G6215.

001 - Recitation 9: Log-linearization

Contents
1 What is log-linearization? Some basic results 2

2 Application: the baseline RBC model 4

1
G6215.001 - Recitation 9: Log-linearization

1 What is log-linearization? Some basic results

Log-linearization is a first-order Taylor expansion, expressed in percentage terms rather than in levels differ-
ences.
In Economics, since units are not always well defined or consistent, we prefer to think in terms of percentage
deviations from reference values. This reference value will very often be a steady-state of a model that we are
studying.
To fix ideas, consider the following relationship between three level variables (x, y, z):

z = f (x, y)

Taking a first order Taylor expansion around some point (x0 , y0 , z0 ), where z0 = f (x0 , y0 ), expressed in levels,
would consist in writing:

z − z0 = f1 (x0 , y0 )(x − x0 ) + f2 (x0 , y0 )(y − y0 ) + o(kx − x0 k, ky − y0 k)

This is a first-order approximation of the level change in z around z0 , as a function of the level changes x − x0 ,
y − y0 .
Instead, we want to look at the percentage change in z, relative to z0 , as x and y change around (x0 , y0 ). To
do this, define:  
z z − z0
ẑ = log ≈
z0 z0
 
x
x̂ = log
x0
 
y
ŷ = log
y0
Taking a first-order Taylor expansion of:

f (x, y) = f (x0 exp(x̂), y0 exp(ŷ))

around (x̂, ŷ) = (0, 0), we get:

f (x, y) − f (x0 , y0 ) = x0 f1 (x0 , y0 )x̂ + y0 f2 (x0 , y0 ) + o(kx̂k, kŷk)

2
G6215.001 - Recitation 9: Log-linearization

Dividing through by z0 = f (x0 , y0 ), we get:

z − z0 x0 f1 (x0 , y0 ) y0 f2 (x0 , y0 )
ẑ ≈= = x̂ + + o(kx̂k, kŷk)
z0 f (x0 , y0 ) f (x0 , y0 )

We have obtained our log-linear approximation. The coefficients in front of x̂ and ŷ are the elasticities of f with
respect to x and y, evaluated at (x0 , y0 ).
We can generalize this result to many variables.

Result 1 (Log-linearization). The first-order log-linear approximation of the relationship:

z = f (x1 , ..., xn )

around the point (x01 , ..., x0n ), z0 = f (x01 , ..., x0n ) is: X
ẑ = f,i x̂i
i

where: 
xi
x̂i = log
x0i
 
z
ẑ = log
z0
and where:
fi (x01 , ..., x0n )x0i fi (x01 , ..., x0n )x0i
f,i = =
f (x01 , ..., x0n ) z0
is the elasticity of f with respect to its i − th argument, evaluated at (x01 , ..., x0n ).

This corrolary may serve as a shortcut in many computations.

Result 2 (Some useful relations).

• If
z = y α xβ
then
ẑ = αŷ + β x̂

3
G6215.001 - Recitation 9: Log-linearization

• If X
z= α i xi
i

then
X α x0 X α i x0
ẑ = P i i 0 x̂i = i
x̂i
i i α i x i i
z 0

2 Application: the baseline RBC model

The REE of the baseline RBC model can be summarized as the set of following relations:

uc (ct , ht ) = λt (MU = SV of income)

uh (ct , ht ) = −λt wt (MDL = SV of income × Wage)
wt = at ktα ht1−α (MPL = Wage)
ct + kt+1 − (1 − δ)kt = at ktα ht1−α (Ressource constraint)
α−1 1−α


λt = βEt λt+1 1 − δ + αat+1 kt+1 ht+1 (Euler equation)

Using our previous results, it is straightforward to log-linearize these relations around the deterministic
steady-state of the model. In what follows, any elasticity is evaluated at the deterministic steady-state of the
model, and any variable without a time subscript is a steady-state value.
The log-linearization of the first-order condition with respect to consumption follows from the main result
directly:
ucc c uch h
λ̂t = ĉt + ĥt
uc uc
The right-hand side of the FOC with respect to hours can be log-linearized using the first part of the corrolary,
to obtain:
λ̂t + ŵt
while the log-linear form of the left hand side is obtained from the main result. The log-linear form of the FOC
with respect to hours is thus:
uch c uhh h
λ̂t + ŵt = ĉt + ĥt
uh uh

4
G6215.001 - Recitation 9: Log-linearization

The log-linear form of the labour demand schedule follows directly from the first part of the corrolary:

ŵt = ât + α(k̂t − ĥt )

The right hand side of the ressource constraint is log-linearized using the first part of the corrolary as well.
For the left-hand side, we use the second additional part of the corrolary. The log-linear form of the ressource
constraint is then:
c k
ĉt + (k̂t − (1 − δ)k̂t+1 ) = ât + αk̂t + (1 − α)ĥt
y y
where y = c + δk.
The Euler equation is slightly more difficult to log-linearize. Rewrite is as:

λt α−1 1−α


= Et λt+1 1 − δ + αat+1 kt+1 ht+1
β

The left hand side is log-linearized as:

λ̂t
Consider log-linearizing:
α−1 1−α


λt+1 1 − δ + αat+1 kt+1 ht+1
We can use either the first result, or a combination of the additional results, to get the log-linear form:
1h  i
λ 1 − δ + αk α−1 h1−α λ̂t+1 + λαk α−1 h1−α ât+1 + (1 − α)(ĥt+1 − k̂t+1 )


Γ
where:
Γ = λ(1 − δ + αAk α−1 h1−α ) = λ/β
Simplifying and replacing into the expectation, the log-linear form of the Euler equation is:
  
α−1 1−α
λ̂t = Et λ̂t+1 + βαk h Et ât+1 + (1 − α) Et ĥt+1 − k̂t+1

In log-linear form, the REE of the RBC model can therefore be written as:

5
G6215.001 - Recitation 9: Log-linearization

ucc c uch h
λ̂t = ĉt + ĥt (MU = SV of income)
uc uc
uch c uhh h
λ̂t + ŵt = ĉt + ĥt (MDL = SV of income × Wage)
uh uh
ŵt = ât + α(k̂t − ĥt ) (MPL = Wage)
c k
ĉt + (k̂t − (1 − δ)k̂t+1 ) = ât + αk̂t + (1 − α)ĥt (Ressource constraint)
y y
  
λ̂t = Et λ̂t+1 + βαk α−1 h1−α Et ât+1 + (1 − α) Et ĥt+1 − k̂t+1 (Euler equation)

Note that the block of the first two FOC can be inverted to express (ĉt , ĥt ) as a function of (ŵt , λ̂t ). This is the
point of the next homework.