Plan for Today Continue with auctions Sponsored Search The VCG - - PowerPoint PPT Presentation

SLIDE 1

Plan for Today

• Continue with auctions
• Sponsored Search
• The VCG auction
• The FCC Incentive Auction

SLIDE 2

Last

time

A

Iii

17g

c

I

price auctioni

winner

is highest bidder

winner pays her bid

2nd price archeni

winner is highest bidder

winner pays 2ndhighest bid

How to

bid

Model

each bidder bids to

maximize

expected

utility

givenbeliefs aboutothers

ViIneceivesitem

pi

utilityof

bidders

value

payment

Vickrey auction

2nd price

anchen

truthful

also called

strategy

proof

dominantstrategy incentivecompatible

Dsic

it is

a dominant strategy to

bid your value

SLIDE 3

also

individually rational

IR

bidder's utility guaranteedto

be nonnegative

maximizes

afar

also

called

surplus

total

happiness of all

participants

sum of

utilities

IE viIiwmsien pi tiEpi

utilityof bidder i utilityof

auctioneer

n

I

wi Ii

wins

F

I

SLIDE 4

1st price auction

no

dominant strategies

analyzed byassuming each player's

value

was

drawn from

distribution

F

Showed that if

F

is

uniform

0 too

then

par

L

is

a

Bayes

Nash

equilibrium far h

2

F

utility ofbidding pCyj1ohenp6yp

vidagJJzEfuteeygb.dd.gov

• tmgfYfbjds

pay

RevenneEquivalence

Thm

µ

If

Vin

F

fi then

for any anchor

sit

in

equilibrium

item

allocated

to

bidden uf

highest

value and payment of

a

player

cuh

valve Oiso

has

same expected

revenue

SLIDE 5

SLIDE 6

Advertising – how it used to be

Posters newspapers magazines Billboards television

SLIDE 7

Pay-per-impression

Price depends on how many people your advertisement is shown to. (whether or not they look at it, or care about it) “Half the money I spend on advertising is wasted; the trouble is, I don’t know which half” Andrew Wanamaker, advertising pioneer

SLIDE 8

Complicated negotiations with high monthly premiums, forms a barrier to entry for small advertisers.

How is the price determined?

SLIDE 9

Modern Advertising

On the web, Many different kinds of ads …

SLIDE 10

Sponsored Search Ads

SLIDE 11

Price determined by auction – per keyword

Slot One Slot Two Ad A Ad B Ad C

“Most people don’t realize that all that money comes in pennies at a time” Hal Varian, Google Chief Economist

SLIDE 12

How are these ads different than the ads in the offline media?

In In many cases: Pay-pe per-cl click ck

Model

perkeyword

auction

Bidders

advertisers who have

astanding bid

• n

keyword

Auction

• ffer

some

K

1

• f slots

Slots

notidentical

CTRchukthroughrat

Cj

Probof aclick

• n the jh

slot

j

l

K

vii

adverts's

value

for

a

click

9i

CTR rate

allqualities

are sated

due to gualitygad

f

d

l

l

SLIDE 13

• verall

ipf put

bidder i's ad

instotj

Prob of

click

Cj'gi

Value

to

bidder

i

f

gets

his

ad

placed

in

slot's

vi

Cj

exp

exp

utility

v

g

pig

Generalized Second

Price

GSP

aH

loBi

g

bid

• n

a

click to

slot j payment

is

next

highest bid

SLIDE 14

Model

A B C

Slots

Advertisers

Click-through rates

1 0.4

Value per click

7 6 1

1 2

No slot

values

PPC

utility

bids

c

6

At

cz

I

6

1

0.4

Cz

SLIDE 15

Example

bidder

I

2

3

values

• 0.05

E

0.99

Ijg

in this example

Y bidder

1

bids

below9

then his utility will

be higher than it

was

biddingtruthfully

Truthful anchor

VCG

auction

chooses

• utcome to

maximize socialwelfare

V S Vy S

social welfare of

an

allocation

EIvicspcis

slot i

is assigned to

C

Cz Cz

YIximizing

allocation

charge each

bidder

the

externality

her presence

Imposes

• n others

IBidduliT

Whats

w

be

way

bidder1 in archer

I

i

a

SLIDE 16

VCE paymentfor bidder I

charge bidder

loss

• thers

incur becauseof her Ktl

cute

O

ai

i

c c

Ci

i

2

VCG payment for

bidder

j

KH

ci

i

a

Vi in jts

CG auction for

sponsored search T

aipi

ftp.bjccji 5

Geta bid from each bidder

w

Relabel bids

so

b

by

bn

for i

I

K

assignbidderi

to

ihslots

bit GiI bitafinicity

for i

h K

for each click

chargebidderi

KH

PPC

pi

I

b Kj

c

j

itithm

VCG.is

truthful

IR

andwelfare maximizing

Progg tmtfeness

Fix

• ther

bids

let

bj

denotejthhighest

bid except

me

my value

slot k

Ckbk

K S

Ca Ca be 1 cable

V

Li

A value

as

movefrom i

to slot

i 1

i

i

l

cm

c

bi I 0 price

want to

do

this

if

v

bi

i

SLIDE 17

bi

S

V

bit

• therbids

P

b by

valuations

CTRs

the GSP

auction

has

an equilibrium

y

same allocation

payments

as

VCG

VCGpaying

p

PI

PIE

no

sweeping

revenue comparison between

GSP

VCE

SLIDE 18

Facebooky what bidders

can bid on

Various

events

placement

size

format ofads

evaluating quality

• f ads

vi Cj

more appropriate model

bidder has

value far

each eventtype difficult

for

advertisers

to figure out CTRs

high quality outcomes

advertisers happy biddig

easy

CG fully general

n

participants

finite set of

• utcomes of auction

I

each agent i

has private

valuation

v

w

Fw c R

vilwt fauegsjntjpr.e.int

aemsw

F

VCI

choose

• utcome

C

argmax Ebilw

w C I 5

1

Lettnis

• n

Mw

07

SLIDE 19

O

P

mEq

j bjlwT

ui

view'T

pi

View t.F.bjw9

wmeoxfpbo.tw

choose bin t.wo.GL

to

maximize

vi

w t

bjlw

J

X

wEqgrax

Userinterface for bidding

allow bidders

band eventsthey

allow

bidders

to specify budget

Computational requirements

SLIDE 20

Pa

total

value others valve O

getif a matinee

theyget

5

wana

5

Pb

3 1

4