DIMACS Workshop on Markets as Predictive Devices (Information Markets)

February 2-4, 2005
DIMACS Center, CoRE Building, Rutgers University

Robin Hanson, George Mason University,
John Ledyard, California Institute of Technology,
David Pennock, Yahoo! Research Labs,
Presented under the auspices of the Special Focus on Computation and the Socio-Economic Sciences, and the following sponsors:

Microsoft Research: http://research.microsoft.com

Newsfutures: http://www.newsfutures.com Hosting PM2, the Prediction Market Market

Yahoo! Research Labs: http://research.yahoo.com


John Ledyard, Robin Hanson, and Takashi Ishikida, Caltech, GMU, Net Exchange

Title: An experimental test of combinatorial information markets

While a simple information market lets one trade on the probability of each value of a single variable, a combinatorial information market lets one trade on any combination of a set of variables, including any conditional or joint probability. In laboratory experiments, we compare the accuracy of simple markets, two kinds of combinatorial markets, a call market and a market maker, and isolated individuals who report to a scoring rule. We consider two environments with asymmetric information on sparsely correlated binary variables, one with three subjects and three variables, and one with six subjects and eight variables (and so 256 states). We also introduce a distraction, a rewarding activity subjects can choose instead of trading. JEL: C92,D82,G14

George Neumann, University of Iowa

Title: Operating With Doctors: Results From the 2004 and 2005 Influenza Markets

National and international surveillance networks collect vast amounts of information about influenza each year. However, these agencies lack the means to aggregate this information and predict the course of influenza activity. An accurate forecast of influenza activity would be helpful on global, national and, regional levels. If local health officials could predict an outbreak, even 1-2 weeks in advance, they could initiate measures to modify the severity of the outbreak. For example, public health officials could allocate resources to increase vaccination rates among both high-risk individuals and healthcare workers. Prophylactic medication, such as neuraminidase inhibitors, could be administered to persons in high-risk groups, for example residents of nursing homes and other closed populations. In addition, if increased influenza activity could be predicted, hospital administrators could plan ahead and ensure that adequate staff and resources are available to care for an increased number of patients admitted for complications of influenza. As a tool for generating such predictions we conducted two prediction markets designed to forecast the severity of influenza in the state of Iowa. In 2004 the markets ran for 4 months, covering 5 two week intervals of flu. All participants were medical personnel (M.D.'s, nurses, or Ph. D.'s) employed at the University of Iowa Hospitals and Clinics. Rewards were made in ersatz currency. The 2005 market ran for 25 weeks and was open to any health care professional in the U. S., and rewards were made in cash. This paper describes how well the market mechanism works in a non-traditional setting and contrasts the use of actual cash payments with ersatz currency.

Justin Wolfers, Business, University of Pennsylvania

Title: Interpreting Prediction Market Prices as Probabilities

While most empirical analysis of prediction markets treats prices of binary options as predictions of the probability of future events, Manski (2004) has recently argued that there is little existing theory supporting this practice. We provide relevant analytic foundations, describing sufficient conditions under which prediction markets prices correspond with mean beliefs. Beyond these specific sufficient conditions, we show that for a broad class of models prediction market prices are usually close to the mean beliefs of traders. The key parameters driving trading behavior in prediction markets are the degree of risk aversion and the distribution on beliefs, and we provide some novel data on the distribution of beliefs in a couple of interesting contexts. We find that prediction markets prices typically provide useful (albeit sometimes biased) estimates of average beliefs about the probability an event occurs.

Previous: Program
Workshop Index
DIMACS Homepage
Contacting the Center
Document last modified on January 12, 2005.