For the past two and a half years, Google has been asking its employees to make their predictions and bets on everything from their perception of projects, to the predicted number of Gmail users by the end of the year. Googlers make the bets anonymously with fake Google “money”, which allows Google to track the opinions of its employees, as well as give them insight into the way information flows internally.
In the time that the program has been running, 1,500 employees bet on the outcome of 280 questions as a part of their corporate predictions program. Questions may have included things like “Will Google open its Russian office on schedule?”, “Will Harriet Miers be confirmed as Supreme Court Justice?”, and so on. Top earners in the market get their name posted on the leader board, and are awarded cash prizes or t-shirts for having made correct predictions.
The purpose of the program is to determine employee opinions on questions about the demand for a future Google product, or how the company will perform during a specific time period. By allowing employees to participate anonymously, Google believes that they eliminate any reason to lie or kiss up in responses, and that their employees are more likely to give honest answers.
The results of the project are quite interesting. According to Bo Cowgill who runs in the internal prediction market, what Google found is that proximity beats out all other factors in determining how people bet. Those who work together, or even share an office, are more likely to bet in similar patterns. Other factors like job rank, internal email list subscriptions, and projects that people have worked on in the past just don’t seem to be as much of an indicator as one’s location and proximity to others. They also found that Google employees are more likely to make the right bet than the wrong one.
Google also found some biases among employees based on their betting patterns. New hires were found to be more likely to overprice bets on matters that would be good for Google, and bets made after a rise in Google’s stock price tended to find the bettors predicting positive things for Google.
Next up for the project, Google would like to explore how one’s knowledge of or work on specific products affects their betting patterns.
Cowgill discussed the prediction market on Tuesday at the O’Reilly ETech conference on emerging technology in San Diego, CA. Along with economists from Dartmouth College and the University of Pennsylvania, Cowgill published a paper on the experiment in January. If you’re interested in reading more of the finer details, you can check out their paper here.