Professional Documents
Culture Documents
Chapter 2
1. The economic model we are considering consists of the equations
1 6
yt = E t f yt +1 , xt +1 , ut ,
1 6
xt = g xt -1 , yt -1 ,
x 0 = x 0 , y 0 = y0 .
To linearize this model we must first find a suitable point around which to take a Taylor
series approximation. A good candidate for such a point is a stable steady state of the
non-stochastic version of the model.
1.a We first assume that there exists a pair of values x , y that solves the system of ; @
equations.
1 6,
y = f y, x , u
x = g1 x , y 6
(1.a-1)
f f f
f1 = , f2 = , f3 =
y y , x ,u
x y , x ,u u y , x ,u
(1.a-2)
g g
g1 = , g2 =
y y , x ,u
x y , x ,u
dy = B dy + Cdu ,t +1
dx dx
t
(1.a-3) A t
t t +1
where
1 ,
0 f f , f .
A
g g
B
0 1 C
0
1 2 3
(1.a-4)
1 2
For the linear approximation to be good this steady state should be stable. This implies
that the roots of the matrix:
(1.a-5) J B -1 A ,
Finally, in order for the stochastic model to stay close to the fixed point of the
nonlinear model, we require that the random variable ut should have a small bounded
support.
1.b ; @
The point x , y , u as defined in part (a) is a good point around which to linearize
the model because sequences that begin close to this point will remain close; for that
purpose we needed a stable fixed point. The continued proximity to this point ensures
that the approximation will remain constant as time progresses, i.e. as iteration of the
dynamic system goes on.
1.c To derive a log linear model write the stochastic system in the form:
1 6 1 6
ln yt = ln f yt +1, xt +1 , ut ,
ln1 x 6 = ln g1 x , y 6 .
(1.c-1)
t +1 t t
yf1 xf 2 uf 3
f$1 = f$2 = f$3 =
(1.c-2)
1 6f y, x , u
,
1
f y, x , u
,
6 1 6
f y, x , u
,
yg xg2
f 1 y, x , u 6 1 6
g$ =
1
1
, g$ 2 = ,
g y, x , u
1 6 1 6 1 6 16
ln yt = Et f$1 ln yt +1 + f$2 ln xt +1 + f$3 ln ut + k1,
ln1 x 6 = g$ ln 1 x 6 + g$ ln 1 y 6 + k ,
(1.c-3)
t 2 t -1 1 t -1 2
1.d Since the model delivers two initial conditions, x0 and y0 any steady state (as long
as one exists) is always determinate. Solutions that begin close to a steady state may
diverge away from it and remain bounded. If local solutions diverge from the steady state
then the linear approximation will be inaccurate but local solutions to the dynamic
equations could still be determinate equilibria. To find if any particular solution is a
determinate equilibrium we would need to know more about the non-linear map in order
to know whether a particular isolated trajectory remains bounded.
1 6
a) ln yt =
1
q
3 1 6 8
ln axt q + 1 - a zt q
1 6
b) ln yt =
g
q
3 1 6 8
ln axt q + 1 - a zt q
1 6 1
c) ln yt = ln axt + bzt 6
d) ln1 y 6 = a ln1 x 6 + b ln1 z 6
t t t
1 6 3
e) ln yt = ln zt a xt b + 1 8
The elasticities of these functions are logarithmic partial derivatives:
x
= a ,
q
z
= 11 - a 6 ,
q
a) e fx
y e fz
y
x ,
= ag
q
z ,
= 11 - a 6g
q
b) e fx
y 1/ g
e fz
y 1/ g
x
= a ,
z
= b ,
c) e fx
y e fz
y
d) e fx = a , e fz = b ,
y - 1 y - 1 .
e) e fx = b
y e fz = a
y
A=
4 3"#
!3 $
.
2
(3.a-1) 16
p l = A - lI =
4-l 3
3 2-l
=0
l2 - 6l + 5 = 0
thus l = 1 or l = 5.
To obtain eigenvector x1 of A (associated with the eigenvalue 1 = 1), one solves the
equations: NB (superscripts index eigenvalues and eigenvectors):
(3.a-2) Ax1 = l1 x 1 .
! 3 2 $!x $ !x $
1 1
(3.a-3) 1 1
2 2
%K4 x + 3x = x
1 1 1
&K
1 2 1
(3.a-4)
' 3x + 2 x = x
1
1
1
2
1
2
Since this system is singular we can choose either of the equations to solve for the
elements of x1 . Selecting the second row gives x11 in terms of x21 :
Similarly, to solve for the eigenvector x2 associated with eigenvalue 2 = 5, one solves the
equations:
(3.a-6) Ax 2 = l2 x 2
%K4 x + 3x
2 2
= 5x12
&K
1 2
(3.a-8)
' 3x + 2 x
2
1
2
2 = 5x22
Q=
y z "# = Q -1
(3.a-10)
!z - y$
.
y 2 + z 2 = 1, - 3 y = z ,
(3.a-11)
y = 1 / 2, z = m 3 / 2.
Selecting the values y = 1/2 and z = 3 / 2 gives matrices Q and which diagonalize A:
Q=
-1 / 2 3/2
, L=
"#
1 0 "#
(3.a-12)
! 3/2 1/ 2 $
0 5
.
! $
3.b Using the same procedure as above, the matrices Q and that diagonalize
A=
3 2 "#
! 2 2$
are
Q=
-1 / 3 2 / 3"#, L=
1 0"# .
(3.b-1)
! 2 / 3 1/ 3 $ !0 4 $
16 1 61
p l = 1 / 2 - l 1 / 2 - l - 1 / 16 = 06
(4.a-1) l - l + 3 / 16 = 0
2
l 1 = 1 / 4, l 2 = 3 / 4.
Its eigenvectors solve:
where we have normalized the first element of each eigenvector to equal 1. The matrix Q
of stacked eigenvectors is then:
Q=
1 1 "#
(4.a-4)
!-1 / 4 1 / 4$
Q -1 =
1 / 2 -2"#
(4.a-5)
!1 / 2 2$
A=
1 1 "#1 / 4 0"#1 / 2 "#.
-2
(4.a-6)
!-1 / 4 1 / 4$ ! 0 3 / 4$ !1 / 2 2$
where
z "# = 1 / 2
1
-2 yt "# "# w "# 1 / 2
1
"# "#
-2 ut
!z #$ !1 / 2 $! $ !w #$ !1 / 2 $! $
t t
(4.a-8) 2
, 2
.
t
2 xt t
2 vt
4.b To find the support of x and y first note that the supports of w1 and w2 are
w2
0 0
A E B0
H0
w1
F0
D0 G0 C0
Figure 2.1 shows the support of w1 and w2. The shaded area is the minimal product space
which covers the joint support, shown as the white rectangle. The corners of the joint
support are convex combinations of neighboring extreme points of the minimal product
space. The weights in these convex combinations will play an important role in what
follows and are 1/5 and 4/5 in our case. For instance, E 0 = 4 / 5 A0 + 1 / 5 B 0 .
The supports of the invariant distributions of z1 and z2 are easily found by solving
equations of the following form for zbi .
where li are the eigenvalues and wbi denotes upper and lower bounds of the errors in the
diagonalized system. This is just the approach explained in the textbook on p. 20. We
find:
(4.b-3) z l1 = 10 / 3 , z u1 = 10 / 3 , z l2 = 10 , z u2 = 10 .
This defines the minimal product space that covers the joint support.
Figure 2.2 shows the support of z1, z2. The joint support is the white parallelogram, whose
corners are obtained as convex combinations of the extreme points A, B, C, D, again
using the weights 1/5 and 4/5.
z2
E
A B
z1
D G C
Equation (4.a-8) tells us how we can map (back) from z1 and z2 into y and x.
y 1 1 z1
(4.b-4) x = Qz = 1 / 4 1 / 4 2
z
In order to obtain the support of the invariant distribution of y and x (4.b-4) is applied to
all points of the support of the invariant distribution of z1 and z2.
1 1 10 / 3 20 / 3
(4.b-5) A = =
1 / 4 1 / 4 10 10 / 3
1 1 10 / 3 40 / 3
(4.b-6) B = 10 = 5 / 3
1 / 4 1 / 4
1 1 10 / 3 20 / 3
(4.b-7) C = =
1 / 4 1 / 4 10 10 / 3
1 1 10 / 3 40 / 3
(4.b-8) D = 10 = 5 / 3
1 / 4 1 / 4
The joint support is the white parallelogram, whose corners are obtained as convex
combinations of the extreme points A, B, C, D, again using the weights 1/5 and 4/5.
We get: E = (8,3); F = ( -2.6& ,-2.3& ); G= ( -8,-3); H = ( 2.6& ,2.3& ) .
A'
E
H B'
D' F
G
C'
Figure 2.3 shows the joint support of the invariant distribution of y and x as the white
parallelogram inside the gray one.
A more straightforward1, however less instructive (as judged by the applications we are
interested in in the next chapters), way of calculating the support starts with the linear
economic model
y = A y + u ,
t -1
x x v
t t
(4.b-9)
t t -1 t
y = ( I - A) u ,
x v
-1
(4.b-10)
which can be used to map the extreme points of the support of the error terms into the
support of y and x, to yield the same result as above for E, F, G, H.
5. Equation (2.38) in the text can be rearranged to express future expectations in terms
of current observations of x and y:
1 b 1
yte+1 = yt - xt - ut .
a a a
5.a We can substitute out the unobservable expectations, yt+1e and yte, from equation
(2.40) in the text, to give a difference equation in terms of observable variables yt and xt
and the error terms ut and ut-1:
yte+1 = lyte + (1 - l ) yt
1 b 1 l lb l
(5.a-1) yt - xt - ut = yt -1 - xt -1 - ut -1 + (1 - l ) yt
a a a a a a
yt = q lyt -1 + bxt - blxt -1 + ut - lut -1 ,
1
(5.a-2) q=
1- a 1- l1 6
.
5.b The support of the distribution of x can be obtained from its law of motion,
(5.b-1) xt = gxt -1 + d + vt .
1
This and an error in a previous version of the solutions manual, which has led to the white
parallelograms inside the gray ones, was pointed out to us by Stefan Weber, graduate student at the
European University Institute (EUI), Florence, Italy.
Iterating this equation backwards, (note that < 1), we can express xt as a function of
past values of vts and the initial value, x0:
t t
(5.b-2) xt = g t x0 + dg j -1
+ g j -1
vt - j +1 .
j =1 j =1
The lower and upper bounds of the support of x are calculated by taking the limits of
(5.b-2) assuming that v takes on its lowest and highest values in every period. Thus,
(applying the rule for the summation of a geometric series), we can show that x x , x ,
where:
a
x= ,
1
(5.b-3)
+a
x= .
1