The U.S. presidential elections offer social scientists and statisticians many avenues for dissecting the mood of the nation. Among the well-publicized polls and surveys conducted by well-known and well-funded organizations, a lower-key method of capturing the likely outcome of the electionprediction marketsis steadily gaining attention from academic researchers and business leaders for use beyond elections, movie box-office earnings, and sporting contest outcomes.
Like other futures markets, prediction markets offer participants the opportunity to trade on their hunches, the difference being that a prediction market offers payout odds based on aggregate hunches of forthcoming events instead of prices.
Prediction markets are gaining interest because the Internet allows greater worldwide access to them, as well as to the ever-increasing amount of data stored on any topic imaginable (which theoretically allows participants to make more informed predictions, individually and in aggregate). These factors, plus the enormous amount of computing power that will make it possible to instantly calculate exponentially small odds, are stimulating new research on advanced computational models in prediction markets. These models could be capable of analyzing entire events such as the annual NCAA collegiate basketball tournament, which begins a 63-game schedule with 263 possible outcomes by the tournament's end.
"I still think it's a growth area," says David Pennock, a principal researcher at Yahoo!, who is working on expanding the capabilities of prediction market outcomes. "Yes, prediction markets get lots of attention every four years during a presidential election, but every election cycle, they get more attention than they did the previous one. The perception of them is growing, startup companies using prediction markets are emerging, and there are a lot of research questions and industry growth still."
In fact, Pennock says, the U.S. Commodities Futures Trading Commission (CFTC) is considering expanding the use of prediction markets beyond low-budget research functions or "play money" markets to regulated public exchanges similar to the world's largest prediction market, Ireland-based Intrade.
"The request for comments was actually very well written and it's clear they understand a lot of the issues," Pennock says. Even if public prediction markets for substantial sums are not approved in the U.S., the markets offer considerable promise for enterprise planners who want the latest information on questions such as a product's likely launch date or revenue projections, and public policy forecasters, who can design markets exempt from CFTC oversight.
Growing opportunities in internal private-sector prediction markets are also revealing divergent philosophies among the markets' designers. Many of the public markets feature price-adjustment algorithms built around answering discrete multiple-choice outcomes, such as which candidate will win an election or if a product will launch in month x, y, or z. However, Mat Fogarty, CEO of prediction markets startup Xpree, says enterprise clients need to address questions expressed as continuous variables, such as a date range in which a product will launch or how many units will sell, and those markets need to feature an intuitive interface that encourages participation among those without a great interest in financial or mathematic complexities. The front end of these new prediction markets, as designed by Xpree, will feature interfaces inspired by computer game design, while the back end will replace multiple-choice algorithms with automated market makers based on Bayesian probability, enabling participants to place bets on a range of options.
The pioneering, modern public-policy prediction market, the University of Iowa's Iowa Electronic Markets (IEM), is now 21 years old and still offering new events for traders to forecast. First used in the 1988 U.S. presidential election, the IEM has offered markets on congressional elections, federal monetary policy, and inspired university colleagues to run a prediction market on national influenza infection trends. The IEM's unique design also inspired the latest corporate prediction market, a virtual-money internal market operated by Google.
IEM steering committee member Thomas Rietz, a professor of finance at the university, says the aggregate zero-risk design of the IEM allows the markets to perfectly reflect the aggregate forecast opinions of its participants. By aggregate zero-risk, Rietz explains that when a trader enters a particular bilateral (either/or) market, he or she must buy one share of each choice, called a bundle, for a total cost of $1. If the trader holds the bundle until the market concludes, there is neither profit nor gain. If the trader guesses the outcome successfully, and sells the losing unit of the bundle to another trader while the market is running, he or she picks up the original $1 bet plus whatever price was agreed upon for the losing share that was sold. If the trader chooses to hold onto the loser and sell the eventual winner, however, they will incur the $1 loss at payout time. At any given time, the number of eventual winning shares and losing shares is equal and held by the traders. So, the university bears no counterparty risk and there is no need to provide hedging margins that irrationally affect outcomes.
The most visible enterprise use of prediction markets is to help companies improve product and process development.
"The price you would be willing to buy or sell for today is your expectation of its value in the futurethe prices can be directly interpreted as a forecast," Rietz says. "In ordinary futures markets, there is a long-lasting debate, going back to John Maynard Keynes in the 1930s, over whether prices can legitimately be used as forecasts, and it all hinges on whether or not people demand a return or face a risk in aggregate when they're investing in these contracts."
The enterprise markets are offering intriguing design opportunities, as expressed by Xpree's Fogarty, as well as possible benefits beyond mining collective beliefs of what may make a successful product. The Google prediction market, for example, was examined by Bo Cowgill of Google, Justin Wolfers of the University of Pennsylvania's Wharton School of Business, and Eric Zitzewitz of Dartmouth College as a vehicle for the way information flows within an organization. Prediction markets, they assert, provide employees with incentives for truthful revelation and can capture changes in opinion at a much higher frequency than surveys, allowing one to track how information moves inside an organization and how it responds to external events. Proactive managers can use the analysis of those information flows to reorganize the company, if necessary, says Wolfers.
"A problem for economists is you can't measure information flows, and a market actually kind of makes those flows measurable," Wolfers says. "I would never suggest you set up a prediction market just to learn about the sociology of your organization. But it tracks, and can also change, how organizations operate."
Although Wolfers concedes the most visible enterprise use of prediction markets is to help companies improve product and process development, he also says, "As an economist, I am much more enthusiastic about how prediction markets could help in producing better public policies."
One public policy market that is gaining momentum is the University of Iowa's Iowa Health Prediction Market, funded by a $1.1 million grant from the Robert Wood Johnson Foundation. The market supplies invited healthcare professionals with $100 to begin trading their forecasts on flu activity in the coming season (winners are allowed to spend their trading earnings on professional advancement, thereby reducing public opprobrium about people profiting from others' illness).
Improving the flu markets' utility will entail expanding the regions the markets cover, and also tackling the most challenging computational issues facing prediction market designerscreating combinatorial markets that allow a much wider range of possible outcomes, and more granular expression of them, than the traditional win-lose, bilateral markets such as election markets. Yahoo!'s Pennock is experimenting with multiple examples of these combinatorial markets, which allow both conditional "if" questions and conjunctive "and" questions to be combined in virtually unlimited multiple arrangements.
For the flu market, which Pennock says he has discussed with the Iowa researchers, a combinatorial interface would allow traders to bet on more than the expected severity of outbreaks in one region.
With a combinatorial interface, he says, "you would choose a region of the country and choose a date range, and then also choose an outbreak range. This is a combination of things you think will happen'In this region, during this time frame, flu outbreak level will be red.' And the market will price it for you."
Combinatorial markets allow a wider range of outcomes and a more granular expression of them than traditional bilateral markets.
One enduring research problem on combinatorial markets is mitigating the effects a virtually unlimited spectrum of outcomes will have on creating markets that are so thin in trades they do not serve their purpose of aggregating information.
In such markets, which might bear a resemblance to an enterprise prediction market in that there are not enough participants to provide a statistically valid spread of opinion, Pennock says a market-maker algorithm might serve as a price setter within widely acceptable limits.
"I believe that approximation algorithms will be fine for the market maker, because people don't really care about making bets on things that are incredibly unlikely, like 106 chance," Pennock says. "But as long as you're betting on something with a 10% chance of happening, we'll be able to approximate pretty quickly with a market-maker price."
Pennock says the continuous increase of computational power is making advanced research into some of these exponentially based markets feasible. "I don't think it would have happened 10 years ago," he says. "The horsepower to do a good approximation is somewhat more recent."
©2009 ACM 0001-0782/09/0300 $5.00
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee.
The Digital Library is published by the Association for Computing Machinery. Copyright © 2009 ACM, Inc.
No entries found