,„MIT
LIBRARIES
3 9080 01917 6582
vm
iSiMliiiliiil
Digitized by the Internet Archive
in
2011 with funding from
Boston Library Consortium IVIember Libraries
http://www.archive.org/details/networksmigratioOObane
UtVVCT
.M415
0$
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
NETWORKS, MIGRATION
AND
AND INVESTMENT:
INSIDERS
OUTSIDERS IN TIRUPUR'S PRODUCTION CLUSTER
Abhijit Banerjee,
MIT
Kaivan Munshi, Univ. of Pennsylvania
Working Paper 00-08
June 2000
Room
E52-251
50 Memorial Drive
Cambridge, MA 02142
This paper can be downloaded without charge from the
Social Science Research Network
Paper Collection
http://papers.ssm.coin/paper.taf?abstract
id=XXXXXX
at
MSSACHUSEHS
INSTITUTE
OF TECHNOLOGY
OCT
2 5 2000
LIBRARIES
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
NETWORKS, MIGRATION
AND
AND
INVESTMENT: INSIDERS
OUTSIDERS IN TIRUPUR'S PRODUCTION CLUSTER
Abhijit Banerjee,
MIT
Kaivan Munshi, Univ. of Pennsylvania
Working Paper 00-08
June 2000
Room
E52-251
50 Memorial Drive
Cambridge, MA 02142
This paper can be downloaded witiiout ciiarge from the
Social Science Research Network Paper Collection at
http://papers.ssm.com/paper.taf? abstract id=XXXXXX
Networks, Migration and Investment:
Insiders
and Outsiders
in Tirupur's Production Cluster
Abhijit Banerjee^
*
Kaivan Munshi*
March 2000
Abstract
in
This paper studies the effects of social network based lending. This is a pervasive phenomenon
Access to such network capital has an obvious influence on
most of the developing world.
would prefer
show that
under reasonable conditions such lending will generate a rather specific pattern of migration and
investment. In particular, migrants to locations where they do not have access to their communty's
lending networks will tend to have higher ability than the traditional residents of that location, but
will invest less relative to their abihty. Under some conditions this generates the possibility that
migrants have higher ability but invest less in absolute terms than the local people. We test this
impUcation using data from the knitted garment industry in the South Indian town of Tirupur.
Comparing the growth rate of output (which, we argue, proxies well for ability) with investment
between garment firms owned by migrants to Tirupur and local people, we find that local people
have slower output growth but invest substantially more at all levels of experience. We also find a
positive correlation between investment and growth within any single community, consistent with
the view that capital access does not vary within each group.
investment.
It also
influences the pattern of migration since, ceteris paribus, migrants
to be in locations where they have access to their community's lending network.
We
'This project could not have been completed without the support and assistance that we received from the Export
(ECGC) and the PSG Institute of Management. Professor A. Govindan and T.
Sivan organized the survey and supervised the data collection. We thank Susan Athey, Esther Duflo, Anjini Kochar,
Credit Guarantee Corporation of India
J.
Nick Souleles and Petra Todd for helpful comments. Munshi's research was funded by
responsible for any errors that may remain.
^Massachusetts Institute of Technology
^University of Pennsylvania
NIH
grant R01-HD37841.
We
are
Introduction
1
There
a growing awareness that social networks play an important role in facilitating economic
is
when markets
activity
that this
is
function imperfectly, particularly in the developing world.
is
usually argued
because long-term relationships, the possibihty of social sanctions, and the smooth flow
make
of information within the network
view
It is
possible transactions that
would not otherwise happen. This
supported by a ntmiber of studies, including Greif (1993), Udry (1994), and Townsend (1994)
in the context of capital markets.
In this paper
we take a more macro approach:
we focus on the
these networks
eflFects
of networks
instead of looking at the internal functioning of
on the functioning of the wider economy. In
this
we
follow a growing recent literature which emphasizes the possible distortions that arise from a reliance
on networks.^ To the best of our knowledge however there are no papers that
these distortions can be of a magnitude that would
attempted here
make them worth taking
tell
us whether or not
seriously.
we
will
argue there
is
clear evidence of a large deviation
At the heart of the case we make
is
more
what
is
from the
70%
of India's
first-best allocation.
the following set of observations. First, in a first-best world,
under the standard assumption of complementarity between
will invest
is
Using panel data that we gathered from the
for the specific case of lending networks.'^
knitted garment industry in Tirupur, a town in Southern India that produces about
exports,
This
ability
and
capital, higher ability people
Second, lending networks tend to be local - in other words, they
in their firms.^
provide privileged credit access to their
members but only
if
their
members
locate where the network
has control. Therefore, in order to take fuU advantage of the lending network, one might have to invest
where one's network
is
established, rather than
choose to invest in an area where their network
best suited to their
skills,
at the cost of being
observation can be rephrased as follows:
do so because
this allows
them
to
all
where one
is
weak,
will
more limited
those
who
is
most productive. Conversely, those who
presumably benefit from being
in the place
in their access to credit. Third, the previous
choose to invest where the network
do what they are good
at,
while
some
of those
who
is
weak,
invest
where
'See for example Greif (1993), Banerjee (1996), Kranton (1996), Banerjee and Newman (1998), Kranton and Minehart
and Jackson and Wolinsky (1996). Many of these papers are more after the subtle (though perhaps equally
important) distortions that arise out of the process of formation of networks or the pecuniary externalities arising from
(1998, 1999),
the effect of networks on market prices.
Lending networks are networks whose main economic function is to facilitate credit market transactions. A reliance
on network-based lending is a ubiquitous feature of Indian corporate finance, particularly in the small-business sector,
and we will argue later that Tirupur is no exception to this rule.
'This observation is standard and goes back to Lucas (1978).
their
those
network
who
strong do so despite being more productive at some other location. This implies that
is
where
invest
their
network
weak are
is
likely to be,
being better matched to their chosen occupation) than those
Finally, the previous observation tells us that migrants in
network
same
is
yet to be established, are likely to be
reason, they
may
also invest less
with non-migrants, in such a location,
ability will also
first-best case:
than the
it is
be the group that invests
it
will
on average, more able
who
invest
(in
the sense of
where their network
is
strong.
a new location, where their community's
more able than the
local people.
local people.
However,
for the
In other words, comparing migrants
possible that the group (the migrants) with higher average
less.
This
only be observed in the data
to outweigh the standard complementarity effect.
if
is
we would
the opposite of what
find in the
network-generated distortions are large enough
It is this
prediction of network-based lending that
forms the basis of our empirical analysis.
Most of the observations that we provided above are in themselves well-known and widely accepted.
For example,
to their
it is
generally recognized that people choose to migrate to places where they have access
community
networks.'*
The
fact that
a lack of social networks limits out-migration has also
been written about. ^ Finally, the fact that migrants have higher
ability
than the locals
place in the migration literature. Borjas (1987) calls this "positive selection".^
is
common-
Our framework
unifies
these observations and uses their joint implications to provide empirical insights into the nature of
networks and their influence on migration and investment.
The reason
for choosing
Tirupur as the venue
for the empirical
work
in this
paper
is
that
we can
take advantage of a recent change in the sociological composition of Tirupur's production cluster.
Until the late 1980s Tirupur was dominated by the Gounders,
from the
area.
However, in the
last
(in the
are traditionally agriculturalists
decade a number of people from
entered the Tirupur industry, attracted by
time when we observed them
who
its
all
over the rest of India have
success as an export center.
mid 1990s) the Gounders had access
We
will
argue that at the
to well-established lending in
Tirupur but the newcomers from outside probably had much more limited access to capital in Tirupur
^See Piore (1979) and Massey, Alarcon, Durand and Gonzalez (1987) for studies of Latin American migrants to the
U.S. that argue that migration tends to flow to areas where past migrants have already established a foothold. Carrington,
Detragiache and Vishwanath (1996) describe the support networks that were put in place for the later arrivals from the
American South during the Great Migration. Similarly, Timberg (1978) describes how social ties framed the expansion
of the Marwari community into specific cities in nineteenth-century India.
^Dasgupta (1987), among others, makes this point.
Borjas derives conditions under which a model where migration is constrained by the physical cost of moving will
generate positive selection. Our theory of positive selection differs from his in linking it to the distortionary effects of
networks.
than they would have had in their home basest
It is this difference in
the access to credit networks,
between insiders and outsiders, that we exploit in the empirical analysis.
The data shows
siders,
that Gounders start their business with substantially
and use substantially more capital per unit of production
with the view that they have better access to capital.
for the Outsiders
than
for the
more
capital than the Out-
at every level of experience, consistent
We also find that
Gounders.^ Starting with lower output
output trajectories are steeper
the Outsiders ultimately
levels,
surpass the Gounders after about five years of export experience, yet they continue to maintain lower
levels of capital stock.
The
fact that they invest less,
fact that
we take
output grows faster
for Outsiders
than
for
Gounders despite the
as evidence that Outsiders have higher ability: this
our theory predicts.^ Thus outsiders with higher ability invest
less
than the
certainly
is
what
which suggests
insiders,
in turn that there are substantial distortions in this industry.
The plan
of the paper
gration/location choice.
is
The
as follows:
first
Section 2 develops a model of lending networks and mi-
goal of this section
is
to identify conditions under
networks lead to positive selection. While our analysis shows that
ditions under which
we
selection given above
is
get positive selection,
incomplete - there are
The second
positive selection.
it
it is
possible to find simple con-
also points out that the simple intuition for positive
many
goal of this section
is
quite convincing situations where
to identify conditions under
higher ability by comparing the growth trajectories of firms in the industry.
empirical work
is
described next in Section
regression results.
3.
which lending
we may not
get
which we can identify
The data used
for the
Section 4 discusses identification issues and presents the
Here we also present some evidence supporting the complementarity assumption
that underlies our empirical work. This section ends with a discussion of other possible interpretations of our results. Section 5 concludes the paper with a discussion
results.
We
on the poUcy implications of our
recognize that demonstrating that there are distortions does not automatically establish
the feasibility of something better, and certainly does not of
itself
make a
case for government inter-
^This assumption is supported by the historical and case-study evidence, mentioned above, which suggests that
communities tend to expand quite slowly into new locations.
*The fact that migrants have steeper earnings profiles is standard in the migration literature. Borjas (1987) calls
this one of the most convincing findings of the empirical literature on migration and cites studies by Chiswick (1978),
Carliner (1980) and De Preitas (1980). It is worth emphasizing, however, that these studies mainly focus on migration
from poor to rich countries. In such cases, there is a natural element of catch-up - the new migrants get educational and
other opportunities that they have never had. For example, migrants to the U.S. learn English, which makes them more
employable. This kind of catch-up is less important with internal migration, and our results are therefore more likely to
be due to positive selection. What is striking about our case is that the steeper output profiles are associated with less
investment.
^But
see the discussion in sub-section 2.7.
we argue that
vention. However,
may be a
in this specific context there
strong case for certain types
of government interventions in the financial sector.
A
2
Simple Model of Location Choice and Investment
we formally develop the model suggested
In this section,
that
we make
will
be inspired by the way the garment industry
We
be to use the model to interpret data from Tirupur.
will
in the introduction.
first
a given interest rate and level of
and investment decision
for
location choice decision
when networks
are present.
Tirupur functions, as our goal
in
characterize the firm's production
We
ability.
then go on to analyze the
we bring the investment
Finally,
location choice together to derive the dynamics of output
Specific assumptions
and investment
for insiders
decision
and
and outsiders
in
the industry.
2.1
Production and Investment
There
is
a large population of firms producing the same good.
price of the
lose
good
is 1.
We
adopt the normalization that the
Firms at any point of time have a stock of regular buyers. In each period firms
a fraction of this stock but acquire a random fraction of new buyers as
well.
writing the following equation for Xt+i, which should be thought of as the
firm
sure to get in period
is
*
+1
This
amount
is
captured by
of sales that the
:
Xt+i=fXtXt{l+et).
We
and
think of
4>{et) is
1.
itself is
€f
as a
random shock
the distribution of
et
distributed on
and
in period
t.
Then
is
this equation says that the
time which seems to
fit
Our impression from
and
stick to
them
minimum
way
of
if
e
and ^( as a number between
in all locations, while the shock
same person
we think
it is
also
assumed to be
of Xt(l+et) as the reahzed sales
possible sales in the next period
modeUng
sales builds in
a
is
a fraction
lot of persistence over
the anecdotal evidence about the knitted garment export industry in Tirupur.
talking to industry experts
unless they have a
'"We can assume that
for the
easy to interpret
of the current period's actual sales. This
with mean
assumed to be the same
assumed to be independent across people, and
independent over time.^° This equation
/i(
is
[0, e*]
(/>(•) is
is
that foreign buyers typically pick their exporters
bad experience.
the same for everyone because any fixed difference in the
to a difference in the ability a, which
is
introduced in the discussion below.
<^(-) is
completely equivalent
Capital
of period
is
t
is
the only input used for production. Capital stock for period
(when Xt
is
chosen at the beginning
t is
known but not Xt(H-ef)). Denote it by Kt. The technology used
for
production
described by the equation
where a
is
the abihty of the entrepreneur in that particular industry.
oo>B>^>B_>0 and
Qrj.ftx'i.
more
capital per unit of sales retains
more
capitalized firm as being
allows better quahty control
better
skills. ^^
The
Finally,
firm maximizes
more
0.
more
>
assume that -gjAx
0,
This equation represents the idea that a firm that uses
This
of its buyers.
vertically integrated.
and increases buyer
we assume
its
<
We
is
easy to interpret
if
we think
of a
We are then saying that vertical integration
retention,
and so does having an entrepreneur with
that there are diminishing returns to capital.
expected discounted
V{Xt)
profits,
which can be written
as:
= maxE f2s'[Xt{l+et)-rKt]
t=i
under the assumption that the owner of the firm
r
risk-neutral
the interest rate that applies to the firm, which
is
takes as given
Xi
,
the assured
demand
is
is
credit rationed or faces
an increasing
and maximizes
profits
specific to the firm.
standard) way of modeling a capital market imperfection:
is
all
t.
It follows
we could have
This
is
"For
instance,
That
is
my
Then
I
will
have
all
as
a specific (though
alternatively
assumed that
interest rate schedule.
^
V
will
double
if
we
XtV{l,r,a).
Cawthorne (1995) quotes one of the Tirupur exporters
ambition.
much
that
ViXt,r,a)
here.
and
under the constraint
Observe from the structure of the maximization problem that the value of
simply double Xt and Kt for
at rate 6,
the assumption that the firm can borrow as
wants at an interest rate of r which, however, may be
the firm
and discounts the future
assumed to be constant over time. The firm
in the first period,
Imphcit in this way of writing the problem
it
is
as saying, "I
want to be
the stages (of production) as one operation.
you know exactly what is going on." While Cawthorne does not share
spoke with expressed the same sentiment.
his view, a large
.
.
like
.
number
the spinning mills
With a
large factory
of exporters that
we
.
It is
evident that V{l,r,a) must be increasing in
differentiating the expression defining
Using
this
decreasing in r (this can be verified by
V{l,r,a) and applying the envelope theorem).
decomposition of V{Xt,r,a), we can write the entrepreneur's maximand as
+ et) - rKt] + 6Xt+iV{l,r,a)}
E{[Xt{l
We
a and
maximize the maximand subject to the constraint above. Substituting
in the expression for
Xt+i
from the constraint equation, we can write the maximand as
+ et)-r^^+6f^{^^-^^ya){l + et)V{l,r,a)]}
E{Xt[{l
Writing
^=
2;^,
we
write this as
Xt[{l+-e)-rzt
where
=
Ji{zt,a)
Maximizing
E^i{j^^,a){l
this
+ ^{zt,a)V{l,r,a)]
+ et).
with respect to
zt gives
r
^
us
^^{zt,a)V{l,r,a).
This determines z*{a,r), the optimal capital-output ratio for this particular firm.^^ Note that z*
constant over time.
It follows
The
effect of
an increase in a both
raises JJ^
in r leads to fall in z*.
since
from the second-order condition
may be
either positive or negative
dominate the negative
Note that
effect
on
an increase
in
a on
z*,
maximization that an increase
for this
when
JI^^
>
and V{1) (higher a means higher
when
Jl^^
<
0,
is
is
unambiguously
lifetime income).-'^
since the positive effect
positive,
The
effect
on V(l) may or may not
fl^
this is telling us that ability
must be complements as long as
Jiaz is
and capital can be complements even
small in absolute value relative to
72^.
if Ji^^
This
is
<
and indeed
because higher
abihty people have a brighter future and therefore are willing to invest more in building up their
customer base.
Strictly this
'
We
is
will find it
not the capital-output ratio but rather the ratio of capital to the guaranteed amount of output.
convenient to state our conditions in terms of derivatives of the Ji function but it is worth noting that
these derivatives inherit the sign of the corresponding derivatives of the
fj.
function.
Locations and Populations
2.2
We now
extend the model to allow
each with
This
is
its
his
own
for location choice.
Each person
industry.
community. Each person
abihty in industry
1
and a^
We
economy
a'^) i
—
assume that there two
is
1,2,
which
is
The population
defined
The consequences
and
in
where a^ represents
each community
on the unit square and
assume that the two populations are exactly
for all pairs (0^,0^).
locations, 1
2,
associated with one of the two locations.
also associated with a pair {a^,a'^),
his ability in industry 2.
a distribution function ^*(a\
no mass points.
is
in the
We
identical,
of allowing the populations to
i.e.,
be
is
is
his
described by
assumed to have
^^(a^, a^)
different are
=
^'^{a^,a'^)
examined
in
sub-section 2.5.
We
will
assume that the two locations are equivalent
their capital
markets function.
the two locations,
equal to
1 in
Xl and Xi,
Specifically,
are equal.
we assume that the
We
perhaps in the way
in all respects except
initial
guaranteed level of demand in
have already set the price of the goods produced to be
both locations.
These assumptions have the immediate impUcation that
in a first-best world,
where everybody
faces
the same interest rate irrespective of location and community, people will simply choose the location
where
makes
their ability
it
(i.e.,
their a)
is
higher. This particularly simple structure of the first-best
outcome
The impHcations
easy to identify the distortions generated by network-based lending.
of
relaxing these assumptions are discussed in sub-section 2.5.
Lending Networks
2.3
We
have in mind a very simple model of lending networks.
Enforcing credit contracts
is
costly.
easier to enforce contracts
when the borrower
belongs to the same community network as the lender. However, this only works
when the borrower
Networks can lower the cost of lending because
is
located where the network
is
at hand. In other words, there
it
is
based: the network only has enforcement power
is
when the borrower
is
no advantage to lending to a community member who has located
away from the network. ^^
'
This says basically that networks are
spread
all
local,
which
is
consistent with the historical evidence.
While the Marwaris
over India during the nineteenth century, a careful reading of their migration patterns reveals that each caste,
within the Marwaris, located in only a few locations (Timberg, 1978).
For example, the Shekhavati Aggarwals were
dominant in Malwa, and had a substantial presence in Calcutta, Assam and the Indo-Gangetic plain. In contrast,
the Oswals dominated the Bombay Deccan, the Calcutta jute industry and had a significant presence in Bangalore,
Hyderabad and Tamil Nadu. Similarly, while the Nakarattars engaged in business throughout Southern India and
Southeast Asia, their internal capital markets had a local aspect as well. For example, Rudner (1994) describes how the
rate on community capital in Rangoon was set at a council meeting, once a week at a fixed time in the local temple.
The combination
population
borrows at location
1
he to locate in location
in location
and
=
rf
We make
1.
rl-
by assuming that when a borrower from
of these assumptions can be captured
1,
he pays rj, which
than
is less
r^,
which
what he would pay were
is
r2
>
r^
the assumption that the locations are symmetrical in the sense that rj
=
rj
2.
Likewise, a borrower from population 2 pays r^ in location 2
The consequences
of not
making
this
and
assumption are discussed in sub-section
2.5.
Location Choice
2.4
Before they choose
how much
model have to decide where to
to invest, people in our
who was born
focus on the decision of someone with ability vector (0^,0^),
locate.
in location 1
us to dispense with the marker for the location of birth. Strictly, this comparison can only
the movers and stayers from location 1 while
we
compare with those who move
from location
to location 1
how
are really interested in
We
will
which allows
tell
us about
the stayers in location
1
However, under the assumption, made
2.
above, that the two locations are perfectly symmetrical, the two comparisons coincide.
When
two
individuals choose their location they
locations, but
do not have
know
own
their
their actual orders for period
abilities
All they
1.
output in the two locations are given by the random variables Xl{l
are
Xf
are
known (and equal
and the
know
+ e\)
is
interest rates in the
that their first-period
and X^(l
+ e^),
where
Xl
to each other), but e\ and ef are only revealed after location decisions
have been made and capital stocks have been chosen.
In order to understand the location decision
in the
two locations.
We
will, for
we need
compare
to
function.
The consequences
same person
the time being, assume that in either location the person would
choose to participate in the local industry. This would be true
fi{-)
lifetime payoffs for the
an Inada-like condition holds
if
for the
and therefore taking the participation
of relaxing this assumption
decision seriously are briefly discussed in a later sub-section.
Denote by
V the individual's expected lifetime payoff
if
he were to choose location
assume that the production technology does not vary across
person
who
is
marginal between going to 2 and staying in
V{Xl,r^,a^), which, from above, implies X^V{l,r^,a^)
This implicitly defines a function a^
someone with attributes of
(a^,
o:'^)
=
chooses
1,
it
tells
=
Because we
F(X|,r%a*). For the
must be true that V{Xi,r'^,a'^)
=
or
us the lowest value of a^ for which
Since V{l,r^,a'')
8
V^
= XlV{l,r^,a^),
f{a^), which
2.
industries,
i.
is
increasing in
«%
it
must be the
case that /(•)
is
an increasing function.
In a first-best world, since r^
those
who move and
those
In the case where
for this case
industry
2,
is
r"^
who
>
=
r^,
f{a^)
=
As a
a^.
result,
stay will be identical. There
F(l,r%a*)
r^, since
represented in Figure
1
is
is
no positive
selection.
decreasing in r\ f{a^)
>
The /()
a^.
by the curve AB. Those who are above the
while the rest choose industry
among
the distribution of abiUties
function
/(•) curve
choose
1.
Insert Figure 1 here.
evident from Figure
It is
people
who have low
1
that the people
ability along
who
both dimensions,
are in the left-bottom corner of the square,
all
remain in their home location. This
essence of the intuitive argument for positive selection offered in the introduction - those
do so because they are
those
who
extreme
(as described
cases.
to migrate, while
we
not enough to guarantee positive selection, except, as
will
need to impose restrictions on the distribution of
by the distribution function ^^(a^,a^)), as well as the shape of the
To understand the
The
In general,
it is
the
who migrate
stay back do not have to be particularly good at the local occupation. And, indeed,
force for positive selection. However,
see, in
them
significantly better at the occupation that requires
is
i.e.,
it is
we
a
will
abilities
/(•) curve.^^
role of the distribution of abilities, consider the case represented in Figure 2.
distribution of abilities in this case has a substantial degree of concentration in the top right
corner of the unit square, suggesting that the correlation of abilities
people.
From the
position of the /(•) curve,
high ability wiU stay back in industry
1.
it is
the highest for high ability
is
clear that in this scenario
The average abihty
most of the people with
of movers will clearly be lower than that
of the stayers.
Insert Figure 2 here.
Figure Sexplains
is
why
the /(•) curve matters. In this case,
we assume
that the distribution function
uniform over the unit square, thereby eliminating any correlation between the two types of
The
/(•)
that
it is
area
ACBQ. The
curve as
Strictly,
we have drawn
it
(ACB)
almost coincides with the diagonal until a^
almost vertical. The average ability of a mover in this case
average ability of a stayer
the /() curve only
the stayers compare with those
tells
is
is
to location 1
from location
locations are perfectly symmetrical, the two comparisons coincide.
2.
1
0.5, after
the average of a^ over the
the average of a^ over the area
us about the movers and stayers from location
who move
—
ability.
PACBRS which is the
while
we
are interested in
how
However, under the assumption that the two
average of the averages over
CB
the diagonal and
average of a^ over
over
PACDS
PACDS.
However, the average of a^
coincides with
ACBQ essentially the same as the
over CBRD
clearly higher than its average
is
is
movers
follows that the average ability of the
it
AC almost
Note that since the hne
almost vertical, the average of a^ over
is
and therefore
than that of the
PACDS and CBRD.
is
unambiguously lower
stayers.
Insert Figure 3 here.
One
case where the /(•) curve will have the shape
either industry
do not use any
has in Figure 3
2 that
no one uses
result, productivity
it;
from their higher
fully
whereas in location
a higher rate of
1,
grows much faster with
vertical /(•) curve. Intuitively, the
complementarity.
As a
result,
interest,
where low
ability people in
is
irrelevant
location where they have higher ability - the /(•)
curve therefore coincides with the 45° line in this range.
need capital to benefit
is
capital. In this case, the difference in the price of capital
They simply choose the
for this set of people.
it
abilities.
capital
is
High
on the other hand,
ability people,
However, capital
is
so expensive in location
cheap, which helps high ability people.
ability in location 1
problem here comes from the
than in location
fact that there
is
2,
As a
generating a nearly
strong ability-capital
high ability people are severely penalized for moving to a location with
which naturally reduces the share of high
ability people
among
the movers.
In the rest of this sub-section we develop two sets of sufficient conditions which guarantee that the
movers from a certain location
will
have higher abiUty than those
who
stay there. Because
interested in comparing average productivity rather than average abiUty, our goal will be to
the distribution of abilities of migrants
abilities
location
those
among
1,
of those
who
first
/(•)
will
be
show that
distribution of
stay back. While our formal arguments apply to movers and stayers from
to location 1 with those
who
terms of the comparison of migrants and
The
(FOSD) the
because the two locations are perfectly symmetric the result also holds
who move
directly to
order stochastically dominates
we
stay back in location
local people at the
same
1.
We
when we compare
therefore state our results in
location, as this corresponds
more
our empirical analysis.
intuition for the first set of conditions
is
very simple: increasing r^, keeping r^ fixed, moves the
curve up. As a result, for a high enough value of r'^ (with r^ fixed), the /(•) curve will pass through
the top-left-hand corner of the unit square {A'B' in Figure
of a^
among
those
who choose
This actually assumes that the
to
/(•)
move
to 2
is
1).
When
this
entirely concentrated at
curve does not become vertical.
10
The reader can
a^
happens, the distribution
=
verify
1.^^
By
by looking
contrast, since
at the expression
almost no one moves, the distribution of a^ among those
identical to the ex ante distribution in the population
a^ that
Clzdm
is
1
choose to stay in industry
1 is
almost
the marginal distribution over values of
(i.e.,
generated by the joint distribution, ^{a^jO^)). This argument implies:^^
For a
fixed value of r^ , there exists a high
ability in industry 1
among migrants
same industry among
the
who
first
enough value of r^ such that the distribution of
order stochastically dominates the distribution of ability in
local people.
This result formalizes the argument for positive selection that was suggested in the introduction.
When
The
the interest rate differential
rest,
including everyone
is
very large, only those
who has medium
who have
very high abihty will migrate.
or low abiUty in both occupations, will stay back,
depressing the average abiUty of stayers relative to migrants.
A
limitation of this result
The next
that
it
has precise predictions only about relatively extreme cases.
result formalizes the intuition, implied
becomes very steep as
cases where the /() curve
first
is
by the example in Figure
ability increases.
shows that under the conditions given below, the
Une and then uses
this property to
Claim 2 Assume
that djl{z,a)/da
respect to z
greater than 1
is
Then
unit square.
Note that while
Section 2.1
it
to avoid
proof, given in the Appendix,
everywhere
flatter
than the 45°
^^ {a^ a^)
,
and^ {a^,Q^)
represent uniform distributions on the
among migrants
same industry among
the condition that djl{z, a) /da
is
first
order stochastically
local people.
a constant
is
there to limit the
amount
and investment which helps us avoid a case such as the one
this condition obviously
imphes that d'^Jl{z,a)/dadz
=
0, as
in
pointed out in
does not necessarily imply the absence of complementarity between ability and capital.
The assumption
of uniform distribution of abilities avoids the case in Figure
that djl{z,a)/dz
inelastic.
is
we want
a constant, that the elasticity of the function dji{z,a)/dz with
ability in the
of complementarity between ability
3.
is
the distribution of ability in industry 1
Among these assumptions,
curve
that
prove the stated result.
and that
dominates the distribution of
Figure
/(•)
The
3,
This
is
is
elastic
to
make
enough with respect to
2,
2. Finally,
directly implies that the
sure that the higher interest rate at location 2
is
the assumption
demand
for capital
is
sufficiently costly for the
Appendix that under the assumption that dEn{z^/{l + et),a^)/da is bounded above and below
(which has already been assumed), the /(•) curve remains bounded away from becoming vertical.
"^^The proof is obvious and is omitted.
for /'(•) derived in the
11
movers. Otherwise the selection effect
may
not be strong enough to guarantee
first
order stochastic
It is
therefore worth
dominance.
Discussion
2.5
We
imposed a rather lengthy
of conditions in order to get this result.
list
emphasizing that the conditions are more stringent than we actually need. For example, we do not
need the property that / ()
our current proof does
2 that in the case
is
everywhere
make use
where
/'(•)
=
of this property.
1 (i.e.,
2 continues to hold as long as r^
positive,
but
is
that as long as
flatter
>
the /
By
r-^.
(•)
than the 45°
r-^
>
r^,
prove Claim
2,
though
can be directly checked from the proof of Claim
It
curve
is
continuity,
bounded above (once again as long
line in order to
as
a straight line parallel to the 45°
it
also holds
>
r"^
r^).
when
/'(•)
—
1 is
As a consequence,
it
line).
Claim
allowed to be
can be shown
each of the other conditions of Claim 2 can be relaxed without changing the
result.
Prom the
point of view of using this model to interpret our data,
it is
examine the
also useful to
consequences of relaxing the strong symmetry assumptions that we have so far imposed.
We now relax
each of the main assumptions in sequence.
1.
Different ability distributions in the two locations: In general, more or
can happen in this case. One interesting special case
is
when one
of the populations has
along both dimensions than the other. This would be the case, for example,
a better outside option than population
1:
in this case the
less
if
anything
more
ability
(say) population 2 has
low ability people in that population would
take their outside option and only relatively high ability people would invest in either of the industries.
It is clear
that this will tend to widen the ability gap between those
location 1
and those who stay
Asymmetric
2.
rates
is
interest rates:
=
r^
—
rf
>
population 2 pay higher rates both
The
effect of relaxing
if
the assumption of symmetric interest
community has systematically higher
- the net
effect
More
interest rates
than the
can in principle go either way since people from
they migrate and
the decision to take the outside option.
will take the outside
location 2 to
in location 1.
potentially quite complex. If one
other - say, r2— r\
who move from
if
they stay. The one unambiguous
effect is
on
of the people in population 2 (which faces higher rates)
option and, as discussed in the previous paragraph, this has the effect of raising
the average ability of movers from that population. -^^
One
resison
why
interest rates
may be
higher in one population than in the other
12
is
that ability levels are higher in
by stayers
may
also
it
makes
selected for industry
3.
Xi
^ Xl,
only people
rj
— r2 we
i.e.,
who move
and the net
industry 2
/ (•) curve
is
a
lot
more
To
small for widely dispersed communities.
raise rf
— rj and
reduce rj
(because
— r^.
less
fix
This
people
to location 1 from location 2 are strongly
In other words, both the movers and the stayers
1.
who would choose
other words, the
it is
— rj =
those
likely that
it less
are Hkely to have lower ability
1
r^
the differential to be large for communities
likely that stayers in location 1 are strongly selected for industry 1
it less
industry
home base and
where starting with
ideas, consider the case
move), but
One might expect
vary across populations.
that have few connections outside their
makes
by movers and the rate paid
possible that the differential between the interest rate paid
It is also
among
those
who
finally join
may be ambiguous.
effect
profitable than industry
1:^^ In this case the
to stay in location 1 would be people with very high values of a^ .^° In
move
will
to position like
A"B"
in Figure
also true that only people with very high values of a^ will
Therefore the net effect of changes in
Xi/X\ on
move
1.
However, by the same logic
to industry 1 from location
2.
the extent of positive selection can be either positive
or negative.
Investment eind Productivity
2.6
The main
lesson from our investigation of location choice
who move
to a particular location
of this selective migration
may be more
on investment
is,
who were
"born" there.
result,
effect
quite possible that migrants, despite
it is
ability, invest less.
The
net effect
may be
in either direction,
Movers have higher abihty but may
effects.
but as we stated in the introduction, we are
interested in the case where movers invest less but have a higher average productivity.
shows that under the conditions of Claim
moving and staying (the proof
Claim 3 Assume
is
2, this
The next Claim
property holds for the person on the margin between
in the Appendix).
that djl{z,a)/da
that population: this raises the
The
The
however, ambiguous, since the ability bias exists precisely
Turning next to productivity, there are clearly two
invest less.
that under reasonable conditions those
able than those
because the movers face higher interest rates. As a
having higher
is
demand
is
a constant and that the elasticity of the function djl{z,a)/dz
for capital
and therefore
interest rates.
case where the good produced by industry 2 carries a higher price than the
good produced by industry
1 is
exactly the same.
The
raising
the
effect of raising
Xl/Xl
first effect
X?
,
keeping
X\
fixed, is actually
also flattens the /(•) curve,
somewhat
subtle:
which acts as a countervailing
must, however, dominate.
13
it
has the effect described in the text, but
effect.
In the extreme cases, as in Figure
1,
with respect to z
is
greater than
have a lower z but a higher
This
is,
1.
Jl{z,
a)
Then
if
the person
who
of course, at best illustrative:
it
moving and staying
applies only to the marginal person, while
movers and
stayers.'^^
facie plausibility of a negative relation between ability
will
However,
we
are interested
does establish the prima
it
and investment.
Dynamics
With the
results
about investment and location choice in hand, we proceed to derive the capital
stock and production trajectories for a firm with ability
accumulation process of someone
Xi
indifferent between
he moves.
in the average for the entire population of
2.7
is
to vary across firms).
who
a
facing an interest rate r. Begin with the
Xi
starts out expecting sales of at least
(note that
we now
allow
Then
X2 = Xi(l + 6i)/z(^^^,a).
+ ^1
1
Iterating
on
this
formula we get
,^
+ ei)
Xt=Xi{l
/-.
\
/-,
(l
N
/•2*(a,r)
+ et_i)M
/
'
,z*(a,r)
.
.
;x(—^^^,a).
\ a)
Therefore,
t-l
t-l
logXt
=
logXi
,.
V
+ 5]log(l + 6,) + ^log^(^-^,
s=l
a).
s=l
Taking expectations, we get the dynamic path of evolution of output:
ElogXt = £;iogXi + itTo
l)£;iog(l
+ es) +
get the corresponding equation for capital stock,
ElogKt = ElogXi + E\ogz*{a,r)
Moreover, as we will
see,
work - the correct measure
is
n
is
+ (t -
(t
-
1)
E\ogtx{^^^^,
we observe that Kt
l)^log(l
+ e,) +
{t
-
a).
= Xf z*{a,r),
1)
which gives us
£;iog^(^J^^,
a).
not exactly the right measure of productivity from the point of view our empirical
E{logfj.{-^,a'')}.
14
Because £^log(l + es) ought to be the same
and only
even
if
if
E log fj,{^j^f^
a)
,
is
higher.
for everyone,^^
As observed
they have a lower z*. For the same reason,
both capital stock and output grow
above,
/l(2*, a),
E log ij.{^-^^f^
may be
if
higher for the outsiders
may
a),
,
faster
be higher
also
for the
outsiders.
Since those
abiUty,
if
we observe
conclude that
work reported
3
who have
it
has
The
The
less ability
but faces lower interest rates. This idea
they have higher
is
we should
the basis for the empirical
in the following sections.
Background
Setting
setting for the empirical analysis
we
the next,
if
that one group has a higher z* but a flatter trajectory for output
Institutional
3.1
a lower z* can have a higher Elogfi^^j;^^, a) only
try to explain
why
is
the South Indian town of Tirupur. In this sub-section and
Tirupur's production cluster
is
ideally suited to test the implications
of network-based lending developed in the previous section.
Tirupur
is
traditionally
located in Coimbatore district, in the
known
as
Kongunad, one of the
to the arrival of the British.
an
elite cultivator caste, in
is fertile
and there are
Kongunad
is
modern Indian
state of Tamil
Nadu. This area was
the Tamil-speaking country, prior
five big sub-divisions of
believed to have been colonized by the Vellala Gounders,
the twelfth century (Beck, 1972). While
significant reserves of subsoil water.
Where
Kongunad
quite dry, the soil
is
well irrigation
was
available, high
Tamil Nadu in the
last
quarter of the
agricultural yields have been obtained from early times.
Kongunad emerged
as the
most commercialized region
in
nineteenth century with the advent of the railways, as cultivation shifted predominantly into cash
crops (particularly cotton).
which
is
By
the 1950s Coimbatore had
20%
of
its
land allocated to cash crops,
the largest share of any district in Tamil Nadu, and this land was
in the state
among
the most valuable
(Coimbatore District Gazetteer, 1951; Baker, 1984). While the Kongu Vellala Gounders
had always been wealthy, the
cultivation of cash crops transformed this
community
into
one of the
wealthiest in Tamil Nadu.
While Tirupur's association with the cotton trade goes
century, the
tars,
first textile
at least as far
back as the nineteenth
manufacturing unit was only established in the town in 1935. The Nakarat-
a community traditionally involved in trading, initially dominated the industry. However, after
This
is
almost by construction: differences among people have been swept into the
15
fj,
function.
a prolonged period of labor unrest in the mid-1960s, they were largely replaced by the Gounders
(Swaminathan and Jeyaranjan, 1994). The Gounders are a
so-called "right
so they were traditionally confined to land-based activities; this
venture outside agriculture.
For the next twenty years or
so,
was
hand" (valangkai)
their first significant
caste,
commercial
the industry was dominated by the
Gounders and catered almost exclusively to the domestic market.
of knitted garments from Tirupur started to grow very rapidly around 1985,
The export
the early 1990s the annual growth rate was above 50%. This generated an inflow of
from outside Tirupur. By the mid-1990s, which
is
exporters were Gounders while the rest were from
Gounders,
9%
are Mudaliars,
10%
are Chettiars,
when we observe
all
and
in
new entrepreneurs
the industry, about half of the
over India. In our sample of exporters,
58%
and the remaining 23% are from outside South
are
India,
mainly from traditional trading communities such as Marwaris, Gujaratis and Khattri Punjabis.
Prom
the point of view of our framework, Gounders in Tirupur are almost the perfect example
of local people.
industry,
outsiders,
They have
substantial presence in the area
and extensive experience
in the local
both of which strengthen network lending. In contrast, the other communities are
who
literally
only arrived in Tirupur in the 1990s with the surge in exports. These outsiders belong
to traditional trading communities with well established networks in other parts of the country but
is
probably too soon for these new entrants to have formed their
would expect such networks to ultimately emerge
if
investment in the long-run. Por the time being at
own networks
in Tirupur,
it
though we
the industry continues to provide high returns on
least, it
forego the lower interest rates that they would receive
if
seems plausible that the outsiders have to
they located in one of their more established
business centers elsewhere in the country.
Our model would
therefore tend to suggest that
more, at least relative to their
ability.
Gounders may have lower
ability
and
will invest
Several other factors reinforce this effect. Pirst, the Gounders
have almost no presence in trade or industry outside Tirupur and therefore their access to network
lending, were they to invest outside Tirupur,
is likely
in sub-section 2.5, larger interest differentials
ability of the population that stays back.
to be very limited.
As we have already observed
between home and abroad tend
to lower the average
Second, the outsiders in Tirupur are from traditional
business communities whereas the Gounders are, for the most part,
new
to industry.
At
least in
terms
of understanding the process of exporting, the outsiders probably have an ability advantage. Third,
because the outsiders come from traditional business communities, their capital probably has
alternative uses, while the
Gounders can only
many
invest their substantial agricultural wealth in the local
16
garment industry, or
-
rates
for
in agriculture or in the highly inefficient Indian financial sector.
both movers and stayers - are probably higher
As we observed
for the outsiders
than
The
for the
interest
Gounders.
in sub-section 2.5, this tends to raise the average quality of outsiders in the industry
(and depress the amount they invest), relative to the Gounders.
The Industry
3.2
The industry produces
t-shirts, targeted at
knitted garments and
low-end
retail outlets in
is
largely focused towards exporting.
Europe and the United
States.
Most firms produce
There are essentially three
types of firms in the industry: direct exporters, indirect exporters, and job-workers. Direct exporters
are the ones
who
receive orders
of the order to one or
who
more
from abroad. Once they have an order they often pass on a fraction
indirect exporters. Indirect exporters are independent
garment producers
are entirely responsible for their share of the order, delivering the finished product to the direct
exporter prior to shipment.
Garment production
stitching, while the
direct
and
is
organized in a number of stages: the major stages are knitting, dyeing and
minor stages include calendaring (shrinkage control), printing and curing. The
indirect exporters will typically
them. For the rest of the stages they
machinery
for
will
own machinery
is
some stages of production, but not
all
of
employ job-workers, who are specialized producers owning
a single stage only.
Job-work and the use of indirect exporters allows
and
for
one reason why there can be large variations
However, such decentralization has costs of
its
for decentralization in the
in the output-capital ratio across direct exporters.
own. Quality appears to suffer and delays in shipment,
particularly during the
peak production season, are more frequent.
bankers in Tirupur and
officers in the
agency that insures exporters,
it
production process
From our
conversations with
Export Credit Guarantee Corporation (ECGC), a govermnent
appears that such delays often result in orders being rejected by
foreign buyers.
3.3
The Data
The main data
source for this paper
and job-workers carried out
is
in 1995.
a survey of six hundred direct exporters, indirect exporters
Details of the entrepreneur's background, his access to
financing, as well as export (production)
to 1994, were collected from each firm.
bank
and investment information over a four-year period, 1991
Some supplemental information was
17
collected through a brief
re-survey in 1997.
Before turning to a description of the data,
in the 1995 Survey,
which
is
non-standard.
we
briefly describe the
The Tirupur production
comprising at least two thousand production units.
"list"
of firms in the town. Moreover, accurate
towns,
is
Many
maps
sampUng procedure employed
cluster
is
a complex institution
of the units are unregistered, so there
no
are unavailable; Tirupur, like most small Indian
a maze of lanes and by-lanes. Under these circumstances we were unable to conduct a census
of all production units, which would have allowed us to randomly sample firms for the survey.
also unable to divide the
town
into a sufficiently large
us to survey a sample of clusters. Instead
number
is
we divided
of days to survey firms within each zone.^^
until the entire
number
We were
of clusters, which would have allowed
the town into ten zones and allocated a fixed
We
then proceeded from one zone to the next
town was covered, over a three month period. Ultimately, information was
collected
from 300 indirect exporters and 147 direct exporters. The distribution of firms by community, in our
sample,
is
very even across the ten zones, suggesting that the
sampUng was not biased towards any
community.'^^
Descriptive Statistics
3.4
The
discussion that follows focuses for the most part on the one hundred and forty-seven direct ex-
porters in the 1995 Survey. Since
we
are particularly interested in comparing the investment behavior
and the export performance of the Gounders and the Outsiders, the sample
We
munity.
specified to
also divide firms into
be
five years of
of experience. Note that
that
we
We
Young and Old
units,
is
partitioned by com-
where the cut-off separating these firms
is
export experience. Very few firms in our sample have more than ten years
we have data
over a four-year period, 1991 to 1994, for most of the variables
discuss.
begin with the individual characteristics of the direct exporters in Columns 1-4 of Table
Each exporter provides information on when he
first direct
export order.
The
is
the
arrived in Tirupur as well as
entrepreneur's age and experience can then be
in time over the sample-period: age
while experience
first
number
is
the
number
of years since he
of years elapsed since he
became a
direct exporter.
when he
computed
first
The
1.
received his
at each point
arrived in Tirupur
sample-statistics for
Production units in Tirupur are located
of
homes and production
units today,
for the most part in converted residential structures. There is no segregation
and firms are spread throughout the town. Since we were unable to identify any
we surveyed the entire town.
a single exception - one zone had a relatively low proportion of Gounders. Dropping this zone does not
affect the estimated export and capital stock trajectories that we report later in Section 4.
spatial clustering in production activity,
There
is
18
,
the two communities are fairly similar, with the exception of the Old direct exporters,
than
five
years of direct export experience. In this category,
we
see that the age variable
is
Gounders were established
Tirupur before the Outsiders arrived.
Turning to sources of finance,
all
the entrepreneurs and bankers
we spoke
significantly
to mentioned the impor-
tance of informal sources of capital in Tirupur. Family and community capital
of the financial structure of
is
consistent with the institutional background, which suggests that the
larger for the Gounders. This
in
who have more
most firms that operate
is
a major component
and Tirupur does
in India's small-business sector,
not appear to be an exception to this pattern. Further, nidhis (informal credit institutions) and chit
funds (rotating savings and credit associations) have been used extensively in Kongunad since early
among
times (Baker, 1984), and continue to be an important source of capital today, particularly
Gounder businessmen.
However, when
came
it
to providing actual details of their
firms were generally unwilling to discuss the informal
this
many
because
is
own finances, we found
component of
their finances.
and
of these transactions, including the informal nidhis
that the surveyed
We
speculate that
chit funds, violate tax
laws and/or financial regulations.
They
did, however, provide us
with information about their bank credit. Bank credit
an impor-
is
tant source of credit for financing investments in machinery for both communities (Table
1).
More
experienced exporters also appear to have greater access to bank credit. Notice, however, that there
is little
difference
between the two communities, both among the Young as well as the Old
last observation is useful in ruling
and Outsiders
differently.
The
firms. This
out the possibility that the formal capital market treats Gounders
distinct investment behavior that
we observe
for these
communities
is
evidently due to their differential access to other sources of finance.^^
Partnership
out that as
firms.
is
many
another formal channel through which the firm can expand
as
25%
of the Outsiders
its
capital base. It turns
and 31% of the Gounders do have outside partners
There were on average three outside partners
in these firms (both for
and the proprietor's share of the ownership was 44%
for the Outsiders
Gounders and Outsiders)
and 39%
for the
While there are small differences here that go
in the right direction (Outsiders are
on a
sets of firms basically look very similar.
single person's personal assets), the
The
fact that
Gounders invest more
two
overall but have the
same proportion
of
bank
tends to be complementary to other sources of finance rather than a substitute. This
models of imperfect capital markets
the
amount
(e.g.,
consistent with
Holmstrom-Tirole, 1997) where the amount the bank lends
of private resources that the investor can raise.
19
Gounders.
more dependent
This
capital, suggests that
is
in their
is
bank
is
not
capital
many standard
proportional to
surprising - partners are typically
members
of the extended family
(29% say they
are partners with
an immediate family member and another 51% name members of their extended family) and have
access to the
same community network.
Gounders and Outsiders
will
be in
We
should therefore expect that the main difference between
levels rather in the share
owned by the partners - each Gounder
partner would be expected to bring more capital (in absolute amount) into the firm.
We
also collected additional details
about the exporter's background.
It
turns out that the Out-
more schooling than the Gounders; the average years of education
siders received
for the
munities (with standard deviations in parentheses) are 13.41(2.62) versus 11.90(3.96).
the equality of
means
the Outsiders versus
industry.
There
is
"ability" in the sense of
With
this
variables that
of the
therefore
in
either
and the
and the
for
Note that Section 2 derived the production
interest rate. Total production
indirect exporting that
is
done
is
It is
we use
is
produced
for the
when
direct
may be a better measure of performance than
supported by the evidence, presented
later,
total
that specialized indirect exporters of
machinery of their own than the direct exporters, which
less
much
if
they really intended
indirect exporting.
worth emphasizing, however, that
total production rather
available
roughly measured by the
essentially a fallback
suggests in turn that direct exporters would not have invested nearly as
on
is
for others (very little
a direct exporter indirect exporting
community operate with much
to focus
higher
hand, we turn to the data on the investment and output
orders are unavailable, the volume of direct exports
is
may have
being better prepared to become successful producers.
domestic market). Because
production. This
to suggest that the Outsiders
at the heart of the empirical analysis.
of direct exports
reject
families with previous experience in the textile
some prima facie evidence
trajectory as a function of ability
sum
Gounders belong to
background information
lie
can
two communities with greater than 95% confidence. Further, 74% of
for the
57%
We
two com-
all
the results reported in this paper continue to hold
if
than direct exports as a measure of production. The exact results are
from the authors.
Looking now at direct exports. Table
very similar for
Gounders.
Young
It will,
1
shows that average exports
direct exporters, but
among
for the
two communities are
the Old exporters the levels are higher for the
however, be a mistake to infer from these simple averages that Gounders have
steeper trajectories with respect to experience;
we
will see in the next section that this gets reversed
once we introduce the necessary controls.
Turning next to investment, Gounders own significantly more machinery at each stage of the
20
direct exporter's life-cycle.
particularly for the
do a
Looking at the capital-export
Young exporters,
is
ratio,
with this evidence
striking. Consistent
significantly greater fraction of indirect exporting.
This presimiably
when
exporter will accept indirect orders from other exporters
accept less indirect exports as they
a substantial
level of indirect
We
on
full capacity.
is
the fact that Gounders
reflects the fact that
machinery
AU
is idle.
a direct
exporters
yet Old Goimders continue to maintain
own
orders are never sufficient to
In contrast, the Old Outsiders in our sample focus
direct exporting.
also look at the capital stock that direct exporters
order in Table
The
become more experienced,
his
exporting which suggests that their
keep their machinery running at
exclusively
the difference between communities,
This
1.
distinction
is
available for all direct exporters
between the communities holds
have in the year prior to their
who commenced
first
direct
during the sample-period.
Gounders
for the starting capital stock as well;
start
with nearly three times as much capital as the Outsiders.
Finally, there are clear differences in the extent of vertical integration. Defining vertical integration
as ownership of machinery in all three
we
see that
Outsiders.
15%
of the
major stages of production
Young Gounders
and
(knitting, dyeing
stitching),
are vertically integrated, as opposed to 7.5% for the
These numbers increase when we study partial
vertical integration,
which
is
Young
defined as
ownership in two or more of these stages of production, but the difference between the two communities
remains.
We close this section by briefly describing the characteristics of the indirect exporters, who are very
different
from the direct exporters
differences are readily discernible.
move up and
Looking at Colimins 5-8
in our sample.
they are younger. This
First,
is
in
Table
1,
not surprising as
the following
many hope
receive direct orders of their own. Very few indirect exporters, particularly
among
to
the
Outsiders, remain in the business once they have crossed five years of age. Profit-margins are small for
these producers and most will leave the business
years. Second, the indirect exporters are
much
if
they do not receive direct export orders within a few
less
reUant on bank finance than the direct exporters.
Capital stock and production are also far lower than the corresponding levels for the direct exporters.
Notice that there
is
Httle difference
between Goimders and Outsiders among the indirect exporters.
Furthermore, firm characteristics hardly change with the exporter's age, although this pattern in the
data
may be due
to selected-exit
among
the older indirect exporters.
Third, most of the indirect
exporters are Gounders. In contrast, the direct exporters are evenly divided between Gounders and
Outsiders. Migrants appear to
come with the view that they want
21
to be direct exporters.
Estimation
4
In this section
We begin
we subject the
basic pattern in the data that
we
report above to
by discussing the identification of the export and capital stock
the estimation results.
We
more
careful scrutiny.
We then present
trajectories.
conclude this section by discussing some of the important assumptions
underlying our interpretation of the results.
Export and Capital Stock Trajectories: Identification
4.1
The
trajectories that
we derived
in Section 2
can be expressed, with a change of notation, as the
following:
yu
where yu
firm
i's
is
=
UiEXPit
+ fi + Cit
either logXt in the export regression or logKt in the capital stock regression.
experience in direct exporting, which was denoted by
Prom
of the firm's trajectory.
term which accounts
stock regression.
levels for all
for
ElogXi
We will treat
fi
2, Ilj
in the regression equation.
=
Elog{l
+
Cs)
as
an unobserved
who
—1
+ Elogfi
fi is
a firm-specific
+ Elogz*{a,r)
in the capital
(^jf^,Oi]-
fixed-effect in this section since the starting export
entered during the sample-period, are unobserved by
we introduce eu a mean-zero disturbance term,
first
is
\
EXPit,fi)
its
—
0,
access to
depending on the regression we are estimating.
is
to
compare the slope of the export and capital stock
Gounders and Outsiders. Notice that the equation above, however, allows
which
E{eit
This would measure current shocks to the firm's exports or
Ultimately our intention
Our
is
in Section 2. Ilj represents the slope
and ElogX\
in the export regression
but the youngest exporters,
the econometrician. Finally,
capital,
Section
t
EXPu
task
is
a firm-specific slope
IXj.
consequently to demonstrate that we end up estimating the community-mean for
Ilj,
what we want, when
To
the regression equation.
11, is
AH? =
signifies that
we
Ili
are
—
11'^,
now
for
replaced by the corresponding community- level coefficient n*^ in
see this, rewrite the equation above as
yf,
where
trajectories for
= W^EXP^, + /f +
£'(An?)
=
[An? EXP^t
•
+ 6?,]
by construction. The c superscript
estimating separate regressions for each community.
22
in the equation
above
To show that an unbiased estimate
from the equation above.
firm to derive the
OLS
We
of
11'^ is
obtained,
take advantage of the fact that AII^
estimate of
the
first
on the
var{EXP[)
the same for
right side of the equation
Our next
we
=
£'(An?)
step
is
0,
all
in.
var{EXP^)
are independent.
first
11'^
point to note
level, or z*,
is
is
different for the
each
we
In Section 2
varied across cohorts.
of the second term
We
E{var{EXP[)).
noted
measures the experience-effect
that ff only affects
II'^ if it
varies systematically
said nothing about
Since the industry
would seem reasonable to assume that successive cohorts vary
is
for
obtained.^^
coefficient correctly
over time, since experience grows over time as well.
the starting export
The numerator
above can thus be written as JE(An^)
hence an unbiased estimate of U^
The
a constant parameter
firms in a balanced panel, regardless of their experience in
to verify that the estimated
are interested
is
E{var{EXPi))
year of the sample, AII^ and
earlier that
that
is
differencing out the fixed-effect
n*^ as:
^
Since
we begin by
is
in their ability
in its
how Xi,
growth phase,
and perhaps
this effect
two communities. This would imply that z* and perhaps Xi, and by extension
would vary over cohorts as
well.
11'^
would be biased in
this case
if
fixed-effects
it
/f,
were not included
in
the regression equation.
Further,
when deriving the export and
capital stock trajectories in Section 2
possibility that the entire industry could receive aggregate shocks,
we did not
allow for the
with a time trend. For instance, the
volume of orders from abroad could grow as more buyers learn about the Tirupur production
or as its reputation expands.
cluster,
Because experience also grows over time, the experience-effect could
simply proxy for a time-trend in the year
To
effects.
see the
problem that could
arise,
we introduce a
time-trend in the regression equation below:
With
fixed-effects
sample mean. While
This
is
we
this
are effectively differencing out each variable in the equation above from
procedure takes care of the cohort
not strictly correct for the youngest firms
firms enter later, and therefore have lower
n'^
who
effects, it
introduces a
new problem. With
enter during the sample period. For example,
var(EXPi), then Allf and var(EXPi)
upward.
23
will
its
if
low-ability
be positively correlated, biasing
a balanced panel, (EXP[^
— EXP[) =
year of the sample, in year
t
(here
t
—
EXP^,
t
t,
for all firms, regardless of their experience in the first
are means,
computed over the sample
when
separately identify the experience-effect and the time trend in the year effects
We cannot
period).
fixed-effects are
included in the regression equation.'^^
One way
of getting
around
this
problem
is
year effects (as in Deaton and Paxson, 1994). However,
we
assume that there
to simply
is
no time trend
we noted earUer that demand
in the
shocks, which
associate with the year effects in our application, are very likely to be increasing over time in this
growing industry.
effects is
common
Our
identifying restriction instead
- variations
in
to assume that the time trend in the year
across communities. Since the time trend arises from expansion of foreign
would seem to be a reasonable assumption
this
is
in this setting.
Other community
demand,
specific year effects
each community's supply of credit in a given year, for instance - are
likely to
be
uncorrelated with the time trend and can be included in the error term without biasing our results.
The
differenced regression equation can then be written as
yf,
where y^
,
e?
are sample
- yl -
means
regressions, estimated separately
which
n'^,
is
what ultimately
(n^
+ l){EXP^^ - EXP[) +
as usual.
The
(e^,
-
e^.),
difference in the slope of the trajectory in the
by community, now
two
identifies the difference in the experience-effect
interests us.
Export and Capital Stock Trajectories: Estimation
4.2
We saw in Section 3 that
also
the Gounders held higher levels of capital on average than the Outsiders.
saw that the Older Gounders had higher exports than comparable Outsiders, whereas
between the two communities were insignificant among the Young exporters.
We now
We
differences
subject these
patterns in the data to more careful scrutiny by comparing investment behavior and export outcomes
for the
two communities at each point in the exporter's
included to account for unobserved cohort
life-cycle.
Firm
fixed-effects will also
be
effects.
Exports and capital stock are regressed separately on experience in Table
in these trajectories over the life-cycle, separate coefficients are estimated for
2.
To allow
for variation
Young and Old
exporters.
^^See Deaton (1997), pp. 123-127, for a clear discussion on identifying age-effects (or experience-effects in our context)
with panel data, when cohort effects and year effects are present.
24
The
cut-off separating these categories
is
specified as five years of direct export experience.
We
will
also present the corresponding nonparametric kernel estimates of these trajectories in Figures 4-7.
We
begin with the simplest export regression, without fixed-effects, in Columns
in logs, are increasing over time for
We
Old exporters.
coefficients for the
both communities, although the trajectory
is
cannot reject the null hypothesis, at the 5 percent significance
two communities are the same
at
term, which measures the starting level of exports,
both stages of the
is
Exports,
1-2.
flatter for the
level,
that the
While the constant
Ufe-cycle.
larger for the Gounders, this difference
also
is
not statistically significant. In general, the starting level of exports and the subsequent trajectories
two communities are
for the
This
statistically indistinguishable.
corresponding nonparametric regression, presented in Figure
grow
faster
for the
more
clearly
demonstrated
in the
While the Older Gounders appear to
4.
than the Old Outsiders, consistent with the patterns
two communities overlap throughout the exporter's
is
Table
in
1,
95%
the
confidence-bands
life-cycle.
Insert Figure 4 here.
One explanation
for the concavity in the
failed to control for cohort effects in
export trajectory, for both communities,
Columns
1-2.
If older
therefore lower average exports over the Ufe-cycle, then
even
if
the experience-effect was really linear.
To
The
we cannot
trajectory
we would observe such a pattern
control for these
and other cohort
also steeper for the Outsiders,
constant term in the Gounder regression
for the Outsiders.
Note that
measures the starting
more easy to
now
in
Columns
reject the null that the export-coefficient is constant over time, for
is
is
now
this constant term,
level of exports for
that
we have
cohorts had lower starting exports, and
proceed to include firm fixed-effects in the regression equation.'^^ Notice
2 that
is
in the
data
effects,^®
we
3-4 of Table
both communities.
both among Young and Old exporters. Moreover, the
significantly larger
which
is
computed
each community.
than the corresponding
coefficient
as the average of the fixed-effects,
These patterns
in the
data are once
visualize with the corresponding nonparametric regression, presented in Figure 5.^°
The
See the discussion of cohort effects in the previous sub-section.
Year dummies do not appear in the fixed-effects regression. Recall that the experience-coefficient includes the time
trend in the year effects in this regression. Thus, while we could in principle include year dummies (only two would be
identified for the four years in the sample), these
dummies
dummies would measure the deviation from
this
time trend. The year
are thus orthogonal to the experience variable, by construction, so their omission does not affect the estimated
experience-coefficient.
^"To obtain the nonparametric kernel estimates in Figure 5 we first difference out the fixed-effects, after retaining
the constant term for each community as described above, from the nonparametric series approximation, presented in
Columns
3-4 (following an approach suggested by Porter, 1996).
the deviation from the
mean
What we
are doing essentially
fixed-effect, to allow for the possibility that the fixed-effects
25
is
to diff'erence out
vary systematically across
Gounders begin with higher exports, but the Outsiders surge ahead
after
about
five
years of direct
exporting.
Insert Figure 5 here.
Turning to the capital-stock regression, we
here as well.
much
Starting with the regression without fixed-effects,
that the capital-stock trajectory, in logs,
experience-coefficients for the
for the
will see that
Gounders
increasing
two communities are
and the
larger
is
is
difference
the starting level for the Gounders
is
estimates in Figure
we
and concave
see in
for
Columns
both communities. While the
between the communities
is
just significant.
higher and the slope of the trajectories
life-cycle.
This
5-6 of Table 2
statistically indistinguishable, the constant
two communities, we would expect the Gounders to have a
than the Outsiders throughout the
of the preceding discussion applies
is
is
the
same
term
Because
for the
significantly higher level of capital stock
precisely
what we
see with the nonparametric
except for a brief period around the five-year experience mark.
6,
Insert Figure 6 here.
Introducing fixed-effects in the capital stock regression in Columns 7-8, the difference between the
communities widens. The trajectory
Finally, the constant
term
for the
is
now linear
Gounders
is
for
both communities and steeper
significantly larger
than the corresponding estimate
for the Outsiders. All of these patterns are observed in the corresponding
presented in Figure
7.
The Gounders begin with a
advantage throughout the
fife-cycle,
for the Outsiders.
nonparametric regression
higher level of capital stock and maintain this
although the trajectory
is
steeper for the Outsiders.
Insert Figure 7 here.
The patterns that we observe in the data can be easily interpreted, using the simple model presented
in Section 2.
The export and
capital-stock trajectories are both steeper for the Outsiders which, in
the context of our model, implies that
E log fi{^-^f^
,
a) must be greater for them.
We assume here that the first stage is flexible enough to capture the basic features of
the export trajectory, and indeed the linearity in the kernel estimates is consistent with the patterns that we obtain
with the series estimator. All the nonparametric regressions in this paper utilize the Epanechnikov kernel function.
cohorts within each community.
Pointwise confidence intervals are computed using a method suggested by Hardle (1990). We assume that the estimated
when computing the nonparametric confidence intervals since the kernel estimates converge much
more slowly than the fixed-efi'ects.
fixed-efi'ects are "fixed"
26
Turning to
We
hfe-cycle.
z*,
Table
1
shows that
also regress log{z*)
it is
higher for the Gounders at every stage of the exporter's
on the exporter's experience, separately by community, including
firm fixed-effects to control for cohort effects in
by the constant term,
is
Columns 9-10 of Table
is
much
nonparametric regression in Figure
8,
we see as expected that
starting z*, measured
While z* dechnes with experience
just significantly larger for the Gounders.
both communities, the decline
for
The
2.
sharper for the Outsiders. Turning to the corresponding
z*
is
significantly higher for the
Gounders
at every level of experienced^
Insert Figure 8 here.
Because z*
be that a
is
is
higher for the Gounders, and £^log^( ^^^^''''
a and
lower interest rates and Outsiders have higher ability, which
networks and migration in Section
up investing
2.
Network lending
is
z* are
it
must
complements. Gounders face
precisely the prediction of the
distorts the capital
effect of
an increase
in
a on
z*
is
/x,^^
>
is
model of
market because high
ability
crucial for our interpretation of the results.
unambiguously positive
condition in the firm's investment decision, an increase in
in this case; going
a both
raises Ji^
the interpretation of the export and capital stock trajectories that
Suppose instead, that
effect
and
higher for the Outsiders,
less.
should be clear that the assumption that
The
is
Discussion of the Underlying Assumptions
4.3
It
a)
higher for them. This implies, in turn, that r must be lower for the Gounders in order to
generate a higher z*, under the standard assumption that
firms end
,
on
Jt^.
Ji^^
<
0.
Now
the positive effect of
In this case, an increase in
a
back to the
and
1^(1).
first-order
Consequently,
we provided above goes through.
a on V{1) may
could result in a decline in z*.
not dominate the negative
Now
the shallower export
capital stock trajectory for the Gounders, as well as their higher z*, could be explained
by lower
ability alone.
The reader will note that in contrast with what our model predicts, 2* is declining over the life-cycle for both
Gounders and Outsiders. This suggests that the fi function may not be the same for young and old firms: it is quite
plausible that fi may be less responsive to z for firms with higher levels of experience - because, for example, their
reputation may be much more secure. If we introduce the assumption that
= fj,{ ^ ,A^ ,a, ()),
varies with time (i.e.,
and in particular that /Xj falls as the firm ages, into the model in Section 2, it is easily checked that the basic analysis
still goes through. The value function is still linear in expected exports, and as long as /Xi2 >
continues to hold for
all values of t, we can still explain why the Gounders have a higher z* but both capital and exports grow faster for
the outsiders. However, z' is no longer constant and will in fact decline over time, which fits the facts well. The one
compUcation it introduces is the possibility that the time paths of log-capital and log-exports may not be Unear, though
fj.
it
does not rule out their being linear.
27
fj,
.
There
are,
ability-capital
however, at least three reasons to be skeptical of this explanation of the data. First,
complementarity
ipants in the industry face the
is
a very standard assumption in the literature. Second,
same
capital suggests that the Outsiders
attracted a
Finally,
and
Ji
much
if
must earn huge
and despite the
rents. It
fact that they
seems puzzling that
larger inflow of Outsiders over time.
same kind
substitution effects are the entire story, the
should also be found within each community.
of negative relation between z*
Note that, by contrast,
be the same within and across communities. In the case where
Jt will
spend more on
would not have
this
our story which
in
p
based on different interest rates across communities, the relation between z* and
and
partic-
Gounders seem to be able
interest rate, then the fact that the
to survive in the industry despite their lower ability
if all
is
does not have to
T^a^ is positive, the relation
between z*
be positive within the community (where interest rates ought to be similar across firms),
and yet could be negative across communities, as we have pointed out above.
The problem
in estimating this relationship within the
the econometrician: what
trajectory Elog{\
+
et)
+
we observe
instead
z* /{I
is
them by
ej) in
Elogfx over the sample-period.
correlated, particularly at the level of the firm, in
built into
+
construction.
To avoid
community
period
We
that z*
is
t,
not observed by
is
and the slope of the export
might expect
(1
-I-
et)
to be serially
which case these variables have negative correlation
this potential
problem, we look at the correlation between
the exporter's capital stock before he got his first direct order (and therefore hopefully before he
learned
et)
and the slope of
avoid this source of bias,
The
period.
initial capital
We
it
his export trajectory.
stock
Ki
correlation in
initial capital
the view that the correlation
we
Columns
By
both communities
first
order during the sample-
1-2 of Table 3, separately
contrast,
by
we have already seen that
stock but have flatter trajectories than Outsiders, consistent with
may be
different
between and across communities. While the
also estimated the export slope-z* correlation,
average capital-export ratio for the firm over the sample-period.
for
able to entirely
associated with a steeper slope for both communities, consistent
with the view that abiUty and capital are complements.
are not reported here,
we have not been
available for firms that received their
is
initial capital is
Gounders have higher
if
goes in the direction of making the correlation more negative.
estimate the export slope-
community. Higher
Note that even
A
where z*
is
results
measured as the
positive correlation
was obtained
in this case as well.
Another assumption that
is
key to our interpretation
28
is
that capital
is
the one input whose price
varies across
community
Suppose instead that labor
lines.^^
mattered and, in particular, that
This would be consistent with the view that the social
the Gounders had access to cheaper labor.
ties
also
between the Goimder exporters and the local workers reduce information problems in the labor
market
(as in Chari, 1997).
Now
the higher capital stock for the Gounders could simply reflect their
preferred access to a complementary input, labor, rather than a cheaper supply of capital
One
implication of this alternative view
capital, since the
complementary input
is
is
more per
that the Goimders should produce
eflFectively
To check
cheaper.
itself.
this possibility
irnit
of
we run a
regression with the capital-output ratio as the independent variable, which exactly parallels the capital-
export ratio regression reported in Table 2.^^
This
is
reported in Colmnns 3-4 of Table
corresponding nonparametric regression for the two communities
the nonparametric regression,
we
see instead that the
is
presented in Figure
9.
3.
The
Focusing on
Gounders maintain a higher capital-production
ratio at every stage of the exporter's hfe-cycle, contradicting the cheap labor view.
Insert Figure 9 here.
Another related explanation
rehes on differential access to the foreign buyers.
exporters,
who have aheady
Goimders who are
effectively begin
two commimities
Suppose, for example, that the older Gounder
just beginning to export.
Close community
ties
allow the young Gounders to
with a reputation in this case. This would explain both the larger
and capital stock
trajectories, that
Notice, however, in Figure 5 that the export trajectory
experience.
for the
established themselves with their buyers, provide referrals for other
as well as the shallower export
The Outsiders
we observe
for the distinct trajectories that
is
we observe
initial
for that
investment,
community.
not just steeper for the Outsiders.
actually cross the Goimders in the absolute level of exports after about five years of
So we do not simply see convergence between the two communities, as the Outsiders
gradually build up their
own
reputation, as the preceding argument would imply. Instead, the older
Outsiders achieve higher levels of exports, yet continue to maintain lower levels of capital stock, which
suggests that a
We
more
persistent distortion
close this section
interest rates.
is
present.
by studying an indirect implication of the view that the Gounders face lower
If capital is sufficiently
cheap for that community,
be prepared to accept the lower profit-margins associated with
its
indirect exports,
Our model is consistent with there being several other variable inputs, as long as
both communities.
^^Here production is the sum of direct and indirect exports.
29
members would presumably
all
of
and so maintain a
them have the same
price for
surplus stock of capital. Extra capital
reputation
when he
is
Gounders maintain a
of the life-cycle.
is
particularly advantageous
if
the exporter needs to build a
young. Indeed, we see in Columns 5-6 of Table 3 and in Figure 10 that the
significantly higher level of indirect exports
From Figure
10
we
see that the
Gounders
start
than the Outsiders at every stage
with about
25%
as indirect exporters, with the corresponding statistic for Outsiders around 20%.
life-cycle,
of their production
By
the end of their
the Outsiders focus entirely on direct orders whereas the Gounders remain with about
15%
of their production in the relatively low-paying indirect exports.
Insert Figure 10 here.
Conclusion
5
The evidence presented
in last
two sections seems to support our view that network-based lending
can generate significant distortions in the matching between
ability
and
capital.
demonstrating that there are distortions does not automatically establish the
and certainly does not of
better,
indeed
likely,
that
if
is
make a
fail.
rule,
investment in certain areas would collapse completely.
The importance
of returns paid
arises in part
of credit networks in India
is
something
It is possible,
On
and
the other hand,
because more formal channels of
clearly not
unconnected with the low rate
by the banking sector and the financial sector more generally, which
to the inefficiency of the (largely public) banking sector
rights
feasibility of
case for government intervention.
worth recognizing that network-based lending
lending
recognize that
the government were to ban network-based lending, for example by implementing
a non-discrimination
it
itself
We
in turn
is
related
and the very poor enforcement of property
and low standards of shareholder protection. Taking
seriously the idea that networks distort the
allocation of resources argues for measures to improve the functioning of the formal capital markets.^^
6
We
Appendix: Proof of Claims 2 and 4
begin by characterizing the payoff function.
From
above,
^^A similar argument has been made by La Porta, Lopez-de-Silanes and Shleifer (1998) who suggest that large jointfamily-owned conglomerates - a particular manifestation of lending networks - arise because the formal capital markets
function badly and generate significant efficiency costs.
30
V{Xi,r\a')
= m3xEf]6'-'[Xi{l + et)-r'Kt]
oo
= max E 5]; 6*-^Xi[{l + q) - r'zt]
t=l
If z^
represents the optimal value of
time),
Zt in
location
i
(we have already shown that
it is
constant over
we know from above that
t-i
1
Therefore
(-1
ViXlr\a^)=XiEj26'-'l[^izyil + e,),a')il+e,){l + et-r'z')
1
a
= Xi^S^-^Cp^y-^l + e - r'z')
i=l
where
e is
the
mean
of e^ and 7?
=
Efj,{z^/{1
+ et),a')(l + et).
It
follows that
and
l
+ e-r^z,
K.(Xi,r%aO=Xi^f-—-<574
(1-^)^
which imphes
V„(X;.r'.a')
= nXi.r'..')^ = ^'^^'-jy"^-
We are interested in characterizing the /(•)
equation
(1)
above and using equation
(3)
function. f'{a^)
which gives
31
us:
=
^
.
(2)
can be derived by differentiating
D
^
Since
^
l-(57Z(z(r2,a2),a2)
X^ = Xl and
=
T^(l,r2,a2)
^
1
V{l,r^,a^), this
is
—
67l(2(ri,ai),ai)
equivalent to
74(2(ri,ai),a^)
74(z(r2,a2),a2)
1
-
:/'
^{z{r'^,a^),a'^)
1
—
^(2;(ri,a^),Q:^)'
Using these expressions we can prove:
Lemma
4 Assume that
greater than
Then
1.
Proof: Using the
constant and that the elasticity of the function JT^ with respect to z
JT^ is a
the slope of the /(•) function is everywhere less than
first
assumption,
1.
easy to show that:
it is
{l-e-r^z'^)V{l,r\a^)
(l-e-ri2i)F(l,r2,Q2)
1-67Z2
=
f'
is
1-^^
l-e-r2z2
1 — e — r^z^'
Recall that 2*
is
determined by the equation:
r'
By
It
the assumptions
=
6E{fl,{^^,a')}V'{l,r\a').
we have already made, V^{l,r^,a^)
follows that since r^
>
>
r^, z^
z'^.
=
V^{l,r'^,a'^)
and
independent of
Jl^ is
of.
Moreover,
2
how
Therefore,
z.
If
less
Using
1.
is
greater than
Combined with the
this
E{ji,{^)}zJl
r'z^
E{TiA4^)]z^'
r2z2 compares with r^z^ depends on the elasticity of the function
the elasticity
than
r2z2
Lemma, we now
Proof of Claim
2:
/IJ
^jn bg smaller than r^z^ and conversely
1, r22;2
fact that /'
=
1
—
2
if
with respect to
the elasticity
is
2
Yzfzfifr) this gives us the result.
present:
Denote by
the density of a^ values of those
5(0:'^)
who
the density of a^ values of those
stay.
Now, by the
fact that
[0,1],
g{a^)
= Kf-\a^)
h{a')
=
K'fia')
32
who go
to 2
and by h{a^)
0^,0^ are uniformly distributed on
where K, K' are constants.
It follows that:
ff(a
+ e) =i +
/"''(")e
5(a)
/-H«)
and
h{a)
^
f'{a)
<l<r' (a)
fia)
Now
and
f{a)>f-\a).
Therefore
g(Q
+ e)
h(a
for all a. Therefore
Finally
we
g
FOSD
h.D
present:
Proof of Claim
3:
These conditions immediately
between the two locations would invest more
the fact that under these conditions the
(57iz(2%ci')}V'*(l,r%a*),
it
V{l,r\a^)
=
Jl^
>
7^^
tell
r'
(because the person
us that the person
he were to invest in location
1:
who
is
indifferent
this follows
order condition for investment choice reduces to
r' is
from
r*
—
associated with less investment. Moreover,
has already been shown that
v(i
follows tha,t
first
if
which implies that a higher
since under these conditions
it
+ e)
h{a)
g{a)
r^z'^
>
r^z^
and
an = l±l:ilil
is
indifferent
between the two locations, which impHes
V{l,r^,a'^))n
References
[1]
Baker, Christopher (1984).
An
Indian Rural Economy, 1880-1955: The Tamilnad Countryside,
Delhi: Oxford University Press.
33
[2]
Banerjee, Abhijit (1996). "Notes Towards a Theory of Industrialization in the Developing World,"
mimeo, M.I.T.
[3]
[4]
Banerjee, Abhijit and
(4),
Beck, Brenda (1972). Peasant Society in Konku:
A
New
Delhi: Vikas
[6]
pp
(4),
pp
631-53.
Study of Left and Right Subcastes in South
Pubhshing House.
Borjas, George (1987). "Self-Selection
Review, Vol. 77
and the Earnings
of Immigrants,"
American Economic
531-53.
Carrington, William, Enrica Detragiache and Tara Vishwanath (1996). "Migration with Endoge-
nous Moving Costs," American Economic Review, Vol. 86
[7]
Dual Economy, and Develop-
(1998). "Information, the
ment," The Review of Economic Studies, Vol. 65
India,
[5]
Andrew Newman
(4),
909-30.
pp
Cawthorne, Pamela (1995). "Of Networks and Markets: The Rise and Rise of a South Indian
Town, the Example of Tiruppur's Cotton Knitwear Industry," World Development, Vol. 23
pp
[8]
(1),
43-56.
Chari, Sharad (1997). "Agrarian Questions in the
India:
A
Making
Geography of the Industrial Present,"
Historical
of the Knitwear Industry in Tirupur,
in eds.
David Goodman and Michael
Watts, Globalising Food, London: Routledge.
[9]
Das Gupta, Monica
India,"
[10]
Economic Development
Cultural Change, Vol. 36 (1),
Policy, Baltimore:
The Johns Hopkins
pp
A
Rural
101-20.
Microeconometric Approach to
University Press.
Greif,
Economy,
Avner
Vol. 102 (3),
pp 437-67.
(1993). "Contract Enforceability
and Economic Institutions
Maghribi Traders' Coalition," American Economic Review, Vol. 83
[13]
in
Deaton, Angus and Christina Paxson (1994). "Intertemporal Choice and Inequality," Journal of
Political
[12]
&
Mechanisms and Population Retention
Deaton, Angus (1997). The Analysis of Household Surveys:
Development
[11]
(1987). "Informal Security
(3),
pp
in Early Trade:
the
525-48.
Hardle, Wolfgang (1990). Applied Nonparametric Regression, Econometric Society Monographs,
Cambridge: Cambridge University Press.
34
[14]
Holmstrom, Bengt and Jean Tirole (1997). "Financial Intermediation, Loanable Funds and the
Real Sector," Quarterly Journal of Economics, Vol. 112
[15]
Jackson,
Matthew and Asher Wolinsky
(1996).
works," Journal of Economic Theory, Vol. 71
[16]
Lucas, Robert (1978).
nomics, Vol. 9
[17]
(2),
(1),
Strategic
pp
pp
663-91.
Model
of Social
and Economic Net-
44-74.
Size Distribution of Business Firms,"
The Bell Journal of Eco-
pp 508-23.
Kranton, Rachel (1996). "Reciprocal Exchange:
Review, Vol. 86
[18]
"On the
"A
(3),
(4),
A
Self-Sustaining System,"
American Economic
pp 830-51.
Kranton, Rachel and Deborah Minehart (1998). "A Theory of Buyer-Seller Networks," mimeo.
University of Maryland and Boston University
[19]
Kranton, Rachel and Deborah Minehart (1999). "Competition for Goods in Buyer-Seller Networks," Cowles Foundation Discussion Paper #1232, Yale University.
[20]
Massey, Douglas, Rafael Alarcon, Jorge Durand and Humberto Gonzalez (1987). Return to Azatlan:
The Social Process of International Migration from Western Mexico, Berkeley: University
of California Press.
[21]
Piore, Michael (1979). Birds of Passage: Migrant labor
and
industrial societies, Cambridge:
Cam-
bridge University Press.
MIT
[22]
Porter, Jack (1996). Essays in Econometrics, unpublished
[23]
Rudner, David (1994). Caste and Capitalism in Colonial India: The Nattukottai Chettiars, Berkeley:
[24]
La
University of Cahfornia Press.
Porta, Rafael, Florencio Lopez-de-Silanes and Andrei Shleifer (1998). "Corporate Ownership
Around the World," mimeo, Harvard
[25]
Ph.D. dissertation.
Swaminathan, Padmini and
Industrial District in the
J.
University.
Jeyaranjan (1994). "The Knitwear Cluster in Tiruppur:
Making?" Working Paper No.
Studies.
35
126,
Madras
Institute of
An Indian
Development
[26]
Timberg, Thomas (1978). The Marwaris: From Traders
to Industrialists,
New
Delhi:
Vikas
Publishing House.
[27]
Townsend, Robert (1994). "Risk and Insurance
in Village India," Econometrica, Vol. 62 (3),
pp
171-84.
[28]
Udry, Christopher (1994). "Risk and Insurance in a Rural Credit Market:
gation in Northern Nigeria," Review of Economic Studies, Vol. 61(3),
36
An
Empirical Investi-
pp 495-526
.
v
CO
kl
o
(
o
o
^,—
£
S-H
3
'3
^.
to
o
.-I
«
O
«5
1
1
o
o
o
o
<^
00
CO
lO
I>
oi
CO
to
r-i
O
CIM
<N
OS
in
05
S
CD
00
i-H
CD
CD
CO
CO
c6
s
<t^
3
O
03
ki
<u
/-^
t>
TJ
o
q
in
a
(N
3
o
Ol
00
CO
o
o
d
o
00
O
m
i:
^
CO
0>
-I
4)
13
g
C
"S
05
CO
q
in
CO
cm'
IM
<4^
C
s
o
o
d
o
^
CO
1-1
»H
3
O
en
I.
lU
,
o
o
d
o
V
lO
CO
CO
T3
-*
CO
CO
Si
q
d
3
o
CO
CO
CO
(m'
in
CO
CO
CO
*
*
5? in CO
CO
CO
CO
O
CO
h
*
s
00
CO
s
'S
d
o
q
o
CD
CO
t-^
CO
00
r-i
CO
CM
CO
CO
t>
CO
-*->
3
O
oa
^
*
m
CO
o q 00
q q
<u
CO
c
3
O
o
S
t>'
*
00 Si
CO
CO
CO
i-H
T3
i-H
CO
CO
CO
in
CO
CM
CM
oi
CO
00
CO
CO
CO
*
*
o
o
o o
d d d d
CO
00
CM
05
CO
O)
*
o
S
* o
q
q
d d in CO
tH
00
CO
in
T—
d
in
CO
CO
o
a
en
1^
oo
CO CD 00
iq
CO
i> Tf (>i 1-i
2
C^l
o
p
^
C000t>i^oOio00
O^CQc^]'^
<M
1—
I
'm
-^
3
tH
ON
^o
t-^
CO
in
o
d
2S
q
r-
m
o\
CO
t»
00
^.
u
^
CM
CO
4)
XI
B
o
CO
in
1-H
in
1—
CO
q
q
CO
1-H
CO
3
o
CM
in
CO
fH
w
'^
CO
o
En
r-
r-:
O
C0C0t>00c^CV|c*^O
OC<3
Oit>
coco
^^
cn
d
*
in •*
in
-I
00
in
CM
«
*
O 9
d"
J2
CO «
SS
rH
CM
3,
C3
2
cs
CJ
u
T
I
cS
g,-2
is
u
ee
cc
es
c
c
c
a
X
o
«
<U
S
<D
«:
•c
m
o
a
'-'
3
c
u
c
ca
c
M
<
m
5
LI
(U
a
X
X
c;
c
.5
«
03
d
o
)—
-3
en
CS
eu
d
o
o
to
§
3
5
S.
H
od
TS
t
^
O
Ch
o
C9
Q
B
O
cs
—
>^
'S
^
-
01
W3
S
£
e
u
3
2
U
tn
o
cs
# ^
s
3
2
a
'
u
t
O
o
is
to
i-l
CO
O
O
05
00
^
O
CO
(N
CO
CO
05
odd
9 d
CO
1-1
o
;0
si
9d
o.
1^
00
05
CO
CO
05
CO
tH
1—
O
(D
ci
CO
(N
i-H
t>
;;j
*
O
d d
*
-I
CO
O
do
00
^7
CM
2;
CO 00
in
o
CM
t>
-*
I>
IV
"-HO
o o
CO
1-1
00
CM
8
d d d
tt>
O
1-1
OS
CM
O
o
O
1-t
1-1
•*
CM
m
5D
s
— O 01
2
p.
"
O
*«
•4->
rH o
(M
o
d o
d d d
o
00
CM
in
15
CO
05
1-1
00
s
o
en
tH
a>
s
T3
g
•*
CO
Si
t^
G
o
in
T)"
o
O o
CM
CO
n i>
o <t
CM o
d d d d
"
t;
-I
=^
in
CM
a
in
h
-*
(N
(M
•*
t~
in
in
CO
CO
o>
lO
CO
t>
^ o
co
o
CO
o
d d d
05
00
00
1-1
Ol
in
CO
i-<
CM
CO
CO
CO
CM
CO
rH
CO -*
/^
o * in
o
d d d d
1-1
O
O
00 in
o
d d d d
o
a;
tH
;z;
m
CO
*
I-l
o
d d
CO ;^
" d d
t-H
CD
O
d d
(N
.O
'^
d
«
d
Tf
l>
^ o
00
O
CO
05
CM
d d
I
CM
CM
CO
CM
r-i
m
O"
d
o
1
ST
Q
g"
H
CO
lU
_<u
&
H
O
.a
[14
d
o
O
« m
:z;
•z
a
<y
^
M 2 ^ ^
"o "o "o "o "o
O U O O O
o
^ S
^ S
t.
o
=?
^
CM
^
2 CO
O
d
o
00
lO
d
CO
05
05
00
Tt
d
CO
O
O
^
O 05
C~
o
o 00
o O
1-1
OS
CO
iM
in
"-^
^
o o
:
11 21
9 d
m
o
II
><
o
00
O
^
O 05
m
CO
in
o
CO
2
q §
9 d
^
o
•z
O
O
p
d o
«5
05
CO
oq
(1)
o
2;
o d
'3
a
o^
t.
o
o
d d
o
CO
f- to
a
o
o
•z
'Z
•s
"-I
H
c
-§
S
C
es
3
*
v>
o
o
u
C
c
<u
T3
C
a
Q
•c
a
•c
D
I
0)
X
I
X
o
Q
R
B
1
«2
C
A
P
/
k
/
.5
Figure 3
«7
1
^
Figure 4: Exports
1
1
1
1
1
1
/
6.5 :
y'
y
'
.___^c
a:
i
;
;
^.^
;
V,
«5.5
c
n
..:
:.
;
..-Hf
^^
^
-^
--^^5
-
1?4.5
_
-6
;
\...^.^^l,^^f^....^^^^ll....
,—
Outsiders
rr. ?v.
-
-
0)
L.Zdi'i
:
•c
o
Q.
X
LU
/
3.5 7
;
:
1
;
-
i
:
;
:
;
i
i
4
i
5
Experience:
6
bw=2
1
10
Figure 5: Exports
-
net fixed-effects
7.5
1
1
1
1
_^^'
Outsiders
6.5
:
;
:
^'C^^^:Z.
^
y.
(A
cc
ra
;
/'^///^-'''^
^ '
Gounders
o
-6 5.5
^•^'-
cu
1?
to
.y-yi^:-^
5
Q.
X
^y^ y
UJ
yA^yy^:^.
4.5
^
3.5
;
^—
^
''^
>
>""
y y y/ ^
yy /''
X
y
y y y/
^
i
1
4
1
5
Experience:
6
bw=2
1
10
Figure
1
1
1
Capital Stocl<
6:
1
1
1
1
:
4.5 -
'':
01
-^'
'
'.
'•
:
a
:
/
'
:
:
3.5
:
;
•
/
:
-^"
Gounders
-^
-
"
^
:
^
^^^^
^
/^
^
^,^-^''^
;
_
yr'-
:
:
/
i
''
^
-/^
-^
^-.""""^
/^
T5
'
:
y''
:
/
^^^^
^^^^-^^^''lOutsiders
T3
o
o
w
2.5
'•
^
-
2
q.
ns
O
2-y
r//^
^^-^ -^--
.--''''
1.5
^ -^
"
-^
;
-^
-^
•
;
:
:
•
•
:
:
;
-
;
'
i
i
i
i
4
i
5
Experience:
6
bw=2
i
i
10
Figure 7: Capital Stock
-
net fixed-effects
4.5
i
Gounders
4_.
^--^
''-
01
^
^3.5
-^
T3
.
0)
^-^^""^
^-"•'''^
y^ -'' -^
.
,<.'
^^
y
^
/
/
.'
,rf^
-^'
.^ i^'
'
>-
-^
"
^^
"^
'H^-^''''''^
^^
y
-^
^^
^ "^
"
-^
^-^
o.
^
y'
ro
O
"^
^y
<«^
--^
^'^'~
^^ y -^^ ^^'
^^.-^
""'^ ,-'"'
^'
-.'•''^
1
Outsiders
1
^^/^
^^
/ —^ y y
y
'
X/
y^
...^
z.^^
^' ^
^^^ ^'
•
,
-^
'^
U
o
\
^^
:
^^-^
.--
-
—
^ —
-^
^
"J^
^ ^^^^-^"^ ^
--
'
y'
^^y^^
^.
^
-^
^^
^
<n
T3
r
/
/
y -^j^ •* y
y
^
^/ y^ y
^x^
/
y
/
V-
^.-'•''''^
^^.."^
^'^
j^
^^ /
^""""""^
^ ^.>-f^_
.V-;
2.5
^^
^^
_J^
^^
^
^o^^^-^
'-^
1
^^
1
1
1
1
1
i
1
i
4
5
6
Experience:
bw=2
1
1
1
1
10
:
Figure 8: Capital- Export Ratio
^
net fixed-effects
-
i
1
I
1
—^^
*^
-1.5
v.
'^_
-2
re
Gounders
X.
:
q:
-c
o
x-2.5
•
-^
x-
"V-
LU
I
\
re
'5.
^
\
re
O
\
\
\
.\.
...Xw.
N.;
Outsiders
-3.5
;
~'.^^'-s,^;_-
1
t
i
4
6
5
Experience:
bw=2
-
1
10
"
Figure 9: Capital-Production Ratio
-
-1.6
1
1
.1.8
1
net fixed-effects
1
1
:
1
=—
\
.
^^^^
_. ^N>*-^
^; ~-
-..
^
Gounders
'.,..>».
^
^
-2.2 -
~^
^^^^"-^
< ^ ^^"\
"^^^
^
^
^
^\.
v;
''
"""^^
,
:
>X^X
^
^ '
r--^;
^..^.^-^...-v^
o
0-2.6
o
^^"-^^
^^
:
^-.^^^-.^;
2-2.4
^
^~^^^
""
,^-—-"^
---"'^"~^-.
'
^
m
5.-2,8
re
O
^X^i
1
>.^
-3.2 -
\
:
""^-v
Outsiders
:
-3.4
:
:
:
^ "^
^
-3.6
1
1
4
1
1
5
6
Experience:
bw=2
1
1
-v.
1
10
~
Figure 10: Production as Indirect Exporter- net fixed-effects
c
N.
:
'\
-^^
^^^
:
\^^
Exporter
o
Indirect
:
\
^
:
;
::x^x
X-^x:v
\X
X
^
:..":^
^^
X
^
\
>^....\
;\
_^'
^>
^^^
X^
X
Gounders
"^x
^
:
\
\V
"^
^
^ '^
ooi
Production
1
;
^^
as
I
:
'S
^\
en
1
/
o
(%)
1
1
•
^,
X,
^
'
--^-
-^
.
—
'
:
-
;
:
:
s.
of
N.
Share
en
X'~'--~^^;^\^^X^^^
Outsiders
o
^"^^5::::
-5
4
5
Experience: bw=2.5
6
10
5236
32
Date Due
Lib-26-67
MIT LIBRARIES
3 9080 01917 6582
„
-,
.
i'i!! i
„