MIT Economics Notes OCW

Chapter 7
Endogenous Growth II: R&D and

Technological Change
225
Economic Growth: Lecture Notes
7.1
Expanding Product Variety: The Romer Model
There are three sectors: one for the nal good sector, one for intermediate goods, and one for R&D.
The nal good sector is perfectly competitive and thus makes zero prots. Its output is used either
for consumption or as input in each of the other two sector.
The intermediate good sector is monopolistic. There is product dierentiation. Each intermediate
producer is a quasi-monopolist with respect to his own product and thus enjoys positive prots. To
become an intermediate producer, however, you must rst acquire a blueprint from the R&D sector.
A blueprint is simply the technology or know-how for transforming nal goods to dierentiated
intermediate inputs.
The R&D sector is competitive. Researchers produce blueprints. Blueprints are protected by
perpetual patents. Innovators auction their blueprints to a large number of potential buyers, thus
absorbing all the prots of the intermediate good sector. But there is free entry in the R&D sector,
which drive net prots in that sector to zero as well.
226
G.M. Angeletos
7.1.1
Technology
The technology for nal goods is given by a neoclassical production function of labor L and a
composite factor X :
Yt = F (Xt , Lt ) = A(Lt )1 (Xt ) .
The composite factor is given by a CES aggregator of intermediate inputs:
Xt =
1/
Nt
(Xt,j ) dj
where Nt denotes the number of dierent intermediate goods available in period t and Xt,j denotes
the quantity of intermediate input j employed in period t.
227
In what follows, we will assume = , which implies
Nt
Yt = A(Lt )
(Xt,j ) dj.
Note that = means the marginal product of each intermediate input is independent of the
quantity of other intermediate inputs:
Yt
= A
Xt,j
Lt
Xt,j
1
.
More generally, intermediate inputs could be either complements or substitutes, in the sense that the
marginal product of input j could depend either positively or negatively on Xt .
228
G.M. Angeletos
We will interpret intermediate inputs as capital goods and therefore let aggregate capital be given
by the aggregate quantity of intermediate inputs:
Nt
Kt =
Xt,j dj.
0
Finally, note that if Xt,j = X for all j and t, then

Yt = AL1
Nt X
t
and
Kt = Nt X,
implying
Yt = A(Nt Lt )1 (Kt )
or, in intensive form, yt = ANt1 kt . Therefore, to the extent that all intermediate inputs are used in
the same quantity, the technology is linear in knowledge N and capital K. Therefore, if both N and
K grow at a constant rate, as we will show to be the case in equilibrium, the economy will exhibit
long run growth, as in an AK model.
229
7.1.2
Final Good Sector
The nal good sector is perfectly competitive. Firms are price takers.
Final good rms solve
Nt
max Yt wt Lt
(pt,j Xt,j )dj
where wt is the wage rate and pt,j is the price of intermediate good j.
Prots in the nal good sector are zero, due to CRS, and the demands for each input are given by
the FOCs
wt =
Yt
Yt
= (1 )
Lt
Lt
and
pt,j
Yt
=
= A
Xt,j
for all j [0, Nt ].
230
Lt
Xt,j
G.M. Angeletos
7.1.3
Intermediate Good Sector
The intermediate good sector is monopolistic. Firms understand that they face a downward sloping
demand for their output.
The producer of intermediate good j solves
max t,j = pt,j Xt,j (Xt,j )
subject to the demand curve
Xt,j = Lt
A
pt,j
1
1
where (X) represents the cost of producing X in terms of nal-good units.
231
We will let the cost function be linear:

(X) = X.
The implicit assumption behind this linear specication is that technology of producing intermediate
goods is identical to the technology of producing nal goods. Equivalently, you can think of interme
diate good producers buying nal goods and transforming them to intermediate inputs. What gives
them the know-how for this transformation is precisely the blueprint they hold.
232
G.M. Angeletos
The FOCs give
pt,j = p
1
>1
for the optimal price, and

Xt,j = xL
for the optimal supply, where
1
x A 1 1 .
The resulting maximal prots are
t,j = L
where
(p 1)x =
1
x
233
1
2
1 1
A 1 .
Note that the price is higher than the marginal cost (p = 1/ > (X) = 1), the gap representing
the mark-up that intermediate-good rms charge to their customers (the nal good rms). Because
there are no distortions in the economy other than monopolistic competition in the intermediategood sector, the price that nal-good rms are willing to pay represents the social product of that
intermediate input and the cost that intermediate-good rms face represents the social cost of that
intermediate input. Therefore, the mark-up 1/ gives the gap between the social product and the
social cost of intermediate inputs.
Hint: The social planner would like to correct for this distortion. How?
234
G.M. Angeletos
7.1.4
The Innovation Sector
The present value of prots of intermediate good j from period t and on is given by
Vt,j =
q
=t
qt
,j
or
Vt,j = t,j +
Vt+1,j
1 + Rt+1
We know that prots are stationary and identical across all intermediate goods: t,j = L for all t, j.
As long as the economy follows a balanced growth path, we expect the interest rate to be stationary
as well: Rt = R for all t. It follows that the present value of prots is stationary and identical across
all intermediate goods:
Vt,j = V =
L
L
.
R
R/(1 + R)
Equivalently, RV = L, which has a simple interpretation: The opportunity cost of holding an

asset which has value V and happens to be a blueprint, instead of investing in bonds, is RV ; the
dividend that this asset pays in each period is L; arbitrage then requires the dividend to equal the
opportunity cost of the asset, namely RV = L.
235
New blueprints are produced using the same technology as nal goods: innovators buy nal goods
and transform them to blueprints at a rate 1/. It follows that producing an amount N of new
blueprints costs N, where > 0 measures the cost of R&D in units of output.
On the other hand, the value of these new blueprints is V N, where V = L/R.
It follows that net prots for a research rm are thus given by
prof itsR&D = (V ) N
Free entry in the sector of producing blueprints imposes prof itsR&D = 0, or equivalently
V = .
236
G.M. Angeletos
7.1.5
Households
Households solve
max
t u(ct )
t=0
s.t. ct + at+1 wt + (1 + Rt )at
As usual, the Euler condition gives
u (ct )
= (1 + Rt+1 ).
u (ct+1 )
And assuming CEIS, this reduces to
ct+1
= [(1 + Rt+1 )] .
ct
237
7.1.6
Resource Constraint
Final goods are used either for consumption by households (Ct ), or for production of intermediate
goods in the intermediate sector (Kt = j Xt,j ), or for production of new blueprints in the innovation
sector (Nt ). The resource constraint of the economy is therefore given by
Ct + Kt + Nt = Yt ,
where Ct = ct L, Nt = Nt+1 Nt , and Kt =
Nt
0
Xt,j dj.
As always, the sum of the budgets across agents together with the market clearing conditions reduce
to the resource constraint. Question: what are the market clearing conditions here? Related: what
are the assets traded by the agents?
238
G.M. Angeletos
7.1.7
General Equilibrium
Combining the formula for the value of innovation with the free-entry condition, we infer L/R =
V = . It follows that the equilibrium interest rate is
R=
L
=
1
2
1 1
A 1 L/,
which veries our earlier claim that the interest rate is stationary.
The Euler condition combined with the equilibrium condition for the real interest rate implies that
consumption grows at a constant rate, which is given by
Ct+1
= 1 + = [1 + R] = 1 +
Ct
239
1
2
1 1
A 1 L/
Next, note that the resource constraint reduces to
Nt+1
Ct
Yt
+
1 +X =
= AL1 X ,
Nt
Nt
Nt
where X = xL = Kt /Nt .
It follows that Ct /Nt is constant along the balanced growth path, and therefore Ct , Nt , Kt , and Yt all
grow at the same rate, , where, again,
1 + = 1 +
1
2
1 1
A 1 L/
The equilibrium growth rate of the economy decreases with , the cost of producing new knowl
edge.The growth rate is also increasing in L, or any other factor that increases the scale (size) of the
economy, and thereby raises the prots of intermediate inputs and the demand for innovation. This
is the (in)famous scale eect that is present in many models of endogenous technological change.
240
G.M. Angeletos
7.1.8
Eciency and Policy Implications
Consider now the problem of the social planner. He chooses {Ct , (Xt,j )j[0,Nt ] , Nt+1 }
t=0 so as to
maximize lifetime utility subject to the resource constraint that the technologies.
Obviously, due to symmetry in production, the social planner will choose the same quantity of
intermediate goods for all varieties: Xt,j = Xt = xt L for all j. Using this, we can write the problem
of the social planner as follows:
max
t u(ct ),
t=0
subject to
Ct + Nt Xt + (Nt+1 Nt ) = Yt = AL1 Nt Xt ,
where Ct = ct L.
241
The FOC with respect to Xt gives

Xt = x L,
where
x = A 1 1
represents the optimal level of production of intermediate inputs.
The Euler condition, on the other hand, gives the optimal growth rate as
1 + = [1 + R ] = 1 +
1
1
1 1
A 1 L/
where
R =
1
1
1 1
A 1 L/
represents that social return on savings.
242
G.M. Angeletos
Note that
1
x = x 1 > x
That is, the optimal level of production of intermediate goods is higher in the Pareto optimum than
in the market equilibrium. This reects simply the fact that, due to the monopolistic distortion,
production of intermediate goods is ineciently low in the market equilibrium. Naturally, the gap
x /x is an increasing function of the mark-up 1/.
Similarly,
1
R = R 1 > R.
That is, the market return on savings (R) falls short of the social return on savings (R ), the gap
again arising because of the monopolistic distortion in the intermediate good sector. It follows that
1 + > 1 + ,
so that the equilibrium growth rate is too low as compared to the Pareto optimal growth rate.
243
Policy exercise: Consider four dierent types of government intervention:

a subsidy on the production of nal goods
a subsidy on the demand for intermediate inputs
a subsidy on the production of intermediate inputs
a subsidy on R&D.
Which of these policies could achieve an increase in the market return and the equilibrium growth
rate? Which of these policies could achieve an increase in the output of the intermediate good sector?
Which one, or which combination of these policies, can implement the rst best allocation as a market
equilibrium?
244
G.M. Angeletos
7.1.9
Building on the Shoulders of Giants
In the original Romer (1990) model, the innovation sector uses a dierent technology than the
one assumed here. In particular, the technology for producing a new blueprint is linear in the
eective labor employed by the research rm, where eective means amount of labor (number of
researchers) times the existing stock of knowledge. Hence, for research rm j,
Nj,t = Lj,t Nt
The aggregate rate of innovation is thus given by
Nt = LtR&D Nt
where LR&D
is the total amount of labor employed in the R&D sector. Market clearing in the labor
t
. The private cost of innovation is now proportional to wt , while the
+ LR&D
=L
market is now Lnal
t
t
value of innovation remains as before. The rest of the model is also as before.
245
The key modication here in that the aggregate Nt enters the technology of producing new blueprints
new researchers build on the shoulders of previous researchers.
We have thus introduced a knowledge spillover that was absent in the previous model. When todays
research rms decide how much to spend on R&D, they do not internalize how this will improve the
know how of future innovators.
How does this aect the eciency/policy conclusions we derived in the previous model?
Imagine that the technology faced by research rm j was given by
Nj,t = (Lj,t )1 (Kj,t )2 (Nt )3
What are the restrictions for 1 , 2 and 3 that are necessary for perpetual growth? What are the
restrictions that are necessary for the individual rm to perceive constant returns to scale, and hence
for the R&D sector to be competitive?
246
G.M. Angeletos
7.2
A Simple Variant of Aghion-Howitt
The economy is populated by a large number of entrepreneurs. Each entrepreneur lives (is present
in the market) for 1+T periods, where T is random. Conditional on being alive in the present period,
there is probability n that the entrepreneur will die (exit the market) by the end of that period. n is
constant over time and independent of age. In each period, a mass n of existing entrepreneurs dies,
and a mass n of new entrepreneurs is born, so that the population is constant.
In the rst period of life, the entrepreneur is endowed with the aggregate level of knowledge in the
economy. In the rst period of life, he also has a fresh mind and can engage in R&D activity. In
later periods of life, instead, he is too old for coming up with good new ideas and therefore engages
only in production, not innovation.
Young producers engage in R&D in order to increase the prots of their own productive activities
later in life. But individual innovation has spillover eects to the whole economy. When a mass of
producers generate new ideas, the aggregate level of knowledge in the economy increases proportion
ally with the production of new ideas.
247
7.2.1
R&D Technology
j
Let Vt+1
denote the value of an innovation for individual j realized in period t and implemented in
period t + 1. Let z
tj denote the amount of skilled labor that a potential innovator j employes in R&D
and q(z
tj ) the probability that such R&D activity will be successful.
q : R
[0, 1] represents the
technology of producing innovations and satises q(0) = 0, q > 0 > q , q (0) = , q () = 0.
The potential researcher maximizes
j
q(z
tj )
Vt+1
wt ztj .
It follows that the optimal level of R&D is given by q (ztj )Vtj+1 = wt or
z
tj = g Vtj+1 /wt
where the function g(v) (q )1 (1/v) satises g(0) = 0, g > 0, g() = . Note that z will be
stationary only if both V and w grow at the same rate.
248
G.M. Angeletos
7.2.2
The Value of Innovation
What determines the value of an innovation? For a start, let us assume a very simple structure. Let
A
jt represent the TFP of producer j in period t. The prots from his production are given by
jt = A
jt
where
represents normalized prots. We can endogenize , but we wont do it here for simplicity.
When a producer is born, he automatically learns what is the contemporaneous aggregate level of
technology. That is, A
jt = At for any producer born in period t. In the rst period of life, and only
in that period, a producer has the option to engage in R&D. If his R&D activity fails to produce an
innovation, them his TFP remains the same for the rest of his life. If instead his R&D activity is
successful, then his TFP increases permanently by a factor 1 + , for some > 0.
249
That is, for any producer j born in period t, and for all periods t + 1 in which he is alive,
A
t
j
A =
(1 + )A
if his R&D fails

t
if his R&D succeeds
It follows that a successful innovation generates a stream of dividends equal to At

per period
for all > t that the producer is alive. Therefore,
Vt+1
1n
=
(At
) =
v At
1+R
=t+1
(7.1)
where where R is the interest rate per period and
1n
=1
1+R
.
R+n
Note that the above would be an exact equality if time was continuous. Note also that v is decreasing
in both R and n.
250
G.M. Angeletos
Remark: We see that the probability of death reduces the value of innovation, simply because it
reduces the expected life of the innovation. Here we have taken n as exogenous for the economy. But
later we will endogenize n. We will recognize that the probability of death simply the probability
that the producer will be displaced by another competitor who manages to innovate and produce a
better substitute product. For the time being, however, we treat n as exogenous.
7.2.3
The Cost of Innovation
Suppose that skilled labor has an alternative employment, which a simple linear technology of pro
ducing nal goods at the current level of aggregate TFP. That is, if lt labor is used in production of
nal goods, output is given by At lt . Since the cost of labor is wt , in equilibrium it must be that
wt = At .
251
(7.2)
7.2.4
Equilibrium
Combining (7.1) and (7.2), we infer that

Vt+1
=
v
wt
It follows that the level of R&D activity is the same across all new-born producers:
ztj = zt = g (v) .
The outcome of the R&D activity is stochastic for the individual. By the LLN, however, the aggregate
outcome is deterministic. The aggregate rate of innovation is simply
t = q(zt ) = (
v)
where (x) q (g (x)) .
252
G.M. Angeletos
It follows that the aggregate level of technology grows at a rate
At+1
= 1 + t = 1 + (
v) .
At
An increase in
increases the incentives for R&D in the individual level and therefore results to
higher rates of innovation and growth in the aggregate level. An increase in has a double eect.
Not only it increases the incentive for R&D, but it also increase the spill over eect from individual
innovations to the aggregate level of technology.
What is aggregate output in the economy? Its the sum of the output of all entrepreneurs plus the
output of workers not employed in R&D. Check that aggregate output grows at the same rate as
aggregate knowledge.
253
7.2.5
Business Stealing
Consider a particular market j, in which a producer j has monopoly power. Suppose now that there
is an outside competitor who has the option to engage in R&D in an attempt to create a better
product that is a close substitute for the product of producer j. Suppose further that, if successful,
the innovation will be so radical that, not only it will increase productivity and reduce production
costs, but it will also permit the outsider to totally displace the incumbent from the market.
Remark: Here we start seeing how both production and innovation may depend on the IO structure.
In more general versions of the model, the size of the innovation and the type of competition (e.g.,
Bertrand versus Cournot) determine what is the fraction of monopoly prots that the entrant can
grasp and hence the private incentives for innovation.
254
G.M. Angeletos
What is the value of the innovation for this outsider? Being an outsider, he has no share in the
market of product j. If his R&D is successful, he expects to displace the incumbent and grasp the
whole market of product j. That is, an innovation delivers a dividend equal to total market prots,
, in each period of life. Assuming that the outsider also has a probability of death (or
(1 + )At
displacement) equal to n, the value of innovation for the outsider is given by
out
Vt+1
1n
=
[(1 + )At ]
= (1 + )
vAt
1
+
R
=t+1
Now suppose that the incumbent also has the option to innovate is later periods of life. If he does so,
he will learn the contemporaneous aggregate level of productivity and improve upon it by a factor
1 + . The value of innovation in later periods of life is thus the same as in the rst period of life:
in
Vt+1
1n
=
[At
] = vA
t.
1
+
R
=t+1
255
out
in
Obviously, Vt+1
> Vt+1
. This is because the incumbent values only the potential increase in produc
tivity and prots, while the outsider values in addition the prots of the incumbent. This business
stealing eect implies that, ceteris paribus, that innovation will originate mostly in outsiders.
Remark: In the standard Aghion-Howitt model, as opposed to the variant considered here, only
outsiders engage in innovation. Think why this is the case in that model, and why this might not
be the case here. Then, nd conditions on the technology q and the parameters of the economy
that would ensure in our model a corner solution for the insiders and an interior solution for the
outsiders. (Hint: you may need to relax the Inada condition for q.) We will henceforth assume that
only outsiders engage in innovation.
Remark: Things could be dierent if the incumbent has a strong cost advantage in R&D, which
could be the case if the incumbent has some private information about the either the technology of
the product or the demand of the market.
256
G.M. Angeletos
out
Assuming that only outsiders engage in R&D, and using Vt+1
/wt = (1+)
v, we infer that the optimal
level of R&D for an outsider is

ztout = zt = g ((1 + )v) .
and therefore the aggregate rate of innovation is
t = q(zt ) = ((1 + )v)
We conclude that the growth rate of the economy is
yt+1
At+1
=
= 1 + ((1 + )
v) .
yt
At
257
We can now reinterpret the probability of death as simply the probability of being displaced by a
successful outside innovator. Under this interpretation, we have
n = ((1 + )
v)
and v solves
v =
R + ((1 + )
v)
Note that an increase in

will now increase v by less than one-to-one, because the displacement rate
will also increase.
258
G.M. Angeletos
7.2.6
Eciency and Policy Implications
Discuss the spillover eects of innovation... Both negative and positive...

Discuss optimal patent protection... Trade-o between incentives and externalities...
7.3
Ramsey Meets Schumpeter: The Aghion-Howitt Model

notes to be completed
259
7.4
Romer Meets Acemoglu: Biased Technological Change
7.4.1
Denition
Consider a two-factor economy, with

Yt = F (Lt , Ht , At )
where L and H denote, respectively, unskilled labor and skilled labor (or any two other factors) and
A denotes technology.
We say that technology is Hbiased if and only if
F (L, H, A) /H
F (L, H, A) /L
>0
Note that this is dierent from saying that technology is Haugmenting.
260
G.M. Angeletos
7.4.2
A simple model of biased technological change
We onsider a variant of the Romer model where we split the nal good sector in two sub-sectors, one
that is intensive in L and another that is intensive in H.
Aggregate output is given by
Yt = (YLt )
+ (1 ) (YHt )
where
NLt
YLt = L
(xLt )1 dj
YHt = H
NHt
261
(xHt )1 dj
The dierentiated intermediate input rms use blueprints to transform 1 unit of the nal good to
one unit of the dierentiated intermediate good.
The R&D rms transform nal goods to blueprints. Blueprints
The resource constraint is given by
Ct + Kt + L NLt + H NHt Yt ,
where
NLt
Kt =
xLt dj +
0
xHt dj.
0
262
NHt
G.M. Angeletos
Given the technologies, the skill premium is given by
wH
= const
wL
NH
NL
where
( 1) (1 ) .
263
H
L
The relative value of innovations is given by
VH
VL
pH
pL
H
L
1 1
NH
H
= const
.
NL
L
=
and the equilibrium innovation rates satisfy
NH
NL
= const
264
H
L
1
.
G.M. Angeletos
Hence, once we take into account the endogeneity of technologies, the equilibrium skill premium is
given by
= const
265
H
L
2
.
Finally, it is easy to show that the growth rate is given by
1
1
(1 ) (H H)1 + (L L)1 1 .
266
Chapter 3
The Neoclassical Growth Model
75
In the Solow model, agents in the economy (or the dictator) follow a simplistic linear rule for con
sumption and investment. In the Ramsey model, agents (or the dictator) choose consumption and
investment optimally so as to maximize their individual utility (or social welfare).
3.1
The Social Planner
In this section, we start the analysis of the neoclassical growth model by considering the optimal plan
of a benevolent social planner, who chooses the static and intertemporal allocation of resources in
the economy so as to maximize social welfare. We will later show that the allocations that prevail in
a decentralized competitive market environment coincide with the allocations dictated by the social
planner.
Together with consumption and saving, we also endogenize labor supply.
76
G.M. Angeletos
3.1.1
Preferences
Preferences are dened over streams of consumption and leisure, x = {xt }

t=0 , where xt = (ct , zt ),
and are represented by a utility function U : X R, where X is the domain of xt , such that
U (x) = U (x0 , x1 , ...)
We say that preferences are recursive if there is a function W : X R R (often called the utility
aggregator) such that, for all {xt }
t=0 ,
U (x0 , x1 , ...) = W [x0 , U (x1 , x2 , ...)]
We can then represent preferences as follows: A consumption-leisure stream {xt }
t=0
induces a utility
stream {Ut }
t=0 according to the recursion
Ut = W (xt , Ut+1 ).
77
We say that preferences are additively separable if there are functions t : X R such that
U (x) =
t (xt ).
t=0
We then interpret t (xt ) as the utility enjoyed in period 0 from consumption in period t + 1.
Throughout our analysis, we will assume that preferences are both recursive and additively separable.
In other words, we impose that the utility aggregator W is linear in ut+1 : There is a function
U : R R and a scalar R such that W (x, u) = U (x) + u. Hence,
Ut = U (xt ) + Ut+1 .
or, equivalently,
Ut =
=0
is called the discount factor, with (0, 1).

78
U (xt+ )
G.M. Angeletos
U is sometimes called the per-period utility or felicity function. We let z > 0 denote the maxi
mal amount of time per period. We accordingly let X = R+ [0, z]. We nally impose that U is
neoclassical, by which we mean that it satises the following properties:
1. U is continuous and (although not always necessary) twice dierentiable.
2. U is strictly increasing and strictly concave:
Uc (c, z) > 0 > Ucc (c, z)
Uz (c, z) > 0 > Uzz (c, z)
2
< Ucc Uzz
Ucz
3. U satises the Inada conditions
lim Uc =
and
lim Uz =
and
c 0
z 0
79
lim Uc = 0.
lim Uz = 0.
z z
3.1.2
Technology and the Resource Constraint
We abstract from population growth and exogenous technological change.

The time constraint is given by
zt + lt z.
We usually normalize z = 1 and thus interpret zt and lt as the fraction of time that is devoted to
leisure and production, respectively.
The resource constraint is given by
ct + it yt
80
G.M. Angeletos
Let F (K, L) be a neoclassical technology and let f () = F (, 1) be the intensive form of F. Output
in the economy is given by
yt = F (kt , lt ) = lt f (t ),
where
t =
kt
lt
is the capital-labor ratio.

Capital accumulates according to
kt+1 = (1 )kt + it .
(Alternatively, interpret l as eective labor and as the eective depreciation rate.)
Finally, we impose the following natural non-negativitly constraints:
ct 0, zt 0, lt 0, kt 0.
81
Combining the above, we can rewrite the resource constraint as

ct + kt+1 F (kt , lt ) + (1 )kt ,
and the time constraint as
zt = 1 lt ,
with
ct 0, lt [0, 1], kt 0.
82
G.M. Angeletos
3.1.3
The Ramsey Problem
The social planner chooses a plan {ct , lt , kt+1 }

t=0 so as to maximize utility subject to the resource
constraint of the economy, taking initial k0 as given:
max U0 =
t U (ct , 1 lt )
t=0
ct + kt+1 (1 )kt + F (kt , lt ), t 0,

ct 0, lt [0, 1], kt+1 0., t 0,
k0 > 0 given.
83
3.1.4
Optimal Control
Let t denote the Lagrange multiplier for the resource constraint. The Lagrangian of the social
planners problem is
L0 =
t U (ct , 1 lt ) +
t=0
t [(1 )kt + F (kt , lt ) kt+1 ct ]
t=0
Let t t t and dene the Hamiltonian as
Ht = H(kt , kt+1 , ct , lt , t ) U (ct , 1 lt ) + t [(1 )kt + F (kt , lt ) kt+1 ct ]
We can rewrite the Lagrangian as

L0 =
{U (ct , 1 lt ) + t [(1 )kt + F (kt , lt ) kt+1 ct ]} =
t=0
t=0
or, in recursive form, Lt = Ht + Lt+1 .

84
t Ht
G.M. Angeletos
Given kt , ct and lt enter only the period t utility and resource constraint; (ct , lt ) thus appears only in
Ht . Similarly, kt ,enter only the period t and t + 1 utility and resource constraints; they thus appear
only in Ht and Ht+1 .
Lemma 9 If {ct , lt , kt+1 }

t=0 is the optimum and {t }t=0 the associated multipliers, then
H
(ct , lt ) = arg max H(kt , kt+1 , c, l, t )

c,l
taking (kt , kt+1 ) as given, and

Ht + Ht+1
kt+1
= arg max
H(kt , k , ct , lt , t ) + H(k , kt+2 , ct+1 , lt+1 , t+1 )
taking (kt , kt+2 ) as given.

We henceforth assume an interior solution. As long as kt > 0, interior solution is indeed ensured by
the Inada conditions on F and U.
85
The FOC with respect to ct gives

L0
Ht
Ht
= t
=0
= 0 Uc (ct , zt ) = t
ct
ct
ct
The FOC with respect to lt gives
L0
Ht
Ht
= t
=0
= 0
Uz (ct , zt ) = t FL (kt , lt )
lt
lt
lt
Finally, the FOC with respect to kt+1 gives
Ht+1
Ht+1
L0
t Ht
=
+
= 0 t +
=0
kt+1
kt+1
kt+1
kt+1
t = [1 + FK (kt+1 , lt+1 )] t+1
86
G.M. Angeletos
Combining the above, we get

Uz (ct , zt )
= FL (kt , lt )
Uc (ct , zt )
and
Uc (ct , zt )
= 1 + FK (kt+1 , lt+1 ).
Uc (ct+1 , zt+1 )
Both conditions impose equality between marginal rates of substitution and marginal rate of transfor
mation. The rst condition means that the marginal rate of substitution between consumption and
leisure equals the marginal product of labor. The second condition means that the marginal rate of
intertemporal substitution in consumption equals the marginal capital of capital net of depreciation
(plus one). This last condition is called the Euler condition.
87
The envelope condition for the Pareto problem is
(max U0 )
L0
=
= 0 = Uc (c0 , z0 ).
k0
k0
More generally,
t = Uc (ct , lt )
represents the marginal utility of capital in period t and will equal the slope of the value function at
k = kt in the dynamic-programming representation of the problem.
88
G.M. Angeletos
Suppose for a moment that the horizon was nite, T < . Then, the Lagrangian would be L0 =
T
t
t=0 Ht and the Kuhn-Tucker condition with respect to kT +1 would give
L
HT
= T
0 and kT +1 0, with complementary slackness;
kT +1
kT +1
equivalently
T = T T 0 and kT +1 0, with T T kT +1 = 0.
The latter means that either kT +1 = 0, or otherwise it better be that the shadow value of kT +1 is
zero. When T = , the terminal condition T T kT +1 = 0 is replaced by the transversality condition
lim t t kt+1 = 0,
which means that the (discounted) shadow value of capital converges to zero. Equivalently,
lim t Uc (ct , zt )kt+1 = 0.
89
Proposition 10 The plan {ct , lt , kt }

t=0 is a solution to the social planners problem if and only if
Uz (ct , zt )
= FL (kt , lt ),
Uc (ct , zt )
Uc (ct , zt )
= 1 + FK (kt+1 , lt+1 ),
Uc (ct+1 , zt+1 )
kt+1 = F (kt , lt ) + (1 )kt ct ,
(3.1)
(3.2)
(3.3)
for all t 0, and

k0 > 0 given,
and
lim t Uc (ct , zt )kt+1 = 0.
(3.4)
Remark: We proved necessity of (3.1) and (3.2) essentially by a perturbation argument, and (3.3) is
just the constraint. We did not prove necessity of (3.4), neither suciency of this set of conditions.
See Acemoglu (2007) or Stokey-Lucas for the complete proof.
Note that the (3.1) can be solved for lt = l(ct , kt ), which we can then substitute into (3.2) and (3.3).
We are then left with a system of two dierence equations in two variables, namely ct and kt . The
intitial condition and the transversality condition then give the boundary conditions for this system.
90
G.M. Angeletos
3.1.5
Dynamic Programing
Consider again the social planners problem. For any k > 0, dene
V (k) max
t U (ct , 1 lt )
t=0
subject to
ct + kt+1 (1 )kt + F (kt , lt ), t 0,
ct , lt , (1 lt ), kt+1 0, t 0,
k0 = k given.
V is called the value function.
91
The Bellman equation for this problem is
V (k) = max U (c, 1 l) + V (k )

s.t.
c + k (1 )k + F (k, l)
k 0, c [0, F (k, 1)], l [0, 1].
Let
[c(k), l(k), G(k)] = arg max{...}.
These are the policy rules. The key policy rule is G, which gives the dynamics of capital. The other
rules are static.
92
G.M. Angeletos
Dene k by the unique solution to

k = (1 )k + F (k, 1)
and note that k represents an upper bound on the level of capital that can be sustained in any steady
state. Without serious loss of generality, we will henceforth restrict kt [0, k].
Let B be the set of continuous and bounded functions v : [0, k] R and consider the mapping
T : B B dened as follows:
T v(k) = max U (c, 1 l) + v(k )
s.t.
c + k (1 )k + F (k, l)
k [0, k], c [0, F (k, 1)], l [0, 1].
The conditions we have imposed on U and F imply that T is a contraction mapping. It follows
that T has a unique xed point V = T V and this xed point gives the solution.
93
The Lagrangian for the DP problem is

L = U (c, 1 l) + V (k ) + [(1 )k + F (k, l) k c]
The FOCs with respect to c, l and k give
L
= 0 Uc (c, z) =
c
L
= 0 Uz (c, z) = FL (k, l)
l
L
= 0 = Vk (k )
k
The Envelope condition is
Vk (k) =
L
= [1 + FK (k, l)]
94
G.M. Angeletos
Combining, we conclude
Uz (ct , lt )
= Fl (kt , lt )
Uc (ct , lt )
and
Uc (ct , lt )
= [1 + FK (kt+1 , lt+1 )] ,
Uc (ct+1 , lt+1 )
which are the same conditions we had derived with optimal control. Finally, note that we can state
the Euler condition alternatively as
Vk (kt )
= [1 + FK (kt+1 , lt+1 )].
Vk (kt+1 )
95
3.2
Decentralized Competitive Equilibrium
3.2.1
Households
Households are indexed by j [0, 1]. For simplicity, we assume no population growth.
The preferences of household j are given by
U
0j
t U (cjt , z
tj )
t=0
In recursive form, U
tj = U (c
jt , z
tj ) + Utj+1 .
The time constraint for household j can be written as
z
tj = 1 l
tj .
96
G.M. Angeletos
The budget constraint of household j is given by
cjt + i
jt + x
jt y
tj = rt ktj + Rt bjt + wt ltj + j t ,
where rt denotes the rental rate of capital, wt denotes the wage rate, Rt denotes the interest rate on
risk-free bonds. Household j accumulates capital according to
ktj+1 = (1 )ktj + i
jt
and bonds according to
bjt+1 = b
jt + x
jt
In equilibrium, rm prots are zero, because of CRS. It follows that t = 0. Combining the above
we can rewrite the household budget as
c
jt + ktj+1 + bjt+1 (1 + rt )ktj + (1 + Rt )bjt + wt ltj .
97
The natural non-negativity constraint

ktj+1 0
is imposed on capital holdings, but no short-sale constraint is imposed on bond holdings. That
is, household can either lend or borrow in risk-free bonds. We only impose the following natural
borrowing constraint:
(1 + Rt+1 )bjt+1 (1 + rt+1 )ktj+1 +
q
w .
q
t+1
=t+1
where
qt
= (1 + Rt )qt+1 .
(1 + R0 )(1 + R1 )...(1 + Rt )
This constraint simply requires that the net debt position of the household does not exceed the
present value of the labor income he can attain by working all time.
98
G.M. Angeletos
Simple arbitrage between bonds and capital implies that, in any equilibrium, the interest rate on
riskless bonds must equal the rental rate of capital net of depreciation:
Rt = rt .
If Rt < rt , all individuals would like to short-sell bonds, and there would be excess supply of
bonds. If Rt > rt , nobody in the economy would invest in capital.
Households are then indierent between bonds and capital. Letting ajt = bjt + ktj denote total assets,
the budget constraint reduces to
cjt + ajt+1 (1 + Rt )atj + wt ltj ,
and the natural borrowing constraint becomes ajt+1 at+1 , where

at+1
q w
qt =t+1
99
We assume that {Rt , wt }

t=0 satises
1
q w < M < ,
qt =t+1
for all t, so that at is bounded away from . Note in particular that if
=t+1 q w
at any t, the agent could attain innite consumption in every period t + 1.
100
was innite
G.M. Angeletos
j j
j
Given a price sequence {Rt , wt }

t=0 , household j chooses a plan {ct , l
t , k
t+1 }t=0 so as to maximize
lifetime utility subject to its budget constraints
max
U0j
t U (cjt , 1 l
tj )
t=0
s.t.
c
jt + a
jt+1 (1 + Rt )ajt + wt ltj
c
jt 0, l
tj [0, 1], a
jt+1 at+1
Let
jt = t jt be the Lagrange multiplier for the budget constraint, we can write the Lagrangian as
L
j0
j
j
j
j
j
j
j
U (ct , 1 lt ) +
t (1 + Rt )at + wt lt at+1 ct
=
t Htj
t
t=0
where
t=0
H
tj =
U (c
jt , 1 ltj ) +
jt (1 + Rt )ajt + wt ltj ajt+1 cjt
101
The FOC with respect to c

jt gives
j
Lj0
t H
t
=
=0
cjt
c
jt
Uc (cjt , z
tj ) =
jt
The FOC with respect to ltj gives
j
Lj0
t H
t
=
=0
ltj
l
tj
Combining, we get
Uz (cjt , z
tj ) =
jt wt
Uz (c
jt , z
tj )
= wt .
Uc (c
jt , z
tj )
That is, households equate their marginal rate of substitution between consumption and leisure with
the (common) wage rate.
102
G.M. Angeletos
The Kuhn-Tucker condition with respect to ajt+1 gives
j
j
H
Lj0
H
t
= t
+
jt+1 0
ajt+1
ajt+1
at+1
jt [1 + Rt ] jt+1 ,
with equality whenever ajt+1 > at+1 . That is, the complementary slackness condition is
jt [1 + Rt ] jt+1
ajt+1 at+1 = 0
Finally, if time was nite, the terminal condition would be
jT 0,
a
jT +1 aT +1 ,
jT ajT +1 aT +1 = 0,
where
jt t jt . Now that time is innite, the analogous condition is given by
lim t jt ajt+1 at+1 = 0
t0
103
Using
jt = Uc (cjt , z
tj ), we can restate the Euler condition as
Uc (cjt , z
tj ) [1 + Rt ]Uc (cjt+1 , ztj+1 ),
with equality whenever ajt+1 > at+1 . That is, as long as the borrowing constraint does not bind,
households equate their marginal rate of intertemporal substitution with the (common) return on
capital. On the other hand, if the borrowing constraint is binding, the marginal utility of consumption
today may exceed the marginal benet of savings: the household would like to borrow, but it cant.
For arbitrary borrowing limit at , there is nothing to ensure that the Euler condition must be satised
with equality. But if a
t is the natural borrowing limit, and the utility satises the Inada condition
Uc as c 0, then a simple argument ensures that the borrowing constraint can never bind.
Suppose that at+1 = at+1 . Then cj = z
j = 0 for all t, implying Uc (cjt+1 , ztj+1 ) = and therefore
necessarily Uc (cjt , z
tj ) < [1 + Rt ]Uc (cjt , z
tj ), unless also c
jt =
0 which in turn would be optimal only
if at = at . But this contradicts the Euler condition, proving that a0 > a0 suces for at > a
t for all
dates, and hence for the Euler condition to be satised with equality.
104
G.M. Angeletos
Moreover, if the borrowing constraint never binds, iterating jt = [1 + Rt ] jt+1 implies t jt = qt j0 .

We can therefore rewrite the terminal condition as
lim t jt ajt+1 = lim t jt at+1 = j0 lim qt at+1
But note that

qt at+1 =
q w
=t
and
=0 q w
< implies limt
=t q w
= 0. We thus arrive to the more familiar version of
the transversality condition:

lim t jt ajt+1 = 0,
or, equivalently,
lim t Uc (cjt , ztj )ajt+1 = 0.
105
It is useful to restate the households problem in a an Arrow-Debreu fashion:

max
t U (cjt , ztj )
t=0
s.t.
qt cjt
t=0
qt wt ztj x
t=0
where
x q0 (1 + R0 )a0 +
qt wt < .
t=0
Note that the intertemporal budget constraint is equivalent to the sequence of per-period budgets
together with the natural borrowing limit. The FOCs give
t Uc (cjt , ztj ) = qt
t Uz (cjt , ztj ) = qt wt ,
where > 0 is Lagrange multiplier associated to the intertermporal budget. You can check that
these conditions coincide with the one derived before.
106
G.M. Angeletos
Proposition 11 Suppose the price sequence {Rt , rt , wt }

t=0 satises Rt = rt for all t,
t=0 qt < , and
j j
j
t=0 qt wt < , . The plan {ct , l
t , a
t }t=0 solves the individual households problem if and only if
Uz (cjt , ztj )
= wt ,
Uc (cjt , ztj )
Uc (cjt , ztj )
= 1 + Rt ,
Uc (cjt+1 , ztj+1 )
c
jt + ajt+1 = (1 + Rt )ajt + wt ltj ,
l
tj + z
tj = 1,
for all t 0, with boundary condition
aj0 > 0 given
and
lim
t Uc (c
jt , z
tj )ajt+1 = 0.
j j
j
j
j
j
Given {a
jt }
t=1 , an optimal portfolio is any {kt , b
t }t=1 such that k
t 0 and b
t = a
t k
t .
Remark: For a more careful discussion on the necessity and suciency of these conditions, check
Stokey-Lucas.
107
3.2.2
Firms
There is an arbitrary number Mt of rms in period t, indexed by m [0, Mt ]. Firms employ labor and
rent capital in competitive labor and capital markets, have access to the same neoclassical technology,
and produce a homogeneous good that they sell competitively to the households in the economy.
Let Ktm and Lm
t denote the amount of capital and labor that rm m employs in period t. Then, the
prots of that rm in period t are given by
m
m
m
m
m
t = F (Kt , Lt ) rt Kt wt Lt .
The rms seeks to maximize prots. The FOCs for an interior solution require
FK (Ktm , Lm
t ) = rt .
FL (Ktm , Lm
t ) = wt .
108
G.M. Angeletos
As we showed before in the Solow model, under CRS, an interior solution to the rms problem to
exist if and only if rt and wt imply the same Ktm /Lm
t . This is the case if and only if there is some
Xt (0, ) such that
rt = f (Xt )
wt = f (Xt ) f (Xt )Xt
where f (k) F (k, 1). Provided so, rm prots are zero, m
t = 0, and the FOCs reduce to
Ktm = Xt Ltm .
That is, the FOCs pin down the capital labor ratio for each rm (Ktm /Lm
t ), but not the size of the
rm (Lm
t ). Moreover, because all rms have access to the same technology, they use exactly the same
capital-labor ratio. (See our earlier analysis in the Solow model for more details.)
109
3.2.3
Market Clearing
There is no supply of bonds outside the economy. The bond market thus clears if and only if
Lt
0=
0
bjt dj.
The capital market clears if and only if
Mt
0
Equivalently,
Mt
0
Ktm dm
Ktm dm = kt , where kt = Kt
1
0
ktj dj
ktj dj is the per-capita capital.
The labor market, on the other hand, clears if and only if
Mt
0
Equivalently,
Mt
0
Lm
t dm
Lm
t dm = lt where lt = Lt
Lt
0
110
=
0
Lt
ltj dj
ltj dj is the per-head labor supply.
G.M. Angeletos
3.2.4
General Equilibrium
Denition 12 An equilibrium of the economy is an allocation {(cjt , ltj , ktj+1 , bjt+1 )j[0,Lt ] , (Ktm , Lm
t )m[0,Mt ] }t=0
and a price path {Rt , rt , wt }

t=0 such that
j j
j
j
(i) Given {Rt , rt , wt }
t=0 , the path {ct , lt , kt+1 , bt+1 } maximizes the utility of of household j, for every j.
(ii) Given (rt , wt ), the pair (Ktm , Lm

t ) maximizes rm prots, for every m and t.
(iii) The bond, capital and labor markets clear in every period
Remark: In the above denition we surpassed the distribution of rm prots (or the stock market).
In the Solow model, we had showed that the decentralized market economy and the centralized
dictatorial economy were equivalent. A similar result holds in the Ramsey model. The following
proposition combines the rst and second fundamental welfare theorems, as applied to the Ramsey
model.
111
Proposition 13 The set of competitive equilibrium allocations for the market economy coincide with the
set of Pareto allocations for the social planner.
Proof.
I will sketch the proof assuming that (a) in the market economy, k0j + bj0 is equal across
all j; and (b) the social planner is utilitarian. For the more general case, we need to allow for an
unequal initial distribution of wealth across agents. The set of competitive equilibrium allocations
coincides with the set of Pareto optimal allocations, each dierent competitive equilibrium allocation
corresponding to a dierent point in the Pareto frontier (equivalently, a dierent vector of Pareto
weights in the objective of the social planner). For a more careful analysis, see Stokey-Lucas or
Acemoglu.
112
G.M. Angeletos
a. We rst consider how the solution to the social planners problem can be implemented as a
competitive equilibrium.
The social planners optimal plan is given by {ct , lt , kt }
t=0 such that
Uz (ct , 1 lt )
= FL (kt , lt ), t 0,
Uc (ct , 1 lt )
Uc (ct , 1 lt )
= [1 + FK (kt+1 , lt+1 )], t 0,

Uc (ct+1 , 1 lt+1 )
ct + kt+1 = (1 )kt + F (kt , lt ), t 0,
lim t Uc (ct , 1 lt )kt+1 = 0.
k0 > 0 given, and
Let t kt /lt and choose the price path {Rt , rt , wt }

t=0 given by
Rt = rt ,
rt = FK (kt , lt ) = f (t ),
wt = FL (kt , lt ) = f (t ) f (t )t ,
113
Trivially, these prices ensure that the FOCs are satised for every household and every rm if we set
cjt = ct , ltj = lt and Ktm /Lm
t = kt for all j and m. Next, we need to verify that the proposed allocation
satises the budget constraint of the households. From the resource constraint,
ct + kt+1 = F (kt , lt ) + (1 )kt .
From CRS and the FOCs for the rms, F (kt , lt ) = rt kt + wt lt . Combining, we get
ct + kt+1 = (1 + rt )kt + wt lt .
The budget constraint of household j is given by
cjt + ktj+1 + bjt+1 = (1 + rt )ktj + (1 + Rt )bjt + wt ltj ,
For this to be satised at the proposed prices with cjt = ct and ltj = lt , it is necessary and sucient
that ktj + bjt = kt for all j, t. Finally, it is trivial to check the bond, capital, and labor markets clear.
114
G.M. Angeletos
b. We next consider the converse, how a competitive equilibrium coincides with the Pareto solution.
Because agents have the same preferences, face the same prices, and are endowed with identical level
of initial wealth, and because the solution to the individuals problem is essentially unique (where
essentially means unique with respect to cjt , ltj , and ajt = ktj + bjt but indeterminate with respect to
the portfolio choice between ktj and bjt ), we have that cjt = ct , ltj = lt and ajt = at for all j, t. By the
FOCs to the individuals problem, it follows that {ct , lt , at }
t=0 satises
Uz (ct , 1 lt )
= wt , t 0,
Uc (ct , 1 lt )
Uc (ct , 1 lt )
= [1 + rt ], t 0,
Uc (ct+1 , 1 lt+1 )
ct + at+1 = (1 + rt )at + wt lt , t 0,
a0 > 0 given, and
lim t Uc (ct , 1 lt )at+1 = 0.
From the market clearing conditions for the capital and bond markets, the aggregate supply of bonds
is zero and thus at = kt .
115
Next, by the FOCs for the rms,

rt = FK (kt , lt )
wt = FL (kt , lt )
and by CRS
rt kt + wt lt = F (kt , lt )
Combining the above with the FOCs and the budget constraints gives
ct + kt+1 = F (kt , lt ) + (1 )kt , t 0,
which is simply the resource constraint of the economy. Finally, and limt t Uc (ct , 1 lt )at+1 =
0 with at+1 = kt+1 implies the social planners transversality condition, while a0 = k0 gives the
initial condition. This concludes the proof that the competitive equilibrium coincides with the social
planners optimal plan.
116
G.M. Angeletos
The equivalence to the planners problem then gives the following.

Corollary 14 (i) An equilibrium exists for any initial distribution of wealth. The allocation of production
across rms is indeterminate, and the portfolio choice of each household is also indeterminate, but the
equilibrium is unique as regards prices, consumption, labor, and capital. (ii) If initial wealth k0j + bj0 is
equal across all agent j, then cjt = ct , ltj = lt and ktj + bjt = kt for all j. The equilibrium is then given by an
allocation {ct , lt , kt }
t=0 such that, for all t 0,
Uz (ct , 1 lt )
= FL (kt , lt ),
Uc (ct , 1 lt )
Uc (ct , 1 lt )
= [1 + FK (kt+1 , lt+1 )],
Uc (ct+1 , 1 lt+1 )
kt+1 = F (kt , lt ) + (1 )kt ct ,
with k0 > 0 given and limt t Uc (ct , 1 lt )kt+1 = 0. Finally, equilibrium prices are given by
Rt = R(kt ) f (kt ) ,
rt = r(kt ) f (kt ),
117
wt = w(kt ) f (kt ) f (kt )kt .
3.3
Steady State
Proposition 15 There exists a unique (positive) steady state (c , l , k ) > 0. The steady-state values of
the capital-labor ratio, the productivity of labor, the output-capital ratio, the consumption-capital ratio, the
wage rate, the rental rate of capital, and the interest rate are all independent of the utility function U
and are pinned down uniquely by the technology F , the depreciation rate , and the discount rate . In
particular, the capital-labor ratio k /l equates the net-of-depreciation MPK with the discount rate,
f ( ) = ,
and is a decreasing function of + , where 1/ 1. Similarly,
R = ,
r = + ,
y
= f ( ),
l
Uz (c , 1 l )
,
Uc (c , 1 l )
c
y
=
,
k
k
w = FL ( , 1) =
y
= ( ),
k
where f () F (, 1) and () f ()/.

118
G.M. Angeletos
Proof. (c , l , k ) must solve

Uz (c , 1 l )
= FL (k , l ),
Uc (c , 1 l )
1 = [1 + FK (k , l )],
c = F (k , l ) k .
Let k/l denote the capital-labor ratio at the stead state. By CRS,
F (k, l) = f ()l
FK (k, l) = f ()
FL (k, l) = f () f ()
where f () F (, 1). The Euler condition then reduces to 1 = [1 + f ( )] or equivalently

f ( ) =
where 1/ 1. That is, the capital-labor ratio is pinned down uniquely by the equation of the
119
MPK, net of depreciation, with the discount rate. It follows that the gross rental rate of capital
and the net interest rate are r = + and R = , while the wage rate is w = FL: ( , 1). Labor
productivity (output per work hour) and the output-capital ratio are given by
y
= f ( ) and
l
y
= ( ),
k
where () f ()/. Finally, by the resource constraint, the consumption-capital ratio is given by
c
y
= ( ) = .
k
k
The comparative statics are trivial. For example, an increase in leads to an increase in , Y /L, and
s = K/Y . We could thus reinterpret the exogenous dierences in saving rates assumed in the Solow
model as endogenous dierences in saving rates originating in exogenous dierences in preferences.
Homework: consider the comparative statics with respect to exogenous productivity or a tax on
capital income.
120
G.M. Angeletos
3.4
Transitional Dynamics
Consider the condition that determined labor supply:
Uz (ct , 1 lt )
= FL (kt , lt ).
Uc (ct , 1 lt )
We can solve this for lt as a function of contemporaneous consumption and capital: lt = l(ct , kt ).
Substituting then into the Euler condition and the resource constraint, we conclude:
Uc (ct , 1 l(ct , kt ))
= [1 + FK (kt+1 , l(ct+1 , kt+1 ))]
Uc (ct , 1 l(ct , kt ))
kt+1 = F (kt , l(ct , kt )) + (1 )kt ct
This is a system of two rst-order dierence equation in ct and kt . Together with the initial condition
(k0 given) and the transversality condition, this system pins down the path of {ct , kt }
t=0 .
121
3.5
Exogenous labor and CEI
Suppose that leisure is not valued, or that the labor supply is exogenously xed. Either way, let lt = 1
for all t. Suppose further that preferences exhibit constant elasticity of intertemporal substitution:
U (c) =
c11/ 1
,
1 1/
where > 0 is the elasticity of intertemporal substitution.
The Euler condition then reduces to
ct+1
= [(1 + Rt+1 )] ,
ct
or equivalently ln(ct+1 /ct ) (Rt+1 ). Thus, controls the sensitivity of consumption growth to
the rate of return to savings
122
G.M. Angeletos
Proposition 16 The equilibrium path {ct , kt }

t=0 is given by the unique solution to
ct+1
= {[1 + f (kt+ ) ]} ,
ct
kt+1 = f (kt ) + (1 )kt ct ,
for all t, with initial condition k0 > 0 given and terminal condition
lim kt = k ,
where k is the steady state value of capital, that is, f (k ) = + .

Remark. That the transversality condition reduces to the requirement that capital converges to
the steady state will be argued later, with the help of the phase diagram. It also follows from the
following result, which uses information on the policy function.
123
Proposition 17 For any initial k0 < k (k0 > k ), the capital stock kt is increasing (respectively, de
creasing) over time and converges to asymptotically to k . Similarly, the rate of per-capita consumption
growth ct+1 /ct is positive and decreasing (respectively, negative and increasing) over time and converges
monotonically to 0.
Proof. The dynamics are described by kt+1 = G(kt ), where G is the policy rule characterizing the
planners problem. The policy rule is increasing and satises k = G(k) if and only if k = 0 or
k = k , k < G(k) < k for all k (0, k ), and k > G(k) > k for all k > k . (See Stokey-Lucas for
the proof of these properties.) The same argument as in the Solow model then implies that {kt }
t=0
is monotonic and converges to k . The monotonicity and convergence of {ct+1 /ct }
t=0 then follows
immediately from the monotonicity and convergence of {kt }

t=0 together with the fact that f (k) is
decreasing.
We will show this result also graphically in the phase diagram, below.
124
G.M. Angeletos
3.6
Continuous Time and Phase Diagram
Taking logs of the Euler condition and approximating ln = ln(1+ ) and ln[1 +f (kt )]
f (kt ) , we can write the Euler condition as
ln ct+1 ln ct [f (kt+1 ) ].
This approximation is exact when time is continuous.
Proposition 18 Consider the continuous-time version of the model. The equilibrium path {ct , kt }t[0,)
is the unique solution to
ct
= [f (kt ) ] = [Rt ],
ct
k t = f (kt ) kt ct ,
for all t, with k0 > 0 given and limt kt = k , where k is the steady-state capital.
125
We can now use the phase diagram to describe the dynamics of the economy. See Figure 3.1.
(Figure not shown due to unavailable original.)
The k = 0 locus is given by (c, k) such that

k = f (k) k c = 0
c = f (k) k
On the other hand, the c = 0 locus is given by (c, k) such that

c = c[f (k) ] = 0
k = k or c = 0
The steady state is simply the intersection of the two loci:

c = k = 0 {(c, k) = (c , k ) or (c, k) = (0, 0)}
where k (f )1 ( + ) and c f (k ) k .
We henceforth ignore the (c, k) = (0, 0) steady state and the c = 0 part of the c = 0 locus.
126
G.M. Angeletos
The two loci partition the (c, k) space in four regions. We now examine what is the direction of
change in c and k in each of these four regions.
Consider rst the direction of c.
If 0 < k < k [resp., k > k ], then and only then c > 0 [resp., c < 0].
That is, c increases [resp., decreases] with time whenever (c, k) lies the left [resp., right] of the c = 0
locus. The direction of c is represented by the vertical arrows in Figure 3.1.
If c < f (k) k [resp., c > f (k) k], then and only then k > 0
Consider next the direction of k.
[resp., k < 0]. That is, k increases [resp., decreases] with time whenever (c, k) lies below [resp., above]
the k = 0 locus. The direction of k is represented by the horizontal arrows in Figure 3.1.
We can now draw the time path of {kt , ct } starting from any arbitrary (k0 , c0 ), as in Figure 3.1.
Note that there are only two such paths that go through the steady state. The one with positive
slope represents the stable manifold or saddle path. The other corresponds to the unstable manifold.
The equilibrium path of the economy for any initial k0 is given by the stable manifold. That is, for
any given k0 , the equilibrium c0 is the one that puts the economy on the saddle path.
127
To understand why the saddle path is the optimal path when the horizon is innite, note the following:
Any c0 that puts the economy above the saddle path leads to zero capital and zero consumption
in nite time, thus violating the Euler condition at that time. Of course, if the horizon was
nite, such a path would have been the equilibrium path. But with innite horizon it is better
to consume less and invest more in period 0, so as to never be forced to consume zero at nite
time.
On the other hand, any c0 that puts the economy below the saddle path leads to so much capital
accumulation in the limit that the transversality condition is violated. Actually, in nite time
the economy has crossed the golden-rule and will henceforth become dynamically inecient.
Once the economy reaches kgold , where f (kgold ) = 0, continuing on the path is dominated
by an alternative feasible path, namely that of stopping investing in new capital and instead
consuming c = f (kgold ) kgold thereafter. In other words, the economy is wasting too much
resources in investment and it would better increase consumption.
128
G.M. Angeletos
Let the function c(k) represent the saddle path. In terms of dynamic programming, c(k) is simply
the optimal policy rule for consumption given capital k. Equivalently, the optimal policy rule for
capital accumulation is given by
k = f (k) k c(k),
with the discrete-time analogue being
kt+1 = G(kt ) f (kt ) + (1 )kt c(kt ).
Finally, note that, no matter the form of U (c), you can write the dynamics in terms of k and :
t
= f (kt )
t
k t = f (kt ) kt c(t ),
where c() solves Uc (c) = , that is, c() Uc1 (). Note that Ucc < 0 implies c () < 0. As an
exercise, draw the phase diagram and analyze the dynamics in terms of k and .
129
3.7
Comparative Statics and Impulse Responses
3.7.1
Additive Endowment
Suppose that each household receives an endowment e > 0 from God, so that its budget becomes
j
= wt + rt ktj + (1 )ktj + e
cjt + kt+1
Adding up the budget across households gives the new resource constraint of the economy
kt+1 kt = f (kt ) kt ct + e
On the other hand, optimal consumption growth is given again by
ct+1
= {[1 + f (kt+1 ) ]}
ct
130
G.M. Angeletos
Turning to continuous time, we conclude that the phase diagram becomes
ct
= [f (kt ) ],
ct
k t = f (kt ) kt ct + e.
In the steady state, k is independent of e and c moves one to one with e.
Consider a permanent increase in e by e. This leads to a parallel shift in the k = 0 locus, but no
change in the c = 0 locus. If the economy was initially at the steady state, then k stays constant
and c simply jumps by exactly e. On the other hand, if the economy was below the steady state, c
will initially increase but by less that e, so that both the level and the rate of consumption growth
will increase along the transition. See Figure 3.2.
c
ehigh
}e
{
Figure 3.2
131
elow
Figure by MIT OCW.
3.7.2
Taxation and Redistribution
Suppose that the government taxes labor and capital income at a at tax rate (0, 1). The
government then redistributes the proceeds from this tax uniformly across households. Let Tt be the
transfer made in period t.
The household budget is

cjt + ktj+1 = (1 )(wt + rt ktj ) + (1 )ktj + Tt ,
implying
Uc (cjt )
= [1 + (1 )rt+1 ].
Uc (cjt+1 )
That is, the tax rate decreases the private return to investment. Combining with rt = f (kt ) we infer
ct+1
= {[1 + (1 )f (kt+1 ) ]} .
ct
132
G.M. Angeletos
Adding up the budgets across household gives

ct + kt+1 = (1 )f (kt+1 ) + (1 )kt + Tt
The government budget on the other hand is
Tt =
j
(wt + rt ktj ) = f (kt )
Combining we get the resource constraint of the economy:

kt+1 kt = f (kt ) kt ct
Observe that, of course, the tax scheme does not appear in the resource constraint of the economy,
for it is only redistributive and does not absorb resources.
133
We conclude that the phase diagram becomes

ct
= [(1 )f (kt ) ],
ct
k t = f (kt ) kt ct .
In the steady state, k and c are decreasing functions of .
A. Unanticipated Permanent Tax Cut
Consider an unanticipated permanent tax cut that is enacted immediately. The k = 0 locus does not
change, but the c = 0 locus shifts right. The saddle path thus shifts right. See Figure 3.3.
A permanent tax cut leads to an immediate negative jump in consumption and an immediate positive
jump in investment. Capital slowly increases and converges to a higher k . Consumption initially is
lower, but increases over time, so soon it recovers and eventually converges to a higher c .
c
Thigh
Tlow
Figure by MIT OCW.

Figure 3.3
134
G.M. Angeletos
B. Anticipated Permanent Tax Cut

Consider a permanent tax cut that is (credibly) announced at date 0 to be enacted at some date
t > 0. The dierence from the previous exercise is that c = 0 locus now does not change immediately.
t and shifts right only for t >
t. Therefore, the dynamics of c and k will
It remains the same for t <
be dictated by the old phase diagram (the one corresponding to high ) for t <
t and by the new
t,
phase diagram (the one corresponding to low ) for t >
At t =
t and on, the economy must follow the saddle path corresponding to the new low , which
t, the economy must follow a path
will eventually take the economy to the new steady state. For t <
dictated by the old dynamics, but at t =
t the economy must exactly reach the new saddle path. If
t, which would violate the
that were not the case, the consumption path would have to jump at date
Euler condition (and thus be suboptimal). Therefore, the equilibrium c0 is such that, if the economy
follows a path dictated by the old dynamics, it will reach the new saddle path exactly at t =
t. See
Figure 3.4.
135
Tlow
Thigh
Figure 3.4
Figure by MIT OCW.
Following the announcement, consumption jumps down and continues to fall as long as the tax cut is
not initiated. The economy is building up capital in anticipation of the tax cut. As soon as the tax
cut is enacted, capital continues to increase, but consumption also starts to increase. The economy
then slowly converges to the new higher steady state.
136
G.M. Angeletos
3.7.3
Productivity Shocks: A prelude to RBC
We now consider the eect of a shock in total factor productivity (TFP). The reaction of the economy
in our deterministic framework is similar to the impulse responses we get in a stochastic Real Business
Cycle (RBC) model. Note, however, that here we consider the case that labor supply is exogenously
xed. The reaction of the economy will be somewhat dierent with endogenous labor supply, whether
we are in the deterministic or the stochastic case.
Let output be given by
yt = At f (kt )
where At denotes TFP. Note that

rt = At f (kt )
wt = At [f (kt ) f (kt )kt ]
so that both the return to capital and the wage rate are proportional to TFP.
137
We can then write the dynamics as

ct
= [At f (kt ) ],
ct
k t = At f (kt ) kt ct .
Note that TFP At aects both the production possibilities frontier of the economy (the resource
constrain) and the incentives to accumulate capital (the Euler condition).
In the steady state, both k and c are increasing in A.
A. Unanticipated Permanent Productivity Shock
The k = 0 locus shifts up and the c = 0 locus shifts right, permanently.
c0 may either increase or fall, depending on whether wealth or substitution eect dominates. Along
the transition, both c and k are increasing towards the new higher steady state. See Figure 3.5 for
the dynamics.
138
G.M. Angeletos
c
low EIS
high A
high EIS
low A
k
Figure 3.5 (A)
Figure by MIT OCW.
t
k,y
y
k
t
L = k + Sk
i
t
Figure 3.5 (B)
Figure by MIT OCW.
G.M. Angeletos
B. Unanticipated Transitory Productivity Shock
The k = 0 locus shifts up and the c = 0 locus shifts right, but only for t [0,
t] for some nite
t.
Again, c0 may either increase or fall, depending on whether wealth or substitution eects dominates.
I consider the case that c0 increases. A typical transition is depicted in Figure 3.6.
high A
low
low A
high A
Figure 3.6 (A)

Figure by MIT OCW.
139
c
t
k,y
y
k
t
c
i
Figure 3.6 (B)
Figure by MIT OCW.
3.7.4
Government Spending
We now introduce a government that collects taxes in order to nance some exogenous level of
government spending.
A. Lump Sum Taxation
Suppose the government nances its expenditure with lump-sum taxes. The household budget is
cjt + ktj+1 = wt + rt ktj + (1 )ktj Tt ,
implying that the Euler condition remains
Uc (cjt )
= [1 + rt+1 ] = [1 + f (kt+1 ) ]
Uc (cjt+1 )
That is, taxes do not aect the savings choice.
The government budget is Tt = gt , where gt denotes government spending.
140
G.M. Angeletos
The resource constraint of the economy becomes

ct + gt + kt+1 = f (kt ) + (1 )kt
Note that gt absorbs resources from the economy.
We conclude
ct
= [f (kt ) ],
ct
k t = f (kt ) kt ct gt
In the steady state, k is independent of g and c moves one-to-one with g. Along the transition, a
permanent increase in g both decreases c and slows down capital accumulation. See Figure 3.7.
Clearly, the eect of government spending nanced with lump-sum taxes is isomorphic to a negative
endowment shock.
141
B. Distortionary Taxation
Suppose the government nances its expenditure with distortionary income taxation. The household
budget is
cjt + ktj+1 = (1 )(wt + rt ktj ) + (1 )ktj ,
implying
Uc (cjt )
= [1 + (1 )rt+1 ] = [1 + (1 )f (kt+1 ) ].
Uc (cjt+1 )
That is, taxes now distort the savings choice.
The government budget is

gt = f (kt )
and the resource constraint of the economy is
ct + gt + kt+1 = f (kt ) + (1 )kt .
142
G.M. Angeletos
We conclude
ct
= [(1 )f (kt ) ],
ct
k t = (1 )f (kt ) kt ct .
In the steady state, k is a decreasing function of g (equivalently, ) and c decreases more than
one-to-one with g. Along the transition, a permanent increase in g (and ) drastically slows down
capital accumulation. See Figure 3.7. (Figure not shown due to unavailable original.)
Clearly, the eect of government spending nanced with distortionary taxes is isomorphic to a neg
ative TFP shock.
143
3.8
Beyond Growth
3.8.1
The Phase Diagram with Endogenous Labor Supply
Suppose separable utility, U (c, z) = u(c) + v(z), with CEIS for u, and let l(k, c) be the solution to
v (1 l)
= FL (k, l)
u (c)
Note that l increases with k, but less than one-to-one (or otherwise FL would fall). This reects the
substitution eect. On the other hand, l falls with c, reecting the wealth eect.
Substitute back into the dynamic system for k and c, assuming CEIS preferences:
ct
= [f (kt /l(kt , ct )) ],
ct
k t = f (kt , l(kt , ct )) kt ct ,
which gives a system in kt and ct alone.
144
G.M. Angeletos
Draw suggestive phase diagram. See Figure 3.8. (Figure not shown due to unavailable original.)
Note that the c is now negatively sloped, not vertical as in the model with exogenously xed labor.
This reects the wealth eect on labor supply. Lower c corresponds to lower eective wealth, which
results to higher labor supply for any given k (that is, for any given wage).
3.8.2
Impulse Responses Revisited
Note that the endogeneity of labor supply makes the Euler condition (the c locus) sensitive to wealth
eects, but also mitigates the impact of wealth eects on the resource constraint (the k locus).
Reconsider the impulse responces of the economy to shocks in productivity or governement spending.
Government spending.... If nanced with lump sum taxes, an increase in g has a negative wealth
eect, which increases labor supply. This in turn leads an increase in the MPK and stimulates more
investment. At the new steady state the capital-labor ratio remains the same, as it is simply the one
that equates the MPK with the discount rate, but both employment and the stock of capital go up...
145
Note that the above is the supply-side eect of government spending. Contrast this with the demandside eect in Keynesian models (e.g., IS-LM).
Productivity shocks....
3.8.3
The RBC Propagation Mechanism, and Beyond
Just as we can use the model to explain the variation of income and productivity levels in the
cross-section of countries (i.e., do the Mankiw-Romer-Weil exercise), we can also use the model to
explain the variation of income, productivity, investment and employment in the time-series of any
given country. Hence, the RBC paradigm.
The heart of the RBC propagation mechanism is the interaction of consumption smoothing and
deminishing returns to capital accumulation. Explain....
This mechanism generates endogenous persistence and amplication. Explain...
Enogenous persistence is indeed the other face of conditional convergence. But just as the model
146
G.M. Angeletos
fails to generate a substantially low rate of conditional convergence, it also fails to generate either
substantial persistence or substantial amplication. For the model to match the data, we then need
to assume that exogenous productivity (the Solow residual) is itself very volatile and persistent. But
then we partly answer and partly peg the question.
Hence the search for other endogenous propagation mechanisms.
Discuss Keynesian models and monopolistic competition... Discuss the potential role nancial mar
kets...
147
Chapter 1
Introduction and Growth Facts
1.1
Introduction
In 2000, GDP per capita in the United States was $32500 (valued at 1995 $ prices). This high income
level reects a high standard of living.
In contrast, standard of living is much lower in many other countries: $9000 in Mexico, $4000 in
China, $2500 in India, and only $1000 in Nigeria (all gures adjusted for purchasing power parity).
How can countries with low level of GDP per person catch up with the high levels enjoyed by the
United States or the G7?
Only by high growth rates sustained for long periods of time.
Small dierences in growth rates over long periods of time can make huge dierences in nal outcomes.
US per-capita GDP grew by a factor 10 from 1870 to 2000: In 1995 prices, it was $3300 in 1870
and $32500 in 2000.1 Average growth rate was 1.75%. If US had grown with .75% (like India,
1
Let y0 be the GDP per capital at year 0, yT the GDP per capita at year T, and x the average annual growth rate over that
period. Then, yT = (1+x)T y0 . Taking logs, we compute ln yT ln y0 = T ln(1+x) T x, or equivalenty x (ln yT ln y0 )/T.
G.M. Angeletos
Pakistan, or the Philippines), its GDP would be only $8700 in 1990 (i.e., 1/4 of the actual one,
similar to Mexico, less than Portugal or Greece). If US had grown with 2.75% (like Japan or Taiwan),
its GDP would be $112000 in 1990 (i.e., 3.5 times the actual one).
At a growth rate of 1%, our children will have 1.4 our income. At a growth rate of 3%, our children
will have 2.5 our income. Some East Asian countries grew by 6% over 1960-1990; this is a factor
of 6 within just one generation!!!
Once we appreciate the importance of sustained growth, the question is natural: What can do to
make growth faster?
Equivalently: What are the factors that explain dierences in economic growth, and how can we
control these factors?
In order to prescribe policies that will promote growth, we need to understand what are the deter
minants of economic growth, as well as what are the eects of economic growth on social welfare.
Thats exactly where Growth Theory comes into picture...
3
1.2
The World Distribution of Income Levels and Growth Rates
As we mentioned before, in 2000 there were many countries that had much lower standards of living
than the United States. This fact reects the high cross-country dispersion in the level of income.
(in what follows, all gures are reproduced from Daron Acemoglus textbook)
Figure 1.1 shows the distribution of (the log of) GDP per capita in 1960, 1980, and 2000 across the
147 countries in the Summers and Heston dataset.
In 2000, the richest country was Luxembourg, with $44000 GDP per person. The United States
came second, with $32500. The G7 and most of the OECD countries ranked in the top 25 positions,
together with Singapore, Hong Kong, Taiwan, and Cyprus. Most African countries, on the other
hand, fell in the bottom 25 of the distribution. Tanzania was the poorest country, with only $570 per
personthat is, less than 2% of the income in the United States or Luxemburg! In 1960, on the other
hand, the richest country then was Switzerland, with $15000; the United States was again second,
with $13000, and the poorest country was again Tanzania, with $450.
4
.4
G.M. Angeletos
1960
Density of coutries
.2
.3
2000
.1
1980
8
9
log gdp per capita
10
11
Figure 1.1: Estimates of the distribution of countries according to log GDP per capita (PPP-adjusted) in
1960, 1980 and 2000.
Courtesy of K. Daron Acemoglu. Used with permission.
The cross-country dispersion of income was thus as wide in 1960 as in 2000. Nevertheless, there were
some important movements during this 40-year period. Argentina, Venezuela, Uruguay, Israel, and
South Africa were in the top 25 in 1960, but none made it to the top 25 in 2000. On the other hand,
China, Indonesia, Nepal, Pakistan, India, and Bangladesh grew fast enough to escape the bottom
25 between 1960 and 1970. These large movements in the distribution of income reects sustained
dierences in the rate of economic growth.
Figure 1.2 shows the distribution of the growth rates the countries experienced between 1960 and
2000. Just as there is a great dispersion in income levels, there is a great dispersion in growth rates.
The mean growth rate was 1.8% per annum; that is, the world on average was twice as rich in 2000
as in 1960. The United States did slightly better than the mean. The fastest growing country was
Taiwan, with a annual rate as high as 6%, which accumulates to a factor of 10 over the 40-year period.
The slowest growing country was Zambia, with an negative rate at 1.8%; Zambias residents show
their income shrinking to half between 1960 and 2000.
Figure 1.3 shows an example of how persistent the dierences in growth rates were across countries.
6
20
G.M. Angeletos
1960
Density of coutries
10
15
1980
2000
.1
.05
0
average growth rates
.05
.1
Figure 1.2: Estimates of the distribution of countries according to the growth rate of GDP per worker
(PPP-adjusted) in 1960, 1980 and 2000.
10
UK
South Korea
Spain
Brazil
Singapore
Guatemala
Botswana
India
log gdp per capita

8
9
USA
Nigeria
1960
1970
1980
year
1990
2000
Figure 1.3: The evolution of income per capita in the United States, United Kingdom, Spain, Singapore,
Brazil, Guatemala, South Korea, Botswana, Nigeria and India, 1960-2000.
G.M. Angeletos
Most East Asian countries (Taiwan, Singapore, South Korea, Hong Kong, Thailand, China, and
Japan), together with Bostwana (an outlier for sub-Saharan Africa), Cyprus, Romania, and Mauritus,
had the most stellar growth performances; they were the growth miracles of our times. Some OECD
countries (Ireland, Portugal, Spain, Greece, Luxemburg and Norway) also made it to the top 20 of the
growth-rates chart. On the other hand, 18 out of the bottom 20 were sub-Saharan African countries.
Other notable growth disasters were Venezuela, Chad and Iraq.
1.3
Unconditional versus Conditional Convergence
Figure 1.4 graphs a countrys GDP per worker in 1988 (normalized by the US level) against the
same countrys GDP per worker in 1960. Clearly, most countries did not experienced a dramatic
change in their relative position in the world income distribution. Therefore, although there are
important movements in the world income distribution, income and productivity dierences tend to
be very persistent.
12
LUX
log gdp per worker 2000

9
10
11
IRL
USA
NOR NLD
ITA BEL
CAN
AUS
AUT
FINFRA DNK
SWE CHE
ISL
GBR
ESP ISR
NZL
JPN
KOR
GRC
PRT
BRB
MUS
MYS
TTO
ARG
CHL
MEX
SYC
ZAF
URY
BRA IRN
VEN
GAB
DOMPAN JOR
SYR
TUR
CRI
EGY
GTM SLV
MAR ECU COL
PRY PER
PHL
JAM
BOL
HND
GIN
NIC
HKG
THA
CPV
ROM IDN
LKA
PAK
BGD
IND
CHN
ZWE
CIV
CMR
COG
COM
NPL
SEN
LSO
GHA
GMB
ZMB
TCD
BEN
KEN
TGO MOZ
MLI
BFA
MDG
MWI UGA
NER
RWA
GNB
ETH
NGA
TZA BDI
8
9
10
Figure 1.4: Log GDP per worker in 2000 versus log GDP per worker in 1960.
10
G.M. Angeletos
This also means that poor countries on average do not grow faster than rich countries. And another
way to state the same fact is that unconditional convergence is zero. That is, if we ran the regression
ln y20001960 = + ln y1960 ,
the estimated coecient is zero.
On the other hand, consider the regression
ln y196090 = + ln y1960 + X1960
where X1960 is a set of country-specic controls, such as levels of education, scal and monetary
policies, market competition, etc. Then, the estimated coecient turns to be around 2% per
annum. Therefore, poor countries tend to catch up with the rich countries with a group of countries
that share similar characteristics. This is what we call conditional convergence.
Conditional convergence is illustrated in Figure 1.5, for the group of OECD countries.
11
JPN
IRL
PRT
annual growth rate 19602000

.02
.03
.04
LUX
ESP
GRC
AUT
ITA
FIN
FRA
BEL
NOR
ISL
GBR
USA
DNK
NLD
AUS
CAN
.01
SWE
CHE
NZL
9.5
10
10.5
Figure 1.5: Annual growth rate of GDP per worker between 1960 and 2000 versus log GDP per worker in
1960 for core OECD countries.
12
Chapter 2
The Solow Growth Model
13
2.1
Centralized Dictatorial Allocations
In this section, we start the analysis of the Solow model by pretending that there is a dictator, or
social planner, that chooses the static and intertemporal allocation of resources and dictates that
allocations to the households of the economy We will later show that the allocations that prevail in
a decentralized competitive market environment coincide with the allocations dictated by the social
planner.
2.1.1
The Economy, the Households and the Dictator
Time is discrete, t {0, 1, 2, ...}. You can think of the period as a year, as a generation, or as any
other arbitrary length of time.
The economy is an isolated island. Many households live in this island. There are no markets and
production is centralized. There is a benevolent dictator, or social planner, who governs all economic
and social aairs.
14
G.M. Angeletos
There is one good, which is produced with two factors of production, capital and labor, and which
can be either consumed in the same period, or invested as capital for the next period.
Households are each endowed with one unit of labor, which they supply inelasticly to the social
planner. The social planner uses the entire labor force together with the accumulated aggregate
capital stock to produce the one good of the economy.
In each period, the social planner saves a constant fraction s (0, 1) of contemporaneous output, to
be added to the economys capital stock, and distributes the remaining fraction uniformly across the
households of the economy.
In what follows, we let Lt denote the number of households (and the size of the labor force) in period t,
Kt aggregate capital stock in the beginning of period t, Yt aggregate output in period t, Ct aggregate
consumption in period t, and It aggregate investment in period t. The corresponding lower-case
variables represent per-capita measures: kt = Kt /Lt , yt = Yt /Lt , it = It /Lt , and ct = Ct /Lt .
15
2.1.2
Technology and Production
The technology for producing the good is given by

Yt = F (Kt , Lt )
(2.1)
where F : R2+ R+ is a (stationary) production function. We assume that F is continuous and

(although not always necessary) twice dierentiable.
16
G.M. Angeletos
We say that the technology is neoclassical if F satises the following properties

1. Constant returns to scale (CRS), or linear homogeneity:
F (K, L) = F (K, L),
> 0.
2. Positive and diminishing marginal products:
FK (K, L) > 0,
FL (K, L) > 0,
FKK (K, L) < 0,
FLL (K, L) < 0.
where Fx F/x and Fxz 2 F/(xz) for x, z {K, L}.

3. Inada conditions:
lim FK = lim FL = ,
K 0
L0
lim FK =
17
lim FL = 0.
By implication, F satises
Y = F (K, L) = FK (K, L)K + FL (K, L)L
or equivalently
1 = K + L
where
K
F K
K F
and
F L
L F
Also, FK and FL are homogeneous of degree zero, meaning that the marginal products depend only
on the ratio K/L.
And, FKL > 0, meaning that capital and labor are complementary.
Finally, all inputs are essential: F (0, L) = F (K, 0) = 0.

18
G.M. Angeletos
Technology in intensive form: Let y Y /L and k K/L. Then, by CRS

(2.2)
y = f (k) F (k, 1)
By denition of f and the properties of F, we have that f (0) = 0,
f (k) > 0 > f (k)
lim f (k) =
k0
FK (K, L) = f (k)
lim f (k) = 0
FL (K, L), = f (k) f (k)k
Example: Cobb-Douglas technology

F (K, L) = K L1
In this case, K = , L = 1 , and
f (k) = k
19
2.1.3 The Resource Constraint, and the Law of Motions for Capital and
Labor (Population)
The sum of aggregate consumption and aggregate investment can not exceed aggregate output. That
is, the social planner faces the following resource constraint:
Ct + It Yt
(2.3)
ct + it yt
(2.4)
Equivalently, in per-capita terms:
We assume that population growth is n 0 per period:

Lt = (1 + n)Lt1 = (1 + n)t L0
We normalize L0 = 1.
20
(2.5)
G.M. Angeletos
Suppose that existing capital depreciates over time at a xed rate [0, 1]. The capital stock in
the beginning of next period is given by the non-depreciated part of current-period capital, plus
contemporaneous investment. That is, the law of motion for capital is
Kt+1 = (1 )Kt + It .
(2.6)
Equivalently, in per-capita terms:

(1 + n)kt+1 = (1 )kt + it
We can approximately write the above as
kt+1 (1 n)kt + it
(2.7)
The sum + n can thus be interpreted as the eective depreciation rate of per-capita capital.
(Remark: This approximation becomes exact in the continuous-time version of the model.)
21
2.1.4
The Dynamics of Capital and Consumption
In most of the growth models that we will examine in this class, the key of the analysis will be to
derive a dynamic system that characterizes the evolution of aggregate consumption and capital in
the economy; that is, a system of dierence equations in Ct and Kt (or ct and kt ). This system is
very simple in the case of the Solow model.
Combining the law of motion for capital (2.6), the resource constraint (2.3), and the technology (2.1),
we derive the dierence equation for the capital stock:
Kt+1 Kt F (Kt , Lt ) Kt Ct
(2.8)
That is, the change in the capital stock is given by aggregate output, minus capital depreciation,
minus aggregate consumption.
kt+1 kt f (kt ) ( + n)kt ct .
22
G.M. Angeletos
2.1.5
Feasible and Optimal Allocations
2
Denition 1 A feasible allocation is any sequence {ct , kt }
that satises the resource constraint
t=0 R+
kt+1 f (kt ) + (1 n)kt ct .
(2.9)
The set of feasible allocations represents the choice set for the social planner. The planner then
uses some choice rule to select one of the many feasible allocations.
Later, we will have to social planner choose an allocation so as to maximize welfare (Pareto eciency).
Here, we instead assume that the dictator follows a simple rule-of-thump.
Denition 2 A Solow-optimal centralized allocation is any feasible allocation that satises the resource
constraint with equality and
ct = (1 s)f (kt ),
for some s (0, 1).
23
(2.10)
2.1.6
The Policy Rule
Combining (2.9) and (2.10) gives a single dierence equation that completely characterizes the dy
namics of the Solow model.
Proposition 3 Given any initial point k0 > 0, the dynamics of the dictatorial economy are given by the
path {kt }
t=0 such that
kt+1 = G(kt ),
(2.11)
for all t 0, where

G(k) sf (k) + (1 n)k.
Equivalently, the growth rate is given by
t
kt+1 kt
= (kt ),
kt
where
(k) s(k) ( + n), (k) f (k)/k.
24
(2.12)
G.M. Angeletos
G corresponds to what we will call the policy rule in the Ramsey model. The dynamic evolution of
the economy is concisely represented by the path {kt }
t=0
that satises (??), or equivalently (2.11),
for all t 0, with k0 historically given.
The graph of G is illustrated in Figure 2.1.

Remark. Think of G more generally as a function that tells you what is the state of the economy
tomorrow as a function of the state today. Here and in the simple Ramsey model, the state is simply
kt . When we introduce productivity shocks, the state is (kt , At ). When we introduce multiple types
of capital, the state is the vector of capital stocks. And with incomplete markets, the state is the
whole distribution of wealth in the cross-section of agents.
25
kt+1
45o
G(k)
k0
k1
k2
kt
k3 k*
Figure 2.1: The policy rule kt+1 = G (kt) of the Solow model.
26
Figure by MIT OCW.
G.M. Angeletos
2.1.7
Steady State
A steady state of the economy is dened as any level k such that, if the economy starts with k0 = k ,
then kt = k for all t 1. That is, a steady state is any xed point k of G in (2.11). Equivalently,
a steady state is any xed point (c , k ) of the system (2.9)-(2.10).
A trivial steady state is c = k = 0 : There is no capital, no output, and no consumption. This would
not be a steady state if f (0) > 0. We are interested for steady states at which capital, output and
consumption are all positive and nite. We can easily show:
Proposition 4 Suppose + n (0, 1) and s (0, 1). A steady state (c , k ) (0, )2 for the dictatorial
economy exists and is unique. k and y increase with s and decrease with and n, whereas c is nonmonotonic with s and decreases with and n. Finally, y /k = ( + n)/s.
Proof. k is a steady state if and only if it solves
0 = sf (k ) ( + n)k ,
27
Equivalently
(k ) =
where (k)
f (k)
.
k
+n
s
(2.13)
The function gives the output-to-capital ratio in the economy. The properties
of f imply that is continuous and strictly decreasing, with

(k) =
f (k)k f (k)
FL
= 2 < 0,
2
k
k
(0) = f (0) = and () = f () = 0,

where the latter follow from LHospitals rule. This implies that equation (2.13) has a unique solution:
k =
+n
s
Since < 0, k is a decreasing function of ( + n)/s. On the other hand, consumption is given
byc = (1 s)f (k ). It follows that c decreases with + n, but s has an ambiguous eect.
28
G.M. Angeletos
2.1.8
Transitional Dynamics
The above characterized the (unique) steady state of the economy. Naturally, we are interested to
know whether the economy will converge to the steady state if it starts away from it. Another way
to ask the same question is whether the economy will eventually return to the steady state after an
exogenous shock perturbs the economy and moves away from the steady state.
The following uses the properties of G to establish that, in the Solow model, convergence to the
steady is always ensured and is monotonic:
Proposition 5 Given any initial k0 (0, ), the dictatorial economy converges asymptotically to the
steady state. The transition is monotonic. The growth rate is positive and decreases over time towards
zero if k0 < k ; it is negative and increases over time towards zero if k0 > k .
29
Proof. From the properties of f, G (k) = sf (k) + (1 n) > 0 and G (k) = sf (k) < 0. That
is, G is strictly increasing and strictly concave. Moreover, G(0) = 0 and G(k ) = k . It follows
that G(k) > k for all k < k and G(k) < k for all k > k . It follows that kt < kt+1 < k whenever
kt (0, k ) and therefore the sequence {kt }

t=0 is strictly increasing if k0 < k . By monotonicity, kt
converges asymptotically to some k k . By continuity of G, k must satisfy k = G(k), that is k

must be a xed point of G. But we already proved that G has a unique xed point, which proves that
k = k . A symmetric argument proves that, when k0 > k , {kt }
t=0 is stricttly decreasing and again
converges asymptotically to k . Next, consider the growth rate of the capital stock. This is given by
t
kt+1 kt
= s(kt ) ( + n) (kt ).
kt
Note that (k) = 0 i k = k , (k) > 0 i k < k , and (k) < 0 i k > k . Moreover, by diminishing
returns, (k) = s (k) < 0. It follows that (kt ) < (kt+1 ) < (k ) = 0 whenever kt (0, k ) and
(kt ) > (kt+1 ) > (k ) = 0 whenever kt (k , ). This proves that t is positive and decreases
towards zero if k0 < k and it is negative and increases towards zero if k0 > k .
30
G.M. Angeletos
Figure 2.1 depicts G(k), the relation between kt and kt+1 . The intersection of the graph of G with the
45o line gives the steady-state capital stock k . The arrows represent the path {kt }
t=
for a particular
initial k0 .
Figure 2.2 depicts (k), the relation between kt and t . The intersection of the graph of with the
45o line gives the steady-state capital stock k . The negative slope reects what we call conditional
convergence.
Discuss local versus global stability: Because (k ) < 0, the system is locally stable. Because is
globally decreasing, the system is globally stable and transition is monotonic.
31
kt+1- kt
kt
kt
k*
(k)
- (+n)
Figure 2.2. The growth rate
kt+1- kt
= (kt) in the Solow model.
kt
Figure by MIT OCW.

32
G.M. Angeletos
2.2
Decentralized Market Allocations
In the previous section, we characterized the centralized allocation dictated by a social planner. We
now characterize the competitive market allocation
2.2.1
Households
Households are dynasties, living an innite amount of time. We index households by j [0, 1], having
normalized L0 = 1.
The number of heads in every household grow at constant rate n 0. Therefore, the size of the
population in period t is Lt = (1 + n)t and the number of persons in each household in period t is
also Lt .
We write c
jt , k
tj , b
jt , i
jt for the per-head variables for household j.
33
Each person in a household is endowed with one unit of labor in every period, which he supplies
inelastically in a competitive labor market for the contemporaneous wage wt . Household j is also
endowed with initial capital k0j . Capital in household j accumulates according to
j
(1 + n)kt+1
= (1 )ktj + it ,
which we approximate by
ktj+1 = (1 n)ktj + it .
(2.14)
Households rent the capital they own to rms in a competitive market for a (gross) rental rate rt .
The household may also hold stocks of some rms in the economy. Let tj be the dividends (rm
prots) that household j receive in period t. It is without any loss of generality to assume that
there is no trade of stocks (because the value of stocks will be zero in equilibrium). We thus assume
that household j holds a xed fraction j of the aggregate index of stocks in the economy, so that
tj = j t , where t are aggregate prots. Of course, j dj = 1.

34
G.M. Angeletos
The household uses its income to nance either consumption or investment in new capital:
c
jt + i
jt = y
tj .
Total per-head income for household j in period t is simply
y
tj = wt + rt ktj +
tj .
(2.15)
Combining, we can write the budget constraint of household j in period t as
c
jt + i
jt = wt + rt ktj +
tj
(2.16)
Finally, the consumption and investment behavior of household is a simplistic linear rule. They save
fraction s and consume the rest:
c
jt = (1 s)y
tj
35
and
i
jt = syti .
(2.17)
2.2.2
Firms
There is an arbitrary number Mt of rms in period t, indexed by m [0, Mt ]. Firms employ labor and
rent capital in competitive labor and capital markets, have access to the same neoclassical technology,
and produce a homogeneous good that they sell competitively to the households in the economy.
Let Ktm and Lm

t denote the amount of capital and labor that rm m employs in period t. Then, the
prots of that rm in period t are given by
m
m
m
m
m
t = F (Kt , Lt ) rt Kt wt Lt .
The rms seek to maximize prots. The FOCs for an interior solution require
FK (Ktm , Lm
t ) = rt .
(2.18)
FL (Ktm , Lm
t ) = wt .
(2.19)
36
G.M. Angeletos
Remember that the marginal products are homogenous of degree zero; that is, they depend only on
the capital-labor ratio. In particular, FK is a decreasing function of Ktm /Lm
t and FL is an increasing
m
m
function of Ktm /Lm
t . Each of the above conditions thus pins down a unique capital-labor ratio Kt /Lt .
For an interior solution to the rms problem to exist, it must be that rt and wt are consistent, that
is, they imply the same Ktm /Ltm . This is the case if and only if there is some Xt (0, ) such that
rt = f (Xt )
(2.20)
wt = f (Xt ) f (Xt )Xt
(2.21)
where f (k) F (k, 1); this follows from the properties FK (K, L) = f (K/L) and FL (K, L) =
f (K/L) f (K/L) (K/L), which we established earlier. That is, (wt , rt ) must satisfy wt = W (rt )
where W (r) f (f 1 (r)) rf 1 (r).
If (2.20) and (2.21) are satised, the FOCs reduce to Ktm /Lm
t = Xt , or
Ktm = Xt Lm
t .
37
(2.22)
That is, the FOCs pin down the capital-labor ratio for each rm (Ktm /Lm
t ), but not the size of the
rm (Lm
t ). Moreover, all rms use the same capital-labor ratio.
Besides, (2.20) and (2.21) imply
rt Xt + wt = f (Xt ).
(2.23)
It follows that
m
m
m
m
rt Ktm + wt Lm
t = (rt Xt + wt )Lt = f (Xt )Lt = F (Kt , Lt ),
and therefore
m
m
t = Lt [f (Xt ) rt Xt wt ] = 0.
(2.24)
That is, when (2.20) and (2.21) are satised, the maximal prots that any rm makes are exactly
zero, and these prots are attained for any rm size as long as the capital-labor ratio is optimal. If
instead (2.20) and (2.21) were violated, then either rt Xt + wt < f (Xt ), in which case the rm could
make innite prots, or rt Xt + wt > f (Xt ), in which case operating a rm of any positive size would
generate strictly negative prots.
38
G.M. Angeletos
2.2.3
Market Clearing
The capital market clears if and only if
Mt
0
Ktm dm
Equivalently,
Mt
Lt
0
=
0
(1 + n)t ktj dj
Ktm dm = Kt
where Kt
(2.25)
ktj dj is the aggregate capital stock in the economy.
The labor market, on the other hand, clears if and only if
Mt
0
Equivalently,
Lm
t dm
Mt
0
(1 + n)t dj
Lm
t dm = Lt
39
(2.26)
2.2.4
General Equilibrium: Denition
The denition of a general equilibrium is more meaningful when households optimize their behavior
(maximize utility) rather than being automata (mechanically save a constant fraction of income).
Nonetheless, it is always important to have clear in mind what is the denition of equilibrium in any
model. For the decentralized version of the Solow model, we let:
Denition 6 An equilibrium of the economy is an allocation {(ktj , cjt , ijt )j[0,1] , (Ktm , Lm
t )m[0,Mt ] }t=0 , a dis-
tribution of prots {(t j )j[0,1] }, and a price path {rt , wt }

t=0 such that
j
j j j
(i) Given {rt , wt }
t=0 and {t }t=0 , the path {kt , ct , it } is consistent with the behavior of household j,
for every j.
(ii) (Ktm , Lm
t ) maximizes rm prots, for every m and t.
(iii) The capital and labor markets clear in every period
40
G.M. Angeletos
2.2.5
General Equilibrium: Characterization
Proposition 7 For any initial positions (k0j )j[0,1] , an equilibrium exists. The allocation of production
across rms is indeterminate, but the equilibrium is unique with regard to aggregates and household allo
cations. The capital-labor ratio in the economy is given by {kt }
t=0 such that, for all t 0,
kt+1 = G(kt )
with k0 =
(2.27)
k0j dj given and with G(k) sf (k) + (1 n)k. Equilibrium growth is

t
kt+1 kt
= (kt ),
kt
(2.28)
where (k) s(k) ( + n), (k) f (k)/k. Finally, equilibrium prices are given by
rt = r(kt ) f (kt ),
(2.29)
wt = w(kt ) f (kt ) f (kt )kt ,
(2.30)
41
Proof. We rst characterize the equilibrium, assuming it exists. Using Ktm = Xt Ltm by (2.22), we
M
M
can write the aggregate demand for capital as 0 t Ktm dm = Xt 0 t Lm
t dm. From the labor market
Mt m
Mt m
clearing condition (2.26), 0 Lt dm = Lt . Combining, we infer 0 Kt dm = Xt Lt , and substituting
L
in the capital market clearing condition (2.25), we conclude Xt Lt = Kt , where Kt 0 t ktj dj denotes
the aggregate capital stock. Equivalently, letting kt Kt /Lt denote the capital-labor ratio in the
economy, we have
Xt = kt .
(2.31)
That is, all rms use the same capital-labor ratio as the aggregate of the economy.
Substituting (2.31) into (2.20) and (2.21) we infer that equilibrium prices are given by
rt = r(kt ) f (kt ) = FK (kt , 1)
wt = w(kt ) f (kt ) f (kt )kt = FL (kt , 1)
Note that r (k) = f (k) = FKK < 0 and w (k) = f (k)k = FLK > 0. That is, the interest
42
G.M. Angeletos
rate is a decreasing function of the capital-labor ratio and the wage rate is an increasing function
of the capital-labor ratio. The rst properties reects diminishing returns, the second reects the
complementarity of capital and labor.
Adding up the budget constraints of the households, we get
Ct + It = rt Kt + wt Lt +
tj dj,
where Ct cj
t dj and It ijt dj. Aggregate dividends must equal aggregate prots, tj dj =
m
t dj. By (2.24), prots for each rm are zero. Therefore, tj dj = 0, implying Ct + It = Yt =
rt Kt + wt Lt . Equivalently, in per-capita terms,
ct + it = yt = rt kt + wt .
From (2.23) and (2.31), or equivalently from (2.29) and (2.30),
rt kt + wt = yt = f (kt )
43
We conclude that the household budgets imply

ct + it = f (kt ),
which is simply the resource constraint of the economy.
Adding up the individual capital accumulation rules (2.14), we get the capital accumulation rule for
the aggregate of the economy. In per-capita terms,
kt+1 = (1 n)kt + it
Adding up (2.17) across households, we similarly infer it = syt = sf (kt ). Combining, we conclude
kt+1 = sf (kt ) + (1 n)kt = G(kt ),
which is exactly the same as in the centralized allocation.
44
G.M. Angeletos
Finally, existence and uniqueness is now trivial. (2.27) maps any kt (0, ) to a unique kt+1
(0, ). Similarly, (2.29) and (2.30) map any kt (0, ) to unique rt , wt (0, ). Therefore, given
any initial k0 = k0j dj, there exist unique paths {kt }

t=0 and {rt , wt }t=0 . Given {rt , wt }t=0 , the
allocation {ktj , cjt , ijt } for any household j is then uniquely determined by (2.14), (2.15), and (2.17).
Finally, any allocation (Ktm , Lm
t )m[0,Mt ] of production across rms in period t is consistent with
equilibrium as long as Ktm = kt Lm
t .
Corollary 8 The aggregate and per-capita allocations in the competitive market economy coincide with
those in the dictatorial economy.
We can thus immediately translate the steady state and the transitional dynamics of the centralized
plan to the steady state and the transitional dynamics of the decentralized market allocations.
Remark: This example is just a prelude to the rst and second welfare theorems, which we will
have once we replace the rule-of-thumb behavior of the households with optimizing behavior given
a preference ordering over dierent consumption paths: in the neoclassical growth model, Pareto
ecient and competitive equilibrium allocations coincide.
45
2.3
Shocks and Policies
The Solow model can be interpreted also as a primitive Real Business Cycle (RBC) model. We can
use the model to predict the response of the economy to productivity, taste, or policy shocks.
2.3.1
Productivity (or Taste) Shocks
Suppose output is given by
Yt = At F (Kt , Lt )
or yt = At f (kt ), where At denotes total factor productivity.

Consider a permanent negative shock in A. The G(k) and (k) functions shift down, as illustrated
in Figure 2.3. The economy transits slowly from the old steady state to the new, lower steady state.
If instead the shock is transitory, the shift in G(k) and (k) is also temporary. Initially, capital and
output fall towards the low steady state. But when productivity reverts to the initial level, capital
and output start to grow back towards the old high steady state.
46
G.M. Angeletos
kt+1
Figure 2.3: The impact of a productivity shock on G.
kt
Figure by MIT OCW.

The eect of a prodictivity shock on kt and yt is illustrated in Figure 2.4. The solid lines correspond
to a transitory shock, whereas the dashed lines correspond to a permanent shock.
Taste shocks: Consider a temporary fall in the saving rate s. The (k) function shifts down for a
while, and then return to its initial position. What are the transitional dynamics? What if instead
the fall in s is permanent?
47
kt
kt
t
yt
yt = Atf (kt)
t1
t2
Figure 2.4: The impact of a productivity shock on G.
Figure by MIT OCW.

48
G.M. Angeletos
2.3.2
Unproductive Government Spending
Let us now introduce a government in the competitive market economy. The government spends
resources without contributing to production or capital accumulation. The resource constraint of the
economy now becomes
ct + gt + it = yt = f (kt ),
where gt denotes government consumption. The latter is nanced with proportional income taxation:
gt = y t .
Disposable income for the representative household is (1 )yt . We continue to assume agents consume
a fraction s of disposable income: it = (1 s)(yt gt ).
Combining the above, we conclude that the dynamics of capital are now given by
t =
kt+1 kt
= s(1 )(kt ) ( + n).
kt
where (k) f (k)/k. Given s and kt , the growth rate t decreases with .
49
A steady state exists for any [0, 1) and is given by
k =
+n
.
s(1 )
Given s, k decreases with .
Policy Shocks: Consider a temporary shock in government consumption. What are the transitional
dynamics?
2.3.3
Productive Government Spending
Suppose now that production is given by

yt = f (kt , gt ) = kt gt ,
where > 0, > 0, and + < 1. Government spending can thus be interpreted as infrastructure
or other productive services. The resource constraint is ct + gt + it = yt = f (kt , gt ).
50
G.M. Angeletos
Government spending is nanced with proportional income taxation and private consumption is a
fraction 1 s of disposable income: gt = yt , ct = (1 s)(yt gt ), it = s(yt gt ).
Substituting gt = yt into yt = kt gt and solving for yt , we infer
yt = kt1 1 kta b
where a /(1 ) and b /(1 ). Note that a > .
We conclude that the dynamics and the steady state are given by
t =
kt+1 kt
= s(1 ) b kta1 ( + n).
kt
s(1 ) b
k =
+n
1/(1a)
.
Consider the rate that maximizes either k , or t for any given kt . This is given by = b/(1+b) = .
The more productive government services are, the higher their optimal provision.
51
2.4
Continuous Time, Long-Run Growth, Convergence
2.4.1
The Solow model in continuous time
We now consider the model in continuous time and also allow for exogenous growth.
The resource constraint of the economy is
C + I = Y = F (K, AL).
Population growth and exogenous technological change are given by
L
=n
L
and
=x
A
and the law of motion for aggregate capital is
K = I K
52
G.M. Angeletos
Let k K/ (AL) , y = Y / (AL) , etc. Then,

kk =
K
LL
A
. Substituting from the above, we infer
K
A
k = i ( + n + x)k.
Combining this with i = sy = sf (k), we conclude
k = sf (k) ( + n + n)k.
Equivalently, the growth rate of the economy is given by

k
= (k) s(k) ( + n + x).
k
(2.32)
This is the same (k) function as in the discrete-time version.
Exogenous productivity growth x simply contributes to the eective depreciation of k, the capitalto-eective-labor ratio.
53
2.4.2
Log-linearization and the Convergence Rate
Dene z ln k ln k . We can rewrite the growth equation (2.32) as
z = (z),
(z) (k ez ) s(k ez ) ( + n + x)
Note that (z) is dened for all z R. By denition of k , (0) = s(k ) ( + n + x) = 0. Similarly,
(z) > 0 for all z < 0 and (z) < 0 for all z > 0. Finally, (z) = s (k ez )k ez < 0 for all z R.
We next linearize z = (z) around z = 0 :
z = (0) + (0) z
z = z
where we substituted (0) = 0 and let (0).
54
G.M. Angeletos
Straightforward algebra gives

(z) = s (k ez )k ez < 0
f (k)k f (k)
f (k)k f (k)
(k) =
= 1
k2
f (k)
k2
sf (k ) = ( + n + x)k
(0) = (1 K )( + n + x) < 0
where K FK K/F = f (k)k/f (k) is the elasticity of production with respect to capital, evaluated
at the steady-state k.
We conclude that

k
k
= ln
k
k
where
= (1 K )( + n + x) < 0
The quantity is called the convergence rate.
55
In the Cobb-Douglas case, y = k , the convergence rate is simply

= (1 )( + n + x),
Note that as 0 as 1. That is, convergence becomes slower and slower as the income share
of capital becomes closer and closer to 1. Indeed, if it were = 1, the economy would a balanced
growth path.
Around the steady state
y
y
= K
k
k
and
y
y
= K
k
k
y
= ln
y
It follows that
y
y
Thus, is the convergence rate for either capital or output.
Quantitative: If = 35%, n = 1%, x = 2%, and = 5%, then = 6%. This clearly contradicts the
data. But if = 70%, then = 2.4%, which does a better job in matching the data on conditional
convergence. (Hint: think of a broad denition of capital.)
56
G.M. Angeletos
2.5
Cross-Country Dierences: Mankiw-Romer Weil
The Solow model implies that steady-state capital, productivity, and income are determined by A, , x
and s, n. Assuming that countries share the same technology up to a level dierence, if the Solow
model is correct, observed cross-country income and productivity dierences should be explained
by observed cross-country dierences in s and n.
Mankiw, Romer and Weil (1992) tests this hypothesis against the data. In its simple form, the Solow
model fails to predict the large cross-country dispersion of income and productivity levels. They then
consider an extension of the Solow model, that includes two types of capital, physical capital (K)
and human capital (H). Output is given by
Yt = f (Kt , Ht , At Lt ) = Kt Ht (At Lt )1 ,
where > 0, > 0, and + < 1. Equivalently,
yt = f (kt , ht ) = kt ht .
57
The dynamics of capital accumulation are now given by

k t = sk f (kt , ht ) (k + n + x)kt
h t = sh f (kt , ht ) (h + n + x)ht
where sk and sh are the investment rates in physical capital and human capital, respectively, and k
and h are the respective depreciation rates. We henceforth assume k = h = .
The steady-state levels of k, h, and y then depend on both sk and sh , as well as and n. Letting j
index a country, and assuming that and x are the same across countries, while sk , sh and n dier,
we have
kj =
hj =
sk,j
nj + x +
sk,j
nj + x +
58
sh,j
nj + x +
sh,j
nj + x +
1
1
1
1 1
G.M. Angeletos
Along the steady state,
Y
= Aj,t f kj , hj .
L j,t
Letting Aj,t = exp (xt + aj ) , where aj is the level of technology for country j, we get
Y
ln
= aj + xt + b1 ln skj + b2 ln shj + b3 ln (nj + x + )
L j,t
b1 =

,
1
b2 =
,
1
b3 =
(2.33)
+
.
1
The idea is then to run a regression like (2.33) in cross-country data. To do this, one needs (i)
to specify a measure for shj , (ii) to assume that x is common across countries, and (iii) to assume
that aj is orthogonal to (skl , shj , nj ) . The last assumption is particularly problematic. As for what
measure of sh to use, thats another tricky point. But lets ignore these issues for a moment and,
as Mankiw-Romer-Weil did, let us proxy sh with the fraction of working-age population that is in
school (and let us also set + x = .05). Also, we will measure sk with the investment-to-GDP ratio
and Y /L with GDP per worker.
59
Now, let us rst run the MRW regression with the restriction that = 0 (i.e., no human capital):

Y
ln
= constant + b1 ln (sk,j ) + b2 ln (nj + x + ) + aj .
L j,t
Then the results are as in Table 1 (reproduced from Acemoglu 2007).
Table 1: MRW for the basic Solow Model
MRW
Updated data
1985
1985
2000
1.42
1.01
1.22
(.14)
(.11)
(.13)
-1.97
-1.12
-1.31
(.56)
(.55)
(.36)
Adj R2
.59
.49
.49
Implied
.59
.50
.55
No. of observations
98
98
107
ln(skj )
ln(nj + x + )
60
G.M. Angeletos
Note that cross-country variation in investment rates sk and population growth rates n can account
for a bit more than 50% of the observed cross-country variation in productivity levels.
However, this is acheived with close to 0.6, which is inconsistent with the income share of capital
observed in the data ( between 0.3 and 0.4). If we were to impose = .35 (b1 = b2 = .54), and
see how the calibrated Solow model ts the data, then cross-country dierences in sk and n would
account for only 1/4 to 1/3 of the observed cross-country productivity dierences. (This is half than
before, but still its a pretty good t.)
But, what if we use the augmented Solow model? Then the results look better, as in Table 2.
61
Table 2. MRW for augmented Solow model

MRW
Updated data
1985
1985
2000
.69
.65
.96
(.13)
(.11)
(.13)
.66
.47
.70
(.07)
(.07)
(.13)
-1.73
-1.02
-1.06
(.41)
(.45)
(.33)
Adj R2
.78
.65
.60
Implied
.30
.31
.36
Implied
.28
.22
.26
No. of observations
98
98
107
ln(sk )
ln(sh )
ln(n + g + )
62
G.M. Angeletos
Therefore, the augmented Solow model can account for 60% to 80% of the observed cross-country
dierences in productivity levels, with an that is comfortably within the range of empirically
plausible values.
However, there are important caveats.
First, if one uses alternative measures for human capital, such as secondary-eductation attainment
levels, then the t is signicantly reduced (Klenow and Rodriguez, 1997).
Second, it is not clear how one should interpret any of these results, given that there are good reasons
to expect (as we will see in the models that follow) that technological dierences are likely to be
correlated with saving rates and education levelsindeed with the causality probably going both
ways.
63
Third, whereas the estimated is empirically plausible, the estimated is not. In particular, the
estimated in MRW implies much higher returns to education than the ones estimated with standard
Mincerian regressions on micro data (Klenow and Rodrigez, 1997; Hall and Jones, 1999). If one
calibrates the augmented Solow model so that is consistent with Mincerian estimates, then the
model accounts for about one half of the observed cross-country output-per-worker dierenceswhich
is leaves the rest one half explained by technological dierence (the augemented-Solow residuals
aj ).
If we calibrate the augmented Solow model as in Hall and Jone (1999), use it to predicted the outputper-worker levels for each country, and and plot them against the actual levels of each country,
then we get Figure 2.5 (reproduced from Acemoglu 2007). Clearly, there is signicant unexplained
variation. Moreover, the errors are strongly correlated with actual levelsthe model particularly
overestimates the output-per-worker levels of poor countries.
64
G.M. Angeletos
PER
FJI
PAN
PHL
URY
THA
CHL
MEX
VEN
TTO
JAM
CRI
BRA
IRNMYS
JOR ZAF PRT
ZWE
PRY
LKANIC DOM
MUS
SYR
BOL
COL
BWA TUR
TUN
HND
IND
IDN
PNG
SLV
EGY
CMR PAK
BGD
GTM
GUY
ZMB
CHN
COG
LSO
MWI
KEN
NPL
MRT
TGO
SEN
MLI BEN CAF
BDI
GMB
SLE
9
10
11
UGA
MOZ
7
7
SEN
BEN
GMB
MLI
NER
RWA
UGA MOZ
45
NOR
JPNCHE
NZL
CAN USA
SWE
KOR
GER
FIN
AUS
DNK
AUT
ISR ISL
FRA
NLD
BEL
HKG
HUN GRC
GBR
ESPITA
PAN
ARG
MYS
PERTHA
IRL
CHL
ECU
MEX
JAMPHL
LSO
JOR URY
PRT
BRB
VEN
ZAF
CRI
BRATTO
ZMB
ZWE
DZA TUR
CHN
IRN
NIC LKA
IDN PRY
HND
COL
BOL
MUS
COG
SYR DOM
TUN
IND
KEN
SLV
TGO GHA NPL
EGY
PAK
CMR
GTM
BGD
MWI
Predicted log gdp per worker 2000

9
10
11
NOR
NZL CHE
DNK
GER
FIN
AUS
CANUSA
JPN
SWE
AUT
NLD
GRC
ISR
BEL
ISL
ARG
HUN
KOR
FRA
CYP GBR
BRBHKG
SGPITA
IRL
ESP
UGA
GMB
MOZ
RWA
SLE
45
9
10
11
NZL CHE
NOR
AUS
DNK
ISR SWE
USA
AUT
JPNFIN
NLD
CAN
BEL
GBR
ISL
GRC
CYP
FRA
ARG
ITA
BRBIRL
KOR
GUY
ESP
PER VEN
HKG
FJIPAN
JAM
SGP
ZMB
ECU
CHL
THA
ZWE PHL
URY
ZAF
MUS TTO PRT
COG
CRI BRA MEX
MYS
BOL IRN
MWI CHN
LKA
COL
TUR
DOM
GHA
NIC TUN
KEN
SYRPRY
IND
HND
SLV
BWA
PAK PNG
GTMJOR
SEN
IDN
LSOBGD
NER
EGY
NPL TGO
CMR
BEN MLICAF
HUN

9
10
11
45
9
10
11
9
10
11
12
Figure 2.5: Calibrated GDP per worker from augmented Solow model versus actual GDP per worker (in
logs).
65
2.6
Conditional Convergence: Barro regressions
Recall the log-linearization of the dynamics around the steady state:
y
y
= ln .
y
y
A similar relation will hold true in the neoclassical growth model a la Ramsey-Cass-Koopmans. < 0
reects local diminishing returns. Such local diminishing returns occur even in endogenous-growth
models. The above thus extends well beyond the simple Solow model.
Rewrite the above as

ln y = ln y ln y
Next, let us proxy the steady state output by a set of country-specic controls X, which include
s, , n, etc. That is, let
ln y X.
66
G.M. Angeletos
We conclude
ln y = ln y + X + error
The above represents a typical Barro-like conditional-convergence regression: We use cross-country

data to estimate (the convergence rate), together with (the eects of the saving rate, education,
population growth, policies, etc.) The estimated convergence rate is about 2% per year. The esti
mated coecients on X are of the expected sign.
These types of regressions are quite informative: they identify important correlations in the data.
However, inferring causality is much trickier. (See Barro and Sala-i-Martins and Acemoglus text
books for further discussion.)
67
2.7
The Golden Rule
The Golden Rule. Consumption at the steady state is given by
c = (1 s)f (k ) = f (k ) ( + n)k
Suppose society chooses s so as to maximize c . Since k is a monotonic function of s, this is equivalent

to choosing k so as to maximize c . Note that
c = f (k ) ( + n)k
is strictly concave in k . The FOC is thus both necessary and sucient. c is thus maximized if and
only if k = kgold , where kgold solves
f (kgold ) = n.
Equivalently, s = sgold , where sgold solves sgold (kgold ) = ( + n) This is called the golden rule
for savings, after Phelps.
68
G.M. Angeletos
Dynamic Ineciency. If s > sgold (equivalently, k > kgold ), the economy is dynamically inecient:
if the saving rate is lowered to s = sgold for all t, then consumption in all periods will be higher!
On the other hand, if s < sgold (equivalently, k > kgold ), then raising s towards sgold will increase
consumption in the long run, but at the expense of lower consumption in the short run; whether such
a trade-o is desirable depends on how one weighs current generations vis-a-vis future generations.
The Modied Golden Rule. In the Ramsey model, this trade-o will be resolved when k satises the
f (k ) = n + ,
where > 0 measures impatience, or discounting across generations. Clearly, the distance between
the Ramsey-optimal k and the golden-rule kgold increases with .
Testing for dynamic ineciency. The golden rule can be restated as r = Y /Y ; dynamic eciency
is ensured if r > Y /Y. Abel et al. use test this relation with US data and nd no evidence of
dynamic ineciency.
69
2.8
Poverty Traps, Cycles, Endogenous Growth, etc.
The assumptions we have imposed on savings and technology implied that G is increasing and
concave, so that there is a unique and globally stable steady state. More generally, however, G
could be non-concave or even non-monotonic. Such pathologies can arise, for example, when the
technology is non-convex, as in the case of locally increasing returns, or when saving rates are highly
sensitive to the level of output, as in some OLG models.
Figure 2.6 illustrates an example of a non-concave G. There are now multiple steady states. The two
extreme ones are (locally) stable, the intermediate is unstable versus stable ones. The lower of the
stable steady states represents a poverty trap.
Figure 2.7 illustrates an example of a non-monotonic G. We can now have oscillating dynamics, or
even perpetual endogenous cycles.
70
G.M. Angeletos
kt + 1
45o
G (k)
kt
Figure 2.6: Non-concave G and poverty traps.
71
Figure by MITOCW.
kt + 1
G (k)
kt
Figure 2.7: Non-monotonic G and cycles.
72
Figure by MIT OCW.
G.M. Angeletos
What ensures that the growth rate asymptotes to zero in the Solow model (and the Ramsey model
as well) is the vanishing marginal product of capital, that is, the Inada condition limk f (k) = 0.
Continue to assume that f (k) < 0, so that (k) < 0, but assume now that limk f (k) = A > 0.
This implies also limk (k) = A. Then, as k ,
t
kt+1 kt
sA (n + )
kt
If sA < (n + ), then it is like before: The economy converges to k such that (k ) = 0. But if
sA > (n + x + ), then the economy exhibits deminishing but not vanishing growth: t falls with t,
but t sA (n + ) > 0 as t .
Jones and Manuelli consider such a general convex technology: e.g., f (k) = Bk + Ak. We then get
both transitional dynamics in the short run and perpetual growth in the long run.
In case that f (k) = Ak, the economy follows a balanced-growth path from the very beginning.
We will later endogenize A in terms of externalities, R&D, policies, institutions, markets, etc.
73
74

MIT Economics Notes OCW

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MIT Economics Notes OCW

Uploaded by

Copyright:

Available Formats

Chapter 7

Endogenous Growth II: R&D and

Economic Growth: Lecture Notes

Expanding Product Variety: The Romer Model

Yt = F (Xt , Lt ) = A(Lt )1 (Xt ) .

The composite factor is given by a CES aggregator of intermediate inputs:

Economic Growth: Lecture Notes

In what follows, we will assume = , which implies

Finally, note that if Xt,j = X for all j and t, then

Economic Growth: Lecture Notes

Final Good Sector

(pt,j Xt,j )dj

for all j [0, Nt ].

Intermediate Good Sector

where (X) represents the cost of producing X in terms of nal-good units.

Economic Growth: Lecture Notes

We will let the cost function be linear:

The FOCs give

for the optimal price, and

Economic Growth: Lecture Notes

The Innovation Sector

Equivalently, RV = L, which has a simple interpretation: The opportunity cost of holding an

Economic Growth: Lecture Notes

s.t. ct + at+1 wt + (1 + Rt )at

As usual, the Euler condition gives

And assuming CEIS, this reduces to

Economic Growth: Lecture Notes

Economic Growth: Lecture Notes

Next, note that the resource constraint reduces to

Eciency and Policy Implications

Economic Growth: Lecture Notes

The FOC with respect to Xt gives

represents that social return on savings.

Economic Growth: Lecture Notes

Policy exercise: Consider four dierent types of government intervention:

Building on the Shoulders of Giants

Economic Growth: Lecture Notes

A Simple Variant of Aghion-Howitt

Economic Growth: Lecture Notes

It follows that the optimal level of R&D is given by q (ztj )Vtj+1 = wt or

The Value of Innovation

Economic Growth: Lecture Notes

if his R&D fails

if his R&D succeeds

It follows that a successful innovation generates a stream of dividends equal to At

where where R is the interest rate per period and

The Cost of Innovation

Economic Growth: Lecture Notes

Combining (7.1) and (7.2), we infer that

It follows that the aggregate level of technology grows at a rate

Economic Growth: Lecture Notes

Economic Growth: Lecture Notes

level of R&D for an outsider is

Economic Growth: Lecture Notes

Note that an increase in

Eciency and Policy Implications

Discuss the spillover eects of innovation... Both negative and positive...

Ramsey Meets Schumpeter: The Aghion-Howitt Model

Economic Growth: Lecture Notes

Romer Meets Acemoglu: Biased Technological Change

Consider a two-factor economy, with

Note that this is dierent from saying that technology is Haugmenting.

A simple model of biased technological change