Why are prediction markets more
accurate than polls at making long range predictions and who do they favor to win the U.S. Presidential Election in 2024?
Two of the prediction markets running trading for
presidential elections are the Iowa Electronic Market (IEM) and PredictIt. These aren’t the only ones; however, the data (market prices) for these are easy to come by and publicly available via
their websites.
IEM was started in 1988 by the University of Iowa. Since then, over 100 other universities around the world have enrolled including Harvard, MIT, Michigan and
Northwestern. According to Crunchbase, PredictIt is a venture-backed private company founded in 1996.
Time after time, academic research has shown prediction markets to be more accurate than polls when predicting election
outcomes.
advertisement
advertisement
To understand why prediction markets are more accurate than polls you need to look at the methodologies. The former is based on the wisdom of crowds, the latter on
probability theory.
Probability requires random sampling and it’s getting more difficult for pollsters to be able to survey random populations of people. According to the Pew
Research Center, cooperation rates for phone surveys have declined to 6 or 7 percent. In 1997, they were closer to 36 percent. This causes pollsters to have to rely on sample weighting or other
hybrid-collection methods to make samples more representative.
Prediction markets aren’t reliant on random samples as they’re not based on probability. It’s better
for them to have respondents that are good at making predictions than it is to have a representative market. They’re also Darwinian -- poor predictors lose their money and leave the market while
good predictors stay longer and have more money with which to transact.
With a poll, people answer survey questions. With a prediction market, people buy and sell shares based on
whether they think something will happen. The trading activity provides market prices that are the basis for predictions. As a result, prediction markets are less likely to be impacted by response
error.
People aren’t always honest in polls or surveys, nor can they accurately recollect events in every case. There have been studies where people were asked questions about
whether they watched ads on TV and when their answers were compared to actual TV viewing information it showed them to be wrong around 30 percent of the time.
Prediction markets have
consequences. If people provide incorrect information, they lose money. When they’re right, they win money. There is incentive to be honest and accurate. This isn’t the case with a poll or
a survey.
Polls ask a different question than prediction markets. Polls often ask, “if the election were held today, who would you vote for?” Prediction markets will more
often ask “who will win the election.” These are two fundamentally different questions.
In the case of polls, the pollsters are asking what would happen in the
present. With prediction markets, the researchers are asking what will happen in the future.
As is so often the case in research, the poll and prediction market results aren’t
completely aligned as it relates to the 2024 U.S. presidential election. Most of the polls we’re looking at have Trump leading or Trump and Biden running neck and neck. These include Harvard
CAPS / The Harris Poll, Beacon Research / Shaw & Co. Research / Fox News and YouGov / The Economist polls. PredictIt prices agree with the polls for its candidate specific market.
However, when you look at both PredictIt and IEM generic ballot market prices to predict whether the Democrats or the Republicans will win the 2024 U.S. presidential election exclusive of
the candidates, both come out in favor of a Democratic win: IEM with a 0.68 DEM vs. a 0.32 REP and PredictIt with a 0.54 DEM and a 0.47 REP.
Without more information, it’s
difficult to definitively know the reason for this anomaly. We can speculate that some people may not think Biden will make it another term due to his advanced age and another Democrat could win in
his place. According to a recent Harvard CAPS / Harris Poll Survey (Dec 13-14, 2023), 62% of respondents had doubts about Biden’s mental fitness to serve as president.
One
reason why you might question the polls as well as PredictIt is that academic research shows that IEM is always better than polls at predicting election outcomes when the election is 100+ days into
the future. The 2024 election is not only a 100+ days out, but also almost a year away.
It doesn’t make a lot of sense to get too upset about the polls or the prediction
markets at this stage in the run up to the election. A lot is going to happen between now and then and it is very difficult for either methodology to make accurate predictions this far in advance.
On the other hand, if you have decisions to make and must guess one way or the other, it would be prudent not to bet against the prediction markets, and more specifically, not IEM.