The Use of Prediction Markets to Forecast Influenza Activity
Phil Polgreen (University of Iowa)
(May 30, 2006 4:30 PM - 5:30 PM)
Although influenza occurs annually, unique characteristics particular to each influenza season make forecasting difficult. Each year the geographical locations, rates of increase and decline, duration, and size of each outbreak vary considerably. Statistical models using historical data may accurately describe the typical pattern for a particular year, but they do not predict departures from the norm. However, it is the deviations that are of the most concern and, therefore, the most important to predict. Nurses, physicians, epidemiologists, pharmacists and microbiologists all have access to unique data that could help predict future influenza activity. However, because of the disparate nature of this information, standard research and statistical methods cannot be used to aggregate and analyze it rapidly enough to ensure clinical relevance. In order to address this shortcoming, we ran an influenza prediction market in Iowa for the 2004-05 influenza season. Traders, who included a diverse mix of healthcare workers, were each given a $100 grant with which to buy and sell contracts that reflected their views on short-term future influenza activity. By aggregating expert opinion, we predicted the epidemic curve, up to 4 weeks in advance, more accurately than forecasts based on historical data. We believe that these methods would be of use in predicting the course and timing of other infectious disease outbreaks.