JOURNAL OF GLOBAL TRADE, ETHICS AND LAW
Volume 3 Issue 2, 2025
PREDICTION MARKETS
Emilio Barone1* and Federico Carli2
1 Department of Economics and Finance, LUISS-Guido Carli University, Rome, Italy, ebarone@luiss.it.
2 Department of Economics, Business, Law and Political Sciences, Guglielmo Marconi University, Rome, Italy,
federico.carli@rcs.it.
*Corresponding Author
Abstract. This paper provides an English version of a study originally circulated
as a working paper in November 2005 on prediction markets and subsequently
published in Italian in Mondo Bancario (January–February 2006). The paper
reviews the structure, functioning, and empirical performance of prediction
markets, with particular emphasis on the Iowa Electronic Markets and other
platforms active in the early 2000s. The analysis covers electoral markets, cur-
rent events, entertainment markets, and macroeconomic derivatives, and
discusses issues related to information aggregation, forecasting accuracy, and
regulation.
Keywords: Prediction Markets, Information Markets, Event Markets.
JEL: D8, G14.
© 2025 Durham & Thunmann, London
Journal of Global Trade, Ethics and Law
ISSN 2977-0025 (Online). Published under Creative Common (CC) BY 4.0 license.
https://doi.org/10.70150/xrt21s70
Durham &
Thunmann
Emilio Barone & Federico Carli, 2025
56
“Prediction is very difficult, especially if it’s about the future”
- Nils Bohr
1. Introduction
“Prediction is very difficult, especially if it’s about the future.” These words are
commonly attributed to Nils Bohr (1885-1962), the Danish physicist who, for his
contributions to quantum theory, was awarded the Nobel Prize in Physics in 1922, one
year after Einstein.
In the difficult task of forecasting the future - or, more precisely, of assigning
probabilities to future events - prediction markets, also known as information markets or
event markets, have increasingly attracted attention. Their archetype is the Iowa
Electronic Market (IEM), an experimental market cited by Vernon Smith in his 2002
Nobel Prize Lecture:
1
“What evidence do we have that the laboratory efficiency properties of continuous
double auction trading apply also in the field? One of the best sources of evidence, I
believe, is found in the Iowa Electronic Market (IEM) used widely around the world.
(Forsythe, et al., 1991, 1999) These markets are used to study the efficacy of futures
markets in aggregating widely dispersed information on the outcomes of political
elections, or any well defined extra-laboratory event, such as a change in the discount
rate by the FED. The ‘laboratory’ is the internet. The ‘subjects’ are all who log on and
buy an initial portfolio of claims on the final event outcomes; they consist of whom ever
logs in, and are not any kind of representative or ‘scientific’ sample as in the polls with
which they are paired. The institution is the open book double auction.
In the IEM, traders make a market in shares representing pari-mutuel claims on the
popular vote (or winner-take-all) outcome of an election, referendum, etc. For example
the first IEM was on the 1988 Presidential Election. Each person wanting to trade shares
deposits a minimum sum, $35, with the IEM and receives a trading account containing
$10 cash for buying additional shares, and ten elemental portfolios at $2.50 each,
consisting of one share of each of the candidates - Bush, Dukakis, Jackson, and ‘rest-of-
field.’ Trading occurs continuously in an open-book bid-ask market for several months,
and everyone knows that the market will be called (trading suspended) in November on
election day, when the dividend paid on each share is equal to the candidate’s fraction of
the popular vote times $2.50. Hence if the final two candidates and all others receive
popular vote shares (53.2%, 45.4%, 1.4%), these proportions (times $2.50) represent the
payoff to a trader for each share held. Consequently, at any time t, normalizing on $1,
the price of a share (÷$2.50), reflects the market expectation of that candidate’s share of
the total vote. A price, $0.43 means the market predicts that the candidate will poll 43%
1
Smith (2002).
Emilio Barone & Federico Carli, 2025
57
of the vote. Other forms of contract that can be traded in some IEMs include winner-take-
all, or number of seats in the House, and so on.
The IEM data set includes 49 markets, 41 worldwide elections and 13 countries.
Several results stand out: the closing market prices, produced by a non-representative
sample of traders, show lower average absolute forecasting error (1.5%) than the
representative exit poll samples (1.9%); in the subset of 16 national elections, the market
outperforms the polls in 9 of 15 cases; in the course of several months preceding the
election outcome, the market predictions are consistently much less volatile than the
polls; generally, larger and more active markets predict better than smaller, thinner
markets; surveys of the market traders show that their share holdings are biased in favor
of the candidates they themselves prefer.
In view of this last result why do markets outperform the polls? Forsythe, et al. (1991)
argue that it’s their marginal trader hypothesis. Those who are active in price ‘setting,’
that is, in entering limit bids or asks, are found to be less subject to this bias, than those
traders accepting (selling and buying ‘at market’) the limit bids and asks. Polls record
unmotivated, representative, average opinion. Markets record motivated marginal
opinion that cannot be described as ‘representative.’
For decades, researchers in artificial intelligence have attempted to construct intelligent
systems by writing software that replicates human cognitive processes. More recently -
as the IEM clearly demonstrates - researchers have sought to solve complex problems by
building networks of autonomous agents that interact with one another. The key idea
behind intelligent system design increasingly appears to be that of leveraging individual
agents - connected through information systems - and exploiting their cognitive
capabilities in innovative ways.
Issues of public interest can therefore be addressed through mechanisms analogous to
those used in markets for financial securities. The proposed solution is to create markets
in which both correct and incorrect answers carry economic consequences. The
underlying hypothesis is that such markets will amplify the influence of those who are in
the best position to know the correct answer. Moreover, the very existence of these
markets may induce participants to seek out information that improves the quality of their
decisions.
The reason this approach should work is that less informed participants tend to lose
money and are eventually driven out of the market, while better informed participants
tend to profit and increasingly guide the market, thereby determining prices.
2. Review of Markets and Contract Design
The most widely known prediction market is the Iowa Electronic Market (IEM),
established in 1988. The first contract traded on the IEM paid 2.5 cents for each
percentage point of the popular vote received by each candidate in U.S. presidential
elections (Bush, Dukakis, and others).
Emilio Barone & Federico Carli, 2025
58
Universities in other countries have also begun to operate their own event markets.
2
Examples include the Austrian Electronic Market at the Vienna University of Technology
and the Election Stock Market at the University of British Columbia.
There are also prediction markets operated by firms active in the sports betting
industry.
3
Notable examples include Trade-Sports, Betfair, and the World Sports
Exchange.
According to information published on the Trade-Sports website (August 2005), the
platform - managed by an Irish company whose financial statements were audited by
Deloitte - reported cumulative trading volume of approximately $1 billion, around 70
million contracts traded, and more than 30,000 registered users over roughly three years
of operation.
In some cases (e.g., Newsfutures or the Hollywood Stock Exchange), trading takes
place using virtual currency. These exchanges define the contracts, and participants either
submit their own offers or accept those posted by others. On the Newsfutures platform -
founded in 2000 and active until 2004 - participants could speculate on a wide range of
current events, including politics, sports, cinema, economics, and technology. As an
incentive, prizes were awarded to the top performing participants. Participants in the
Hollywood Stock Exchange use virtual currency to speculate on questions related to the
film industry, such as a movie’s box-office revenue, the number of spectators during its
opening weekend, or the allocation of Academy Awards. Insider trading is entirely legal
in this market and is, in fact, explicitly encouraged. Film studios often rely on the
forecasts generated by these markets.
The Foresight Exchange also operates using virtual currency. The range of contracts
offered is broad, spanning traditional financial contracts as well as contracts related to
disasters, news events, politics, and scientific developments. For example, at the end of
August 2005, the contract “NASDAQ drops below 1000 by 2008” traded at 24, “Whites
US Minority by 2060” at 75, “Cold Fusion by 2015” at 19, “Human Organ Farms by
2015” at 26, and “Moonbase by 2025” at 35.
More recently (in 2002), Goldman Sachs and Deutsche Bank launched a new market -
Economic Derivatives - in which the events to be predicted concern the release of
macroeconomic data, such as employment, retail sales, gross domestic product, consumer
confidence indices, and inflation.
In June 2005, Goldman Sachs reached an agreement with the Chicago Mercantile
Exchange to create the CME Auction Markets, where “a series of innovative event-driven
economic derivatives” would be traded. The auctions were to be conducted using
proprietary software developed by Longitude. The agreement specified that trading would
initially take place via the web starting in September 2005, before migrating to the Globex
platform in January 2006.
This CME initiative illustrates how the phenomenon of prediction markets - initially
confined to over-the-counter markets - began, at that time, to extend into organized
exchange-traded markets.
4
2
See Brüggelambert (1988), Ortner (1996), Beckmann, & Werding (1996), Jacobsen et al. (2000).
3
See Wolfers & Leigh (2001).
4
For a survey, see Goldfinger (2004).
Emilio Barone & Federico Carli, 2025
59
A list of existing prediction markets is provided in Error! Reference source not
found., while a sample of contracts traded on the Foresight Exchange is reported in
Error! Reference source not found..
3. Iowa Electronic Markets
The Iowa Electronic Markets (IEM) are small-scale markets - managed by the College
of Business at the University of Iowa - in which contracts are traded using real money.
5
The most prominent of these markets are the Iowa Political Markets.
6
Contracts traded in
these markets are designed so that prices can be used to generate forecasts of electoral
outcomes.
7
The IEM operate continuously, 24 hours a day, and employ a continuous double-
auction trading mechanism. Traders invest their own funds, execute trades autonomously,
and gather information independently. As such, the IEM occupy a niche between highly
stylized and tightly controlled “laboratory” markets and fully fledged “real” markets.
Owing to their design, the IEM provide researchers with data that are otherwise
unavailable.
In the IEM, data collection and aggregation rely on a mechanism that differs
substantially from that used in opinion polls. Polls are based on representative samples of
potential voters, the reliability of survey responses, and statistical inference. In contrast,
IEM traders face explicit monetary incentives and do not constitute a representative
sample of the electorate. The vast majority of participants are young (with an average age
close to 30), predominantly male, and well educated. Traders are not required to satisfy
the eligibility criteria for voting in elections.
3.1. Market Mechanism
Each market is linked to a specific future event - such as a presidential election - and
allows the trading of contracts whose final value (payoff) is determined by the outcome
of that event. Contracts enter circulation when traders purchase a bundle of contracts from
the exchange, known as unit portfolios. Conversely, contracts are withdrawn from
circulation when unit portfolios are sold back to the exchange.
5
The maximum investment allowed per trader is $500. Average investment is less than $50. The number
of traders actively participating in a given market typically ranges from a dozen to more than 500. In the
market on the 1992 U.S. presidential elections, 78,007 contracts were traded, for a total notional value of
$21,445.
6
The Iowa Electronic Markets are composed of several segments: electoral markets based on vote shares
(vote-share markets), markets based on the number of seats (seat-share markets), and markets based on the
election winner (winner-takes-all markets), as well as markets based on other political outcomes, economic
indicators, corporate earnings, and rates of return of selected firms.
7
The information is drawn from Berg, Forsythe, Nelson, and Rietz (1998).
Emilio Barone & Federico Carli, 2025
60
Unit portfolios consist of one unit of each contract available in the market. They are
purchased from and sold to the exchange at a fixed price equal to the final value of the
entire portfolio. The rate of return associated with a unit portfolio is equal to the risk-free
interest rate, which is zero in the case of the IEM.
The use of unit portfolios ensures that the market operates as a zero-sum game and that
the supply of contracts is determined endogenously by the net number of unit portfolios
purchased by traders. Unit portfolios serve solely to introduce contracts into circulation.
Transactions among traders take place at prices determined by participants for each
individual contract.
Traders may submit market orders, which require immediate execution at the prevailing
market price, or limit orders, which specify quantities to buy or sell at given bid or ask
prices within a specified time horizon. Limit orders are ranked according to price and the
time at which they are submitted. They may be canceled at any time before being executed
(hit) or before expiration.
Figure 1: IEM: Trading Screen.
Source: Berg et al. (2000).
Note: The trading screen is divided into three sections. The upper section reports - for each contract (e.g., DemVS,
ReformVS, and RepVS) - the best bid and ask quotes, the price of the most recent transaction (last trade price), the trader’s
current holdings (quantity held), and the trader’s outstanding buy (YourBids) and sell (YourAsks) orders. The middle section
contains a menu that allows traders to enter quantities (Qty) for market orders or limit orders, specifying, in the latter case,
both the price (Price) and the expiration date (Expire). The lower section provides a menu enabling traders to access other
activities.
Emilio Barone & Federico Carli, 2025
61
The information set made available by the exchange to traders consists of the most
recent transaction price (last trade price) and the best bid and ask quotes. The screen that
appears to traders is shown in Error! Reference source not found..
Traders do not observe the quantities available at the best bid and ask prices, nor do
they have access to the remainder of the order book, which comprises all outstanding
limit orders. With respect to historical data, traders may access daily information on
trading volume and contract value, as well as daily minimum, maximum, average, and
closing prices.
8
3.2. Types of Markets
Winner-Takes-All Markets
Contracts traded in winner-takes-all markets pay $1 if the associated event occurs. For
example, if an election is won by a given candidate, the contract linked to that candidate
yields a payoff of $1.
Vote-Share Markets
In vote-share markets, the final value of each contract is determined by the share of
votes received by the corresponding candidate. The final value of a contract equals the
product of $1 and the candidate’s vote share. To ensure that the sum of vote shares equals
unity, one of two methods is employed:
1. A contract associated with the residual vote share (rest-of-the-field
contract) is traded; or
2. Vote shares are computed as fractions of a restricted total (for example,
the Democratic vote share is calculated as the number of votes received by
the Democratic candidate divided by the total votes received by
Democratic and Republican candidates).
The method adopted in each case is described in the prospectus prepared for the specific
market.
8
Although markets operate continuously, information is aggregated and reported over daily 24-hour
periods.
Emilio Barone & Federico Carli, 2025
62
Figure 2: IEM Election Futures.
Source: Wolfers, & Zitzewitz (2004).
Seat-Share Markets
In seat-share markets, the final value of contracts is determined by the number of
legislative seats obtained by each party in an election. The payoff of each contract equals
the product of $1 and the party’s share of seats.
3.3. Efficiency
A recurring question in the analysis of these new markets concerns their efficiency. Are
prediction markets truly efficient? Do they effectively aggregate information? What is
the informational content of market prices? Are the prices generated in these markets
genuinely useful for forecasting purposes? According to several empirical studies, the
answer appears to be affirmative.
Berg et al. (2000) argue that prices observed in the Iowa Electronic Market yield highly
accurate forecasts, outperforming those derived from opinion polls.
9
Error! Reference source not found. reports data for four U.S. presidential elections
(1988-2000). The horizontal axis shows the number of days remaining until election day,
9
See also Granberg & Brent (1983), Oliven & Rietz (1995), Bondarenko & Bossaerts (1999), Slemrod
& Greimel (1999).
Emilio Barone & Federico Carli, 2025
63
while the vertical axis reports the mean absolute forecasting error computed on the basis
of IEM prices. As the figure shows, the accuracy of market-based forecasts improves as
election day approaches, in parallel with the arrival of new information. Error!
Reference source not found. also indicates that, during the final week before the
election, markets predicted the vote shares of Democratic and Republican candidates with
an average absolute error of approximately 1.5 percentage points. By comparison, the
forecasting error of the final Gallup polls for the same four elections was 2.1 percentage
points (Error! Reference source not found.).
Table 1: Final Gallup Polls: Forecasting Errors (1988-2000).
3.3.1. Bush vs. Kerry
The outcome of the most recent U.S. presidential election was accurately predicted by
the IEM. In the 2004 U.S. Presidential Vote Share Market (Pres04_VS), two contracts
were traded: BU|KERR, which paid $1 if Bush won, and KERR, which paid $1 if Kerry
won. The election took place on November 2, 2004. On the day preceding the election,
the quoted prices were as follows:
Year
Final
Poll
(%)
Election
Outcome
(%)
Forecasting
Error
(%)
Average
Forecasting Error
(%)
2000
48.0
47.9
0.1
46.0
48.4
-2.4
4.0
2.7
1.3
1.3
1996
52.0
50.1
1.9
41.0
41.4
-0.4
7.0
8.5
-1.5
1.3
1992
49.0
43.3
5.7
37.0
37.7
-0.7
14.0
19.0
-5.0
3.8
1988
56.0
53.9
2.1
44.0
46.1
-2.1
2.1
Total
2.1
Symbol
Units
Volume ($)
Low
High
Average
BU|KERR
215
109,249
0.480
0.519
0.508
KERR
337
337,490
0.475
0.510
0.492
Emilio Barone & Federico Carli, 2025
64
It should be noted that the sum of the average prices of the two contracts (column
“Average”) equals 1 (= 0.508 + 0.492). This is not the case for closing prices (column
“Last”), which may reflect slight timing mismatches. A graphical representation of the
“Last” prices is provided in Figure 1.
Figure 1: IEM Election Futures: Bush vs. Kerry.
Source: Wolfers, & Zitzewitz (2004).
In the second market (Pres04_WTA), four contracts were traded: DEM04_G52 and
DEM04_L52 (Kerry receiving more or less than 52% of the vote), and REP04_G52 and
REP04_L52 (Bush receiving more or less than 52% of the vote). The quoted prices were
as follows:
By summing the “Average” prices of the first two contracts, one obtains the market-
implied probability of a Kerry victory: 49.5% (= 14.7% + 34.8%). Summing the
“Average” prices of the last two contracts yields the probability of a Bush victory: 50.3%
(= 15.8% + 34.3%). The resulting values do not sum to one and are marginally different
from those obtained previously. This inconsistency would likely not have arisen had mid-
market prices - that is, the averages of bid and ask quotes - been used; however, historical
series of such prices are not available on the IEM website.
Symbol
Units
Volume ($)
Low
High
Average
Last
DEM04_G52
9,227
1,358,635
0.130
0.164
0.147
0.155
DEM04_L52
10,365
3,606,583
0.300
0.474
0.348
0.330
REP04_G52
13,069
2,067,415
0.070
0.239
0.158
0.150
REP04_L52
11,609
3,978,678
0.309
0.393
0.343
0.362
Emilio Barone & Federico Carli, 2025
65
3.3.2. Vote-Share and Seat-Share Markets
In
Figure 2, electoral forecasts based on 237 contracts (49 markets across 13 countries)
are compared with actual election outcomes. In the figure, the horizontal axis measures
actual outcomes, while the vertical axis reports predicted outcomes. If forecasts were
perfectly accurate, all points would lie on the 45-degree line. Forecasts that overestimate
outcomes appear above the line, while underestimates appear below it.
Figure 2: IEM: Comparison Between Forecasts and Actual Outcomes.
Source: Berg et al. (2000).
Note: VOTE-SHARE MARKETS: Austria (Federal Parliamentary Elections ’95; Styria ’95; Vienna ’95; European
Parliamentary Elections ’96), Canada (Parliamentary Elections ’93, ’96), Korea (Presidential Election ’92), Denmark
(Parliamentary Election ’91), Finland (Presidential Elections × 2 markets), France (Presidential Election ’95), Germany
(Parliamentary Elections ’90 × 3 markets - Bonn, Frankfurt, and Leipzig; Federal Elections ’91, ’94, and ’98; State Election
’98), Norway (Parliamentary Election ’95), Netherlands (Parliamentary Elections ’91, ’94 - Second Chamber; European
Parliamentary Election ’94; Municipal Council Election ’94), Sweden (European Parliamentary Election ’94), Turkey
(Parliamentary Election ’91), United States - Gubernatorial Elections (NY ’94, TX ’94), United States - House of
Representatives (UT ’94), United States - Presidential Elections (’88, ’92, ’96 × 2 markets, ’00), United States - Presidential
Primaries (IL ’92, MI ’92), United States - Senate Elections (IL ’90, IA ’90, AZ ’94, NJ ’94, PA ’94, TX ’94, VA ’94).
SEAT-SHARE MARKETS AND OTHERS: Australia (Parliamentary Election ’93), Canada (Parliamentary Elections ’93,
’96), Netherlands (Parliamentary Election ’94 - Second Chamber; European Parliamentary Election ’94), United States -
House of Representatives (’94), United States - Senate (’94).
Emilio Barone & Federico Carli, 2025
66
Figure 2 shows no significant forecasting errors. On average - particularly in markets
related to major U.S. elections - forecast accuracy is very high.
The variability in forecasting accuracy across U.S. elections can be explained by three
main factors:
1. markets associated with presidential elections perform better than markets
associated with lower-profile elections at the congressional, state, or local
level;
2. markets with higher trading volume close to election day perform better
than those with lower trading volume;
3. markets with a smaller number of contracts (i.e., fewer candidates or
parties) perform better than those with a larger number of contracts.
3.3.3. Opinion Polls
Figure 3 compares the relative performance of prediction markets and opinion polls.
Because market prices change continuously, a choice must be made regarding which price
should be used for forecasting purposes.
Emilio Barone & Federico Carli, 2025
67
Figure 3: Presidential Elections (Vote-Share Markets): Forecasting Errors (Polls and Markets).
Source: Berg et al. (2000).
Note: The forecasts derived from the major opinion polls conducted during the week preceding the presidential elections are
compared with those generated by the vote-share markets over the same week and at midnight on the day preceding the
election.
Two measures are reported in Figure 3:
(i) the price observed at midnight on the day preceding the election; and
(ii) the average price (weighted by trading volume) observed during the week
preceding the election.
The first measure incorporates all information available to traders as of midnight on
the day before the election, but it is subject to high variability due to the thinning of the
order book on the final trading day.
The second measure reflects trades occurring contemporaneously with major opinion
polls.
Market performance - measured using both pricing methods - was superior to that of
opinion polls in 9 out of 15 cases. The mean absolute forecasting error of opinion polls
was 1.93 percent, while that of markets was 1.49 percent using the first measure and 1.58
percent using the second.
In some cases - specifically, the U.S. presidential elections of 1988 and 1992 - the
performance of prediction markets was clearly superior to that of opinion polls. In most
other cases, markets performed approximately on par with polls, sometimes slightly
worse but more often slightly better.
3.4. Educational Uses
The IEM have been used to study a wide range of issues. They help bridge the gap
between traditional laboratory markets and real-world markets and provide information
not typically available in financial markets, including individual traders’ orders, detailed
transaction data, complete order-book information, portfolio compositions of individual
traders, and demographic characteristics of participants.
The IEM also allow researchers to interview traders at any point in time, record their
responses, and link those responses to other available information, thereby
complementing existing research methodologies.
4. Trade-Sports
4.1. Elections
Emilio Barone & Federico Carli, 2025
68
Prediction markets allow researchers to assess the relative popularity of potential
candidates in elections. In this respect, contracts on prospective Democratic and
Republican candidates for the 2008 U.S. presidential election traded on Trade-Sports are
particularly informative. For example, at the end of August 2005, the contract on Hillary
Clinton traded in the range of 41.5-41.9 (Figure 4), significantly outperforming other
potential candidates.
Figure 4: Contracts on Democratic Candidates in the 2008 U.S. Presidential Election.
Source: Trade-Sports, August 23, 2005 (2:24 pm).
4.1.1. Arnold Schwarzenegger
Figure 5 reports bid and ask quotes recorded at four-hour intervals on two markets –
Trade-Sports and the World Sports Exchange – both of which traded a contract on the
election of Arnold Schwarzenegger as Governor of California. Opportunities for risk-free
arbitrage, achieved by purchasing the contract at a lower ask price on one market and
selling it at a higher bid price on the other, were virtually nonexistent.