CS 364A Algorithmic Game Theory Stanford University

CS364A Exercise Set 1

Austin Liu SUNet ID: yliu2
April 19, 2019

AGT - Algorithmic Game Theory by Nisam, Roughgarden, Tardos, Vazirani

Exercise 1
Olympic badminton tournament format: There are four groups (A,B,C,D) of four teams each. The tourna-
ment has two phases. In the first “round-robin” phase, each team plays the other three teams in its group,
and does not play teams in other groups. The top two teams from each group advance to the second phase,
the bottom two teams from each group are eliminated. In the second phase, the remaining eight teams play
a standard “knockout” tournament (as in tennis, for example): there are four quarterfinals (with the losers
eliminated), then two semifinals (with the losers playing an extra match to decide the bronze model), and
then the final (the winner gets the gold, the loser the silver).
The eight winners from the round-robin phase are paired up in the quarterfinals as follows: The team with
the best record from group A plays the second-best team from group C in the first quarterfinal; similarly
with the best team from group C and the second-best team from group A in the third quarterfinal; and
similarly with the top two teams from groups B and D in the second and fourth quarterfinals.
So there is an incentive to intentionally lose a match if highly ranked team is ranked second-best in a
group.
Possible solutions:

A. Start with knockout immediately, eliminate the round-robin phase.

B. Four groups of four teams play round robin but only highest ranked team in each group advances to
semifinal.

Exercise 2
The Nash equilibrium captures the notion of a stable solution, a solution from which no single player can
individually improve his or her welfare by deviating.
A. One shot-simultaneous move game (AGT 1.2), 4 men, 5 women choices, one woman is preferred more
by all the men. The John Nash character proposes that his four friends go for the woman’s four friends.
This is not a Nash equilibrium since any one of the men can improve his utility by going for the blonde.
B. The solution is a Nash equilibrium in a sequential game in which, for instance, each woman can only
talk to one person at a time for five minutes. If all of the four men go for the blonde, then the blonde
only talks to one of them, the other three talk to the remaining women. If the blonde is not impressed,
the blonde’s friend whom no one talks to will not be happy with being second choice (model this as
reduction in probability of success or otherwise). Possible for that man to go home himself. (Refine
with more rigor)

Austin Liu: [email protected] 1

CS 364A Algorithmic Game Theory Stanford University

Exercise 3
Rock-Paper-Scissors game

Rock Paper Scissors

Rock 0,0 -1,1 1,-1
Paper 1,-1 0,0 -1,1
Scissors -1,1 1,-1 0,0

From Lecture 1: If both players randomize uniformly in Rock-Paper-Scissors, then neither player can
increase their expected payoff via a unilateral deviation - indeed, every unilateral deviation yields zero
expected payoff to the deviator. A pair of probability distributions with this property is a (mixed-strategy)
Nash equilibrium.
Consider 1/3,1/3,1/3 for R,P,S strategy for player. Expected payoff against 1/3,1/3,1/3 for R,P,S for
opponent is zero. Consider 1/3,1/3+,1/3- for R,P,S strategy for player. Expected payoff is still zero.
Conclude without loss of generality that the pair with 1/3,1/3,1/3 is a mixed strategy Nash equilibrium.

Exercise 4
eBay auction: “proxy bidder” increases your bid on your behalf until either your maximum bid is reached
or you are the highest bidder, whichever comes first.

sealed-bid second-price auction: one round in which highest bidder wins and pays a price equal to the
second highest bid.

This implies that the eBay auction is effectively a second price auction, since you only pay the highest
other bid (the second highest overall), plus a small increment. Bids are sealed, there is a preset time when
bids are opened, so even in presence of “snipers” the previous proofs for second-price auctions work.

Exercise 5
Single item auction with at least three bidders. Want to show that awarding the item to a bidder at price
equal to third-highest bid is not a dominant strategy incentive compatible auction.
A dominant strategy maximizes one’s utility regardless of what others do.
Every truthful bidder is guaranteed non-negative utility (incentive compatibility).
Consider following case: (b1 , b2 , b3 ) = (v1 , v2 , v3 ) where v1 > v2 > v3 .
2 gets no utility from bidding truthfully, but utility v2 − v3 from overbidding v1 . This implies an incentive
to overbid.

Exercise 6
k identical copies of a good and n > k bidders. Each bidder can receive at most one good. Analog of
second-price auction is a k + 1-th price auction.
Auction is indeed DSIC.
Claim 1: Setting bid bi equal to private valuation vi maximizes utility of bidder i regardless of what
other bidders do.
Let B be the k-th highest bid, if bi < B, then i loses and receives utility 0. If vi < B, this is the highest
possible utility, obtained by bidding truthfully and losing. If bi ≥ B, then i wins at price B and receives
utility vi − B, obtained by bidding truthfully and winning.
Claim 2: Every truthtelling bidder is guaranteed non-negative utility
Losers get utility 0, winners get utility vi − B only when vi ≥ B.

Austin Liu: [email protected] 2

CS 364A Algorithmic Game Theory Stanford University

Exercise 7
Procurement/reverse auction:
Contractors report their cost valuation vi < 0, select second lowest bid. All is as before.

Exercise 8
Sponsored search: bidder i has valuation Pn vi per click. k click-through-rates α1 ≥ α2 . . . αk . Social surplus
of an assignment of bidders to slots is i=1 vi xi . xi is CTR of slot to which i is assigned or 0 if bidder not
assigned any slot.
Want to show that the social surplus is maximized by assigning the bidder with the jth highest valuation
to the jth slot for j = 1, 2, . . . , k (set xi = αi where v1 ≥ v2 ≥ · · · ≥ vn )
Suppose there exists a better valuation with i, j swapped, αi ≥ αj . Change in valuation is:

αj xi + αi xj − αi xi − αj xj = (αj − αi )(xi − xj )

But since αi ≥ αj , above quantity is negative, contradicting claim that social surpus can increase in this
ordering. Hence initial ordering maximizes social surplus.

Austin Liu: [email protected] 3