Econ 620 2007

2. Growth Theory
2.1 Motivation and some facts.
Canadian real per capita GDP grew by a factor of 13 from CAN $1,694 in 1870
to CAN $22,300 in 2001. The average growth rate over this period was 1.98%1.
Canadian GDP per capita

2000

1990

1980

1970

1960

1950

1940

1930

1920

1910

1900

1890

1880

1870

25,000
20,000
15,000
10,000
5,000
0

What if the growth rate was 0.8% (like the Peru from 1960-1997)?
In that case current real per capita GDP would have been CAN $4,700 only 2.8 times the
income of 1870. Canadian average income will be close to the level of Brazil or Russia.
What if the growth rate was 2.93% (like Israel from 1960-1997)?
In that case current real per capita GDP would have been CAN $76,400, 43 times the
income of 1870 and more than 3 times its current income.
Comparing with your grandparents (born 50 years / 2 generations ago)
- Taiwan (5.6%)
16 times richer
- Zaire (-1.8%)
had 40% of what your grandparents had
- Canada (1.6%)
2.2 times richer
1

Most of the calculations are based on the data that I have posted in WebCT. When we look at
comparisons within a country across time data is always expressed in real terms, i.e. adjusted for changes in
prices (and to a minor extent quality adjusted). When comparing different countries at a same point in time
data is Purchasing Power Parity Adjusted. This is the best comparable data available. For those of you
interested on how this type of data is constructed Berthold Herrendorf provides a good description in his
first
CESifo
internet
lecture
that
can
be
found
at
http://www.cesifogroup.de/portal/page?_pageid=36,164276&_dad=portal&_schema=PORTAL&item_link=lectherrendorf2005.htm

Econ 620 2007

Small differences in growth rate lead to vast differences in per capita income when
compounded over long periods of time.
Comparing real per capita GDP over the last century involves multiples as high as 20.
But comparing levels at a point in time across countries exhibit even larger (and
increasing) multiples.

GDP per capita/ CAN GDP per capita

1.5
1
0.5

ZA
R
BF
A
M
D
G
G
M
B
ZM
B
LS
O
C
H
N
PH
L
ID
N
G
AB
TH
A
C
R
I
JO
R
G
R
C
KO
R
N
ZL
G
BR
AU
S
U
SA

Zaire
Canada

1950
1,026
19,473

2004
531
35,281

19

66

Growth
-1.83%
1.60%

Another way to think about it is that an average citizen of Zaire has to live for a
whole year with the income that an average Canadian worker earns in less than 6 days
(we always work with real values, so everything is adjusted for price differences across
countries). These stunning differences in income per capita are the result of wide
differences in the growth rates (data between 1960 and 1997).
Average per capita GDP growth 1960-1997
8%
6%
4%
2%

-4%

ID
N
JP
N
KO
R

ZA
R

-2%

M
L
N I
ER
ZW
E
PE
R
C
IV
ET
H
AR
G
D
ZA
N
AM
C
H
L
C
O
L
BG
D
BE
L
EG
Y
SY
R

0%

Econ 620 2007

Average per capita GDP growth 1960-2000

35
30
25
20

Zambia
Nicaragua
Mozambique

Bolivia
Venezuela
Nigeria

Australia
Chile
Canada
US
UK

Maurtitius
France
Israel
India
Brazil

Spain
China
Greece
Indonesia

15

Malaysia
Thailand
Japan
Hong Kong
Singapore
S. Korea

10
5
0
<0%

0%-1%

1%-2%

2%-3%

3%-4%

>4%

2.1.1 The importance of growth.

a) Growth vs. business cycles.
During the 1990-92 recession, Canadian per capita RGDP fell by 6%, over a
period of 3 years. This is equivalent to a permanent reduction2 in the growth rate of
one twentieth of one percentage point (i.e. 0.06 percentage points). RGDP at the lowest
point in the recession was higher than anytime in history before the previous 5 years. If
this is the correct way of measuring well-being it seems that the welfare consequences of
growth are substantially more important than the effects of the business cycle.
Lucas (1987) formalized this point in the following way. The present value of
intertemporal utility depends on the evolution of consumption, which in turn depends on
the initial level of consumption, its growth rate and its variance.
1. How much is an agent willing to pay in terms of current consumption to increase
growth permanently by 1%?
C1
U = E
dt ;
0 1

= 2.5

Assume that 0 is the historical average of consumption growth recovered from the data,
and 2 its variance

By permanent I mean for 100 years.

Econ 620 2007

where
U C0 , , 2 = U C0 , 0 , 2
= 0 + 1%

= .7 , so the agent is willing to reduce current consumption by 30%

2. How much is an agent willing to pay in terms of current consumption to get rid of all
the business fluctuations?
U [ C0 , 0 , 0] = U C0 , 0 , 2

= .996 , so the agent is willing to reduce current consumption by less than 0.05%
These results are difficult to believe. They are driven partly from the assumption
about preferences (separable across time and households), the use of per capita data that
hides individual variation But surprisingly the results do not change that much when
you allow for financial market imperfections, absence of unemployment benefits
b) Other measures of welfare.

Income per capita is positively correlated with life expectancy, the inverse of
child mortality, access to drinking/potable water, literacy and other welfare
indicators (the direction of causality is unclear).
% of population with access to
sanitation facilities vs. GDP per
capita

% of population immunized for

measles vs. GDP per capita
100

120

90
80

100

70

80

60
50

60

40

40

30
20

20

10

0
100

1000

10000

100

100000

% of population with improved water

access vs. GDP per capita

10000

100000

Years of life expectancy vs. GDP per

capita

120

90
80

100

70

80

60
50

60

40

40

30
20

20

10

0
100

1000

0
1000

10000

100000

100

1000

10000

100000

(From left to right). Sub-Saharan Africa, East Asia and Pacific, Middle East and North Africa, Latin
America and Caribbean, High Income.

Econ 620 2007

c) Growth, Inequality and Poverty.

A rising tide lifts all the boats. There is a growing agreement that it is difficult
to eradicate poverty by redistributing existing wealth, therefore creation of new
wealth seems to be the most promising avenue to improve the standard of living of the
poor. Redistribution does not only affect the allocation of the pie, but also its size
sometimes in good ways, access to opportunities, but other times in harmful ways
distorting incentives.
What are the effects of growth on the world poorest? What is the evolution of the world
income distribution?
Lets use the World Bank definition of the poverty line, $1 a day 1985 prices, although
we can think that its level is arbitrary, most people still think that its evolution is
informative. Lets see what has happened over the last 40 years.
* The world income distribution has shifted to the right (cumulative growth). World
average per capita income increased from $4,250 to $7,200.

* Across country inequality has increased very little (China and India account for 1/3
of the world population), with a considerable increase in the density of relatively rich
countries, while many countries still remain quite poor3. Within country inequality
has increased (China, US).

It is natural to look at the log of variables that grow over time. Since growth rates are approximately
proportional, i.e. leading to exponential growth, when we take the log of income the resulting variable is
approximately linear.

Econ 620 2007

Source: Acemoglu (2007)

* Between 1970 and 1998 the percentage of people below the poverty line moved
from 17% to 7%, from 575 million to 375 million4, even when the world population
increased by 2.5 billions.

P o v e r ty R a t e s ( $ 5 7 0 )
60%

50%

40%

30%

20%

10%

0%
1970

1975
A f r ic a

L a t in A m e r ic a

1980
Ea s t A s ia

1985
S o u t h A s ia

1990
M id d e l Ea s t a n d N A

1995

2000

Ea s te r n Eu r o p e a n d C A

Sala-i-Martin data recovers each countrys income distribution by combining per capita GDP (to match
the mean) with income shares per quintiles extracted from household surveys (to match the dispersion).
The World Bank figures are substantially higher but they exhibit a similar tendency (see More or Less
equal)

Econ 620 2007

P o v e r ty C o u n ts ($ 5 7 0 )
4 0 0 ,0 0 0
3 5 0 ,0 0 0
3 0 0 ,0 0 0
2 5 0 ,0 0 0
2 0 0 ,0 0 0
1 5 0 ,0 0 0
1 0 0 ,0 0 0
5 0 ,0 0 0
0
1970

1975
A f r ic a

L a t in A m e r ic a

1980
E a s t A s ia

1985
S o u th A s ia

1990
M id d e l E a s t a n d N A

1995

2000

Ea s te r n Eu r o p e a n d C A

Source: Sala-i-Martin (2002) (counts are in 1,000s) (is the equivalent to the 1985 prices $1 poverty line)

In terms of poverty reduction the data suggests that the positive effect of growth
more than overcomes the negative effect of increasing inequality. The largest
reductions in poverty took place in countries that experienced high growth (Asia) while
the largest increases in poverty took place in low growth areas (Africa).

Source: Acemoglu (2007)5

2.1.2. Kaldors Stylized Facts

Kaldor (1958) summarized the following stylized facts about the behavior of
industrialized economies6.
1. Per capita output grows at a more or less constant rate over fairly long periods
of time.
5

Notice that an increase in 1 unit in log GDP is equivalent to an increase in GDP by a factor of 2.71. Since
there is a difference in log GDP between the Western Offshoots and Africa of 3, then the Western
Offshoots GDP is around 20 times ( e3 ) higher than the African one.
6
Notation: is aggregate profits, W is aggregate wages, r is the return to capital, w is the wage rate, Y is
aggregate income (gross national product), K is aggregate capital stock (machines, structures,
computers), L is the labor force/population. The sub index i representents per capita (per worker)
quantities. Notice that the ratio of two aggregates is equivalent to the ratio of two per capita variables,
X X L Xi
=
=
Z
Z L Zi

Econ 620 2007

2. The capital-output ratio shows no trend.
3. The return to physical capital shows no trend.
4. A wide variety of growth rates (and levels) of income per capita. (This one was
added afterwards, but we have already seen the cross country evidence)
Canadian K/Y

Canadian Output Growth

20.00
15.00
10.00
5.00
0.00
-5.00
-10.00
-15.00
-20.00
-25.00

3
2.5

19
31
19
38
19
45
19
52
19
59
19
66
19
73
19
80
19
87
19
94
20
01

2
1.5

1986

1982

1978

1974

1970

1966

1962

1958

1954

1950

Canadian Real Interest rate

2000

1996

1992

1988

1984

1980

1976

1972

1968

-6

1964

-1

1960

-11

K
Combining (2) i
Yi
constant.

K i = Yi > 0 and more or less

= 0 , with (3), Yi > 0 , we reach

This implies that the net investment rate (savings rate in a closed economy) is also
trendless.

I
I K
I
K i i
K i i i
i is more or less constant
Ki
Yi Yi
Yi
It also implies that the capital labor ratio grows at a positive rate, more or les constant.

K
Yi = K i > 0
L

Econ 620 2007

(3) + (4) imply that the factor income shares exhibit no trend.
Y

is also constant.
K Y K
Y
W
, the labor income share is also constant.
Furthermore, since Y W + therefore
Y
r

This last fact implies that the wage grows at a positive and stable rate.
^

W WY
Y
w
=

w = 0 + > 0
L Y L
L

2.1.3. Modeling economic growth

-

Demand Side Preferences

Supply Side Production Technology
Time dimension

In general most of the models in the growth literature share the same underlying
structure. Households in the demand side of the economy decide how much of their
income they are willing to consume (or how many hours to work, or how many children
to raise) to satisfy their current needs and how much are they willing to save/invest to
increase their future production possibilities.
This saving-consumption decision cannot be understood without the introduction
of the supply side of the model, where the amount saved is transformed into future
output. The interaction between the rate of transformation of saving into future output
(production function) and willingness of household to sacrifice current consumption for
future one (preferences) determines the optimal decision rule to allocate current output
into its alternative uses, consumption and saving7.
Finally it is important to notice the dynamic nature of the growth problem. Our
current choices will affect our future production and consumption possibilities.
Through our models we are trying to capture some generalities that are common
across economies. Each economy or each period in time has a set of particularities that

All the models we are going to work with in the growth part are supply-side models, where Says
law fully applies. In these models supply of loanable funds, saving, creates its own demand, investment.
Supply of labor, i.e. population growth, creates its own demand, determining the level of employment. In
any of these models there are no idle resources, so we can think of all these theories as theories for the
evolution of potential output (at any point in time the business cycle will account for the deviation between
potential and actual output). Is this assumption realistic in general? What about for our particular purpose?
The 10-fold difference in output per capita we experienced in the last century is a supply or a demand
phenomenon? The difference between an average citizen of Zaire and the average Canadian one resides in
their willingness to buy or in their ability to produce?

Econ 620 2007

need to be taken into account for a complete understanding of any issue. This is a point
we should always keep in mind.

10

Econ 620 2007

2.2. The Neoclassical Growth Model

2.2.1 The Solow model without technological change.
2.2.1.1. Description of the model
The demand side is summarized by an exogenously given saving rate, s. Each
period a fixed proportion of output is saved/invested.
In the supply side we need to specify,
a) The production technology.
b) The laws of motion of the inputs.

- The production function.

Our model economy produces only one homogeneous commodity, output (Y ( t ) ) .

There is a supply of homogeneous labor ( L ( t ) ) which is used, together with the stock of

capital ( K ( t ) ) according to the following neoclassical technology, that for simplicity we

assume to be of the Cobb-Douglas form 8
Y = F [ K , L ] = K L1

1. F(.) exhibits positive and diminishing marginal products,

F
2 F
> 0, 2 < 0
K
K
So holding L constant, Y is increasing in K but at a decreasing rate. The same applies for
labor.
2. Inada conditions,

F
F
= 0; lim
=
K K
K 0 K
lim

An equivalent condition applies for labor.

8

I have in mind a broad definition of capital that includes tools, machinery, buildings, public and private
infrastructure one of the key features of capital is its reproducible nature, capital as opposed to most
natural resources could be produced. To think about economic growth before the industrial revolution will
require the introduction of land as an additional input, for modern capitalist economies this distinction
seems unnecessary. For instance, in Britain, where long time series are available, the value of non-urban
land as a percentage of total wealth has decreased from 65% at the beginning of the 18th century to 2%
nowadays. For the U.S. in 1995 the share of oil purchases over GDP was 3.5% according to the Energy
Information Administration, an agency of the U.S. Government.

11

Econ 620 2007

3. Constant returns to scale in capital and labor. We want a technology that if we have
twice as much labor and capital we produce twice as much output. Mathematically the
production function is homogeneous of degree 1 if,
F [ cK , cL ] = cF [ K , L ]

Furthermore, for functions homogenous of degree 1 Eulers theorem implies that

FK K + FL L = Y . As a result if markets are competitive, since factors are paid their
marginal products, output is exhausted by the payments to the factors of production.
Since the stylized facts we are interested in explaining involve per capita output it
will be convenient to re-write the production function in per capita terms,

Y=K L

K
then
= L = LK i
Yi = K i
L

Where the subscript i denotes per capita variables.

- Evolution of the inputs.
Labor/Population is assumed to evolve exogenously at the exponential rate, n.
L = n, given L ( 0 ) = L0
L ( t ) = L0 e nt
Final output can be consumed or invested (saved). The fraction of output devoted to
investment, s, is exogenous and constant, and the existing capital stock depreciates at a
rate , so

K& = I K = sY K

K (0) = K 0 > 0

Since we are interested in the evolution of income per capita, it is convenient to rewrite
this expression in per capita terms,
Ki =

& LK
&
K
KL
K&
sY K

K& i =
= nK i =
nK i = sYi ( + n ) K i
2
L
L
L
L

which is the law of motion of per capita capital.

2.2.1.2. Steady state of the model

In the first chapter we defined a steady state as a situation where the variables of
our model where not changing. In the case of Newtons law of cooling a steady state

12

Econ 620 2007

happens when the temperature of the object is equal to the temperature of its surrounding
environment; in the case of the dynamics of the market a steady state is a situation where
the price is such that desired supply is equal to desired demand; in the case of rabbits and
mice a steady state is a situation where both populations are constant. In the Solow model
a steady state is a situation where income per capita (and capital per capita) remains
constant.
Combining the law of motion of capital with the production function in per capita
terms we obtain the basic equation of the Solow model,
K& i = sK i ( + n ) K i

Notice that we have a single differential equation on one variable, K i . It says that
the change in capital per worker is the difference between actual investment and
replacement investment. With the labor force growing and depreciation taking place
some investment is needed, to provide capital to the new workers that join the labor force
and to replace the machines that break down, to prevent capital per capita to fall.

When actual investment is larger than replacement investment, capital per capita
is growing, when actual investment is less than replacement investment, capital per capita
is falling. In this context a steady state is a situation in which that capital per capita is
constant, so the economy reaches a steady state when K& i = 0 , and therefore the steady

state level of capital, denoted by *, must satisfy s ( K i* ) = ( + n ) K i* , so9

In any model we have: Endogenous variables which evolution is determined within the model, like
capital or output in this model; Exogenous variables which change over time but according to some law

13

Econ 620 2007

1 1

s
K =

n +
*
i

s
Yi * = ( K i* ) =

n +

Higher savings rate or lower population growth rate are associated with
higher income per capita. The steady state of this model fits several of our stylized
facts, differences in parameters across countries could potentially explain the crosscountry variation in income per capita, the capital-income ratio and the return to physical
capital (the real interest rate) are both constant, but what about the growth rate of
output per capita in steady state?

2.2.2. Technological change in the Solow Model

In order to generate growth in output per capita in our model we introduce laboraugmenting technological progress, A ( t ) , modifying our production technology as
follows,

Y = K ( AL )

where AL is often known as effective labor. This formulation implies that as the available
technology improves, A ( t ) increases, each worker can produce more output with the
same level of capital10. Notice that the production function still exhibits constant returns
to scale in capital and labor, the private inputs, but increasing returns to scale in the three
factors.
As before it will be convenient to re-write the production function in per capita
terms,

determined outside the model, like population in this model; and Parameters which remain fixed by
assumption, n, in this model. Solving any model means describing the behavior of the endogenous
variables in terms of the exogenous variables and the parameters.
10
This type of technological change is known as labor augmenting or Harrod-neutral. Alternatively we
could use capital augmenting technological change Y = F ( AK , L ) or Hicks-neutral technological
change, Y = AF ( K , L ) . Under our Cobb-Douglas assumption the three specifications are consistent with
the existence of a steady state. But notice that intuitively if we want an equilibrium where capital and
output grow at similar rates (stylized fact 2) and output grows faster than labor (stylized fact 1), then it
seems natural to require that technological change takes the form of increases the amount of effective
labor.

14

Econ 620 2007

K
K ( AL ) = A1 L = LK i A1
L
then
Y = LK i A1
Yi = K i A1

as before the subscript i denotes per capita variables.

Capital and labor follow the same laws of motion than before, and technology is
assumed to grow at the exogenous rate, g

A = g ,

given A ( 0 ) = A0
A ( t ) = A0 e gt

2.2.2.1 Steady State of the model

Using the production function in per capita terms, Yi = K i A1 , we can logdifferentiate it reaching,
Yi = K i + (1 ) A = K i + (1 ) g

(1)

the capital accumulation equation,

Y
K& i = sYi ( + n ) K i
K i = s i ( + n )
Ki
implies that the growth rate of capital is constant if and only if the output-capital ratio is a
Y
constant, which is consistent with stylized fact #3. Since we need i to be a constant,
Ki
capital and output both need to grow at the same rate, combining this finding with (1) we
obtain the common growth rate of capital, output and technology,
Yi = K i = A = g

This situation is the equivalent to the steady state in a growing economy and is
usually called a balanced growth path or stable growth path.
Given that along a stable growth path our endogenous variables, Yi and K i , will
be growing at the rate of technological change, it will prove useful for our analysis to
rewrite the basic equations in terms of variables that remain constant along the stable
growth path, i.e. in terms of variables that achieve a steady state. Defining the following
Y
K
variables in units of effective labor y = i , k = i , we can rewrite the production
A
A
function and the law of motion of capital as,

15

Econ 620 2007

Yi = K i A1
y = k
sY ( + n ) K i
K& A K A& K&
K
K
k& = i 2 i = i g i = i
g i = sy ( + n + g ) k
A
A
A
A
A

These adjusted variables are not of interest in their own right, but rather are a
convenient transformation for our analysis11. The variables of interest, per capita and
aggregate, can be recovered easily since the level of labor and technology are determined
(and known) at any point in time outside of the model. You should become familiar with
this type of transformation and you should figure out how to do it in different models.
1
Given our production technology, Y = K ( AL ) , output can grow for 3 reasons: more
capital, more labor, more technology. Since the evolutions of labor and technology are
determined outside of the model, it is convenient to define new variables that are adjusted
for these exogenous changes. In a sense we are reducing the dimensions of our problem,
abstracting from the changes on those variables that we know in advance, and
concentrating on the endogenous variable.
Combining the previous results we reach the basic equation of the Solow model
with technological change,
k& = sk ( + n + g ) k

The intuition is the same than in the previous model. When actual investment is larger
than replacement investment, capital per unit of effective labor is growing eventually

11

Technically we have transformed a differential equation that depended on time

1
explicitly K& i = sK i ( A0 e gt ) ( + n ) K i , into an equivalent equation that is autonomous. This allows us
to use all the techniques introduced in the math review.

16

Econ 620 2007

reaching its steady state level, where by definition k& = 0 . As before we can calculate the
steady state levels of output and capital
1 1

s
k =

n + + g
*

s
y =

n + + g
*

The steady state growth rate of Yi follows from12 y* =

Yi

Yi* = Ay *
Yi * = g > 0
A

Y * = LYi *
Y * = n + g
L
The steady state of this model fits qualitatively all of our stylized facts,
differences in the rate of population growth and saving across countries could potentially
explain the cross-country variation in income per capita, the capital-income ratio and the
return to physical capital (the real interest rate) are both constant, and the growth rate of
output per capita is positive, constant and equal to the rate of technological change.
Now we turn to analyze what happens when our economy begins out of steady
state.

The steady state growth rate of Y follows from Yi =

2.2.2.2 Out of the Steady State behavior: Transitional dynamics.

From the basic equation,
k& = sk ( + n + g ) k

Since the rate of growth of output is proportional to the rate of growth of capital nothing
is lost by focusing in the behavior of the latter. Dividing both sides by capital per unit of
effective labor we reach,
k = sk 1 ( + n + g )

For low levels of k, its marginal product is very high (recall the Inada conditions) and
therefore actual investment is larger than replacement investment so capital is growing.
On the other hand for large levels of k, the marginal product of capital is very low and
actual investment is not enough to compensate for depreciation, population growth and
technological change, and therefore capital per unit of effective labor falls.

12

Notice that outside the steady state y& 0 .

17

Econ 620 2007

Once we have checked qualitatively the stability, we can turn to a more formal
analytical evaluation. Linearizing k& around k * , we reach

( )

k& s k *

( g + n + ) k k * k k *
k ( t ) = k * + e t k ( 0 ) k *

If the eigenvalue, , is negative, the steady state is stable.

= s ( k * )

( g + n + )

From our steady state relation we know

( ) = (g + n + )k

s k*

( )

s k*

( )

*
= ( g + n + )
s k *

= (g + n + )

And then we can rewrite the eigenvalue as = (1 )( g + n + ) < 0 , and therefore the
steady state is stable, as we have already seen in the phase diagram.

What are the effects of an increase in the saving rate?

18

Econ 620 2007

Since the initial level of capital (associated with the initial saving rate) is below
the steady state level of capital associated with the new/higher level of savings, after the
increase in saving actual investment is above replacement investment and capital per unit
of effective labor begins to increase. As capital increases, its marginal product decreases
and each additional unit of capital is translated into smaller increases in output and
saving. Eventually the new steady state is reached where the output produced by the last
unit of capital provides just enough investment to cover depreciation, technological
change and population growth. What happens with consumption?
An increase in saving only has temporary growth effects, with the economy
returning after a while to the original balanced growth path, along which output grows at
the same exogenous rate of technological change, g. Nonetheless the increase in saving
has permanent level effect, with output being permanently above the level that would
have prevailed at the lower saving rate.

From the previous three graphs some interesting features of the model emerge,
1. No matter where the economy begins eventually achieves a steady state.
2. The further an economy is below its steady state level of capital per unit of
effective labor the faster is its rate of growth. The further is an economy above
its steady state level of capital per unit of effective labor the faster its capital and

19

Econ 620 2007

output decrease. The key element behind this phenomenon, known as
convergence, is the role played by diminishing returns to capital. If capital is
low its marginal product is very high and an additional unit of capital is very
productive leading to a substantial increase in output, i.e. to high growth. Think of
two identical economies one starting from k H ( > k * ) and the other initially in

k L ( < k * ) . The former will decrease net investment and output, the latter will

grow, and both of them will eventually converge to k * .

3. Policy changes have no long-run growth effects, only level effects. And
therefore the only source of long term growth is the productivity of labor, which
is determined outside of the model. The Solow model highlights two sources of
growth; steady state growth (permanent) driven by exogenous technological
change, and transitional dynamics growth (transitory) driven by endogenous
capital accumulation.
So far we have assumed that the saving rate is an exogenous constant, in the next
section we are going to allow households to choose how much to save and how much to
consume every period. We have seen that the saving rate, together with the rate of
population growth, are the two key variables identified by the model as the main sources
of variation in income per capita across space and time. Our goal is to evaluate whether
the basic predictions of the model are robust in this more realistic environment13

2.2.3. Optimal Saving in the Solow Model: The Ramsey-Cass-Koopmans

model.
There are two alternative approaches to present this model. First, a fully
decentralized approach. Households own all the inputs of production, they sell labor
and capital to the firms and their goal is to maximize their lifetime utility subject to the
resources they own. In making their choices they are competitive in the sense that they
take the relevant prices, the wage the interest rate and the price of output, as given. Firms
hire labor and capital to produce output, which in turn sell to the households and other
firms. Firms are also competitive. Since only relative prices matter we can express the
other two prices in terms of the price of output, these prices are determined at the
aggregate level, the real wage is determined by the interaction between the overall supply
13

A major objection raised against the Solow model is the lack of connection between technological
change and investment. In reality, investment is the most common vehicle for introducing new techniques
into the process. Solow (1961) in his vintage capital model modified his framework to allow for
embodied technological change, i.e. in order to incorporate technological innovations into the production
process you need to replace your capital since technology is embodied in new capital. His results were
surprisingly identical to what we have discussed in class. Since technology is embodied in machinery the
rate of growth of the economy is determined by the average age of the production facilities, which
happens to be constant in steady state. In the case of an increase in saving, this leads to a decrease in the
average age of the capital stock. In the vintage model this modernization of the capital stock raises the
average level of productivity and the rate of technological change as the economy is moving to the new
steady state, once it reaches this point, the growth rate of output and technology remain constant, as in the
basic model.

20

Econ 620 2007

of labor provided by the households and the aggregate labor demand of all the firms in
our economy, something similar occurs with the real interest rate. A second approach
assumes that economic choices are made by an omniscient social planner, that knowing
preferences and technologies does not need markets or prices to allocate resources. Under
certain conditions both approaches yield identical results, and therefore when possible we
will follow the central planner approach which is simpler. Nonetheless, in the next
section we will solve the decentralized problem for expositional purposes.

2.2.3.1 Description of the model.

Households
Consider an economy populated by L identical households, indexed by i. Since
our households are identical, we can represent their choices as if they were made by a
single household, a representative household14. Population grows at the exogenous rate
n so

L = n, L ( 0 ) = L0
L ( t ) = L0e nt
Lets define C ( t ) as the aggregate consumption at time t of our economy, then

C (t )
is consumption per capita.
L (t )

Ci ( t )

Instantaneous preferences of one of these (identical) agents are well represented

by the Constant Relative Risk Aversion (CRRA) utility function,

u( Ci ( t ) )

(C (t))
=

uc > 0, ucc < 0

>0

lim uc = 0, lim uc =
c

c 0

where , the coefficient of relative risk aversion or the inverse of the intertemporal
elasticity of substitution, is a measure of the degree of curvature of the utility function15.

14

Our economy trivially admits a representative agent. In general, if we consider an economy populated by
a finite number of (non-identical) households and preferences of each one of them can be represented by an

indirect utility function of the form, vi p, y i = ai ( p ) + b ( p ) yi , where i is household specific, then these
preferences can be aggregated and represented by those of a representative household with indirect utility
v ( p, y ) =

a ( p ) + b ( p ) y . This is just a re-statement of Gormans Aggregation Theorem. The most

i

common class of preferences used in macroeconomics, the Constant Elasticity of Substitution (CES),
admits a representative agent.
15
Risk aversion measures the willingness to smooth consumption across states of nature, and intertemporal
substitution measures the willingness to smooth consumption across time, they are conceptually different.
Nonetheless under our preference specification both are governed by the same parameter.

21

Econ 620 2007

Each household is assumed to live forever and maximizes the present value of
intertemporal utility, , given by16

= u ( Ci ( t ) )e
0

dt =

(C (t ))

(1 ) g > 0 17

e t dt

(1)

where is a weighted average of present and future utility, with > 0 the rate of
time preference being the weight (measure of impatience).
A natural question to ask is whether our infinite life assumption is relevant for
the results we will derive given that most individuals we know eventually die. There are
two lines of justification for this assumption. First, most people know they are finitelylived although they are not aware of when they will die. A potential way to model the
uncertainty surrounding such a process is to introduce a constant probability of death at
each point in time in (1). The perpetual youth model18 exactly does that and the
behavior of households in that model is identical to the behavior captured by (1). A
second justification for our assumption is based on intergenerational altruism. We can
imagine that each individual not only derives utility from his own consumption but also
from the bequest he leaves to his offspring. The preferences representing households that
have a finite life-span but leave bequests to their offspring are also well approximated
by (1).
Per capita assets of household i , ai ( t ) , are in the form of ownership on capital or
as loans to other households, which we measure in units of the consumption good
(normalized by price of consumption). Since both assets are perfect substitutes, they
16

In Romers footnote 1 in page 48, he explains how to move from his notation to the one used in these
notes. We can rewrite our discount factor as notes = romer n . In footnote 22 I rewrite our system of
differential equations using this fact and I reach equations (2.23) and (2.24) from the book. So both
methods yield identical implications for the behavior of the economy.
17
Since along the stable growth path per capita variables will be growing at the exogenous rate of

technological change, g , then

(C e )

gt

dt =

( C0 )1 e g (1 )t e t dt = ( C0 )1
1

( g (1 ) ) t dt .

For the integral to be bounded (1 ) g > 0 is required.

18

This is easily seen in discrete time. Lets take the discrete time counterpart of (1), =

u (C )
t

(1),

t =0

where is the rate of time preference, i.e. the discrete time equivalent of . Lets assume the probability
of death each period is given by . Without loss of generality lets normalize the utility of death u ( d ) = 0 .
Then mortal = u ( C0 ) + (1 ) u ( C1 ) + u ( d ) + (1 ) (1 ) u ( C2 ) + u ( d ) + ... and collecting

terms we can rewrite it as mortal =

( (1 ) ) u ( C ) , which differs from (1) only in the rate of time

t

t =0

preference.

22

Econ 620 2007

command the same rate of return, r ( t ) . Households are competitive in the sense that they
take the return on assets and the wage rate, W(t), as given. Since we ignore the laborleisure choice we can just normalize each household labor supply to 1. Finally lets
assume that the government taxes any income at a constant rate, , and makes per capita
lump-sum transfers, Ti . This is the conventional, though unrealistic, way of introducing
taxes, this will allow us to focus on the substitution effects associated with the tax, while
the income effect will be absent (since the transfer will offset the tax in terms of income).
Under these assumptions, at any point in time household is budget constraint is
given by19
a&i = (1 )(W + rai ) Ci nai + Ti

Setting up the Hamiltonian, where i ( t ) is the co-state variable associated with

wealth of household i , i.e. its shadow price,

(C (t ))
Ha =

e t + i e t ( (1 )(W + rai ) Ci nai + Ti a&i )

With the following first order conditions (FOC)

uci Ci = i

(1 ) r n =

(1)

&i
i

(1 )(W + rai ) Ci nai + Ti = a&i

(2)

(3)

along with the transversality condition (TVC),

lim i ai e t = 0
t

(4)

The first FOC equates the marginal utility of consumption to the shadow value of
wealth. Since output can be either consumed or accumulated as wealth an optimal
allocation requires an additional unit of income to be equally valued in its two alternative
uses. The second FOC is the intertemporal allocation condition equating through time
the after tax return to wealth, adjusted for population growth, and the return to
consumption (a combination of impatience and the change in marginal utility as
19

In per-capita terms a fixed level of wealth now needs to be redistributed among (1+n) members of the
household the next period.

23

Econ 620 2007

consumption changes,

&i ucc
c& ) . To interpret this second condition lets focus first in
=
i uc

the case where utility is linear. Under linear utility, marginal utility is constant and
therefore from (2), is constant. In this case condition (3) states that along an optimal
solution the after tax return to capital should be equated to the level of impatience. In a
sense the amount of extra per capita consumption you can achieve by investing
today ( (1 ) r n ) is equated to the cost associated with delaying consumption from
today to tomorrow ( ) . Now when preferences are concave, there is an additional cost

since increasing consumption does not increase utility proportionally, but rather at a
decreasing rate (diminishing marginal utility). This last effect is captured by the term
&i ucc
c& , that measures the curvature of our utility function. In practice increasing the
=
i uc
degree of concavity of the utility function has the same effect than increasing the
impatience of our representative agent20. The third is just a restatement of the constraint,
while the forth states that at the end of the planning horizon either wealth should not be
valued or should be zero.

What are we looking for? The two endogenous variables are Ci and ai , therefore
the solution of the household problem amounts to find a system of two differential
equations on Ci and ai . We already have one of them, (3). To obtain the other lets log
differentiate (1) with respect to time, and replace the result in (2) as follows
1
in (2)
C i = i
r n = + C i
C i = ( (1 ) r n )
(5)

(5) is known as the Euler equation and describes the optimal behavior of the
growth rate of consumption. If the return to saving (the after tax interest rate adjusted for
population growth) is larger than the rate of impatience, it is worth to forgo current
consumption, increasing current saving and accumulating wealth, in order to increase
future consumption. As a consequence consumption will grow over time.
Given the resource constraint facing each agent, consumption growing is
equivalent to low initial consumption. Think of 3 agents that only differ on their degree
of impatience, facing the same constraints and owning the same resources, their
respective consumption paths will look like,

There is conceptual difference between and . The first captures the fact that the more we consume
the lower is the additional satisfaction we obtained from additional consumption (diminishing marginal
utility) while the second captures that todays consumption is preferred to tomorrows consumption. In our
simple model the only choice is between current and future consumption of a single composite commodity
(instead than between different consumption goods in a given period and across time) and therefore both
parameters lead to qualitatively similar effects.

20

24

Econ 620 2007

Firms
There are J identical21 firms indexed by j. Any of those firms hires inputs, labor
and capital, combining them with the freely available level of technology to produce final
output, according to the familiar Cobb-Douglas specification

Yj = ( K j )

( AL )

(6)

Firms are assumed to maximize profits. The representative firms profit at any
point in time are given by, where everything is normalized by the price of final output
(real wage, real interest rate) and capital is assumed to depreciate at a rate, .
j = (K j )

( AL )

( r + ) K j WL j

The j -th firm hires labor and capital up to the point where their marginal products are
equated to the given market prices,
1
1
F
FK j = r
( K j ) ( AL j ) = r
K j

F
FL j = (1 ) ( K j ) ( AL j ) A = W
L j

Notice that since W , r , , A are given, all firms choose the same capital-labor ratio,
i.e. the same degree of capital intensity (this will be true if we had firms of different sizes
21

Since all the firms are identical we can represent the aggregate choices of the supply side of our economy
with those of a representative firm. In general, in the absence of production externalities and under
competitive markets, any supply side configuration can be represented by a representative firm.

25

Econ 620 2007

as long as they face the same prices and have access to the same technology, is just a
consequence of the constant returns to scale imposed on the production structure).
If we take the ratio of both first order conditions,

(K j )

( AL )
(1 ) ( K ) ( AL )
1

Lj r +
=
(1 ) K j W

we get the familiar condition equating the marginal rate of technical substitution (the rate
at which the firm needs to substitute labor for capital in order to maintain a constant level
of output, the slope of the iso-quant) equated to the relative price of the factors of
production the rate at which the market is willing to substitute labor for capital (the slope
of the iso-cost lines).

2.2.3.2. Macroeconomic Equilibrium

We began with households choosing consumption taking their wage and return on
wealth as given, we continued with firms maximizing profits for given input prices. We
can now combine the aggregate behavior of firms (demand of inputs / supply of goods)
and the aggregate behavior of households (supply of inputs / demand for goods), to
determine those equilibrium prices.
Aggregating across firms, K j = JK j =K ,
j

L
j

= JL j = L and22 Y j = JY j = Y ,
j

the equilibrium prices are determined as the marginal product of the aggregate capital
stock and the marginal product of the aggregate labor force,
r + = FK = ( K )

( AL )

W = FL = (1 )( K ) ( AL ) A
1

In the previous section, we found that each firm hires capital up to the point where
its marginal product is equal to the given market price. Now, we determine this market
price as the marginal products of the aggregate stocks of labor and capital.
Finally these prices are consistent with an aggregate market clearing condition,
national income equals national product, NI = NP . Assuming that the government
rebates all the taxes it raises in the form of lump transfers23,
GNI = ( r + ) K + WL = ( K )

22

( ) ( AL j )

JY j = J K j

= JK j

( AL )

) ( AJL j )

K + (1 )( K )

= (K )

production function is homogeneous of degree 1.

G stands for gross.

23

26

( AL )1

( AL )

AL = Y + (1 ) Y = Y = GNP

= Y where we used the fact that our

Econ 620 2007

Since the representative household cant be either a net lender or a net borrower
(i.e. the average loan is 0) and since all the factors of production are ultimately owned by
K
the households, then Lai = JK j = K , so ai = = K i the representative agent budget
L
constraint becomes,
K& i = (1 )(W + rK i ) Ci nK i + Ti

& ,
Then aggregating across households, K = K i =LK i , then K& = LK& i + LK
i
i

& = WL + rK C = WL + ( r + ) K K C = Y K C
K& = L (1 )(W + rK i ) Ci nK i + Ti + LK
i
where we used again the fact that the government budget is balanced every period.
As a result the evolution of the aggregate capital stock is determined by,

1
K& = ( K ) ( AL ) K C

And the evolution of aggregate consumption is determined as,

1
1
1
1
C = C i + n = ( (1 ) r n ) + n
C =
(1 ) ( K ) ( AL ) n + n

a. The Steady State

Given our previous experience we know it is convenient to rewrite the system in
terms of variables that eventually achieve a steady state24. Defining
K
C
Y
k
,c
,y
we can rewrite our system as
AN
AN
AN

( K ) ( AL )
k = K n g =

c = C n g =

K C

(1 ) ( ( K )
(
1

n g = k 1

( AL )

c
n g
k

n + n n g

k& = k c ( + n + g ) k
c =

(1 ) k 1 n g

24

(7)

Is it clear why do we need to do this? First notice that is just a convenient transformation of the original
system, that eliminates the complications introduced by steady state growth (which is just a constant)
allowing for a more clear analysis of the transitional dynamics. Our original variables are easily recovered
multiplying the new variables by the amount of population and the level of technology that evolve
exogenously through time.

27

Econ 620 2007

which is a system of differential equations that together with the initial condition
k ( 0 ) = k0 and the TVC (which intuitively is equivalent to an initial condition for
consumption) fully describe the dynamic evolution of our economy25. Imposing the
steady state condition, k& = c& = 0 ,

(1 )
c& = 0
(1 ) k 1 n g = 0
k * =
n + + g + (1 )

c* = k * ( + n + g ) k *
k& = 0

1
1

( )

We can use the tools we developed to evaluate stability of this steady state,
c

c& 0
(1 ) ( 1) k 2
=

&
k 1
k 1 n g

c c*
0 Tr > 0

Saddle
*
k = k * k k
+ Det < 0

c = c*

With the stable (negative) eigenvalue associated with k and the unstable (positive) with c.
When k& = c& = 0 , per capita ( K i , Ci , Yi ) and aggregate variables ( K , C , Y ) are
growing along the Stable Growth Path, at the constant rates g and g + n respectively.
That implies that the capital-output ratio, the saving rate and the marginal product of
capital are constant along such a path. The real wage is growing at a rate g.
So even when saving is endogenous growth in the efficiency of labor is the only
source of persistent per capita growth (it makes sense since increases in saving only lead
to temporary increases on the growth rate and the saving rate cant grow forever, it
shouldnt be the source of permanent growth).

b. Transitional Dynamics.
Now we turn to analyze our economy outside the steady state. For that purpose
we will use the phase diagram in the c-k space.
The c& = 0 line
c
c0
c& =
(1 ) ( k 1 ) n g = 0 so
(1 ) ( k 1 ) n g = 0
0

it is a vertical line
c& c
= (1 ) ( 1) k 2 < 0 as capital increases consumption decreases.
k

25

If we set = = 0 and as we said notes = romer n . Then k& becomes k& = k c ( n + g ) k , which is the

Cobb-Douglas equivalent to (2.23). And c& becomes c& =

( ( k )
1

romer

g , which is again

equivalent to (2.23). So both approaches yield identical implications for the behavior of the economy.

28

Econ 620 2007

The k& = 0 line

dc
c = k ( + g + n ) k

k& = k c ( + g + n ) k = 0
dk

k&
= 1 < 0
c

The phase diagram

29

= k 1 ( + g + n )
k& = 0

Econ 620 2007

For any given level of k, there is a unique initial level of c that is consistent with
stability. Any of the trajectories in the phase diagram satisfies (7), but only the stable
arm, i.e. the trajectory that leads to steady state, satisfies the transversality condition.
The initial level of capital will be determined by an initial condition, k ( 0 ) = k0 ,
inherited from the past. Assume that k0 > k *

Given that at any point in time k ( t ) 0 , if the economy begins in a point above
the stable arm, for (7) to be satisfied c must be decreasing, although not fast enough to
prevent k from falling, until we cross the c& = 0 line, thereafter consumption increases as
capital continues falling. k falls first due to its low marginal product and later due to high
c. Eventually capital has to become negative for (7) to be satisfied (or c must jump to
zero, which violates (7)), and therefore that path cannot be optimal. We can rule out all
those paths.

30

Econ 620 2007

If the economy starts below the stable arm, consumption is low and to satisfy (7)
capital needs to be increasing. This path converges to the intersection of k& = 0 line and
the horizontal axis, A, that violates the TVC. Intuitively the TVC prevents the agent from
leaving anything behind when she dies, at point A consumption is very valuable (it is
very low so its marginal utility is very high) and we keep accumulating capital, this
behavior does not make sense26.
We can conduct the transitional dynamics analysis analytically, find eigenvalues,
find the arbitrary constants (the previous analysis allows us to set the constant associated
with the unstable eigenvalue equal to zero, i.e. to choose the c ( 0 ) in such a way that the
economy is in the stable arm), find measures of the speed of convergence, initial response
of consumption

c. A decrease in the level of taxes: Permanent vs. Transitory shocks.

As we already know differences in saving behavior, together with variation in the
rate of population growth, are the two potential candidates identified by the Solow model
as the source for the observed variation in growth rates, both across time and space. This
model suggests that countries with high savings and low population growth rates will
have higher income per capita, and at least temporarily, will grow faster.
Starting from steady state, if the government decreases the tax on capital income27
agents will find optimal to increase the amount of capital they hold, since owning capital
has become cheaper or equivalently the net return to capital has increased. This requires
26

Formally, we can use the transversality condition to rule out paths towards A,

(1 ) g ]t
lim i ai e t = lim Ci Ki e t = lim ce gt
ke gt e t = lim c ke[
=0
t

( ) ( )

Noticing that around A, k is almost constant (we cant say the same about c , since c is small and
decreasing and therefore its growth rate might be non-trivial). Then log differentiating, the TVC can be
expressed as,
c + 0 + (1 ) g < 0
Using our expression for the rate of growth of consumption evaluated at A, satisfying the
transversality condition is equivalent to,

(
or (1 ) ( A

)
)
) n g > 0

(1 ) A 1 n g + (1 ) g = (1 ) A 1 + n + g < 0
1

( )

Since the golden rule level of capital satisfies, k GR

1

then ( A )

< n + g

and

(1 ) ( A 1 ) n g < 0 ,

= n + g , n, g , > 0 and

A > k GR ,

which implies that the transversality

condition is not satisfied in A , and we can rule out all the trajectories leading to that point.
The last thing we need to check is that the stable arm satisfies the terminal condition,
bounded utility

c + k + (1 ) g < 0
+ (1 ) g < 0
lim c ke(1 ) g t = 0
t

27

In fact our government taxes both labor and capital income (wages and interest rate), but since the laborleisure choice is not modeled agents are not allowed to respond to changes in labor income taxation, on the
other hand since the trade-off between consumption and saving is endogenous our agents will adequate
their behavior to changes in the rate of capital taxation.

31

Econ 620 2007

an increase in the amount they save, or equivalently a decrease in the rate at which they
consume. As a consequence differences in fiscal policy that lead to differences in saving
behavior could be at the source of the observed variation in income per capita across
countries.
In this section we are going to illustrate the difference between the adjustment of
our economy to a permanent shock as opposed to a transitory shock (in both cases I
will assume that the tax increase is unanticipated, i.e. agents learn about it at the instant
that it takes place)
c&
0 , the
Lets begin with a permanent (unanticipated) decrease in taxes. Since

k&
c& = 0 line will shift after a decrease in the tax rate. Since
= 0 the k& = 0 line will be

unaffected28. A decrease in taxes increases the steady state level of capital, so the
c& = 0 shifts to the right.
When does consumption respond? Consumption responds at the time when the
new information, in this case the decrease in taxes, is released. At the time of the shock,
agents re-optimize using the new information and they find the new path for consumption
based on the new tax. This is the only behavior consistent with the satisfaction of the first
order conditions represented by (7). Think of oil prices, when do they change? And why?
The same with consumption. This implies that a jump can only take place when new
information is released, in the absence of new information agents try to smooth
consumption (i.e. to have something as close as possible to a constant marginal
utility of consumption) and they will avoid expected sudden changes in
consumption.

28

It is worth noticing that since our tax is rebated as a lump sum transfer there is no income effect
associated with a change in taxes. This income effect is at the root of the Keynesian multiplier after a tax
cut, which leads to an increase in consumption as a consequence of the increase in disposable income. In
our framework this income effect is not present and we only have a substitution effect, when taxes change
the relative price of consumption/saving changes leading to changes in the allocation of resources between
consumption and investment.

32

Econ 620 2007

At the lower level of taxes, the capital stock is below its steady state level, and
therefore its after-tax marginal product is higher than the rate of return on consumption.
Under these circumstances households find optimal to substitute current consumption for
future consumption (less current-more future) in order to accumulate the now more
profitable capital, so they increase saving. The high level of investment leads to a
transition along which capital is accumulating. This process leads to progressive
increases in output and consumption. With capital being accumulated its marginal
product has to be falling, and the incentives to postpone consumption are being reduced.
Eventually (ad infinitum) the new steady state will be reached, at that point the
after tax marginal product of capital net of depreciation is equated to + n + g , and
consumption and capital cease to increase.

A temporary decrease in taxes,

Since the agents populating our economy are forward looking they realize that the
potential gains from a temporary tax cut in terms of the additional return earned on

33

Econ 620 2007

capital will last only for a few periods and therefore they do not cut their consumption by
as much as they do when the tax cut is permanent.

When the tax rate is back in the originally higher level the economy exactly hits
the stable arm corresponding to the initial steady state, after that point capital and
consumption decrease until the new steady state is reached.
Now we can recover the time paths of any variables of interest; capital,
consumption, the interest rate, investment, the wage rate29
In this section we have extended the Solow model to a framework where savings
is the result of optimal decisions of the households. Nonetheless the implications of the
model are very similar to those of the original Solow model, since in steady state the
saving rate is constant. The economy eventually achieves a steady state, there is
convergence, parameter changes have only level, but not growth, effects and there are
two sources of growth: technological change along the balanced growth path and capital
accumulation along the transitional dynamics. As a result we will proceed our analysis
using the simpler version with exogenous savings since little is lost by doing so.

2.2.3.3 A Digression: The Central Planner Solution

The golden rule level of capital, the level of capital that maximizes the level of

steady state consumption is given by ( k GR )

= + n + g . Lets compare the level of

capital achieved by our market solution with this ideal level of capital. Abstracting from
taxes, = 0 , in steady state, ( k * )

= + n + + g . Given that (1 ) g > 0 then

+ g > g and therefore the steady state level of capital is below the golden rule one30.
Is there an alternative allocation that would make our representative agent better off, is
the market equilibrium Pareto efficient?
29

See the book for similar analysis of a change in the rate of impatience and in the level of government
purchases.
30
The steady state level of capital in this model is known as the modified golden rule level of capital.

34

Econ 620 2007

From the First Welfare Theorem we know that if markets are competitive,
complete and there are no externalities or transaction costs (and agents are finite) then the
competitive equilibrium is Pareto-efficient. In general Pareto efficiency means that there
is not other allocation that reaches a higher value of satisfying the constraint for at
least one agent, without lowering the value of for other agent (in principle the market
will allow for all those welfare improving reallocations to take place in the form of trade).
From our analysis of the Solow model, if the economy was above k GR , you can
reduce capital and increase both current and future consumption. The situation is
different when the initial level of capital is below k GR , in that case you can only increase
future consumption and the cost of current consumption. With endogenous saving, the
steady-state level of capital is reduced by a factor that reflects the fact that present
consumption is preferred to future consumption (impatience)31. Even if you begin with
k= k GR , the economy will converge to k * (because impatient agents with this high level
of capital will find optimal to consume at an unsustainable high rate decreasing capital
until the modified golden rule is reached).
An alternative way to inspect the efficiency of our solution is to solve the problem
from the perspective of a benevolent central planner. He knows preferences of the
households and the production technology available for the firms. To maximize social
welfare he allocates the resources without the need of a price system. The problem he
faces differs in the fact that he takes into account that w and r are functions of the capitallabor ratio, as opposed to the decentralized/competitive problem where households and
firms took them as given. The planner has to choose the paths for per capita consumption
and capital to solve,

max = max u ( Ci )e

max A0
0

(c)

(1 ) g t

dt = max
0

dt = max
0

( Ci )

(c)

dt = max
0

( cA e )

gt 1

e t dt =

e 't dt

Subject to
k0 given

k (t ) , c (t ) 0

And the law of motion of capital that is also a market clearing condition, that states that

all output should be either consumed or saved, k& = ( k ) c ( + n + g ) k

Setting the Hamiltonian,

31

g picks the fact that tomorrows MU of consumption will be reduced due to the growth of consumption

at the rate g.

35

Econ 620 2007

(c)
Ha =

e 't + e 't

(( k )

c ( + n + g ) k k&

The FOC are

c =

(k )

(1)
n g = '

&

(2)

along with the transversality condition,

lim ke 't = 0
t

As before, it is helpful now to think at this point on what are we looking for? The
evolution of this economy is determined by the paths of the 3 inputs (capital, technology
and population), output and consumption. Given the 3 inputs, the evolution of output is
straight-forward. Since population and technology are exogenous, the first thing we need
is the path for capital. Therefore we are looking for the paths of capital and
consumption. Knowing that, we need to think on how to transform our conditions to
reach a system on those two variables.
Log diff (1) and replacing in (2) to get rid of

(k )

c =

n g = '+ c
c =

( ( k )

1
1
( k ) n g + (1 ) g
suuuuuuuuuuuur

'

n g which are identical to the dynamic equation describing

the behavior of the competitive economy.

Since, under the assumptions need for the First Welfare Theorem to hold, it is
equivalent to solve the competitive problem or the central planners problem, we can
choose which solution method to follow.

2.2.4. Evaluation of the Neoclassical Growth Model

One of the most interesting features of the Solow model is the amount of testable
implications that it generates. In this section we are going to evaluate the implications of
the model along several dimensions.

2.2.4.1. The Solow Model and Kaldors stylized facts.

36

Econ 620 2007

1. Per capita output growth at a positive rate which shows no tendency to diminish.
Once we allow for exogenous technological change, output per capita grows
permanently at a constant and positive rate that coincides with the exogenous rate
of technological change.
2. The capital-output ratio shows no trend. In steady state output and capital grow at
the exogenous rate of technological change and therefore the capital-output ratio
is constant.
3. The return to physical capital shows no trend. The marginal product
capital, Y K , is constant in steady state.
4. A wide variety of growth rates and of income levels. The Solow model provides a
trivial, non-informative, way to account for differences in growth that relies in
differences in the non-modeled rate of technological progress across countries. A
second mechanism is related to the convergence properties of the model, with
countries away from steady state growing faster than countries close to steady
state.

When we turn to differences in the levels of output, since Yi * Ay* = A ( k * ) , the

Solow model points to differences in capital per capita as the source of cross country
variation in income per capita. Countries where workers are equipped with more
capital will have higher income per capita.

50,000

Yi

40,000
30,000
20,000
10,000
0
0

20,000

40,000

60,000

80,000

100,000 120,000

Ki

Correlation: 0.94
The obvious next step is to evaluate the determinants of capital suggested by the
1

s
model using the fact that Yi = A ( k ) = A
. According to the Solow

n+ + g
model differences in capital per capita stem from differences in savings and population
growth, and therefore the ultimate determinants of the international variety of income
levels are those two variables.
*

37

45,000
40,000
35,000
30,000
25,000
20,000
15,000
10,000
5,000
0
0.00%

45,000
40,000
35,000
30,000
Yi

Yi

Econ 620 2007

25,000
20,000
15,000
10,000
5,000

10.00
%

20.00
%

30.00
%

40.00
%

-2.0%

0
0.0%

Saving

Correlation: 0.69

2.0%

4.0%

6.0%

Correlation: -0.3

So far a relatively simple model seems to do an outstanding job at capturing key

features of growing economies.

2.2.4.2 Convergence.
In general we define convergence as the tendency for countries with lower initial
income to grow faster and catch up with higher income countries. The Solow model
predicts that the farther an economy is from its own steady state the fastest will be its rate
of growth. This is just a consequence of the key role played by capital in this model
combined with a production technology that exhibits diminishing returns to this input. As
a result in countries with low levels of capital (which also have low levels of income), an
additional unit of capital is very productive leading to a substantial increase in output, i.e.
to higher growth.
Lets turn to data on OECD (as Baumol (1986)) countries to check the validity of
this prediction.

38

Econ 620 2007

Average Growth 60-9

OECD Convergence
5.00%
4.00%
3.00%
2.00%
1.00%
0.00%
0

10

15

20

25

30

Thousands

Yi 1960

The data seems pretty consistent with the predictions of the model. Countries that
were already rich in 1950 tend to grow slower than countries that were poorer at that
date. Convergence allowed those relatively poor countries to catch up with the leading
ones.
When the same prediction is checked against a broader data-set that includes not
only high income countries but also middle and low income ones, the signs of
convergence disappear.

Average Growth 60-9

Convergence
8.00%
6.00%
4.00%
2.00%
0.00%
-2.00% 0

10

15

20

25

30

Thousands

-4.00%
Yi 1960

39

Econ 620 2007

The problem is that the sample of countries in our first exercise is not random,
only successful countries, the ones that meet the requirements to become part of the
OECD, were included and therefore is not surprising to find convergence, since those
countries are rich nowadays wherever they began in 1950. Once we use the whole
population of countries the evidence of convergence disappears.
But there is still a potential solution to this problem. If you were a zoologist
researching on the relation between age and height and you had a mixed sample of mice
(the height of an adult mouse is a few centimeters) and elephants (the height of an adult
elephant is several meters), you will end up finding no relation between height and age32.
You will have adults very tall (elephants), adults very short (mice), youngsters very short
(mice) and youngsters relatively tall (elephants). This is something similar to what might
be happening in the previous figure, we are mixing elephants and mice. What is the
solution? For the zoologist it is evident and straight forward, normalize the height of each
of the animals by the adult (steady state) height of its species, and then conduct the
analysis, you will find that there is a positive relation between height and age (at least for
a certain age interval).
Similarly, the Solow model implies that (only) if countries share the same steadystate (if they are the same species) they will convergence to the same level of income and
thereafter all of them will grow at the exogenous rate of technological change, g.
Nonetheless the model allows the steady state to vary with n and s, which exhibit
substantial variability across countries. We can then talk about conditional
convergence33, i.e. the Solow model predicts that countries will catch up only after
controlling for the determinants of the steady state. In a sense we are going to have to
normalize the income of each country by its steady state income (which is related to its
saving rate and its rate of population growth).
Formally convergence could be tested using the following equation,
ln yt ln y0 = ln y * (1 e t ) ln y0 (1 e t )

( 1) ( + n + g ) < 0

is often referred as the speed of convergence, and is a measure of the rate at

which economies move towards their steady states. Given that (1 et ) is positive (since

< 0 ), this equation implies that the growth rate between 0 and t ( ln y ( t ) ln y ( 0 ) ) is

positively correlated with the steady state level of income and negatively correlated
with the initial level of income. What is even more important we can empirically
measure it, having information on growth rates, initial and steady state levels of income34.

A first set of regressions, which are econometric equivalent to the last graph,
tested for absolute convergence, i.e. they did not control for the steady state. These
results are from Mankiw, Romer and Weil (1992).
32

For the sake of exposition, I am assuming that both species have similar life spans, even though I have
never heard of a 70 year old mouse.
33
If convergence was absolute, then independently of their steady states, lagging countries will grow faster
than leading ones.
34
See the Appendix for a formal derivation.

40

Econ 620 2007

implied
ln yt ln y0 = ln y0 (1 e t )
= 0.00017

(0.00218)

And as we expected from the graph, they found no signs of convergence.

Barro and Sala-i-Martin (1995) conduct this same type of regressions for
economic areas that most likely share the same steady states (this is an alternative way to
control for the steady state, choose a sample that most likely shares the same steady
state). Looking at Japanese prefectures they find = 0.0279 ( 0.0033) , looking at
convergence between states in the U.S. they find = 0.0174 ( 0.0026 ) , looking at
European regions they find = 0.019 ( 0.002 ) .

Annual growth rate vs. log per capita personal income for U.S. states.

A second set of regressions tested for conditional convergence, i.e. they did
control for the country specific determinants of the steady state.
y
implied
ln t = ln y * 1 e t ln y0 1 e t
= 0.013
y0

(0.004)

1
s
where y =

+n+ g
*

The results are more supportive of the model, the convergence parameter has the
right sign and it is statistically significative, although from an economic point of view is

41

Econ 620 2007

very small, i.e. convergence takes place but a very slow speed. Imagine that we have two
countries Rich and Poor that share the same steady state, and the income of Rich is twice
as much as the income of Poor. At the estimated rate of convergence it will take Poor
around 53 years to reach an income equal to 75% of the income of Rich and around 106
years to reach an income level of 87.5% of the income of Rich.
To summarize our findings, during the last 50 years we have not observed a
tendency for economies to converge, in fact the standard deviation of income per capita
between 1960 and 1997 for our sample of countries more than doubled, increasing from
5,595 to 11,584. When we look at economies that have similar underlying parameters
and therefore similar steady states, like US states, Japanese Prefectures or European
regions, we find that the initial level of income plays an important role in explaining the
subsequent growth performance; areas with low initial incomes tend to grow faster
reducing their distance to the leaders.
Once we control for the determinants of the steady state (specifically,
investment in physical and population growth) we recover some signs of convergence,
what the literature calls conditional convergence, i.e. the fact that economies that are
further from their respective steady states tend to move towards those steady states taking
larger steps.
In general the estimates of the speed of convergence are low, in the range of 1%2%, that implies that in one year the economy reduces by 1%-2% the distance to its own
steady state, and it takes between 35 and 50 years to eliminate half of the initial gap35.
Overall the reported evidence on convergence tends to lean (weakly) towards models
that stress the role diminishing returns to capital.

2.2.4.3. Case Studies.

a. The reconstruction of Europe after WWII: An economy after a large
destruction of capital.
Wars can be used as natural experiments to test the implications of economic
theory. The following example uses data from post-war Europe to evaluate the
performance of the Solow model.

39

Y 45 46
Y 45 Y 39
K

45

39

France

Germany

Italy

Austria

Holland

4,793

5,406

3,521

4,096

5,544

2,573

2,217

1,922

1,725

2,686

53%

41%

54%

42%

48%

65-80%

70%

75-80%

70%

60-73%

Table 1. European output and capital stock before and after WWII.

35

e t y y* =

( y y ) t = ln 2 t = .69 t = .69
*

42

Econ 620 2007

Casualties
Population Armed Forces Military Civilian %Milit. %Civil Tot.%

Country
France
Italy
Netherlands
Germany/Austria
U.S.
Canada
Australia

42,000
43,800
8,700
78,000
129,200
11,100
6,900

4,600
400
17,900

122
227
14
3,250

470 2.65% 1.26% 1.41%

60
0.14% 0.66%
150 3.43% 1.81% 1.88%
2,350 18.16% 3.91% 7.18%

16,354
1,100
1,340

405
39
29

2.48% 0.00% 0.31%

3.57% 0.00% 0.35%
2.19% 0.00% 0.43%

Table 2. The effects of the war in European population. Figures in thousands, population measured in 1940.

The data suggests that our group of European nations suffered an important
reduction in capital per capita as a consequence of the war. The next figure graphs the
average growth of per capita GDP in the post-war period,
^

Yi
.08
.07
.06
.05
.04
.03
.02
.01
.00
50

52

54

56

58

War economies

60

62

64

66

68

70

Control group

Figure 2. Post World War II per capita growth.

This behavior is broadly consistent with the transitional dynamics generated by

the Solow model. Low capital, relative to its steady state level, will lead to a period of
high and decreasing growth.

b. The New Industrialized Economies. An economy after a large

increase in saving.

43

Econ 620 2007

GDP per capita growth

Saving rate
0.16

68

0.14
58

0.12
48

0.10
CAN
SNG

0.08

38

HK
KOR

0.06
28

0.04
18

0.02

0.00

8
1960

1970

1980

1960

1990

1970

1980

1990

Again as the Solow model predicts, a large increase in the saving rate, like the one
experienced by some East-Asian economies in the 60s, will be followed by a period of
high (but decreasing) growth.

2.2.4.4. Quantitative Implications.

At the qualitative level the predictions of Solow model seem to be consistent
with Kaldors stylized facts, data about cross-country income differences, differences in
growth rates and some recent growth episodes. Nonetheless we are going to push the
model one step further, evaluating its ability to reproduce the quantitative features we
have just described.
In order to evaluate the quantitative implications of our model we will begin by
pinning down some of the key parameters from the data.
Parameter
36

g
s
n

Value
1/3
0.05
0.02
Country specific
Country specific

Source
Gollin (2002)
Mankiw (1995)

PWT
PWT

a. Cross-country variation of output per capita.

36

Since the Solow model assumes perfectly competitive markets the elasticity of output to capital coincides
with the capital share in national income, which can be recovered from the data.

44

Econ 620 2007

Lets take two countries with a 20-fold income gap in 1997, for instance Canada
($35,281) and Gambia ($1,633). This twenty-fold gap in income per capita implies that

1
K
Yic ( K ic ) ( Ac )
=
= ic

1
K ig
Yig ( K ) ( A )

ig
g

A
c
Ag

K
= 20
ic
assume
K ig

Ac = Ag

K
1/
= 20 ic = 20 = 8,000
K ig

Using the same data source we can calculate the ratio of the average per capita
capital for the two countries reaching 121. Since capital is a minor ingredient in the
production of output (only 1/3), explaining the large variation in income per capita on
the basis of variations in capital per capita requires a huge spread in the distribution of
this variable, which is almost two orders of magnitude higher than the observed in the
data.
Assuming the economies are in steady state, we can translate the observed
differences in income per capita into implied differences in saving rates or population
growth rates37. When we do this exercise we obtain differences in the order of 400, i.e. in
order to explain the observed difference in output per capita an average Canadian must
have a saving rate around 400 times larger than the saving rate of an average Gambian. In
our data set savings in Canada are around 24.6% less than 3 times larger than the 8.5%
savings reported for Gambia.
Another way of looking at the problem is to calculate the implied differences in
returns to capital.
rg
rc

( kg )

( kc )

k
= g
kc

( yg )
=
1
( yc )

Y A
= ig g
Yic Ac

AY
= g ic
AcYig

c
g

( 20 )
assume

A =A

= 400

The model implies huge differences in the returns to capital required to explain
the observed variability across time or across countries in income per capita. Measuring
returns to capital tends to be difficult, you can only look at companies that are quoted in
stock markets or government bonds, but in a world with some capital mobility the
suggested differences in rates of return will lead to huge capital flows from rich to poor
countries, but in 1990, the richest 20% of the world received 92% of the portfolio capital
inflows; the poorest 20% only received 0.1%. The riches 20 percent of the world
population received 79% of FDI; the poorest 20% received 0.7% of FDI. It seems that
capital is flowing in the other direction.
What can we conclude from the previous calculations? If the share of capital in
national income is a good measure of the elasticity of output to capital ( ) (i.e. if
markets are competitive and there are no externalities), then since capital is a
relatively unimportant input of production the observed variation in the level (growth
rate) of capital per capita cannot account for the observed variation in the level (growth
rate) of income per capita across countries in the world.
37

In the case of population I assumed away the effects of capital depreciation.

45

Econ 620 2007

b. Quantitative effects of a destruction of capital.

We can quantify the effects of a 50% destruction of capital in the growth rate of
an economy that is moving along a stable growth path, lets assume a saving rate of 20%
(this is roughly the average saving rate of continental European countries over the last
half century),
y = k
y = k
%k
k& = sk ( n + g + ) k
k = sk 1 ( n + g + )

Yi = Ay
Yi = g + y = g + k = g + sk 1 ( n + g + )

Yi = g + sk 1 ( n + g + )

As you can see the previous formula captures the two elements of growth in the
Solow model, steady state growth and transitional dynamic growth. Using parameter
values consistent with post-war European data38 this exercise generates a growth rate of
per capita income close to 3.5%, of which 2% corresponds to the exogenous rate of
technological change and the remaining 1.5% to the transitional dynamics generated by
the destruction of capital. This figure is only half of the growth rate observed in postwar Europe.

c. Quantitative effects of an increase in the saving rate

In order to get a sense of the quantitative relevance of changes in saving on
subsequent growth lets analyze the effects in steady state income per capita of a change
in the savings rate,

( )

y = k
*

1
1
y*
s

s
1
=

s 1 n + g +
n + g +
n + g +

It is convenient to rewrite it as an elasticity

s
y s
=

*
s y 1 n + g +
*

/
1
1/ /

1
s

y*
s
=
= 0.5
* = 0.5
*
1
n + g + y/
y
s

So changes in savings lead to relatively small changes in steady state output.

Revisiting the growth performance of East Asian Economies the Solow model predicts
1 *
k in the formula for per capita growth,
2
as the initial capital stock, and trace the evolution of the capital stock, output and their growth rates.
38

Given n, g , , and s , we can calculate k . Then we can plot

*

46

Econ 620 2007

that after doubling their saving rate, steady state income per capita will increase only by
50%, while their growth experiences are characterized by a doubling of income per capita
every 7-10 years, well above the predictions of the model39.

d. Growth Accounting
Solow (1957) provides a model free framework to empirically investigate the
contribution of different inputs to growth, this type of exercise is generally known as
growth accounting.
For simplicity assume the following production function, where technological
change is Hicks neutral,
Y = AF [ K , L ]

Taking time differentials and dividing by Y

Y & Y & Y &
A
K
L
Y K& K Y L& L (1)
K&
L&

Y=
+
+

Y = A +
+
A + K + L
Y
Y
Y
K Y K L Y L
K
L
Y = A + K K + L L
where (1) assumes that markets are competitive and therefore factors are paid their
Y K
is the capital income share, which we
marginal products. Under that assumption
K Y
Y L
denote as K and
is the labor income share, which we denote as L . Since both
L Y
shares need to add up to one, we can rewrite the previous expression as,

( )

Y = A + K K + (1 K ) L
Yi = A + K K i

This equation decomposes the increase in output per capita into two
components: increases in productivity, and increases in capital per capita, i.e. capital
deepening. Since we can observe the rate of growth of output and the rate of growth of
capital we can rearrange the previous expression to estimate the increase in productivity
as

( )

A = Yi K K i

39

Average Canadian GDP for year 2004 (base year 2002) is in the order of CAN$40,000. Canadas saving
rate is around 20%, notice that an increase in saving of two percentage points will bring a (permanent)
increase in average income per capita of CAN$ 2,000. Since this difference grows forever at the exogenous
rate of technological change, it seems ridiculous to regard the effects of such a policy as unimportant.

47

Econ 620 2007

A is the rate of growth of technology, the rate of growth of Total Factor
Productivity or the Solow residual. It captures everything that contributes to growth in
income per capita that is not capital or labor, and to some extent is a measure of our
ignorance.
Solow (1957) took time series of output per unit of labor, capital per unit of labor
and the capital share for the U.S. from 1909 to 1949, he found that around 1/8 of the
increase in output per capita was a consequence of capital deepening. Using US data
from 1950 to 2000, and assuming a constant capital income share of one third, we
reproduce Solows calculations for this new sample period,

1950-1973
1974-1990
1995-2000

Yi

K i

4.0%
2.7%
4.2%

3.0%
0.9%
3.0%

3%
2%
2.7%

2%
0.24%
2.1%

Contrib.
of

K i to Yi
33%
73%
30%

Contrib.
of
A to Y
i
67%
27%
70%

I have chosen somewhat arbitrary break-points, but we can see that the last 50
years of US growth could be divided into 3 periods: (i) post WWII characterized by high
productivity growth and strong capital deepening, with the capital-labor ratio growing at
3% and productivity growing at a rate close to 2%, (ii) the productivity slowdown period
1973-1995, during which the capital-labor ratio grew at the slower rate of 2% per year
and productivity almost remained constant; and (iii) the recent five years, with again high
productivity growth and high capital deepening.
The observed fluctuations in the rate of capital deepening are important, but
more important than that is the degree of change in total factor productivity growth;
the swings in this variable over the three periods are huge. The contribution of the Solow
residual to per capita growth exhibits great variation, from 27% of the observed growth
during the productivity slowdown to 70% in the late nineties.

2.2.4.5. Conclusion.
Robert Solow, in his 1956 and 1957, articles was mainly concerned with the
growth process of a mature capitalist economy, the U.S. All the emphasis was placed in
capital accumulation as the main engine behind growth and his results were striking; in
the presence of diminishing returns to capital, capital accumulation cannot be the
ultimate source of long run growth. In a sense capital accumulation is not the source
of growth but rather growth in itself.
Given the limited role played by capital in the production of output, observed
variations in capital across countries or in the same country across time are insufficient
to explain the variations in income across countries or over time. As a result Solow
pointed to technological change40 as the driving force behind the sustained increase in
40

Notice that differences in technology are; first, more difficult to measure since by technology we mean
ideas, methods of production, institutions i.e. anything that affects production and is not capital or labor

48

Econ 620 2007

income per capita experienced by capitalist economies, and therefore differences in
technological level could be at the root of the cross country differences in income per
capita.
Even if this simple version of the neoclassical model was able to match
quantitatively the data, it still would leave important questions unanswered. If
differences in income levels and growth rates where the consequence of differences in
saving choices and population dynamics we still need a theory to understand what
causes those differences, we will need models of savings (we already presented one) and
fertility choices to account for those differences.

2.3 Endogenous Growth Theory

If we are interested in the sources of long run growth, neoclassical models provide
a negative answer, specifically if markets are competitive and there are no externalities
(i.e. the share of national income paid to capital is a good proxy for the contribution of
capital to output) then physical capital accumulation cannot account for a large part of
either long-run growth or cross-country income differences. The main source of longrun growth, the rate of technological progress, was assumed to be exogenous and
therefore determined outside the model.
A group of growth theorists led by Paul Romer (1986) beggn to develop a class of
models in which the rate of technological progress is the result of intended/unintended
actions of the agents populating the economy, this stream of literature received the name
of endogenous growth theory.
In the absence of exogenous technological change, the standard neoclassical
model achieved a steady state along which per capita variables did not grow. The
diminishing returns to capital that characterize Solows technology imply that as
capital accumulates its marginal product (its additional contribution to output) decreases.
This continuous decrease in the marginal product of capital eventually leads to a situation
where investment is just enough to cover depreciation and provide enough capital for the
new born workers as a result growth ceases. This result was robust to changes in the
demand side of the economy, and therefore changes in our modeling strategy for the
supply side of the economy, the production technology, seem the most promising avenue.
A common feature of all the models in the new growth literature is the fact that
the marginal product of capital (defined in a broad sense to include both physical,
human, organizational) remains bounded above zero. As we will see the elimination
of diminishing returns opens the possibility for per capita growth even in the absence of
exogenous technological change.
In the following section we will introduce a rather rudimentary technology, aK ,
to illustrate the behavior of the economy under constant returns to scale for the
reproducible factors. Then we will briefly present some of the alternative production
structures introduced in the endogenous growth literature and finally we will explore an
endogenous growth model with human capital.
and second, it is difficult to understand what prevents the best ideas or methods of production to be used
across the world. When I use a pencil you cannot use it at the same time, but when I use a mathematical
formula you can be using it at the same time, without affecting my use of it (ideas are non rival, we will
return to this point in chapter 6).

49

Econ 620 2007

2.3.1. A model with Constant Returns to Scale in reproducible factors.

The key property of endogenous-growth models is the absence of diminishing
returns to capital at the aggregate level. The simplest version of a production function
with the marginal product of capital bounded above zero is a technology where output is
linear in capital
Y = aK

(1)

Where a is a positive constant that reflects the level of technology, and that in a
fully fledge model is endogenously determined. Notice that (1) violates the Inada
conditions.
Final output can be consumed or invested (saved). The fraction of output devoted
to investment, s, is exogenous and constant, so

K& = I K = sY K

K (0) = K 0 > 0

Rewriting the model in per capita terms, and assuming that population grows at an
exogenous rate, n, we obtain the following differential equation for the evolution of the
per capita capital stock,
Yi = aK i
K& i = sYi ( + n ) K i
K& = saK ( + n ) K
i

(2)

K i = sa ( + n )
Since Yi is proportional to K i , then Yi = K i = C i , given that consumption is a
fixed proportion (1-s) of output. Furthermore since (1) is a linear differential equation on
K i we can find the paths that the capital stock (and output and consumption) will follow,
aggregate
K i ( t ) = K i 0 e( sa n )t
K ( t ) = L ( t ) K i ( t ) = L0 e nt K i 0 e( sa n )t = K 0 e( sa )t

Using a diagram similar to the one we used for the Solow model, under the assumption
that sa >

50

Econ 620 2007

From this analysis we can see that the economy does not achieve a steady state,
since there is permanent per capita growth. Nonetheless we can use a similar definition of
equilibrium. In this model, we consider equilibrium to be a solution where the growth
rate is constant, and instead of steady-state we refer to it as a stable growth path (as we
did with the Solow model with exogenous technological change).

2.3.2. Alternative Specifications of the aK technology.

Arrow (1962) and Lucas (1988) present a model where learning-by-doing leads to
technological change as an accidental by-product of the economic activity. Lets
assume population is constant. We can model the level of technology as an increasing
function of the level of capital,
A ( t ) = BK ( t )

There are J identical firms indexed by j. Any of those firms hires inputs, labor and
capital combining them with the freely available level of technology to produce final
output, according to the familiar Cobb-Douglas specification
Yj = ( K j )

( AL )

Each firm, being just one of many, ignores its individual contribution to the
accumulation of knowledge (this is an externality), taking the evolution of A as
exogenous. As you can see in the aggregate we have a technology that exhibits constant
returns to capital, summing across firms,

K
j

= JK j =K ,

L
j

= JL j = L

so Y = Y j = JY j = J ( K j )

( AL )

= (K )

( AL )

= (K )

( BKL )

= aK

Barro (1990) emphasizes the importance of the government provision of public

infrastructure. Assuming a constant labor force, in his set up output is produced

51

Econ 620 2007

combining capital, labor and a non-rival publicly available good, G , according to the
following technology,
Y j = ( K j ) ( GL j )

The government devotes a constant fraction of output, Y , to the provision of

infrastructure. Summing across firms
Y = Y j = JY j = J ( K j ) ( GL j )

= ( K ) ( GL )

That combined with the fact that G = Y yields,

Y = ( K ) ( YL )

Y = ( K ) ( L )

Y = ( L )

K = aK

In this type of models the marginal product of capital is bounded above zero. As a
consequence of this any policy that affects either saving or the marginal product of
capital will have permanent growth effects. In contrast with Solow type models, as
capital accumulates, its marginal product tends to zero, so any policy will have only
temporary effects. Policies that induce changes in the allocation of resources to the
activity that causes technological change will have permanent effects on the growth rate
of the economy.

As we can see from the graph, higher saving lead to a permanently higher rate of
growth. The transmission mechanism is simple a higher level of saving implies higher
investment, and a higher accumulation of knowledge, and higher growth.
A final important issue with this type of models concerns the optimality of the
market solution. In the presence of externalities the First Welfare Theorem no longer
holds since there is a gap between the private marginal product of capital
1
1
( ( K ) ( KL ) = a in the learning by doing model) and the social marginal product
of capital ( ( K )

( KL )

= a in the same model). This is just a consequence of the

external effect, firms when making their investment decisions do not take into account
52

Econ 620 2007

the (negligible) effect of their individual decisions in the aggregate capital stock and in
the expansion of knowledge and therefore they invest below the social optimum.

2.3.3 An Endogenous Growth Model with Human Capital

The neoclassical growth model is at the same time a remarkable success and a
remarkable failure. The success is, of course, that the model is able to account for each
and every one of the stylized facts of growth in developed countries. In all appearance,
the model captures some key characteristics of the actual economic system of advanced
industrial countries. For this reason, the neoclassical growth model is the underlying
structure used for most macroeconomic research concerning developed countries.
The neoclassical growth model is also a failure, however, because it does not
provide any explanation for the lack of convergence between rich and poor countries
during the last 50 years. Evidently, there is something that keeps most of the world from
growing, and whatever it is, it is not captured in the neoclassical growth model.
The question arises whether we need to abandon the standard growth model
completely in order to understand stagnation in developing countries. In this section, we
will see that it is possible to extend the standard model in a way that keeps all its
implications for rich countries intact, while also offering an explanation for stagnation
in the poorer countries of the world. The key ingredient for this explanation is the
endogenous growth driven by the accumulation of both physical and human capital.
This possibility is empirically relevant since the correlation between the average
years of schooling and per capita GDP is close to 0.85. The following graph plots per
capita GDP for 1997 against the average years of schooling for our sample of countries,

50,000

Yi

40,000
30,000
20,000
10,000
0
0

10

12

14

Years of Schooling

As we can see more educated countries tend to enjoy higher income per
capita, and therefore at some point higher growth (or alternatively we could think that
richer countries devote more time to education).

2.3.3.1 Description

53

Econ 620 2007

Since endogenous savings seems not to play a crucial role in the results of our
analysis, we will summarize the demand side of our economy by an exogenously given
saving rate, s, and therefore each period a fixed proportion of output is saved/invested.
In the supply side we need to specify,
c) The production technology for final output.
d) The law of motion for physical capital.
e) The production technology for human capital and its law of motion.
- The production function for final output.

As before our economy produces an homogeneous good, Y ( t ) , combining

capital, K ( t ) , and skilled labor, H ( t ) h ( t ) L ( t ) , and labor augmenting technology,
according to the following Cobb-Douglas technology,
Y = K ( uhLA )

Our formulation of H implies that a worker with human capital41 h is the productive
1
equivalent of two workers with h each, or a half-time worker with 2h . Finally since
2
the accumulation of human capital requires our representative agent to spend time at
school, we noted by u the fraction of time devoted to the production of output. We
assume that this time allocation is exogenous and constant with the remaining time being
devoted to the acquisition of skills.
- The law of motion for physical capital.
As in the previous models the labor force is assumed to evolve exogenously at the
exponential rate
L = n, given L ( 0 ) = L0
L ( t ) = L0 e nt
Technology is assumed to grow at the exogenous rate, g A
A = g A , given A ( 0 ) = A0
A ( t ) = A0 e g At

Final output can be consumed or invested (saved). The fraction of output

devoted to investment, s, is exogenous and constant, and the existing physical capital
stock depreciates at a rate , so
K& = I K = sY K

K (0) = K 0 > 0

41

Notice that H is an aggregate measure of human capital and h is the average measure of human capital of
the population. In the last section of this chapter we will see how can we map the data on years of
schooling with our measure of human capital.

54

Econ 620 2007

- The production function for human capital and its law of motion.

The production technology for (per capita) human capital takes a different form.
Unlike physical capital, human capital is not just a part of output that gets converted into
productive capital. Rather, producing new human capital requires the use of existing
human capital and time. Human capital is produced through education (and
experience), and existing knowledge is the key ingredient in the production of education.
We therefore assume that future human capital is produced using only human, but not
physical, capital42. The amount of future human capital depends on the fraction of time,
1 u , spent on education and on the current level of human capital
h& = B (1 u ) h

(1)

Lucas (1988) points out that this production function is consistent with the
evidence we have on individual earnings. In our formulation B is a measure of the overall
productivity of the educational sector. Notice that if no effort is devoted to the
accumulation of human capital, then none accumulates and per capita human capital
remains stationary at its initial level.

2.3.3.2 Steady State of the model

The evolution of per capita human capital is easy to determine from its law of
motion, dividing both sides by h we reach,
h&
= h = B (1 u ) g h
h
To analyze the steady state of our model, it is once again useful to rewrite the
model in terms of variables that will achieve a steady state. Since labor, per capita
human capital and technology all of them grow at known rates, we can infer that a
variable that would satisfy this condition would be of the form,
x

X (t )
, where X = K , Y , C
A (t ) h (t ) L (t )

Using our newly defined variables, the production function and the law of motion
of physical capital become,

42

Mulligan and Sala-I-Martin (1993) analyze a similar model where both inputs are used in the production
of both types of capital and their result are consistent with those presented in this chapter, as long as the
degree of physical capital intensity is higher in the production of output than in the production of human
capital.

55

Econ 620 2007

K (t )

Y (t )

A (t ) h (t ) L (t )

( uh ( t ) L ( t ) A ( t ) )

A (t ) h (t ) L (t )

y = k u1

Y
k = K g A g h n = s ( + g A + g h + n )
k& = sy ( + g A + g h + n ) k
K

combining both results we reach the basic equation of our model with human capital,
k& = sk u1 ( + g A + g h + n ) k
which takes a very familiar form that we can solve for the steady state level of adjusted
capital,
1/1

s
k = u
k& = 0

+ g A + gh + n
*

( )

y = k
*

/1

s
= u

+ g A + gh + n

Remember that our adjusted variables are just a convenient transformation but
have little economic content so for practical purposes it is easier to think in terms of per
capita variables so,
1/1

s
Ki ( t ) = A ( t ) h ( t ) k = A ( t ) h ( t )

+ g A + gh + n
*

/1

s
Yi ( t ) = A ( t ) h ( t ) y = A ( t ) h ( t )

+ g A + gh + n
*

and their steady state growth rates,

K i = A ( t ) h ( t ) k *
K i* = g A + g h g A + B (1 u )
Yi ( t ) = A ( t ) h ( t ) y*
Yi * = g A + g h g A + B (1 u )
In terms of income per capita we can get a better grasp of the different factors at work.
/1

s
Yi ( t ) =

+ g A + B (1 u ) + n

uA ( t ) h0e B (1 u )t

As in the Solow model, countries with higher saving rate or lower rate of population
growth enjoy higher per capita income. The effects of education are somehow more
complex. On one hand the time devoted to the acquisition of skills reduces the time

56

Econ 620 2007

available to produce commodities and therefore the level of per capita output, that is
captured by g h = B (1 u ) in the denominator and the last u . But on the other hand, the
time devoted to the acquisition of skills affects the growth rates of human capital,
physical capital and output, e B (1 u )t . Since the growth effect is amplified every period, the
positive effect of education on income per capita soon dominates its negative effect,
leading to a positive association between income per capita and years of schooling.
Notice that even though the growth rate is constant, it does depend on an action
of the agents populating our economy, namely, their choice of the fraction of time
devoted to education, 1 u . This is one of the crucial differences with the Solow model
where the choices of the representative agent did not affect the steady state growth rate of
the economy.

2.3.3.3 Transitional Dynamics of the model

What does the human capital model imply for growth?43 Consider first the case of
two countries with the same amount of human capital. Let us also assume that the two
countries are identical in every respect apart from their initial level of physical capital.
Since the growth rate of human capital is constant and identical in the two economies, we
have the immediate conclusion that the level of human capital in the two economies will
always be the same. Lets look at the steady-state ratio between physical and human
capital,
1/1

K (t ) k * A(t ) h (t ) L (t )

s
= k * A(t ) =

h (t ) L (t )
+ g A + gh + n
H (t )
*

uA ( t )

so since both economies have the same parameters and exogenous variables, they only
differ on the initial levels of physical capital, both economies reach the same steady state
level of adjusted capital and therefore both economies eventually have the same ratio of
physical to human capital. As a result in the steady state both economies have identical
amounts of human and physical capital (since they have the same ratio and the same
amount of physical capital), and therefore the same level of output. If one country
happens to start with a lower level of physical capital, it will grow faster during the
transition to the steady state and ultimately catch up the other country. In other words, for
countries with the same level of human capital the model works just like the Solow
model. The model implies convergence along the physical capital dimension. This
feature is one of the main advantages of the human capital model: it preserves the
successes of the Solow model in terms of explaining growth in rich countries.
The situation is different if two countries start out with different initial levels of
human capital. Assume, for example, that the first country has twice as much human
capital as the second country (as before assume that the rest of the parameters and the
43

Given the similarities between the structure of the dynamic equations in this model and in the Solow
model with technological change, most of the conclusions we derived for that model are still valid in this
environment, but we will find some additional implications. Besides we could draw a similar version of the
graphical analysis conducted in chapter 3.

57

Econ 620 2007

initial level of physical capital are identical for both countries). Since human capital
grows at the same constant rate in both countries, the ratio of human capital between the
countries will always be two to one. Since in each country, the ratio of physical to human
capital is the same, the long-run ratio of physical capital between the two countries will
be two to one as well. In other words, the country that starts out with less human capital
never catches up to the rich country: convergence does not occur along the human
capital dimension.
We therefore see that the human capital model is, in principle, consistent with
most of the evidence on the world income distribution. The model does predict
convergence among rich countries (under the assumption that they have similar levels of
human capital), but also relative stagnation of poor countries that start with little human
capital.
One of the implications of the Solow model is that given the observed differences
in capital per capita across countries if nothing prevents the best technology to be adopted
by all the countries we should observe large capital flows from poor to rich countries, but
it seems that these patterns are not observed in the data. What are the implications of the
human capital model for international capital flows? The best approach to answer this
question is to calculate the implied return to capital in this model,
MPK =

Y
1
1
= K 1 ( uhLA ) = K i 1 ( uhA )
K

Since the marginal product of capital is increasing in the average level of education, h ,
this model is consistent with capital flowing to rich countries where the level of education
is generally higher.
The predictions for poor countries are not entirely pessimistic, however. Since
the growth rate is endogenous, the model points to a potential path out of poverty: a
country that invests a higher share of resources into education (or spends more time
acquiring skills) can increase its long-run growth rate, and thereby converge to rich
countries even if their initial level human capital is lower.

2.3.3.4 Empirical Evaluation of the aK Model

a. Kaldors stylized facts.
1. A wide variety of growth rates (and of income levels). In these models differences
in growth rates arise from differences in technology. As we have seen the level of
technology is endogenously determined by the rate of investment in physical
capital, the effort devoted to the accumulation of skills or the number of
inventors.
2. Per capita output growth at a positive rate which shows no tendency to diminish.
As we pointed out in the first section, Yi = K i = sa n
K 1
3. The capital-output ratio shows no trend. Since Y = aK = , where a is an
Y a
endogenously determined constant.
58

Econ 620 2007

4. The return to capital shows no trend. The marginal product of capita, a, is
constant.

b. Convergence.
We have defined convergence as the tendency for countries with lower initial
income to grow faster and catch up with higher income countries. We know that the
engine for convergence is based in diminishing returns to capital. In the presence of
diminishing returns, at low levels of capital its marginal product is high and so it is output
growth. The elimination of the diminishing returns to capital in the aK technology
removes the engine for convergence, opening up a set of interesting possibilities where
economies might temporarily converge, diverge or move along parallel paths. In general
this model predicts at best, lack of divergence.

c. Limitations.
To achieve a balanced growth path, any aK model requires exactly constant
returns to scale in the factors that could be reproduced. In general form we can write,
Y = aK
Using our basic differential equation in per capita terms,
K& i = saK i ( + n ) K i
K i = saK i 1 n

59

Econ 620 2007

If = 1 , we are fine, if < 1 , the model exhibits diminishing returns and
eventually growth will cease, we are back in the Solow world. Finally if > 1 growth will
take place at an increasing rate. Solow (1994) criticizes the very strong conditions,
exactly constant returns to scale to reproducible factors, required by this class of models
to generate stable growth.

d. Quantitative evaluation of the human capital model.

d.1. Regression Analysis

Mankiw, Romer and Weil (1992) used regression analysis to take a version of our
human capital model to the data. Under the assumptions that economies are in steady
state, the rate of technological change, g , is common across countries and that the
initial level of technology, Ai ,0 , is orthogonal to the independent variables, i.e. it is
uncorrelated with savings, education and population growth, they reached the following
estimation equation,
ln yi* = 0 + 1 ln ( sk ,i ) + 2 ln ( sh ,i ) + 3 ln ( ni + g + ) + i

(1)

This specification requires a proxy for their measure of investment in education, sh,i .
They used the fraction of the working age population that is enrolled in school. The
results of their exercise are summarized in the following table,

ln ( sk ,i )
ln ( sh ,i )
ln ( ni + g + )

Adjusted R 2

Estimates of the Human capital model

MRW 1985
Updated to 2000
.69 (.13)
.96 (.13)

.66 (.07)

.70 (.13)

-1.73 (.41)

-1.06 (.33)

.78

.60

Standard errors in brackets

These results are very impressive. Both measures of investment come out with the
right sign and so does the effect of population growth. In particular the adjusted R 2
suggests that close to 80% of the income per capita differences across countries can be
explained by differences in their physical capital, human capital and fertility. The
immediate implication is that technology differences, ( Ai ) , have a somewhat limited
role. If this conclusion was correct, it would imply that, as far as the causes of prosperity
are concerned, we can limit our attention to differences in human and physical capital
accumulation and abstract from technological differences across countries.
It was soon noticed some of the limitations of Mankiw, Romer and Weils work.
The assumption that the levels of technology are orthogonal is very difficult to justify.
We not only expect Ai to vary across countries but also to be correlated with physical

60

Econ 620 2007

capital, sk ,i , and human capital, sh ,i , investment. It seems sensible to assume that
countries that are more productive, high Ai , will also have higher levels of investment in
physical and human capital and therefore the key right hand variables in (1) are correlated
with the error term. As a result, OLS estimates of 1 , 2 and the adjusted R 2 will be
biased upwards. This potential overestimation of the role of human and physical
capital casts serious doubts over the previous analysis. Since the physical and human
capital coefficients are incorrectly capturing most of the productivity differences,
uncovering the true contribution of capital accumulation requires some assessment of the
productivity differences across countries.
d.2. Calibration Analysis

A potential alternative approach to explore the relative importance of physical capital,

human capital and technology is to calibrate the model, in a similar fashion that with the
standard Solow model. Lets begin with the stable growth path level of income per capita
in this model,
/ 1

s
Yi ( t ) = A ( t ) h ( t ) y =

+ g A + B (1 u ) + n
*

uA ( t ) h ( t )

(1)

According to this expression, countries with higher savings, lower population growth
or higher educational achievement will enjoy higher welfare and growth. As in the
simple Solow model we need to find the empirical counterparts to the parameters and
variables of the model, to evaluate how it fits the observed variation in income per capita
across the world. We have data on income per capita for 1997, average saving, and
average years of education for 1995 and average per year population growth rate from
1980-97 and therefore a possible strategy to calibrate our model is as follow
1. We assume that technology is the same across countries, so Ai ( t ) = Aj ( t ) . Since
our model does not provide a framework to understand the differences in
technology across countries, it seems sensible as a first step to ignore them.
2. We assume that the common rate of growth of technology is 2% and the rate of
depreciation 5%.
3. We have defined h ( t ) as a measure of the average years of schooling per worker,
now we need to map our data on years of schooling into its productive
counterpart, h ( t ) . The labor literature suggests that additional years of schooling
increase earnings proportionally. Labor economist explain earnings using what it
is known as the Mincerian wage equation,
ln wi = X i + Si
where wi is wage earnings of the individual in some labor market, X i is a set of
controls that might affect wages apart from education (family status, age,
experience) and X i is years of schooling. The estimate of the coefficient is the

61

Econ 620 2007

rate of return to education, measuring the proportional increase in earnings
resulting from one more year of education. This literature provides estimates of
that range between 0.06 and 0.1, i.e. it suggests that an increase in one year of
education increases wages somewhere between 6% and 10%. Now if we assume
that labor markets are competitive this implies that one more year of schooling
increases worker productivity by about 6 to 10 percent. We take the midpoint 8%,
and we therefore estimate h ( t ) ignoring the other covariates as
hi = exp ( 0.08* Si ) .

4. Finally, we normalize B = 1 and then we assume that the average

working+schooling life is around 55 years. We can combine the data on years of
schooling with this assumption to infer the values of (1 u ) for different
countries in the following way,

Cameroon
Brazil
Canada
U.S.A.

Y of E
3.37
4.45
11.39
11.89

(1-u)
3.3/55 0.06127
4.4/55
0.08
11.3/55 0.2070
11.8/55 0.2161

Assuming economies are in steady state we can use (1) to calculate the values for
the relative income (income of a country over Canadian income) predicted by our
model, as follows,
/1

si
ui hi

*
+ g + B (1 u ) + n
Yi* A ( t ) hi yi
A
i
i

=
=
/1
Yc* A ( t ) hc yc*

sc
uc hc

+ g A + B (1 uc ) + nc

And compare the resulting series with the relative income recovered from the data.

62

Econ 620 2007

Pre Yi/Yc

1.5

0.5

0
0

0.5

1.5

Yi/Yc

At first sight, the qualitative predictions of the model are consistent with our data.
Countries that devote a higher fraction of their time to the accumulation of skills tend to
enjoy higher income per capita. But at the same time our calibration of the human capital
model tends to overstate the role of human capital for low income countries, countries
with relative incomes below 1 tend to have a predicted income always higher than actual
income. This suggests that the level of human capital is not low enough in poor countries
to explain their poverty or equivalently high enough in rich countries to account for their
wealth neither, the observed variation in years of schooling is not enough to explain the
observed variation in income.
Accounting for human capital is an important first step towards answering the
questions about international income differences left open by the Solow model, but it
seems that investment in education is not enough to explain the performance of poor
countries. As a result we still have important cross-country differences in income
levels that are not explained either by physical or by human capital differences.
Lets see how can we incorporate technology into the analysis, at least in a very
naive way, we can use (1) to solve for the level of technology to exactly match the levels
of income per capita observed for each country in 1997. We choose the level of
technology for each observation in such a way that the predicted income of the country
equals its actual income, in a sense we call technology the ratio between the observed
income and the predicted income (this predicted income was calculated under the
assumption that technology was equal across countries and normalized to some constant).
We can now use our estimates of Ai to calculate productivities relative to the level
of Canadian productivity. The next graph plots these relative technology levels against
relative incomes,

63

Econ 620 2007

1.5

Pre Ai/Ac

0.5

0
0

0.5

1.5

Yi/Yc

Notice first that by construction Ai is capturing any source of variation in income

per capita beyond physical and human capital. The first interesting observation is that
rich countries tend to have high levels of technology and poor countries generally have
low levels. High income countries not only have more physical capital and human capital
but they also combine these inputs in a much more productive way. Second, our estimates
of Ai are similar to the Solow residual introduced in the previous chapter, so they capture
anything that is not physical capital or human capital (measured as years of
schooling). Our measure of technology is therefore capturing differences in the quality of
education, experience acquired in the work place, health of the labor force, differences in
geography and institutions across nations that can have effects on production and
distribution of goods, variation in working hours44 and therefore some authors prefer to
refer to this estimates as total factor productivity, a term that already introduced.
In this section we have presented an endogenous model to allow for human
capital to play an active role in production. This exercise enhances the ability of the
model to account for differences in income per capita and growth rates across countries
and provides an explanation for the lack of convergence between rich and poor countries
observed over the last half century. On the downside the introduction of human capital
still falls short to explain the huge gap in income per capita across nations. Nonetheless
the model points towards differences in total factor productivity as the key element to
understand the process of growth and development.

44

An average employed Canadian works 1,782 hours a year while an average worker in Norway works less
than 1,400 hours. That is a difference of around 30% that tends to underestimate Total Factor Productivity
in Norway relative to the Canada. This problem can be solved using output per hour. When we look at
history, in 1870 an average Canadian manufacturing worker spent around 65 hours per week in the factory,
nowadays less than 40 hours per week.

64

Econ 620 2007

2.4. Beyond K, H and Ideas: Fundamental Causes of Growth

We began this course pointing out the incredible differences in income per capita
across countries. Focusing on the five wealthiest countries and the five poorest ones, the
following table reproduces their income levels and some of the variables we have
identified as potential sources of those differences,
Yi
35,439
34,941
33,337
30,965
29,858

Ki
87,330
107,283
82,443
107,870
88,076

h
11.8
9.5
10.7
9.8
10.6

s
0.20
0.27
0.25
0.30
0.25

n
0.010
0.008
0.012
0.007
0.014

Tanzania
Uganda
Burundi
Malawi
Niger

1,124
1,123
1,055
1,048
1,011

1,171
988
1,118
1,185
1,153

4.8
4.6
4.7
5.0
3.8

0.10
0.02
0.07
0.07
0.08

0.024
0.027
0.021
0.030
0.033

Top 5
Bottom 5

32,908
1,072

94,600
1,123

10.48
4.59

0.25
0.07

0.01
0.03

30.7

84.2

2.3

3.7

0.4

U.S.A.
Luxembourg
Canada
Switzerland
Australia

Ratio

All the data is from Hall and Jones (1999), instead than from our original data set.

On average the wealthiest countries have 30 times more income per worker than
the poorest ones, their workers are equipped with almost 85 times more capital and they
have been educated for almost 10.5 years as opposed to the 4.5 years of the workers in
the poorest countries.
Following Hall and Jones (1999) we are going to perform a growth accounting
exercise in levels rather than in growth rates. As they argue levels capture the
differences in long-run economic performance, are more relevant to well-being as
measured by consumption, and with technology flowing relatively free across countries
most models predict that all countries will converge to a common growth rate in the
long run. As a consequence long-run differences in levels are the relevant differences to
explain.
Assume output is produced according to the following Cobb-Douglas production
function,
Y = K ( AhL )

where K is the capital stock, A is a measure of total factor productivity, h , is the average
level of human capital and L is the size of the labor force. Output per capita is given by,

65

Econ 620 2007

Yi = K i ( Ah )

Taking logs in both sides yield the following expression that decomposes
increases in output per capita into increases in capital per capita, increases in
human capital per capita and increases in total factor productivity
ln Yi = ln K i + (1 ) ln h + (1 ) ln A

Consider the following thought experiment, the level of A rises without any
change in saving or education per worker. With a constant saving rate, the resulting
higher output increases the amount of physical capital and when the country reaches its
new balanced growth path, K i and Yi will be higher by the same proportion. The previous
decomposition will attribute a fraction of the higher log of income per capita to
increase in capital per capita and the remaining (1 ) portion of the increase in the log
of output per capita to total factor productivity, while we know that all the increase comes
from an increase in total factor productivity. The way proposed by Hall and Jones (1999)
to address this issue is to subtract from both sides of the previous equation ln Yi ,
reaching

ln Yi ln Yi = ln K i + (1 ) ln h + (1 ) ln A ln Yi

(1 ) ln Yi = ( ln Ki
ln Yi =

ln Yi + (1 ) ln h + (1 ) ln A

ln i + ln h + ln A
(1 ) Yi

The last equation decomposes output per worker into capital intensity, measured
by the capital-output ratio, education and technology. Furthermore, using this equation
in our previous thought experiment we get the answer we were looking for. Assuming
= 1/ 3 and using the previous expression to decompose output per capita for the top-five
and bottom-five economies we get,

Top 5 countries45
Bottom 5 countries46
Difference in logs
Relative contribution

ln Yi

ln i
(1 ) Yi

ln h

ln A

10.4
6.98
3.42
100%

0.53
0.02
0.51
14.91%

0.84
0.37
0.47
13.74%

9.03
6.59
2.44
71.35%

45

Canada, US, Ireland, Norway and Singapore. Mincerian equations are used to map years of education
into productive human capital.
46
Ginea-Bissau, Haiti, Malawi, Togo and Uganda.

66

Econ 620 2007

Only around 15% of the difference in GDP is explained by differences in capital per
worker, education accounts for another 15% of the difference, with the remaining 70%
of the difference left unexplained, what we have labeled as total factor productivity.
Notice that after half a century of research we reach a result fully consistent with Solow
(1957), where he found that around one eighth of US output per capita growth between
1909 and 1949 was a result of physical capital investment.
But the most important finding from Hall and Jones (1999) is that the
contributions of physical capital, schooling and total factor productivity are not
independent. Rich countries not only have more capital per worker, higher saving and
lower fertility rates, a better educated labor force but they also combine them in a more
efficient way. They find a correlation coefficient of .52 between the log of education and
the log of total factor productivity and of .25 between the log of capital per worker and
the log of total factor productivity. The next graph borrowed from the cited paper
reproduces the strong positive correlation between output per worker and total factor
productivity,

These strong correlations between the different engines of growth lead scholars to think
that there is some underlying force that is driving the accumulation of inputs and the
level of productivity. In words of Hall and Jones (1999) capital accumulation, education
and technology are only proximate causes for growth, while the fundamental
determinants of the wide variation in income levels across countries are still to be
determined.

Output per worker

( K i , H i , A )

Institutions (Acemoglu et al., 2001)

Geography (Bloom et al., 1998)
Property Rights (De Soto, 1989)
Culture (Landes, 1998)
Incentives (Easterly, 2002)
Luck, multiple equilibria, others

67

Econ 620 2007

If one found that inputs of production are able to account for most of the
differences, then growth economics could focus on explaining low rates of factor
accumulation, there would be ample scope for controversy over the policies better suited
to engineering higher investment rates in various types of capital but some consensus
could be reach in what are the intermediate goals of growth theory. But unfortunately
most of the evidence suggests that differences in efficiency are at the root of the
differences in output per capita, and therefore growth theory faces the additional task of
explaining why with similar level of inputs some countries are able to produce much
more output than others.
Lets briefly discuss the main arguments supporting each of the hypothesis behind
this fundamental determinants of income47.

8.1. Luck and multiple equilibria

The basic flavor of multiple equilibria models is captured by the following game
of investment. Assume the economy is populated by N investors indexed by i with the
payoffs of the k -th individual and the rest of the economy summarized in the following
table.

High investment i = k
High investment i k
Low investment i k

y ,y

yL , yL

Low investment i = k
yL , yL
yL , yL

If y H > y L > > 0 , we have two pure strategy symmetric equilibria in this game. If the
k -th individual expects all other agents to choose high investment, it is optimal for him
to invest at the high rate. In the other hand, if the k -th individual expects all the others to
choose low investment, low investment is his best response. Since the same analysis
applies to any of the other individual agents, we end up with two symmetric equilibria.
These types of models highlight the importance of complementarities in many
economic activities. For instance a relatively simple production process, manufacturing
T-shirts, requires success along a surprisingly large number of different dimensions. Our
T-shirt firm needs the basic inputs of production; cotton and polyester, machines, a
competent and healthy workforce with the required know-how, a factory building, a
continuous supply of energy and water Apart from these the production process must
be kept secure from theft and expropriation, the firm must acquire licenses and obtain
regulatory approval for production, a good transportation network that allows access to
consumer market Finally, the managers of the firm need to know how to organize the
production process, how to manage the work force These different activities are
complementary (failure along any single dimension can have an important impact in
output) and sometimes external to any individual firm (the market, as a coordination
device, might play a very limited role).

47

This discussion draws on Acemoglu (2007)

68

Econ 620 2007

8.2. Geography
Nature, that is, the physical, ecological and geographical environment of nations
can potentially play an important role in their economic performance. In words of Nobel
Prize winner Gunnar Myrdal
as serious study of the problems of underdevelopment should take into
account the climate and its impact on soil, vegetation, animals, humans
and physical assets.
A more actual version of the geography hypothesis links poverty in many areas of
the world to their disease environment, as pointed out by the World Health Organization
in todays world, poor health has particularly pernicious effects on
economic development in sub-Saharan Africa, South Asia, and pocket of
high disease and intense poverty elsewhere and extending the coverage
of crucial health services to the worlds poor could save millions of lives
each year, reduce poverty and spur global development
Along these lines it is well-documented in the microeconomics literature that
unhealthy individuals are less productive, less able to learn and accumulate less physical
capital48.

48

The adult survival rate is the fraction of 15 years old that will survive to age 60. Several authors have
estimated the effects of height (a health outcome) on wages (a proxy for productivity). Their results account
for the genetic variability in height and the potential correlation of health with other determinants of wages
by instrumenting height with inputs into health in childhood that are not related to family income, such as
distance to health facilities. Their results suggest that a 1 cm increase in wage is associated with an 8%
increase in wage in Colombia (Ribero and Nunez (2000)), a 9% in Ghana (Schultz (2002)), and a 7.8% in
Brazil (Schultz (2002)) for countries such as Brazil, Ghana, Mexico and Colombia.

69

Econ 620 2007

Weil (2006) proposes a growth accounting framework that incorporates health
capital. Using micro-estimates to measure the contribution of this capital to productivity
he concludes that around 20% of the variation in income per capita is explained by
differences in health.
Nonetheless it is important to notice that the fact that the burden of disease is
heavier in poor nations today is as much a consequence as a cause of poverty.

8.3. Culture
Culture is viewed, by some social scientists, as a key determinant of the values,
preferences and beliefs of individuals and societies that shape their economic
performance. Weber (1930) argued that English piety, in particular, Protestantism, was
an important driver of capitalist development. In his view, Protestantism led to a set of
beliefs that emphasized hard work, thrift and saving. It also interpreted economic
success as consistent with, even as signaling, being chosen by God. Many historians and
scholars have argued that not only the rise of capitalism, but also the process of economic
growth and industrialization are intimately linked to cultural and religious beliefs. Similar
ideas are also proposed as explanations for why Latin American countries, with their
Iberian heritage, are poor and unsuccessful, while their North American neighbors are
more prosperous thanks to their Anglo-Saxon culture.
In my view theories of economic growth based on culture do not stand a serious
evaluation.
Although no former Spanish or Portuguese colony has been successful
economically as British colonies such as the United States, many former British colonies,
such as those in Africa, India, Bangladesh and the Caribbean, are poor today; Why did
the Protestant ethic have such a large impact in North America and Australia and not in
Africa or the Caribbean49?
If some Asian cultural values are responsible for the successful growth experience
of South Korea why do these same values not spur growth in North Korea? If Asian
values are important for the current Chinese growth why did they not lead to a better
economic performance during the first three decades that followed World War II?

8.4. Institutions
Douglas North that has widely explored the importance of institutions in the
historical development process, offers the following definition,
Institutions are the rules of the game in a society, more formally, are the
humanly devised constraints that shape human interactions In

49

Some authors claim that what matters is the European heritage. While North American population is
mainly of European descent, the population of former British colonies in Africa is not of European descent.
Again it is not difficult to come with counter-examples, the high rates of population with European descent
of Uruguay and Argentina are associated with poor economic performance, while Singapore or Honk Kong
have almost no European heritage and their economic record is outstanding.

70

Econ 620 2007

consequence institutions structure incentives in human exchange,
whether political, social, or economic
This definition makes institutions a rather broad concept. We can think of the structure of
property rights, the presence and functioning of markets (especially capital and
insurance markets), the contractual framework available to individuals as some
easily identifiable economic institutions. The proponents of this hypothesis expect
societies with economic institutions that facilitate and encourage factor accumulation,
innovation and the efficient allocation of resources to prosper relative to societies that do
not have such institutions. Identifying the key institutional features that provide the
right incentives50, exploring the interaction between different institutions,
understanding the conflicts of interest at the source of institutional inertia are some of
the most promising areas of research.

50

Most likely there is not a single mapping between institutions and economic outcomes. For instance,
scholars often highlight the importance of private property arrangements as a requisite for development.
But the introduction of private property in the former Soviet Union preceded the collapse of the Russian
economy, while Chinas mixed-property system, where businesses are co-owned by private actors and local
governments, has recently produced remarkable outcomes.

71

Econ 620 2007

Appendix I.

Formally, we can combine Solows dynamic equation with the production function,
%k
k& = sk ( n + g + ) k
k = sk 1 ( n + g + )
y = k
y = k

to reach,

y = sk 1 ( + n + g )
1

k = y
k
Since y = k

=y

using this result

y = sy ( + n + g )

Which we can rewrite in terms of ln y,

1 ln y

ln y
= se
( + n + g )
t

and that we are going to linearize around ln y*

1 ln y*

1 1 ln y*
ln y
*
se
( + n + g ) + s
ln y ln y
e
t

1 *
ln y
ln y
ln y ln y*
( 1) se
t
ln y
( 1) ( + n + g ) ln y ln y*
t

For convenience lets define ( 1) ( + n + g ) < 0

ln y
solve _ diff _ equation
rearrange
ln y ln y*

ln yt = ln y* + ln y0 ln y* e t

substract _ ln( y0 )
ln yt = ln y* (1 et ) + ln y0e t
ln yt ln y0 = ln y * (1 e t ) ln y0 (1 e t )

Given that 1 et is positive (since < 0 ), this equation implies that the growth
rate between 0 and t ( ln y ( t ) ln y ( 0 ) ) is positively correlated with the steady state

72

Econ 620 2007

level of income and negatively correlated with the initial level of income. What is
even more important we can empirically measure it, having information on growth rates,
initial and steady state levels of income51.

51

Before proceeding with the estimation one additional step is needed, since we dont have data in units of
effective labor, we need to rewrite it in terms of per capita variables. Since Yit = At yt yt = Yit At , then our
equation becomes.

Y
Y
ln it = ln y* 1 et + ln i 0 et
At
A0

Since

Y* Y* Y*
y* = it = i1 = i 2
At
A1 A2

is constant, for convenience we evaluate it at the initial period.

Y*

i 0 1 e t + ln Yi 0 e t = ln Y * 1 e t + ln Y e t ln A
= ln

i0
0
i 0
A0

A0
A
ln Yit = ln Yi*0 1 e t + ln Yi 0 e t ln t = ln Yi*0 1 e t + ln Yi 0 e t + gt


A0

Y
ln it
At

ln Yit ln Yi 0 = ln Yi*0 1 e t + ln Yi 0 e t ln Yi 0 + gt

Y
ln it = gt + ln Yi*0 1 e t ln Yi 0 1 e t

Yi 0

Y
ln it = a + ln Yi 0 + j
j
Yi 0 j

73