Here’s the second part of my notes from the paper, INFORMATION AGGREGATION MECHANISMS: CONCEPT, DESIGN AND IMPLEMENTATION FOR A SALES FORECASTING PROBLEM, by Charles R. Plott of CalTech and Kay-Yut Chen of Hewlett Packard Laboratories,which describes how they set up a prediction market for sales forecasts at HP.
Advantages of Prediction Market Over Other Forecasting Methods
· The methodology is flexible. It can be used to aggregate any type of information possessed by different people. It involves a natural methodology for quantifying subjective, qualitative, information and giving weights to the opinion of different people for the purpose of information aggregation. The task is performed giving not only a point forecast but also a complete probability over the range for which the value of some unknown variable is to be predicted.
· The methodology is scalable by number of participants, timing of participants and location of participants. There are no practical limits to the number of people that can participate. With markets conducted over the Internet, hundreds and even thousands of people can participate either at the same time or at different times. Traditionally, businesses collect and aggregate information through a process of meetings, which not only limits the number of participants but also the time frame for information collection.
· The methodology tends to be incentive compatible. Incentives to hide information, misrepresent information or simply ignore requests for information are either eliminated or limited. Furthermore the markets are designed to give incentives to the participants’ to acquire information about future events and use this information wisely in the market.
· Theoretical arbitrage profits existed. In all the experiments, prices summed to be greater than the winning payoff. However, to take advantage of the arbitrage conditions, individuals needed to execute multiple trades when fluctuations of prices were substantial. So it is likely that there were actually no practical arbitrage opportunities. Why in all 12 experiments was the sum of the prices always above the winning payoff?
· No significant trends in the sequences of predictions are observed. So it doesn’t appear there was any response to changing market information during the trading. Maybe all the information aggregated quickly at the beginning.
· How is the performance of the system related to the psychology and decision biases of individuals?
· How can one deal with incentive problems in which individuals might large incentives to conceal or misrepresent what they know?
· What rules and mechanisms might be needed for different underlying information structures?
· If markets are thin or the number of participants few, how will the performance of the system be affected?
· How can we find the people with the relevant information and how do we know that they knew something of relevance anyway? If the participants know nothing, the mechanism will produce nothing.
· Can a prediction market not only produce a prediction but also simultaneously help management ascertain which participants have information. That is, can it be designed to attract those with good information and discourage those with bad information?
The experimental demonstration is first found in Plott and Sunder (1982, 1988). This early paper demonstrated that the ability of markets to aggregate information is sensitive to the market architecture. In particular, this early work demonstrated that compound securities are not as reliable as indicators as a complete set of state dependent instruments. The conditions under which a single compound security is reliable are isolated in Forsythe and Lundholm (1990) The need for selecting proper instruments is underlined by demonstrations of markets that can equilibrate at patterns that are not fully revealing of information such as cascades (Anderson and Holt, 1997; Hung and Plott, 2001) or misleading such as mirages (Camerer and Wiegelt, 1991) or bubbles ( Smith et al, 1988; King et al. 1993; Porter and Smith, 994; Lei et al, 2001). In fact, some types of market organization facilitate no information aggregation at all as is the case of the winners curse in sealed bid auction markets (Kagel and Levin, 1986; Lind and Plott, 1991). See Sunder (1995) for a summary, or aspects of search (Sunder, 1992).