes’ or ‘no’ answers and becoming regulated by the Commodity Futures Trading Commission (CFTC). As a finance company, it can offer contracts in all 50 states without tweaking its availability to suit each state.
The United States faces cultural barriers to political betting. Kalshi does not offer political betting because it’s a cultural taboo in the United States. In contrast, Canada and the United Kingdom have thriving election betting markets.
Prediction markets have the advantage of aggregating responses from large groups of individuals who only have snippets of information. James Surowiecki, author of Wisdom of Crowds, emphasizes that information individuals possess must be relevant to the question they are answering for the prediction to be accurate. In a large enough group, even egregious errors can cancel each other out.
Surowiecki cites a study in his book in which a group of college students try to guess the correct number of jellybeans in a jar. There were 850 beans in the jar, and the students collectively guessed 871, only 2.5% away from the real answer.
Kalshi uses this same principle to offer prediction markets ranging from the Fed interest rate to the end of the Writer’s Strike. By pricing contracts from $0 to $1, customers into probabilities. Customers use their intuition to make their ‘yes’ or ‘no’ contract purchases, and the publicly available market prices show the probabilities of the ‘yes’ and ‘no’ outcomes.
Aggregating political opinions can work the same way. However, public prediction markets would require controls to ensure accuracy at scale.
Prediction Market Flaws And Political Poll Weaknesses
If prediction markets or polls want to accurately predict election outcomes, they must pay attention to their audiences.
In Strength in Numbers, that during the 2016 election, one of the biggest reasons that presidential polls underestimated Trump’s chance of victory was a failure to factor in education. Education was correlated with voter decisions in a way that it wasn’t in previous elections. This voting pattern emerged because of Trump’s populist campaign and the perception that Clinton was out of touch with white working-class voters.
Morris cited a New York Times analysis that found “nearly half of respondents in a typical national poll had at least a bachelor’s college degree. But the percentage of college graduation among the actual population is only 28%.” Many polls failed to include enough likely voters without college degrees, leading the final University of New Hampshire poll to overestimate Clinton’s New Hampshire victory by 10.6 points.
Morris also found that traditional polls struggled to account for ‘low-trust’ voters. Voters with low trust in government institutions were less likely to respond to pollsters. Trust in institutions wasn’t measured with any single demographic. Morris pointed out that “a low-trust voter might be a 21-year-old college-educated woman from Iowa now living in Philadelphia” as much as it could be “a 65-year-old Republican man in the Deep South.” The low-trust voters were more likely to vote Republican, so polls systematically overestimated Trump’s chances.
Prediction Markets And Their Audiences
Prediction markets aren’t immune from error. PredictIt, an election market platform, found during the 2020 presidential election that bettors were wagering on who they wanted to win instead of who they thought would win. Slate reported that the “day after the Electoral College voted for Biden, people were still backing Trump on PredictIt.”
Even after the election was called for Biden, PredictIt users still bet on a Trump victory. Markets are only as rational as the crowds that compose them . Prediction markets must adjust to meet that challenge to successfully make predictions.
Prediction markets are also subject to the type of market manipulation that the NASDAQ monitors. Pollsters can correct for market irrationality by adjusting their samples. While pollsters can make mistakes, prediction markets can’t unilaterally make people improve their predictions. Prediction markets could attract new customers with particular characteristics, but they could run into the same difficulties with low-trust respondents that traditional polls did.
Prediction markets can’t replace traditional polls, but they could complement them. A thriving political market could show real-time changes in polling results. Transparency of audience metrics could help poll readers decide whether a poll accounted for fast-moving political trends, like populist appeals in 2016 and loss of institutional trust in 2020.
Final Thoughts On Reading And Using Polls
Morris ends his book by recommending a new reading of margin of error. Margins of error help the reader understand the range of possibilities that the poll is predicting. Morris found that a poll’s ‘true’ margin of error can be double the sampling error. Taking the poll as a range of possibilities instead of concrete predictions would improve the reading and reporting of polls. Reporting ranges is also a useful lesson for prediction markets hoping to offer election betting in the United States.
There’s no crystal ball that perfectly predicts human behavior. Polls do a good job of measuring elections overall, even factoring in high-profile mistakes in 2016 and 2020. Pollsters can miss important factors that impact voting patterns. However, they also have process lessons to draw on to avoid making 2020’s mistakes in 2024.
Prediction markets won’t automatically solve polling’s predictive issues. Political prediction markets would face the same challenges of building representative audiences that pollsters do. The wisdom of crowds can produce actionable and accurate predictions, but the future isn’t a math problem to be solved. Even the best predictive models can only narrow down a range of futures rather than identify the one that will come to pass.