In essence, all markets are prediction markets. The value of assets traded in a market depends on information that is not fully revealed and will not be known for some time, if ever. Market valuations are explicitly or implicitly predictions of that unknown information, perhaps the future value of a commodity, the expected cash flows generated by a firm, or the outcome of a potential corporate merger.

While commodities futures are often used as financial instruments to hedge long or short positions, the markets also reward traders with better information while punishing those with worse information. Giving traders incentives to reveal good information is the core function of prediction markets, even where no underlying assets, in the traditional sense, are available to be traded. Prediction markets trade future events or outcomes, and the settling process amounts to using a documented and published formula to determine winners and losers and to pay out incentives. Many experiments and real-world tests show that market mechanisms can be implemented simply to create predictions and that these systems work rather well.

Perhaps the best known of all prediction markets are the Iowa Electronic Markets, which enable traders to forecast the outcomes of future elections. The power of these markets to generate forecasts accurate and stable enough to inform decision makers has been demonstrated for nearly two decades [5]. Another set of experiments at Hewlett Packard demonstrated the ability of prediction markets to call future sales more effectively than traditional forecasting processes [6].

In our research at Intel we are extending the idea of prediction markets to create "forecasting markets," which are essentially prediction markets or similar IAMs integrated into the company's standard, ongoing forecasting processes. Participants reveal not just an expected outcome but a series of expected outcomes for the same variable over time. So, the forecasting market captures individual and collective assessments about trends such as increasing or decreasing demand just as weather forecasts anticipate warming and cooling trends.

We believe that three factors enable markets to outperform other types of forecasting systems and more effectively move information from source to decision maker. First, the features of anonymity and incentives work together to draw out good information. The experiments of Kay-Yut Chen and Charles Plott at Hewlett Packard suggest that people provide the best information when rewarded to do so and when protected from potential ramifications of expressing their honest opinions. Incentives encourage participants to search for the best information they can find and reward trading behavior that is unbiased. Anonymity helps prevent biases created by the presence of formal or informal power, the social norms of group interaction, and expectations of management. We found many individuals at Intel who told us that their opinions sometimes differ from stated targets or unstated expectations. Looking back at forecasts that were off substantially, we have been told that teams sometimes did not believe the forecast they published but were pressured, perhaps overtly, to adjust forecasts upward or downward. To the extent anonymity and incentives curb bias and motivate the hunt for good information, they should improve the signal created by market mechanisms.

Second, the simple mechanism of aggregating data through a survey or market has two remarkable properties. It smoothes results over time, which is great for guiding supply, and it tends to produce a group forecast more accurate than the forecast of at least a majority of individual participants. A study by Scott Page demonstrated that even among a fairly homogeneous group this effect holds true. In the context of forecasting selection order in professional sports drafts, he found that averaging the individual forecasts of experts soundly outperformed the forecasts of any individual [7].

Finally, in many forecasting examples it has been found that increasing the diversity of a pool of participants increases the accuracy of the collective forecast. As long as each additional participant brings some information, adding more, diverse opinions improves the collective judgment. This condition holds true in many cases because good information tends to be positively correlated and sums, while errors are often negatively correlated and cancel [8].