econstor

A Service of

zbw
Leibniz-Informationszentrum
Wirtschaft

Make Your Publications Visible.

Leibniz Information Centre
for Economics

Sala-i-Martin, Xavier

Working Paper
Lecture Notes on Economic Growth: Introduction to
the Literature and Neoclassical Models, Volume I

Center Discussion Paper, No. 621

Provided in Cooperation with:

Economic Growth Center (EGC), Yale University

Suggested Citation: Sala-i-Martin, Xavier (1990) : Lecture Notes on Economic Growth:

Introduction to the Literature and Neoclassical Models, Volume I, Center Discussion Paper, No.
621, Yale University, Economic Growth Center, New Haven, CT

This Version is available at:

http://hdl.handle.net/10419/160543

Standard-Nutzungsbedingungen: Terms of use:

Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Documents in EconStor may be saved and copied for your
Zwecken und zum Privatgebrauch gespeichert und kopiert werden. personal and scholarly purposes.

Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle You are not to copy documents for public or commercial
Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich purposes, to exhibit the documents publicly, to make them
machen, vertreiben oder anderweitig nutzen. publicly available on the internet, or to distribute or otherwise
use the documents in public.
Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen
(insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, If the documents have been made available under an Open
gelten abweichend von diesen Nutzungsbedingungen die in der dort Content Licence (especially Creative Commons Licences), you
genannten Lizenz gewährten Nutzungsrechte. may exercise further usage rights as specified in the indicated
licence.

www.econstor.eu
 I.
'

I.

I
ECONOMIC GROWTH CENTER

YALE UNIVERSITY

Box 1987, Yale Station

New Haven, Connecticut 06520

CENTER DISCUSSION PAPER NO. 621

LECTURE NOTES ON ECONOMIC GROWTH:

INTRODUCTION TO THE LITERATURE AND NEOCLASSICAL MODELS

VOLUME I

Xavier Sala-i-Martin

Yale University

December 1990

Notes: Center Discussion Papers are preliminary materials circulated to

stimulate discussion and critical comments.

These notes were developed while I was assisting Robert J. Barro to

teach the first year graduate macroeconomics class at Harvard in
1988-89 and 1989-90. The two vintages of first year students suffered
the first draft of these notes and made all kinds of useful
comments/amendments. Thank you kids. I am especially grateful to
Susan Guthrie for carefully reading them and helping me, think about
section 1. I also received comments from Serge Marquie and (Kid) Casey
Mulligan from Harvard, and Julie Lee from Yale.

This paper is the revised version of the first draft, November 1988.

See Discussion Paper No. 622 for Volume II of Lecture Notes on Economic
Growth.
 LECTURE NOTES ON ECONOMIC GROWTH:

INTRODUCTION TO THE LITERATURE AND NEOCLASSICAL MODELS

VOLUME I

Abstract

· This is a survey· of· the literature on Economic ·.Growth. ·.In. the ....·

introduction we analyze the main differences between exogenous and

endogenous growth models using fixed savings rate analysis. We argue that

in order to have endogenous growth there must be constant returns to the

factors that can be accumulated. A graphical tool is then developed to

show that changes in the savings rate have different effects on long run

·growth in the .two kinds of models; we show that only endogenous growth

models are affected by shifts in the savings rate. We then explore two

versions of the Ramsey-Cass-Koopmans neoclassical model where savings are

determined optimally; one with exogenous productivity growth and one

without.

KEY WORDS: Economic Growth, Increasing Returns, Externalities,

Endogenous Growth

. .,. ·;..:..
 "The consequences for human welfare involved in questions . .Like
these are simply staggering: once one starts to ·think.about them, it.is ihard
to think about anything else". Lucas (1988), p. 5.

(1) INTRODUCTION TO GROWTH MODELS.

(a) Exogenous versus Endogenous Growth models: An Introduction

Most of the recent economic growth literature deals with

"optimizing growth models" where consumers choose a consumption path by
maximizing some kind of utility function subject to some intertemporal
1
budget constraint The ·:complicated mechanics of dynamic optimization,
however, obscures.· ·some of the importantc. points and issues. Hence, before.
studying such models it will be convenient to start with the assumption that
the savings rate is an exogenous constant: people save a constant fraction
"s" of their income. This is what Solow (1957) and others, following the
Keynesian •multiplier hypothesis, do. Within an intertemporally optimizing
framework, there is a configuration of parameters that will yield a constant
2
savings rate . Hence, economists that do not believe in Keynesian

Early economists used to confine the intertemporal optimization

analysis to normative issues. The celebrated Ramsey 1928 paper (which deals
with intertemporal optimizing economies) starts with the sentence "The first
problem I propose to tackle is this: how much of its income should a nation
save?" (p.543). ·· Contemporaneous economists use intertemporal optimizing
models for descriptive or positive analysis as 'Well. Following Barro
(1974), the representative agent is assumed to be a family or group of
individuals linked to each other through bequests.
2
Kurtz (1968) showed that if the production function is Cobb
Douglas, necessary and sufficient conditions for· constant transitional
optimal savings rates are
(1) the utility fuIJ.c:_~on) be Constant Elasticity of Intertemporal
Substitution of the form c ;s /(1-(1/s)), wheres is the savings rate,

, . .- .
~-.
 multipliers may want to think of an economy described by such configuration.
Suppose also that the only asset in this (closed) economy is something we
r c'all Kt.· ·You···may want to think' of K as being physical··· CAPITAL· but .it may
also .include other inputs .that can be accumulated,,; such as knowledge". or"
skills. Now imagine .that the ..production function is .Cobb-Douglas and. that
there are two aggregate inputs. One ·of them, Kt, can be purposely
accumulated and the other·Lt, cannot be accumulated, or it grows at ar-ate
which is independent -of individual choices ~(think of 'L as 'labor but···ut·"may -··
·also include other unreproducible resources such as land or energy).

(1.1) y

The increase in K over time, which we will call 3 K=dK/dt is

aggregate net INVESTMENT In a'closed economy net investment must equal to
SAVINGS minus·. DEPRECIATION. 'Using (1.1) and the fixed savings Tate
assumption:

Where o is the (constant) depreciation rate. Population is

assumed "to be equal to .employment (so we abstract .from unemployment and.
labor force participation issues) and is assumed to grow at an exogenous

constant rate, L/L=n. Let us define lower case k as the capital-labor ratio
(or capital per worker) K/L. By taking derivatives of kt with respect to

.(2) ·the discount rate be related to ::the ..parameters ·· of.Jthe model ;through·
p=f3-s, where p is the discount rate, and f3 is the share of capital in the
production function.
See also Barro and Sala-i-Martin (1990) chapter 1 for an extension
of this result.
3
Throughout these notes we will denote time derivatives by "dots"
on top of variables.

2
 4
time we can rewrite (1.2) in per capita terms as

(1.2)' kt

Let us divide both sides of (1.2)' by kt and define the growth rate of

capital per worker kt/kts-yk. We will call STEADY. STATE .the state where all.
variables grow, at a,, constant· (possibly ·zero),; rate. Thus; in ·steady·"stlate ''.'fk:' ,., -
is constant. Take logarithms and derivatives of both sides and get

(1.3) 0=(/3-1)-yk+n(a+/3-l)

This KEY equality deserves further attention. In the original

Neoclassical ·growth model (Solow (1956) and Swan (1956)) the production
function is assumed to exhibit Constant •Returns .to Scale in capital and
labor (ie, · a+/3=1) but Decreasing Returns to Capital alone (/3<1). Notice
that by virtue of the CRS assumption (a+/3=1), the second term in the right
hand.side ,of (1.3). is zero.so:we are left with

(1.3), 0=(/3-1)-yk

but due to the Decreasing Retur:ns to Capital assumption (/3<1), equality

(1. 3)' says that the only sustainable steady state growth rate is -yk=O.
That is, in the CRS neoclassical model, the only possible steady state
growth rate is zero. If the only possible growth rate is zero, how did the
neoclassical theorists of the 50's and 60's explain long run growth?. They
basically assumed that the economy gets (exogenously) more productive over
time. ., In other words, they extended . the .. technology in _( 1.1) to ,a more. ~,

4
Notice that the difference between expressing the accumulation
equation' in levels or 'in per capita terms, is the ·term nk added to ok. We ·
can in fact think of nk as some extra depreciation since it represents the
loss of capital per person due to the fact that, when population grows, we
- have to share capital with an increasing number of people.

3
 general

(1.1), yt

where A(t) reflects the 'level of the technology which is asswned to . be

growing at the constant rate g, so A(t)=A(O)egt. The parameter "g" is often
5
referred as the· "exogenous ·productivity ,growth.rate"; In section l.:·3-,we·
will present· an optimizing version «Of this model. We will see that ·income
· per capita, capital per capita, and investment per capita will end up
growing at this exogenously given rate. We will also expand on the term
A(t) and on different ways to model productivity growth.
A second (and possibly more interesting) way to read equation
(1. 3) is the following: "In a CONSTANTS RETURNS TO SCALE model (a+,8=1) in
order to have positive steady state growth (1k>O), the production function
must exhibit CONSTANT. RETURNS TO .THE INPUTS THAT CAN BE ACCUMULATED, ,8=1. 116
This simple fact underlies the CONSTANT RETURNS ENDOGENOUS GROWTH models
developed in the late 80' s. The implied production function is the
following:

(1.1)'' Y = AK
t t

The simplest growth model using this type of production function

(Rebelo (1990)) is outlined in section 1. 4. Notice that this t}'Tpe of
production function does not give any role to exhaustible or non
reproducible resources such as raw labor or land. One could argue, however,
that what matters for production is not raw labor but, rather, quality

5
It is called exogenous because it is unaffected by any of the
parameters of the model such as the capital share or the savings rate.
6
· Notice that we are saying CR to K and no~ C(t ~ Scale. The distinction
is important: the production function Y=K L - with 0<,8<1, exhibits
constant returns to scale (if we multiply all inputs by A>l we get A times
as much output) but Decreasing Returns to Capital (since if we multiply
capital by A we get less than A times as much as output).

!+
 adjusted labor. The quality of the labor force (often called Human Capital)
is accumulated as, each generation is .more knowledgeable than the one- before.
When one eombine·s "physical· -'and, 1.human - -capital• into _,·some, -broad measure -rn~f

capital, the aggregate production function will look l·ike the linear AK
,,
function postulated above. This is ·the· approach taken' by ·Lucas (1988') and
Uzawa (1956). A version of these models is presented in section 1.7.
Barro (1990) and its extensions ' outlined in Barro "and
Sala-i-Martiri (1990) - assume that' the two 'inputs of production are private
physical capital and publicly provided inputs such as roads, infrastructure
or law enforcement. Output exhibits constant returns to both inputs. In
section 1.5 we will show that this setup ends U? being similar to
postulating an AK production function where K must again be interpreted as a
broad measure of capital.
Notice that (as can be seen from equation (1.3)', the steady state
growth rate -yk derived from these models is positive without assuming
exogenous productivity growth. As we will see in the next subsection, the
parameters"''of the modeJ. (in-particular the savings rate) will determine this
growth. rate/ Because the growth' rate is determined within the model, (in
other words, it depends on the other parameters of the model) these are
often,cal-led "ENDOGENOUS' GROWTH MODELS".
Finally, equation (1.3) allows for one more possibility. If the
population growth rate is zero (n=O), we can have nonreproducible inputs
(a>O) together with ENDOGENOUS GROWTH (-yk>O) if there are CONSTANT RETURNS
TO THE INPUTS THAT CAN BE ACCUMUI.ATED (fi=l). But notice that this implies
OVERALL INCREASING RETURNS TO SCALE (a+fi>l). This possibility gives rise to
the so called "INCREASING RETURNS ENDOGENOUS GROWTH MODELS 117 •
Of course, if we plainly postulate an Increasing Returns to Scale
(IRS) production function we run into trouble since we cannot find a set of

7
As can be seen from equation ( 1. 3) , when the population growth
rate is positive, the inc,reasing returns -to scale models· (a+fi>l) run· into
trouble since there is no -yk that satisfies the key equality. What happens
in this. circumstances'· is that' the ·growth rate is never constant but, rather,
it increases over time.

5
t
prices to support a general competitive equilibrium. There are at least two
ways to get around this problem.
(a) The first one (due to Alfred:Marshall) is to introduce IRS.at
the aggregate level but CRS at ' the firm level. This can be formulated
through production externalities or spillovers: each fir:m' s decisionc.:affects
all other firms output, but none of the firms takes this into account.
Hence, all the . firms face a, "concave''. problem -which has a competitive
solution. .The ··economy as. a whole, :however; ·.faces an IRS production.function
.which, .·;under ..some .conditions . that we will mention in·. a second, generates
endogenous growth. The Cobb Douglas version of this production function is

(1.1.)" yt

where Kt is private capital ,and ~t is the aggregate capital in the economy.

Individual firms do not think they can affect ~ so they take it as given.
Notice that under these circumstances firms face a perfectly defined concave
problem so the Kuhn-Tucker theorems .apply. In the aggregate, however, total
capital will equal. the sum .. of. individual capitals. and therefore ~=K. . Thus
the aggregate production function will be

(1.1)'''' y = AKP+~Ll-P
t t t

Notice that if the size of the externality is "correct" (that is

if p+~=l) we will have CONSTANT RETURNS TO CAPITAL in an INCREASING RETURNS
TO SCALE world. Thus, by modeling IRS through externalities we get around
the problem of inexistence of competitive equilibrium. As it is well known,
however, these competitive equilibrium models with externalities will be NON
OPTIMAL. In section 1. 6 we show how Romer-:: (1986), · following Arrow (1962)
and Sheshinski (1967), postulates capital spillovers (externalities) in the
aggregate production function and finds that the ··model generates· steady
endogenous growth when p+~=l.

(b) A second way to get around the existence of the competitive

equilibrium problem is to drop the assumption of competitive behavior. This
is sometimes called the Chamberlinian approach to increasing returns. This

6
approach is interesting for a variety of reasons, one of the main ones being
that under imperfect competition the rewards to all inputs of production
does not .·exhaust, total.o.utput,:>,.~rHe.nce; .. there.. are rEHlts ...that ,can be, ass:i,gned_
to activities not. di·rectly productive but ;that . may, contribute .to- the
expansion of the··- frontiers of -knowledge• such as R&D. No~ surprisingly,
therefore, this approach has been extensively- used by economists --that ,th-i,nk"
that R&D is an important source of economic growth. .In sec.tion 8 we explore
8
a model of R&D =and. ,growth ·taken from Barro-'and ,Sala-i-Martin (1990 ;a) ·where
firms invest in R&D in search.of new-varieties of capital goods. In that
model, there are NO decreasing returns to the introduction of new varieties
so the incentive to perform R&D never diminishes, which keeps the economy
9
growing
Of course one could have models with both imperfect competition
and externalities. In fact •there - is an important line of research that
combines R&D (with imperfect competition) with externalities. It emphasizes
R&D as some activity exercised by firms in search for new varieties of
products· or higher quality products; · As a side product, R&D . increases the
general stock of .'knowledge. which -has. two effects. First, it decreases the
cost of further research (so the incentive to perform R&D remains positive
and Knowledge grows at a constant rate forever). And second it .increases
the productivity of --other inputs (such as -labor) in ·-·the production of a
manufacturing good. Therefore, given that the stock of knowledge grows at
constant rate, so does the manufacturing good. Models of this type include
Aghion and Howitt (1989), Grossman (1989) and Grossman and Helpman (1989 d
and e).
Before showing the mechanics of all these models, let us introduce

8
This model, in turn, -is •an extension of Romer (1987) and Grossman
and Helpman (1989, a).
9
There is a third way to model increasing returns in a model of
perfectly competitive firms and that is to introduce imperfect financial
markets. This approach has been taken by Greenwald, Salinger and Stiglitz
(1990).

7
 a graphical device that will further clarify the basic difference between
exogenous and endogenous growth models. It will also help,us understand why
,the ,,, savings , (or, investment),, rate-·,does not ''"af,£ec t',~long,,"run growthw-in "the '''''·1

first one•· and does so in the latter.

(b) The Role of Saving and Investment: a Graphical Exposition.

We ·-can O'ften he'ar economic advisors to third world countries •say ·

that one"of the necessary conditions for economic growth and development is
the increase in national savings rate. Higher savings will lead to higher
investment (since in a ,closed economy the two must be equal) and higher
investment will lead to more rapid economic growth. In this section we will
analyze•under what conditions this policy recommendation is valid.
Let us keep assuming that people save a constant fraction of their
income and .that the, government can influence it (for instance through
distortionary income taxes). Suppose that, for whatever reason, the
government manages to increase the economy's savings rate. What will the
, long run effects of such policy be?'.-
, In order to answer this question, let us start, by assuming that
the production -function c,is constant returns to scale< (a+,8=1) and dividing
bo.th sides _of the (per c·ap·ita) .capital accumulation equation (1.2)' to, get

(1.4) k /k = sAk-(l-,8) - (o+n)

t t t

The left hand side of this equation is the instantaneous growth

rate. Equation (1.4) says that the growth rate is the difference between
sAk~ (l-,8) and (o+n). We depict these two functions in Figure 1. The
function o+n is independent of k S,O it is a flat line. In the neoclassical
model ,8<1 applies. , This implies that the function sAk- (l-,8) is downward
t
sloping in k, and approaching zero asymptotically. Notice that the two
*
curves cross at a point k , the steady state capital labor ratio. Let us
now consider an economy with an initial level of capital k
0
lower than k . *
The initial growth rate of capital will be very large (notice that,
accor:ding to (1.4), the growth rate is the vertical difference between the
two curves) and it will be decreasing over time. Imagine for a second that

8
 we are in the steady state and, suddenly, the savings rate s, increases.
Figure 1 suggests that the curve sAk~ (l-{3) will shift to the right and
.: nothing will:,.happen to the (o+n) .line. We can see .that."the following things
are true:
(a) the growth rate·. will immediately increase.
(b) the growth• rate will be falling over time until, eventually, it
goes back to zero.
(c) the steady state capital. labor ratio is higher.

Hence, an increase in the savings rate generates a short term

increase in the growth rate and an increase in the steady state LEVEL of
capital per worker. It does not affect, however, the long run or steady
state growth rate, which is still zero. Under normal parameterizations, the
speed of convergence towards t;he new steady state is quite fast. For
instance, Barro and Sala-i-Martin (1990) suggest that the model predicts
that half the distance between k
0
*
and k disappears in less than 6 years!.
As it was mentioned above, Figure 1 suggests that the growth rate
for an economy'·which ·starts below ·the steady state is high and decreasing.
This, of course, implies that if economies differ ONLY on the initial
capital labor ratic:>' we·· should •'observe poor economies grow faster than rich
ones (in Figure 1, ·. different economies would be represented by• different
stocks of k
0
but all of them would have the same steady state k ) . *
Economists call this the. "convergence hypothesis". This hypothesis is
certainly true, but notice that .there is a big ONLY on it. That is,
economies may differ NOT ONLY in the capital labor ratio but also in the
level of technology (A), the savings rate (s), the depreciation rate (S), or
the population growth rate (n). If countries differ in one or more of these
parameters, they will end up in different steady states.
In Figure 2 we show the behavior.• of ·two economies, one called P
(poor) and one called R (rich). The poor economy has a lower initial
capital stock k p<kOR' (that is why it is called poor). We assume that the
0
poor economy also has a smaller savings rate so it converges to a smaller
steady state capital labor ratio, * *
kp<~. Notice that in this particular
example, it happens that the poor economy grows less than the rich one so

9
 there is no convergence in the absolute sense. Yet there is CONDITIONAL
convergence in the sense that each country converges to its own steady state
at· dimini'Shing growth rates. : :c Empirically,· this means that if w~ hold
··constant the steady state, poor -;·countries will . . grow faster ;than rich, one.s ..
<[f we don't, however, we will not. see poor ec.onomies growing faster unless
they are very similar (in the sense that they converge to similar steady
states).
Barro and Sala-i-Martin (1990) find that this feature of the
neoclassical model can be found in.the data. They find that the States of
the U.S. display absolute convergence while countries in the world do not.
Holding constant the steady state, however, there is convergence across
countries also. This makes sense if we think that the states of the U.S.
are similar .in the sense. of having. the same tastes and technology so they
converge to the same steady state. This is certainly not true for the large
cross section of countries, so they display conditional convergence only.
For related studies on convergence see Baumol (1986), Delong (1988), Dowrick
and NGuyen (1989), Manki~, Romer andWeil (1990), and Sala-i-Martin (1990).
Let us now expand the· neoclassical production function by
introducing exogenous productivity growth. Recall that the production
function now looks like

(1.1)' yt = A(t)k~

where A(t)=A(O)egt. Notice that, in terms of Figure 1, this specification

implies that the curve sA k-(l-,B) keeps shifting over time at a rate g.
t
This implies that the steady state capital labor ratio k
* keeps shifting at
the same rate. This is how the neoclassical model explains long run growth.
In Figure 3 we show that the implications from changing the
savings rate are very different when we consider a simple endoge·nous growth
model . . If the capital share is 1 (,B=l), the sAk-(l-,B) curve is a flat line
at sA. If· we assume that the economy is productive enough so as to have
sA>S+n, then the growth rate (difference between the two lines) is constant.
In other words, the economy grows at a constant rate equal to sA- (S+n).
Notice also that in this case, . an. exogenous increase in the savings rate

10
increases both the short run and the steady state growth rates. Hence,
contrary to the neoclassical predictions, policies directed to increases in
the savings (and. investment).·• -rates ·will have ·long ru,n ..growth effects.
Further, notice that· if economies differ in the•initial capital stock ONLY,
it is not true anymore·' that poor ·ones will· grow faster·· than rich. ones.
Finally,· this model predicts .that a, temporary· recession will have-permanent
effects. That is, if the capit'al stock temporarily falls for some exo.genous
reason (an earthquake, a•natural tragedy or . a . war that destroys part.of the.
capital stock), the economy will not grow temporarily faster so as to go
back to the prior path of capital accumulation. The endogenous model
described here predicts that after such a temporary reduction in the capital
stock, the growth rate will still be the same so the loss will tend to be
10
permanent
Figure 4 depicts the case where fi>l (IRS in the inputs that can be
accumulated11 ). The curve sAk-(l-fi) is upward sloping (and if fi>2 its slope
is increasing!). Notice that this implies growth rates that increase over
time. We will refer to this case again in section 6 (Romer (1986)).

(c) The Harrod-Damar Model.

Long before the neoclassical theory came to life in the mid 50's,
the most popular model of economic growth was the so called the Harrod-Damar
mode·l (developed by >Harrod (1939). and Damar (1946)). We can use the
graphical tool developed in the last subsection to learn about this older

10
There are unbelievable amounts .of papers on the existence of a
unit root in macroeconomic aggregates such ·as GNP. There seems . to some
evidence that, for the United States, GNP is non stationary, which is what
this simple model would predict. See Blanchard and Fischer (1989) Ch. 1 for
discussion of these issues.

11
In this case the assumption of GRS o:+fi=l must be dropped since a
negative labor share has little economic sense. Think of this case as one
where ·a=O {so •all. inputs· can be accumulated) and fi>l (so there are both IRS
and IR to capital.

11
 growth model.
Harrod and Domar tried to put together two of the key features of the
·<, ·Keynesian···economics . -·the··_><:multiplier. and the .accelerator-. in a model~ .that_.·· ·,_v
explained 1ong•i run economic gr.owth. We have been. using the multiplier
assumption (savings is >a ·fixed proportion of income) .alL along. so let .us
describe the differential feature of the Harrod-Domar model: the
accelerator. The increase in capital required to produce a given increase
in output is .assumed to be a constant number. In .particular,· . it .. is
independent of the capital labor ratio. That is

where A is constant. Notice that one production function that satisfies

this relationship is ·'the line~r AK production function used by the CRS
Endogenous Growth models. Thus one could be tempted to identify the
Harrod-Domar model with the new Endogenous Growth Models. Yet that would be
a mistake. The reason is· that Harrod and Domar were very concerned about
12
the effects.of growth on.long run employment and unemployment (their study
could be though to be '-an explanation for the then existing long run
unemployment of the Great Depression). Although they never introduced a
specific production function, the fact that they worried so much about
employment seems to indicate that they were not talking about a function
such as "AK", where there is no role for inputs such as labor.
Another production function which satisfies the accelerator
principle and which is closer to the spirit of what Harrod and Domar had in
mind is the Leontief Fixed Coefficients function. Output is assumed to be
produced by a fixed proportion of capital and labor. Given this proportion,
an increase in the level of one of the inputs without a corresponding
increase in the other leaves output unchanged. Thus, we should replace the

12
In fact, Domar's paper is called "Capital Expansion, Rate of
Growth, and Employment".

12
production function (1.1) by

where A and B are• exogenous production parameters. After rewriting this

function in per capita terms -y=min(Ak,B)- we plotted it in Figure 5. We
see that there is a capital labor ratio k *=B/A that has the following
property: * Ak is smaller than·B so
for capital labor ratios smaller·than k,
output is· determined by Ak. For. capitaL labor ratios.,larger than k * , .Ak .is
larger than B so output is determined by B. In other words, this production
function can be rewritten as

for all kt<k*=B/A

(1.1)
for all kt>k*=B/A

Notice that this technology is similar to the Ak model ·but only for
small capital labor ratios. For large ·ones, 'however, the production function
is flat so the Marginal Product of Capital is equal to zero. We can now
a.pply the basic .savings equal investment equality (1.5.) to this technology.
to get.

JsAkt+(l-6-n)kt for kt<k*=B/A

(1. 7) kt+l
+(1-6-n)kt for kt>k*=B/A
=
lsB

As Harrod and Domar pointed out, there are three possible

configurations of parameters each of which will yield different implications
for growth and employment.

CASE 1: sA<6+n
When the savings rate and/or the marginal productivity of capital
are very small compared to the aggregate depreciation rate (which includes
population growth), there is no possible steady state. This is pictured in
Figure 6. Notice that the economy converges to a point where the logarithm
of the capital labor ratio is minus infinity· (so the capital labor ratio

13
 converges to zero) . In this case not only there will be unemployment
(because AK<BL) but it will grow over time. Harrod and Domar thought that
~.·t • this was 'a "good de·scr:i:ption of ..;the observed large .and1 growing unemployment _,
rates of the 30's.
CASE 2: sA=o+n
When, by chance, the exogenously given savings rate and marginal
product of capital are such that sA=o+n, the economy will reach a steady
state where all the per capita variables grow at a zero rate. In Figure- 7
we show· that,, in this case, the initial. capital labor . ratio .will .b,e .,the
steady state one.
CASE 3: sA>o+n
The third case, depicted in Figure 8, is one where the marginal
product of capital or/and the savings rate are very large relative to the
depreciation rate. We see incFigure 8 that, for small capital labor ratios,
this case looks very ·much like the Rebelo model. But as the capital labor
ratio grows, the labor requirement gets binding (that is we hit kt =B/A at
" ·•s01ne. 'finite point .ini,vtime). :-;. After this. point,.,~the- marginal product. of.,.
capital is zer·o and .the ·per. capita« growth process stops. The steady state
capital labor ratio, k* will. be one where there will be. unemployed
machinery.
Two out of the three configurations of parameters yield long run
equilibria were there are idle •resources and the only that does not, would
be achieved-•only by chance: remember ,that all the relevant parameters -A, s,
o and n- were given by mother nature. The question is why in the world
would mother nature be so kind as to give us exactly that configuration of
parameters?. In other words, the chance of them being such that the
equality above is satisfied are quite small.
At the time, the Neoclassical approach was seen as a way of
solving this knife edge property of the Harrod-Domar model. That is, the
neoclassical production function achieves the equality between sA and o+n by
13
allowing for A (the marginal product of capital) to be variable in k . We

13
And we know that there will be a level of capital k such that the

14
 should just mention that there are other non neoclassical ways of achieving
this equality. One of them, proposed by, the old Cambridge School in England
'was to argue that '.the savings ,,rate was .endogenous. . ·They- thought that.
workers had a different marginal propensity to save from capitalists.
Hence, so they argued, ; ;in the process. of., economic growth there will be
changes in the distribution of income that will lead to changes in the
aggregate savings rate in such a way that the equality between sA and S+n
14
will be guaranteed We will not talk about the Cambridge school of thought
anymore.

(d) The "Sobelow" Production Function.

Finally, with this graphical approach we can see that the growth
paths are not limited·to the cases seen up to now. ·We could find functions
that behave in some other ways, ..we may discover new growth models and new
transitional dynamics . towards steady states.·. ·Consider Figure 9: The steady
· ~- -state:. is: sim:ilar .. ·.~~to,. ~the ...,one::· . de.s.crib~d ;~.,by ·.-the Rebelo ·. mo.del b.ut "···the·,
transitional dynamics •are different. · One production function that exhibits.
such dynamics is the following:

(1.9) Yt = AKt +BK~

This production function was first. proposed by Kurtz (1968) and

Gale and Sutherland (1968) and later reintroduced in the endogenous growth

marginal product of capital is equal to (S+n)/s since the marginal product

is assumed to range from zero (f'(~)=O) to infinity (f'(O)=~) in a
continuous fashion.
14
This was one of the main differences between the Cambridge (U.S.) and
the Cambridge (U.K.) school of thought. The other main difference was that
the british rejected the Neoclassical production function and, in
particular, they rejected the notion of aggregate capital stock. They
thought of capital as a number of different machines which, combined with
different types of workers yielded different types of output. Such a
heterogeneous set of objects, they argued, is impossible to aggregate into a
single variable called Aggregate Capital stock. See Robinson (1954).

15
literature by Jones and Manuelli (1990). Notice that this function is half
fi ' 15
way between Solow (BK ) and Rebelo (AK) . It has all the nice properties
·required by the Kuhn Tucker theorem so we can apply straightforward ·
optimization techniques to find solutions.
In per capita terms the Solow production function is concave and,
as k tends to infinite, the marginal product of capital approaches zero.
The Rebelo production function in per capita terms is linear with. slope
equal to A for all values- of k. The Sobelow production function is also
concave for all ·capital-labor ratios . . As k goes to infinity, however, the
slope of the production function does not go to zero but to A. For large
levels of k, therefore, it gets arbitrarily close to the Rebelo production
function. Hence, the only difference between the Solow and the Sobe low
functions is the latter does not satisfy the Inada condition.
We observe in fig~re 9 that sf(k)/k now is not going to zero
asymptotically but to A. As Kurtz (1968) noted, if A is sufficiently large
(in this case this means if sA>o+n), then the steady state growth rate is
positive, even though-, there' is a transition, period" where growth rates -are
decreasing monotonically. , It .is worth noticing that -if the economy has
been· going on for a while, the decreasing returns part of the production
function will· be almost ··'-irrelevant so we might as well ·deal with the
(simpler) -linear technology described above.

(e) Poverty Traps.

Another possibility could be the one in Figure 10. Here we see

the function sf(k)/k crossing the horizontal line (o+n) twice so there are
two steady states. The lower crossing represents a "stable poverty trap".
That is, countries whose initial "capital" (here we define capital in a
broad sense that includes all inputs that can be accumulated) is very low
will tend to this zero growth-low income trap. In fact all countries whose

15
That we.call it the Sobelow production function.

16

-- .:•..:.. ..
,.·
initial capital lies to the left of k * will fall into this trap. Countries
2
that start to the right of this trap will tend to a constant growth steady
state a la Rebelo.
In the next two •sections we .will present the optimizing versioTl.s
of the Neoclassical models we have been, talking about in this introduction.
In sections 4 through 7 we present the "new" growth models of the 80's. It
is useful to think about them in terms of·being optimal saving versions of
the /3=1 model we just presented in this section.

(2) The Ramsey-Cass-Koopmans model

(a) The Model.

All optimizing growth models we will assume that consumers choose

a path of consumption so as to maximize a utility function of the form:

(2.0) U(O)
0

Where p is the discount rate, ct is consumption per capita at time

t and Lt is population. We can think about horizons being infinite (despite
the fact that, obviously, lifetimes are not) if, following Barro (1974) we
think that individuals care about their utility AND about their children 1 s
utility. In this sense, we must think of the agent as being a dynasty or
family the number of individuals of which grows 'over time. Under this
interpretation, the discount rate (which was described by Ramsey (1928) as
"ethically indefensible and arises only from the weakness of the
1116
imagination , (p. 543) at the individual level) represents the fact that
individuals care more for their own utility than the one of their children.

16
Ramsey was considering the optimal choice from a government's
point of view. He thought that introducing a discount rate was ethically
indefensible because that meant that the government was giving a larger
weight to current than to future generations.

17
Since ct is consumption per capita, u(ct) is the instantaneous per capita
felicity. Hence, the instantaneous felicity for the whole family is equal to
.the individual times the number of people in the family.
We assume that there is only one good (cookies). We will assume
17
that households OWN the firms (or that there is only household production )
so they can consume this good or they can nail it to the floor. The reason
why anyone would .·do such a horrendous thing· is that cookies nailed to. :the
floor can be used to , produce. more cookies in the . future ... For lack . of a
better name, all cookies nailed to the floor will be called "capital" and
will be represented by Kt. We assume that there is nobody else in the
universe, so all the cookies produced will have be consumed or nailed. Hence
the increase in existing capital (called investment) must be equal to
saving. If we let kt be per capita capital (Kt/Lt), the following resource
constraint must be satisfied:

(2.0)' k f(k) - c -nk -6k

Notice that n is like a· "depreciation rate" because it represents

the fraction of resources that we need to give to new generations. The key_
Neoclassical assumption is a production function that expresses NET output
in per capita terms as a function of capital per capita with the following
properties: .twice differentiable, with f(O)=O, f' (k)>O, f' '(k)<O, f' (O)=oo
18
and f' (oo)=0 .. A simple Neoclassical production function that we will be
using throughout is the Cobb Douglas: f(k)=k.B with 0<,B<l. Population is

17
As we will show in the next section, the results will be the same
we would get if we assume that households own capital and labor and sell
their services to competitive firms in exchange for wages and rents.
18
The last two conditions (the Inada conditions) are often swept
under the rug. They are of crucial importance because, as Kurtz (1968)
showed, the mathematical difference between an endogenous and exogenous
growth model is the condition lim f' (k)=O. This point has been emphasized
k->oo
also by Jones and Manuelli (1990).

18
assumed to grow at the (exogenously given) rate n so we can rewrite the
program as:

(2.1) MAX U(O) =

J
oo e -(p-n)t[c 1-a -1] dt

0
-t - -
1-a
s.t. k f(k) - c -nk - ok
k(O) >0 given

For U(O) to be bounded (U(O)<oo, and the program to be meaningful

at all) we need the term inside the integral to go to zero as t goes to
infinity. This implies

. e -(p-n)t[c 1-a -1·]

Lim 11·m ~
-(p-n)t
c
1-a
- lim e
-(p-n)t
1/(1-a)
t t
t~OO - --
1- a 1-a

In steady·state ct will be constant (we will show later). Hence,

if this' limit has to be zero, ·it must be the case that

(2.2) p>n.

To solve the model, we set up the corresponding Hamiltonian

(2.3) H()

where v is the dynamic Lagrange multiplier (or shadow price of

investment). The first order conditions are the following:

-(p-n)t -a
(2 .. 4) H 0 ~ e c - v = 0
c
(2.5) Hk -v ~ v = -v(f'(k)-n-o)
(2.6) TVC lim (ktvt) = 0
t~OO

Equation (2.4) says that at the margin, the value we will give to

19
consume one more unit will be equal to the value we will give to invest one
more unit (that is, we, will ,,be" indifferent between consuming and investing
the unique good) . Take logs of (2.4) to get -(p-n)t-alog(ct)=log(v). Now
take the derivative with respect to time to get:

(2.7) -(p-n)-a(c/c) v/v

so

-1 .
(2.7)' c/c a (-p+n-v/v)

We can now plug this in (2.5) to get,the traditional condition for

consumption growth:

-1
(2.8) 1=c/c =a (f'(k)-p-o)

This"',equation' can be ,r,ewrittem'as p+a(c/c)=f' (k)-o and interpreted

as follows: The left hand side represents .,the return to consumption. The
discount rate represents the gain .in utility from consuming today since we
prefer consumption for ourselves rather than for our children. The return

to conswnption also includes ac/c. If we want to smooth consumption over

time (a>O), then we want to increase consumption today, whenever we expect

consumption to be higher in the future (ie, when c/c>O). The return to

saving (and investment) is the marginal product of capital minus the
depreciation rate, o. Optimizing individuals should, at the margin, be
indifferent between consuming and investing. This indifference is the one
represented by equality (2.8).
Using the Cobb-Douglas technology, (y=kP), equation (2.8) can be
written as

-1 -(1-P)
(2.8)' 1 =a (pk -p-o)
t

20
If we define steady state as the state where all the variables grow at a
constant {and possibly zero) rate,- equations (2.8) together with the capital
accumulation equation (2. 0)' say that there is steady state k *
a unique
. h ensures t h at capita
wh ic . 1 and consumption
. .
per capita d o not grow19 Hence
this model says that, in the steady state, all variables in per capita terms
do not grow at all. Alternatively, all "level" variables grow at the same
rate as population, which is assumed to be exogenous.

{b) Competitive Solution.

Since this model is concave (concave preferences and technology)

and there are no externalities of any kind, the OPTIMAL PROGRAM (command
economy ·solution) will yield .. the same solution as the COMPETITIVE
EQUILIBRIUM PROGRAM,· provided· that consumers and firms have RATIONAL
EXPECTATIONS (since ..• these models do not. have uncertainty, rational
expectations implies PERFECT FORESIGHT). We can show that the competitive
solution is the same as the' one we solved. On the· consumption side,
individuals maximize (2.0) subject to

(2.9) kt = w + r k - c
t t t t

19
We can show that the only sustainable growth rate is zero:
take the constraint k=k,B-c-nk-ok and divide it by k. Define k/k=1k which in
steady state will be a constant (by definition of steady state!!). Realize
that k(,8-l)=ba+p)/,B. Rearrange to get c/k=ba+p)/,B-1k-n-o=constant. Take

logs and derivatives to conclude that c/c=k/k=1k=1. Now consider again the
equality k(,8-l)=(1a+p)/,B. The RHS is a constant. Take logs and derivatives
of both sides to conclude that (,8-l)1k=O. This is another way to show what
we saw in section 1: if there are DR to k (,B<l), then the steady state
growth must be zero. The only way to achieve nonzero growth rates is to have
CR to k (,8=1).

21
 where wt is the return to labor (wage) and rkt is the return to
capital (we are abstracting again from depreciation and population growth).
In the other side, competitive firms will price factors at marginal costs
so:

f' (kt)-&k
f(kt)-ktf' (kt)

Notice that w+rk= f(k)-kf' (k)+kf' (k)-&k=f(k)-&k so substituting

(2.10) into the individual budget constraint will give the original resource
constraint (2.0)'.

(c) Transitional Dynamics, Golden Rule, and Dynamic Efficiency

The neoclassical model just outlined is NOT a very interesting

model of steady state growth (because steady state growth ··is zero). It is
nevertheless an interesting ,·-model of the transition towards the steady

state. This transition is·shown in Figure 11. The vertical line is the c=O

locus·. The upward sloping line is the k=O locus representing the· resource
constraint (2.0)'. Notice that the economy can converge to the steady state
from below or from above. The interesting case is the one where we converge

from· below so we '.actually grow. Along this• path k/k>O. Per capita capital
grows, but it does so at a decreasing rate (which ends up being zero in
steady state). As the capital labor ratio increases, the marginal product
of capital falls and, therefore so does the interest rate.
It is worth noticing that, in Figure 11, there is a level of
capital called kgold (for Golden Rule). This is the capital level that
maximizes steady state consumption. Froni the budget constraint we see that

when k,,,;O, steady state consumption is equal to c *=f(k)-(&+n)k. The capital

labor ratio that maximizes c * is the one that satisfies f' (kgold)=(n+o).
This level capital divides the set of capital labor ratios in two. Capital
levels above the Golden Rule have the property that in order to achieve
higher steady state consumption, the economy needs to get rid of some

22
capital. In other words, in order to achieve higher consumption in the
future the economy would need to dissave (which of course means higher
consumption today). Therefore, if the economy were to find itself in one of
such capital levels, everybody could increase consumption at all points in
time. The points above kgold are called the DYNAMIC INEFFICIENT REGION
because some generations could be made better off without making any
generation worse off. Notice that for capital levels below the Golden
Rule, if the economy ·wants to increase the steady state consumption, , it
needs to· accumulate· or save: higher consumption tomorrow would have to be
traded for lower consumption today. This region is called DYNAMIC EFFICIENT
REGION.
We can integrate (2.5) forward between 0 and t and get

e-I(f'(ks)-o-n)ds
(2.5)' lit =110

which, after substituting in the TVC yields

_ r(f' (ks)-o-n)ds
(2.6)' lim 11
0e ~ kt 0
t->oo

Since is positive, it must be the case that the second term in

11
0
(2. 6)' is equal to zero. Notice also that this implies that in the steady
state, the marginal product of capital must be larger than o+n. This
condition is always satisfied in steady state if we assume that utility is
bounded. Recall that this condition required p>n and the steady state
implies f'(k)=p+o so this ensures that f'(k)>n+o. Notice how this
inequality implies that the capital per capita in the steady state will be
20
dynamically efficient (to the left of the golden rule) .

20

23
 I
I
I
(d) Ruling out explosive paths.
I
It just remains to be shown that, given the saddle-path stability I
property of the model, the economy will find itself on the stable arm. To
show this we must rule out all other possible paths. Suppose that we start
with the capital stock k in Figure 11. Let c the consumption level that
0 0
corresponds to the ,, saddle path. Let 1us· imagine; first that the <initial
consumption level is c >c .
0 0
If this is the case, the economy will follow
the path depicted in Figure 11: at first both c and k will be growing. At

some point the economy will hit the k=O schedule and, after that,
consumption will keep growing yet capital will be falling. Hence, the
economy will hit the zero capital axes in.finite time. At this point, there
will be a jump in c (because with zero capital there is zero output and
therefore zero, 'consumption) which will violate the first order condition
(2.8). In order to show that the economy will hit the k=O axes in finite
time just realize that kt can be rewritten as

(2.11) k ds
s.

Suppose that T . is the time at which we hit the k=O schedule.

After that moment, kt evolves according to (2.11). If we show that dk/dt is

negative, we will have that k is negative and falling so k is falling at

increasing rates. This, of course implies that there is a time T' at which

it will be zero. The derivative of k with respect to time is ((from 2.0)')

Recall that k * is such that f' (k* )=p+6 and that the bounded
utility condition (2. 2) implies that p>n. Therefore
f' (k )=p+6>n+6=f' (k · ) .. Since the production function is concave (f' '<0)
gold*
it follows that k <kgold"

24
(2.12) dk/dt [f'(k). - (n+cn]k: - c <O

notice that since kt<k* , we know that f'>n+o. We also know that k<O and c>O
so overall, (2.12) is negative which implies that k is falling at increasing

rates. '.Hence, if:we--are ·in. this region:•in·.finite- time: (ie·.if we .hit;:the-... k=<Oc,...
schedule in finite time), then kt will-hit zero in finite time.· Therefore,

it ONLY remains to be shown that we will hit the k=O schedule in finite

time). We can show that this is the case because around the k=O schedule,

consumption increases at increasing rates so it will reach the k=O schedule

in finite time. Notice that the derivative of c with respect to time is

(2.14) d~/dt = (l/a)[f'(k)-(o+p)J~ + (c/a)[f''(k)Jk

noti:ce::.·,.::that''' the ;first.·· terms.is' positive.· and, around the k=O schedule it.·.

dominates-the second term so overall dc/dt>O. Hence, if initial consumption

is larger than the one required by the stable arm we will first hit the k=O

schedule in .finite-time and then hit the k=O axes in finite time. This will
imply a finite time jump in consumption which will violate the first order
condition (2.8). Hence, it is not optimal to start above the stable arm.
Let us imagine next that we start below the stable arm. The
dynamics in Figure 11 tell us that we will converge to k ** Notice that
this path will violate the transversality conditions since k **>kgold' That
is

- (f'(k** )-o-n)ds
lim k
t->oo
** e I >0

which is positive since the term inside the integral is negative.

25
Hence, initial consumption levels below the stable arm are not optimal
either. We are left, therefore, with the stable arm as the UNIQUE optimal
path of this model.

(e) Convergence and Convergence Regressions.

The Neoclassical model just described has . the ·. additional

implication that, if ·all >Countries share the , same. ,.production. and utiLity.
parameters, then poor countries tend to grow; at a faster rate than· rich
ones. In other words, income or output levels will converge over time.
Following Sala-i-Martin (1990), we can show this important implication we
can linearize the two key differential equations (2. 8)
and the capital
21
accumulation equation (2.0)' budget constraint around the steady state If
we express all variables in log~rithms the system becomes

-(1-a)ln(k )
(1/a)(ae t -(p+o))
(2.15)

e
-(1-a)ln(kt) - e (ln(ct)-ln(k t )) -
( n+o~)

In steady state the two equations are equal to zero so

-(1-a)ln(k* )
e (p+o)/a
(2.16)

e
*
(ln(c )-ln(k*)) e-(1-a)ln(k* ) _ (n+o) h >0

where c* and k* are the steady state values of ct and kt respectively and
h=(p+o(l-a)-an)/a. We can now Taylor-expand the system (2.15) around (2.16)

21
See King and Rebelo (1990) and Barro and Sala-i-Martin (1990) for
a discussion of convergence when the economy is far away from the steady
state.

26
and get

-µ[ln(kt)-ln(k* )]
(2.17)

-h[ln(ct)-ln(c * )] + (p-n)[ln(kt)-ln(k* )]

where µ=.(1-a) (p+o)/a>O.. or

.
ln(c )]
(2.18)
[
ln(k:)
notice'''that the·. determinant of the matrix is detA=-hµ<O which implies
that the system is saddle path stable. The eigenvalues of the system are

2 )(1/2)
-A = (l/2)(p-n - ( (p-n) +4µh ) <0
1
.(2.19)
2 (1/2)
A2 = (l/2)(p-n + (<p-n) +4µh) ) >0

The solution for ln(kt) has the usual form

.~le-Alt
'I' + .~'1'2e A2t

where and .,µ are two arbitrary constants. To determine them, we notice
.,µ
1 2
that since A is positive, the capital stock will violate the transversality
2
condition unless .,P =0. The initial conditions help us determine the other
2
constant since at time 0 the solution implies

(2.20)' ln(k ) - ln(k*)

0

27
 Hence the final solution for the log of the capital stock has the form

(2.21) ln(kt) - ln(k ) * *

[ln(k ) - ln(k )]e ->. 1 t
0

If we realize that ln(kt)=ln(y t) /a: and we subtract ln(y 0) from both

sides of equation (2. 21) we will get what is known as the "convergence
equation"

*
where a=ln(y )(1-e
->. t .
1 )/t and ~=(1-e
->. t
1 )/t. This equation says that if a
set of economies have the same deep parameters (discount rate, coefficient
of intertemporal elasticity of substitution, capital share, depreciation and
population growth rates,· etc} sd they converge to the same steady state, the
cross section regres·sion of growth on the·· log of initial income should
display a negative coefficient. In other words, poor countries should tend
to grow faster. The· .reason for- that is that countries with low .initial
capital would have high .initial. marginal product of capital. That would
lead them to save, invest and therefore grow a lot.
If countries converge to different steady states, however, there should
be no relation between growth and initial income, unless we hold constant
the determinants of the steady state. Sala-i-Martin (1990) and Barro and
22
Sala-i-Martin (1990) use a slightly more complicated version of (2.22) to
show that the states of the U.S. (which we may think are described by
similar production and utility parameters) converge to each other exactly
the way equation (2.22) predicts. They also show that, once they hold
constant the determinants of the steady state, large sample of countries
ALSO converge to each other the way equation (2 .. 22) predicts.

22
It is a slightly more complicated version because they include
exogenous productivity growth.

28
 (3) EXOGENOUS PRODUCTIVITY AND GROWTH

(a) Classification of Technological Innovations.

As we just mentioned, the simple neoclassical model predicts that

the long run growth rate is zero. In order to explain observed long run
· growth neoclassical economists amended the model and incorporated exogenous
productivity growth. ·In section 1 we saw that, in •the fixed saving ·rate
models, the introduction· of productivity growth lead to long run economic
growth. The question ·is what kind of technological progress should be
introduced. Some inventions "save" capital relative to labor (capital
saving technological progress), some save labor relative to capital (labor
saving technological progress) and some do not save either input relative to
the other (Neutral or,unbiased technological progress).
Notice cthat the 1definition of neutral .innovations depends on what we
mean by "saving". The two most popular definitions of unbiased or Neutral
>technological .progress~are. due i to ,Hicks and Harrod resp.ec.tive,ly.
Hicks says that a technological innovation is Neutral (Hicks-Neutral)
with respect to capital and labor if and only if the ratio of marginal
products remains unchanged for a given capital labor ratio. Consequently, a
technological innovation is labor (capital) saving if the marginal product
of capital (labor) increases by·. more than the marginal product of labor
(capital) at a given capital labor ratio. Notice that Hicks neutrality
amounts to renumbering the .isoquants. Production functions with Hicks
Neutral technological progress can be written as

where A(t) is an index of the state of . technology at moment t evolving

according to At= A egt (ie, A/A=g) and were F() is still homogeneous of
0
degree one. The second definition of technological unbias is due to Harrod.
He ·says that a technical innovation is neutral (Harrod Neutral) if the
relative shares (KFk/LFL) remainunchanged for a given capital OUTPUT ratio.
Robinson·· (1938) ;and .. Uzawa. (19.61) . showed that this implied a production

29
function of the form

where, again, A(t) is an index of technology at time t, A/A=g and F is

homogeneous of degree one. Notice that this production function says that,
with the same amount of capital, we.need less:and less .labor to·.-produce ..,the,
same amount of output. - Therefore, this .function •is also , known as labor
augmenting technological progress. By symmetry we could have thought .of
technological change as being "capital augmenting", ie Y=F(BtKt, Lt). This
would mean that, for a given number of hours of work (Lt), we need
decreasing amounts of capital to achieve the same isoquant.
The reason why we care about what kind of technological progress
we should postulate is that,- as Phelps showed, a necessary and sufficient
condition -.for.- the existence of a steady state in .an economy with exogenous
technological progress -is for this technological progress to be Harrod
Neutral -or Labor ··Augmenting;··Notice,- however;· that., when· we >work with· Gobbo--· -
Douglas utility functions the.two types of progress are identical since

Y(K,AL) BY(K,L)

(b) The Irrelevance of Embodiment.

All types of technological change we have been talking up to now

are "DISEMBODIED" in the sense that, when a technological innovation occurs,
ALL existing machines get more productive. An example of this would be
improvements in computer software: it makes-all existing computers better.
There are a lot of inventions, however, that are not of this type. When one
invention occurs, only the NEW-. machines are more productive (as it is the
case with computer hardware). Economists call this, "EMBODIED TECHNOLOGICAL
PROGRESS".
In the 60's, when the neoclassical model of exogenous productivity
growth was being developed, there was a debate on the importance of

30
 embodiment in economic growth. Proponents of what at the time was called
"New Investment Theory" (embodied technological progress) said that
,. investment in new .machines had the usual effect. of increasing the capital
stock and the additional effect of modernizing the average capital stock.
Proponents of the "unimportance of the embodiment question" argued that this
new effect was a level effect but that it did not affect the steady state
rate of growth. In a couple of important papers Solow (1969) and Phelps
(1962) showed the following:
(1) The neoclassical model with· embodied technological progress and
perfect competition (so the marginal product of labor is equal for all
workers no matter what the vintage of the machine they are using is) can be
rewritten in a way that is equivalent to the neoclassical model with
disembodied progress (Solow (1969)).
(2),The Steady State growth is independent of the fraction of progress
that is embodied (it depends on the total rate of technical progress but not
on its composition) (Phelps (1962)).
(3) The convergence or speed of adjustment to the steady state growth
rate is faster the larger the fraction of embodied progress (Phelps (1962)).

Thus, the distinction between embodied and disembodied progress seems"

unimportant when studying long run issues but might be crucial when studying
23
short run dynamics The modeling of embodied technological progress is
quite complicated because one· ·has . to keep track of all old vintages of
capital and ·associated labor. , Yet ·a simple way to think about it is to
postulate a technology-free production function Y=F(K,L) and an accumulation

function of the form K=A(t)(Yt-Ct) where A(t)/A(t)=g and K(t) is a measure

of aggregate capital. This function reflects the fact that a unit of saving

23
The importance of embodiment in modeling business cycles can be
seen from the fact that an embodied shock affects the marginal product of
capital but does NOT affect the marginal product of labor or current output
supply. This is a key difference with respect to a disembodied shock,
especially as far as ·the implications for the procyclicality of real wages
and real interest rates is concerned.

31
(Y-C) in a later period generates a larger increase in capital than a unit
of saving in earlier time. This is like saying that later vintages of
capital are more productive.

(c) The Neoclassical Model with Technological progress.

Let us go back .to . the labor. augmenting , foxm as depicted ..,in .

equation (3 .1)'. To solve this' model it is going to be useful -to define· .>the
concept ·of "effective labor", L.

(3.2) Lt Loe nt ---->LA t =L0 e(n+g)t

In words, for a given size of physical population we get more

effective labor as time passes by. Since, on the other hand, the number of
physical ·bodies increases at the constant rate n, the effective labor force
grows at rate g+n. Notice that using this definition we can rewrite the
production function as follows.

Let's divide both sides of (3.3) by Lt, define y=Y/(L) and

k=K/(L). The CRS assumption implies:

I\ A

(3.4) y f(k)

Again, the closed economy assumption implies that domestic savings

equal gross domestic investment so Y=K+C-oK. Divide both sides by L and get
I\ • /\ I\ I\ A • A

y=(K/L)+c-ok. By the definition of k, we know that k=K/L - (n+g+o)k, which

we can plug in savings equal investment equality to yield:

(3.5) k f(k) - (n+g+o)k - c

Consumers maximize a utility function of the form (2.0) subject to

32
(3.5). Notice that the utility function is defined in consumption per capita
(per physical body) while the budget constraint is ·defined in terms of

(3.6) U(O) J e - (p-n) t [ (~tegt) i-u -1] Lodt

o 1-a

, We have to choose ct so as to maximize (3.6) subject to (3.5) and

subject to K , L and A . Set up the Hamiltonian:
0 0 0

(3. 7) H() ~ - (n+g+6 )k]

The F.O.C. are the following:

(3.8) H" 0 ~ e - (p-n) t e gt ("ce.gt)-a - v 0

c
(3.9) H" -v ~ v -v(f'(k)-n-g-o)
k "
(3.10) TVC lim (ktvt) = 0
t~OO

By following the same steps as in the previous section, we will

find that:

" "
(3.11) v/v -(p-n) + g -ac/c - ag f' (k) + n + g +o

" "
by setting c/c=O we will get the steady state condition:

(3 .12) f' (k) p+ag+o

Observe.that this result is exactly parallel to the one in section

two ·(equation (2. 8)). ·The .difference here is that the growth rate relates
to· consumption per unit of efficient labor. This means that, since

33
variables in efficiency units do not grow, variables in per capita terms
grow at the constant rate g.

(c) Bounded Utility Condition.

For U(O) to be bounded, again, we need the expression inside the

integral to tend to zero as t goes to infinity.

(3.14) lim e-(p-n-g(l-a)c~-a/(1-a) - lim e-(p-n)t/(1-a)

Note that if p > n+g(l-a) > n, the second term goes to zero. Since c
will end up growing at rate g, the first term also goes to zero if the above
condition holds. Notice, finally, .that this condition implies that the TVC
is satisfied and that we will end up at a point to the left of the golden
rule (dynamically efficient region).
Finally, let's analyze the saving rate.

(3.27) s/y=(k/y)+(nk/y)=(k/k)(k/y)+n(k/y)=(-y+n)(k/y)=
(-y+n)/(k-(l-,8)eg(l-,8)t)=(-y+n)/[(p+a1)/,B]=(g+n),B/(p+ga).

A patient society (low p) will save more and end up with a higher
output LEVEL along the balanced ,path than an impatient one. She will not,
however, grow at a faster rate. we have seen that the growth rate depends
on g and n only. This is an important implication of the neoclassical model
of economic growth.

34
 Figure 1 : The Neoclassical Model
B-1 B<1
' \ sAk , o+n

Growth Rate

'\

l.;.)
V1

6+n I•_. .... ".,

,/
I ! .. m. . .

sAk B- 1

' \

ko k* kt

' \
:.·
' \

Figure 2: Conditional Convergence in the Neoclassical Model

B-1
sAk , cS+n
Poor Country-Low Growth Rate

Rich Country-High Growth Rate

w
°'

cS+n ~~~~+---'-~r--~~__,__~~~~~~~~~~

B-1
s~kP B-1
S~kR

~p kOR ~ k* kt
R
cu
"O
0
E
~

4=0
cu
_a
cu ,-
a: II cu
cu en
..c co
a:
I-
.. ..c
C')
~
Q)
1...
0
1...
:J
O> CJ
·-
LL

c
+
\.Q
... c
<(
en +
\.Q

37

. .,,. ~· :: ; .:.. ,.
 Figure 4: Increasing Returns and Increasing Growth Rates

sAkB-1, c5+n B> 1

'/Increasing Growth Rates

w
00

c5+n I ! .... ... ... ...

k kt
0

-------"·--·----.----- --- -----~------ - -··,-----,--...,......-

 c
0
·-
t5
c
::J
LL
c
0
f5::J
"O
0Lo
a..
Lo
cu
E
0
0I
"O
0
Lo
Lo
cu
I

t
(])
.c
1-
LO
(]) ~
Lo <(
::J
O> II
·-
LL

_ _ _ _ __ j __ _ _ _---'I Q,'

39
 Figure 6: The Harrod-Damar Model
Case 1: sA<(6+n)
sf(k)/k, 6+n

.p-.
0

Negative Growth Rate

6+n~~~~--r-+~~~~~~~~~~~~~~~

sA
~sf(k)/k

ko 8/A kt
 Q)
"O
.o
:?
"-
co
oc
E
0I +
t.()

e
"O ..........,
II
"- <(
~ .en.
Q) C\I ···················· ~
co
..c Q)
J- en
. . (.)
co
I"-
Q)
"-
::J
Cl
·-
LL

c
+
t-0

--
~

2
-
..........,
'+-
en
<(
en
II
c
+
t-0

41

.,, - · - ,.-. v
... - .
~ . ,.·. .
 Q)
"'O
0
~
co
'--

E ............
0 c
0I +
"'O .._....
\.<)
0
'-- /\
co
'--
<(
en
I
Q) CW')
..cQ)
I- en
co
co (.)
Q)
'--
::J
Cl
·-
LL

c
+
\.<)

--
""
~
............ <( c
.._....
~
'+-
en +
\-0
en

42
 Figure 10: Stable Poverty Trap
sf(k)/k, o+n

<
\

Stable Poverty Trap Threshold Point

.;::...
w

o+n I ..;> 1~ .... .... .... .... ?'/..... • ..

k* k*
1 2 kt
 Figure 11 : The Ramsey-Cass-Koopmans Phase Diagram

c
t
6=0

.
k=o
"
''·

+:-
+:-

c'0 .
l .............../ ........... ,

co
c"
0
·~~k*-;---~k~~===~~ ko gold k** kt

-~------,---~------------~-.----·---------. -----~-.----·----·----~-----· ·~·~-~----------