PhD Micro II: Game Theory

Problem Set 1:
Strategic and Extensive Form Games
Solutions
Lecturer: Amrish Patel
Teaching Assistant: Conny Wollbrant

Class date: Friday 12th November

1a. Let p be the probability 1 chooses A and q be the probability 2 chooses X.

Assume p and q are independent.
Player 3’s payo¤ from playing K is

u3 (K; p; q) = 4pq + 4 (1 p) (1 q)
= 8pq 4p 4q + 4:

Suppose K is a best-response for some p and q. This implies the following

inequalities

8pq 4p 4q + 4 u3 (J; p; q) = 8pq; (1)

8pq 4p 4q + 4 u3 (L; p; q) = 8 (1 p) (1 q) ; (2)

8pq 4p 4q + 4 u3 (M; p; q) = 3: (3)

Simplify (1)

8pq 4p 4q + 4 u3 (J; p; q) = 8pq (4)

4p 4q + 4 0
4 4p + 4q
1 p + q:

Now simplify (2)

1
 PhD Micro II: Game Theory

8pq 4p 4q + 4 u3 (L; p; q) = 8 (1 p) (1 q) (5)

8 (1 q p pq)
8 8q 8p 8pq
0 4 4q 4p
4 (p + q) 4
p+q 1:

Combining (4) and (5) we get p + q 1 p + q; so we conclude that p + q = 1

(q = 1 p): Use this in (3) so that we have

8pq 4p 4q + 4 3: (6)
8p(1 p) 4p 4(1 p) + 1 0
8p 8p2 4p 4 + 4p + 1
8p2 8p + 3 0 (7)

LHS of (6) is now quadratic in p: The discriminant = b2 4ac becomes

2
8 4(2)(3) = 32; such that it has no real roots. The LHS is always positive
–contradiction. Therefore K is never a best-response.

1b.Consider a mixed strategy, , for player 3 where (J) = p, (L) = q and

(M ) = 1 p q. Suppose strictly dominates K. (We only need to compare
against opponents’ pure strategies – see lectures) If 1 plays A and 2 plays X,
this implies

4 < 8p + 3 (1 p q) (8)
0 < 5p 3q 1

If 1 plays B and 2 plays X, this implies

0 < 3 (1 p q) (9)
0 < 3 3p 3q

If 1 plays A and 2 plays Y , this implies

2
 PhD Micro II: Game Theory

0 < 3 (1 p q) (10)
0 < 3 3p 3q

If 1 plays B and 2 plays Y

4 < 8q + 3 (1 p q) (11)
0 < 5q 3p 1

The inequality in (8) implies

3 1
p> q+ ; (12)
5 5

and the inequality in (11)

5 1
p< q (13)
3 3

Clearly both (12) and (13) cannot hold –contraction. Therefore K is not strictly
dominated.

1c. For games of more than two players the set of rationalisable strategies
(i.e., those that are not never-best responses) is not equal to the set surviving
iterated dominance. In this game we have a strategy that is a never best-
response, i.e., not rationalisable, but is not strictly dominated i.e., survives
iterated dominance. Therefore the set of rationalisable strategies is a subset of
the set that survive iterated dominance.

2a: Split the dollar: Continuum of NE, any pair of numbers that sum to 100.

2b: Beauty contest, NE: (0; 0) if assume that ties are broken by a lottery, or if
everyone gets a high pass if students tie then any strategy pro…le where everyone
says the same number is an NE.

3. Delete Y (strictly dominated by W ). Delete B (is now a never-best re-

sponse). Delete Z (is now strictly dominated by W ), leaving a 2x2 game where
all strategies are BRs and no pure strategy NE

W X
A 2, 5 4, 2
C 3, 2 2, 3

3
 PhD Micro II: Game Theory

MSNE: Player 1 plays A with probability p and player 2 plays W with proba-
bility q

q (2) + (1 q) (4) = q (3) + (1 q) (2)

2q + 4 4q = 3q + 2 2q
q = 2=3
p (5) + (1 p) (2) = p (2) + (1 p) (3)
5p + 2 2p = 2p + 3 3p
p = 1=4

4. Payo¤ function of player i, separating out choice variable from other players

0 1
n
X n
Y
@ 2
hj + hj A wi (hi ) (14)
j=1 j=1
X Y
hi + h i + hi h i wh2i

Di¤erentiating (14) with respect to the decision variable hi and setting it equal
to zero yields the …rst order condition (F.O.C)

Y
+ h i 2whi = 0:

Solving the F.O.C for hi yields the best response function

Y
2whi = + h i
Q
+ ( h i)
hi =
2w

I.e., the more others lobby, the more i lobbies for all > 0: But for a particular
level of hi to be a dominant strategy, we must play the same hi regardless of
others’strategies. For this to be the case, we must have that = 0, such that
hi = =2w, i.e., increasing in subsidy bene…t and decreasing in e¤ort cost (but
does not depend on h 1 :)

5a. A strategy is a number of cows, Si : ni 2 [0; 1): The payo¤ function is:

4
 PhD Micro II: Game Theory

ui (n1 ; :::; ni; :::; nF ) = ni v (n1 + ::: + ni + ::: + nF ) ni c (15)

= ni v(N ) ni c

5b. Given n i ; we need to …nd i’sPbest-response (BR). BR is either solution

to the F.O.C
P where ni 2 [0; N j6=i nj ) or boundary solution ni = 0, (if
ni = N n
j6=i j , then i makes negative pro…ts).

Di¤erentiating the payo¤ function ui with respect to ni (the decision variable)

yields

@ui
= v (N ) + ni v 0 (N ) c; (16)
@ni

and setting the expression equal to zero yields the F.O.C:

@ui
= 0 ! v (N ) + ni v 0 (N ) = c (17)
@ni

Let bi (n i ) denote solution to F.O.C, if it exists, otherwise set ni = 0. In a

symmetric NE, all players have the same number of cows grazing such that

bi (n ; :::; n ) = n ; (18)

which must satisfy the F.O.C in (17), so we have

v (F n ) + n v 0 (F n ) = c (19)

We have v (0) + 0v 0 (0) > c; since v (0) > c (by assumption). Therefore, n > 0:
Since by assumption, v(N ) = 0 if N N ; a solution must satisfy n 2 (0; N =F ).
Thus all farmers graze this n number of cows.

5c. Social welfare is given by

W = F (ni v(N ) cni ); F ni = N; (20)

W = N v(N ) cN

5
 PhD Micro II: Game Theory

(i.e., only total number of cows matters, not its distribution.) Di¤erentiating
(20) and setting equal to zero yields the F.O.C

@W
= N v 0 (N ) + v(N ) c
@N
0 ! N v 0 (N ) + v(N ) = c
0
N v (N ) + v(N ) = c (21)

In a symmetric equilibrium we must have that

F n v 0 (F n ) + v (F n ) = c (22)

Thus, the socially optimal n is the one satisfying the condition in (22).

5d. To show that the social optimum involves a lower production level than
does the Nash equilibrium, we need to show that W 0 (n ) < 0: Intuitively, if
the value of the derivative of the social welfare function is less than zero at the
Nash solution n , then the function W itself cannot be at its maximum.

@W
= F n v 0 (F n ) + v (F n ) c
@N Fn
= F n v 0 (F n ) + v (F n ) c + n v 0 (F n ) n v 0 (F n )
= (F 1) n v 0 (F n ) < 0:

and so the Nash level of production is socially ine¢ cient.

6a.
Win 4, 4, 4
C
Win
B Mac 2, 2, 0
Win Win 2, 0, 2
A Mac
B C Mac
Mac 0, 2, 2

win mac

0, 0, 0 2, 2, 0

6b. Anne chooses the row, Bjorn chooses the column and Caroline chooses
the matrix. Bjorn’s strategies are of the form (Xy) where X denotes his action

6
 PhD Micro II: Game Theory

choice if Anne plays W and y denotes his action choice if Anne plays M . Payo¤s
in the matricies are in the order (Anne, Bjorn, Caroline).

Bjorn Bjorn
Ww Wm Mw Mm Ww Wm Mw Mm
Anne W 4, 4, 4 4, 4, 4 2, 0, 2 2, 0, 2 Anne W 2, 2, 0 2, 2, 0 0, 2, 2 0, 2, 2
M 0, 0, 0 2, 2, 0 0, 0, 0 2, 2, 0 M 0, 0, 0 2, 2, 0 0, 0, 0 2, 2, 0
Caroline plays W Caroline plays M

6c. BRs highlighted, note they coincide for 6 strategy pro…les, so there are 6
NE.
Bjorn Bjorn
Ww Wm Mw Mm Ww Wm Mw Mm
Anne W 4, 4, 4 4, 4, 4 2, 0, 2 2, 0, 2 Anne W 2, 2, 0 2, 2, 0 0, 2, 2 0, 2, 2
M 0, 0, 0 2, 2, 0 0, 0, 0 2, 2, 0 M 0, 0, 0 2, 2, 0 0, 0, 0 2, 2, 0
Caroline plays W Caroline plays M

6d. 3 subgames

6e Note that if player 1 chooses W in the simultaneous game there are two NE:
M M and W W: Use each of these in turn to identify the SPNE: W W W; M M M

6f. Same logic as lectures/textbooks –"empty threats".

7a. Player 1 cannot condition moves on anything, so his strategy set is equal to
his action set m1 . Player two can condition on player 1’s choice, so his strategy
m
set is equal to (m1 ) 2

7b. Assume the contrary, i.e., for (a1 ; a1 ) and (a01 ; a02 ) (where either a1 6= a01
or a2 6= a02 ) and u1 (a1 ; a2 ) 6= u1 (a01 ; a02 ) or u2 (a1 ; a2 ) 6= u2 (a01 ; a02 ) and suppose
there exists multiple SPNE. We have a …nite game of perfect information so we
can use Zermelo to identify SPNE. Player 2 always has a unique best-response
at each node given the lack of indi¤erence. Backwards induction and lack of
indi¤erence for player 1 means player 1 also has a unique best-response. There-
fore there is a unique SPNE, contradicting the supposition. In order to have
multiple SPNE, at least one player needs to be indi¤erent at some node.

8. Even if each …rm could earn monopoly pro…ts, A would leave the market after
t = 20 and …rm B would leave after t = 25. Consider t = 20 (the last period
that A might be in). If B continued to stay in for the next 5 periods, it gets
P25
t=21 (51 2t) = 25: If A stays in t = 20, then B gets 10:5 20 = 9:5 < 25.
Therefore, regardless of what A does, B will stay in t = 20:If A stays in t = 20,
A gets 105 10 (20) = 95, thus A will stay out from t = 19: Similarly, we can
keep going backwards to …nd a unique SPNE or speed things up as follows:

7
 PhD Micro II: Game Theory

Firm B will clearly exit after …rm A. Firm A will never make monopoly
pro…ts, so we need to identify the last period in which …rm A makes non-
negative pro…ts. This will be the last period in which there is a duopoly. So we
have that

105 10t 0; t 10:

Solving …rm A’s pro…t for t yields tha last period in which he stays. Firm 2 will
then stay in market alone until t = 25: SPNE: Both …rms in for t < 11, only
…rm B in for 11 t 25, no …rm in for t > 25:

9a The father’s optimal choice of x and t maximizes W (U; V )

W (U; V ) = W (u (x) t; v (x) + t)

@W
= W1 + W2 = 0 ! W1 = W2
@t
@W
= W1 u0 (x) + W2 v 0 (x) = 0
@x
Sub one into the other
0 0
W1 u (x) + W1 v (x) = 0; W1 cancels out...
u0 (x) + v 0 (x) = 0

9b. We solve using backwards induction. Hence, we start with the parent’s
decision. What t will he give?

@W=@t = 0 ! W1 = W2

This FOC implicitly de…nes the optimal t: Now allow t to be conditioned on x.

The child thinks about the transfer that he is going to receive given his choice
of x, i.e.,he applies 1-level of backwards induction, thus t = t (x) (this is a best
response function for the parent - he chooses transfer given kid’s e¤ort.)

To solve the kid’s problem, we need to know t0 (x) ; i.e., how the transfer will
change with x: We need to totally di¤erentiate our implicit de…nition of parent’s
choice of t; W1 + W2 = 0 (remember W (u (x) t (x) ; v (x) + t (x)).)

dt dt dt dt
W11 u0 (x) + W12 v 0 (x) + = W21 u0 (x) + W22 v 0 (x) +
dx dx dx dx
dt
dt (W22 W12 ) v 0 (x) +
u0 (x) = dx
(23)
dx (W11 W21 )

8
 PhD Micro II: Game Theory

Kid’s maximises v (x) + t (x), so FOC is:

dt
v 0 (x) + =0
dx

Note that this is the same as the …nal parentheses on the top of the fraction on
the RHS of equation (23). Thus RHS of equation (23) is zero. Thus, from the
LHS we have

dt
u0 (x) = ;
dx

telling us how the parent’s optimal choice of transfer is changing with x. Now
we revisit the kid’s problem. We substitute into his FOC to show that his choice
of x is

u0 (x) + v 0 (x) = 0
,

i.e., the same as in part (a)! (Becker’s Rotten kid theorem: despite being
sel…sh, the kid will still maximise family welfare with the right incentives. The
optimal transfer is set such that the kid internalises the positive externality of
working).

10a. We solve by backwards induction. The father chooses t as to maximize

U (s; t) = u(y t) + V (s; t); for > 0:

We substitute in the daughter’s utility function

U (s; t) = u(y t) + [v1 (z s) v2 (s(1 + r) + t)]; for > 0 and > 0:

The father chooses t, so the F.O.C becomes:

9
 PhD Micro II: Game Theory

Ut (s; t) = u0 (y t) + v20 (s (1 + r) + t) = 0

It is also useful to note

Utt = u00 (y t) + v200 (s (1 + r) + t) < 0

Uts = (1 + r) v200 (s (1 + r) + t) < 0

We allow the father to condition his choice of t on the daughter’s choice of s, so

we have t = t (s). Substitute this into the father’s F.O.C:

Ut (s; t (s)) = u0 (y t (s)) + v20 (s (1 + r) + t (s)) = 0

Totally di¤erentiate w.r.t. s

dUt (s; t (s)) dt

= Uts + Utt =0
ds ds

Since we know that Utt ; Uts < 0, then dt=ds < 0: I.e., the more the daughter
saves, the less of a transfer she will get. Finally, we solve the daughter’s prob-
lem. given t = t (s) (again, the father’s best response function). The daughter
maximises w.r.t s

V (s; t) = v1 (z s) + v2 (s (1 + r) + t (s)) ;

The F.O.C then becomes

dt
v10 (z s) + (1 + r) + v20 (s (1 + r) + t (s)) = 0
ds

This condition implicitly de…nes the daughter’s savings rate.

10b. Recall that in previous part dt=ds < 0 and the daughter’s FOC was

dt
v10 (z s) + (1 + r) + v20 (s (1 + r) + t (s)) = 0
ds

10
 PhD Micro II: Game Theory

Ifthe father has precommitted, then clearly dt=ds = 0 (since the father no
longer conditions his choice on the daughter’s savings rate.) The daughter’s
F.O.C. then becomes

v10 (z s) + ((1 + r)) v20 (s (1 + r) + t ) = 0

But then s is no longer optimal. @V (s ; t) =@s > 0, (because previously dt=ds <
0) therefore, she increases s above s . Pre-commitment allows both her and
father to be better o¤. She saves to little without precommitment because
knows the father will make up the shortfall.

11