In defense of prediction markets
Alexander Tabarrok
In late July, word of a new Pentagon program headed by John Poindexter - allowing people to bet money on the odds of
such events as political assassinations and acts of terror - caused quite a
stir in the U.S. Senate. It was quickly "disestablished" by the military. The
concept, however, is little understood by the public.
The concept springs from a long history of open markets - from public stocks to
commodities exchanges. The insurance industry itself is a kind of market hedge
against tragic events.
ADVERTISEMENT | | We asked Alexander Tabarrok,
associate professor of economics at George Mason University in Fairfax,
Virginia, to explain why he thinks the idea should have been given a chance,
and why similar markets for predicting the future may still play a big role in
business and government.
In light of the massive failure of the U.S. intelligence agencies to avert the
events of 9/11, one would think that politicians would be eager to try to
develop better institutions for collecting data. Yet, when one such innovative
endeavor - the Policy Analysis Market (PAM) - was recently proposed, Senators
Ron Wyden (Democrat, Oregon) and Byron Dorgan (Democrat, North Dakota) labeled
it a "terror market." And before the idea itself could be discussed or
rationally debated, the program was terminated.
However, the concept of "information markets" or "prediction markets," has been
proven successful by industry. The best known is the
Iowa Political Stock Market. The Iowa market lets traders use real
money to buy and sell "shares" of political candidates. A share in George W.
Bush, for example, pays $1 if President Bush wins the next election and nothing
otherwise. If the market price for a Bush share is 75 cents, this suggests that
market participants think that President Bush has a 75 percent chance of
winning the next election.
The Iowa markets have passed the market test. The Iowa market has proven to be
more accurate than polls (see
"Results from a Dozen Years of Election Futures Markets Research" and
"Accuracy and Forecast Standard Error of Prediction Markets" for
details) in nearly 14 years of predicting U.S. and foreign elections,
primaries, and other political events. In tight elections, professional bond
traders - who often have millions of dollars riding on post-election economic
policies - monitor the Iowa markets for clues about future events.
Hewlett-Packard has used a similar market approach to help predict future
hardware sales. Members of HP's sales team bought and sold shares that paid off
when sales fell within a certain range. A typical security would pay out $1, if
and only if, future sales were, say, between 10,000 and 15,000 units. Another
might pay off if sales fell between 15,000 and 20,000 units. The market
contained 10 types of securities - a range broad enough to include all the
relevant possible sales outcomes.
By examining the prices of all 10 shares, HP could assign a probability to any
combination of outcomes - a more nuanced analysis than that available from a
questionnaire. They could determine, for example, the probability of sales
falling anywhere below 10,000 units or anywhere above 25,000 units.
HP subsidized participation in the information market, so the traders could not
lose money. But the traders could keep any money that they made, giving them
substantial incentive to trade carefully. HP compared the implicit forecasts
made by the prediction market with its own official forecasts and with actual
sales figures. In 15 out of 16 trials, the mean market-based prediction was
significantly closer to the actual sales figure than the official forecast. In
the one remaining trial, the mean market prediction and the official forecast
were equally close. Encouraged by these results, HP created its own
experimental economics laboratory.
Another advantage is that these markets encourage traders to put their money
where their mouths are not. The members of a company's sales team rarely
has incentive to tell their bosses that next quarter's sales are going to be
lower than expected.
Siemens used a similar internal prediction market to help forecast when a
project would be completed. The market correctly predicted a longer delay than
the official forecast, revealing a comparable reluctance on the part of
Siemens's staff to "tell it like it is." Prediction markets can help to
overcome the "yes-man" phenomenon, which makes it difficult for information to
rise from the field to the decision makers.
The Hollywood Stock
Exchange (HSX) is also proving that the innovative use of markets can
be profitable. The Hollywood Exchange lets traders buy and sell shares and
options in movies, music, and Oscar contenders. Trading on the Hollywood
exchange is conducted in make-believe "Hollywood Dollars," but the goal of HSX
- which is owned by a subsidiary of the trading firm Cantor Fitzgerald - is
profit. Some 800,000 people trading on HSX for fun have proven to be reliable
indicators of future film profits, and HSX is now selling its data to Hollywood
studios eager to improve their predictions about future blockbusters.
In Entrepreneurial Economics: Bright Ideas from the Dismal Science (Oxford
University Press, 2002), economist Robin Hanson of George Mason University
explains how information markets could be used to help answer public policy
questions. A market might be formed, for example, to predict the crime rate if
a piece of gun-control legislation passed or failed. (The PAM market was going
to forecast events such as total yearly casualties from terrorism if,
for example, the U.S. invaded Syria or withdrew its troops from Saudi Arabia.)
Such political information markets could replace the less accurate
information-gathering mechanisms that we use to make political decisions today.
Marketocracy could replace democracy.
Markets, of course, are not perfect. Instead of predicting future revenues, the
market prices of many dot.com stocks seemed to reflect belief in the
sucker-born-every-minute principle. Occasionally, markets can even be cornered
and turned to the advantage of particularly powerful investors. (One concern
about PAM was that terrorists could use it to make money off their acts,
although there are far better - and more anonymous - ways to do that, such as
shorting airline stocks before a hijacking or shorting beef prices before
introducing mad cow disease.) Yet, one of the most intriguing aspects of the
research being done in economics laboratories at HP, IBM, and universities
around the world is the potential to discover rules and procedures that make
markets less susceptible to irrational exuberance and cornering. Careful design
of prediction markets could enhance their forecasting ability.
Business people know that markets are important institutions of a well-run
society. But until recently, few thought of them as important tools for running
their businesses. The uncanny ability of markets to usefully aggregate and
quantify large amounts of dispersed information, however, indicates that such
markets should be taken more seriously as tools to predict everything from
quarterly sales to consumer preferences.
And if information markets can help business, why not the CIA, FBI and other
security risk analysts as well? Despite being attacked by sound-biting
senators, the evidence strongly suggests that this innovative approach to
intelligence gathering and processing could have improved the capability of the
intelligence services before 9/11. Although abandoned by government, we may yet
see private firms using the information processing power of prediction markets
to help us decide contentious political questions and forecast geopolitical
events, including those impacting terrorism.
Alexander
Tabarrok is research director for
The Independent Institute, associate professor of economics at George
Mason University, and editor of Entrepreneurial Economics: Bright Ideas from the
Dismal Science.
|