Prediction Markets and the Flow of Information at Google

For 2.5 years, Google operated the largest experiment with internal corporate prediction markets in existence. Find out what they learned about their markets, their business, and their organization.

A preview of the talk was shared in a Google blog post on the subject:

• Traders in the same location tend to make the same trades at the same time. The trades of cubemates within a small radius is the best predictor we found. By using a record of historical office changes, we could observe that the correlation begins shortly after people are seated nearby. It makes sense, because the physical proximity enables easy communication. As Eric Schmidt (Google CEO) and Hal Varian (Google Chief Economist) advised in 2005: “The best way to make communication easy is to put team members within a few feet of each other. No telephone tag, no e-mail delay, no waiting for a reply.” As you can see below, our finding about the importance of proximity holds, even once we account for many other factors.

• Although we did find strong correlations among professional and social contacts, these were substantially weaker than the correlations for micro-geography. We also measured the influence that people on similar projects, in similar places in the organization, and with similar demographic characteristics exert on each other. This helped establish that geographic proximity—and not some other type of similarity—was responsible for the correlations we saw.

• Despite the markets’ strong forecasting abilities, there is a slight optimistic bias driven mainly by new employees. On average, outcomes that were good for Google were overpriced by 20%. This bias was strongest on days after appreciations in Google stock and, ironically, for outcomes under our own control! We also find biases against extreme outcomes and short selling. Given a range of five outcomes, the middle ones were typically overpriced and unprofitable by comparison with the outliers.