Algorithmic Game Theory 2, Exercises - Computer Science, Exercises of Game Theory

Prof. Sebastian Lehaie, Computer Science, Algorithmic Game Theory, Columbia, Lecture Notes

Typology: Exercises

2010/2011

Uploaded on 11/05/2011

yorket
yorket 🇺🇸

4.4

(38)

276 documents

1 / 2

Toggle sidebar

This page cannot be seen from the preview

Don't miss anything!

bg1
CS6998-3: Problem Set # 2
Due in class, on Monday October 27
Problem 1
AGroves mechanism is a sealed-bid auction where each agent ireports a valuation ˜vi, the mechanism selects
an efficient allocation Rwith respect to ˜v, and charges each agent ia payment of
qi=hivi)X
jNi
˜vj(Rj),
where hiis a function of the other agents’ reports. In the following, assume that agent valuations are drawn
from the domain of general valuations.
(a) Show that in a Groves mechanism, an agent maximizes its utility by reporting its true valuation.
(b) Show that among all individually-rational Groves mechanisms, the VCG mechanism maximizes the
payment of each agent.
Problem 2
Consider a set of nagents whose valuations v1,v2, . . . , vnare single-minded.
(a) Show that the following nonlinear, anonymous (i.e., order 2) prices are competitive equilibrium prices.
p(S) = max
iNvi(S).
That is, given any efficient allocation R, the pair hR, piis a competitive equilibrium.
(b) Exhibit a linear program for the efficient allocation problem such that the dual variables corresponding
to the “supply equals demand” constraints are nonlinear, anonymous prices. (You do not need to
exhibit the dual, but might need to derive it for yourself to ensure your primal is correct.)
Problem 3
Consider a model with a set of agents Nand a set of items Mwhere the number of items equals the number
of agents, say n. For each agent iNlet vij be the agent’s value for item jM. Each agent has a
unit-demand valuation, meaning that the value of bundle Sto a typical agent iis
vi(S) = max
jSvij.
In words, given a bundle S, the agent keeps the item it values the most in Sand discards the rest (hence
“unit-demand”). Assume further that vij is of the form vij =aibj, where a1> a2> . . . > an>0 and
b1> b2> . . . > bn>0. This means that each agent strictly prefers item 1 to item 2, item 2 to item 3, and
so on.
1
pf2

Partial preview of the text

Download Algorithmic Game Theory 2, Exercises - Computer Science and more Exercises Game Theory in PDF only on Docsity!

CS6998-3: Problem Set # 2

Due in class, on Monday October 27

Problem 1

A Groves mechanism is a sealed-bid auction where each agent i reports a valuation ˜vi, the mechanism selects an efficient allocation R with respect to ˜v, and charges each agent i a payment of

qi = hi(˜v−i) −

j∈N −i

˜vj (Rj ),

where hi is a function of the other agents’ reports. In the following, assume that agent valuations are drawn from the domain of general valuations.

(a) Show that in a Groves mechanism, an agent maximizes its utility by reporting its true valuation.

(b) Show that among all individually-rational Groves mechanisms, the VCG mechanism maximizes the payment of each agent.

Problem 2

Consider a set of n agents whose valuations v 1 , v 2 ,... , vn are single-minded.

(a) Show that the following nonlinear, anonymous (i.e., order 2) prices are competitive equilibrium prices.

p(S) = max i∈N

vi(S).

That is, given any efficient allocation R, the pair 〈R, p〉 is a competitive equilibrium.

(b) Exhibit a linear program for the efficient allocation problem such that the dual variables corresponding to the “supply equals demand” constraints are nonlinear, anonymous prices. (You do not need to exhibit the dual, but might need to derive it for yourself to ensure your primal is correct.)

Problem 3

Consider a model with a set of agents N and a set of items M where the number of items equals the number of agents, say n. For each agent i ∈ N let vij be the agent’s value for item j ∈ M. Each agent has a unit-demand valuation, meaning that the value of bundle S to a typical agent i is

vi(S) = max j∈S vij.

In words, given a bundle S, the agent keeps the item it values the most in S and discards the rest (hence “unit-demand”). Assume further that vij is of the form vij = aibj , where a 1 > a 2 >... > an > 0 and b 1 > b 2 >... > bn > 0. This means that each agent strictly prefers item 1 to item 2, item 2 to item 3, and so on.

(a) Show that the unique efficient allocation is to give item 1 to agent 1, item 2 to agent 2, and so on.

(b) Let (p 1 , p 2 ,... , pn) be linear, anonymous (i.e., order 1) prices. These are competitive equilibrium prices if and only if the following hold. (You should convince yourself of this.)

vii − pi ≥ vij − pj ∀i ∈ N, j ∈ M pi ≥ 0 ∀i ∈ N

Show that this system of inequalities can be simplified to the following.

vii − pi ≥ vii+1 − pi+1 for i = 1,... , n − 1 vii − pi ≥ vii− 1 − pi− 1 for i = 2,... , n pn ≥ 0

(c) Show that if p and p′^ are order 1 competitive equilibrium prices, then so are prices p ∧ p′. The latter is the price vector whose ith component is min{pi, p′ i}.

(d) The previous part implies that there is a unique minimal competitive equilibrium price vector ¯p, such that pj ≥ p¯j (for all j ∈ M ) for any other linear, anonymous competitive equilibrium prices p. Verify that p¯i =

j>i

aj (bj− 1 − bj ).

(Note that this construction proves that linear, anonymous competitive equilibrium prices always exist.)

(e) What is the VCG payment of agent i, in terms of the ai’s and bj ’s?