Content uploaded by George R. Neumann
Author content
All content in this area was uploaded by George R. Neumann
Content may be subject to copyright.
www.sciencemag.org SCIENCE VOL 320 16 MAY 2008
877
POLICYFORUM
P
rediction markets are forums for trad-
ing contracts that yield payments based
on the outcome of uncertain events.
There is mounting evidence that such markets
can help to produce forecasts of event out-
comes with a lower prediction error than con-
ventional forecasting methods. For example,
prediction market prices can be used to
increase the accuracy of poll-based forecasts
of election outcomes (1) (see the figure), offi-
cial corporate experts’ forecasts of printer
sales, and statistical weather forecasts used
by the National Weather Service.
Several researchers emphasize the poten-
tial of prediction markets to improve deci-
sions (2–5). The range of applications is vir-
tually limitless—from helping businesses
make better investment decisions to helping
governments make better fiscal and mone-
tary policy decisions.
Prediction markets have been used by
decision-makers in the U.S. Department of
Defense (6), the health care industry (7), and
multibillion-dollar corporations such as
Eli Lilly, General Electric, Google, France
Telecom, Hewlett-Packard, IBM, Intel, Micro-
soft, Siemens, and Yahoo (8). The prices in
these markets reflect employees’ expecta-
tions about the likelihood of a homeland
security threat, the nationwide extent of a flu
outbreak, the success of a new drug treat-
ment, the sales revenue from an existing
product, the timing of a new product launch,
and the quality of a recently introduced soft-
ware program.
These markets could assist private firms
and public institutions in managing economic
risks, such as declines in consumer demand,
and social risks, such as flu outbreaks and
environmental disasters, more efficiently.
Unfortunately, however, current federal
and state laws limiting gambling create sig-
nificant barriers to the establishment of
vibrant, liquid prediction markets in the
United States. We believe that regulators
should lower these barriers by creating a legal
safe harbor for specified types of small-
stakes markets, stimulating innovation in
both their design and their use (9).
How and Why Prediction Markets Work
An example will help to clarify the prediction
market concept. Consider a contract that pays
$1 if Candidate X wins the presidential elec-
tion in 2008. If the market price of an X con-
tract is currently 53 cents, an interpretation is
that the market “believes” X has a 53%
chance of winning. Prediction markets reflect
a fundamental principle underlying the value
of market-based pricing: Because informa-
tion is often widely dispersed among eco-
nomic actors, it is highly desirable to find a
mechanism to collect and aggregate that
information. Free markets usually manage
this process well because almost anyone can
participate, and the potential for profit (and
loss) creates strong incentives to search for
better information. To be sure, a lively debate
has arisen about whether prediction market
prices are subject to various biases, which
might diminish their accuracy as an aggrega-
tion mechanism (10–14). However, predic-
tion markets have been used with success in a
variety of contexts.
Legal Impediments
The use of prediction markets has been
greatly deterred by state and federal laws
restricting Internet gambling because at
least some of these laws are plausibly under-
stood to cast serious doubts on prediction
The ability of groups of people to make
predictions is a potent research tool that should
be freed of unnecessary government restrictions.
Kenneth J. Arrow,
1
Robert Forsythe,
2
Michael Gorham,
3
Robert Hahn,
4
* Robin Hanson,
5
John O. Ledyard,
6
Saul Levmore,
7
Robert Litan,
8
Paul Milgrom,
1
Forrest D. Nelson,
9
George R. Neumann,
9
Marco Ottaviani,
10
Thomas C. Schelling,
11
Robert J. Shiller,
12
Vernon L. Smith,
13
Erik Snowberg,
14
Cass R. Sunstein,
7
Paul C. Tetlock,
15
Philip E. Tetlock,
16
Hal R. Varian,
17
Justin Wolfers,
18
Eric Zitzewitz
19
ECONOMICS
The Promise of Prediction Markets
1
Department of Economics, Stanford University, Stanford,
CA 94305, USA.
2
College of Business, University of South
Florida, Tampa, FL 33620, USA.
3
Stuart School of Business,
Illnois Institute of Technology, Chicago, IL 60661, USA.
4
Reg-Markets Center at American Enterprise Institute,
Washington, DC 20036, USA.
5
Department of Economics,
George Mason University, Fairfax, VA 22030, USA.
6
Department of Economics, California Institute of
Technology, Pasadena, CA 91125, USA.
7
University of
Chicago Law School, Chicago, IL 60637, USA.
8
Kauffman
Foundation, Kansas City, MO 64110, USA.
9
Department of
Economics, University of Iowa, Iowa City, IA 52242, USA.
10
Kellogg Graduate School of Management, Northwestern
University, Evanston, IL 60208, USA.
11
School of Public
Policy, University of Maryland, College Park, MD 20742,
USA.
12
Department of Economics, Yale University, New
Haven, CT 06520, USA.
13
Chapman University School of
Law, Orange, CA 92866, USA.
14
Graduate School of
Business, Stanford University, Stanford, CA 94305, USA.
15
Yale School of Management, New Haven, CT 06520,
USA.
16
Haas School of Business, University of California,
Berkeley, CA 94720, USA.
17
School of Information,
University of California, Berkeley, CA 94720, USA.
18
The
Wharton School, University of Pennsylvania, Philadelphia,
PA 19104, USA.
19
Department of Economics, Dartmouth
College, Hanover, NH 03755, USA.
*Author for correspondence. E-mail: rhahn@aei.org.
Information Revelation Through Time
0
2
4
6
0306090120150
Days until election
Average absolute forecast error
(Vote share %)
Information revelation through time. Data are from the Iowa Electronic Markets for markets predicting the
two-party vote shares from the 1988, 1992, 1996, and 2000 presidential elections (19). The vertical axis
plots the average absolute difference between the market prediction and the actual vote share. In the week
immediately before the election, the market erred by an average of 1.5 percentage points compared with an
average error of 2.1 percentage points for the final Gallup poll. The longer-run forecasting performance of
the market is also impressive, with an average error of only 5 percentage points 150 days before the election,
a time when polls have much larger errors when interpreted as predictions. Calculations are based on data
available at www.biz.uiowa.edu/iem.
Published by AAAS
on May 15, 2008 www.sciencemag.orgDownloaded from
16 MAY 2008 VOL 320 SCIENCE www.sciencemag.org
878
POLICYFORUM
markets. Currently, eight states bar Internet
gambling outright. In 2006, President Bush
signed the Unlawful Internet Gambling
Enforcement Act, designed to crack down
on such gambling.
The legal questions here are complex, but
to create a prediction market in the United
States that is unambiguously legal, one must
run a regulatory gauntlet (15). In principle,
these difficulties could be avoided by creat-
ing prediction markets outside the United
States, but this approach could suppress inno-
vation and reduce opportunities to aggregate
information and improve decisions. It would
be better for U.S. authorities to clarify the cir-
cumstances under which prediction markets
are plainly legal.
Breaking the Legal Impasse
We suggest that two steps should be taken to
facilitate the use of prediction markets while
still meeting the legitimate concerns of law-
makers and regulators.
(i) The Commodity Futures Trading
Commission (CFTC), the federal regulatory
agency that oversees futures market activity,
should establish safe-harbor rules for selected
small-stakes markets. One limited safe harbor
is the no-action letter, in which the CFTC mar-
ket oversight staff confirms in writing that it
will not recommend enforcement action if the
recipient acts in specified ways. The only pre-
diction market to receive a no-action letter (in
1992) is the Iowa Electronic Markets (16),
which is run by professors at the University of
Iowa and which initially focused on presiden-
tial elections. Although such no-action letters
reduce the chances of legal action under other
state and federal laws, they may not be ade-
quate. We would therefore urge the CFTC to
explore other approaches to ensuring safe har-
bors, for example, formal rules or guidance
approved by the commission.
We suggest that three types of entities be
eligible for safe harbor treatment. The first
would be not-for-profit research institutions,
including universities, colleges, and think
tanks wishing to operate exchanges similar to
the Iowa Electronic Markets. The second
would be government agencies seeking to do
research similar to that of nongovernmental
research institutions. The third group would
consist of private businesses and not-for-
profits that are not primarily engaged in
research, which would only be allowed to
operate internal prediction markets with their
employees or contractors.
In all cases, markets would be limited to
small-stakes contracts. Although the defini-
tion of small stakes is somewhat arbitrary, we
use the term to mean an exchange in which
the total amount of capital deposited by any
one participant may not exceed some modest
sum, perhaps something like $2000 per year.
The exchanges themselves would be not-
for-profit but would be allowed to charge
modest fees to recoup administrative and reg-
ulatory costs. Brokers and paid advisers
would be barred, reducing the risks that con-
tracts would be sold to inappropriate or vul-
nerable customers or that customers would be
charged fees above the amounts needed to
maintain the markets. Exchanges would be
self-regulated, leaving them with the respon-
sibility to make reasonable efforts to keep
markets free from fraud and manipulation.
For its part, the CFTC should allow con-
tracts that price any economically meaningful
event. This definition could allow for
contracts on political events, environmental
risks, or economic indicators, such as those
offered by the Iowa Electronic Markets, but
would presumably not include contracts on
the outcomes of sports events.
The contracts qualifying under this safe
harbor would also create opportunities
for more efficient risk allocation (17).
Although the small-stakes nature of these
markets would necessarily limit their use-
fulness for hedging risk, they could serve as
proofs of concept for larger-scale markets
that could be developed under alternative
regulatory arrangements.
The CFTC should allow researchers to
experiment with several aspects of prediction
markets—fee structures, incentives against
manipulation, liquidity requirements and the
like—with the goal of improving their design.
Prediction markets are in an early stage, and if
their promise is to be realized, researchers
should be given flexibility to learn what kinds
of design are most likely to produce accurate
predictions. Of course, exchanges would
need to inform their customers so that they
are aware of the risks and benefits of partici-
pating in these markets.
(ii) Congress should support the CFTC’s
efforts to develop prediction markets (18). To
the extent that the CFTC incurs costs in pro-
moting innovation, Congress should provide
the necessary funding. More fundamentally,
Congress should explore alternative ways of
securing a legal framework for prediction
markets if the CFTC’s existing authority
proves inadequate. In particular, Congress
should specify that a no-action letter, or simi-
lar mechanism, preempts overlapping state
and federal antigambling laws. Because
Congress did not intend the CFTC to regulate
gambling, it is important to design new regu-
lations so that socially valuable prediction
markets easily qualify for the safe harbor but
gambling markets do not.
Conclusion
We have suggested some modest reforms at
the federal level that we hope will facilitate
the development of prediction markets.
These markets have great potential for im-
proving social welfare in many domains.
American leadership in this area is likely to
encourage parallel efforts in other countries,
speeding the development of this tool. The
first step in helping prediction markets
deliver on their promise is to clear away regu-
latory barriers that were never intended to
inhibit socially productive innovation.
References and Notes
1. J. Berg, F. Nelson, T. Rietz, Int. J. Forecast., in press.
2. R. Hanson, IEEE Intell. Syst. 14, 16 (1999).
3. R. Hahn, P. Tetlock, Harvard J. Law Pub. Pol. 28, 213
(2005).
4. C. Sunstein, Infotopia: How Many Minds Produce
Knowledge (Oxford Univ. Press, New York, 2006).
5. E. Snowberg, J. Wolfers, E. Zitzewitz, Q. J. Econ., 122,
807 (2007).
6. R. Hanson, T. Ishikida, J. Ledyard, C. Polk, Proceedings of
the ACM International Conference on Electronic
Commerce, Pittsburgh, PA, 30 September to 3 October
2003 [Association for Computing Machinery (ACM),
New York, 2003], p. 272.
7. P. M. Polgreen, F. Nelson, G. Neumann, Clin. Infect. Dis.
44, 272 (2007).
8. B. Cowgill, J. Wolfers, E. Zitzewitz, “Using prediction
markets to track information flows: Evidence from
Google,” Dartmouth College (2008); www.bocowgill.com/
GooglePredictionMarketPaper.pdf.
9. K. Arrow et al., “Statement on prediction markets,”
AEI-Brookings Joint Center Related Publication No. 07-11
(May 2007); available at Social Science Research
Network (SSRN), http://ssrn.com/abstract=984584.
10. J. Wolfers, E. Zitzewitz, “Interpreting prediction market
prices as probabilities,” Stanford Graduate School of
Business (2005).
11. C. Manski, Econ. Lett. 91, 425 (2006).
12. C. Sunstein, Infotopia: How Many Minds Produce
Knowledge (Oxford Univ. Press, New York, 2006).
13. M. Ottaviani, P. N. Sørensen, J. Eur. Econ. Assoc. 5, 554
(2007).
14. P. Tetlock, “Liquidity and prediction market efficiency”
(March 2008); available at SSRN: http://ssrn.com/
abstract=929916.
15. R. Hahn, P. Tetlock, J. Regul. Econ. 29, 265 (2006).
16. No-action letter from Andrea M. Corcoran, Director,
Commodity Futures Trading Commission (CFTC) Division
of Trading and Markets, to George R. Neumann, Professor
of Economics, University of Iowa (5 February 1992);
www.cftc.gov/files/foia/repfoia/foirf0503b002.pdf.
17. R. Shiller, Macro Markets: Creating Institutions for
Managing Society’s Largest Economic Risks (Oxford Univ.
Press, Oxford, 1993).
18. On May 1, 2008, the CFTC requested public comment on
the appropriate regulatory treatment of prediction mar-
kets. “Commodity Futures Trading Commission, Concept
release on the appropriate regulatory treatment of event
contracts” (May 2008); www.cftc.gov/stellent/groups/
public/@lrfederalregister/documents/file/e8-9981a.pdf.
19. J. Wolfers, E. Zitzewitz, J. Econ. Perspect. 18, 107 (2004).
20. The views expressed here represent those of the authors
and do not necessarily represent the views of the institu-
tions with which they are affiliated. Support provided by
the Reg-Markets Center at AEI (www.reg-markets.org).
10.1126/science.1157679
Published by AAAS
on May 15, 2008 www.sciencemag.orgDownloaded from