Handout Class 3
Advanced Macroeconomics, Fall 2011 - Alessandro Di Nola
October 13, 2011
1
1.1
The model economy
The representative household problem
We assume that the representative household owns the capital stock, takes sequences of wages
and interest rates as given and in every period chooses consumption and capital to be broght into
tomorrow (intertemporal trade-o). It also faces an intratemporal trade-o between leisure and
consumption. We also assume that households own the rm, i.e. are claimants of the rms prots
t.
Formally, the representative household solves the following problem:
(1
)
X
t
max 1 E0
[log Ct
Ht ]
fCt ;Ht ;Kt+1 gt=0
t=0
s.t.
Ct + Kt+1 = Wt Ht + rt Kt + (1
Ct ; Kt+1
)Kt +
K0 > 0 given
The solution to the households problem is represented by the following equations:
1
Ct
Et
1
[rt+1 + 1
Ct+1
Wt
;
Ct
= Wt Ht + rt Kt + (1
] ;
(1)
=
Ct + Kt+1
(2)
)Kt +
t:
(3)
1.2
The representative rms problem
The representative rms problem is, given a sequence of prices fwt ; rt g1
t=0 ,
(1
)
X
[Yt rt Kt Wt Ht ]
max 1 E0
0 =
fYt ;Kt ;Ht gt=0
t=0
s.t.
t
Yt = zt Kt
log(zt+1 ) = (1
where
Ht
) log(z) + log(zt ) + "t+1
denotes labor-augmenting technological progress. Actually this is not a true dynamic
problem since the variables chosen in period t, (Yt ; Kt ; Ht ) do not aect the constraints nor returns
at later periods. Therefore the above problem is equivalent to
max zt Kt
fKt ;Ht g
Ht
rt K t
W t Ht
and the optimality conditions are
(1
Yt
Ht
Yt
Kt
= Wt ;
(4)
= rt ;
(5)
Yt = zt Kt
Ht
(6)
Note that the constant returns to scale assumption together with perfect competition implies that
the rms prots in period t are
t
= Yt
= Yt
rt K t W t H t
Yt
Kt (1
Kt
Yt
Ht
Ht
= 0
It follows that the representative household, as owner of the rm, does not receive any prots in
equilibrium.
1.3
Denition of Equilibrium
Now we are ready to state formally what we mean by competitive equilibrium.
Denition 1 A competitive equilibrium (C.E.) is a sequence of prices q = fwt ; rt g1
t=0 , allocations
s
for the representative household h = Ct ; Hts ; Kt+1
f=
1
Htd ; Ktd t=0
such that given q:
2
1
t=0
and allocations for the representative rm
1. h solves the household maximization problem;
2. f solves the rm maximization problem;
3. all markets clear, i.e. Hts = Htd , Kts = Ktd , and Ct + Kt+1
(1
)Kt = Yt .
Now we can state the two fundamental theorems of welfare economics. Remember that we are
in a perfect competition environment with a convex production technology, so both theorems apply.
Denition 2 A feasible allocation x = fKt+1 ; Ct ; Ht g1
t=0 is Pareto optimal if there is no other
feasible allocation that yields higher utility to the agent.
Theorem 3 (First Welfare Theorem). A C.E. allocation is Pareto optimal.
Proof. (By contradiction). Assume that x is a CE (with prices q) and assume that it is not
Pareto-optimal, i.e. there exists another feasible allocation x0 6= x that yields higher utility to
the agent. Since, given q; x was maximizing the households utility, it must be that x0 violates
households budget constraint, i.e.
0
Ct0 + Kt+1
> Wt Ht0 + rt Kt0 + (1
0
Moreover, since x0 is feasible it must satisfy Ct0 + Kt+1
(1
)Kt0
)Kt0 = Yt0 But then,rearranging terms
we have that
0
= Yt0
Wt Ht0
rt Kt0 > 0
This contradicts that x is a CE since x does not maximize rms prot (recall that
= 0).
Theorem 4 (Second Welfare Theorem).A Pareto optimal allocation x can always be obtained
as a C.E. for some prices.
Proof. Omitted.
Collecting all equations together the CE allocation of the model can be characterized by the
following non-linear equations:
1
Ct
Et
Ct+1
[rt+1 + 1
Wt
;
Ct
= Wt Ht + rt Kt + (1
] ;
Ct + Kt+1
Yt
(1
)
= Wt ;
Ht
Yt
= rt ;
Kt
(1
(8)
)Kt ;
(9)
(10)
(11)
Yt = zt Kt
Ct + Kt+1
(7)
Ht
(12)
)Kt = Yt ;
(13)
log(zt+1 ) = (1
) log(z) + log(zt ) + "t+1 :
(14)
We have 8 equations and 7 variables: Ct ; Ht ; Kt ; Yt ; Wt ; rt ; zt . Is there something wrong? Actually not, since Walras Law tells us that once the labor and capital market are in equilibrium
also the goods market clears, and therefore the goods market equilibrium equation (13) can be
safely omitted. In fact equation (13) can be obtained by combining (9), (10) and (11), hence it is
redundant.
1.4
Stationary system
The equilibrium characterized above is non-stationary due to labour-augmenting technological
progress. We need to transform the system into a stationary one by using per e ciency unit
variables instead of per capita ones:
1
ct
Et
ct+1
wt
;
ct
= yt + (1
[rt+1 + 1
] ;
(16)
ct + kt+1
yt
= wt ;
(1
)
ht
yt
= rt ;
kt
yt = zt kt h1t
log(zt+1 ) = (1
(15)
)kt ;
(17)
(18)
(19)
(20)
) log(z) + log(zt ) + "t+1 :
where ct = Ct = t ; yt = Yt = t ; kt = Kt = t ; wt = Wt =
variables and ht = Ht .
4
(21)
are the transformed per e ciency unit
1.5
Steady-state
In a stationary steady-state without any shocks the steady-state value of zt is z. The Euler equation
gives
1
e
r=
1+ ;
where e = . From (9) and (11) we can nd k and c as functions of y:
k= y
r
c= 1+
(1
) y
Then the steady-state level of output is
"
y= z
1
1
and
w= c
h=
To simplify the notation, let
1.6
1+ )
1+ .
Linearized system
A rst-order Taylor expansion about the steady-state yields the following system of linear equations:
Et ct+1
rEt rt+1 = ct ;
(22)
0 = wt ct ;
k
k
kt+1 = yt + (1
) kt
y
y
0 = yt kt rt ;
0 = yt
0 =
zt+1 =
ht
(23)
c
ct ;
y
(25)
wt ;
yt + zt + kt + (1
zt + "t+1 :
(24)
(26)
)ht ;
(27)
(28)
1.7
System in matrix form
Now we try to solve the model following the same procedure of handout 2. First we recast equations
(22) to (28) in following matrix form:
AEt s0t+1 = Bs0t +Czt
(29)
where s0t = [kt ; yt ; rt ; wt ; ht ; ct ]T and the matrices A; B and C have dimension 6 6, 6 6 and
6
1 respectively. Then the subsequent step would be to premultiply both sides of (29) by A
provided that A is non singular, and to apply the Jordan decomposition to D
diagonalize the system. In this case we have:
2
0 0
6
60 0
6
6 k
6 y 0
A=6
6
60 0
6
60 0
4
0 0
2
0
0
0
6
6
0
0
0
6
6
k
6(1
0
)y 1
B=6
6
1
1
1
6
6
6
0
1
0
4
1 0
1.8
3
r 0 0 1
7
0
0 0 07
7
7
0
0 0 07
7
7
0
0 0 07
7
0
0 0 07
5
0
0 0 0
3
0
0
1
7
1
0
17
7
c7
7
0
0
y7
7
0
0
0 7
7
1
1
0 7
5
0 1
0
1B
1,
in order to
2 3
0
6 7
607
6 7
6 7
607
7
C=6
6 7
607
6 7
607
4 5
1
Blanchard-Kahn under singularity problem
At this point a serious problem arises: in the baseline RBC model A is not invertible (it has
det = 0). This is a general feature when you include "static variables" like yt or ht in your model.
For the static equations (such as production function or FOC for labor), in fact, the t+1 coe cients
are all zero by construction. This implies that the matrix A is singular.
In the literature on solving linear RA models there are basically three ways to deal with such
non-invertibility problem:
Eliminate all static variables from the model so that you are left with two dynamic equations
for ct and kt (plus the stochastic process for zt )
Rewrite the model the other way:
AEt (xt+1 ) = Bxt
6
Then, even if A is not invertible, B is generally invertible, so you can write
1
xt = B
AEt (xt+1 )
and apply the eigenvalue-eigenvector decomposition to B
1 A.
Use the generalized Schur decomposition.
In the subsequent analysis I will follow the rst approach, namely eliminating all static variables
from the model. For the readers convenience I report down the equilibrium equations of the model:
Et ct+1
rEt rt+1 = ct ;
(30)
wt = ct ;
k
kt+1 = yt + (1
y
yt rt = kt ;
yt
wt
yt + (
Let the matrix form be:
A
(31)
k
) kt
y
c
ct ;
y
(32)
(33)
ht = 0;
(34)
1)ht = zt + kt ;
(35)
zt+1 =
(36)
"
kt+1
ft+1
zt + "t+1 :
=B
" #
kt
ft
+ Czt
where ft = [yt ; rt ; wt ; ht ; ct ]T is the vector of endogenous non-predetermined (or forward-looking,
or control) variables, kt is the only endogenous predetermined (or state) variable and zt is the
exogenous state variable. Clearly A is not invertible. I do the following steps:
1. Divide the variables into a vector of dynamic variables (i.e. indexed by t and t + 1) and static
variables (i.e. indexed only by t)
2. Substitute out the static variables
Let the vector of dynamic variables be s0t = [kt ; ct ]T and the vector of static variables be
ft0 = [yt ; rt ; wt ; ht ]T . The dynamic equations are
Et ct+1
rEt rt+1 = ct
k
kt+1 = yt + (1
y
k
) kt
y
c
ct
y
We can collect them in
0
DEt s0t+1 +FEt ft+1
= Gs0t +Hf 0t
hence
D=
"
0
k
y
#
1
F=
"
r 0 0
and
"
H=
G=
0 0
#
0 0 0 0
1 0 0 0
"
1
) ky
(1
c
y
The static variables are instead:
yt
wt = ct ;
(37)
yt
rt = kt ;
(38)
wt
ht = 0;
(39)
yt + (
1)ht = zt + kt ;
(40)
which can be written as:
Af 0t = Bs0t +Czt
where
2
0
6
61
6
A=6
61
4
1
7
0 7
7
7
1 7
5
1
2
3
2 3
0 1
0
6
7
6 7
61 07
607
6
7
6 7
B=6
7 C=6 7
60 07
607
4
5
4 5
0
1
Hence the matrix form of the RBC model (step 1) is:
0
DEt s0t+1 +FEt ft+1
= Gs0t +Hf 0t ;
(41)
Af 0t = Bs0t +Czt ;
(42)
zt+1 =
(43)
zt + "t+1 :
To implement step 2 delined above, we have now to get rid of the static equations. Premultiplying
both sides of (42) by A
yields:
ft0 = A
Bs0t +A
(44)
Czt
Plugging (44) into (41) delivers:
DEt s0t+1 +F A
BEt s0t+1 +A
CEt zt+1 = Gs0t +H A
Bs0t +A
Czt
Using the property Et zt+1 = zt and rearranging gives:
Et s0t+1 = Ks0t +Lzt
where K = D + FA
G + HA
B and L = D + FA
(45)
1
that (45) is made only of dynamic variables and dynamic equations.
8
HA
FA
C . Note
1.9
Parameter values
The following parameter values are used:
= 0:99
= 0:0045
= 1:0039
= 0:2342
z = 6:0952
= 0:025
= 0:9983
2
e
= 0:00025
It implies that system (45) is:
"
# "
kt+1
1:1373
Et
=
ct+1
0
#" #
0:6802 kt
0:8882
ct
"
#
0:7090
0:1458
zt
You can check this by running the code handout_class3:m (see my webpage). Note that (45) and
(43) can be written together as:
Et xt+1 = Vxt
where
xt =
" #
s0t
zt
= kt ct zt
or, using the parameter values given above,
2
3 2
kt+1
1:1373
6
7 6
6
7
6
Et 4 ct+1 5 = 4 0
zt+1
0
The Jordan decomposition V = M M
iT
V=
"
#
K L
0
32 3
0:6802 0:7090
kt
76 7
7
6
0:8882 0:14585 4 ct 7
5
0
0:9983
zt
gives the following representation of the system:
Et xt+1 =
Let Mij denote the (i; j) element of matrix M
1.
decoupled system:
which can also be written as
(46)
Et x
et+1 =
Et x
ei;t+1 =
ei;t ,
ix
9
xt
Letting furthermore x
et = M
x
et
8i = 1; 2; 3
xt we have the
Clearly if j i j > 1 it is the case that limt!1 x
ei;t =
1 unless you set x
ei;t = 0 8t, which gives you
the equation for the saddlepath. In this model we have
3 2
2
32
3
x
e1;t+1
1:1373
0
0
x
e1;t
7 6
6
76
7
7=6 0
7 6x
7
Et 6
x
e
0:8882
0
e
2;t+1
2;t
5 4
4
54
5
x
e3;t+1
0
0
0:9983 x
e3;t
which implies the restriction
x
e1;t = 0 () M11 kt + M12 ct + M13 zt = 0
from which
ct =
where
ck
M11
M12 ; cz
M13
M12 .
ck kt
cz zt
(47)
To get the solution for capital, simply substitute (47) into the rst
row of (46):
kt+1 = V11 kt + V12 ct + V13 zt
= (V11 + V12
Letting
kk
= V11 + V12
ck
and
"
kz
kt+1
zt+1
= V13 + V12
# "
=
kk
ck ) kt
+ (V13 + V12
cz ) zt :
the state-space form is:
#" # " #
kt
0
kz
+
"t+1 ;
zt
1
cz
or, more compactly,
st+1 =
st +W"t+1 ;
where st = [kt ; zt ]T . Finally it is easy to nd the vector equation relating the static variables to
the states:
ft = Ust ;
where ft = [yt rt wt ht ct ]T and U is a 5 2 matrix.
Exercise 5 Given the parameter values I provided you, compute the matrices
; W and U: (Hint:
run the le handout_class3.m).
Exercise 6 Given the state space form you get from previous exercise, compute impulse responses
to a one-percent technology shock for capital, consumption and the interest rate. Compute also the
impulse responses to a one-percent deviation of capital from its steady-state value (Hint: recall that
if kt = k, then b
kt = 0).
10