Prediction Diversity If PD>0 then CE < AIE +

+
An introduction to
Prediction Markets and
the wisdom of the crowds
Prof. Luca Iandoli
University of Naples Federico II
+
Can groups be smarter than their
average members?

Francis Galton study
+
Can groups be smarter than their
average members?

Who wants to be a millionaire: better ask the expert or the
crowd?
According to the
show stats, the
expert guesses it
right 65% of the
time, the crowd
91%, why is that?
+
Can groups be smarter than their
average members?

Which task? Collective Prediction

Conditions for a crowd to be smart:

people have diverse predictive
models;

Independency: crowd members are
not allowed to influence each other or
do so limitedly;

prediction process is decentralized.
+
Diversity and collective
intelligence

“If you want to have good ideas you must have many ideas” –
Linus Pauling
+
Diversity and collective
intelligence

The importance of the signal: Why diversity works and why
it leads to convergence (which ultimately is diversity
suppression): exploration + exploitation
+
Theories of collective intelligence

The Condorcet’s theorem (1785)
In majority voting
probability for a crowd of n to be right -> 1
As n -> ∞ and
p(each individual) to be right > 0.5
The theorem also says that when this
individual probability is less than .5 the
probability of the crowd to be wrong
approaches 1 as n increases.
+
Theories of collective intelligence

The Diversity Prediction theorem (Scott Page, 2007)
Collective Error = Average Individual Error – Prediction
Diversity
If PD>0 then CE < AIE
+

The collaboration dilemma
The collective prediction models assume independency. What
happens when people collaborate? Two very different things
Collective creativity
Group thinking (polarization,
information cascades, hidden
profiles, …)
Diversity suppression
+
Group thinking

Balkanization and polarization:

crowds tend to disarticulate in groups holding opposite beliefs
(balkanize); the members of a faction of like-minded people tend
to develop even more extreme opinions than the one they held
individually prior to joining the group;

Information cascades: information propagates quickly in
social network by word of mouth and imitation

Hidden profiles: individuals may not share all the information
they have when they are in a group, but will focus on the
information they have in common.
+
Markets and collective intelligence

The case of prediction markets
+
Markets and collective intelligence
 PREDICTION
MARKETS (PMs) are electronic
markets that leverage the wisdom of the crowds.
 They
come through virtual trading software
platforms and are used in distributed networks of
decision makers, sometimes even in the form of
public markets on the Internet

E.g. tradesport, holliwood stock exchange,
ideafutures, inkling, University of Iowa Prediction
markets, …
+
Markets and collective intelligence
 How
they work? Pretty much as real markets do.
In a PM, payoffs are tied to the outcomes of future
events.
 Participants trade contracts associated to the
occurrence of a given event (Who will win the UEFA
Champions league?)
 The market exchange of contracts determines their
price: in general, the higher the price of a contract,
the higher the confidence of the market in the future
occurrence of the associated event.
 Participants trade with real or virtual money; in any
case they bet on the outcome they think is more likely

https://www.intrade.com/v4/misc/howItWorks/theBasics.jsp
+
Markets and collective intelligence
 How
they work: the winner-take-all case
Contract has payoff of $0 or $1 based on outcome
(assumption: event has a clear outcome)
 Participants trade win/loose contracts
 The market price is the probability of the event to occur
 Profit = Payoff –cost = (1 – cost of single contract)*n°-ofcontracts-you-own

REAL MADRID Wins
Probability = p
Payoff=$1
REAL MADRID Loses
Probability=1-p
Payoff=$0
+
Prediction Markets

Are PMs accurate? Evidence shows that PMs can predict
better than polls and sometimes better than experts
J. Berg, R. Forsythe, F. Nelson and T. Rietz, Results from a Dozen Years of Election Futures Markets
Research, 2001.
+
Prediction markets
Why
they work?
Efficient
market hypothesis
Incentives
Uncertainty and diversity
+
Prediction markets

Why they work?
 ASSUMPTIONS FOR THE Efficient market
hypothesis
 Many buyers
 Information from various and random
sources
 Prices adapt quickly to new information
Prices reflect all the available information. It
is impossible to beat the market
+
Prediction markets
Why
they work?
Incentives
 Profit opportunity is an
incentive for information
seeking (the p to make correct
in Condorcet’s theorem
increases)
+
Prediction markets
 Why
they work?
 Uncertainty and diversity
 Diversity: different
opinions about what is the
right answer (otherwise everybody bets on the
same contract and nobody profits)
 Not all the opinions are equally weighted
(unlike in polls): more informed buyers bet
more and lead the market
+
References

Scott Page (2007), The difference: how the power of diversity
creates better groups, firms, schools and societies. Princeton
University Press.

James Surowiecki (2004). The Wisdom of Crowds. New York:
Doubleday Press.

Justin Wolfers and Eric Zitzewitz (2004). Prediction Markets.
Journal of Economic Perspectives—Volume 18, Number 2, Pages
107–126.

J. Berg, R. Forsythe, F. Nelson and T. Rietz, Results from a Dozen
Years of Election Futures Markets Research, 2001.

B. Cowgill, J. Wolfers, and E. Zitwewitz. Using Prediction Markets
to Track Information Flows: Evidence from Google. 2008.