You are on page 1of 88

Lecture 1 (March 3, 2014)

1. Definition of a game

Definition 1 : A game is a formal representation of a situation in which each player at-

tempts to have partial influence for taking advantage in a setting of ex-ante preference of
consequences.

• Traders partially influence on a game due to the effects of other players.

2. Elements of a game

• Player

• Strategy

• Payoff

– The payoff of one player is the function of other players’ strategies as well as his strategy.
– The payoff of player i can be denoted by ui (si , s−i ) where si is a particular strategy of
player i and s−i is that of other players.

3. Representation

• normal form

• extensive form

• Player i’s payoff function: πi (s1 , s2 , . . . , sn ) where si ∈ Si .

1
• Formally, we write G = hI, (Si )i∈I , (π)i∈I i.

Player 2
L R

U a, b c, d
Player 1
D e, f g, h

Figure 1. Payoff matrix for two-person game

4. Solution Concept

Definition 4 : A strategy profile s∗ ≡ (s∗1 , s∗2 , . . . s∗n ) is a dominant strategy equilibrium

(DSE) if s∗i is the dominant strategy ∀ i ∈ I.

• If payoff matrix follows Figure 2, DSE is given by (Confess, Confess).

Prisoner 2
Confess Deny

Confess 0, 0 10, −5
Prisoner 1
Deny −5, 10 5, 5

2
Figure 2. Payoff matrix for Prisoners

Example 3: Battles of Sexes

• If payoff matrix follows Figure 3, NE is given by (Ballet, Ballet) and (Soccer, Soccer).

• In this case, players have no incentive to depart form the equilibrium (self-enforcing agree-
ment).

Female
Ballet Soccer

Ballet 5, 10 0, 0
Male
Soccer 0, 0 10, 5

Figure 3. Payoff matrix for the male and female

3
Lecture 2 (March 5, 2014)

1. Types of a game

• Non-cooperative game

– Assuming individual payoff

– Solved by Nash equilibrium
– Default notion of games

• Cooperative game

– Assuming the payoff of whole group

– Solved by Sharply value

2. Elements

• Player

• Strategy

• Payoff

• Normal form (strategic form)

• Extensive form

4. Equilibrium concept

• Dominant strategy equilibrium (DSE): This does not exist in almost every case.

1
• may not exist

• may not be unique

– hard to predict
– Ex) battle of sexes

• may not be plausible

– Refinement is required.

Player 2
H T

H 1, −1 − 1, 1
Player 1
T − 1, 1 1, −1

Figure 4. Payoff matrix of matching pennies

Definition 6. : Given player i’s (finite) pure strategy set Si , a mixed strategy for player i,
σi : Si → [0, 1], assigns to each pure strategy si ∈ Si a probability σi (si ) ≥ 0 that it will be
P
played, where si ∈Si σi (si ) = 1.

• Let

– (Σi )i∈I be mixed strategy set,

– (πi∗ )i∈I be expected payoff, and
– mi be the number of pure strategies

2
of player i.

• In general, for a game,

G = hI, (Si )i∈I , (π)i∈I i

the mixed extension of G is

r = hI, (Σi )i∈I , (πi∗ )i∈I i

where
mi
X
mi
Σi = {σi ∈ R | σij = 1, σij ≥ 0, ∀ j}
j=1
X
πi∗ (σi , . . . , σn ) = πi (S1 , . . . , SI )σ(S1 , . . . , SI ).
S1 ,...,SI

• If we assume independence of each player’s strategy, we have

X
Σi = σ1 (s1 )σ2 (s2 ) · · · σI (SI ).
S1 ,...,SI

Example 5: Matching Pennies (NE does not exist)

• Let

Si = {H, T },
σi : Si → [0, 1],
α = probability that Player 1 chooses H,
β = probability that Player 2 chooses H,
payoff matrix of each player be given as Figure 4.

– α∗ ∈ (0, 1) and β ∗ ∈ (0, 1), and

– Expected payoff when player 1 chooses H = Expected payoff when player 1 chooses T

• For player 1,
β − (1 − β) = −β + (1 − β).

3
• Then β ∗ = 1/2 and symmetrically, α∗ = 1/2.

• Thus mixed strategy Nash equilibrium is given by

 
∗ ∗ 1 1
(α , β ) = , .
2 2

• One interesting feature is that one player’s equilibrium strategies is determined by other
player’s decision.

4
Appendix

• We denote the sequence by {xm }m=∞ m m

m=1 or, simply, by {x } or even x .

Definition A.1. : The sequence {xm } converges to x ∈ RN , written as limm→∞ = x, or

xm → x, if for every ε > 0 there is an integer Mε such that k xm − x k< ε whenever m > Mε .
The point x is then said to be the limit point (or simply the limit) of sequence {xm }.

Definition A.2. : Consider a domain X ∈ RN . A function f : X → R is continuous if for

all x ∈ X and every sequence xm → x (having xm ∈ X for all m), we have f (xm ) → f (x). A
function f : X → Rk is continuous if every coordinate function fk (·) is continuous.

• A function is continuous if, when we take a sequence of points x1 , x2 , . . . converges to x, the

corresponding sequence of function values f (x1 ), f (x2 ), . . . converges to f (x).

• Intuitively, a function fails to be continuous if it displays a “jump” in its value at some point
x.

Definition A.3. : Fix a set X ∈ RN .

(i) We say that a set A ⊂ X is open (relative to X) if for every x ∈ A there is an ε > 0 such
that k x′ − x k< ε and x′ ∈ X implies x′ ∈ A.

(ii) A set A ⊂ X is closed (relative to X) if its complement X \ A is open (relative to X).

5
• Given a point x ∈ RN , a set B = {x′ ∈ RN :k x′ − x k< ε} for some scalar ε > 0 is called an
open ball around x.

• A set A ⊂ X is open (relative to X) if, for every point x in A, there is an open ball around x
all of whose elements (in X) are elements of A.

Theorem A.1. : Fix a set X ⊂ RN . In what follows, all the open and closed sets are relative
to X.

(i) The union of any number, finite or infinite, of open sets is open. The intersection of a
finite number of open sets is open.

(ii) The intersection of any number, finite or infinite, of closed sets is closed. The union of a
finite number of closed sets is closed.

(iii) A set A ⊂ X is closed if and only if for every sequence xm → x ∈ X, with xm ∈ A for all
m, we have x ∈ A.

• Property (iii) of Theorem A.1 is noteworthy because it gives us a direct way to characterize a
closed set:

– a set A is closed if and only if the limit point of any sequence whose members are all
elements of A is itself an elements of A.

• Points (in X) that are the limits of sequences whose members are all elements of the set A
are known as the limit points of A.

– Property (iii) says that a set A is closed if and only if it contains all of its limit points.

Definition A.4. :

(ii) The set A ⊂ RN is compact if it is bounded and closed relative to RN .

6
• Given a function f : X → Rk , the image of a set A ⊂ X under f (·) is the set f (A) = {y ∈
RK : y = f (x) for some x ∈ A}.

Theorem A.2. : Suppose that f : X → Rk is continuous function defined on a nonempty set

X ⊂ RN .

(i) The image of a compact set under f (·) is compact: That is, if A ⊂ X is compact, then
f (A) = {y ∈ RK : y = f (x) for some x ∈ A} is a compact subset of RK .

(ii) Suppose that K = 1 and X is compact. Then f (·) has a maximizer: That is, there is
x ∈ X such that f (x) ≥ f (x′ ) for every x′ ∈ X.

Definition A.5. : Given a set A ⊂ RN , a correspondence f : A → RK is the rule that assigns

a set f (x) ⊂ RK to every x ∈ A.

• For every x ∈ A, f (x) is composed of precisely one element, then f (·) can be viewed as a
function in the usual sense.

Definition A.6. : Given A ⊂ RN and Y ⊂ RK , the graph of the correspondence f : A → Y is

the set {(x, y) ∈ A × Y : y ∈ f (x)}.

Definition A.7. : Given A ⊂ RN and the closed set Y ⊂ RK , the correspondence f : A → Y

has a closed graph if for any two sequences xm → x ∈ A and y m → y, with xm ∈ A and
y m ∈ f (xm ) for every m, we have y ∈ f (x).

7
Definition A.8. : Given A ⊂ RN and the closed set Y ⊂ RK , the correspondence f : A → Y
is upper hemicontinuous (uhc) if it has a closed graph and the images of compact sets are
bounded, that is, for every compact set B ⊂ A the set f (B) = {y ∈ Y : y ∈ f (x) for some x ∈
B} is bounded.

Theorem A.3. : Given A ⊂ RN and the closed set Y ⊂ RK , suppose that f : A → Y is a single-
valued correspondence (so that it is, in fact, a function). Then f (·) is an upper hemicontinuous
correspondence if and only if it is continuous as a function.

Definition A.9. : Given A ⊂ RN and a compact set Y ⊂ RK , the correspondence f : A → Y

is lower hemicontinuous (lhc) if for every sequence xm → x ∈ A with xm ∈ A for all m, and
every y ∈ f (x), we can find a sequence y m → y and an integer M such that y m ∈ f (xm ) for
m > M.

• Roughly speaking, upper hemicontinuity is compatible with “explosions”, while lower hemi-
continuity is compatible with “implosions”.

• When correspondence is both upper and lower hemicontinuous, we say that it is continuous.

Theorem A.4. : Brower’s Fixed Point Theorem Suppose A ⊂ RN is nonempty, compact,

convex set, and that f : A → A is a continuous function from from A into itself. Then f (·) has
a fixed point; that is, there is an x ∈ A such that x = f (x).

Theorem A.5. : Kakutani’s Fixed Point Theorem Suppose that A ⊂ RN is a nonempty,

compact, convex set, and that f : A → A is an upper hemicontinuous correspondence from A
onto itself with the property that the set f (x) ⊂ A is nonempty and convex for every x ∈ A.
Then f (·) has a fixed point; that is, there is an x ∈ A such that x ∈ f (x).

8
Theorem A.5. : Tarsky’s Fixed Point Theorem Suppose that f : [0, 1]N → [0, 1]N is a
nondecreasing function, that is, f (x′ ) ≥ f (x) whenever x′ ≥ x. Then f (·) has fixed point; that
is, there is an x ∈ A such that x = f (x).

• Tarsky’s theorem differs from Brouwer’s in three respects.

– The base set is not any compact, convex set, but rather special one - an N -product of
intervals.
– The function is required to be nondecreasing.
– The function is not required to be continuous.

– I consumers (indexed by i = 1, . . . , I),

– J firms (indexed by i = 1, . . . , J), and
– L goods (indexed by ℓ = 1, . . . , L).

• Consumer i’s preferences over consumption bundles xi = (x1i , . . . , xLi ) in his consumption
set Xi ⊂ RL are represented by the utility function ui (·).

• The total amount of each good ℓ = 1, . . . , L initially available in economy, called the total
endowment of good ℓ, is denoted by wℓ ≥ 0 for ℓ = 1, . . . , L.

• Each firm j has available to it the production possibilities summarized by the production set
Yj ⊂ R L .

• An element of Yj is a production vector yj = (yj1 , . . . , yLj ) ∈ RL .

• If (y1 , . . . , yJ ) ∈ RLJ are the production vectors of the J firms, the total (net) amount of good
P
ℓ available to the economy is wℓ + j yℓj .

9
Definition A.10. :

(i) An economic allocation (x1 , . . . , xI , y1 , . . . , yJ ) is a specification of a consumption vector

xi ∈ Xi for each consumer i = 1, . . . , I and a production vector yj ∈ Yj for each firm
j = 1, . . . , J.

(ii) The allocation is feasible if

I
X J
X
xℓi ≤ wℓ + yℓj for ℓ = 1, . . . , L.
i=1 i=1

Definition A.11. : A feasible allocation (x1 , . . . , xI , y1 , . . . , yJ ) is Pareto optimal (or Pareto

efficiency) if there is no other feasible allocation such that ui (x′i ) ≥ ui (xi ) for all i = 1, . . . , I
and ui (x′i ) > ui (xi ) for some i.

• Pareto optimal means that there is no alternative way to organize the production and distri-
bution of goods that makes some consumer better off without making some other consumer
worse off.

• The criterion of Pareto optimality does not insure that an allocation is in any sense equitable.

• Pareto optimality serves as an important minimal test for the desirability of an allocation; it
does, at the very least, say that there is no waste in the allocation of in society.

References

Mas-Collel, A., M. Whinston, and J. Green (1995): “Microeconomic Theory”.

10
Lecture 3 (March 10, 2014)

1. Continuity of Correspondence

• Notations

– A ⊂ Rm
– B ⊂ Rk
– xq : a sequence on A such that xq → x0 ∈ A
– y q : a sequence on B such that y q → y 0
– ϕ : A → B: a correspondence from A to B

Definition 7 :

(i) For any two sequences xq and y q with y q ∈ f (xq ) for every q, if we have y 0 ∈ ϕ(x0 ),
correspondence ϕ is upper hemicontinuos at x0 .

(ii) For every sequence xq and every y 0 ∈ ϕ(xq ), if there exists y q ∈ ϕ(xq ), correspondence ϕ
is lower hemicontinuos at x0 .

(iii) Correspondence ϕ is continuous at x0 if ϕ is both upper hemicontinuous and lower

hemicontinuous.

Theorem 1 : Graph
Φ ≡ {(x, y)|x ∈ A, y ∈ ϕ(x)}

Theorem 2 : (Kakutani’s fixed point theorem) Let A ⊂ Rm be nonempty, compact an convex

set and the correspondence ϕ : A → A be nonempty, convex-valued and upper hemicontinu-
ous. Then there exists x∗ ∈ A such that x∗ ∈ ϕ(x∗ ).

1
Theorem 3 : (Debreu) Let G = hI, (si )i∈I , (πi )i∈I i. Suppose the following hold.

Then there exists a pure-strategy Nash equilibrium.

P ROOF : Let the best response correspondence BRi (s−i ) = {x ∈ Si |πi (x, s−i ) ≥ πi (y, s−i ), ∀ y ∈ Si }.
1. Show ‘BRi (s−i ) is nonempty.’
Since Si is compact and πi is continuous in s, there exists maximum and minimum values of πi by
Weierstrass theorem.

2. Show ‘BRi (s−i ) is convex-valued.’

Let x1 , x2 ∈ BRi (s−i ) with x1 6= x2 . Then x1 , x2 ∈ Si and πi (x1 , s−i ) = πi (x2 , s−i ). Let x3 =
tx1 + (1 − t)x2 for some t ∈ (0, 1). The qausiconcavity of π implies

πi (x3 , s−i ) ≥ πi (x1 , s−i ) = πi (x2 , s−i ).

However, since x1 and x2 are best response for s−1 , πi (x3 , s−i ) cannot be greater than πi (x1 , s−i )
or πi (x2 , s−i ). Thus we have

3. Show ‘BRi (s−i ) is upper hemicontinuous.’

Let sq−i → s0−i on S−i , sqi ∈ BRi (sq−i ), and sqi → s0i . We want to show that s0i ∈ BRi (s0−i ). We know
that
π(sqi , sq−1 ) ≥ πi (y, sq−1 ), ∀ y ∈ si .
By the continuity of π,
π(s0i , s0−1 ) ≥ πi (y, s0−1 ), ∀ y ∈ si .
Thus s0i ∈ BRi (s0−i ).

4. Construct mapping BR such that

BR : S1 × S2 . . . × Sn → S1 × S2 · · · × Sn .

There exists
BR(S1 , . . . , Sn ) = BRi (S−1 ) × BRi (S−2 ) · · · × BRi (S−n ).
Then correspondence BR is nonempty, convexed and upper hemicontinuous.

2
Appendix

(i) For every real number a, if

Ca+ ≡ {x ∈ U : f (x) ≥ a}

(ii) For every real number a, if

Ca− ≡ {x ∈ U : f (x) ≤ a}

is convex set, f is quasiconvex.

Theorem A.2. : Let f be a function defined on a convex set U in Rn . Them the following
statements are equivalent to each other:

f (tx + (1 − t)y) ≥ min{f (x), f (y)}.

Definition A.2. : Let {an }∞ n=1 be a sequence. Let f be a strictly increasing function from P
into P. The sequence {af (n) }∞ ∞
n=1 is called a subsequence of the sequence {an }n=1 .

Theorem A.2. : (Bolzano-Weierstrass Theorem) Every bounded real sequence has a conver-
gent subsequence.

3
P ROOF : Let {an } be a bounded sequence. Then there is a closed interval [c, d] such that an ∈ [c, d]
for every positive integer n.
Consider the two closed intervals
   
c+d c+d
c, , ,d .
2 2

One of these intervals must contain an for infinitely many positive integers n. We denote this
interval by [c1 , d1 ].
We repeat this process with the interval [c1 , d1 ]. One of the intervals
   
c1 + d1 c1 + d1
c1 , , , d1 .
2 2

must contain an for infinitely many positive integers n. We denote this interval by [c1 , d2 ]. Contin-
uing process, we obtain a sequence
[c1 , d1 ], [c2 , d2 ], . . .

of closed intervals such that

[c1 , d1 ] ⊃ [c2 , d2 ] ⊃ [c3 , d3 ] ⊃ · · ·

and
d−c
dk − ck = , k = 1, 2, . . . (A.1)
2k
and each interval [ck , dk ] contains an for infinitely many positive integer n.
Choose a positive integer n1 such that an1 ∈ [c1 , d1 ]. Since [c2 , d2 ] contains an for infinitely many
positive integers n, there exists a positive integer n2 > n1 such that an2 ∈ [c2 , d2 ]. Continuing this
process we obtain elements an1 , an2 , . . . satisfying

ank ∈ [ck , dk ], k = 1, 2, . . . (A.2)

and
n1 < n2 < n3 < · · · .

Therefore, {ank }∞
k=1 is a subsequence of {an }. We will show that {ank } converges.
The sequence {ck } is monotone and bounded so is convergent. Let L = limk→∞ ck . Similarly,
the sequence {dk } converges to some number M. Using equation (A.1), we have

d−c
M − L = lim dk − lim ck = lim (dk − ck ) = lim = 0.
k→∞ k→∞ k→∞ k→∞ 2k

ck ≤ ank ≤ dk .

4
References

Fundenberg, D. and J. Tirole (1991): “Game Theory”.

Johnsonbaugh, R. and W. Pfaffenberger (2002): “Foundations of Mathematical Analysis”.
Mas-Collel, A., M. Whinston, and J. Green (1995): “Microeconomic Theory”.
Simson, C. and L. Blume (1994): “Mathematics for Economists”.

5
Lecture 4 (March 12, 2014)

Theorem 3 : (Debreu) Let G = hI, (si )i∈I , (πi )i∈I i. Suppose the following hold.

Then there exists a pure-strategy Nash equilibrium.

P ROOF : Let the best response correspondence BRi (s−i ) = {x ∈ Si |πi (x, s−i ) ≥ πi (y, s−i ), ∀ y ∈ Si }.
1. Show ‘BRi (s−i ) is nonempty.’
Since Si is compact and πi is continuous in s, there exists maximum and minimum values of πi by
Weierstrass theorem.

2. Show ‘BRi (s−i ) is convex-valued.’

Let x1 , x2 ∈ BRi (s−i ) with x1 6= x2 . Then x1 , x2 ∈ Si and πi (x1 , s−i ) = πi (x2 , s−i ). Let x3 =
tx1 + (1 − t)x2 for some t ∈ (0, 1). The qausiconcavity of π implies

πi (x3 , s−i ) ≥ πi (x1 , s−i ) = πi (x2 , s−i ).

However, since x1 and x2 are best response for s−1 , πi (x3 , s−i ) cannot be greater than πi (x1 , s−i )
or πi (x2 , s−i ). Thus we have

3. Show ‘BRi (s−i ) is upper hemicontinuous.’

Let sq−i → s0−i on S−i , sqi ∈ BRi (sq−i ), and sqi → s0i . We want to show that s0i ∈ BRi (s0−i ). We know
that
π(sqi , sq−1 ) ≥ πi (y, sq−1 ), ∀ y ∈ si .

By the continuity of π,
π(s0i , s0−1 ) ≥ πi (y, s0−1 ), ∀ y ∈ si .

4. Construct mapping BR such that

BR : S1 × S2 . . . × Sn → S1 × S2 · · · × Sn .

1
There exists
BR(S1 , . . . , Sn ) = BRi (S−1 ) × BRi (S−2 ) · · · × BRi (S−n ).
Then correspondence BR is nonempty, convexed and upper hemicontinuous.

(i) Nonempty
Since BRi is nonempty, we can choose element such that

Take s′ = (s′1 , . . . , s′n ). Then s′ ∈ s′1 ∈ BR(s).

(ii) Convex-valued
Since, for every s′i and s′′i , ti s′i + (1 − ti )s′′i ∈ BRi , ∀ i and ∀ ti ∈ (0, 1), we have every s′ ∈ BR
and s′′ ∈ BR. Thus



because

(iv) There exists a Nash equilibrium

By Kakutani’s fixed point theorem, there exists

s∗ ∈ S1 × · · · Sn s.t. s∗ ∈ BR(s∗ ).

Since s∗i ∈ BRi (s∗−i ), s∗ is a Nash equilibrium.

Theorem 4 : (Nash) There exists at least one mixed strategy Nash equilibrium for any finite
game.

P ROOF : Let G = hI, (Si )i∈I , (πi )i∈O i. Then we have the mixed extension of G = hI, (Σi )i∈I , (πi )∗i∈I i
where
mi
X
mi
Σ = {σi ∈ R | σij = 1, ∀ i σij = 1},
j=1
X
πi∗ (σ1 , ·, σn ) = σ(s1 , . . . , sn )πi (s1 , . . . , sn ).
σ1 ,...,σn

2
Then Σi is nonempty, compact and convex, and πi is linear in σi and that πi is linear in σi implies
that πi∗ is continuous and quasi concave in σi . Therefore there exists a mixed Nash equilibrium.

Theorem 5 : Let σ ∗ be a mixed strategy equilibrium (mse) of Γ (mixed extension of G).

∗ ) < π (σ ∗ ), then σ ∗ (s0 ) = 0.
(i) If πi (s0i , σ−1 i i i

∗ ) = π (σ ∗ ).
i

P ROOF : (1) Suppose σi∗ (σi0 ) > 0. Then

X
πi (σ ∗ ) = πi (s0i , σ−i

)+ ∗
πi (si , σ−i )σi∗ (si )
si 6=s0i
X
< πi (σ ∗ )σi∗ (s0i ) + ∗
πi (si , σ−i )σi∗ (si )
si 6=s0i
X
≤ πi (σ ∗ )σi∗ (s0i ) + πi (σ ∗ )σi∗ (si )
X
= πi (σ ∗ )σi∗ (si )
si ∈Si
= πi (σ ∗ ),

which contradicts to assumption.

∗ ) > π (σ ∗ ). Then σ ∗ is note an equilibrium. Suppose π (s0 , σ ∗ ) < π (σ ∗ ).
(2) Suppose πi (s0i , σ−i i i i −i i
∗ 0 0 ∗ ∗
Then σi (si ) = 0 by (1). Thus πi (si , σ−i ) = πi (σ ).

with strict inequality for at least one s−i .

3
• In a single person strategy, weakly dominate strategy dose not used in equilibrium.

• In a multi person strategy (game theory), weakly dominate strategy would be used in equi-
librium.

• Strictly dominated strategy is eliminated in equilibrium.

Remark 1 : Strictly dominated strategy is never used in any mixed strategy equilibrium.

Example 6: Iterative Elimination

t1 t2 t3 t4

s1 1, 1 5, 2 0, 0 5, 1 2

s2 2, 5 6, 6 2, 8 4, 3 4

s3 0, 0 8, 2 4, 4 1, 1

s4 1, 5 3, 4 1, 1 3, 3 2

3 3 1
Figure 5. Iterative elimination in two person game

Definition 9 : The set of strategy for each player left after iterated elimination of strictly
dominated strategy is called sophisticated equilibrium and if the set is singleton, the game is
called a dominance solvable game.

Example 7: Strictly Dominance by Mixed Strategy

4
Player 2
t1 t2 t3

s1 1 3 5

Player 1 s2 5 3 1

s3 2 2 2

5
References

Fundenberg, D. and J. Tirole (1991): “Game Theory”.

Mas-Collel, A., M. Whinston, and J. Green (1995): “Microeconomic Theory”.

6
Lecture 5 (March 17, 2014)

Definition 9 : Consider G = hI, (Si )i∈I , (π)i∈I i. Let

Si0 = Si
Σ0i = Σi
Sik+1 = {si ∈ Sik | ∄ σi ∈ Σki s.t. πi (σi , s−i ) > πi (si , s−i ), ∀ s−i ∈ S−i
k
}
Σk+1
i = {σi ∈ Σi |σi (si ) > 0, only if si ∈ Sik+1 }

(i) If Sit = Sit+1 for all i ∈ I and for some t, the set S t = S1t × . . . Snt is the set of sophisticated
equilibrium

done.

Example 8: Elimination of Weakly Dominated Strategy

L R

U 1, 1 1, 1

D 1, 1 0, 0 1

1
L R

U 1, 1 1, 1

D 1, 1 0, 0

1
Figure 5. Elimination of weakly dominated strategy by changing order 1

L M R

U 10, 2 2, 2 -1, 0

D 10, 4 -1, 0 3, 1

1 1

L M R

U 10, 2 2, 2 -1, 0 2

D 10, 4 -1, 0 3, 1

1
Figure 6. Elimination of weakly dominated strategy by changing order 1

• Figure 5 and Figure 6 show that remained strategy would be difference when weakly domi-
nated strategies are eliminated spontaneously and when eliminated sequentially.

Definition 10 : Information ω is common knowledge among payers in I if for all sequence i,

i−i from I, the statement “i1 knows that i2 knows that . . . that ik knows ω for all k” is true.

• The statement “I saw that Chulsu saw the accident“ is not the case of common knowledge
since Chulsu does not know that I know what Chulsu knows.

2
Implicit Assumption of Game Theory : Platers have common knowledge about

Example 9: Strictly Dominance by Mixed Strategy

Player 2
a b c d

A 4, 10 3, 0 2, 6 2, 6
Player 1
B 0, 0 1, 10 10, 3 3, 6

Figure 7. Rationalizability in two person game

• Is A rationalizable?

– A is the best response to a and vice versa.

• Is B rationalizable?

– B is the best response to c and c is never a best response to any strategy of I.

– B is the best response to d and d is the best response to mixed strategy of A and B where
σ1 (A) = σ1 (A) = 1/2.
X A is the best response to a.
X B is the best response to d again.
– Is b rationalizable?
X b is the best response to B and B is the best response to d and d is best response to
mixed strategy of A and B where σ1 (A) = σ1 (A) = 1/2.

3
Sophisticated Equilibrium

Rationalizability

NE

Figure 8. Relation among Nash equilibrium, rationalizability and sophisticated equilibrium

Definition 11 :

(i) si ∈ Si is 1-rationalizable strategy for player i if si is the best response to some probability
distribution over strategies for other players.

(ii) si ∈ Si is a k-rationalizable strategy for player i if si is the best response to some proba-
bility distribution over (k − 1) rationalizable strategy for other players.

(iii) si ∈ Si is a rationalizable strategy if it ks k-rationalizable for all k.

Theorem 5 :

(i) In two-person game, the set of rationalizable strategy equals to the set of sophisticated
equilibrium.

(ii) If |I| ≥ 3, the set of rationalizable strategy is the subset of sophisticated equilibrium.

Home Assignment 1

1.“Divide-the dollar” game. There is \$1. Player 1 and Player 2 bid x and y, respectively, and

4
their payoff function is given by

(x, y) if x + y ≥ 1
g(x, y) =
(0, 0) if x + y > 1.

2. The payoff matrix of Player 1 and Player 2 is given by

Player 2
L R

T 1, 1 1, 0

Player 1 M 3, 0 0, 3

B 0, 1 4, 0

5
References

Fundenberg, D. and J. Tirole (1991): “Game Theory”.

Mas-Collel, A., M. Whinston, and J. Green (1995): “Microeconomic Theory”.

6
Lecture 6 (March 19, 2014)

0 1 x

NE = {x, y|x + y = 1, orx ≥ 1 and y ≥ 1}.

(2) Strategy T of Player 2 is strictly dominated by mixed strategy σ1 = (0, 1/2, 1/2) of Player 1.
Thus we have

σ1∗ = {0, 1/4, 3/4}

σ2∗ = {4/7, 3/7}.

• Henceforth, ‘dominated’ means ‘strictly dominated.’

1
• Selten (1975) points that (0, 0) is not sensible and suggests trembling hand perfect equilibrium.

Player 2
L R

T 1, 1 0, 0
Player 1
M 0, 0 0, 0

Figure 11. Implausibility in Nash equilibrium

Definition 12 : For any normal form game G(I) = hI, (Σi )i∈I , (πi )i∈I i, we can define per-
turbed game G(ε) = hI, (Σi )i∈I (ε), (πi )i∈I i where

Σi (ε) = {σi ∈ Σi |σi (si ) ≥ εi (si ) ∀ σi ∈ Si and εi (si ) > 0}.

Definition 13 : A Nash equilibrium (σ1 , σ2 , . . . , σn ) of game G(I) = hI, (Σi )i∈I , (πi )i∈I i is
trembling-hand perfect equilibrium if there is some sequence of perturbed games such that

lim (σ1ε , σ2ε , . . . , σnε ) = (σ1 , σ2 , . . . , σn ).

ε→0

• Note if
lim πi (σiε , σ−i
ε ε
) ≥ πi (σi′ , σ−i )
ε→0
then
πi (σi , σ−i ) ≥ πi (σi′ , σ−i )

by continuity of πi .

Example 10: Perfect Equilibrium

2
Player 2
L R A2

T 1, 1 0, 0 -1, -2

Player 1 M 0, 0 0, 0 0, -2

• Let probability that Player I choose (L, R, A2 ) by mistake be (ε1 , ε2 , ε3 ).

• Let probability that Player II choose (U, D, A1 ) by mistake be (ε′1 , ε′2 , ε′3 ).

• If we exclude from payoff matrix in Figure 12, strategies A1 and A2 , (U, L) is perfect equilib-
rium.

Definition 14 : (Myerson) A perfect equilibrium is said to be proper equilibrium if

ε ε
πi (si , σ−i ) < πi (s′i , σ−i )

then σiε (si ) < εσiε (s′i ).

NE

Perfect Equilibrium

Proper Equilibrium

3
Figure 13. Relation among Nash equilibrium, perfect equilibrium and proper equilibrium

4
References

Fundenberg, D. and J. Tirole (1991): “Game Theory”.

Mas-Collel, A., M. Whinston, and J. Green (1995): “Microeconomic Theory”.

5
Lecture 7 (March 24, 2014)

• Consider the game in Figure 14.

• Suppose Player 2 chooses L with probability q and R with probability 1 − q and then we have

u(T ) = 1
u(M ) = 3q
u(B) = 4(1 − q),

4

 B if q < 7

BR1 (q) = 4
B or T if q = 7

4
T if q > 7,

• It is clear that strategy T is eliminated from Player 1’s choice option.

Player 2
L R

T 1, 1 1, 0

Player 1 M 3, 0 0, 3

B 0, 1 4, 0

1
u(·)

u(M )

u(T )
u(B)
0 4 q
7

BR1

BR2
p

Figure 16. Best responses for Player 1 and Player 2

2
Lecture 8 (March 26, 2014)

Extensive Form Representation

• Major difference between normal form game and extensive game is that the former can im-
pose commitment while the latter cannot.

– normal form: matrix

– extensive form: tree

• Tree is

– a collection of nodes and arcs (links, lines)

– connected figure with no cycle

Figure 17. No tree

• Terminology

– Starting node
– B follows A
– Terminal (node): no following node

• Description

(1) Player: nodes

(2) Strategy: related to arc
(3) Payoff: Noting at terminal node
(4) information: using information set

1
(−1, −1)
fight

Monopolist
enter

not enter
(0, 2)

Example 12: Entry Deterrence II

(−1, −1)
fight
Monopolist

seriously
information set

enter
accommodate (1, 1)
(−1, 0)
fight
casually
enter
Entrant

accommodate (1, 1)

not enter
(0, 2)

• perfect information game: Every information set consists of singleton.

• imperfect information game: There exist information sets which does not consist of singleton.

2
(1, −1)
H
II
H
T (−1, 1)
I
(−1, 1)
H
T

T (1, −1)

– Action: defined at each node

– Strategy: Contingent plan for each information set

Si : Hi → Ai
|{z} |{z}
information set action set

(1, −1)
H
II
H
T (−1, 1)
I
(−1, 1)
H
T

T (1, −1)

• Player I

3
– Strategy: H, T
– Action: H, T

• Player II

– Strategy: HH, HT, T H, T T

– Action: H, T

Player II
HH HT TH TT

H 1, −1 1, −1 1, −1 − 1, 1
Player I
T − 1, 1 1, −1 − 1, 1 1, −1

• Sometimes, Nash equilibrium is not plausible in extensive form game.

– In Figure 23, Nash equilibriums are given by (enter, accommodate) and (not, fight).
– The former is not plausible since this depends on Monopolist’ empty threat.

Monopolist
fight accommodate

enter − 1, −1 1, 1
Entrant
not 0, 2 0, 2

• Selten suggests subgame perfect equilibrium (NE + backward induction).

4
Lecture 9 (March 31, 2014)

Definition 15 : Player i’s information set is defined by partition of Hi such that

(i) all nodes in the same hji ∈ Hi must have the same number of immediate successors.

(ii) in the same hji , no node can follow another.

Definition 16 : A subgame of an extensive form game is a subset of the game having the
following properties:

(ii) It includes all the following nodes.

1. It does not cut nay information set, i.e., if a node is contained in a subgame, all other
nodes in the information set containing the node must be contained in the subgame.

Definition 17 : A Nash equilibrium is subgame perfect if for all subgames, the restriction of
the strategies to the subgame constitutes a Nash equilibrium in the subgame.

Example 15: Worker Vs. Employer Game

• In Figure 24, worker’s strategy set is given by {train, not train} and employer’s is given by
{hh, hn, nh, nn}.

• We see from Figure 25 that Nash equilibriums are given by (not train, hh), (train, hn) and
(not train, nh).

• However, the last two depend on the empty threat of employer.

• In this game, subgame perfect Nash equilibrium (SPNE) is given by (not train, hh).

1
(5, 8)
hire
Employer

train
not (−1, 0)
Worker
(6, 4)
hire
not

not (0, 0)

Figure 24. Extensive form worker vs. employer game

Worker
hh hn nh nn

train 5, 8 5, 8 − 1, 0 − 1, 0
Employer
not train 6, 4 0, 0 6, 4 0, 0

Mixed Strategy Vs. Behavioral Strategy in Extensive Form Games

Definition 18 : Behavioral strategy for player i is a probability distribution over the action
available on each information set of player i.

Example 16: Mixed Strategy and Behavioral Strategy

• Strategy sets of Player I and Player II are given by {H, T } and {HH, HT, T H, T T }, respec-
tively.

• Suppose two mixed strategies of Player II:

 
1 1 1 1
m1 = , , ,
4 4 4 4
 
1 1
m2 = , 0, 0, .
2 2

2
• Suppose Player II chooses m1 .
1 1 1
Pα (H) = P (HH) + P (HT ) = + =
4 4 2
1 1 1
Pα (T ) = P (T H) + P (T T ) = + =
4 4 2
1 1 1
Pβ (H) = P (HH) + P (T H) = + =
4 4 2
1 1 1
Pβ (H) = P (HT ) + P (T T ) = + =
4 4 2

 
1 1
b1 (α) = ,
2 2
 
1 1
b1 (β) = ,
2 2

 
1 1
b2 (α) = ,
2 2
 
1 1
b2 (β) = ,
2 2

(1, −1)
H
II
H α
T (−1, 1)
I
(−1, 1)
H
T β

II
T (1, −1)

Figure 26. Extensive form matching pennies II

• Behavioral strategy assumes independence between strategies while mixed strategy permits
correlation between them.

3
• Thus behavioral strategy is contained in mixed strategy.

Mixed Strategy

Behavioral Strategy

Figure 27. Relation between behavioral strategy and mixed strategy

Definition 19 : Player i has perfect recall if and only if for every h1i , h2i ∈ Hi if a node x ∈ h2i
follows a choice z at h′i then every node y ∈ h2i follows the same choice z.

Theorem 6 : (Kuhn) There is no loss of generality in restriction attention to behavioral

strategy in game of perfect recall.

4
Lecture 10 (April 2, 2014)

• In Figure 27, Player I cannot remember his choice at information set h11 at the time he reaches
information set h21 .

1 1
u
h21 2 4

T 1
d 0 4
II

u 0 0
U B
Player II d 0 0
h11
I
1
u 0 4

D T 1 1
d 2 4

u 0 0
B
Player I 0 0
d
m1 b 1

 
1 1
m1 = , 0, 0, .
2 2

– Probability Player I chooses U and chooses D are given by

1 1
P (U u, U d) = +0=
2 2
1 1
P (Du, Dd) = 0 + = ,
2 2
respectively.

1
• Suppose Player I attempts to make a choice at h11 .

– Probability Player I chooses u and chooses d are given by

1 1
P (U u, Du) = +0=
2 2
1 1
P (U d, Dd) = 0 + = ,
2 2
respectively.

• Thus Player I’s behavioral strategy at h11 and h21 are given by
 
1 1 1
b1 (h1 ) = ,
2 2
 
2 1 1
b1 (h1 ) = , ,
2 2

respectively.

• As shown in Figure 27, probability distribution generated by mixed strategy m1 and this
generated by behavioral strategy b1 .

– Mixed strategy m1 permits correlation between choices U and D at h11 and that u and d
at h21 , while behavioral strategy b1 does not.

• Recall Example 15 in Figure 28.

– If worker chooses ‘not train’ and employer chooses ‘hire’, we call this on-the-equilibrium
path (depicted with blue line).
– Other pathes are called off-the equilibrium path.

• In SPNE, we should specify choices both cases when they lies on-the-equilibrium path and
when off-the equilibrium path.

2
(5, 8)
hire
Employer

train
not (−1, 0)
Worker
(6, 4)
hire
not

not (0, 0)

• Player I’s strategy set is given by {U u, U d, Du, Dd}.

• Since Du and Dd have exactly the same payoff regardless of Player II’s choice, we can obtain
reduced normal form game as in Figure 29.

• Reduced normal form game does not affect Nash equilibrium but it could affect trembling
hand perfect equilibrium.

u
I

II d
U
r
I

3
Player II
ℓ r

Uu ·, · ·, ·

Ud ·, · ·, ·
Player I
Du ·, · ·, ·

Dd ·, · ·, ·

Figure 29. An extensive form game and its reduced normal form

Example 17: Entry Deterrence II

(−1, −1)
fight
Monopolist

seriously
enter
accommodate (1, 1)
(−1, 0)
fight
casually
enter
Entrant

accommodate (1, 1)

not enter
(0, 2)

4
Monopolist
accommodate fight

Entrant Casually enter 1, 1 −1, 0

Not enter 0, 2 0, 2

Figure 30. Entry deterrence II

• Entrant has strategy set {SE, CE, N E} and monopolist has strategy set {a, f }.

• However, (N E, f ) is not reasonable since monopolist would choose a if entrant chooses SE

or CE.

Definition 20 : A pair of (behavior) strategy profile and beliefs ((σi∗ )i∈I , µ) constitutes a weak
perfect Bayesian equilibrium if and only if

(i) (sequential rationality) at each information set h, player’s actions is optimal given strat-
egy of others and beliefs,

(ii) (consistency) beliefs are computed from strategy by using Bayesian law whenenver pos-
sible, i.e.,

pσ (t)
µ(t) = σ∗
p (h)

where pσ (h) > 0 is the probability that a path reaches information set h in equilibrium
∗ ∗
and pσ (t) is the probability a path reaches t in h. If pσ (h) = 0, we place a arbitrary
belief.

• See Figure 31.

5
– Claim 1: Nash equilibrium (N E, f ) is not a weak PBE.

P ROOF : Given Information set 1, the expected payoff of monopolist is 1 if he chooses

accommodate and −µ if he chooses fight. Since 1 > −µ for every µ ∈ [0, 1], choice of
fight contradicts sequential rationality.
– Claim 2: Nash equilibrium (CE, a) is a weak PBE. P ROOF : We set µ = 0. Then (CE, a)

is a weak PBE.
– Claim 3: Nash equilibrium (SE, a) is a weak PBE. P ROOF : We set µ = 1. Then (SE, a)

is a weak PBE.

(−1, −1)
fight
Monopolist
µ
seriously
Information set 1

enter
accommodate (1, 1)
(−1, 0)
fight
casually
enter
Entrant
1−µ

accommodate (1, 1)

not enter
(0, 2)

Figure 31. Entry deterrence II and Bayesian update

Home Assignment 2

6
(1, −2)

II

r
t (−3, −1)
(3, 1)

m
I

r
(−2, −1)
b

(0, 2)

Figure 32. Home assignment 2

7
Lecture 11 (April 7, 2014)

(1, −2)

II
µ
r
t (−3, −1)
(3, 1)

m
I
1−µ
r
(−2, −1)
b

(0, 2)

• NEs are given by (m, ℓ) and (b, r).

Player 2
ℓ r

t 1, −2 −3, −1

Player 1 m 3, 1 −2, −1

b 0, 2 0, 2

1
3. Find wPBE.

(i) Player II’s optimal choice

• Expected payoff of Player II is given by

−2µ + (1 − µ) = 1 − 3µ if Player I chooses ℓ,
u2 (·) =
−µ − (1 + µ) = −1 if Player I chooses r

2

 ℓ if µ < 3

2
r if µ > 3

ℓ or r 2
if µ =

3

(ii) Player I’s optimal choice

• Suppose µ < 2/3 and then Player II will choose ℓ.
– Payoff function of Player I can be written as

 1 for t,

u2 (·) = 3 for m,

0 for b.

– Thus  
2
m, ℓ, µ <
3
can be a candidate of wPBE.
– Since Player I must choose m, wPBE is given by

(m, ℓ, µ = 0) ,

which satisfies consistency constraint.

• Suppose µ > 2/3 and then Player II will choose r.
– Payoff function of Player I can be written as

−3 for t,

u2 (·) = −2 for m,

0

for b.

– Thus  
2
b, r, µ >
3
can be a candidate of wPBE.

2
– Since arbitrary belief is available on off-the-equilibrium path, Player I’s another
wPBE is given by  
2
b, f, µ > ,
3
which satisfies consistency constraint.
• Suppose µ = 2/3 and then Player II will choose some mixed strategy.
– Let p denote the probability that Player II chooses ℓ.
– Payoff function of Player I can be written as

 4p − 3 for t,

u2 (·) = 5p − 2 for m,

0

for b.

– Since 5p − 2 > 4p − 3 and thus Player I never choose t.

– Hence wPBE is given by
 
2 2
b, 0 < p ≤ , µ = .
5 3

Example 18: Entry Deterrence III

• Claim: (ne, se, f ) is still wPBE. P ROOF : If we set µ = 1, we directly see this.

(1, −2)
a
M
µ
se
f
(−3, −1)
E

e (3, 1)
a
ce
E
1−µ
f
ne (−2, −1)
(0, 2)

3
Figure 34. Extensive form game of entry deterrence III

Player 2
a f

se 1, −2 −3, −1
Player 1
ce 3, 1 −2, −1

• Example 18 shows that wPBE SPNE ⊂ NE.

Definition 21 : Perfect Bayesian Equilibrium (PBE) is wPBE which includes wPBE in every
subgame.

Definition 22 : (Kreps and Wilson) Let σ = (σ1 , . . . , σn ) be strategy profile and µ =

(µ1 , . . . , µn ) be system of belief. Then (σ, µ) is a sequential rationality if

(i) (sequential rationality) for every i and for every information set h, σi is optimal to (σ−i , µ)

(ii) (consistency) µ is a limit of {µn } such that µnh (x) = pσ (x(h)) for some a sequence of
n

• Let

– λn : Probability that the entrant chooses e

– αn : Probability that the entrant chooses se given that he chooses e.

• Thus λn αn is the probability the entrant chooses e and se.

4
• We have
λn αn
µn (α) = = αn
λn
and limαn →0 µn = αn = 0 which satisfies consistency.

βn

α
αn
1 − βn
E

λn
βn
1 − αn β
E

1 − βn
1 − λn

Figure 36. Extensive form game of entry deterrence III

• We see that
SE ⊂ PBE ⊂ SPE ⊂ NE.

5
Lecture 13 (April 14, 2014)

Incomplete Information Game (Bayesian Game)

• Two features

(i) Nature moves first (Player, strategy and payoff are chosen).
(ii) All players are not perfectly informed of the nature’s choice choice for other players
(Asymmetric information occurs).

– Imperfect information game: unobservable actions of other players

– Incomplete information game: unobservable characteristic of other players.

• Player II’s type is t1 or t2 .

Player 2 Player 2
L R L R

U 3, 1 3, 0 U 3, 0 2, 1
Player 1 Player 1
D 0, 1 4, 1 D 0, 0 4, 1

t1 t2

1
L (3, 1)

U
R (2, 0)
II
t1 L (0, 1)
D

R (4, 0)
N I
L (3, 0)

t2 U
R (2, 1)
II
L (0, 0)
D

R (4, 1)

• Game structure

(i) a type of player i:

– t1 ≡ (si , πi ) ∈ Ti where Ti is the set of type.
– t1i = t2i ⇐⇒ s1i = s2i and πi1 = πi2 .
(ii) strategy rule: si : Ti → Si
(iii) payoff function:
πi : S1 × S2 · · · Sn × T1 × T2 · · · Tn → R

Definition # : A Bayesian Nash equilibrium (BNE) is a set of strategy rule {s∗i (ti )} such that
each player maximized his expected payoff givne his type and other player’s strategy rule.

• Suppose there are Player I and Player II then BNE is formally defined as follows:

– BNE is a pair of strategy rule ({s∗1 (ti )}, s∗2 (t2 )) such that ∀ t1 ∈ T1 and ∀ s1 ∈ S1 ,
Z Z
π1 (s1 (t1 ), s2 (t2 ), t1 , t2 ) p(t2 |t1 )dt2 ≥ π1 (s1 , s∗2 (t2 ), t1 , t2 ) p(t2 |t1 )dt2 ,
∗ ∗

and ∀ t2 ∈ T2 and ∀ s2 ∈ S2 m
Z Z
π2 (s∗1 (t1 ), s∗2 (t2 ), t1 , t2 ) p(t1 |t2 )dt1 ≥ π2 (s∗1 (t1 ), s2 , t1 , t2 ) p(t1 |t2 )dt1 .

2
Example # BNE of Example #

Player 2 Player 2
L R L R

U 3, 1 3, 0 U 3, 0 2, 1
Player 1 Player 1
D 0, 1 4, 1 D 0, 0 4, 1

t1 t2

Figure #(a). Nash equilibrium for t1 and t2 in game of example #

• Nash equilibrium:

(U, L) if t = t1
(D, R) if t = t2

• Expected payoff of Player I for choosing U and D:

U : 3p + 2(1 − p) = p + 2
D : 0 + 4(1 − p) = 4(1 − p).

• BNE:

U if p > 2/5
s1 =
D if p < 2/5

L if t = t
1
s2 =
R if t = t
2

3
Example # BNE of the Modification of Normal Form Game in Example #

Player 2 Player 2
L R L R

U 3, 1 3, 0 U 3, 0 2, −1
Player 1 Player 1
D 0, 1 4, 1 D 0, 0 4, −1

t1 t2

• BNE:

s1 = 3

L if t = t1
s2 =
L if t = t2

• Separating equilibrium: BNE which classifies the types of players (Example #)

• Pooling equilibrium: BNE which does not classify the types of players (Example #)

Example # Entry Deterrence x

f (−1, −1)
M
enter
E a (1, 1)
tough
not (0, 2)

N f (−1, 0)
enter
soft a (1, 1)

E
not (0, 2)

4
Figure # Entry deterrence x

• BNE

– Pooling equilibrium:
(i)

s1 (t) = enter ∀ t
s2 = a

(ii)

s1 (t) = not enter ∀ t

s2 = f

– Separating equilibrium:

• Let p = 1/2.

h (9, 5)
H
train
W n (−1, 0)

high h (7, 3)
not
n (0, 0)
N
h (6, −1)
train
low n (−4, 0)

W h (−1, 1)
not
H n (0, 0)

Figure #. Home assignment III

5
Lecture 19 (May 14, 2014)

Forward Induction

(−3, −3)
R

R
C (−1, 1)
E
M
enter
(1, −1)
R
C
E

C (−2, −2)
not
(0, −2)

M
R C

R − 3, −3 −1, 1
E
C 1, −1 −2, −2

(not enter, (R, C)) ,

(enter, (C, R)) .

• The SPE (not enter, (R, C)) is not plausible.

1
• We can iterative eliminate of weakly dominated strategy as follows.

M
C R

not enter 0, -2 0, -2 3

E enter C -2, -2 1, -1

• Man has the option to burn money (self-injury)

F
s t

s 3, 1 −2, 0
burn M
t −2, 0 2, 2

M
F
s t
not burn s 5, 1 0, 0
M
t 0, 0 1, 5

2
F
ss st ts tt

bs 3, 1 3, 1 -2, 0 -2, 0 5

bt -2, 0 -2, 0 -1, 5 -1, 5 1

M
nb 5, 1 0, 0 5, 1 0, 0

nt 0, 0 1, 5 0, 0 1, 5 3

4 2 2
Figure #. Battle of the sexes with burning money

• Man has the option to burn money (self-injury)

M
R C

R −1, −1 1, 3
enter 1
E
C 3, 1 −1, −1

E
M
R C
enter 2 R 0, 0 2, 1
E
C 1, 2 0, 0

• Equilibrium: (enter 1, (C, R)).

3
Lecture 20 (May 19, 2014)

burn not burn

s t s t

s 3, 1 −2, 0 s 5, 1 0, 0

t −2, 0 −1, 5 t 0, 0 1, 5

Figure #. Burning money in battle of sexes

• Forward induction

1. Eliminate weakly dominated strategy (iteratively)

2. Eliminate equilibrium dominated strategy
3. Eliminate equilibrium never a weak best response (interior strategy)

Essential Equilibrium

Definition # : σ ∈ G is an essential equilibrium if for every ε > 0, there exists δ > 0 such
that d(G, G′ )δ then there exists a Nash equilibrium σ ′ ∈ G′ such that d(σ, σ ′ ) ∈ ε.

Example # Not a essential equilibrium

L M R

U 1, 1 1, 0 0, 0

D 1, 1 0, 0 0, 1

1
Figure #. Not a essential equilibrium

• Pure strategy NEs: (U, L), (D, L)

• If we set some perturbation of given NEs such as (U, L) = (1+ε, 1+ε) or (D, L) = (1+ε, 1+ε),
the NEs are not stable.

• Thus (U, L) and (D, L) are not essential equilibrium.

• Essential equilibrium is set-valued solution concept while previous studied equilibriums are
pint-valued solution concept.

Strategic Stability (Kohlberg and Merton)

• Axiomatic approach

– existence
– backward induction
– invariance of inessential trasform
– admissibility (not weakly dominated)
– iterative dominance

• Axioms of Kohlberg and Merton

– existence
– backward induction
– invariance
– iterated dominance
– forward induction
– connectedness

2
Theorem # : The set of NE consists of finitely many connected components

Theorem # : At least one component of Nash equilibrium contains a stable set in the same
that for any perturbation in payoffs, a Nash equilibrium is close to the component.

Definition # : (Rough definition) A stable set of equilibrium is a minima set of equilibrium

in the normal form of a game that in the sense of p.

(−1, −1)
fight

Monopolist
enter

not enter
(0, 2)

– Reduced normal form

M
a f

enter 1, 1 − 1, −1
E
not 0, 2 0, 2

Figure #. Reduced normal form of entry deterrence

3
– Find NE sets (including mixed strategy equilibrium)

 1 if q > 1/2

p = [0, 1] if q = 1/2

0

if q < 1/2

[0, 1] if p = 0
q =
1 if p > 0

NEs are given by (p = 0, 0 ≤ q ≤ 1/2) and (p = 1, q = 1).

– Stable set: (p = 1, q = 1)

Remark # :

2. Every point in stable set is perfect equilibrium.

4
Lecture 21 (May 21, 2014)

• Axioms of Kohlberg and Merton

1. existence
2. backward induction
3. invariance
5. iterated dominance
6. forward induction
7. connectedness

Definition # : A closed set S of Nash equilibrium is a hyper stable set if it is minimal with
respect to the following property:

• For every perturbed normal form game, there exists a Nash equilibrium close to S.

• Sequential equilibrium

– violate 3 and 4

• Perfect equilibrium

– violate 6

• Hyper stable set

– violate 4 and 5

1
T B

T 0, 0 3, 1

in B 1, 3 0, 0

out
(2, 2)

T B

T 0, 0 3, 1

B 1, 3 0, 0

O 2, 2 2, 2

Figure #. Not a forward induction equilibrium

• It is noted that (OB, T ) is SPE, PE and proper equilibrium but it is not a forward induction
equilibrium.

Definition # : A closed set S on Nash equilibrium is a stable set if it is minimal with respect to
the following property: for every η > 0 and for some ε′ > 0, ∀ ε < ε′ and for any perturbation
(ε(si )), every perturbed game G(ε) with 0 < ε < ε′ has an equilibrium within η of some
equilibrium in the set.

• Note that stable set ⊂ fully stable set ⊂ hyper stable set.

Example #

2
1

a 2 , −1

h
f 1

2 , −1
E II
enter a (1, 1)

I f (−1, −1)

not
(0, 2)

a f
1 1
Eh 2 , −1 2 , −1

Eℓ 1, 1 −1, −1

not 0, 1 0, 2

• I’s best response

3
 p =1 if q >
 1 4

3
p1 + p2 = 1 if q = 4

p2 = 1 3
if q <

4

• II’s best response

q = 1 if p2 > 0
q ∈ [0, 1] if p2 = 0

3

• NEs are given by (p2 = 1, q = 1) and p1 = 1, q < 4 .

Example #

3
(2, 0, 0)
T ℓ (0, 3, 3)
L
r (5, 1, 1)
I
ℓ (5, 1, 1)
M
R
NT r (1, 0, 0)
II III
ℓ (1, 0, 0)
L
B (5, 1, 1)
r
ℓ (5, 1, 1)
R
r (0, 3, 3)

• Stable set: {(T, L, ℓ) ∪ (T, R, r)}

• Sequential set: 0, 21 , 12 , 12 , 21 , , 1 1
   
2, 2

4
mixed strategy와 benavioral strategy
수업 시간에 예를 든 perfect recall game에서 player 1의 순수전략: Uuu, Uud, Udu, Udd,
Duu, Dud, Ddu, Ddd
1. 내가 예로 들려고 했던 mixed strategy는 처음에 U하면 나중에도 u하는 전략과 처음에
 
D하면 나중에도 d하는 전략을 각각 1/2로 택하는 전략         
 
즉 Uud는 처음에 U하고 이처럼 U하면 나중에 u, D하면 나중에 d를 하는 것이므로 수업시
간에 예를 든 Uuu처럼 처음에 D를 해도 나중에 u를 하는 전략보다 내가 생각하는 perfect
correlation strategy의 의미에 보다 잘 부합한다. 그러면 player 2가 무조건 확률 1로 T라는
순수전략을 택한다고 할 때 terminal node에 induce하는 확률분포는 위부터 (1/2,
0,0,0,0,1/2,0,0)이 된다. 이제 이와 동일한 확률분포를 induce하는 behavior strategy를 구
하면 우선 player 1의 첫 information set 에서 U를 택할 확률은 Uuu, Uud, Udu, Udd를
택할 확률의 합이므로 1/2이 되므로    이다. 이제 두 번째 information set
 
 에서 u를 어떤 확률로 택해야 위와 같은 확률분포를 induce하는가?  ××  가 되
 
려면   이 되어야 한다. 따라서     임. 마지막으로 세 번째 information set 
 
에서 u를 택할 확률은  ×  ×      이어야 하므로    , 따라서    이 되면
 
위와 동일한 확률분포를 induce하게 됨

 
2. 이제 수업 시간에 예를 든 mixed strategy        를 생각해 보자. 수
 
업 시간에 설명한 방식대로 풀어보면 위와 같은 논리로    ,      ,
    를 얻게 됨. 따라서 두 개의 서로 다른 mixed strategy  과  에 하나의
behavior strategy가 대응되는 1대1의 관계가 성림하지 않는 (many to one) 예가 됨. (수업
시간에 어떤 학생의 질문에 대한 대답임)

3. 이제  와 같은 확률분포를 유도하는 behavior strategy를 처음에 수업시간에 구하려

했던 방식으로 구하면 왜 안 되는가?   를 구할 때 player 1의 strategy 중 두 번째
action이 u인 strategy, Uuu, Uud, Duu, Dud 네 가지 경우의 확률을 더해서 1/2이라고 생각
하면 안 됨. 왜냐하면 두 번째 information set 에서 u를 택할 확률은 에 도달했다는 조
건 하에 첫 번째 terminal node에 도착할 확률을 구해야 함. terminal node들을 위에서부
터 1, 2, 3, ..., 8이라 부르기로 하면 우리가 구해야 할 것은
 ∩           
            
                     
따라서     가 됨. 같은 방법으로   를 구할 때는
 ∩          
           
                   
따라서    가 됨.

4.  에 대해서도 같은 논리를 적용하면     ,    를 구할 수 있음.

게임이론과 응용
연세대학교 경제학부 2006년 봄학기
교수 김영세 중간시험

N
1. 어떤 과점시장에 N개의 잠재적 기업이 있다. 시장수요함수는 P  A   qi 이며
i 1

양(+)일 경우에만 소요된다.

(1) 내쉬균형을 구하라.
(2) 완전담합이 이루어질 경우 생산량, 가격, 결합이윤을 구하라.
(3) 경쟁상황에서는 일부 기업은 퇴출하고 나머지 기업만 조업하는 결과가 나오는
반면 완전담합 하에서는 N개의 기업이 모두 생존하는 경우를 찾고 해석하라.

2. Downs 모형에서와는 달리, 후보들이 집권확률에는 관심이 없고 집권자의 정책

포지션(position)에만 관심이 있는 상황을 고려하자. 두 후보가 있다. 각 후보는
나름대로 가장 선호하는 정책포지션이 있으며 실제 정책이 어느 방향으로든 자신이
가장 선호하는 정책으로부터 멀어질수록 보수가 낮아진다. 진보성향의 후보1이 가장
이상적으로 여기는 정책포지션 d1은 중위유권자의 그것(m)보다 좌측에 위치한다. 반면
보수성향의 후보2가 가장 선호하는 정책포지션 d2는 m보다 우측에 위치한다. (즉, d1 <
m < d2임)

후보1는 전략 x1을 선택하고 후보2는 전략 x2를 선택한 경우 더 많이 득표한

후보가 선거에서 이기며 그 후보가 약속한 전략이 실행된다. 두 후보가 똑 같이
득표할 경우에는 타협정책인 1
2 ( x1  x 2 ) 이 실행된다. 이 게임에서 내쉬균형을 구하라.
(힌트: 다음 순서로 한 경우 한 경우 따져보아라. (i) x1  m, x 2  m , 또는
x1  m, x 2  m , (ii) x1  m  x 2 또는 x 2  m  x1 , (iii) x1  m, x 2  m 또는
x1  m, x 2  m , (iv) x1  m  x 2 .).

3. 두 국가 1, 2가 있고 각국에는 독점기업 1, 2가 있으며 재화는 하나밖에 없다. 기업은

단일재화를 생산하여 내수용으로 국내에 hi 단위를 공급하고 수출용으로 해외에 ei
단위를 파는데 생산비는 전혀 들지 않는다고 하자. 국가i의 시장수요함수를

1
Pi  Ai  Qi (단 A1  A2 는 둘 상수, Qi  hi  e j 임) 이라 하자. 당연히 국가의

목적함수는 소비자잉여, 생산자잉여, 정부수입을 합한 사회후생함수이며 기업의

목적함수는 이윤이다.
(1) 완전폐쇄경제체제 하에서 부분게임완전균형을 구하라.
(2) 완전개방경제체제 하에서 부분게임완전균형을 구하라.
(3) 국가규모에 따라 개방의 효과가 어떻게 달라지는지 위의 두 결과를 기초로
해석하라.

4. 다음과 같은 최후통첩협상게임의 변형을 고려하자. 두 경기자 1과 2가 10만원의 돈을

나누려고 한다. 경기자1은 10만원이나 0원은 제안할 수 없으나 그 사이의 어떠한
금액도 제안할 수 있다. 경기자1의 제안을 경기자2가 수락하면 그대로 나누고
거부하면 둘 다 0의 보수를 얻는다. 이제 mi를 경기자i가 받는 금전수입이라 놓으면,

(1) 전개형게임을 나타내라.

(2) 부분게임완전균형을 구하라.
(3) 위 답이 c에 따라서 어떻게 달라지는 해석하라.

5. 투자자(investor)가 1(억)원의 자금 가운데 T(억)원만큼은 수탁자(trustee)에게 맡기고

나머지 (1-T)(억)원은 일단 자신이 갖는다. 그 다음 수탁자가 수탁자금 T를 투자하면
2T억원의 순이익이 발생한다. 수탁자는 2T억원의 금액을 자신과 투자자에게 어떻게
배분할지를 결정한다. 이 때 수탁자 자신이 갖는 금액을 Y라 놓자. 이 게임은
일회성이며 두 사람은 다시는 만날 일이 없다고 가정하자.
(1) 전개형게임을 묘사하라.
(2) 부분게임완전균형과 균형보수를 구하라.
(3) 부분게임완전균형보다 사회적으로 우월한 최적 자원배분을 구하라.
(4) 부분게임완전균형보다 우월한 자원배분이 달성하도록 하려면 어떠한 방안이
있겠는가?

2
6. 두 경기자가 용의자의 딜레마 게임을 무한 반복해서 경기한다고 하자. (단, F>3는 상수)

Coop Fink

Coop 3, 3 0, F

Fink F, 0 1, 1

(1) 두 경기자가 모두 무자비촉발전략을 사용한다고 가정할 때 협조체제 (3,3)이

반복게임의 내쉬균형으로 달성가능하려면 할인인자가 최소한 얼마나 되어야
하겠는가? [6 점]
(2) 두 경기자가 모두 “tit-for-tat”전략을 사용한다고 가정할 때, 협조체제 (3,3)가
반복게임의 내쉬균형 보수로 달성되려면 할인인자가 어떠한 조건을 만족해야
하겠는가? [6 점]
(3) 부분적 협조체제 (2,2)의 달성가능 여부에 대하여 위의 (1), (2) 각각의 경우에
대하여 풀어보고 설명하라. [8 점]

모범답안

1.

 
(1) 경기자-i의 극대화 문제는 max qi
ui   A   q j  qi  (qi ) 2  F  이다. 제1계조건을 구하
 j 

1 
여 정리하면 경기자-i의 최선응수상응 qi   A   q j  이 도출된다. 대칭성에 의해 모든 qi를
4 j i 

A 2 A2
똑같이 놓고 풀면 개별기업 생산량은 qi  이며 개별기업 이윤은 ui  F 로 일
N 3 ( N  3) 2
단 구해진다. 이렇게 구한 이윤이 양(+)수라면 N개의 기업이 모두 조업을 하는 내쉬균형이 성립하

겠으나, 만약 음(-)수라면 일부 기업만 조업하고 나머지 기업은 퇴출될 것이다. 이러한 아이디어에

착안하여 아래 식을 만족하는 非陰(non-negative) 정수 가운데 가장 큰 정수를 N*라고 표기하자.

2 A2
 F  0 혹은 N  A 2
3
( N  3) 2 f

결국 내쉬균형에서 N*개의 기업은 산업에 남아서 조업을 하고 나머지 (N-N*)개의 기업은 퇴출하

A 2 A2
게 된다. 이 때 생존기업의 생산량은 qi *  이고 생존기업의 이윤은 u *  F
N*  3 ( N *  3) 2
i

이다.

3
(2) 결합이윤은 모든 기업의 이윤을 더한 것이므로 카르텔의 최적화 문제는

     
max U   u j   A   q j    q j     (q j ) 2   NF 
( q1 ,, q N )
j  j  j   j  
제1계조건(기업-i의 생산량 qi에 대하여 편미분하여 0으로 놓은 식)을 구하면 최선응수상응

1 
qi   A  2 q j  이 도출된다. 대칭성에 의해 모든 qi를 똑같이 놓고 풀면 개별기업
4 j i 

A  A2 
생산량은 qi  이며 결합이윤은 U  N   F  로 일단 구해진다. 이렇게 구한
2( N  1)  4( N  1) 

이윤이 양(+)수라면 N개의 기업이 담합하는 상황이 성립하겠으나, 만약 음(-)수라면 심지어 독점산

업 하에서도 손실이 발생함으로 의미하므로 산업생산이 0이 될 것이다. 결론적으로

A2 A A2
만약 F 이라면, qˆi  이며 uˆi  F
4( N  1) 2( N  1) 4( N  1)
A2
만약 F 이라면, qˆi  0 이며 uˆi  0 .
4( N  1)

(3) 위의 (1)에 의하여 내쉬균형에서 일부 기업은 퇴출하고 나머지만 조업할 조건은

2 A2
F 이고, 위의 (2)에 의하여 완전담합에서 N개의 기업이 모두 조업할 조건은
( N  3) 2
A2 A2 2 A2
F 이다. 한편  은 언제나 성립함을 쉽게 보일 수 있다. 결론적으로
4( N  1) 4( N  1) ( N  3) 2
초기장치비용 F가 아래 부등식의 범위에 있으면 된다.

A2 2 A2
F
4( N  1) ( N  3) 2
만약 장치비용 F가 지나치게 크면 완전담합 하에서도 손실이 발생하여 아무도 생산에 참여하지

2.

문제를 푸는데 x1≤x2이라고 가정해도 무방하다.

4
<Case i> x1  m, x2  m
어떤 경우에도 후보2가 전략을 바꿀 유인이 존재함. 좀 더 구체적으로

만일 d 2  2m  d1 이면, 후보2는 x2 대신 d2을 선택함으로써 더 높은 보수를 얻을 수 있다.

만일 d 2  2m  d1 이면, 후보2는 x2 대신 (2m-d1-ε)을 선택함으로써 더 높은 보수를 얻을 수
있다.

x1  m, x 2  m 인 경우도 유사한 논리로 후보1이 전략을 바꿀 유인이 있다.

<Case ii> x1  m  x2
Without loss of generality, assume that x2  2m  x1 (즉, 후보1이 승자가 될 상황).
만일 d 2  [m,2m  x1 ] 이면, x2에서 d2으로 전략을 바꿀 유인이 있다.
만일 d 2  2m  x1 이면, x2에서 (2m-x1)으로 전략을 바꿀 유인이 있다.
<Case iii> x1  m, x 2  m
후보1가 m에서 왼쪽으로 옮길 유인이 있다.

있다.

만일 d1  2m  x2 이면, 후보1은 m 대신 d1을 선택함으로써 더 높은 보수를 얻을 수 있다.

x1  m, x 2  m 인 경우에도 유사한 논리를 적용.
<Case iv> x1  m  x2
후보1, 후보2 어느 누구도 다른 전략으로 바꿀 유인이 없다.

3.

(1) 기업-i의 이윤극대화 문제 max Qi

ui  ( Ai  Qi )Qi 를 풀면 균형가격, 기업이윤, 소비자잉여가

각각 다음과 같이 구해진다.

1 1 1
Pi Autarky  Ai , uiAutarky  ( Ai ) 2 , CSiAutarky  ( Ai ) 2
2 4 8

(2) 기업-i의 이윤극대화 문제 max hi ,ei

ui  ( Ai  hi  e j )hi  ( Aj  h j  ei )ei 로부터 제1계조건을

구하여 연립방정식을 풀면 균형가격, 기업이윤, 소비자잉여가 각각 다음과 같이 구해진다.

1 1 2
Pi LF  Ai , uiLF  ( A1 ) 2  ( A2 ) 2  , CSiAutarky  ( Ai ) 2
3 9 9
(3) 폐쇄경제체제에서 완전개방체제로 이행함에 따라 나타나는 효과는 다음과 같다.첫째, 대국이나

5
소국 할 것 없이 가격은 하락하고 소비자잉여는 증가한다.

둘째, 대국의 독점기업의 이윤은 적어지고 소국의 독점기업의 이윤은 두 시장의 상대적인 크

기에 따라 증가할 수도 있고 감소할 수도 있다. 구체적으로

A2 5
만약  (즉, 대국과 소국이 규모에서 큰 차이가 없으면), 소국의 독점기업 이윤도 낮아진
A1 2

A2 5
다. 반면 만약  (즉, 대국의 시장규모가 상대적으로 훨씬 크면), 소국의 독점기업 이윤은
A1 2

높아진다.

셋째, 소국의 사회후생은 언제나 상승하나 대국의 사회후생은 두 국가의 상대적 규모에 따라

A2 2 6
증가할 수도 있고 감소할 수도 있다. 구체적으로 만약  (즉, 대국과 소국의 규모가 상
A1 3

A2 2 6
대적으로 엇비슷하면), 대국의 사회후생도 증가한다. 그러나 만약  (즉, 대국의 시장규
A1 3

모가 상대적으로 훨씬 크면), 대국의 사회후생은 오히려 낮아진다. 여기서 국가-i의 사회후생은

3 1 1
Wi Autarky  ( Ai ) 2 에서 Wi LF  ( Ai ) 2  ( Aj ) 2 으로 변함에 유의하라.
8 3 9

4.

(2) 부분게임에서 경기자2는 m2  cm1 이 0보다 클 경우에는 수락하고 작을 경우에는거부할 것이

10
다. 그런데 수락할 경우 m1  m2  10 이라는 사실에 착안하면, 경기자2는 m1  이면 수락
1 c
하고 그렇지 않으면 거부할 것임을 계산할 수 있다. 이제 전체게임으로 거슬러 올라가면 부분게

 10 10c 
임완전균형에서 경기자1이  ,  를 제안할 것임을 알 수 있다.
 1 c 1 c 
(3) 매개변수 c가 0에 가까우면 부분게임완전균형은 (10,0)에 근접하고 반대로 1에 가까우면 (5,5)

spitefulness를 나타냄에도 불구하고 부분게임완전균형에서 fair division이 나타나는 재미있는 결과

를 얻게 된다.

6
5. Problem Set에서 똑 같은 문제 출제.

(1) 간단

(2) 균형보수 (1,0)

(3) 투자자는 1억원을 전부 신탁하고 2억원의 투자수입이 발생하면 수탁자는 자신이 0과 1사이의

보수를 챙기고 투자자에게 1과 2 사이의 보수를 돌려준다. 둘 다에게 좋을 뿐 아니라 사회적으로

도 최적상태임.

(4) 신탁회사의 도덕적 해이, 그로 인하여 사회적으로 열악한 자원배분이 균형현상으로 관찰, 개인

간의 계약으로 일부 해결 가능하나 한계가 있음, 법제도(예컨대 간접투자운용관리에 관한 법률)의

중요성

6.

F 3
(1) 3  (1   ) F   1   
F 1
F F 3
(2) 3  (1   )( F  0    F    )    
2

1  3
F 2
(3) 무자비촉발전략: 
F 1
F 2
Tit-for-tat 전략: 
2

7
게임이론과 응용
연세대학교 경제학부 2008년 봄학기
교수 김영세 학기말시험

1. [22점] 다음 전개형 게임을 보고 물음에 답하라.

P1 Across
0, 5

Left Right

P2

Up Down Up Down

5, 0 -1, -2 5, 0 -3, -5

(1) [6점] 내쉬균형을 구하라.

(2) [6점] 완전베이즈균형을 구하고 설명하라.
(3) [10점] 보수구조가 아래와 같을 때 완전베이즈균형을 구하라 (힌트: 혼합전략 균형만 하
나 존재한다.)
P1 Across
0, 5

Left Right

P2

Up Down Up Down

5, 0 -2, 2 4, 0 -3, -5

1
2. [20점] 어떤 회사가 근로자와 고용계약을 맺으려고 한다. 근로자가 외부에서 벌 수 있는
유보임금은 W이며 근로자가 떠날 경우 이 회사의 이윤은 0이다. 근로계약 체결 후 근로
자가 열심히 일하면 H단위의 총 수입이 생기고, 근로자가 태만하면 반-반의 확률로 H단
위의 총 수입이 생기든지 아니면 L단위의 총 수입이 생긴다. 근로자가 열심히 일할 경우
근로자는 C단위의 기회비용이 들며 태만할 경우 기회비용이 들지 않는다. (단 H>L>W>C
는 항상 성립한다.)
(1) 회사와 근로자 간의 전개형 게임을 그려라.
(2) 일회게임의 부분게임 완전균형을 구하라.
(3) 무한 반복게임에서 회사가 무자비 촉발전략(grim strategy)을 사용한다고 가정하자.
근로자의 할인인수가 δ 일 때, 실효임금(efficiency wage)과 회사의 기대이윤을 계산
하라. 매개 변수들 간의 크기에 따라 답이 달라지므로 유의하라.

3. [20점] 효용함수가 u(x)=xα (단, 0<α≤1)로 동일한 두 명의 구매희망자가 하나의 물건을 놓

고 경매에 참가한다. 구매희망자의 가치평가는 β의 확률로 1(백만원)이거나 나머지 (1-β)
의 확률로 0이다. 또 두 구매희망자의 가치는 상호독립적이다.
(1) 베이즈게임으로 구성하라
(2) 영국식 경매에서 베이즈내쉬균형을 구하고 경매인(auctioneer)의 기대수익을 구하라.
(3) 최고가밀봉입찰에서 베이즈내쉬균형을 구하고 경매인의 기대수익을 구하라.
(4) 위의 두 경우를 비교하라.

4. [20점] 노동시장에서의 신호 문제를 고려하자. 임의의 노동자의 생산성 θ는 θH이거나 혹

은 θL이며, 동 직종 내에 생산성이 높은 노동자의 비율은 f∈[0,1]이다. 모든 노동자의 유보
임금은 θL로 똑같다. 노동자는 교육수준 x∈[0,x ] (여기서 x 는 충분히 큰 상수)를 선택함으
로써 기업에게 신호를 발송할 수 있다. 기업은 노동시장에서 완전경쟁에 직면하므로 0의
기대이윤을 얻는다고 가정하자. 노동자의 효용함수와 기업의 이윤함수는 각각 다음과 같
다.
U(w, x | θ) = w - c(θ)x
V(w, x | θ) = θ + αx - w
위에서, c(θL) > α > c(θH)는 항상 성립한다.
(1) 완비정보하의 균형을 구하라.
(2) 미비정보하에서 분리 완전베이즈균형을 구하고, 그림으로 나타내라.
(3) 미비정보하에서 공용 완전베이즈균형을 구하고 그림으로 나타내라.
(4) 위 (2), (3)에서 구한 완전베이즈균형 가운데 Cho-Kreps 기준을 만족하는 균형은 어
떤 것인가?

2
5. [25점] 기업의 소유주가 전문경영인을 고용하였다. 전문 경영인은 최선을 다해 경영활동
을 할 수도 있고 연봉이나 받으면서 적당히 회사를 운영할 수 도 있다. 열심히 일할 경우
(Work) 전문경영인에게 드는 수고비용은 γ단위(단, 0<γ<1)이지만 태만할 경우(Shirk) 드는
수고 비용은 없다. 전문경영인이 열심히 일할 경우 총 수입 y∈[0,1]의 누적확률분포함수는

전문경영인은 이 회사에 취직하지 못하더라도 다른 직장에서 일하면 β단위의 유보효용은

얻을 수 있다. 소유주는 위험 중립적이고 전문경영인은 효용함수 U w, x √w x (여기
서, w는 연봉이고 x는 수고비용)을 갖는다.
(1) 소유주가 전문 경영인의 노력수준을 관찰할 수 있는 경우, 최적 계약 형태를 구하라.
소유주 입장에서 전문 경영인이 열심히 일하는 상태를 더 선호할 조건을 적시하라.
이제 소유주가 전문 경영인의 노력 수준을 관찰할 수 없다고 가정하자. 또 위(1)에서 구
한 조건이 아래 모든 문제에서 성립한다고 가정한다.
(2) 성과급 체계의 한 유인 계약 w y , 설계 문제를 구성하라.
(3) 위(2) 최적화 문제에서 유인계약은 어떠한 방정식들을 만족하는 함수인지 제1계 조
건을 구하고 의미를 설명하라.
(4) 유인계약 계약 w y , 을 라그랑지 상수 λ, μ의 함수로 구하고 그래프로 나타
내라.
(5) 위(4)의 결과를 단조우도비율조건(MLRP)과 연결하여 설명하라.

모범답안

1. (1)
Up Down

Left 5,0 -1,-2

내쉬균형은 3개 존재한다. ① (Across, Down if L or R), ② (Left, Up), ③ (Right, Up)

(2) 위의 내쉬균형 중 ①은 완전베이즈균형이 아니다. P1은 Left or Right 을 어떠한 확률로
선택하든 P2에게 Down은 Strictly dominated by Up. 즉 NE 에서 ① 은 sequential
rationality 위배. 나머지는 ②와 ③은 PBE임.
(3) Pure NE 없다. 그런데 PBE ⊆ NE 이므로 당연히 pure PBE는 없음. 특히 P1 이 Across를
선택하는 균형은 존재할 수 없다. (왜냐하면 Right가 Across보다 강우월하므로).

3
Up Down
Across 0,5 0,5
Left 5,0 -2,2
Right 4,0 3,-5

Up (q) Down (1-q)

Left (p) → 5q-2(1-q)
Right (1-p) → 4q+3(1-q) q*=5/6
↓ ↓
0 2p-5(1-p) p*=5/7
그러므로 완전베이즈균형은 {(0, 5/7, 2/7), (5/6, 1/6)} over {(A,L,R),(U,D)}이고 균형보수 행렬은
(23/6, 0)이다. ■

2. (1) 교과서 참조.

(2) If H L W , the Firm offers w and worker accepts the offer and shirks. If H L W , (which
is not possible in this case, because H>L>W), the firm offers less than W and the worker
rejects.
(3) The value when working hard is Vhard = W-C. The value when shirking, Vshirk, is as follows:

that Vhard ≥ Vshirk, we have the efficiency wage W W C.

(Case i) H L 2 C (which comes from the condition that H W H L W.)
(i-a) If H W C then the firm offers W , and the worker accepts the offer and
*

works hard. The expected revenue to the firm is max{0,H-W*}.

(i-b) If H W C then the firm offers less then W, and the worker rejects
(Case-ii) H 2 C
The firm offers W, and the worker accepts and shirks. ■

3, 4. 이전 기출문제 참고.

5.
(1) √w γ β Î w β γ
√w 0 β Î w β
EVhard (the principal’s expected revenue when the agent works hard)

4
= ydF y|Work β γ = β γ
EVshirk (the principal’s expected revenue when the agent shirks)
= ydF y|Shirk β = β
If EVhard ≥ EVshirk, i.e. γ 2β γ , then it is optimal to implement the agent’s working
hard. Otherwise, it is optimal to implement the agent’s shirking.
(2) max ,
EV y w y dF y|Hard
s.t.
(IC) w y dF y|Hard γ w y dF y|Shirk
(IR) w y dF y|Hard γ β
(3) 교과서 풀이 참조하여 라그랑지 극대화 문제를 풀어 제1계 조건을 구하면 다음과 같다.
|S
µ λ 1
|H

그런데 f y|Shirk y , f y|Hard αy 를 대입하면

w y µ λ 1 y (FOC)

위 (2)의 두 식 (IC), (IR)을 등호로 놓고 (FOC)와 함께 세 식을 연립해서 풀면 세 미지수

λ, µ, 그리고 무엇보다도 w y , 를 구할 수 있다.

(4) 위(3)에서 구한 바
w(y)

0 y
y0 1

[[추가: 학부생들에게 이렇게까지 요구하지는 않음. 그러나 대학원생인 경우 아래와 같은 아

이디어에 착안하여 recursive method를 적용하는 안을 제시해야 함.]]
Let y0 be such that

5
1
Max EV = ∫ [ y − w( y )]dF ( y Hard )
{ w ( y )} y0
1
s.t ( IR) ∫
y0
w( y ) dF ( y Hard ) = γ + β
1
( IC ) ∫ y0
w( y ) dF ( y Shirk ) =β
1
1 1 − (α − )
FOC w( y ) = [ μ + λ{1 − 2 y α }] for y ∈ [ y 0 ,1]
2 α
1
1 1 − (α − )
w( y ) = max{0, [ μ + λ{1 − 2 y α }] 2 }
4 α

이상의 method 언급했으면 가산점, y 0 는 (IR),(IC) 조건 이용하여 구함.

1
f ( y Shirk ) 1 − (α − )
(5) 우도비율 = = y α , α > 1 이므로 우도비율은 y에 대해서 감소
f ( y Work ) α2