Predi

tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast

Predi tion Markets

Market Design, Fall 2011

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Logisti s

Friday, November 11  No lass (Veteran's Day)

Friday, November 18  Clayton Featherstone, Al Roth: S hool
Choi e
Friday, November 26  No lass (Thanksgiving)
Friday, De ember 2  Former students Ja ob Leshno and
Eduardo Azevedo; on luding omments from Al
Monday, De ember 12, 11:59PM  Submit proje ts!

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Outline

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design

Information Aggregation

Corporate Predi tion Markets / Crowd ast

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Predi tion Markets vs Finan ial Markets

Question: What is the purpose of the sto k market?

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Predi tion Markets

in entives for truthful revelation

in entives for resear h and information dis overy
me hanism for aggregating opinions

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
A Typi al Contra t

Contra t payos depend on unknown future events

Example: ontra t pays o $1 if Y o urs by time T , 0 otherwise.

Event realizations must be spe i .

Good: The Patriots win Super Bowl XLV
Bad: Tom Brady gets injured in November.
Warning! Seemingly well-dened ontra ts an have holes.

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Winner-Take-All Contra ts

Winner-Take-All Contra t:

Pays out $1 if a spe i event o urs, 0 otherwise.

Market pri e p for this ontra t gives the market's
expe tation of the probability that this event will o ur

Why?

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Interpreting Pri es

Manski (2006) modeled risk neutral bettors with xed budgets.

Continuum of bettors
Ea h better has beliefs about event E
Finding: Equilibrium pri e yields bounds on mean beliefs

Gjerstad (2005) and Wolfers & Zitzewitz (2005)

Extended Manski model to in lude risk aversion
Finding: With logarithmi utility, equilibrium pri e=mean
beliefs.

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Winner-Take-All Contra t: Example

The Saddam Se urity

Paid out $100 if Saddam ousted from power by June 2003.
Traded on website tradesports. om
Pri e on Jan 1, 2003: $55
Pri e on Mar h 1, 2003: $70

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Index Contra ts

Contra ts pay out the value of a spe i future event.

Example:

Contra t pays $1 for ea h per entage point of the 2012 popular

vote won by Sarah Palin.

Market pri es of index ontra ts measure means.

Let y be Palin's per entage of the popular vote.
Market pri e of ontra t reveals E [y ].

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Spread Contra ts

Amount of bet is xed.

Market trades based on utos that determine whether event
o urs

Example:

Contra t pays even money if Sarah Palin wins more than y % of the
popular vote.
Market trades based on y (I'll buy 10 units at y = 43)
Market pri e reveals median value for y . (This is a fair bet if event
is as likely to o ur as not)

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
More Compli ated Contra ts

Carefully onstru ted markets an reveal information about the

distribution of an un ertain future event:
index ontra t 1 pays y

index ontra t 2 pays y

Market pri es reveal E [y

2 ] and

[ ]

E y .

In general, ontra ts an be onstru ted to provide any desired

order statisti about distributions.

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Contingent Contra ts

Cleverly designed ontra ts an predi t ontingent events.

Example: Ele tability
Candidates / Party vote share

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
How are Contra ts Traded?

In many predi tion markets, the me hanism used is the ontinuous

double au tion.

This is the me hanism used on the NYSE.

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Market Design Choi e: IEM Model

The Iowa Ele troni ex hange trades ontra ts that pay o on

mutually ex lusive events.
A set of mutually ex lusive events is alled a basket (or a bundle.)
IEM never loses money!
To enter the market, you may either
Go long on a position
Ex hange $1 for a basket of goods.

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Market Design Choi e: IEM Model

Drawba ks:
Two ways to make the same bet. (if you hold positions)
Arbitrage?
Hard to take a short position. Must rst buy a basket.
Your ash is held by IEM - no interest.

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Market Design Choi e: InTrade Model

To bet against an event, short the asset.

Suppose an asset that pays o $1 if y o urs is trading at pri e p .

Shorting
gives you p dollars now...
...but you are on the hook for $1 if y o urs. That is, $1 of
assets are frozen in your a ount to make sure you an over
your bet.
Like margin trading.

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
A ura y of Predi tion Markets

Hollywood sto k ex hange: Os ar predi tions

Iowa Ele troni Ex hange vs Gallup

HP Printer Sales vs internal sales predi tion methods

Google internal predi tion markets

But...an open question.

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Predi tion Market Challenge 1: Real vs Play Money

Question: Does Money Matter?

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Real vs Play Money: Empiri al Finding
Emile Servan-S hreiber, Justin Wolfers, David M. Penno k & Brian
Galeba h, Does Money Matter?, Ele troni Markets, 2004.
2003 NFL season (14 games)
Comparison of two websites:

TradeSports: Real $$
NewsFutures SportsEx hange, in partnership with
Play $$

USA Today:

Findings:
Neither site systemati ally out predi ted the other
Predi tions from both sites were remarkably a urate

Authors entered aggregate opinions into a third sports betting

site (whi h used play money.)
TradeSports and NewsFutures took 11th and 14th pla e out of
2000 experts, respe tively.

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Unlawful Internet Gambling Funding Prohibition A t (2003)
AN ACT To prevent the use of ertain bank instruments for
unlawful Internet gambling, and for other purposes.
The Congress nds as follows:

Internet gambling is primarily funded through personal use of

bank instruments, in luding redit ards and wire transfers.

The National Gambling Impa t Study Commission (1999)

re ommended legislation to prohibit wire transfers to Internet
gambling sites or the banks whi h represent them.

Internet gambling is a major ause of debt olle tion problems

for insured depository insts. & the onsumer redit industry.

Internet gambling ondu ted through oshore jurisdi tions has

been identied by United States law enfor ement o ials as a
signi ant money laundering vulnerability.

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Statement on Predi tion Markets (Brookings JCRS)

Predi tion markets are markets for ontra ts that yield payments
based on the out ome of an un ertain future event, su h as a
presidential ele tion. Using these markets as fore asting tools ould
substantially improve de ision making in the private and
publi se tors.

We argue that U.S. regulators should lower barriers to the

reation and design of predi tion markets by reating a safe

harbor for ertain types of small stakes markets. We believe our

proposed hange has the potential to stimulate innovation in the
design and use of predi tion markets throughout the e onomy, and
in the pro ess to provide information that will benet the private
se tor and government alike.

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Statement on Predi tion Markets ( oauthors)

Coauthors: Kenneth J. Arrow, Robert Forsythe, Mi hael Gorham,

Robert Hahn, Robin Hanson, Daniel Kahneman, John O. Ledyard,
Saul Levmore, Robert Litan, Paul Milgrom, Forrest D. Nelson,
George R. Neumann, Mar o Ottaviani, Charles R. Plott, Thomas C.
S helling, Robert J. Shiller, Vernon L. Smith, Erik Snowberg,
Shyam Sunder, Cass R. Sunstein, Paul C. Tetlo k, Philip E.
Tetlo k, Hal Varian, Justin Wolfers, and Eri Zitzewitz

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Predi tion Market Challenge 2: Thin Markets

Thin betting on events like

Who will be the next pope?
Who will be the next nobel prize winner in e onomi s?

Why is it risky to pla e a limit order in a thin market?

=> On tradesports, limit orders are free.
Why is it risky to make any trade in a thin market?

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Proper S oring Rules

A proper s oring rule is a method for soli iting a probability

distribution from an individual.
Let X be a random variable that takes on K possible states
(indexed by i ).
Let si () be a mapping that des ribes a state ontingent payo
from a reported probability distribution r for X .
si

()

is a proper s oring rule if it is in entive ompatible for an

individual with beliefs p to report beliefs truthfully. That is,

= arg max
r

"

()

pi si r

Peter Coles

and

"

( ) 0.

pi si p

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Examples of Proper S oring Rules

Quadrati S oring Rule (Brier, 1950)

( ) = a +b

si r

2ri

r2 .
j

Logarithmi S oring Rule (Good, 1952)

( ) = a + b log(r ).

si r

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Market S oring Rules

Robin Hanson (2003) insight: Let market parti ipants orre t ea h

others' probabilities!
Market for variable X whi h an take on K states.
Marketpla e `sponsor' hooses initial distribution q.
Market parti ipants may adjust the distribution to r.
Ea h user makes payment (upon realization of the event) to the
previous user who has made an adjustment:

= s (r ) s (r
i

Peter Coles

).

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Market S oring Rules

= s (r ) s (r
i

).

Market sponsor subsidizes the market

Subsidy is apped

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Market S oring Rules as Automated Market Maker

How to adapt market s oring rules to the notion of buying and

selling assets?
Insight: Let users adjust s !

For ea h asset i , si (p ) represents `shares outstanding.'

Bettor names quantity si she would like to buy or sell; asks
how mu h?
New pri es would given by p

1 (s

+ s )

Amount quoted for bundle given by integral of pri es over

quantity range.

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Market S oring Rules

= s (r ) s (r
i

).

Market sponsor subsidizes market; Subsidy is apped

Has `feel' of buying and selling assets.
Can be operated alongside traditional predi tion market, where
assets payout of 1 when state is i  are traded.
Quoted pri es? Probabilities represent pri es for very small
trades
Predominant market design in internal orporate predi tion
markets.
Open Question: How well do market s oring rules aggregate

information? (See e.g. Ostrovsky 2009)

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Appli ations

HP Printer sales

Avian Bird u

Google Predi tion Markets (HBS Case)

Cambrian House (HBS Case)

Crowd ast (in luding drug predi tion markets)

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Open Resear h Questions
Theory:
How well do various markets aggregate information?
What role does a `balan ed budget' (or apped subsidy) play
in limiting market me hanism?
What happens with predi tion markets are taken seriously to
make de isions? (feedba k loop)

Empiri al:
Data from Crowd ast, other sour es.
How many users do we need for a urate predi tions?
When an we expe t a urate predi tions? (type of market,
type of event, type of user et )

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
No-Trade Theorem Logi

Aumann (1976), Milgrom-Stokey (1982)

No trade amongst partially informed traders
Solution?
Naive Traders
Noise Traders
Non-Strategi Traders
Market S oring Rules

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Ostrovsky (2009)

Logi :
Sequen e of partially informed traders make adjustments to
probabilities; payos a ording to MSR
Players are strategi :

Consider history of adjustments

Consider impa t of adjustments on future pri es
Consider Perfe t Bayesian Eqm:

Do equilibrium pri es aggregate all available information?

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Ostrovsky (2009) ont.

Finding:
Yes, provided the se urity is separable (given an information
stru ture).
A se urity is separable if, roughly, for any prior distribution of
beliefs, there is at least one informed trader.
Example of non-separable se urity: (see Ostrovsky 2009)

Peter Coles

Predi tion Markets

Predi tion Markets Overview + Types of Contra ts

Predi tion Market Design
Information Aggregation
Corporate Predi tion Markets / Crowd ast
Market Design 2011 Predi tion Market

Peter Coles

Predi tion Markets