Macroeconomics

Lecture 5

1 Arayah Preechametta 10/15/2018

 Nash equilibrium (Time consistency)

• Consider a two-player dynamic game with no

randomness

Player i has an objective to

∑ i ( t it jt )
T
max β t
r x , u , u + β T +1
V0i ( xT +1 ) ,
{uit }t =0
T
t =0

0 < β < 1; =
i 1, 2; i ≠ j.
s.t. xt +1 = g ( uit , u jt ) ,
x0 given.
For player i, {u jt }
T
is given.
t =0
2 Arayah Preechametta 10/15/2018
 The L arg rangian function for player=
i, i 1, 2., i ≠ j , is

∑ β r ( x ,u ,u ) + β ( x ) + ∑ β t λt'  g ( uit , u jt ) − xt +1 ,

T T
T +1
=
Li i t it
t
jt 0i T +1 V
t 0=t 0

x0 given, and {u jt }
T
is given for player i.
t =0

1, 2., i ≠ j , are
For the case of player i, the 1st − order cond . w.r.t. uit , i =
∂ri ( xt , uit , u jt ) ∂g ( uit , u jt )
+ λt =
0, t =
0,..., T − 1, (I )
∂uit ∂uit
Differentiate Li w.r.t. xt to get ,
∂ri ( xt , uit , u jt )
β t
− β t −1λt −1 =
0, t =
0,..., T − 1, ( II )
∂xt
Solving ( II ) for λt −1 and shifting forward one period ,
∂ri ( xt +1 , uit +1 , u jt +1 )
λt = β , ( III )
∂xt +1
From ( III ) and ( I ) , one has
∂ri ( xt , uit , u jt ) ∂g ( uit , u jt ) ∂ri ( xt +1 , uit +1 , u jt +1 )
+β =
0 ( IV )
3 ∂uit ∂uit ∂xt +1 10/15/2018
Hence, for the case of player 1, F .O.C equations and transition functions are
∂r1 ( xt , u1t , u2t ) ∂g ( u1t , u2t ) ∂r1 ( xt +1 , u1t +1 , u2t +1 )
+β =
0, t =
0,1,..., T − 1, (1.a )
∂u1t ∂u1t ∂xt +1
xt +1 = g ( u1t , u2t ) , t = 1,...., T , {u2t }t =0 (1.b )
T
s.t. x0 given, given,
∂r1 ( xT , u1T , u2T ) ∂g ( u1T , u2T ) ∂V01 ( xT +1 )
+β =
0, (1.c )
∂u1T ∂u1T ∂xT +1
∂V01 ( xT +1 )
and β − λT =
0. (1.d )
∂xT +1
For the case of player 2,
∂r2 ( xt , u1t , u2t ) ∂g ( u1t , u2t ) ∂r2 ( xt +1 , u1t +1 , u2t +1 )
+β =
0, t =
0,1,..., T − 1, (2.a)
∂u2t ∂u2t ∂xt +1
( u1t , u2t ) ,
g= {u1t }t =0 ( 2.b )
T
xt +1 t 1,...., T , s.t. x0 given, given,
∂r2 ( xT , u1T , u2T ) ∂g ( u1T , u2T ) ∂V02 ( xT +1 )
+β =
0, ( 2.c )
∂u2T ∂u2T ∂xT +1
∂V02 ( xT +1 )
and β − λT =
0. ( 2.d )
4 ∂xT +1 10/15/2018
 At t = T , given xT , eqs. (1.c ) and ( 2.c ) which are
∂r1 ( xT , u1T , u2T ) ∂g ( u1T , u2T ) ∂V01 ( xT +1 )
+β =
0, (1.c )
∂u1T ∂u1T ∂xT +1

∂r2 ( xT , u1T , u2T ) ∂g ( u1T , u2T ) ∂V02 ( xT +1 )

+β =
0, ( 2.c )
∂u2T ∂u2T ∂xT +1

and transition function

xT +1 = g ( u1T , u2T ) ,
all together det er min e the optimal values of u1T , u2T and xT +1
simul tan eously, so that ,
= i ( xT ) ,
uiT h= =
i 1, 2., and ( u1T , u2T ) f ( xT ) .
xT +1 g=
5 Arayah Preechametta 10/15/2018
 Nash equilibrium (Time consistency)
• Next, for t = T-1, T-2,…., 1, 0., one can jointly use
equations (1.a), (1.b), (2.a) and (2.b) to find
(xt+1, u1t , u2t), for t = T-1, T-2, …, 1, 0.
• Note that each player problem is a recursive
problem.
• In summary, the entire system, formed by (1.a)-(1.d)
and (2.a)-(2.d), can be solved by working
backward, starting from date T.
• The optimal policy functions
uit = hi(xt) , t = 0,1,2,…,T., i = 1,2.
have self-enforcing property.

6 Arayah Preechametta 10/15/2018

 Dominant Player Game (Time Inconsistency)

• Player 1 takes player 2’s action as given.

• Player 2 knows that his action has significant
influence on player 1’s action.
• Let us represent (1.a) by an implicit function as
shown in (3.a), and recall (1.c)
φ ( xt , xt +1 , u1t , u1t +1 , u2t , u=
2 t +1 ) 0.=t 0,1,..., T − 1, (3.a)
∂r1 ( xT , u1T , u2T ) ∂g ( u1T , u2T ) ∂V01 ( xT +1 )
+β =
0. (1.c )
∂u1T ∂u1T ∂xT +1
• The decisions of player 1 are determined by
(3.a) together with the terminal condition,(1.c),
given x0 and the actions of the leader, {u2t}t=0T.

7 Arayah Preechametta 10/15/2018

 For player 1, at t = T , where xT and u2T are given.
We can find optimal value of u1T and xT +1 as functions
of xT and u2T , or in other words,
u1T = h1 ( xT , u2T ) ,
xT +1 = f ( xT , u2T ) ,
by u sin g equation (1.c ) and transition function,
∂r1 ( xT , u1T , u2T ) ∂g ( u1T , u2T ) ∂V01 ( xT +1 )
+β = 0, (1.c )
∂u1T ∂u1T ∂xT +1

Hence, by solving the problem of player 1 recursively,

u sin g ( 3.a ) , (1.c ) and transition function, one get
= 1 ( xt , u2 t ) ,
u1t h= t 0,1, 2,...., T .
8 Arayah Preechametta 10/15/2018
 Next , the leader problem ( player 2 ) is to
T
max
{u2 t }t =0
T ∑ 2 t 1t 2t
β t

t =0
r ( x , u , u ) + β T +1
V02 ( xT +1 ) , 0 < β < 1,

s.t. xt +1 = g ( u1t , u2t ) ,

x0 given, and Eq. (3.a ) and ter min al condition, (1.c ) .

T
L ∑ 2 t 1t 2t
β t

t =0
r ( x , u , u ) + β T +1
V02 ( xT +1 )
T
+ ∑ β t λt'  g ( u1t , u2t ) − xt +1 
t =0
T −1
+ ∑ β t µt' φ ( xt , xt +1 , u1t , u1t +1 , u2t , u2t +1 ) 
t =0

 ∂r1 ( xT , u1T , u2T ) ∂g ( u1T , u2T ) ∂V01 ( xT +1 ) 

+ β θ 'T
+β ,
 ∂u1T ∂u1T ∂xT +1 
9 Arayah Preechametta 10/15/2018
The 1st − order condition with respect to u2t is
 ∂r2 ( xt , u1t , u2t ) ∂r2 ( xt , u1t , u2t ) ∂u1t   ∂g ( u1t , u2t ) ∂g ( u1t , u2t ) ∂u1t 
β + .  + β  + .  λt
 ∂u2t ∂u1t ∂u2t   ∂ u 2t ∂ u 2t ∂u 2t 

 ∂φ ( xt , xt +1 , u1t , u1t +1 , u2t , u2t +1 ) ∂φ ( xt , xt +1 , u1t , u1t +1 , u2t , u2t +1 ) ∂u1t ∂φ ( xt , xt +1 , u1t , u1t +1 , u2t , u2t +1 )  ∂xt +1  ∂u1t 
+β + . +  .  µt
 ∂ u 2t ∂ u 1t ∂ u 2t ∂ x t +1 ∂
 1t  2t 
u ∂u
 ∂φ ( xt −1 , xt , u1t −1 , u1t , u2t −1 , u2t ) ∂φ ( xt −1 , xt , u1t −1 , u1t , u2t −1 , u2t ) ∂u1t 
+ + .  µt −1 =
0,
 ∂ u 2t ∂ u 1t ∂ u 2t 

= t 0,1, 2,..., T − 1. ( 4)
Note that , i.e.,
∂φ ( xt −1 , xt , u1t −1 , u1t , u2t −1 , u2t ) ∂ 2 r1 ( xt −1 , u1t −1 , u2t −1 )
=
∂u2t ∂u1t −1∂u2t
∂g ( u1t −1 , u2t −1 )  ∂ 2 r1 ( xt , u1t , u2t ) ∂ 2 r1 ( xt , u1t , u2t ) ∂u1t 
+β  + . ,
∂u1t −1  ∂ xt ∂ u 2 t ∂ x t ∂u1t ∂u 2 t 
∂g ( u1t −1 , u2t −1 )  ∂ 2 r1 ( xt , u1t , u2t ) ∂ 2 r1 ( xt , u1t , u2t ) ∂u1t 
β  + . 
∂u1t −1  ∂ xt ∂ u 2t ∂ x t ∂u 1t ∂u2t 
Which means that player 2' s action at time t ( or u2t ) can manipulate
player1' s action at time t − 1 ( or u1t −1 ) through player1' s f .o.c.
at time t − 1 ( or eq (1.a ) at t − 1) .

10 Arayah Preechametta 10/15/2018

 And , at t = T ,

 ∂r2 ( xT , u1T , u2T ) ∂r2 ( xT , u1T , u2T ) ∂u1T   ∂g ( u1T , u2T ) ∂g ( u1T , u2T ) ∂u1T 
β + . +β  + .  λT
 ∂u2T ∂u1T ∂u2T   ∂u2T ∂u1T ∂u2T 
 ∂φ ( xT −1 , xT , u1T −1 , u1T , u2T −1 , u2T ) ∂φ ( xT −1 , xT , u1T −1 , u1T , u2T −1 , u2T ) ∂u1T 
+ + .  µT −1
 ∂ u 2T ∂ u 1T ∂u 2T 

 ∂ 2 r1 ( xT , u1T , u2T ) ∂ 2 r1 ( xT , u1T , u2T ) ∂u1T 

+ + . θ
 ∂u1T ∂u2T ∂u1T ∂u1T ∂u2T 

2 ∂ g ( u1T , u2T ) ∂V01 ( xT +1 ) ∂ 2 g ( u1T , u2T ) ∂u1T ∂V01 ( xT +1 ) 

 2
+β  + . . θ
 ∂u 1T ∂u 2T ∂x T +1 ∂u 1T ∂ u 1T ∂ u 2T ∂xT +1 
= 0, ( 5)

11 Arayah Preechametta 10/15/2018

 The 1st − order condition w.r.t. λt is
xt +1 ( u1t , u2t ) ,
g= t 0,1, 2,..., T . ( 6)

The 1st − order condition w.r.t. µt is

φ ( xt , xt +1 , u1t , u1t +1 , u2=
t , u 2 t +1 ) 0,=t 0,1, 2,..., T − 1. (7)

12 Arayah Preechametta 10/15/2018

 The 1st − order condition w.r.t. θ is
∂r2 ( xT , u1T , u2T ) ∂g ( u1T , u2T ) ∂V02 ( xT +1 )
+β =
0, (8)
∂u2T ∂u2T ∂xT +1

13 Arayah Preechametta 10/15/2018

 This system of equations (eqs. (4), (5), (6), (7),
and (8)) must be solved simul tan eously,
and hence, it is not block recursive.

Because the follower ' s actions u1s , for s < t ,

depend on u2t , hence, the leader ' s problem fails
to be recursive.

14 Arayah Preechametta 10/15/2018