CS711: Introduction to Game Theory and Mechanism Design Jul-Nov 2018

Project: Assignment 2

Akshat Jindal : 150075

Question 1

Part (a)

We firstly comment that since f is SP, it is MONO too according to the result in the lecture notes. We
construct a new preference profile P ˆ where the - means any legit alternative out of {a,b,c}.
0 0
P1 P2 P1 P2 P1ˆ P1ˆ
a c b a b c
b b a b a -
c a c c - -
Before we move on, we will prove 3 lemmas and then use them later.
0 0 0
Lemma 1.1 Let P : P1 (1) = a 6= b = P2 (1) and P : P1 (1) = a 6= b = P2 (1). Then, if f (P ) = a, then
0
f (P ) = a
Proof: We construct a new preference profile P ˆ where the - means any legit alternative out of {a,b,c}.
0 0
P1 P2 P1 P2 P1ˆ P2ˆ
a b a b a b
- - - - b a
- - - - c c
Please note that the third preference in P ˆ HAS to be c as this is the only alternative left if we fix a and b
in the top 2 positions which we want in our construction.

0
Let’s assume f (P ) = b.

0
A transition form P to P ˆ is clearly monotonic as the best position b can take for Player 1 is at the 2nd
0
spot which it is in P1ˆ and b is in the 1st position in P2ˆ anyway. Thus since f (P ) = b, f (P ˆ = b).

A transition form P to P ˆ is also monotonic as the best position b can take for Player 1 is at the 2nd spot
which it is in P1ˆ and b is in the 1st position in P2ˆ anyway. Thus since f (P ) = a, f (P ˆ ) = a)

0 0
Clearly, we have a contradiction as a 6= b. Thus our assumption that f (P ) = b is incorrect. Thus f (P ) = a.

0 0 0
Lemma 1.2 Let P : P1 (1) = a 6= b = P2 (1) and P : P1 (1) = c 6= a 6= b = P2 (1). Then, if f (P ) = a, then
0
f (P ) = c
Proof: We construct a new preference profile P ˆ where P2ˆ is the same as P2 and where the - means any
legit alternative out of {a,b,c}.

1-1
1-2

0 0
P1 P2 P1 P2 P1ˆ P2
a b c b c b
- - - - a -
- - - - b -
Please note that the third preference in P1ˆ HAS to be b as this is the only alternative left if we fix a and c
in the top 2 positions which we want in our construction.

0
Let’s assume f (P ) = b.

0
Clearly, the transition from P to P ˆ follows Lemma 1.1. Thus, f (P ˆ ) = b.

Realise that aP1ˆ b. However, we see f (P ˆ ) = b and a unilateral deviation of Player 1 to P1 from P1ˆ changes
the outcome to a. Thus Player 1 benefits from unilateral deviation which violates the fact that f is SP. Thus
0
our assumption is wrong and f (P ) = c

0 0 0
Lemma 1.3 Let P : P1 (1) = a 6= b = P2 (1) and P : P1 (1) = a 6= b 6= c = P2 (1). Then, if f (P ) = b, then
0
f (P ) = c
Proof: We construct a new preference profile P ˆ where P1ˆ is the same as P1 and where the - means any
legit alternative out of {a,b,c}.
0 0
P1 P2 P1 P2 P1 P2ˆ
a b a c a c
- - - - - b
- - - - - a
Please note that the third preference in P2ˆ HAS to be a as this is the only alternative left if we fix c and b
in the top 2 positions which we want in our construction.

0
Let’s assume f (P ) = a.

0
Clearly, the transition from P to P ˆ follows Lemma 1.1. Thus, f (P ˆ ) = a.

Realise that bP2ˆ a. However, we see f (P ) = b and a unilateral deviation of Player 2 to P2 from P2ˆ changes
the outcome from a to b. Thus Player 2 benefits from unilateral deviation which violates the fact that f is
0
SP. Thus our assumption is wrong and f (P ) = c
0
Now, we get back to our original table and preferences. Here, we are given f (P ) = a. Let f (P ) = a. Now,
0
the transition from P to P ˆ follows Lemma 1.3. Thus f (P ˆ ) = c. But we also observe that the transition
from P to P ˆ follows Lemma 1.2. Thus f (P ˆ ) = b. Since b 6= c, we reach a contradiction and thus our
0 0
assumption that f (P ) = a is incorrect. Thus f (P ) = b.

Part (b)

No, the earlier conclusion does not hold here.

To see exactly where the proof faulters, observe the proof of Lemma 1.2, which is used in part (a). The
lemma is used for the transition from P tp P ˆ . Just like we had to construct a third profile there, we look at
the third profile that needs to be created here. (We call it P temp )
 1-3

P1 P2 P1ˆ P2ˆ P1temp P2temp

a c b c b c
b b a - a -
c a - - - -
Now, even while we were proving Lemma 1.2, to remark on the transition from the 2nd to the 3rd profile
which is P ˆ to P temp here, Lemma 1.1 had to be invoked and would have required the construction of the
following profile :
P1tp P2tp
b c
c b
a a
Clearly, P2tp is not single peaked and thus, the proof would get stuck here.

A mechanism that can be proposed is the following :

f (P ) = min Pi (1)
∀i∈[2]

i.e, picking the leftmost peak as the social outcome everytime. Clearly by this mechanism both f (P ) = a and
0
f (P ) = a as a › b in the intrinsic ordering.

Question 2

We define a few things :

• A : The set of alternatives which is basically 2|X| , the powerset of X.

0
• Ri : The extension of Pi defined over A as mentioned in the question

0
Claim 1.4 Ri is restricted in its domain, i.e not all preference orderings over A are allowed.
Proof: Let a1 , a2 ∈ X. We then construct the following subsets of X:

• S1 : {a1 }
• S2 : {a2 }
• S3 : {a1 , a2 }

0
Now we observe the ordering S1 › S2 › S3 where › denotes strict preference in accordance to Ri . We know
that either a1 Pi a2 (Case 1) or a2 Pi a1 (Case 2).

0
• Case 1: Then by the definition of Ri , since S3 has a1 as the best project and S2 has a2 as the best
project, S2 ›S3 is impossible.
0
• Case 2: Then by the definition of Ri , since S1 has a1 as the best project and S2 has a2 as the best
project, S1 ›S2 is impossible.
1-4

0
Thus, for any arbitrary pair of projects a1 and a2 , the ordering S1 › S2 › S3 is not allowed in Ri . Thus
0
Ri is domain restricted.
Since we know that the GS theorem needs the preferences to be unrestricted, it cannot be applied in this
scenario.

Question 3

Yes, we claim that the median voter SCF is indeed group strategy proof.

Claim 1.5 Median voter SCF is group strategy proof.

Proof: Let f (P ) = a = median of all agent peaks and phantom peaks.

First, we consider all K ⊆ N s.t i ∈ K where Pi (1) = a.

Clearly, agent i cannot do better than he is doing right now as he is achieving his peak preference. Thus, no
0
matter how the agents deviate to other preferences, the clause f (PK , P−K )Pj f (PK , P−K ) will never be true
for j = i. Thus, in this case, f is group SP.

Now, we consider all K ⊆ N s.t ∀i ∈ K, Pi (1) 6= a.

Let ml + mr =| K | s.t ml denotes the number of agents in the group with their peaks to the left of a and
mr denotes the number of agents in the group with their peaks to the right of a.
0
Now, we can imagine that a new PK forms from PK if atleast 1 agent deviates to a new preference. Now,
within ml , the only way agents can cause a change in outcome is if atleast 1 of them reports a peak to the
right of a. Within mr , the only way agents can cause a change in outcome is if atleast 1 of them reports a
peak to the left of a.

• Case 1 : The number of deviating agents in ml > the number of deviating agents in mr . In this case
0 0
the median, i.e f (P ) = a lies to the right of a. Since preferences are single peaked, all agents in
0 0
ml ,(whos peaks lie to the left of a) prefer a over a . Thus, the clause f (PK , P−K )Pj f (PK , P−K ) will
not be true ∀ j ∈ ml .Thus, in this case, f is group SP.
• Case 2 : The number of deviating agents in mr > the number of deviating agents in ml . In this case the
0 0
median, i.e f (P ) = a lies to the left of a. Since preferences are single peaked, all agents in mr ,(whos
0 0
peaks lie to the right of a) prefer a over a . Thus, the clause f (PK , P−K )Pj f (PK , P−K ) will not be
true ∀ j ∈ mr .Thus, in this case, f is group SP.
• Case 3: The number of deviating agents in ml = the number of deviating agents in mr . In this case
0 0
the median doesnt change and f (P ) = a. Thus, the clause f (PK , P−K )Pj f (PK , P−K ) will not be true
∀ j ∈ K.Thus, in this case, f is group SP.

Thus, ∀K ⊆ N , f cannot be manipulated by K.