Forecasting with Prediction Markets: Capturing the Wisdom of your Employees

Contributed by: Jasmien Lismont, Jan Vanthienen, Wilfried Lemahieu, Bart Baesens

This article first appeared in Data Science Briefings, the DataMiningApps newsletter. Subscribe now for free if you want to be the first to receive our feature articles, or follow us @DataMiningApps. Do you also wish to contribute to Data Science Briefings? Shoot us an e-mail over at and let’s get in touch!

The price traded at the stock exchange is designed to reflect the true value of shares. Traders buy and sell based on whether they belief stocks will go up or down. As such, it’s a continuous forecasting of what a company is worth. But what if we would trade on something else than stocks, e.g. events?

In 2005, Google implemented prediction markets for their employees [1]. This mechanism -which also goes by the name of “information market” or “event futures”- allows people to trade on events such as demand forecasts, internal performance but also external events related and not related to Google (e.g. about the quality of the new Star Wars). In 2002, Hewlett-Packard reported that their prediction market beat the official HP forecasts [4]. And with many others implementing this new way of forecasting, e.g. Best Buy, Eli Lilly, General Electric, Microsoft and Siemens [2], the popularity of prediction markets in the business sector is growing.

How does it work? In the most common system, a winner-take-all contract with continuous actions of bidders and sellers, the market price reflects the probability that an event will occur. For example, imagine we’re betting on whether the Red Devils, the Belgian football team, will win its first match in the European Championship. Upon closing of the market, the winners get a pay-off of €1 per contract and losers get nothing. If I believe chances are low they’ll win, around 20%, then I’ll buy contracts until the price is €0.20. If the price increases, because other participants believe that the chance that the Red Devils will win is higher, I can sell my contracts. A participant will only take action (buy/sell) when the price is different from his belief. As such, the final price will reflect the mean belief of the traders. In this example, this mean belief reveals the probability that the Red Devils will win. It captures the “wisdom of the crowd”.

Off course, other contracts and market designs exist and companies need to carefully reflect on the details. The design and whether participants trade with play or real money will determine the flexibility of your market (think of US gambling laws which might create barriers). The type of contract will tell us what the market price reflects: the probability of occurrence (winner-take-all contract); the mean value of the outcome (index contract); or the market’s expectation of median outcome (spread contract). Finally, the set-up and incentive will also affect whether your employees will be motivated to participate. When Google implemented their prediction market, they decided to work with play money, “Goobles”, which could be exchanged for prizes. This allows some flexibility in comparison to trading with real money and still provides a great incentive for their “Googlers”. If you’re interested in experiencing a prediction market yourself, we recommend a visit to the Iowa Electronic Markets [3]. IEM are small-scale, real-money markets developed for research purposes at the University of Iowa. Starting off in 1998, IEM are also one of the oldest and well-known prediction markets.

Why start with prediction markets? You probably already have some form of forecasting implemented in your company. And this might be a very good tool for doing so. Nevertheless, researchers found that prediction markets outperform polls, expert surveys and Delphi panels in accuracy. Moreover, they are less costly, especially now soft- and hardware costs are decreasing, and easily scalable. They aggregate the knowledge of your employees in a continuous, real-time manner which contrasts with conventional techniques solely based on historical data. Finally, you motivate employees to engage in a new way. However, we need to keep in mind that we’re working with human beings who not always act rational. Participants might overprice their favourites (e.g. an Apple fan bidding on the new IPhone sales), often perceive extreme outcomes as less plausible or might be overly optimistic. But in the end, it offers an interesting way to aggregate the opinion of many. As such, it’s a great technique to complement your traditional decision-making process.

To conclude:

  • Prediction markets are an interesting mechanism to complement traditional forecasting.
  • It captures the knowledge and believes of your employees and, as such, transforms tacit insights into valuable information.
  • It creates new opportunities for continuous, real-time forecasting but also for new product development (“idea markets”) or other forms of information aggregation.
  • Before implementation, companies need to carefully reflect on the design of the market.


  1. Cowgill, Bo, Wolfers, Justin, and Zitzewitz, Eric. 2009. Using prediction markets to track information flows: Evidence from Google.
  2. Graefe, Andreas, Luckner, Stefan, and Weinhardt, Christof. 2010. Prediction markets for foresight. Futures, 42(4): 394-404.
  3. Iowa Electronic Markets,
  4. Plott, Charles R., and Chen, Kay-Yut. 2002. Information Aggregation Mechanisms: Concept, Design and Implementation for a Sales Forecasting Problem. Social Science Working Paper, 1131. California Institute of Technology , Pasadena, CA.
  5. Wolfers, Justin, and Zitzewitz, Eric. 2004. “Prediction Markets.” Journal of Economic Perspectives, 18(2): 107-126.