In the spring of 1988, three economics professors at the University of Iowa — Robert Forsythe, Forrest Nelson, and George Neumann — were teaching a class on financial markets and decided to try an experiment.
They asked their students to bet small amounts of real money — capped at $500 per person — on the outcome of the 1988 U.S. presidential election. The students traded contracts that paid out based on the share of the vote each candidate would receive.
The market opened in June. By Election Day in November, it had outperformed every major polling organization in the country.
This is the story of how prediction markets started — and why a $5 stake from a graduate student turned out to be a better signal than a Gallup survey of 1,500 randomly-sampled Americans.
The setup
The market the professors built was simple. Each candidate had a contract whose final payout was tied to their actual share of the popular vote. If you bought a Bush contract at 53¢ and Bush received 53% of the vote, you broke even. If he received 54%, you made 1¢ per share. If he received 51%, you lost 2¢.
The contracts traded continuously between June and November. The price of each contract was, by construction, the market's live estimate of what each candidate's vote share would be.
And here's the part the professors didn't expect. The price was almost exactly right.
The result
Bush won the 1988 election with 53.4% of the popular vote. Dukakis received 45.6%.
The Iowa market's average prediction in the final week was Bush at 53.2%, Dukakis at 45.2%. The error was less than half a percentage point on the winning candidate.
Gallup, the highest-respected pollster of the era, predicted Bush at 56% and Dukakis at 44%. They were off by 2.6 points.
The New York Times / CBS poll was off by 1.8 points. The Wall Street Journal / NBC poll was off by 1.4 points. The CNN / Time poll was off by 2.2 points.
A market run by university students with five-dollar stakes was more accurate than every professional polling organization in the United States.
Why the professors didn't expect it
The professors had originally set up the market as a teaching exercise. They wanted to demonstrate to their finance students how price formation works in a small, contained environment. They expected the prices to be roughly directionally correct — Bush would be priced higher than Dukakis — but they did not expect the market to beat the polls.
What they discovered, by accident, was that a small group of moderately informed people, betting their own money on calibrated outcomes, produced a more accurate forecast than a much larger group of randomly-sampled people answering survey questions.
The mechanism was the same one that makes any market accurate: information-weighted aggregation. The students who knew the most about politics — who had read more, who watched the debates more carefully, who had family connections to the campaigns — bet more. The students who didn't know much stayed on the sidelines or bet smaller. The market price reflected the views of the most-informed participants more heavily than the views of the least-informed.
Polls don't do this. A poll weights every respondent equally, regardless of what they know.
What happened next
The 1988 result was published as an academic paper in 1992. It received attention in the economics community but not much beyond. The professors continued running the market — renamed the Iowa Electronic Markets, or IEM — for every U.S. presidential election since.
The IEM beat the polls in 1992. It beat them in 1996. It beat them in 2000. The pattern held across nine presidential cycles. By 2008, the IEM had a documented track record of being more accurate than every major polling organization, on average, by a margin of roughly 2 percentage points.
Academic papers piled up. Robin Hanson, a professor at George Mason, started writing about the broader implications. By the early 2000s, the U.S. Department of Defense had quietly explored whether prediction markets could be used to forecast geopolitical events. (The program, called FutureMAP, was killed by Congress when the public learned about it.)
By the time Polymarket and similar platforms launched in the 2020s, the academic case had been made for thirty years. The Iowa market was where it started.
Why the result mattered
The Iowa Electronic Markets did one thing that no amount of theorizing could have accomplished. They produced a clean, repeated, public demonstration that information-weighted markets outperformed information-flat polls.
This wasn't a clever finding from one election. It was a multi-decade record of consistently beating the most credentialed forecasters in American politics, using small-stakes traders.
If you had to design a single experiment to validate the entire intellectual case for prediction markets, you couldn't have designed a better one. The professors did it almost by accident.
The lesson, three decades later
What the Iowa experiment proved is not that "the crowd is wise." That's a lazy summary. The crowd, in the form of a poll, is not particularly wise — it's average.
What's wise is the crowd weighted by capital. When the price of a contract is determined by the people who are willing to put their money on the outcome, the prediction is shaped most strongly by the participants who actually know something. The rest cancel out.
That's the entire intellectual move that distinguishes prediction markets from polls, surveys, and panels. Three economics professors discovered it in a classroom in 1988. The principle has been re-validated thousands of times since.
That same principle — aggregated market prices out-predicting expert consensus — is the foundation that Juno's prediction market platform is built on, bringing real-money forecasting to Thailand for the first time.
It's a useful thing to remember the next time someone tells you that the polls say one thing and the markets say another. The markets are usually right.