Prediction Markets

In 1988, researchers at the University of Iowa made a puzzling discovery. They ran a stock market game where faculty, students and staff traded virtual stocks for presidential candidates in the 1988 U.S. presidential elections. The shares could be freely bought and sold by players and would redeem at a price equivalent to each candidate’s actual vote share on election day.

Participants in this game were not representative of the U.S. electorate. Amongst other differences they were younger and more intelligent than the average voter, they were largely male and Caucasian, they were better educated and typically came from wealthier households (Berg, 2005).

Representativeness did not matter because the researchers’ purpose was a study on trader behavior. What the found was that the traders behaved in ways not always fully rational. An analysis showed that many traders built portfolios biased by their personal preferences, they often did not take advantage of the best prices available in the market, and they quite frequently made trading mistakes.

Several cardinal rules of sound market research were all but ignored. Market participants were not selected randomly, and their participation was not anonymous.

No pollster would have dreamt of questioning a crowd of this composition rather than a carefully stratified random sample and expect an accurate forecast of the election result. The participants were an arrogant bunch, too: 90% believed that they were better informed about politics and more skilled than the other traders.

Imagine the researchers’ bafflement when - despite all these departures from traditional prescriptions - they compared the share prices resulting from the game to the actual election outcome: the prices outperformed the majority of national polls by a significant margin. Incredulous, they repeated the game over and over, for more than a decade (see Figure 1). Slowly it became clear that they were on to something: after analyzing nearly 600 traditional national polls from 1988 to 2000 they found that their prediction market was more accurate in 76% of cases - three times out of four. Further, accuracy increased as traders gained experience in subsequent rounds.

Predictive markets started to make euphoric headlines. Still, it was entirely unclear how and why virtual stock exchanges could achieve this feat. They just worked. Researchers started to realize how little they actually knew about the trading and price mechanism which arguably is the informational backbone of today’s wealthiest market economies.

Next: First commercial promise