Forecasting a Possible Human-to-Human Bird Flu Outbreak Using a Commodities Futures Trading Model
Researchers at the University of Iowa Henry B. Tippie College of Business launched an "Avian Influenza Market," a "prediction market" modeled on the principles of commodities futures trading and designed to aggregate information about the H5N1 virus useful to forecasting the timing and extent of a possible human-to-human bird flu outbreak.
The Avian Influenza Market linked public health experts worldwide via the Internet, who "traded" in futures contracts tied to specific H5N1 milestones (e.g., whether a given number of avian flu cases were confirmed by a certain deadline). Researchers say the contract price reflects the consensus among traders about the likelihood that the event will occur.
- Researchers defined and conducted trading in a total of 52 contracts concerning avian flu-related events beginning in February 2007 and extending for two years until January 1, 2009.
- Contract prices in this market were, according to project researchers, "by and large consistent with the eventual outcomes" of the events they were designed to track.
The Robert Wood Johnson Foundation (RWJF) supported this project with a grant of $187,465 from August 2006 to April 2010.
During the mid-2000s, many experts believed an influenza pandemic involving the H5N1 virus (avian flu) was inevitable (see Webby and Webster, "Are We Ready for Pandemic Influenza?" Science, 302: 1519–1522, 2003). In addition, reports in a number of medical journals indicated that H5N1's pandemic potential was increasing (see "Avian Flu Should Be Ruffling Our Feathers," The Lancet Infectious Disease, 4(10): 595, 2004).
Whereas information was available that could help public health authorities predict the timing of a new flu pandemic, the data were in disparate forms and geographically dispersed, making available methods, such as traditional infectious disease surveillance systems, ineffective in efficiently collecting and interpreting these data. As a result, the public health and medical communities needed alternative tools to help forecast the spread of the H5N1 virus.
The Development of Prediction Markets as a Forecasting Tool
"Prediction markets" are a promising forecasting method modeled on the principles of commodities futures trading. Commodities futures traders buy and sell contracts based on their beliefs about the future value of a commodity, such as oil. Traders profit or lose depending upon the outcomes of their investments—for example, if oil increases or decreases to the value stipulated in the contract.
In prediction markets, there are no commodities or raw materials. Instead, traders buy and sell futures contracts whose ultimate value is based on some future event that may or may not occur. Researchers say the contract price reflects the consensus among traders about the likelihood that the event will occur.
In a weather prediction market, for example, meteorologists and others might bid on a contract with a face value of $1 that reads, "Snow will begin falling by noon tomorrow." If the contract price closes at 75 cents, that can be taken to mean that traders believe there is a 75 percent chance that it will snow and that the contract will be worth $1. It also means that one-quarter of traders take the opposite view.
The concept of prediction markets was first fully developed by investigators at the University of Iowa Henry B. Tippie College of Business. Since 1988, its Iowa Electronic Market (IEM) has run markets for currency prices, stock options, elections around the world and movie box office receipts, and it has achieved a prediction record superior to alternative methods (see Forsythe, Nelson, Neumann and Wright, "Anatomy of an Experimental Political Stock Market," American Economic Review, 82: 1142–1161, 1992).
Prediction markets also have been used successfully as forecasting tools by industry. For example, Eli Lilly Company has used internal markets to predict which developmental drugs have the best chance of advancing through clinical trials. Hewlett-Packard has used experimental markets to forecast sales of its printers, and these markets have outperformed statistical sales forecasts (see Pennock DM, et al., Science, 291: 287–288, 2001).
According to Project Director Forrest Nelson, Ph.D., many health care workers are the first to know about influenza activity in their communities and that information can help predict the course of an infectious disease. A prediction market in avian flu could tap into that dispersed knowledge to provide important new information about the development and spread of the H5N1 virus, and better ready scientists and public health authorities to protect the public both here and abroad.
In summer 2005, the Iowa Electronic Market launched a seasonal flu market with funding from RWJF (Grant ID# 053352). As of March 2010, the market was still operating. For more on the market, see the Pioneering Ideas Blog.
Researchers at the Iowa Electronic Health Market (an offshoot of the Iowa Electronic Market, dealing exclusively in health-related futures) launched the "Avian Influenza Market," an information trading and aggregation system designed to help public health officials around the world collect and analyze information to forecast the timing and extent of a human-to-human bird flu outbreak.
To accomplish this, researchers:
- Identified critical H5N1 events that would help predict the likelihood of an avian flu pandemic. They converted these into a series of "yes/no" questions—for example, "Will there be 300 WHO-confirmed human cases of H5N1 by 7/1/2007?"—that formed the basis of the six-month futures contracts traded on the Iowa Electronic Health Market. (For a description of other contracts, see the Appendix.)
Editors of ProMED-mail (a worldwide Web-based network of health care workers who share and disseminate information on outbreaks of infectious diseases and exposure to toxins) assisted with the identification of critical H5N1 events and with the definition of the six-month futures contracts based on those events.
- Created a password-protected Web site on which trading would take place.
- Recruited traders. Traders came almost exclusively from the 34,000 members of ProMED-mail, representing a dozen different professions including epidemiologists, physicians, microbiologists and veterinarians. Trading was not open to the public.
- Traders received a password to the trading site, instructions in trading and 2,500 "avian dollars" for use in trading (equivalent to about $250). Traders did not receive this currency all at once, and they could earn added avian dollars for trading more frequently.
- The market opened for trading in February 2007. The researchers introduced four waves of the six-month contracts; participants continued to trade for two years.
- Each contract typically asked about an event that would or would not occur over the next six months—for example, "Will a new HIGH-PATH H5N1 case be confirmed by the World Organization for Animal Health (OIE) in a native non-captive avian species in North or South America by July 1, 2007?"
- All contracts had a face value of one avian dollar, although the value of a contract at any one time depended on trading of the contract—that is, the overall market price of the contract was determined after bidding by the traders on each contract. The researchers said that the market price for a particular contract reflected the consensus belief among the traders that the underlying event would come to pass.
- In August 2007, and again after the close of the last contracts in January 2009, traders were "cashed out"—reimbursed for their participation in real dollars in proportion to the avian dollars they held.
Researchers contracted with Larry Madoff, an editor at ProMED-mail, to act as co-investigator on this project. He and other ProMED-mail editors helped assemble the questions that would define the avian flu contracts.
Researchers presented posters about the project work at:
- The 45th Annual Meeting of the Infectious Diseases Society of America, San Diego, October 4–7, 2007
- The International Conference on Emerging Infectious Diseases 2008, Atlanta, March 16–19, 2008
- The International Meeting on Emerging Diseases and Surveillance, Vienna, Austria, February 13–19, 2009
Burness Communications, a Bethesda, Md.-based public relations firm, funded by RWJF's Pioneer team's Communications Fund, prepared and disseminated a press release about this project's avian flu prediction market in March 2007. Some of the resulting coverage may be viewed or heard at http://fluprediction.uiowa.edu/fluhome/Media.html.
According to a 2009 project report to RWJF:
- The researchers defined and conducted an Avian Influenza Market that began in February 2007 and extended for two years until January 2009.
- Fifty-two pairs of six-month yes/no contracts were introduced on the market in waves. The last four waves of contracts overlapped and extended beyond the grant term, ending on January 1, 2009.
- A total of 210 active traders participated in the market. The number of traders who remained active declined over time, dropping from the initial 210 to 60 in the fourth wave of contracts. However, the level of trading remained sufficient to provide a suitable number of observations for the study, according to the researchers.
- Trader recruitment was not easy but was sufficient to the task, in the view of the researchers. Traders came from 33 countries and represented a dozen different professions including epidemiologists, physicians, microbiologists and veterinarians.
- The researchers report that contract market prices were "by and large quite consistent with the eventual outcomes" they were designed to track. In almost all of the cases, the contract representing the actual outcome of each event investigated traded at prices above 90 cents for at least two months before the close of trading—indicating a high degree of consensus among traders that the event would come to pass.
In one pair of contracts—concerning whether 350 human cases of H5N1 would be documented by January 1, 2008—there was little consensus among traders. This was reflected in prices that fluctuated between 17 cents and 73 cents over the last month of trading and closed at 53 cents on the last day of trading, December 31, 2007.
Approaching that last day of trading, in other words, traders as a group came to be divided equally in their opinion that 350 cases would appear by January 1, 2008. It happened that the 349th case occurred on December 30 and the 350th case occurred on January 3, as reported by the WHO two weeks later. According to the researchers, for the market to be so evenly divided "was exactly the correct prediction."
Researchers note that statistical tests of the performance of the market are not possible, given the limited time period and resulting small number of observations.
- Recruit traders from an existing and highly informed network, if possible. ProMED-mail, a 34,000-member group with sufficient interest, access and knowledge of local flu-related events, ensured creation of a high-functioning avian flu prediction market. "Even with access to members of that network, trader recruitment remained one of the most difficult tasks of the project." (Project Director/Nelson)
- Communicate frequently with traders to keep them active. Researchers note that weekly e-mails sent to traders appeared to promote market activity. "That was highly effective. We would get the weekly reminders sent off about Thursday. We noted about 90 percent of our trading was on Thursdays and Fridays!" (Project Director/Nelson)
The Avian Influenza Market ceased operating at the close of its last contracts, January 1, 2009. As of January 2010, the Iowa Electronic Health Market operated predictive markets in U.S. cases of syphilis and seasonal and H1N1 influenza.
GRANT DETAILS & CONTACT INFORMATION
Using electronic prediction markets to forecast avian flu activity
University of Iowa Henry B. Tippie College of Business (Iowa City, IA)
Dates: August 2006 to April 2010
Forrest Nelson, Ph.D.
Avian Flu Contracts Traded
The following avian flu-related contracts traded on the project market site from February 2007 to January 2009. All contracts were six months in duration. Where there is no accompanying number specification, the date below refers to when the contract began trading.
Date, or Number Specifications by Date
|Policy||Will Phase Four of the WHO-defined Pandemic Alert Period be announced before the specified date?||4||7/1/2007
|Will Phase Five of the WHO-defined Pandemic Alert Period be announced before the specified date?||1||7/1/2007|
|Will the World Health Organization issue an emergency travel advisory relating to H5N1 before the specified date?||3||1/1/2008
|Human Numbers||Will the number of WHO-confirmed human cases of H5N1 be as great as the specified number by the specified date?||5||300 by 7/1/2007
350 by 1/1/2008 400 by 1/1/2008
400 by 7/1/2008 425 by 1/1/2009
|Will the number of new WHO-confirmed cases of H5N1 in humans during the specified month be greater than the number of cases during the previous month?||4||Feb. 2007
|Will the number of new WHO-confirmed cases of H5N1 in humans during the specified quarter be greater than the number of cases during the corresponding quarter of the previous year?||4||Qrtr 1, 2008 Qrtr 2, 2008 Qrtr 3, 2008 Qrtr 4, 2008|
|Human Location||Will a new WHO-confirmed human case of H5N1 occur in North or South America before the specified date?||4||7/1/2007
|Will a new WHO-confirmed human case of H5N1 occur in Europe (as defined by the WHO Regional Office for Europe) before the specified date?||4||7/1/2007
|Will a new WHO-confirmed human case of H5N1 occur in Hong Kong before the specified date?||4||7/1/2007
|Will a new WHO-confirmed human case of H5N1 occur in Singapore before the specified date?||4||7/1/2007
|Will a new WHO-confirmed human case of H5N1 occur in Africa, as defined by the WHO Regional Office for Africa, before the specified date?||3||1/1/2008
|Will the total number of new WHO-confirmed cases of H5N1 outside both Indonesia and Vietnam be greater than or equal to the total number of new cases in the two countries before the specified date?||3||1/1/2008
|Animal||Will a new HIGH-PATH H5N1 case be confirmed by the World Organization for Animal Health (OIE) in a native, non-captive avian species in North or South America before the specified date?||4||7/1/2007
|Will a new HIGH-PATH H5N1 case be confirmed by the World Organization for Animal Health (OIE) in commercial poultry in North or South America before the specified date?||4||7/1/2007
|Will a new H5N1 case in domesticated mammals outside the continent of Asia be confirmed by the World Organization for Animal Health (OIE) before the specified date?||1||7/1/2007|
(Current as of date of this report; as provided by grantee organization; not verified by RWJF; items not available from RWJF.)
World Wide Web Sites
http://fluprediction.uiowa.edu/fluhome/Market_AvianInfluenza.html. The "avian flu" page of the Iowa Electronic Market's Web site. It gives information on how prediction markets work, the nature of the avian flu market at the Iowa Electronic Market (now Iowa Electronic Health Market, IEhM), press coverage about the avian flu market as well as how to register as a trader. During 2007 and 2008, when contracts traded in avian flu, the page also permitted entry to a password-protected trading site. Iowa City, IA: University of Iowa Henry B. Tippie College of Business.
http://fluprediction.uiowa.edu/fluhome/AvianFlu_prospectus.html. Prospectus for avian flu prediction contracts. Iowa City, IA: University of Iowa Henry B. Tippie College of Business.
Report prepared by: James Wood
Reviewed by: Richard Camer
Reviewed by: Molly McKaughan
Program Officer: Paul Tarini