Economics 202A

Lecture Outline #5 (version 1.3)

Maurice Obstfeld
Endogenous Growth
We have already seen one crude endogenous growth model, the so-called
AK model. It is crude because it does not give a realistic account of the
channels through which productivity grows over time namely, innovation
and the creation of new knowledge.
We now turn to a class of models that indeed endogenize the innovative
process. The challenge in thinking about these problems is that the creation
of knowledge, which has a public-good aspect, is dierent fromthe production
of other economic goods.
The endogenous growth literature began with contributions of Robert
Lucas and especially Paul Romer in the 1980s and 1990s, although the ideas
certainly had important precursors in the growth literature of the 1960s.
A Model of Endogenous Growth: The Basic Idea
The model builds on some of the ideas about dierentiated products that
also underlie the new trade theory developed by Paul Krugman and others
in the late 1970s and early 1980s. In the model, additional varieties of
dierentiated capital goods will boost productivity, and the process through
which new capital goods are invented is endogenized.
In this economy, production of a nal consumption good is given by
1
t
= 1(1
1;t
. .... 1
At;t
. 1
Y;t
) =
_
At

j=1
1

j;t
_
1
1
Y;t
=
At

j=1
1

j;t
1
1
Y;t
.
where 1
Y;t
is the amount of labor employed in the nal goods sector at t and
, 2 f1. 2. ....
t
g indexes the dierent types of capital that can be used in
production as of t. Labor not devoted to nal-goods production will, as we
shall see, be devoted to research and development into new capital goods.
We assume that the capital depreciation rate is o = 1. so that the price
of a machine is its rental rate.
Note some interesting features of this production setup. At any point in
time, there are constant returns to scale with respect to the existing factors of
1
production, no matter how many there are. But while the marginal product
of an existing capital good is nite, the marginal product of a new capital
good is innite.
A dierent thought experiment gives a good illustration of why the pre-
ceding production function can generate endogenous growth. Imagine com-
bining 1 unit each of ` capital goods with 1 unit of Labor; we get 1 = `.
Instead, imagine we combine `,(` + 1) units each of ` + 1 capital goods
with 1 unit of labor. We get
1 =
N+1

j=1
_
`
` + 1
_

= (` + 1)
_
`
` + 1
_

= `
a
(` + 1)
1
`.
So with more capital goods, the output/labor ratio rises holding constant the
amount of capital input (measured in terms of consumption goods). Thus,
the creation of new capital goods has the potential to raise productivity and
per-worker output over time.
Notice, nally, that if 1
j;t
=
~
1
t
for all varieties , (as is the case in
equilibrium when all goods are symmetric), then
1
t
=
At

j=1
~
1

t
1
1
Y;t
=
t
~
1

t
1
1
Y;t
=
~
1

t
_

1
1
t
1
Y;t
_
1
.
so the production side looks equivalent to what we assumed for the Solow
model. What we will add, as we now show, is a model of how
t
grows
endogenously over time.
Production of Capital Goods and Blueprints for New Goods
To produce one unit of capital (of any kind) you need exactly one unit of
nal output. Capital goods are produced by monopolistic rms. To set up
a rm you need to purchase a blueprint for the specic variety , of capital
good you will produce. (The cost of the blueprint is sunk.) You can then
use a unit of output on date t to yield a unit of your capital good , on date
t + 1, which you sell (rent) at price j
j
.
We will assume that more labor devoted to research and development
(R&D) results in an expanded set of blueprints allowing the production of
more varieties of capital. Specically, if 1
A
is labor input to the R&D sector,

t+1

t
= o
t
1
A;t
. (1)
2
According to eq. (1), labor productivity in R&D is proportional to the exist-
ing stock of knowledge so in eect, we have learning by doing. This as-
sumption captures the important idea that, as a public good, new knowledge
is nonrival (more than one person can use it at the same time) and nonex-
cludable (people cannot be prevented from using knowledge). The learning
by doing is external to rms; each rm in R&D behaves competitively.
1
A blueprint can be put into use the very same period in which it is
developed. The total labor force 1 is constant and fully employed,
1 = 1
Y
+ 1
A
.
Solving the Model: First Steps
The key is to gure out how the labor force is divided between nal-goods
production and R&D. The more labor goes into R&D, the faster the growth
rate of the economy. The level of output of blueprints, in turn, depends on
their price in terms of nal goods, j
A
.
Let us conjecture that in equilibrium we will observe a constant real rate
of interest :, constant relative prices, a constant demand for each type of
capital, and a constant allocation of labor to sectors of the economy. (Later
we show that these guesses are all correct.) Let us start by considering the
demand of nal-goods rms for capital goods, given by the solution to
max
fK
j
g
At

j=1
1

j
1
1
Y

At

j=1
j
j
1
j
n1
Y
.
where j
j
(once again) is the output price of capital of type , and n is the
wage in terms of nal output. The rst-order condition for a maximum for
1
j
is
j
j
= c1
1
j
1
1
Y
. (2)
1
Think of the R&D sector as consisting of individual small competitive rms, each with
the production function
ow of new blueprints = 0
t
/
A;t
,
where /
A;t
is the rms labor input on date t and
t
is the economys stock of exisiting
blueprints. The individual rm takes f
t
g to be exogenous to its decisions.
3
Thus, the demand for a capital good is
2
1
j
=
_
c
j
j
_ 1
1
1
Y
.
What does this imply for producers of the intermediate capital goods?
The (intertemporal) prots of intermediate producer , are

j
=
j
j
1
j
1 + :
1
j
=
c1

j
1
1
Y
1 + :
1
j
.
Maximizing
j
with respect to 1
j
yields:
c
2
1
1
j
1
1
Y
1 + :
1 = 0
or

1 =
_
c
2
1 + :
_ 1
1

1
Y
(where the , subscript has been dropped, as all capital goods are symmetric).
Substituting this equation into eq. (2) yields the (constant) relative price of
a (generic) intermediate capital good:
j = c

1
1

1
1
Y
= c
_
_
c
2
1 + :
_ 1
1

1
Y
_
1

1
1
Y
=
1 + :
c
.
For a constant elasticity demand function, a standard result is that a
monopolists price is a constant markup over cost.
3
Here we see that
Price
Cost
=
j
1 + :
=
1
c
=
1
1
1
1
1
.
The cost of production is 1 on date t 1. and the price obtained (also from
the perspective of date t 1) is j,(1 + :).
2
We also see that (1 c)

At
j=1
1

j
1

Y
= n.
3
If the price elasticity of demand is j, the markup is j,(j 1), which goes to 1 as
j ! 1. In the present model, j = 1, (1 c).
4
Given all this, what is the prot that a capital-good producer earns? We
need to know this because the requirement that the stream of prots covers
sunk cost is what ties the model down. Substitution yields:

=
j

1
1 + :


1 =
_
j
1 + :
1
__
c
2
1 + :
_ 1
1

1
Y
=
_
1 c
c
__
c
2
1 + :
_ 1
1

1
Y
. (3)
There is free entry into producing intermediate goods, so the price of a blue-
print must equal the present discounted value of

above, or
j
A
=
1

t=0

(1 + :)
t
=
1 + :
:

=
1 + :
:
_
1 c
c
__
c
2
1 + :
_ 1
1

1
Y
(4)
An important point: if we did not have monopoly in the capital-producing
sector, there would be no stream of monopoly prots to cover the sunk cost of
blueprints, and so blueprints would never be purchased. In the market setting
we have assumed, monopoly and some degree of monopoly ineciency is
necessary to sustain positive growth.
Equilibrium Growth Rate
Equilibrium growth in the number of capital goods is given by
q =

t+1

t
= o

1
A
.
Production of each specic capital good will remain constant at

1.
What ties down the equilibrium allocation of labor, and hence q, is the
preceding eq. (4) for j
A
. Suppose there are too many workers in nal goods
production (relative to the equilibrium) because workers are paid more in
nal goods than in R&D. Then the demand for capital (to equip those work-
ers) will be high, raising the prots of intermediate producers and causing
them to bid up the price of blueprints j
A
. But that development, in turn will
raise the wages paid in the R&D sector, drawing workers out of nal goods.
The process will continue until wages in the two sectors are equal.
5
We formalize the requirement that workers have the same marginal value
product in both sectors by requiring that
MVPL in R&D = j
A
o = (1 c)1

Y
A

j=1

= (1 c)

1

Y
= n.
The solution is

1
Y
=
_
(1 c)
j
A
o
_1

1
=
_
:(1 c)
(1 + :)

o
_1

1
=
_
_
:(1 c)
(1 + :)
_
1

_ _

2
1+r
_
1
1

1
Y
o
_
_
1

_
c
2
1 + :
_ 1
1

1
Y
)
1 =
_
c:
(1 + :)

1
Y
o
_1

_
c
2
1 + :
_

)
1 =
_
c:
(1 + :)

1
Y
o
_ _
c
2
1 + :
_
1
)

1
Y
=
:
co
.
This is consistent, by the way, with the assumption we made that

1
Y
is
constant. We can now also nd the long-run rate of growth, which is
q = o

1
A
= o(1

1
Y
) = o1
:
c
. (5)
Notice that there is a scale eect here: a bigger work force implies more
innovation and hence faster growth. Higher interest rates retard growth
though we have yet to solve for the equilibrium rate of interest :.
Lets do so next. If the lifetime utility function of the representative
consumer is
l
0
=
1

t=0
,
t
C
1
1

t
1
1

.
then the intertemporal Euler equation is
C
1=
t
= ,(1 + :)C
1=
t+1
.
6
In balanced-growth equilibrium, consumption, like productivity, grows at the
(gross) rate 1 + q. so
1 + q =
C
t+1
C
t
= (1 + :)

.
which tells us the interest rate is
: =
(1 + q)
1

,
1.
Now combine this solution with eq. (5),
q = o1
1
c
_
(1 + q)
1

,
1
_
.
to infer the cni|i/:in: rate of growth as the solution to
c, q + (1 + q)
1

= , (1 + co1) .
The solution is illustrated for an arbitrary value of o by the intersection of
the two schedules in gure 7.14 of the Obstfeld-Rogo book. For example,
when o = 1. we can solve directly and one nds that
q =
c,o1 (1 ,)
1 + c,
.
Growth is higher for higher 1. c. ,. o, and o. (Why?) One also nds that
: =
c(1 + o1 ,)
1 + c,
.
Government policy can certainly aect the economic growth rate in this
model. For example, suppose the government imposes a xed fee t that
new rms have to pay for a license to enter the capital-goods industry. This
will increase the sunk cost of entry into the production of new capital goods.
The break-even condition, based on eqs. (3) and (4), now becomes
j
A
+ t =
1 + :
:
_
1 c
c
__
c
2
1 + :
_ 1
1

1
Y
.
Intuitively, as t rises from 0, j
A
will fall and

1
Y
, will rise. But with

1
A
=
1

1
Y
therefore lower, the pace of productivity growth will be lower as well.
7
Equilibrium versus Optimal Growth
Due to the presence of monopoly in the system, the equilibrium alloca-
tion is not Pareto ecient. Unless there are monopoly prots, producers
of capital goods cannot cover their sunk costs, so there is no demand for
blueprints, no innovation, and no growth. (Indeed, growth can never even
get started.) A second source of ineciency is the externality in the produc-
tion of blueprints. An omnipotent central planner, however, could achieve a
superior allocation by at: researchers would be ordered to produce the opti-
mal ow of blueprints, capital producers to use them to provide the socially
optimal level of each capital good. Growth would not simply be maximized
in the command allocation, however, because that would require too great
a sacrice of consumption. Instead, there is an optimal trade-o between
consumption and R&D. For example, in the case o = 1, the planner would
use the Lagrangian
L =
1

t=0
,
t
_
log
_

t
1

t
_
1 1
A;t
_
1

t+1
1
t+1
_
`
t
(
t+1

t
o
t
1
A;t
)
_
.
It turns out that the optimal growth rate is
q

= ,o1 (1 ,) q =
c,o1 (1 ,)
1 + c,
.
If the government subsidizes capital producers to allow them to produce
protably at social marginal cost, we get a growth rate of
q
subsidy
=
,o1 (1 ,)
1 + ,
< q

.
This exceeds the free-market growth rate because the monopoly distortion
has been xed, but falls short of the optimal growth rate because there is no
subsidy to make the R&D sector internalize the knowledge externality.
Kremers (1993) Paper
In a well-known paper, Michael Kremer examined the corollary of the
preceding type of model that growth is higher when population is higher.
He takes the position that knowledge diuses across borders, so that the
appropriate object of analysis is global population.
8
Technology growth in Kremers setup is

t+1

t
= o
t
1
t
where 1 is the global population/workforce. Here, people can have ideas
even while producing nal consumption goods. In the presence of some xed
factor such as land, output is given by
1
t
=
t
1
1
t
.
Population growth is endogenous in Kremers model and given by a
Malthusian assumption: population instantaneously rises to the point where
per capita output/consumption just equals the minimal level that sustains
life. Let us be a bit more generous and allow the minimal living standard to
be higher, and to rise over time in a way that (crudely) reect technology
C
min
_

t
=
1
t
1
t
.
Combining the last two equations yields

t
= (C
min
1

t
)
2
.
Substituting into
t+1

t
= o
t
1
t
leads to
1
2
t+1
1
2
t
= o1
1+2
t
.
or
1
t+1
1
t
= (1 + o1)
1
2
.
Population growth, according to this relation, should accelerate over time.
Kremer nds support for this prediction on data from 1 million B.C. through
1990. He also nds that between the disappearance of land bridges between
the continents and about 1500 (when the Age of Exploration began), the
larger continents had faster population growth. Assuming initial populations
were proportional to surface area, this prediction too conrms the theory.
9