1

Romer01a.doc
The Solow Growth Model
Set-up
The Production Function
Assume an aggregate production function:
( ( ) ) ( ( ) ) ( ( ) ) ( ( ) ) [ [ ] ] t L t A t K F t Y , (1.1)
Notation:
Y output
K capital
L labor
A effectiveness of labor (productivity)
Technical change is labor-augmenting (also known as Harrod neutral).
The production function exhibits constant returns to scale:
( ( ) ) AL K cF cAL cK F , ) , ( for all 0 c . (1.2)
Setting AL c / 1 in (1.2) yields the production function in intensive form:
( ( ) ) AL K F
AL AL
K
F ,
1
) 1 , ( (1.3)
Now define:
AL
K
k
AL
Y
y
( ( ) ) ( ( ) ) 1 , k F k f
2
Then (1.3) becomes:
( ( ) ) k f y (1.4)
So output per effective unit of labor is a function of capital per effective unit of labor.
We further assume that:
( ( ) ) 0 0 f
( ( ) ) 0 ' > > k f
( ( ) ) 0 ' ' < < k f .
We also note that ( ( ) ) k f ' is the marginal product of capital. By CRS:
( ( ) )

, ,
_ _




, ,
_ _




AL
K
ALf
AL
K
ALF AL K F 1 , ,
Differentiating with respect to K:
( ( ) )
( ( ) ) k f
AL AL
K
ALF
K
AL K F
'
1
1 , '
,

, ,
_ _




, ,
_ _






We also assume that the Inada conditions hold:
( ( ) )

k f
k
' lim
0
( ( ) ) 0 ' lim

k f
k
.
Note: the Cobb-Douglas production function is one that satisfies the conditions assumed
here.
Evolution of Inputs
Labor and capital grow at constant rates;
( ( ) ) ( ( ) ) t nL t L
&
(1.8)
( ( ) ) ( ( ) ) t gA t A
&
(1.9)
3
where the dot notation refers to a time derivative: ( ( ) )
( ( ) )
dt
t dX
t X
&
.
The constant growth assumption also permits us to describe paths for L and A by:
( ( ) ) ( ( ) )
nt
e L t L 0
( ( ) ) ( ( ) )
gt
e A t A 0 .
By definition, output can be divided into consumption and investment. The fraction of
output going to investment is s, which is assumed to be a constant. The constant rate of
saving is a key feature of this model. Capital also depreciates at the rate . Thus the path
of capital must satisfy the equation:
( ( ) ) ( ( ) ) ( ( ) ) t K t sY t K
&
(1.10)
It is assumed that 0 > > + + + + g n .
Analysis of the Model
Since , / AL K k
( ( ) )
dt
AL
K
d
t k

, ,
_ _




&
Using both quotient and product rules for derivatives:
( ( ) )
( ( ) ) ( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) ) ( ( ) ) [ [ ] ]
2
t L t A
t A
dt
t dL
t L
dt
t dA
t K
dt
t dK
t L t A
t k
1 1
] ]
1 1



+ +

&
(1.11)
( ( ) )
( ( ) )
( ( ) ) ( ( ) )
( ( ) )
( ( ) ) ( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) ) ( ( ) )
( ( ) )
( ( ) ) t L
t L
t L t A
t K
t A
t A
t L t A
t K
t L t A
t K
t k
& & &
&
(1.11a)
1
Now recall (1.10):

1
This can also be written as:
A
A
k
L
L
k
K
K
k
k
& & & &


+ +


+ +


.
4
( ( ) ) ( ( ) ) ( ( ) ) t K t sY t K
&
(1.10)
Substituting (1.10) and the given growth rates for A and L into (1.11a):
( ( ) )
( ( ) ) ( ( ) )
( ( ) ) ( ( ) ) ( ( ) ) t gk t nk t k
t L t A
t Y
s t k ) (
&
(1.12)
( ( ) ) ( ( ) ) ( ( ) ) ( ( ) ) t k g n t k sf t k + + + + ) (
&
(1.13)
Equation (1.13) is the key equation in the Solow model. It describes how the capital stock
evolves over time. The first term on the RHS is the amount of investment per worker.
The second term is the amount of investment that would be needed to keep k constant.
Two diagrams are useful here: Figure 1.2 and Figure 1.3 in Romer. (Note that these
diagrams are drawn to satisfy the Inada conditions).
Balanced Growth Path
If
*
k k < < , then 0 > > k
&
and if
*
k k > > , then 0 < < k
&
. Therefore, over time k will approach
*
k .
When
*
k k , 0 k
&
. At such a point L and A are growing at constant rates n and g and K
is growing at rate g n + + . Since both capital and effective labor are growing at rate g n + + ,
by CRS, output is also growing at rate g n + + .
L
K
and
L
Y
are growing at rate g. We see
that each variable in the model grows at a steady rate when 0 k
&
.
At such a point, the economy is on a balanced growth path. On this path, the growth rate
of output per worker depends only technical progress.
Some Stylized Facts
Growth rates of labor, capital, and output are constant.
Output and capital grow at about the same rate, so the capital-output ratio is roughly
constant.
Output and capital grow faster than labor, so output per worker and capital per worker are
rising.
These facts are compatible with the Solow growth model.
Comparative Dynamics
5
We will consider changes in the savings rate, the key model parameter. Policymakers
might be able to influence this parameter by changing tax rates or the amount and/or
composition of government spending.
Suppose the savings rate increases. This causes the ( ( ) ) k sf line to shift upward.
Investment exceeds its break-even level, so k begins to increase and does so until it
reaches a new higher level of
*
k . During the transition, output per worker grows faster
than A, because of the rise in k. So a permanent increase in the savings rate produces a
temporary increase in the growth of output per worker.
The initial impact of the increase in s is to reduce c. Since output is not initially change,
the added saving must reduce consumption. As k grows, y grows, and c grows. However,
when the new balanced growth path is reached, it is questionable whether c is higher
than before (illustrate via diagram).
The golden-rule level of
*
k occurs at the savings rate that maximizes c (illustrate via
diagram).
Calibration Experiments
When the savings rate changes, the economy moves to a new balanced growth path. But
how much does a change in s affect y, and how quickly?
With plausible functional forms and parameter values, Romer concludes that:
A significant (10%) increase in the saving rate has a modest (5%) effect on output
along the balanced growth path.
Following a change in the savings rate, convergence to a new balanced growth
path is slow. Half of the movement toward the new growth path is accomplished
in 18 years.
These results seem to imply that it is difficult to increase an economys standard of living
by way of higher saving.
Implications
The results noted above suggest that variations in s and k will probably account for little
of the variation in growth and output across countries. Instead, variations in the
productivity parameter, A, must account for most of the variation in output. In the Solow
model, variations in A are not explained.
What is A?
The stock of knowledge?
Education of the labor force (human capital)?
6
Quality of infrastructure (public capital)?
Institutions regulating and enforcing property rights?
Solow on Growth Accounting
Also see the Solow Handout (in lieu of the discussion on pp. 26-27 in Romer).
Recall the production function:
( ( ) ) ( ( ) ) ( ( ) ) ( ( ) ) [ [ ] ] t L t A t K F t Y , (1.1)
Differentiate with respect to time:
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) ) t A
t A
t Y
t L
t L
t Y
t K
t K
t Y
t Y
& & & &


+ +


+ +


, (1.28)
where:
A
AL
Y
L
Y

, ,
_ _








and L
AL
Y
A
Y

, ,
_ _








.
Divide on both sides by ( ( ) ) t Y and rearrange to get:
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) ) t A
t A
t A
t Y
t Y
t A
t L
t L
t L
t Y
t Y
t L
t K
t K
t K
t Y
t Y
t K
t Y
t Y
& & & &


+ +


+ +


(1.29a)
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) ) t R
t L
t L
t
t K
t K
t
t Y
t Y
L K
+ + + +
& & &
(1.29b)
Note that the s are elasticities of output with respect to the indicated inputs.
Subtracting ( ( ) ) ( ( ) ) t L t L /
&
from each side, and noting that 1 + +
L K
, we obtain:
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) ) t R
t L
t L
t K
t K
t
t L
t L
t Y
t Y
K
+ +
1 1
] ]
1 1




& & & &

The rate of growth of the output/labor ratio depends on the rate of growth in the
capital/labor ratio and a residual, the Solow residual. Everything in the equation above is
fairly easily measured, except for ( ( ) ) t R . But that means that ( ( ) ) t R can be determined as a
residual. One can use this equation to decompose growth into portions due to changes in
capital per worker and changes due to technical change.
7
Empirical Results from Solows paper:
For the period 1909-29: Technical change accounts for 0.90 percentage points of per
capita growth per year.
1930-1949: Technical change accounts for 2.25 percentage points of per capita growth
per year.
Technical change accounts for 7/8 of per capita growth; increased capital per worker
accounts for just 1/8.
Technical change appears to be highly variable from year to year.
Economic Convergence
Why should we expect convergence across economies (i.e. economies should be
similar in terms of income, capital, etc.)?
Each country should approach its balanced growth path.
Countries with lower capital stocks will have a higher marginal product of capital,
and will attract investment.
The ability to emulate best technology should allow convergence toward a
common level for A.
Empirical evidence on the hypothesis of convergence is partly contradictory, but it seems
clear that not all poor countries are in a process of catching up.
Romer02a.doc
Introduction
We next consider a growth model that differs from that of Solow in an important way.
The savings rate is now endogenously determined as the result of choices made by
maximizing households, rather than exogenously imposed. The savings rate also need
not be constant.
Assumptions
Assumptions About Firms
There are many identical firms.
Each firm produces subject to the CRS production function: ( ( ) ) AL K F Y , .
Firms hire workers and rent capital in competitive markets; they also sell output in a
competitive market.
Firms are profit maximizers.
Profits accrue to households as income.
A grows exogenously at rate g.
Households
There are many households.
The size of each household grows at rate n.
Each member of the household supplies one unit of labor at every point in time. (The
number of people is equal to the quantity of labor).
Households rent all capital they own to firms.
Each household has initial capital holdings ( ( ) ) H K / 0 , where ( ( ) ) 0 K is the initial amount of
capital in the economy and H is the number of households.
There is no depreciation.
Households divide income between consumption and saving in order to maximize
lifetime utility, given below:
( ( ) ) ( ( ) )
( ( ) )
dt
H
t L
t C u e U
t
t





0

(2.1)
Here is an individuals discount rate. Note that C indicates consumption per member
of the household, while H L/ is the number of members (laborers) per household (i.e.
total labor divided by the number of households). So household utility increases with
consumption per member and with the number of members.
We assume a particular for the instantaneous utility function u:
( ( ) ) ( ( ) )
( ( ) )

1
1
t C
t C u , 0 > > , ( ( ) ) 0 1 > > g n .
(This is a constant relative risk aversion utility function).
Parameter determines the willingness of a household to shift consumption between
periods: a small means a household is more willing to tolerate shifts in consumption
between periods. The assumption that ( ( ) ) 0 1 > > g n insures that the household
cannot obtain infinite lifetime utility.
Behavior
Behavior of Firms
Firms employ capital and labor, pay them their marginal products, and earn zero profits
(under perfect competition).
The marginal product of capital is given by ( ( ) ) k f ' , where ( ( ) ) k f is the intensive
production function. With no depreciation, the real rate of return on capital (real rate of
interest) is given by
( ( ) ) ( ( ) ) ( ( ) ) t k f t r ' (2.3)
We can also show that the real wage per effective unit of labor is:
( ( ) ) ( ( ) ) ( ( ) ) ( ( ) ) ( ( ) ) ( ( ) ) t k f t k t k f t w ' .
To show the last result:
( ( ) ) AL K F Y ,

, ,
_ _




AL
K
ALf Y
Now let
*
L AL

, ,
_ _




*
*
L
K
f L Y
( ( ) )
2
*
*
*
*
'
L
K
L
K
f L
L
K
f
AL
Y

, ,
_ _




, ,
_ _






( ( ) ) ( ( ) ) k kf k f
AL
Y
'


Since the marginal product of labor (not effective labor) is ( ( ) ) AL AL K F A / , , a workers
labor income at time t is ( ( ) ) ( ( ) ) t w t A .
The Households Maximization Problem
Each household takes r and w as given.
Define ( ( ) ) t R :
( ( ) ) ( ( ) )

d r t R
t



0
( ) t R
e

serves to discount future flows when the interest rate is not constant over time. This
is a generalization of the case of a constant interest rate, r. In that case multiplication by
rt
e

accomplishes the discounting.

To consider why the generalization works, consider a series of short (1-period)
intervals. Within each of these short intervals, the interest rate is constant.
Repeatedly apply the constant interest rate formula.
Example: At time zero you have $1. At the end of period 1, you have
t r
e
1
1 $ . At
the end of two periods you have
2 1
1 $
r r
e e . At the end of three periods, you have
3 2 1 3 2 1
1 $
r r r r r r
e e e e
+ +
= , etc. For very short periods, the summation is replaced by the
integral.
Since the household has ( ) H t L / members, its labor income is ( ) ( ) ( ) H t L t w t A / and
consumption is ( ) ( ) H t L t C / . The households budget constraint is therefore:
( )
( )
( ) ( )
( )
( ) ( )
( )


=

+
0 0
0
t
t R
t
t R
dt
H
t L
t w t A e
H
K
dt
H
t L
t C e (2.5)
Define ( ( ) ) t c as consumption per effective unit of labor. This can be written as
consumption per person, ( ( ) ) t C , divided by the amount of effective labor per person, ( ( ) ) t A .
So ( ( ) ) ( ( ) ) ( ( ) ) t A t C t c / . Similarly, ( ( ) ) 0 k is the initial capital stock per effective unit of labor,
so that ( ( ) ) ( ( ) ) ( ( ) ) ( ( ) ) 0 0 / 0 0 L A K k .
Using the relationships described above, rewrite (2.5) in terms of consumption and labor
income per effective unit of labor:
( ( ) )
( ( ) )
( ( ) ) ( ( ) )
( ( ) )
( ( ) ) ( ( ) )
( ( ) )
( ( ) ) ( ( ) )
( ( ) )







+ +
0 0
0 0
0
t
t R
t
t R
dt
H
t L
t w t A e
H
L A
k dt
H
t L t A
t c e (2.6)
Next, substitute ( ( ) ) ( ( ) ) ( ( ) ) ( ( ) )
( ( ) )t g n
e L A t L t A
+ +
0 0 into (2.6):
( ( ) )
( ( ) )
( ( ) ) ( ( ) )
( ( ) )
( ( ) )
( ( ) ) ( ( ) )
( ( ) )
( ( ) )
( ( ) ) ( ( ) )
( ( ) )



+ +

+ +



+ +
0 0
0 0 0 0
0
0 0
t
t g n
t R
t g n
t
t R
dt
H
e L A
t w e
H
L A
k dt
H
e L A
t c e ,
then divide each side by ( ( ) ) ( ( ) ) : / 0 0 H L A
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )



+ + + +



+ +
0 0
0
t
t g n t R t g n
t
t R
dt e t w e k dt e t c e . (2.7)
Romer also shows that the budget constraint can be written in an alternative way:
( ( ) )
( ( ) )
. 0 lim


H
s K
e
s R
s
It is also useful to rewrite the households objective function in terms of consumption per
effective unit of labor. Recall that ( ( ) ) ( ( ) ) ( ( ) ) t A t C t c / . We can rewrite the utility function:
( ( ) ) ( ( ) ) ( ( ) ) [ [ ] ]






1 1
1 1
t c t A t C
( ( ) ) ( ( ) ) ( ( ) ) [ [ ] ]

1
0
1
1
1
t c e A t C
gt
( ( ) )
( ( ) )
( ( ) )
( ( ) )

1
0
1
1
1 1
1
t c
e A
t C
gt
(2.13)
Recall the households objective function:
( ( ) ) ( ( ) )
( ( ) )
dt
H
t L
t C u e U
t
t





0

(2.1)
Substitute (2.13) and ( ( ) ) ( ( ) )
nt
e L t L 0 into (2.1):
( ( ) )
( ( ) )
( ( ) ) ( ( ) )
dt
H
e L t c
e A e U
nt
gt
t
t
0
1
0
1
1 1
0
1 1
] ]
1 1

( ( ) )
( ( ) )
( ( ) )
( ( ) )
dt
t c
e e e
H
L
A U
nt gt
t
t

1
0
0
1
1
0
1
( ( ) )
dt
t c
e B U
t
t

1
1
0
(2.14)
where:
( ( ) )
( ( ) )
H
L
A B
0
0
1

and
( ( ) )g n 1 .
Romer02b.doc
Household Behavior: the Maximization Problem
The households problem is to choose a path for ( ( ) ) t c to maximize lifetime utility subject
to the budget constraint:
Recall the objective function:
( ( ) )
dt
t c
e B U
t
t

= =


= =


1
1
0
(2.14)
Also recall the budget constraint:
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )


= =
+ + + +

= =

+ +
0 0
0
t
t g n t R t g n
t
t R
dt e t w e k dt e t c e . (2.7)
Form the Lagrangean:
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) ) ( ( ) )
( ( ) )
( ( ) )






+ + + +

= =
+ +

= =


= =
+ +


= =


dt e t c e dt e t w e k dt
t c
e B L
t g n
t
t R
t
t g n t R
t
t
0 0
1
0
0
1
*

The first order condition for an individual ( ( ) ) t c is:

1
( ( ) )
( ( ) ) ( ( ) )t g n t R t
e e t c Be
+ +
= =

(2.16)
Take logs on each side:
( ( ) ) ( ( ) ) ( ( ) )t g n t R t c t B + + + + = = ln ln ln
Differentiate with respect to t on each side:
( ( ) )
( ( ) )
( ( ) ) ( ( ) ) g n t r
t c
t c
+ + + + = =
&

(In taking the derivative above, recall that ( ( ) ) ( ( ) )

d r t R
t

= =
= =
0
0)).

1
As Romer notes, this step is somewhat informal. See Romer, p. 44, footnote 7.
Finally, solve for
( ( ) )
( ( ) ) t c
t c&
to get:
( ( ) )
( ( ) )
( ( ) )

= =
g n t r
t c
t c&
or, using ( ( ) )g n = = 1 :
( ( ) )
( ( ) )
( ( ) )

g t r
t c
t c
= =
&
(2.19)
Equation (2.19) is the Euler equation for this problem.
The Euler equation describes how consumption must behave over time for a given ( ( ) ) 0 c ;
( ( ) ) 0 c must be chosen so as to satisfy the budget constraint.
Essentially, optimality requires that it not be possible for a consumer to gain by way of
making a small shift of consumption from one time period to another, while satisfying the
budget constraint. The Euler equation is an implication of that optimality requirement
Dynamics of the Economy
Dynamics for c
Since all households are the same, condition 2.19 holds for the entire economy, not just a
single household. All firms are identical also under our assumptions the intensive
production function for a single firm is identical to that for the entire economy.
Recall that ( ( ) ) ( ( ) ) ( ( ) ) t k f t r ' = = and rewrite (2.19):
( ( ) )
( ( ) )
( ( ) ) ( ( ) )

g t k f
t c
t c
= =
' &
(2.22)
By (2.22). we know that 0 = = c& when ( ( ) ) g k f + + = = ' . Let
*
k denote the capital stock
such that the latter condition holds.
When
*
k k > > , ( ( ) ) g k f + + < < ' , so c& is negative.
When
*
k k < < , ( ( ) ) g k f + + > > ' , so c& is positive.
Fig 2.1 in Romer, p. 46, Summarizes the results noted above.
Dynamics for K
Capital accumulation is governed by an equation similar to that in the Solow model
(except that the savings rate need not be constant and depreciation has been assumed to
equal zero):
( ( ) ) ( ( ) ) ( ( ) ) ( ( ) ) ( ( ) ) ( ( ) ) t k g n t c t k f t k + + = =
&
(2.23)
Recall Figure 1.6, Romer, p. 19. For any value of k, k
&
will be equal to zero so long as
consumption is equal to the vertical distance between the ( ( ) ) k f curve and the ( ( ) )k g n + +
line (remember that deprecation is zero). This is so because actual saving is equal to
break-even saving.
So the 0 = = k
&
locus will have an appearance like that shown in figure 2.2, Romer, p. 47.
Above this curve, consumption is higher, hence saving is lower, and . 0 < < k
&
Below this
curve, consumption is lower, hence saving is higher, and . 0 > > k
&
See Figure 2.3, Romer p. 48, for a diagram summarizing the dynamics of both c and k.
Point E indicates a balanced growth path, with both c& and k
&
equal to zero.
Note that the balanced growth path is to the left of the golden rule level for k. This
must be true, as is argued below:
Recall that in this model 0 = = c& when ( ( ) ) g k f + + = =
*
' (again see (2.22)). At the
golden rule point, ( ( ) ) g n k f
GR
+ + = = ' . Concavity of the intensive production
function implies that if
GR
k k > >
*
, then
g n g + + < < + +
or
( ( ) )g n < < 1
or
( ( ) ) 0 1 < < g n .
However, in order to insure that lifetime utility does not diverge, we initially
assumed in (2.2) that :
( ( ) ) 0 1 > > g n .
We conclude that it is not the case that
GR
k k > >
*
; instead
GR
k k < <
*
, as Figure 2.3
shows.
An Initial Value for c
In solving the maximization problem, an individual has a given initial value for k, and
must choose an initial value for c, as well as its subsequent path.
Some possible paths are illustrated in Figure 2.4, Romer p. 49. The trajectories shown
satisfy equations (2.22) and (2.23). To further limit the possible trajectories, we see
which paths also satisfy the budget constraint and the requirement that the capital stock
be positive.
First note that at a point like F, the trajectory ends at the balanced growth point, E.
Suppose instead that we choose an initial level of consumption that is higher, as at point
C. From this point, the trajectory eventually leads to a negative value for k, violating a
requirement of the model.
Alternatively, suppose we start at a lower consumption level, as at point D. From this
point, the economy eventually exceeds the golden rule value for k, and from there follows
a path with c declining, even as k is increasing. Beyond this point, the present value of
the future income stream strictly exceeds the present value of the future consumption
stream. This implies that there would have been room to increase consumption and
utility, so that such a path must not have been optimal.
For an economy starting at a value of k such that
*
k k > > there will similarly be a unique
path leading back to the balance growth path.
1
Romer02c.doc
Balanced Growth in the Ramsey Neoclassical Growth Model
Welfare
Diagram 2.5 in Romer illustrates movement toward the balanced growth path.
Convergence to this point and the subsequent balanced growth path are optimal. A social
planner would choose the same path as utility maximizing individuals competitive
market outcomes are optimal in the absence of externalities, etc.
Properties of the Balanced Growth Path
Along the balanced growth path, consumption and capital per effective unit of labor are
constant (as the diagram directly shows). If the capital to effective labor ratio is constant,
then output per effective unit of labor must also be constant. Since y and c are constant,
the savings rate ( ( ) ) y c y s / = = will also be constant. This implies that on the balanced
growth path, this model economy looks no different than the Solow economy.
Approaching the balanced growth path, the savings rate is not constant, however.
In the Ramsey model, the capital stock is less than the golden rule capital stock (that
would maximize consumption per effective unit of labor). Recall that we earlier showed
that
GR
k k < <
*
, which establishes this result. This is consistent with utility maximization,
since current consumption is valued more highly than future consumption because of
discounting. I.e., if we were on the golden rule balanced growth path, it would be worth
it (in terms of grand utility) to give up some future consumption for current
consumption.
Comparative Dynamics
Suppose falls (the utility value of future consumption will be higher). According to
equation (2.22):
( ( ) )
( ( ) )
( ( ) ) ( ( ) )

g t k f
t c
t c
= =
' &
. (2.22)
Consider the 0 = = c& locus. If falls, then ( ( ) ) ( ( ) ) t k f ' must fall, which means k must be
lower. So the 0 = = c& locus shifts to the right. Now if consumption (at the time of the shift)
were to remain unchanged we would head up and to the left in our phase diagram, away
from the new balanced growth path. Instead, consumption must immediately fall, to put
the economy on a path which will converge to the balanced growth path. Along the path
to balanced growth, k will gradually increase.
2
Skip the section on the Rate of Adjustment and the Slope of the Saddle Path
Government Purchases
Assume that the government buys goods at the rate ( ) t G per unit of effective labor, and
that government purchases do not affect the utility of private good consumption or future
output. Government spending is financed by lump-sum taxes. Initially assume that
current spending must be financed by current taxes,; i.e., ( ( ) ) ( ( ) ) t G t T = = . Here ( ( ) ) t T is taxes
per effective unit of labor.
Equation (2.23) now becomes:
( ) ( ) ( ) ( ) ( ) ( ) ( ) t k g n t G t c t k f t k + =
&
(2.38)
The Euler equation will hold as it did before (it was derived without using either 2.23 or
the budget constraint). The budget constraint (see equation (2.9)) is changed however. It
becomes:
( )
( )
( )
( )
( )
( ) ( ) [ ]
( )


=
+ +

+
0 0
0
t
t g n t R t g n
t
t R
dt e t G t w e k dt e t c e . (2.39)
Suppose that the economy initially is on a balanced growth path with ( )
L
G t G = . Then
suppose that ( ) t G suddenly and permanently increases to ( )
H
G t G = . Government
spending does not affect the Euler equation, from which the 0 = c& locus was derived. So
the 0 = c& locus is unchanged. At any level of k, c is now lower by exactly the amount of
the increase in G on the 0 = k
&
locus. Initial consumption must adjust to get the economy
back on a path towards balanced growth in this case, consumption simply falls by the
amount of the increase in G, and we are immediately back on the balanced growth path.
Since k is unchanged, the marginal product of capital and the real interest rate are also
unchanged.
Now suppose that instead of a permanent increase in G, there is a sudden temporary
increase in G (of known duration). As we have noted before, the 0 = c& locus is
unchanged, but the 0 = k
&
locus falls. If c fell by the full amount of the change in G, then
we would stay at that point until G reverted to its lower level, at which time it would have
to shift back up. But the discontinuous upward shift cannot be optimal there can be no
anticipated discontinuity in consumption because it would imply a discontinuity in
marginal utilities (which would lead to a desired reallocation of consumption from one
period to the next).
Instead, c will fall by less than the full amount of the change in G. At this point, the
economy is above the 0 = k
&
line, so k falls. As k falls, we move to the left of the 0 = c&
locus, so c rises (we are moving northwest in Fig. 2.9a). When G returns to its normal,
3
low level we must be back on the path of convergence to the original balanced growth
path.
So, a temporary increase in G causes a downward jump in consumption (less than the
increase in G). While the higher G is in effect, capital is decumulating. Consumption
immediately begins a gradual return to its original level, however. When G reverts to its
normal level, consumption is still below its balanced growth path level, and capital
begins to accumulate.
During the period of high G, k is declining, so the marginal product of capital is
increasing, as is the interest rate. Once G returns to normal, k begins to increase, and the
interest rate falls. We would therefore expect high interest rates in wartime (when
government spending is temporarily higher than normal).
Bond and Tax Finance
The Government Budget Constraint
We will now permit the possibility that a government might finance current spending
either with taxes or bonds (borrowing). The governments budget constraint says that the
present value of its spending stream must be less than or equal to initial government
wealth plus the present value of its tax revenues. We will assume that the constraint holds
as an equality. So the budget constraint can be written as:
( ( ) )
( ( ) )
( ( ) )
( ( ) ) ( ( ) ) [ [ ] ] ( ( ) ) ( ( ) ) ( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) ) ( ( ) )


= =

= =
+ + + +
+ + = =
0 0
0 0 0 0 0 0 0
t t
t g n t R t g n t R
dt L A e t T e L A b dt L A e t G e
In the above equation, ( ( ) ) 0 b is the initital level of the outstanding stock of government
debt per effective unit of labor. This represents negative wealth for the government.
Dividing by ( ( ) ) ( ( ) ) 0 0 L A on each side yields:
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )


= =

= =
+ + + +
+ + = =
0 0
0
t t
t g n t R t g n t R
dt e t T e b dt e t G e (2.42)
Romer notes that the budget constraint (as an equality) could also be written as:
( ( ) ) ( ( ) )
( ( ) ) 0 lim = =
+ +

s b e e
s g n s R
s
This says that in the limit, the present value of the governments outstanding debt must
approach zero. (The debt itself could converge to a positive number, but the present value
must go to zero). This implies that if I buy a bond, I expect a stream of future payments
from the government equal in present value to what I paid for the bond. (If this were not
true, I would not buy the bond).
4
Households also have a budget constraint. Modify equation (2.7) to include taxes:
( ( ) )
( ( ) )
( ( ) )
( ( ) ) ( ( ) )
( ( ) )
( ( ) ) ( ( ) ) [ [ ] ]
( ( ) )


= =
+ + + +

= =

+ + + +
0 0
0 0
t
t g n t R t g n
t
t R
dt e t T t w e b k dt e t c e (2.44)
Now rewrite (2.42), isolating the present value of taxes on one side:
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )


= =
+ +

= =
+ +
+ + = =
0 0
0
t
t g n t R
t
t g n t R
dt e t G e b dt e t T e (2.42)
Substitute into (2.44) to get:
( ( ) )
( ( ) )
( ( ) )
( ( ) ) ( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )


= =
+ +

= =
+ + + +

= =

+ + + +
0
0 0
0 0 0
t
t g n t R
t
t g n t R t g n
t
t R
dt e t G e b dt e t w e b k dt e t c e
or, canceling the ( ( ) ) 0 b terms,
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) )
( ( ) ) ( ( ) )
( ( ) )
( ( ) )


= =
+ +

= =
+ + + +

= =

+ +
0
0 0
0
t
t g n t R
t
t g n t R t g n
t
t R
dt e t G e dt e t w e k dt e t c e (2.45)
Notice that government spending enters the budget constraint above, but the quantity of
bonds at any time does not. In fact, the latter equation is identical to (2.39), which was
the budget constraint in the case where the government never issued bonds.
( )
( )
( )
( )
( )
( ) ( ) [ ]
( )


=
+ +

+
0 0
0
t
t g n t R t g n
t
t R
dt e t G t w e k dt e t c e . (2.39)
The implication is that only the path of government purchases, and not the path of the
taxes that finance those purchases, affects the economy. This is the Ricardian
Equivalence Theorem.
Ricardian Equivalence
Consider the intuition for the result noted above. For a given government spending plan,
a government can finance todays spending (say $100) by collecting taxes or issuing
bonds. Suppose that the government levies no current tax, but instead issues $100 of
bonds. This implies that households now face a future tax liability which has a present
value equal to $100 -- bondholders must be paid a future stream equal in present value to
the amount they pay for the bonds. Households see no change in wealth they are freed
of the need to pay $100 in taxes, but they are simultaneously burdened with a future
liability of equal present value. Households also have no reason to alter their
consumption plan when a tax is replaced by bond financing a household can maintain
consumption as before. The funds freed up by the reduction in taxes can be used to buy a
bond. When higher taxes must be paid in the future, returns from that bond will provide
5
exactly the funds required to pay the taxes. So the choice of bond versus tax finance has
no affect on the real economy.
Clearly, this result is contrary to the implications of conventional Keynesian models,
which imply that the decisions to alter taxes will affect aggregated demand, output
consumption, interest rates, etc.
Does Ricardian Equivalence Hold?
Our model has made some strong assumptions, which might affect the Ricardian
equivalence result:
Infinitely lived individuals. If individuals can die, they can avoid future taxes.
However, if individuals care enough about descendants to leave bequests, the
governments financing decisions are again irrelevant. The government may cut
my taxes, which apparently redistributes from my child to me. But if I am
planning to leave a given bequest to my child, I can still do so. I can just by a
bond with my current tax savings, and leave the bond to my child when I die.
Proceeds from the bond will just pay his added future tax liability.
Even if individuals are not so calculating, it will often be the case that quantitative
deviations from Ricardian equivalence will be small. Suppose that the
government gives me a one-time tax break of $20,000. Even if I believe that this
is wealth (I dont worry about the future tax burden), I am not likely to spend it all
at once. The optimal consumption plan will typically result that I increase
consumption slightly in many future periods. So a big tax cut will not have big
impacts on current spending.
Liquidity Constraints.
When the government cuts your taxes today (which you must pay back in higher
taxes in the future), they are giving you a loan. Our model supposes that
individuals already had the option to borrow at the prevailing interest rate, so the
governments beneficence is not appreciated.
In the real world, most individuals cannot borrow at the same interest rate as the
government. Given an opportunity to borrow at a lower rate, I might borrow
more, so my consumption is affected by the governments decision to tax or
borrow for me.
There are two counter-arguments to favor the Ricardian position. First, if my
borrowing rate were lower, it is not obvious that I would borrow a lot more
(enough to give a notable Keynesian kick to the economy).
Second, liquidity constraints are not exogenous. When the government borrows
more on your behalf, private lenders may become less likely to lend to you
6
(because of your future tax liability you are seen as a riskier prospect) so that the
impacts of the initial government action are reduced.
Taxes are Not Lump-Sum
When taxes are not lump-sum, there will clearly be some real world effects
associated with changing the time path of tax collections. E.g., people will
reallocate work effort to avoid taxes.
One effect of imposing income taxes instead of lump sum taxes is that uncertainty
about future (after-tax) income is reduced, which may lead to higher current
consumption.
Imposing an income tax may also affect the importance of liquidity constraint
effects. My riskiness as a potential borrower is lessened under an income tax. I.e.,
if my income turns out to be low, my future tax burden will also be low so the
future tax liability will not have a big negative impact on my ability to borrow in
the private sector today.
Non-Optimizing Behavior
Traditional Keynesian models did not posit optimizing behavior the
consumption function could instead reflect rule-of-thumb behavior. Individuals
naively ignore future tax liabilities and increase consumption in response to a tax
cut.
1
Romer04_Notation.doc

Y Aggregate Output

C, I, G, Aggregate Consumption, Investment, Government Spending

Depreciation rate

U Utility (grand utility function)

w Real wage

r Real rate of interest

Utility function discounting parameter

N Population (grows at rate n)

H Number of Households (constant)

c
N
C
c (Not
AN
C
)

l
N
L
l

2
Romer04a.doc

Business Cycle Facts

Business cycles do not exhibit a simple, regular, cyclical pattern. Both amplitude and
duration of cycles are irregular.

Fluctuations in output are distributed unevenly over its components.

Inventory investment is a very small fraction of total output, but it accounts for
1/3 of the shortfall in output in recessions. Investment in general is has high
amplitude and is procyclical.

Over rather long periods, output tends to be slightly above its trend path; over rather short
periods, output tends to be sharply below its trend path.

Fluctuations before the Great Depression and after WWII are similar in character.

Cycles of the magnitude experienced in the Great Depression in the U.S. are unparalleled
in the remainder of the historical record.

Some evidence on co- movements. In a recession:

Employment falls.
Unemployment rises.
Length of average workweek falls.
Declines in employment and hours are small relative to the decline in output (so
measured productivity falls).
There is disagreement about the behavior of inflation over the business cycle.
The real wage falls slightly.
Nominal interest rates and the real money supply fall in recessions.

Theories of Fluctuations

We have already studied Keynesian, New Classical, and New Keynesian Models of
Fluctuations.

Oddly, these models do not seem to depart directly from the neoclassical growth model
(in which individuals maximize utility over long time horizons, firms maximize profits,
markets are competitive, and market imperfections are absent).

We will now study real business cycle models, which do take neoclassical growth theory
as a point of departure.

How will the Ramsey model be modified?

3
We must have random shocks hit the economy (or else we will approach a
balanced growth path). We will emphasize real shocks: shifts to technology
(hence real business cycle models).

We must also permit labor supply to be endogenous (to reflect the fact that labor
is procyclical).

Although real shocks may not turn out to provide the best explanation of business cycles,
we should certainly consider the possibility that they do (before assuming market
imperfections, incomplete nominal adjustment, etc.).

A Baseline RBC Model

Assumptions

Time is discrete.

There are many identical price-taking firms and many identical price-taking households.

Households are infinitely lived.

Inputs into production are capital and labor.

The production function is Cobb-Douglas:

( )

1
t t t t
L A K Y , . 1 0 < < (4.1)

Output is divided into consumption (C), investment (I), and government purchases (G).
Note that C is defined differently than it was in Romer Chapter 2. It is now aggregate
consumption (earlier it was consumption per member of the household).

Fraction of capital depreciates each period. Thus capital evolves according to:

t t t t
K I K K +
+1


t t t t t t
K G C Y K K +
+1
(4.2)

Government purchases are financed with lump-sum taxes (and Ricardian equivalence
holds).

Labor and capital are paid their marginal products, so:

( ) ( )
t t t t t
A L A K w

1

4
( )
t
t t
t
t
A
L A
K
w

,
_

1 (4.3)

and

,
_

1
t
t t
t
K
L A
r (4.4)

A representative household maximizes the expected value of :

( )

0
1 ,
t
t
t t
t
H
N
l c u e U

(4.5)

Here,
t
N is the population, N C c / and N L l / . L represents labor supplied. Each
individual is endowed with 1 unit of time per period, so l 1 is leisure per person.

Population grows exogeously at rate < n :

nt N N
t
+ ln (4.6)

We will assume the following functional form for the utility function:

( )
t t t
l b c u + 1 ln ln , 0 > b . (4.7)

Finally, we need to describe processes for two shock variables, G and A:

t t
A gt A A
~
ln + + (4.8)

t A t A t
A A
, 1
~ ~
+

1 1 < <
A
(4.9)


( )
t t
G t g n G G
~
ln + + + (4.10)

t G t G t
G G
, 1
~ ~
+

1 1 < <
G
(4.11)

The
G
and
A
terms are independently distributed white noise disturbances.

Romer04b

Skip from bottom of p. 154 through top of p. 156.

5
Begin at Romer p. 156 on Household Optimization under Uncertainty

Household Optimization under Uncertainty

In our model, a household does not have information about future rates of return or future
wages. At any point in time, an individual makes optimal decisions given information
currently available.

We follow Romers informal approach to derive first order conditions for our
optimization problem.

Consumption

Consider a household in period t. Suppose that the household reduces consumption per
member by a small amount c and uses the added wealth to increase consumption per
member in period 1 + t . Optimality requires that such a change leave expected utility
unchanged. Recall the utility function described earlier in equations (4.5) and (4.7):

( )

0
1 ,
t
t
t t
t
H
N
l c u e U

(4.5)

( )
t t t
l b c u + 1 ln ln , 0 > b . (4.7)

In the equation above, c is defined by N C c / and l is defined by N L l / .

From these equations, we find that the marginal utility of
t
c is:

( )( )
t t
t
t
c H N e c U / 1 / /



The utility cost of a small discrete change c is then ( )( )
t t
t
c c H N e / /

.

The total consumption given up by the household in period t is ( ) H N c
t
/ . This permits
an increase in total consumption for the household next period of ( )( )
1
1 /
+
+
t t
r H N c
1
.
The added consumption per member next period is then

( )( ) ( ) ( ) ( )( ) ( )
( )
( ) ( ) c r e e N r e N c H N r H N c
t
n t n
t
nt
t t t
+ + +
+
+
+ + + 1
1
1 1 1
1 0 / 1 0 / / 1 / .

From (4.5) The marginal utility of in consumption per member in period 1 + t is:

( )
( )( )
1 1
1
1
/ 1 / /
+ +
+
+

t t
t
t
c H N e c U

.

1
If I save today, my return comes via the marginal product of capital next period,which
must equal the interest rate over that period, which is denoted
1 + t
r .
6

So, the expected utility gain of the added consumption per member is:

( )
( )( ) ( ) [ ] c r e c H N e E
t
n
t t
t
t
+
+

+ +
+
1 1 1
1
1 / 1 /

or

( )
( ) ( ) [ ] c c r e H N e E
t t
n
t
t
t
+
+ +

+
+
1 1 1
1
/ 1 /

.

Equating the marginal gains and losses from this shift of c yields:

( )
( ) c r
c
e
H
N
e E
c
c
H
N
e
t
t
n t t
t
t
t t

1
]
1

+
,
_

,
_

,
_

+
+
+ +
1
1
1 1
1
1

(4.22)

or, since
( )
( )
n
t
t
e H N e

+
+
/
1
1
is not random and since
n
t t
e N N
+1


( )
1
]
1

+
+
+

1
1
1
1 1
t
t
t
t
r
c
E e
c

(4.23)

This is analogous to the Euler equation in the Ramsey model.

Equation (4.23) can again be rewritten, using the formula for the expected value of a
product:

[ ]

,
_

,
_

+ + +
1
]
1

+
+
+
+

1
1
1
1
1 ,
1
Cov 1
1 1
t
t
t t
t
t
t
r
c
r E
c
E e
c

(4.24)


Labor Supply

Each household must also make a labor supply decision in each period.

Optimality will require that the marginal utility of working (to gain additional
consumption) be equal to the marginal utility from leisure. Equivalently, a small
increase in labor supplied should leave utility unchanged.

Again recall the utility function:


( )

0
1 ,
t
t
t t
t
H
N
l c u e U

(4.5)

7
( )
t t t
l b c u + 1 ln ln , 0 > b . (4.7)


The marginal disutility of working is given by:

t
t t
l
b
H
N
e
l
U

,

and the loss associated with a small increase l is then:

l
l
b
H
N
e
t
t t

.

By working more, one gains additional income and consumption. The added
consumption per worker is given by
t t
l w , and gives an added utility gain of:

( )( )
t t t t
t
l w c H N e

/ 1 /

Equating the marginal utility gain with the marginal utility loss yields:

t t
t
t t
t
t t
l w
c H
N
e l
l
b
H
N
e

1
1

(4.25)

or

b
w
l
c
t
t
t

1
(4.26)

Equations (4.23) and (4.26) are the key equations of the model.

Skip to section 4.6

Romer04c.doc

Start on Romer p. 164

Solving the Model

The RBC model we have been developing, like most RBC models, cannot be solved
analytically.

We will instead describe the solution to a log- linear approximation of the model.

8
The Log-Linearized Model

In any period, the state of the economy can be described by level of the inherited capital
stock and current values of government spending (G) and technology (A).

If we log- linearize around around the non-stochastic blanced growth path, solutions for
consumption and labor supply must be given by:

t CG t CA t CK t
G a A a K a C
~ ~ ~ ~
+ + (4.43)

and

t LG t LA t LK t
G a A a K a L
~ ~ ~ ~
+ + (4.44)

where X
~
represents the difference between the log of X and the log of its balanced
growth path level.

To solve the model, we must find values for the as.

It turns out that the key equations, 4.23 and 4.26 impose restrictions that permit us to
identify the a's, hence solve the model.

Intratemporal (Current Consumpt ion vs. Current Leisure) First Order Condition

Recall (4.26):

b
w
l
c
t
t
t

1
(4.26)

Also recall (4.3):

( )
t
t t
t
t
A
L A
K
w

,
_

1 (4.3)

Substitute (4.3) into (4.26) and take logs:

( )
b
A
L A
K
l
c
t
t t
t
t
t

,
_

1
1


( ) ( )
t t t t t
L K A
b
l c ln ln ln 1
1
ln 1 ln ln

+ +
,
_

(4.45)
9

The deviation of the actual value of the RHS of this equation from its balanced growth
path level is simply ( )
t t t
L K A
~ ~ ~
1 + .


On the LHS we will approximate using a first-order Taylors series approximation.
2

Think of the LHS as a function of
t
c ln and
t
l ln ; i.e.:

( )
t t
l c z 1 ln ln

( )
t
l
t
e c z
ln
1 ln ln

Taking the Taylors series approximation around the balanced growth path values of
t
c ln
and
t
l ln :

t t
l
l
l
c z z
~
1
~
*
*
0

,
_

+

Because population is not affected by shocks,
t t
c C
~
~
and
t t
L l
~ ~
and we can rewrite the
ezpression for the LHS of (4.45) as a deviation from its balanced growth path level as:

t t
L
l
l
C z z
~
1
~
*
*
0

,
_

+

To see that
t t
c C
~
~
recall:

N
C
c

so:

N C c ln ln ln

2
For ( ) y x f z , ,

( )
( )
( )
( )
( )
0
0 0
0
0 0
0 0
, ,
, y y
y
y x f
x x
x
y x f
y x f z

+
or

( )
( )
( )
( )
0
0 0
0
0 0
0
, ,
y y
y
y x f
x x
x
y x f
z z

10

Letting asterisks indicated balanced growth path levels:

* * *
ln ln ln N C c

Differencing the last two equations and noting that
*
N N :

C c
~
~


By equating the expressions derived for the LHS and RHS of (4.45) as deviations from
balanced growth path values, we have:

( )
t t t t t
L K A L
l
l
C
~
ln
~
ln
~
ln 1
~
1
~
*
*
+

,
_

+ (4.46)

romer04d.doc

Recall

( )
t t t t t
L K A L
l
l
C
~
ln
~
ln
~
ln 1
~
1
~
*
*
+

,
_

+ (4.46)

Furthermore, recall that we have conjectured solutions:

t CG t CA t CK t
G a A a K a C
~ ~ ~ ~
+ + (4.43)

and

t LG t LA t LK t
G a A a K a L
~ ~ ~ ~
+ + (4.44)

Substituting (4.43) and (4.44) into (4.46) yields:

( ) G a A a K a
l
l
G a A a K a
LG t LA t LK t CG t CA t CK
~ ~ ~
1
~ ~ ~
*
*
+ +

,
_

+ + +
( )
t t
K A
~
ln
~
ln 1 + (4.47)


Equation (4.470 must hold for all values of K
~
, A
~
, and G
~
. This implies that coefficients
for each of these variables must be identical on the two sides of this equation, leading to
three equations which impose restrictions on the s ' . See equations (4.48)-(4.50) in
Romer.
11
The Intertemporal First-Order Condition

Our strategy here will be to use the intermporal first order condition to get three more
restrictions on the s. We outline the procedure for doing so.

Recall the intertemporal first -order condition:

( )
1
]
1

+
+
+

1
1
1
1 1
t
t
t
t
r
c
E e
c

(4.23)

Define the bracketed expression above as
1 + t
Z , i.e.:

( )
1
]
1

+
+
+
+ 1
1
1
1
1
t
t
t
r
c
Z

Let
1
~
+ t
Z be the difference between the log of
1 + t
Z and the log of its balanced growth path
value.

Recall our conjectured solution:

t CG t CA t CK t
G a A a K a C
~ ~ ~ ~
+ + (4.43)

Update this equation to time t+1:

1 1 1 1
~ ~ ~ ~
+ + + +
+ +
t CG t CA t CK t
G a A a K a C (4.51)

Recall (4.4), now updated one period:

+
+ +
+
1
1
1 1
1
t
t t
t
K
L A
r (4.4)

We can substitute (4.4) into the definition of
1 + t
Z given above. Also substitute the
definition
1 1 1
/
+ + +

t t t
N C c .

This would give us
1 + t
Z as a function of
1 + t
A ,
1 + t
K ,
1 + t
C , and
1 + t
L (
1 + t
N is
predetermined).

Take logs and use a Taylors series expansion to express
1
~
+ t
Z as a function of
1
~
+ t
K ,
1
~
+ t
A ,
and
1
~
+ t
C and
1
~
+ t
L .

12
Substitute the conjectured solutions (4.43 and 4.44) for
1
~
+ t
L and
1
~
+ t
C .

Recall (4.43) and (4.44):


t CG t CA t CK t
G a A a K a C
~ ~ ~ ~
+ + (4.43)

and

t LG t LA t LK t
G a A a K a L
~ ~ ~ ~
+ + (4.44)


Substituting updated versions of (4.43) and (4.44) would give
1
~
+ t
Z as a function of
1
~
+ t
K ,
1
~
+ t
A , and
1
~
+ t
G .

To get rid of the endogenous
1
~
+ t
K term, recall equation (4.2):

t t t t t t
K G C Y K K +
+1
(4.2)

Use the production function to eliminate
t
Y , then log- linearize equation (4.2) and write
1
~
+ t
K as a function of
t
K
~
,
t
A
~
,
t
G
~
,
t
L
~
and
t
C
~
. Use (4.43) and (4.44) to substitute for
t
L
~

and
t
C
~
.

We then have
1
~
+ t
K as a function of
t
A
~
,
t
G
~
, and
t
K
~
:

t KG t KA t KK t
G b A b K b K
~ ~ ~ ~
1
+ +
+
(4.52)

Next, we can then get rid of the
1
~
+ t
K term in the
1
~
+ t
Z expression, so
1
~
+ t
Z is now a
function of
t
K
~
,
t
A
~
,
t
G
~
,
1
~
+ t
A , and
1
~
+ t
G .

Finally, use this expression to calculate [ ]
1
~
+ t t
Z E . The
1
~
+ t
A and
1
~
+ t
G terms drop out when
we take expectations, so we will have [ ]
1
~
+ t t
Z E as a function of
t
K
~
,
t
A
~
, and
t
G
~
.

Finally, recall the intertemporal first order condition, (4.23):

[ ]
1
1
+

t t
t
Z E e
c

(4.23)

13
On each side, we wish to take logs, and then and express as deviations from balanced
growth path levels. First substituting
t t t
N C c / , the LHS will become a linear function
of
t
C
~
, which we will replace with our conjectured solution,
t CG t CA t CK t
G a A a K a C
~ ~ ~ ~
+ + .

On the RHS assume that [ ] B Z E Z E
t t t t
+
+ + 1 1
ln ln , where B is a constant. Romer
provides assumptions for which this will be true.
3
Given this, when we express the RHS
as a deviation from the balanced growth path level, we get [ ]
1
~
+ t t
Z E , which we have
already found is a function of
t
K
~
,
t
A
~
, and
t
G
~
.

Equating coefficients on the LHS and RHS of the modified (4.23) will provide 3
additional restrictions on the s. We now have 6 linear restrictions (we had three others
from the other first order condition) and can solve for the 6 s, giving us an
approximate solution to the model. Thereafter we can use our approximate solution to
investigate the properties of the model.

romer04e.doc

Start on Romer p. 168 on Implications

Introduction

Given our approximate solution for the model, we can calculate
t
C
~
and
t
L
~
(and other
variables) when given values for
t
K
~
,
t
A
~
, and
t
G
~
. This permits us to calculate paths for
the models variables following shocks to technology or government spending.

One generally begins by selecting baseline values for the model parameters, solving for
the s in our solution, and then tracing the impacts of shocks. Romer selects plausible
parameter values based on empirical evidence.

Technology Shock

Let 95 . 0
G A
. Then consider the impact of a positive 1% shock to technology.
The qualitative effects of the shock are:

Capital gradually accumulates and then returns to normal:

The marginal product of capital is higher because of the shock.
Higher output and a desire to spread out consumption leads to more saving (and
investment) initially.
Eventually the shock dissipates, and we must return to the balanced growth path.


3
For more on this see Lindgren, Statistical Theory, p. 176.
14
Labor supply jumps, then gradually falls, goes below the balanced growth path, and then
returns to the balanced growth path:

High marginal product of capital (hence interest rate) and high marginal product
of labor both induce high current labor supply after the shock (the interest rate
works via an intertemporal substitution effect).
As the shock dissipates, we are left with an above balanced growth path capital
stock (i.e. higher wealth than on the balanced growth path).
As the capital stock is permitted to return to normal, we enjoy both more
consumption and leisure (relative to the balance growth path), hence below path
labor supply.

Output increases in the period of the shock, then returns gradually to the balanced growth
path.

The shock and intertemporal work effort effects both lead to an initial increase in
output.
The effect persists because of the persistence of the original shock and the
accumulation of capital in the initial periods (higher consumption is spread over a
long horizon).

Consumption rises more slowly than output, then gradually returns to normal:

This reflects the consumption smoothing motive.

The wage rises and then gradually returns to normal:

The shock directly increases the marginal product of labor and then dissipates.
Capital accumulation also contributes to the rising marginal product of labor
initially.

The interest rate immediately rises, then gradually falls below the balanced growth path
level, before returning to it:

Initially, the shock increases the marginal product of capital. When the shock
begins to dissipate and labor supplied is reduced, the capital stock is high relative
to the amount of effective labor, and the marginal product of capital is low. As the
capital stock returns to its balanced growth path level, so does the marginal
product of capital (and the interest rate).

A Less Persistent Technology Shock

If the
A
parameter is smaller, technology shocks are less persistent. This implies that
wealth effects of the shocks will be smaller, and substitution effects larger. We have
shorter, sharper output fluctuations (the period of the shock is now an especially good
15
time to work, but consumption will not be affected greatly). In contrast, if 1 =
A
,
technology shocks are permanent; and the output burst is initially smaller but sustained.

Changes in Government Spending

Other things equal, a positive government spending shock reduces output available for
other uses. Output is scarce in the period of the shock. The desire to maintain
consumption (spread out the effects of the shock) leads to capital decumulation and a rise
in the interest rate. The increase in the interest rate induces more work effort in the initial
periods of the shock. The wealth effect also induces higher work effort and lower
consumption (i.e., lower wealth reduces both consumption and leisure). The wage
declines at the time of the shock (with more labor working, the marginal product is
lower).










1
Romer Chapter 6 on the Lucas Model
The Case of Perfect Information
Producer Behavior
There are many different goods.
A representative producer of a typical good, good i, produces according to the production
function:
i i
L Q (6.1)
where
i
L is the amount the individual works and
i
Q is the amount he produces.
Real consumption, is nominal income divided by a price index:
P
Q P
C
i i
i
.
Utility depends positively on consumption and negatively on hours worked:
1 ,
1
> >

i i i
L C U (6.2)
Substituting the previous equations into (6.2) gives:

i
i i
i
L
P
L P
U
1
. (6.3)
Taking prices as given, an individual maximizes utility by selecting
i
L to satisfy the first
order condition:
0
1


i
i
L
P
P
. (6.4)
Rearranging, we get:
1
1


, ,
_ _





P
P
L
i
i
(6.5)
Letting lowercase letters denote logs:
( ( ) ) p p l
i i



1
1

(6.6)
2
This is a labor supply function (and, indirectly, an output supply function) in which an
individuals hours depend on the relative price of the individuals output price. Note that
this supply function does not include inertial effects like those in the Lucas paper we read
earlier.
Note that if P P
i
, then 0 , 0 , 1 , 1
i i i i
q l Q L (the utility function was designed so
that this would be the result).
Demand
We assume that the demand for good i has the following form (note: this is NOT derived
from a utility maximization problem):
( ( ) ) 0 , > > + + p p z y q
i i i
, (6.7)
where y is the log of a measure of aggregate income,
i
z is a shock to demand for good i
(with mean zero across goods), and is the demand elasticity. More specifically, y is
defined to be the average of the s q
i
' across goods, and p is defined to be the average of
the s p
i
' across goods:
i
q y (6.8)
and
i
p p . (6.9)
Aggregate demand is given by:
p m y (6.10)
This is just a simple way of modeling aggregate demand; the essential property is that the
price level and output are inversely related. While m can be literally interpreted as the log
money supply, it might be thought of more generally as any aggregate demand shifter.
Equilibrium
We require that quantities demanded equal quantities supplied in each market i. From
(6.6) and (6.7) we obtain:
( ( ) ) ( ( ) ) p p z y p
i i i
+ +


1
1
. (6.11)
3
Solving for
i
p yields:
( ( ) ) p z y p
i i
+ + + +
+ +

1
1
(6.12)
Next, average the left- and right-hand sides of (6.12):
p y p + +
+ +

1
1
(6.13)
Solve for y:
. 0 y (6.14)
Recall equation (6.10):
p m y (6.10)
If 0 y , then (6.10) implies:
m p (6.15)
Thus money is neutral in this model. An increase in m leads to a proportional increase in
p. Also, since p is observable and markets clear, the average level of log output is zero.
An increase in aggregate demand does NOT lead to higher aggregate output in the perfect
information version of the model.
Imperfect Information
Producer Behavior
We now consider the case where producers observe the price of the good they produce,
but not the aggregate price level.
Define the relative price of good i as p p r
i i
, we get:
i i
i i
r p p
p p p p
+ +
+ + ) (
(6.16)
Individuals supply choices are motivated by relative prices, but relative prices are not
observed;
i
p is observed, but the individual must make a forecast regarding
i
r .
4
We assume that individuals calculate the expectation of
i
r given
i
p , and then act as if
this expected value were known with certainty (we implicitly have been making this
assumption of certainty equivalence in all of our work with rational expectations). This
implies that with uncertainty, equation (6.6) is modified to give:
[ [ ] ]
i i i
p r E l |
1
1

(6.17)
We must next describe the process generating values for m. We assume that m is
normally distributed with mean [ [ ] ] m E and variance
m
V (this is a bit different and more
general than the demand process specified in the paper by Lucas that was assigned
earlier). We also assume that the s z
i
' , which we earlier assumed had mean 0, are
normally distributed, and that the
i
z and m shocks are independent.
We will next invoke our solution to the signal extraction problem. We wish to forecast
i
r
using our knowledge of
i
p . Recall that
i i
r p p + + ; i.e. we observe a sum, but wish to
forecast a component of the sum. Writing our synthetic regression forecast in a form
where variables are expressed as deviations from means we get:
[ [ ] ] [ [ ] ] ( ( ) ) p E p
V V
V
p r E
i
p r
r
i i

+ +
| , (6.19)
where
r
V is the variance of
i
r and
p
V is the variance of p . Use of this formula requires
that
i
r and p be independent normal variables. At the moment we have not derived
expressions for
r
V and
p
V , but we will be able to do so eventually.
Note here that the s r
i
' have unconditional mean zero, and the s p
i
' have unconditional
mean [ [ ] ] p E .
Recall equation (6.17):
[ [ ] ]
i i i
p r E l |
1
1

(6.17)
Substituting (6.19) into (6.17) yields:
[ [ ] ] ( ( ) ) p E p
V V
V
l
i
p r
r
i

+ +

1
1

(6.20a)
or
5
[ [ ] ] ( ( ) ) p E p b l
i i
. (6.20b)
Averaging across producers yields:
[ [ ] ] ( ( ) ) p E p b y (6.21)
This is the Lucas supply function.
Equilibrium
Now combine aggregate demand, equation (6.10) and the Lucas aggregate supply curve,
(6.21):
[ [ ] ] ( ( ) ) p E p b p m .
Solve for p:
[ [ ] ] p E
b
b
m
b
p
+ +
+ +
+ +

1 1
1
. (6.22)
We know how to solve from this point on using the method of undetermined coefficients.
However, Romer uses a trick that often (but not always) lets us solve rational
expectations models more quickly. Take expectations on both sides of (6.22):
[ [ ] ] [ [ ] ] [ [ ] ] p E
b
b
m E
b
p E
+ +
+ +
+ +

1 1
1
(6.24)
This equation can be solved for [ [ ] ] p E :
[ [ ] ] [ [ ] ] m E p E . (6.25)
Substituting into (6.22):
[ [ ] ] m E
b
b
m
b
p
+ +
+ +
+ +

1 1
1
or , using the fact that [ [ ] ] [ [ ] ] ( ( ) ) m E m m E m + + :
[ [ ] ] [ [ ] ] ( ( ) ) m E m
b
m E p
+ +
+ +
1
1
. (6.26)
Recall equation (6.21):
[ [ ] ] ( ( ) ) p E p b y (6.21)
6
Now substituting (6.25) and (6.26) into (6.21) gives a solution for output:
[ [ ] ] [ [ ] ] ( ( ) ) [ [ ] ]

, ,
_ _




+ +
+ + m E m E m
b
m E b y
1
1
[ [ ] ] ( ( ) ) m E m
b
b
y
+ +

1
. (6.27)
Equations (6.26) and (6.27) illustrate the basic features of the Lucas model. Expected
variations in money affect prices in proportion. Money surprises affect both prices and
output, with the division of the impacts depending on underlying variances of relative and
general prices.
From equations (6.20), recall that


, ,
_ _




+ +


, ,
_ _






p r
r
V V
V
b
1
1

.Clearly, the larger the variance

in relative prices, the larger is b. With larger b values money shocks have bigger impacts
on output and small impacts on the aggregate price level.
Finally, to tie up loose ends, we need to go back and figure out the variances of
r
V and
p
V .
Recall equation (6.26):
[ [ ] ] [ [ ] ] ( ( ) ) m E m
b
m E p
+ +
+ +
1
1
. (6.26)
This equation implies that
( ( ) )
2
1 b
V
V
m
p
+ +
.
To find
r
V , first recall equations (6.7), (6.20b), and (6.21):
( ( ) ) 0 , > > + + p p z y q
i i i
, (6.7)
[ [ ] ] ( ( ) ) p E p b q l
i i i
. (6.20b)
[ [ ] ] ( ( ) ) p E p b y (6.21)
Substitute (6.21) into (6.7) to get:
[ [ ] ] ( ( ) ) ( ( ) ) p p z p E p b q
i i i
+ +
7
In (6.20b), add and subtract bp to the right-hand side:
( ( ) ) [ [ ] ] ( ( ) ) p E p b p p b q
i i
+ + .
Combining the last two equations yields:
[ [ ] ] ( ( ) ) ( ( ) ) ( ( ) ) [ [ ] ] ( ( ) ) p E p b p p b p p z p E p b
i i i
+ + + +
( ( ) ) ( ( ) ) p p b p p z
i i i

( ( ) )( ( ) ) p p b z
i i
+ +
b
z
p p
i
i
+ +

Since p p r
i i
, the variance of
i
r is:
( ( ) )
2
b
V
V
z
r
+ +

.
Again recall from (6.20) that b is given by:
p r
r
V V
V
b
+ +

1
1

where
( ( ) )
2
1 b
V
V
m
p
+ +

and
( ( ) )
2
b
V
V
z
r
+ +

.
Substituting the last two equations into the definition of b, one gets:
( ( ) )
( ( ) ) 1 1
1 1
1 1
1 1
1 1
] ]
1 1







+ +
+ +
+ +


m z
z
V
b
b
V
V
b
2
2
1
1
1

,
which implicitly defines b in terms of known parameters.
8
Finally, note that since:
[ [ ] ] [ [ ] ] ( ( ) ) m E m
b
m E p
+ +
+ +
1
1
. (6.26)
and
b
z
p p r
i
i i
+ +

,
that p and
i
r are respectively linear functions of m and
i
z . This implies that p and
i
r are
normal and independent, as required when we invoked the solution to the signal
extraction problem.
9
Discussion of the Lucas Model
Policy shocks affect both output and prices positively, leading to a Phillips curve
relation.
There is no exploitable tradeoff between output and inflation only the money supply
process error term has an effect on output.
Changes in policy rules change expectations, which change apparent aggregate
relationships (i.e., observed Phillips curve tradeoff parameters). If policymakers try to
take advantage of statistical relationships, effects operating through expectations may
cause those relationships to break down. This is the Lucas critique.
Policy ineffectiveness: Systematic stabilization policy must be ineffective. This is a
general result. [ [ ] ] m E can be inferred on the basis of a very complicated policy rule, and
still only the error term [ [ ] ] m E m appears in (6.27).
Suppose that governments observe aggregate demand shocks (from a source other than
money) but the public does not. This would permit the government policymaker to
potentially stabilize output. However, this is not a good justification for stabilization
policies. First, most stabilization policy choices are generally based on observable
performance indicators. Second, if the public did not observe those indicators, it would be
easier to simply announce them than to carry out a counter-cyclical stabilization policy.
Empirical evidence on impacts of expected and unexpected money on output: Barro,
Romer and Romer.
Is it plausible that there is substantial imperfect information about m and p?
Are labor supply elasticities large enough that we can explain output fluctuations as
responses to relative price misperceptions?
1
Romer Chapter 6 Part B Staggered Price Adjustment
The Lucas model showed that with rational expectations and market-clearing
assumptions, systematic countercyclical stabilization policy would be ineffective. Early
criticism from Keynesians focused on the implausibility of the rational expectations
assumption. Over time, the assumption of rational expectations has become more
acceptable (at least as a modeling assumption), and criticism has shifted to the
assumption that markets quickly adjust to market clearing equilibria.
Initially Keynesians responded by assuming some nominal price or wage rigidity.
We will follow this approach, and will later come back to the issue of why it is that prices
might be rigid.
It turns out that modern New Keynesian models generally rely on the existence
imperfect competition. This helps out later in explaining why prices might be rigid.
Given that this is so, we will develop a model which incorporates imperfect competition.
A Model of Imperfect Competition and Price Setting
Assumptions
The economy consists of a large number of individuals.
Each individual is the producer of a good and sets the price of that good.
Labor is the only input: the output of good i is equal to the amount of labor employed in
its production.
One major difference from the Lucas model: there is a competitive labor market where
individuals may sell and hire labor in order to produce goods (in the Lucas model, one
had to use own labor to produce your good).
Demand equations for each good are as follows (same as the Lucas model except for the
absence of market specific shocks):
( ( ) ) p p y q
i i
= = , 1 > > ,
where the lower case letters represent logs, p is the average of the s p
i
' , and y is the
average of the s q
i
' . Converting from logs to levels, the demand equation can be written:







= =
P
P
Y Q
i
i
.
Sellers with market power will sell at a price above marginal cost. If they cannot adjust
price, they will be willing to satisfy small increases in demand.
2
We employ the same utility function as the Lucas model did:

i i i
L C U
1

Income now includes profit as well as wage income, so utility can be written as:
( ( ) )

i
i i i
i
L
P
WL Q W P
U
1

+ +
= = . (6.37)
Once again, aggregate demand is given by:
p m y = = .
Unlike the Lucas model, here we assume that the money supply is observed.
Individual Behavior
Substituting the demand equation,







= =
P
P
Y Q
i
i
, into the utility function, yields:
( ( ) )

i
i
i
i
i
L
P
WL
P
P
Y W P
U
1

+ +






= =

(6.38)
The individual chooses his price,
i
P , and his work effort,
i
L . The first order conditions
for the utility maximization problem lead to the following results (after some
manipulation):
P
W
P
P
i
1
= =

(6.40)
1
1







= =

P
W
L
i
(6.42)
The first of these is a price mark-up equation; the second is an individuals labor supply
function.
3
Equilibrium
By symmetry, in equilibrium, each individual works the same amount and produces the
same amount of output. From (6.42) the real wage is:
1
= =

Y
P
W
.
Substituting into (6.40) yields an expression for each producers desired relative price as
a function of aggregate output:
1
*
1


= =

Y
P
P
i
. (6.44)
In logs:
( ( ) )y p p
i
1
1
ln
*
+ +









= =

or
y c p p
i
+ + = =
*
. (6.45)
Since producers are symmetric, all prices are the same, and each price is equal to the
average price. We can then use (6.44) to solve for the equilibrium level of output:
1
1
1









= =

Y . (6.46)
From the aggregate demand equation, P M Y / = = , we find the equilibrium price:
1
1
1









= =

M
P .
Implications
Output is less than 1, which was the competitive market level of output. This reflects the
suboptimality of the monopoly outcome. This implies:
4
Recessions and booms have asymmetric welfare effects. Booms are good,
recessions are bad.
Pricing decisions have external effects. Suppose that starting from an equilibrium
point, each producer cuts price by a small amount. The average price declines,
and aggregate output rises, shifting out the demand curve for each product. The
demand shift makes everyone better off (even though each individual is privately
better off not cutting price).
The existence of imperfect competition is consistent with monetary neutrality.
The solution for output in (6.46) does not depend on M.
Predetermined Prices
This is the Fischer model of staggered pricing.
Assumptions
Price setters cannot freely adjust prices in each period. Instead, each price setter sets
prices every other period for each of the next two periods.
Example, at time 0 = = t , I set a price for time 1 = = t and a price (possibly a different
one) for time 2 = = t . At 2 = = t , I would set prices again, this time for times
3 = = t and 4 = = t .
Recall (6.45):
y c p p
i
+ + = =
*
. (6.45)
Since p m y = = , this becomes:
( ( ) ) m p c p
i
+ + + + = = 1
*
.
For simplicity, normalize the constant c to be zero (one can think of this as simply
redefining the LHS as a deviation from c):
( ( ) )p m p
i
+ + = = 1
*
No specific assumptions are made about the process for m.
We will assume that price setters set future log prices at the expected profit maximizing
log prices, given information available at the time prices are set. Expectations are
assumed to be rational.
5
Solution
The average price at time t is given by:
( ( ) )
2 1
2
1
t t t
p p p + + = =
where
1
t
p is the price set for time t by those setting prices at time 1 t and
2
t
p is the
price set for time t by those setting prices at time 2 t .
Since individuals select expected profit maximizing prices,
*
1
1
it t t
p E p

= and
*
2
2
it t t
p E p

= . From these we get:
*
1
1
it t t
p E p

=
( ) [ ]
t t t t
p m E p + =

1
1
1
(6.50)
( ) ( )
2 1
1
1
2
1
1
t t t t t
p p m E p + + =


*
2
2
it t t
p E p

=
( ) ( )
2 1
2 2
2
2
1
1
t t t t t t
p p E m E p + + =

(6.51)
Solve (6.50) for
1
t
p :
2
1
1
1
1
1
2
t t t t
p m E p

+
+
=

. (6.52)
Take expectations at time 2 t on both sides of (6.52):
2
2
1
2
1
1
1
2
t t t t t
p m E p E

+
+
=

(6.53)
Substitute (6.53) into (6.51)
( )

+
+

+
+
+ =

2 2
2 2
2
1
1
1
2
2
1
1
t t t t t t t
p p m E m E p

(6.54)
6
Solve for
2
t
p :
t t t
m E p
2
2

=
Substitute (6.55) into (6.52) and simplify to get:
( )
t t t t t t t
m E m E m E p
2 1 2
1
1
2


+
+ =

Recall that ( ( ) )
2 1
2
1
t t t
p p p + + = = and that
t t t
p m y = , so:
( )
t t t t t t t
m E m E m E p
2 1 2
1


+
+ =

( ) ( )
t t t t t t t t
m E m m E m E y
1 2 1
1
1

+
+
=

Implications
Aggregate demand shifts have real effects; see the
t t t
m E m
1
term.
Anticipated demand shifts (that become anticipated after the first prices are set) affect
output; see the
t t t t
m E m E
2 1
term. Illustrate via numerical example.
It is possible for (some) policy rules to stabilize output. The policymaker can observe the
effect of
t t t t
m E m E
2 1
when choosing a target for
t
m .
The impacts of an anticipated demand shock depend on real rigidity, as measured by
parameter . If is low, producers are reluctant to allow relative prices to get out of
line. So when setting the one-period-ahead price, they set it close to those already set for
the next period, which increases overall price rigidity.
7
This handout describes a staggered price-setting New Keynesian model, as presented by
David Romer, Advanced Macroeconomics, pp. 265-270.
Set-up
There are two groups of firms (1 and 2). Every two periods, one group sets a price that
then stays in effect for two periods (the current period and the next one). Groups 1 and 2
alternate in setting prices; i.e., group one may set prices in periods 1, 3, 5 etc., while group
2 sets prices in periods 2, 4, 6, etc. Firms in the two groups are in other respects similar,
but they are not perfect competitors -- different groups may charge different prices
without having sales go to zero. Rather, firms are imperfectly competitive.
There are five basic equations:
t t it
y p p = =
*
i = 1 2 , (6.45)
The equation above is the supply relation. The variable p
it
*
is the log price that a firm in
group i would ideally like to charge in period t (if it could change prices in every period,
which it cant). The variable p
t
is the log aggregate price level in period t. The variable
y
t
is the log of aggregate output. This is equation (6.45) in the text, with the
normalization requiring that 0 = c .
The equation says that a firm would desire a higher price for its output when the general
price level is high. One can think of firms as imperfect competitors who are sensitive to
rivals prices. When rivals are charging high prices, I can also charge high prices; when
rivals charge low prices, I must also keep prices low. In other words, my price should not
be too far out of line with others prices. The equation also says that I am willing to see
my relative price be higher when aggregate output is high. I.e., if demand is strong overall,
then I would be willing to charge a somewhat higher price than my rivals, because sales
will remain strong anyway.
Romers book provides a slightly different justification for this equation. Essentially, in an
imperfectly competitive market, firms charge prices that are a markup over costs (with the
markup depending upon the elasticity of demand). In this model, firms must set prices and
then satisfy demand. But if aggregate demand turns out to be unusually high, firms must
then produce more output in the aggregate. But to produce more output, they must hire
more labor. In this model, the supply of labor depends on the real wage, so wages (and
costs) must rise with output. Because of the markup in pricing, with higher costs firms
would wish to be charging a higher price for their product. So, other things equal, firms
prefer to be charging a higher relative price when aggregate output is high.
( )
x p E p
t it t it
= +
+
1
2
1
* *
(6.60)
The variable x
t
is the actual price set by firms in the group that sets a price at time t. This
price must prevail for two periods (i.e. periods t and t+1). It is equal to an average of this
periods optimal price and next periods expected optimal price.
( ) p x x
t t t
= +

1
2
1
p. 266
As noted, p
t
is the aggregate price level in period t. In the equation above, it is shown as
an average of the price set last period (by one group) and the price set this period (by the
other group).
y m p
t t t
= (6.10)
Equation (6.10) is a conventional, simple aggregate demand relation with m
t
representing
the money supply.
m m u
t t t
= +
1
(6.59)
Equation (6.59) is a policy rule. It says that the money supply is a random walk.
Analysis
Our strategy will be to obtain a solution for x
t
, the price set by firms who choose a price
at time t. Clearly, once we have a solution for x
t
, we will also be able to solve for p
t
.
Substitute (6.10) into (6.45):
( ( ) )
t t t it
p m p p + + = =
*
( ( ) )
t t it
p m p + + = = 1
*
Now substitute the equation above into (6.60):
( ( ) ) [ [ ] ] ( ( ) ) [ [ ] ] { { } }
1 1
1 1
2
1
+ + + +
+ + + + + + = =
t t t t t t t
p E m E p m x
( ( ) ) ( ( ) ) ( ( ) )
t t t t
t t
t t
x x E m
x x
m x + + + + + +




+ +
+ + = =
+ +

1
1
2
1
1
2
1
2
1
2
1
2
1
2
1

( ( ) )( ( ) )
1 1
2 1
4
1
+ +
+ + + + + + = =
t t t t t t
x E x x m x
Notice that if we regard m
t
as exogenous, then the preceding equation includes only one
endogenous variable, x
t
, and also the expectational term E x
t t +1
. We now have the model
in a form we know how to handle. Following Romer, I will isolate x
t
on the LHS of this
equation before following our usual procedures. After some algebra we obtain:
( ) ( ) x A x E x A m
t t t t t
= + +
+ 1 1
1 2 (6.62)
where
( ( ) )
( ( ) )

+ +

= =
1
1
2
1
A .
Now conjecture a solution:
t t t
m x x + + + + = =
1
Update one period:
1 1 + + + +
+ + + + = =
t t t
m x x
Take expectations:
t t t t
m x x E + + + + = =
+ +1
Substituting our conjectured solution for x
t
into the equation above gives:
( ( ) )
t t t t t
m m x x E + + + + + + + + = =
+ + 1 1
.
Now substitute for x
t
and E x
t t +1
in equation (6.62):
( ( ) ) [ [ ] ] ( ( ) )
t t t t t t t
m A m m x A Ax m x 2 1
1 1 1
+ + + + + + + + + + + + = = + + + +

.
Now equate coefficients on the two sides of the equation. This yields the following three
equations:
(1) A A + + = =
(2)
2
A A + + = =
(3) ) 2 1 ( A A A + + + + = =
The second equation is quadratic and can be solved for . The solution is:
( ( ) )( ( ) )
A
A A
2
2 1 2 1 1 + +
= =
After substituting for A and doing lots of algebra, we obtain:

+ +

= =
1
1
or

+ +
= =
1
1
Only the first solution is reasonable (because the second solution would give an unstable
solution for x
t
).
Now that we have a solution for getting solutions for and should be
straightforward. However, in this case there is a complication. In equation (1) it appears
that any value of will satisfy the equation.
1
To find we will take a short-cut. We can pin that parameter down using an analysis of
the steady state of this model when there are no shocks. In such a steady state, we will
have a stable price and a stable level of output. I.e., the sequence of chosen prices will
have identical values in each period. Moreover, the chosen prices will also be the same as
the desired prices in each period. But if desired price equals actual price in each period,
then y
t
= 0 by equation (6.45). And if y
t
= 0, then equation (6.10) implies that p m
t t
=

1
For any non-zero value of , the first equation implies that :
A A + + = = 1
or
( ( ) )
( ( ) )

+ +
= =

= =
1
1
2
1
A
A
However, this value of is not compatible with the value we solved for earlier:
( ( ) )
( ( ) )

+ +



+ +
1
1
1
1
2 .
This implies that if we are to satisfy both (1) and (2), then 0 = = . (Any non-zero value
leads to the contradiction noted). The text gives an alternative route for finding this value
for .
in each period. But if actual and desired prices are equal in each period, then equation
(6.60) implies that
1 1 1
= = = = = = = = = =
t t t t t t
m p x m p x , etc. Now recall that our conjectured
solution was:
t t t
m x x + + + + = =
1
The preceding analysis implies that
t t t
m m m + + + + = = .
But this is generally true only if 0 = = and 1 = = + + . We now have a solution for .
Moreover, the solution for will be given by = = 1 .
Given this solution for p
t
, and given that y m p
t t t
= , a solution for output can be
found:
t t t
u y y
2
1
1

+ +
+ + = =

Implications
The aggregate price level adjusts gradually to a shock. Because price adjusts gradually, the
shock to output is persistent. The intuition is that when there is a demand shock, one
group of firms has preset its price. The group which gets to chose price does not want a
price too different from the other group, for competitive reasons. So price does not fully
and immediately move to the new long-run price level. In the next period the next group
chooses price, but they are similarly constrained by the need to keep their relative price in
line. So prices adjust slowly. With prices adjusting slowly, the demand shock affects
output instead (firms with market power are willing to produce the additional units
demanded because price is in excess of marginal cost). When price adjusts slowly, output
effects will last a long time.
1
Romer06D.doc
Introduction
There are problems with the Lucas model and with models where prices are assumed to
be rigid. The money supply and the aggregate price level are easily and quickly
observable, contrary to the assumption of the Lucas model. Prices can usually be changed
at low cost, apparently undermining the rigid price models.
Agents are presumably interested in real magnitudes. Nominal magnitudes matter only in
minor ways to them. If nominal imperfections are macroeconomically important, one
must show that small frictions at a microeconomic level have large impacts at a
macroeconomic level.
We will assume the existence of menu costs, small costs associated with changing prices.
We investigate whether such costs might have important macroeconomic consequences.
Market power, Menu Costs, and Price Adjustments
Consider a firm with market power, which currently sets a price at the profit maximizing
level.
Now suppose that there is a reduction in the demand facing this firm.
Momentarily suppose that the firm does not lower its price to the new lower optimal
level.
Fig. 6.3 in Romer shows that:
The profit consequence of a reduction in demand is large.
The profit consequence of failing to cut price is small.
We can illustrate the same point with an alternative diagram (show as a function of P),
or with a Taylors series expansion of the profit function.
If the menu cost (the cost incurred when one changes a price) is larger than the gain from
changing price, then the firm will leave its price unchanged.
How Aggregate Demand Externalities Arise
For simplicity suppose that AD is given by:
Y = M/P.
If prices are rigid, then a reduction in M reduces real demand for all firms in proportion.
2
But for each of many small firms (each imperfectly competitive), the losses from failing to
adjust prices are very small (as shown above).
If all firms did adjust prices, they would all be better off (because their demand curves
would be shifted back out to the right). But no firm can make its demand curve shift back
out by unilaterally being flexible on price.
Thus one can view a recession as a result of market failure -- the failure to overcome
externalities because of free-rider problems.
Some Welfare Implications.
In some models (Lucas, Fischer, McCallum) booms as well as recessions are associated
with welfare losses.
In this model monopoly equilibrium is suboptimal (with output too low) so booms increase
welfare (by causing higher output). Recessions result in welfare losses, consistent with
everyday views.
It is not obvious, but in such models fluctuations (i.e. demand induced business cycles) are
inefficient. There can be a role for stabilization policy.
Some Problems for the Market Power/Price Rigidity Model
The analysis of price rigidity for an individual firm was conducted in a partial
equilibrium environment.
It turns out that a general equilibrium analysis reveals some problems with the preceding
analysis.
Romer Section 6.6 Revisited
First, recall that the first order condition for utility maximization describing labor
supply behavior was given by:
1
1







= =

P
W
L
i
(6.42)
Because of symmetry, in equilibrium each individual works and produces the
same amount, so Y L
i
= = and:

1
Y
P
W
= = (6.42a)
3
where
1
1

= =

. This is the elasticity of labor supply with respect to the real

wage.
Each producer faces a demand function:







= =
P
P
Y Q
i
i
(6.7a)
Each firms (real) profit is given by (again recall that one unit of output requires
the use of one unit of labor input):
( ( ) )
P
Q W P
i i

= = (6.37a)
or
i
i
Q
P
W
P
P






= = (6.37b)
Subsituting (6.7a) and (6.42a) into (6.37b) we get:














= =

1
Y
P
P
P
P
Y
i i
i
or, since
P
M
Y = =

1
1


















= =

P
M
P
P
P
M
P
P
P
M
i i
i

+ +



















= =
P
P
P
M
P
P
P
M
i i
i
1
1
(6.84)
Remember that each firm regards M and P as exogenous, so 6.84 expresses profits
as a function of
i
P .
Price Adjustments
In section 6.6, we used this model to solve for an equilibrium under the assumption that
all firms adjusted prices to optimal levels.
4
We will now consider the profit consequences of failing to adjust prices when demand
(M) changes.
Profits when Price is not Adjusted
By assumption, in the original flexible price equilibrium other firms are charging P and
firm i is charging P P
i
= = . Now, when M changes, if firm i does not change price, then we
still have P P
i
= = and by equation (6.84) profits for firm i are:

+ +






= =
1
P
M
P
M
FIXED
(6.85)
Profits when Price is Adjusted:
When firm i adjusts its price, holding P fixed, it sets it to the profit maximizing value.
We also recall that the utility maximization problem for this model also involved
choosing price,
i
P . The first order condition implied this mark-up pricing result:
P
W
P
P
i
1
= =

(6.40)
Successively substituting (6.42a),

1
Y
P
W
= = , and
P
M
Y = = into this equation yields:

1
1







= =
P
M
P
P
i
(6.40a)
Recall (6.84):

+ +



















= =
P
P
P
M
P
P
P
M
i i
i
1
1
(6.84)
Now substituting

1
1







= =
P
M
P
P
i
into (6.84) yields:

+ +










































= =
1 1
1
1
1 1 P
M
P
M
P
M
P
M
ADJ
5
which simplifies to:

+ +
















= =
1
1 1
1
P
M
ADJ
(6.86)
We can now compare profits gained by the firm when it does and does not adjust price
when M changes.
A Numerical Example
Set the labor supply elasticity at 1 . 0 = = .
This implies that labor supply is not very elastic, in conformity with most micro
evidence.
Set the demand elasticity at 5 = = .
This implies that the price markup factor,
1

, is 1.25. See equation (6.40) on

the markup.
For the original equilibrium level of output, recall that
P
M
Y = = , impose the condition
P P
i
= = ,and use (6.40a):

1
1







= =
P
M
P
P
i

1
1
1 Y

= =

= =
1
Y
For these parameters, 978 . 0 = = Y .
Now suppose that there is a 3% fall in M. A comparison of (6.85) and (6.86) implies that:
253 . 0
FIXED ADJ
.
6
This implies that when a firm fails to change its price in response to a 3% decrease in M,
it loses profit that amounts to about one-fourth of its revenue! Menu costs will not be this
large, so the model seems to imply that prices will always be adjusted. What happened?
The problem is that when M rises and P is held fixed, output must rise. However, to
produce more output more labor must be supplied. To induce more work effort, the real
wage must rise. How much must the real wage rise? A 1.0% change in the real wage
induces a 0.10% change in labor suppled (and output produced). To get a 3.0% change in
labor supply (i.e., a 30 times bigger change), we therefore need a 30.0% change in the
real wage. Obviously it would be very costly to ignore a 30.0% change in the real wage
when setting prices so firms should not do it.
So how do we explain observed price inflexibility and business cycle effects which might
result from them?
More Model Alterations
The model we have employed above suggests that we are unlikely to observe price
stickiness as a consequence of imperfect competition when there are menu costs. The key
problem is that in this model real wages were flexible, and demand shifts have impacts on
wages that will not be ignored by firms considering price adjustments.
If we are to rationalize sticky prices, the determination of wages in our model will require
some modification. One possibility is simply that labor supply elasticities are much larger
than assumed in the quantitative example. In fact, short-run elasticities might be rather
large because of intertemporal substitution effects.
A second possibility is that workers are off of their labor supply curves at various times
over the business cycle. Romer Chapter 10 takes up models of the labor market that
imply that wages are acyclical over the business cycle.
A Quantitative Example with Real Wage Stickiness
Suppose that firms pay workers a wage that is above the market clearing level (this may
make workers value their jobs, hence they may shirk less, so it may not be irrational).
This assumption will imply that workers are willing to work more hours when asked to
do so by firms.
Furthermore, suppose that the real wage is now determined by:

AY
P
W
= = (6.87)
Note that this looks a lot like (6.43), but the parameters are now different. Using (6.87)
instead of (6.43) throughout the model, we can again derive an expression for profit, and
7
for the difference between profit when prices are allowed to adjust or are held fixed,
given a change in demand.
Set 1 . 0 = = , implying that real wages are not sensitive to the quantity of labor employed.
Set 806 . 0 = = A , so that the flexible price equilibrium output is 0.928 (this is 95% of
output in our previous calculation; output should be lower here because of the
supracompetitive wage assumption).
Now, if we assume a decline in M of 3%, and if other firms do not adjust its price, the
gain to firm i from adjusting price is 0.000168, a tiny fraction of its revenue.
So the combination of imperfect competition and real wage rigidity is sufficient to
explain nominal price rigidity even when menu costs are small.
Romer Chapter 9 pp. 388-398
Steady Inflation
I. Introduction to Inflation
A. Causes of Price Changes
1. G
2. T
3. M
4. Supply Shocks
B. Money is the only plausible explanation for sustained inflation.
II. Money Growth and Inflation in the Keynesian Framework
A. First recall the impact of a one-time increase in the money supply:
1. P and W go up in proportion to M.
2. Real variables are unchanged.
B. Now suppose that money is growing at a constant rate, .
C. Conjecture a steady state with the following properties:
1. Output is constant at Y (Assume classical aggregate supply).
2. P and W also grow at the rate .
3. Expected inflation,
e
, is equal to .
4. The real rate of interest,
e
i r = = , is invariant to money growth and
inflation.
D. Recall the equations of the classical model:
1. ) , ( Y i L
P
M
= =
2. ( ( ) ) T G i Y E E
e
, , , = =
3. Y E = =
4. ( ( ) ) L F Y = =
5. ( ( ) )
P
W
L F = =
6.





= =
P
W
L L
s
E. Analysis of the Steady State:
1. So long as W and P are growing together at rate , the real wage will be
constant, and it must be at the level that equates labor quantities demanded
and supplied (i.e. the same level as in the static classical model).
2. This implies that L and Y are also equal to the levels that prevail in the static
version of the model.
3. The real interest rate must be unchanged if the IS equation is to be satisfied
(Y, G, and T are unchanged, so
e
i must also be unchanged).
4. The nominal interest rate must vary point for point with expected inflation.
Thus, compared to the zero inflation case, the nominal interest rate is higher
by
e
units.
(1) Show this result in the IS-LM diagram. IS shifts up with an increase in
e
.
5. The LM equation tells us that
P
M
must be lower in the inflationary steady
state.
F. Moving from zero inflation to steady inflation:
1. Suppose that it is announced that the money growth rate will rise from 0 to
from this moment on, and that this is immediately believed.
2. P jumps up immediately as expectations change, then continues upward at rate
thereafter.
3. Note that there are two effects on P:
(1) An effect related to the change in expectations, which results in the higher
nominal rate of interest.
(2) The direct effects of money on prices.
4. If inflation expectations adjust gradually, the price level will gradually adjust
to its new steady state path. At some point inflation must exceed its steady
state value, however. This is inflation rate overshooting.
More on Efficiency Wages From Romer Chapter 9, pp. 417-421

Generalize the effort function to:

( , , )
a
e e w w u (9.9)

Recall the problem:

Max ( ) F eL wL

First order conditions (for choices of L and w) take the form:

( ) ( )
( )
' , ,
, ,
a
a
w
F e w w u L
e w w u
(9.10)

( )
( )
1
, ,
1
, ,
a
a
we w w u
e w w u
(9.11)

Example:

Let
if
0 otherwise
w x
w x
e
x

_
>

'
,

where

( ) 1
a
x bu w

For this functional form, the first order condition for w corresponding to (9.11) becomes
(show this!):

1
x
w

or

1
1
a
bu
w w

(9.15)

In equilibrium, each firm pays the prevailing wage, so
a
w w . Substituting into (9.15):

1 1 bu

bu

u
b

EQ
u
b

(9.17)

This is the equilibrium, or natural rate of unemployment.

According to (9.15), each firm would want to pay a wage (w) lower than that prevailing
(
a
w ) if u were higher than
EQ
u .

One can further show that the equilibrium effort and wage are:

1
EQ
e

,
(9.18)

(This condition above comes from the effort function, substituting the equilibrium
unemployment rate in the expression for x, and then simplifying )

( )
1
'
EQ EQ
EQ EQ
e u L
w e F
N
_

,
(9.19)

(This condition comes from 9.10, letting
( ) 1 u L
L
N

for each firms number employed;

where L is the labor force (fixed) and N is the number of firms).

Implications of this model:

Equilibrium unemployment depends only on the parameters of the effort function (not on
population or the production function) . See (9.17).

A modest can lead to a reasonable value for the natural rate of unemployment. E.g., if
1 b and 0.06 , then unemployment is 6% in equilibrium. (Note that is the
elasticity of effort with respect to the wage premium).

If unemployment rises in a recession, firms do not have much incentive to alter the real
wage rate (even though effort depends on unemployment). Romer shows this with a
numerical example, but the logic is familiar. If we start at the profit- maximizing w, then a
small change in w has a small impact on profit. Suppose unemployment rises. The firm
(optimally) would lower the wage it pays. If it fails to do so, it is paying a wage that is
too high. But by paying a wage that is too high workers choose to exert more effort,
which results in costs being lower. So the higher wage is largely offset by the cost-
reducing impact of more effort, and the gains from adjusting wages are small.

This then leads to a fairly reasonable explanation for real wage inflexibility in the face of
business cycle fluctuations. Recall that this is just what we needed for our New
Keynesian models, real wage inflexibility to accompany nominal price rigidity.

Unlike the simpler efficiency wage model, the chosen wage now does depend on
unemployment. Also the earlier efficiency wage model implied that the real wage and
employment were invariant to technical change and population growth, which would, in
turn, suggest that the unemployment rate might persistently rise or fall over time.

Romer on Time Inconsistency Problem

Lucas Supply Curve is

( )
e
y y b = +

Policymaker objective function is:

( ) ( )
2 2
* *
1 1
2 2
L y y a = + where
*
y y > .

Substitute the supply function into the objective function:

( ) ( ) ( )
2
2
* *
1 1
2 2
e
L y b y a = + +

First order condition is

( ) ( )
*
* 0
e
y b y b a

+ + =



Require
e
= (by rational expectations):

( )
*
* 0 y y b a + =



or

( )
*
* a y y b =


*
*
b
y y
a
=


*
*
b
y y
a
=


*
*
b
y y
a
= +