Instituto de Estudios Avanzados en Desarrollo (INESAD)

Report Part Title: Theory of Convergence

Report Title: Theory, History and Evidence of Economic Convergence in Latin America
Report Author(s): Paola Andrea Barrientos Quiroga
Instituto de Estudios Avanzados en Desarrollo (INESAD) (2007)

Stable URL: http://www.jstor.com/stable/resrep00574.4

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Instituto de Estudios Avanzados en Desarrollo (INESAD) is collaborating with JSTOR to digitize,

preserve and extend access to this content.

This content downloaded from

102.128.76.42 on Thu, 08 Dec 2022 11:33:14 UTC
All use subject to https://about.jstor.org/terms
 6

Chapter 2

Theory of Convergence

This chapter introduces in detail, the theoretical growth model that will be used in the
further estimations of convergence and compares it to other models that also study con-
vergence. The purpose of this chapter is to make clear all concepts of convergence with an
emphasis on those concepts that are used in this study.
The chapter contains four sections. The …rst section speci…es the basic foundations
of a neoclassical growth model (NGM), used in Barro and Sala-i-Martin (2004), which is
based on the growth model of Ramsey (1928) and optimal growth model of Solow (1956).
Furthermore, it discusses the di¤erences between absolute and conditional convergence, and
beta ( ) and sigma ( ) convergence. The second section explains the main characteristics
of other three models that also study convergence. The …rst one is the endogenous growth
model. The second, the distribution dynamics model that derives convergence within
clubs of countries that emerge over time. The third is the technology di¤usion model
which is based on microfoundations and derives the catching-up convergence concept. The
third section explains brie‡y the link between the theory of economic integration and
convergence. Finally, a summary of the most important points and concepts is presented.

2.1 The Neoclassical Growth Model

The neoclassical growth models try to predict stylized facts of economic growth and one
of them is convergence. It seems, empirically, that conditional on relevant characteristics
for economic growth, there is a negative relation between initial income levels and growth
rates of income of a certain period. This means that rich countries tend to grow less than
poor countries, once some conditions are settled. This kind of convergence is known as

This content downloaded from

102.128.76.42 on Thu, 08 Dec 2022 11:33:14 UTC
All use subject to https://about.jstor.org/terms
 7

conditional convergence and it is well forecasted by the NGM whenever economies have
similar technologies and preferences.
The NGMs are based on an economy with a speci…c production function and a util-
ity function that represents its preferences. Under some assumptions, the economy will
eventually arrive at an equilibrium called the steady-state, where it cannot grow anymore.
If the economy is approaching its steady-state, there is convergence but if it is moving
away from the steady-state, there is divergence. The NGM used here, is able to calculate
the speed of convergence at which the economy gets closer to its own steady-state. The
procedure is explained below

2.1.1 Production Function

The production function is a Cobb-Douglas with labour augmenting technological progress1 :

Y = F (K; LA) = K (LA)1 ; (2.1)

where Y is the output, K; L and A are the capital, labour and technology respectively,
and is the share of capital (0 < < 1). The technology A grows at a constant rate x.
The former equation can also be written in e¤ective labor terms (since is homogenous
of degree one):
^ = Ak^ ;
y^ = f (k) (2.2)

where y^ = Y =AL and k^ = K=LA. In a closed economy, under equilibrium, savings

equals investments, implying that k^ moves dynamically according the following equation:

k^ = Ak^ c^ ^
( + x + n)k; (2.3)

where c^ is the consumption in e¤ective labor terms, is the depreciation rate of capital
and n is the constant growth of the population (or labor).

1
As usually a neoclassical production funtion, assumes constant returns to scale in capital and labour,
positive but diminishing marginal product of inputs, essenciability of inputs and satisfaction of the Inada
conditions.

This content downloaded from

102.128.76.42 on Thu, 08 Dec 2022 11:33:14 UTC
All use subject to https://about.jstor.org/terms
 8

2.1.2 Utility Function

Households are assumed to have in…nite time horizon and they maximize the following
utility function:
Z h i
U= c1 1)=(1 ent e t
dt; (2.4)

where c is consumption per capita, is the elasticity of marginal utility and is the
rate of time preference2 The solution of the maximization problem yields to the dynamic
equation for consumption (Ramsey, 1928):

h i
c=c
_ = (1= ) k^ 1
: (2.5)

2.1.3 Steady-State Equilibrium

De…nition 1 The steady-state equilibrium is the situation in which various quantities grow
at constant rates (Barro and Sala-i-Martin, 2004)

The steady-state is reached when output, capital and consumption in terms of e¤ective
labor no longer grows (y^ = k^ = c^ = 0), so variables in terms of per capita grow at a
constant rate, x; and level variables grow at rate x + n:
From Equation(2.3), the Steady-State is given by:

sf (k^ ) = ( + x + n)k^ ;

where y^ = f (k^ ) is the steady-state value of y^ and s is the saving rate (s < 1):
When an economy starts with a level of capital per unit of e¤ective labor lower than
^
the one in the steady-state (k(0) < k^ ), the capital level will monotonically increase until
it reaches its steady-state value. This means that its growth rate declines monotonically3 .
Since output varies together with capital, the output growth rate will also be monotonically
declining when its level is below its steady-state level. In other words poor countries

2
and by the transversality condition, that assures utility maximization in the long run, is higher than
n + (1 )x:
3
The proof is in Appendix 2D of Barro and Sala-i-Martin (2004).

This content downloaded from

102.128.76.42 on Thu, 08 Dec 2022 11:33:14 UTC
All use subject to https://about.jstor.org/terms
 9

will grow faster than rich countries, assuming that both have the same technologies and
preferences, until they converge to the same steady-state.

2.1.4 The Logic of the Speed of Convergence

In order to quantify the speed of convergence it is necessary to log-linearize the Equations(2.3)

and (2.5) around the steady-state. This consists in taking the …rst-order Taylor expansion
of both Equations4 . The result, which is saddle-path stable, is:

t t
log [^
y (t)] = e log [^
y (0)] + (1 e ) log(^
y ); (2.6)

where is the negative eigenvalue from the process of Taylor expansion and 0 < <1
. The condition < 1 rules out leapfrogging or overshooting, where poor economies are
systematically predicted to get ahead of rich economies at future dates and when >0
means there is convergence. Thus can be interpreted as the speed of convergence.
After doing some transformations, Equation(2.6) can be written as:

(1 e T)
(1=T ) log [y(T )=y(0)] = x + log [^
y =^
y (0)] ; (2.7)
T

where the left-hand side of the equation is the average growth rate in the interval from
0 to T . Notice that the left hand side is not anymore in terms of e¤ective units (in terms
of y^). This is because the "e¤ective" part, the technology, now appears in the right-hand
part as the growth rate of technology x because of the transformations.
Equation(2.7) says that the average growth rate of output in the interval T is related
negatively to the initial output y(0) in relation to the steady-state output y^ , and to
the technology growth rate x, while keeping and T constant. This implies conditional
convergence in the sense that a poor country A will grow faster than a rich country B,
understanding that country A is poorer because it is away from the steady-state than
country B is and that both have the same steady-states.

4
This is done in Appendix 2D of Barro and Sala-i-Martin (2004).

This content downloaded from

102.128.76.42 on Thu, 08 Dec 2022 11:33:14 UTC
All use subject to https://about.jstor.org/terms
 10

If the period taken into account is quite long, and is taken as given, the average
growth rate of output will tend to approach the technological growth rate and the negative
T
e¤ect of the initial position, y(0), will be diminished. So as T ! 1; ( 1 e
T ) ! 0. And
T
when the period is quite short T ! 0; ( 1 e
T )! (by l’hopital rule), the negative e¤ect
of y(0) will be .
When increases, taking T as given, the term (1 e t) increases as well, which
means that at high levels of ; the response of the average growth to the di¤erence between
log y and log y(0) is greater.

2.1.5 Theoretical Behavior of the Speed of Convergence: an Unrealistic pre-

diction

The speed of convergence determination, is also a result from the log-linearization and by
assuming a constant saving rate as in Solow (1956), yields5 :

= (1 ) (x + n + ): (2.8)

Here, depends on parameters not connected to the utility function; it is increasing

with the exogenous technology growth, population growth and the capital depreciation,
and decreasing with the share of capital, . In the extreme case when = 1, convergence
is zero. This would be the case of the endogenous growth model with AK technology.

5
When assuming a varying saving rate, is:
1=2
1 2 1 + + x 1
= 2
+4 ( ) ( + + x) (n + x + ) 2
;

where = n (1 ) x>0
This complex de…nition of speed of convergence varies with a lot of parameters linked and not linked to
the utility function and with the behavior of the saving rate during the transition towards its steady state,
which is:

s = (x + n + )=( + + x):
The transitional dynamic of the gross saving rate is monotonic; when the economy begins with a low
^ then the saving rate increases if s > 1= , decreases if s < 1= and it is constant if s = 1= .
value of k,
The way parameters in‡uence depends on the interaction with the saving rate. A higher value of , the
^ and therefore
intertemporal substitution coe¢ cient, increases the chance that the saving rate rises with k
reduces the speed of convergence (because we are closer to the steady state). In (Barro and Sala-i-Martin,
2004), it is shown that when , and/or x increases the speed of convergence increases as well, but when
n increases the impact on is not clear.

This content downloaded from

102.128.76.42 on Thu, 08 Dec 2022 11:33:14 UTC
All use subject to https://about.jstor.org/terms
 11

To have some idea of the theoretical speed of convergence one can substitute believable
values for the parameters in Equation(2.8) . In Barro and Sala-i-Martin (1992) the exercise
was done with values of growth rates for technology and population of 2% per year for each
variable, capital depreciation rate of 5% per year, constant saving rates and a capital share,
, of 0.35. The result was = 0:059 per year which means a half-life for the logarithm of
output per e¤ective worker of 11.8 years (= ln(2)=0:059). A speed of convergence of 0:059
is a rapid rate compared to 0:02, which is the empirical estimate normally found by
applied research (see chapter 3)
To get the empirical estimates of 0:02, according to the theory, would imply a
capital share of 80%, which is extremely high. A more ample de…nition of it as human
and physical is one way to explain the high share of capital. A capital share of 80% (and
x = n = 0:02; = 0:05) implies that economies tend to stay far from the steady-state for
long periods: every 34.7 years an economy will be in half way closer to reach its steady-
state.
Thus to explain through theory the 2% of convergence found empirically, requires high
levels of share of capital, . Under variable saving rates, this is possible if the intertemporal
substitution parameter, , acquires high values and if the depreciation rate of capital, , is
almost zero. Unfortunately these values are not realistic at all (Barro and Sala-i-Martin,
2004).

Relaxing Assumptions

In order to …nd an explanation of the high theoretical speed of convergence or a solution

to "cool it down", di¤erent variations were introduced and some assumptions relaxed.
For instance, when permitting for perfect factor mobility, unrealistic results were found.
Among them were in…nite theoretical speeds of convergence, and that the most patient
country asymptotically owned everything and consumed almost all of the worlds output
(Barro and Sala-i-Martin, 2004).
An additional variable was added, the costs of installing extra capital, which indeed

This content downloaded from

102.128.76.42 on Thu, 08 Dec 2022 11:33:14 UTC
All use subject to https://about.jstor.org/terms
 12

brought …nite speed of convergence. These adjustment costs cannot by themselves explain
the low speed of convergence observed empirically. Then, to relax the assumption of in…nite
horizons and to open the economy, both together, do not solve the problem either.
Barro, Mankiw, and Sala-i-Martin (1992) extended the neoclassical growth model by
distinguishing between two types of capital: the …rst one is mobile and can be used for
borrowing and lending from one country to another whereas the second one cannot be used
as collateral (the second type can be interpreted as human capital is not mobile). Under
these extensions, the theoretical value of the speed of convergence is close to the empirical
of 0:02. The estimations are still based on high proportions of ample capital, 80%, but
from this, some can be used as collateral and some not. The proportion that can be used
as collateral is from 0 75% which yields a convergence speed of 0:014 to 0:035.

2.1.6 Absolute vs. Conditional -Convergence

Above, the concept of conditional convergence was explained, which in short says that
when the growth rate of a country is positively related to the distance from its initial level
of income to its own steady-state. In other words, countries grow more if they are initially
further away from their own steady-state.
Meanwhile absolute -convergence exists when poor economies grow faster than rich
ones, regardless of whether they have a common steady or not. So poor countries tend to
"catch up" when time passes.

Equations

The two concepts can be described through equations. For the conditional convergence,
the average growth rate from Equation(2.7) can be written for each country i and include
a random disturbance ui0;T :

(1 e T) (1 e T)
(1=T ) log [yiT =yi0 ] = xi + log [^
yi ] log [^
yi0 ] + ui0;T : (2.9)
T T

If > 0 there is conditional convergence.

This content downloaded from

102.128.76.42 on Thu, 08 Dec 2022 11:33:14 UTC
All use subject to https://about.jstor.org/terms
 13

absolute -convergence still de…nes a negative relation between the average growth
rate and the initial income per capita, but excludes explicitly the steady-state income and
the technology growth rate:

(1 e T)
(1=T ) log [yiT =yi0 ] = a log [^
yi0 ] + ui0;T ; (2.10)
T

where a is a constant, that may or may not include the steady-state and technology
growth rate, and if > 0 there is absolute convergence
Both concepts can refer to the same de…nition and can be measured by the same
equation, in some cases. But in others they contradict each other and measuring can raise
some problems.

Di¤erent Concepts When Di¤erent Steady-States

When economies reach conditional but not absolute convergence, means that a rich econ-
omy can grow faster than a poor if the rich is further below its own steady-state than
the poor is in relation to its own steady-state. So both, rich and poor, have di¤erent
steady-states. In this case the correct equation to measure convergence is Equation(2.9)
because it will measure the convergence of each country to its own steady-state. If instead
Equation(2.10) is used, it will be misspeci…ed and the error term will be:

(1 e T)
wi0;T = ui0;T + log [^
yi ] + xi
T

and if y^i0 is related to y^i , as estimated will be biased. But if y^i0 is not correlated
to y^i , will still be correctly estimated though Equation(2.10) misspeci…es the underlying
process.

Equivalent Concepts When Similar Steady-States

Conditional and absolute convergence have the same de…nition for country A and B if both
converge to the same steady-state. This will be the case if both have equal technologies

This content downloaded from

102.128.76.42 on Thu, 08 Dec 2022 11:33:14 UTC
All use subject to https://about.jstor.org/terms
 14

and preferences. In this case, when estimating Equation(2.10) the constant a will be:

(1 e T)
a = xi + log [^
yi ] (2.11)
T

and is consistent.

2.1.7 vs Convergence

The two concepts studied above refer to -convergence. The s from Equation(2.10) and
Equation(2.9) try to measure the mobility of income within the same distribution. On the
other hand -convergence studies how the distribution of income varies over time.
-convergence analyzes the dispersion of income of diverse economies and convergence
occurs if the dispersion is diminishing over time. Usually it can be measured as the standard
deviation of the logarithm of income per capita across di¤erent economies.
It can be seen from Appendix A, that when there is divergence it does not necessarily
lead to -convergence. But in order to have -convergence it is necessary to have -
convergence. Therefore, -convergence is a necessary but not su¢ cient condition for -
convergence.

2.2 Other Models

2.2.1 Endogenous Growth Models

The endogenous models di¤er from the NGM in that they do not assume diminishing re-
turns to capital. For example the one sector AK model is based on the following production
function:
Y = AK;

which can be compared to a Cobb-Douglass when the share of capital is 1, = 1. This

can be interpreted as a taking a broad de…nition of capital that includes human capital.
This type of production function violates the Inada conditions and does not exhibit the
diminishing returns to capital. Replacing = 1 in Equation(2.8), leads to = 0. So it
does not matter if a country is poor or rich, the poor will never catch up to the rich.

This content downloaded from

102.128.76.42 on Thu, 08 Dec 2022 11:33:14 UTC
All use subject to https://about.jstor.org/terms
 15

In general, the one sector models of endogenous growth fail to predict any kind of con-
vergence, even when perfect or partial capital mobility is introduced. However variations in
the AK model can explain convergence. That is the case when the saving rate, the depre-
ciation rate and/or the population growth rate are allowed to be endogenous (determined
by the level of capital in the economy). But as pointed out by Sala-i-Martin (1996a), these
variations are often seen as implausible, not deeply explored or fail to explain convergence.
In the case of two sector models, Mulligan and Sala-i-Martin (1992) have showed that
these models are able to predict conditional convergence in the same way that the NGM
does.

2.2.2 Models of Distribution Dynamics

In the literature, this type of models is also referred to as polarization, persistent poverty (or
poverty traps), strati…cation, and/or clustering models (Quah,1996; Quah, 1997; Azariadis
and Stachurski, 2005; Durlauf and Johnson, 1995).
In the models of distribution dynamics, each economy has multiple locally-stable steady-
states. Transitory shocks may have permanent e¤ects whereas the NGM assume that
economies have a unique, globally stable steady-state equilibrium and that the transitory
shocks a¤ect the income ranking of an economy in the short run but do not have lasting
e¤ects (Galor, 1996)
The idea of the models of distribution dynamics can be explained by …gure 2 1, which
shows hypothetical income distribution across countries. The GDP per capita is showed
on the vertical axis and time on the horizontal. At time t0 the income distribution has
one peak, while later, say at t1 there are two peaks: one for countries with high levels of
income and another for low. Countries with middle income levels have disappeared. Some
went to the low income group and others to the high income group. At least it is clear that
the ones that were at t0 among the richest ended up in the top group in t1 and viceversa.
This process of changing distribution in income into two is called "emerging twin peaks".
This means that at t1 , countries converge to two di¤erent groups, even though they were

This content downloaded from

102.128.76.42 on Thu, 08 Dec 2022 11:33:14 UTC
All use subject to https://about.jstor.org/terms
 16

t0 t1

Figure 2 1: Distribution Dynamics. Hypothetical income distributions

in two periods. One-peak income distribution in period t0 and two-peak income
distribution in period t1 . Taken from Quah (2007).

at t0 in the same group. The number of groups at t1 can also be more than two, in which
case the process is called "strati…cation" and each group is called a "club".
From this analysis, the concept of "Club Convergence" arises. Club convergence among
a group of countries exists, when they have a similar steady-state and when they depart
from similar initial conditions (Galor, 1996). Quah (1996) emphasizes that what matters
most is how a single economy performs relative to others, rather than to its own history.
According to Galor (1996) the NGM can generate both the conditional and the club
convergence concept, particularly the neoclassical model of overlapping generations. He
a¢ rms that in practice both concepts can be estimated by the inclusion of empirical sig-
ni…cant variables like human capital, income distribution and fertility in the NGM, along
with capital imperfections, externalities, and non convexities.

This content downloaded from

102.128.76.42 on Thu, 08 Dec 2022 11:33:14 UTC
All use subject to https://about.jstor.org/terms
 17

2.2.3 Technology Di¤usion Models

These models predict convergence based on microfoundations. It is argued that poor

countries will "catch-up" slowly to the rich ones (leaders), because the followers can imitate
the products invented by the leaders. This idea was introduced by Nelson (1966), where
it was stated that the technological progress for a country is a function of the distance
between its level of technology and the world leader:

A_ i
= (Aleader Ai )
Ai

By assuming that imitation and implementation costs are lower than innovation costs,
when very little has yet been copied, the lagging countries can imitate and converge gradu-
ally to the leaders level of technology (Barro and Sala-i-Martin, 2004).When more products
are being copied, the imitation and implementation costs will rise. Therefore convergence
is created by a diminishing returns to imitation.
In practice, these models can be estimated in a similar way as the NGM. Once conver-
gence is found, it is hard to distinguish whether the reason of its existence is the neoclassical
hypothesis of diminishing returns to capital or the hypothesis of positive (but slow) rates of
technological di¤usion unless one can model how technology evolves in the leading countries
(Barro and Sala-i-Martin, 2004).
The process of technology di¤usion is still in debate. Some researchers argue that
the technology di¤usion is via foreign investments (Barro and Sala-i-Martin, 2004), trade
(Romer,1990; Aghion and Howitt, 1992), ‡ows of people (Barnebeck and Dalgaard, 2006)
or the type of institutions and geography (Acemoglu et al., 2004).

2.3 Theory of Integration

Economic integration among countries is linked to economic convergence. The theory of

economic integration studies the creation of a common market as a process that goes
together with economic growth. The deepening of this process tends to be deepened, via

This content downloaded from

102.128.76.42 on Thu, 08 Dec 2022 11:33:14 UTC
All use subject to https://about.jstor.org/terms
 18

monetary and political integration, coupled with growth is related directly to the idea of
convergence among countries and regions (Sotelsek, 2001).
However, the process of integration can have obstacles that delay the process of con-
vergence. One obstacle is that the integration process may have a di¤erent impact in
certain economic sectors and regions. Another is the possible absence of considerable ef-
forts of harmonization of institutions and policies. These obstacles postpone one of the
main objectives of the integration process, which is to increase the standards of living of its
population and diminish the existent disparities in the standards of living of their citizens,
(Sotelsek, 2001).
Conceptually, economic convergence can be de…ned broader than discussed so far
as a process, spontaneous or with some intention, that leaves economies in a similar degree
of development. Economic integration can be de…ned as a process that is intended
to follow speci…c objectives and rules previously agreed to by its members that lead to
economic convergence.
Alternatively, full economic convergence is the last resulting phase of economic integra-
tion. The integration process itself has three phases, (Heirman, 2001):

1. Convergence of basic instruments like common external tari¤s and commercial regu-
lation among its members and to other countries

2. Convergence in macroeconomic, …scal and social policies

3. Convergence in real terms

These phases are connected to the stages of economic integration, which total six, and
for every two stages, one phase of convergence is reached. The stages are:

1. Preferential trading area

2. Free trade area

3. Customs Union

This content downloaded from

102.128.76.42 on Thu, 08 Dec 2022 11:33:14 UTC
All use subject to https://about.jstor.org/terms
 19

4. Common market

5. Economic and Monetary Union and

6. Complete economic integration.

Currently, the only integration process that is in the last phase of convergence and last
stage of economic integration is the European Union. Nevertheless, Walz (1999) tested
the hypothesis that integration has promoted economic convergence among the European
countries and found that it is rejected. It seems that the economic convergence created by
the European Union is across regions rather than countries.
In LA the most advanced economic integration processes are still in the third stage
of economic integration: custom unions. Consequently, they are in the second stage of
convergence; in macroeconomics, …scal and social policies. Still, this study will test for
convergence in real terms among countries in each of the custom unions in Chapter 5.

2.4 Summary

This chapter has explained the main characteristics of the neoclassical growth model used
in Barro and Sala-i-Martin (2004), which is the base model for further estimations, and
has presented other important theories about convergence.
The theory of convergence, in general, has de…ned four concepts of convergence: ab-
solute , conditional , and catch up convergence. Absolute -convergence exists when
per capita incomes of a number of economies converge to one another in the long run,
independently of their initial conditions. Conditional -convergence exists when per capita
incomes of economies that have identical their structural characteristics (e.g. preferences,
technologies, rates of population growth etc.) converge to one another in the long run
independently of their initial conditions. Club convergence is conditional convergence
conditioned also on having similar initial conditions. -convergence across a group of
economies exists if the dispersion of their real per capita GDP tends to decrease over time.

This content downloaded from

102.128.76.42 on Thu, 08 Dec 2022 11:33:14 UTC
All use subject to https://about.jstor.org/terms
 20

Finally the concept of catching up, means that poor economies will reach the rich economies
when time passes, which is the same concept as absolute convergence.
Once convergence is found, under the NGM, the theoretical reasons can be two: di-
minishing returns to capital and/or lower costs of technology imitation than innovation.
Additionally, convergence can also be due to a result of an intended process, in the case of
integration agreements. However economic convergence is expected only in the last stage
of the integration process.

This content downloaded from

102.128.76.42 on Thu, 08 Dec 2022 11:33:14 UTC
All use subject to https://about.jstor.org/terms