Southern Economic
2006, 72(3). 578-596
Journal
Finance and Income Inequality: What Do the Data Tell Us? and Heng-fu Zou{
George R. G. Clarke,* Lixin Colin Xu,|
there are distinct conjectures about the relationship Although their explanatory little empirical research compares power.
between
finance
and income
the relationship 1995. Because financial
We
examine
inequality, between
for 83 countries between 1960 and and income inequality develop we use instruments to be endogenous, from the literature on law, finance, and growth might is less when financial development control for this. Our results suggest that, in the long run, inequality and Newman is greater, consistent with Galor and Zeira the (1993) and Banerjee (1993). Although at very sector development that inequality might increase as financial increases results also suggest as suggested sector development, low levels of financial and Jovanovic this (1990), by Greenwood
finance ment
is not
result results
thus
robust. We suggest
that financial development reject the hypothesis to improving that in addition financial growth,
benefits
the rich. Our
only
also
development
reduces
inequality.
JEL Classification: D3, G2, Ol
1. Introduction Recent
have
studies
shown
Because
financial
are
markets
sector
that financial
fraught
with
adverse
selection
financial
as collateral
are well
markets might
as
benefit
In contrast,
developed.
sector
the financial
for the rich, but not the poor, itmight worsen But excluded
this might from with
people revolution
getting talents,
not
loans,
the
Department,
case.
markets
World
As gain
might
ambition,
in financial
* Research
be
and
Bank,
(Levine
growth
borrowers
problems,
who
If financial
do have
that can be used
property
development
access
improves
inequality. sector
to it. In this Rajan
persistence.
is "opening
hazard
and moral
the rich
develops.
the financial access
economic
benefits only the rich and powerful.
therefore, find it difficult to get loans even
need collateral. The poor, who do not have this, might, when
boosts
development
that financial development
1997b).1 But many people worry
the
gates
and of
grows, respect, Zingales the
1818 H Street, NW, Washington,
the poor, finance
DC
might
(2003,
aristocratic
20433;
were
who
p. clubs
E-mail
be 92) to
previously
an equalizer argue
that
for the
everyone,"
as
[email protected];
corresponding author. t Research Department, World
DC 20433; Guanghua School of Management, Bank, 1818 H Street, NW, Washington, Peking University, Beijing 100871, China; E-mail lxu 1(g)worldbank.org. DC 20433. E-mail
[email protected]. | Research Department, World Bank, 1818 H Street, NW, Washington, We are grateful for suggestions of two anonymous referees. We thank Jerry Caprio, Robert Cull, Ross Levine, and Mattias Lundberg for comments. We are especially grateful to Thorsten Beck for early collaboration in this project, help in collecting and compiling data, and many useful discussions. The findings, interpretations, and conclusions expressed herein are those of and Development/The World the authors and do not necessarily reflect the views of the International Bank for Reconstruction Bank and its affiliated organizations, or those of the Executive Directors of The World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. Received October 2003; accepted June 2005. 1 For the relationship between financial development and growth see, among others, Beck, Levine, (2000). Loayza, and Beck (2000), and Rousseau and Wachtel
578
and Loayza
(2000), Levine,
Finance and Income Inequality
579
witnessed by the observation that, "in 1929, 70% of the income of the top 0.01% of income earners in the United
came
States
from
In 1998,
of capital....
holding
and
wages
income
entrepreneurial
made
up 80% of the income of the top 0.01% of income earners in the United States, and only 20% came from
capital."
the idea that financial development might benefit the poor, several theoret that income inequality will be lower when financial markets are better devel suggest show that when 1993; Galor and Zeira 1993). These models oped (Banerjee and Newman investments are indivisible, financial market imperfections perpetuate the initial wealth distribution, Consistent with
ical models
in a negative
resulting
between
relationship
financial
and
development
income
even
inequality
in the
run.
long
the relation between
Although that
possible
different
to a nonlinear
leading
at different
of
financial
development
and
levels
sector
financial
between
relationship
could be linear, it is also
inequality and financial development dominate
mechanisms
sector
development, Greenwood
inequality.
and Jovanovic (1990) show how financial and economic development might give rise to an inverted U-shaped relationship between income inequality and financial sector development. In their model, income inequality first rises as the financial sector develops but later declines as more people gain access
to the
system.
The relation between financial development makers?policy
growth. Although economic
want
makers
how
and income distribution affect
policies
is important for policy
as well
inequality
as how
they
affect
recent work has established a robust link between financial sector development and
and
less
1997b),
(Levine
growth
development
to know
inequality.
work
Understanding
on
has
focused
this
relationship
the
relation
will
allow
between
to assess
makers
policy
sector
financial
whether financial development will improve inequality and when it might be useful in doing so. Because different theoretical models give different predictions about the distributional impact of financial development on inequality, empirical investigation is needed to distinguish between the competing
conjectures.2
This paper analyzes the relation between the distributional impact of financial intermediary development and income distribution using data from developing and developed countries from between 1960 and 1995. Specifically, we analyze whether financial intermediary development affects income
inequality
different allow sector
and whether
mechanisms the
might to be
relationship to
development
endogeneity
using
results
impact
inequality
literature
(see,
show
that
inequality
for
level
of financial
levels causation
to financial
sector
example,
decreases
Zeira (1993) and Banerjee and Newman
2
because
inequality
financial
the
at different
Further,
or from for
on
depends
powerful
nonlinear.
instruments
development-growth Our
the be more
development
Levine as financial
(1993). Although
1997a, markets
could sector
Because
development.
of financial
sector run
in
from we
development,
suggested
development,
either
the
the we
financial control
financial
for sector
1999). deepen,
consistent
with
Galor
and
some weak evidence suggests that at low
in regressions looking at factors Li, Squire, and Zou (1998) and Li, Xu, and Zou (2000) include financial sector development that affect income inequality. This paper, however, differs from Li, Squire, and Zou (1998) and Li, Xu, and Zou (2000) in several ways. First, neither of these earlier papers is primarily concerned with the impact of financial sector development on inequality. Li, Squire, and Zou (1998) focus on explaining international and intertemporal variations in income inequality, whereas Li, Xu, and Zou (2000) focus on the relationship between corruption and inequality (and growth). They do not try to distinguish the various hypotheses as we do here; that is, they assume a linear relationship, and given their focus, they do not run a battery of specifications to examine the robustness of their results. In addition, they do not deal with the endogeneity of use a different measure of financial sector development financial development, that measures financial development less over and only include results from a pooled cross section. (M2 GDP), precisely
Clarke, Xu, and Zou
580
of financial
levels that
is, that
development is an
there
inequality
inverted
relation
U-shaped
as financial
increase
might
between
sector sector
financial
increases,
development
and
development
income
inequality, as suggested by Greenwood and Jovanovic (1990), this second result is not highly robust. We strongly reject the hypothesis that financial development benefits only the rich: We do not find a positive and significant relation between financial development In the next
we
section,
and
inequality
sector
of financial
the endogeneity
development. review
briefly
sector
financial
and inequality after controlling for
the
theoretical We
development.
on
literature
then
the
discuss
the
relation
between
that we
data
income
use
to
test
the
theoretical hypotheses in Section 3. After discussing the empirical specification and some estimation issues in Section 4, we present empirical results in Section 5 and conclude in Section 6.
2. Theoretical
on Finance
Perspectives
and
Inequality
Although most economists would not expect financial development in the long
the popular
run,
middlemen
serve
who
common
press,
of
the rich
a recent
of
chapter
and Marxist
literature,
interest
the
only
the first
that
some
book
theory
and well
to widen
often
depict
connected. the
defending
income inequality
system
are
views
these
Indeed,
free-market
as greedy
financiers
two
by
so
famous
economists, Rajan and Zingales (2003), is entitled "Does finance benefit only the rich?" One plausible reason why financial development might benefit the rich, especially when institutions are weak, is that the financial system might mainly channel money to the rich and well are
to offer
who
excluding
the poor.3 As financial sectors become more developed,
households
but
continue
sector
develops,
financial
able
collateral
to neglect
and who
be more
connected,
the poor
the poor
who
remain
might
are unable
to provide
to migrate
to urban
unable
to repay
likely
they might
while
to rich
even
as the or start
in education,
invest
areas,
loan,
a result,
As
collateral.
the
lend more
new businesses. This tendency might be reinforced if the rich are able to prevent new firms from getting access
to finance,
economic
income
and
development
the
Although
sector
financial
the
from
case,
and
entering,
we
would
see
to
expect
least at some
inequality?at
the ability
reducing
levels
of
a positive
of financial
to improve
the poor relation
development.
then
between
financial
We
this story
call
hypothesis of financial development.
the inequality-widening from
them
preventing If this were
lot.
arguments
previous
suggest
than
development
that
low-income
households
high-income
this
households,
is not
benefit
might
more
the case.
necessarily
As
financial markets become deeper, and access to finance improves, households that did not previously have access to finance might be the main beneficiaries. Because poor households cannot invest in human only
and their
households
physical own are
capital
resources, able
or bear they
to draw
on
will
the be
their
start-up
to do
unable
own
costs
resources
associated
with
so unless for
investment
they
starting can whatever
a new
business
the
using
In contrast,
borrow.
level
of
rich
financial
sector development. Therefore, capital constraints might be less binding for rich households at any level of financial sector development, and so they might gain less when these constraints are
loosened. Several
recent
market
theoretical
models
have
imperfections might and Newman (1993) and Galor and Zeira 3
This paragraph mainly
this
formalized
increase income
draws from Rajan and Zingales
intuition,
suggesting
that
capital
inequality during economic development. Banerjee (1993) suggest that capital market imperfections and
(2003), chapter
1.
581
Finance and Income Inequality
in investment in human or physical capital may lead to divergence of income for the rich and the poor even in the long run. Further, depending on the initial wealth distribution, these indivisibilities
imperfections
and Zeira
who
agents
given
inequality
persists
a two-sector
model
through
will
be
to the next
bequests
can
this
run.
long
between
bequests
capital
to make
able
in the
This
results
market imperfections and an initially unequal distribution of wealth will maintain more
grow
and Newman
larly, Banerjee indivisible
require
a similar
than
slowly
Because
of
inequal
with
capital
of wealth.
two
of
rich
only
imperfections,
invest
this inequality and
distribution
in which
model,
market
capital
initial
equitable
a three-sector
construct
(1993)
investment.
a more
with
economy
the
in income
an economy
In their model,
generation.
sector.
than
larger
bequests
investment.
where
generations,
in a skill-intensive
work
with
individuals
only
imperfections,
can borrow
even with
in human
investment
market
capital
is perpetuated
ity that
income
construct
indivisible
or who
amount
that
(1993)
an
make
However, ment
mean
might
Galor
the
Simi
technologies can
agents
borrow
enough to run these indivisible, higher-return technologies. Once again, the initial distribution of has
wealth
on
long-run
effects
With
all else
imperfections.
income
remaining
distribution
and
these
equal,
in
growth
models
the presence
of
suggest
market
capital
with
that countries
larger
capital
market imperfections, that is, higher hurdles to borrow funds to finance indivisible investment, should have
income
higher
financial thesis
inequality. and
development of financial
(1990)
but
a theoretical
present
should
We
inequality.
a
observe
relationship
between
inequality-narrowing
hypo
negative
this hypothesis
call
the
development.
a related,
Offering
we
Consequently,
income
different,
that has
model
on
perspective elements
these
of both
basic
ideas,
ideas.
In their model,
and
Greenwood
agents
Jovanovic the
operate
more profitable, but more risky, of two technologies only when they can diversify risk by investing in financial
coalitions.
intermediary
less
thus
and of
members
accumulate
intermediary
inequality.
However,
litions,
resulting
in
Jovanovic's ity and
(1990)
eventually
hypothesis
inverted
are,
thus,
and income
in or
hump with
in the
long
different
the
relationship first
as more of financial
among
coa
these
join
and
Greenwood
between
income
and
increasing
then
inequal decreasing
financial
coalitions.
We
relation
between
financial
intermediaries
these three hypotheses the access
improving
in
increase
call
this
development. the
about
and
join
people
eventually
Consequently,
inequality
predictions
in rich
trend.
upward
(high-income) in an
resulting
all agents
U-shaped
distinguishing
households
is fixed,
fee
with
that poor individuals
between
widen,
income run
is correct,
will
inverted
hypothesis
U-shaped
hypothesis
low-income
a
predicts
quite
(low-income)
reversal
differences
outsiders
associated
fees)
membership
income
slowly,
the entrance
development,
inequality. Yet
inequality-narrowing and benefit
eventual
stabilizing
the
There
and
because
model sector
financial
before
an
more
wealth
coalitions
income
(e.g.,
individuals from joining them. Assuming
these coalitions prevent low-income save
costs
the fixed
However,
in poor
countries
to finance alike.
is important. If the
would
reduce
In contrast,
if the
inequality inverted
U
shaped hypothesis
is correct, improving the access to finance might initially worsen income inequality
in poor
improving
countries,
it only
the
after
has
country
passed
a certain
stage
of
financial
sector
development. Finally, if the inequality-widening hypothesis is true, some countries might be trapped in a high-inequality world that would be only worsened by financial sector development. In what follows,
we
use
data
from
a broad
cross
section
of
countries
between
1960
and
1995
to assess
the
empirical validity of the different hypotheses. It is perhaps useful to note that the inverted U-shaped hypothesis concerns a situation in which the empiricist observes the evolution of income inequality and financial development during short-
the
development
or medium-run
process. time-series
Thus, or
panel
the
relationship
data.
In
would
contrast,
be testing
most the
likely
to
show
inequality-widening
up
in and
Clarke, Xu, and Zou
582
inequality-narrowing time
long
hypotheses
might
require
as cross-sectional
such
data,
long-run
data
on
based
series.
3. Data This section describes our indicators and data for financial intermediary development and income inequality as well as the set of conditioning information. Table 1 presents descriptive statistics and correlations.4 The income inequality data are based on a new data set of Gini coefficients compiled by Deininger and Squire (1996) and extended by Lundberg and Squire (2000). Although the original data set contained over 2600 observations, Deininger and Squire (1996) and Lundberg and Squire (2000) limited the data set by imposing several quality conditions. First, all observations had to be national
from
household
for
surveys
the national
of
sentative
population. own
accounted
for,
including
To
explore
whether
or
expenditure all
Third,
income.
sources
the
Second,
income
of
of
to be
had
coverage
uses
and
repre to be
had
expenditure
consumption.5 is an
there
inverted
U-shaped
between
relationship
economic
development
and income inequality, as proposed by Kuznets (1955), we regress the logarithm of the Gini coef ficient on the log of real per capita GDP and its square. Figure 1 shows the result for the panel sample. The
for alternative
of an
the existence
suggests
graph
income
of
explanations
inverted
curve.
U-shaped such
inequality,
this graph
However,
as financial
not
does
control
depth.
recent literature on the relationship between financial intermediary development and economic growth has developed several indicators to proxy for the ability of financial intermediaries to The
identify
profitable
resource
mobilization.
GDP banks ment
monitor
and
concentrate
on
projects, We
and nonbank
private
sector
access
to loans
a measure
intermediaries
enterprises have
agents
access
sector
of financial
of financial
sector
a good
seems intermediation
and
management, financial
by
development showing
on
central
facilitate over
intermediaries
economic
that growth
as
banks
proxy
and
growth
have
and
the extent
used
in countries
govern
or
1990)
that have
studies this
from
to which
Jovanovic
recent
1993). Many
is faster
lenders
for
variable
(as in Greenwood and Zeira
1993, Galor
development,
sector
excludes
(but which
to financial
risk
comprises credit to private firms and households
as borrowers),
and Newman
(as in Banerjee
at the effect
looked
financial
state-owned
ease
managers, to the private
credit
indicator, which
(private credit). This and
control
as
variable
where
private
credit is higher (see, for example, Beck, Levine, and Loayza 2000; Levine, Loayza, and Beck 2000). To
assess
the robustness
of results,
we
use
a second
measure
of financial
development:
claims
on
sector by deposit money banks divided by GDP (bank assets). In contrast to private credit, this measure excludes credits by nonbank financial intermediaries and includes the nonfinancial domestic
credit
4
to governments
and
state-owned
enterprises.
Australia, Austria, Belgium, Burkina Faso, the Bahamas, Bolivia, Botswana, Brazil, Cameroon, Canada, Chile, China, Colombia, Costa Rica, Cote d'Ivoire, Cyprus, Denmark, Dominican Republic, Ecuador, Egypt, El Salvador, Finland, France, Gabon, Gambia, Germany, Ghana, Greece, Guatemala, Guinea Bissau, Guyana, Honduras, Hong
The sample includes Algeria, Argentina,
India, Ireland, Italy, Jamaica, Japan, Jordan, Kenya, Korea, Luxembourg, Madagascar, Malawi, Mali, Netherlands, New Zealand, Nicaragua, Niger, Nigeria, Norway, Pakistan, Panama, Paraguay, Peru, Philippines, Portugal, Senegal, Sierra Leone, Singapore, South Africa, Spain, Sri Lanka, Sudan, Sweden, Thailand, Trinidad and Tobago, Tunisia, Turkey, Uganda, United Kingdom, United States of America, Venezuela, Zambia, and Zimbabwe.
Kong
(China), Indonesia,
Malaysia, Mexico, Morocco,
5
sampling methods, we adjust the data using a method suggested by Deininger and Squire (1996) and also applied by Li, Squire, and Zou (1998) and Lundberg and Squire (2000). Specifically, Deininger and Squire (1996) find Gini coefficients. We, a systematic difference of 6.6 points between the means of income-based and expenditure-based therefore, add 6.6 points to the expenditure-based Gini coefficients.
To account for different
Finance and Income Inequality Table 1. Descriptive
Number
583
Statistics Gini Coef.
Private Credit
Bank Assets
205 38.4 22.4 61.1
205 44.9 1.6 202.8
205 42.4 2.5 132.1
Initial GDP per Capita
Risk of Ethno-linguistic Fract.
Gov't Cons.
Exprop.
Inflation Mod Sect. Rate Val. Add.
of
observations Mean Minimum Maximum Gini coefficient Private credit
1.00 -0.38
1.00
Bank assets
(0.00) -0.48
0.86
Initial GDP
(0.00) -0.59
(0.00) 0.69
0.61
per capita Risk of
(0.00) -0.59
(0.00) 0.64
(0.00) 0.66
205 5552 160 20,367
(0.15) -0.48
Inflation
(0.00) 0.33
0.32
rate
-0.28
0.78
(0.00) 0.37
(0.00) 0.61
sector
-0.25
0.55
1.00
(0.00) 0.50
-0.36
(0.00) (0.00) -0.22
-0.30
1.00
(0.00) -0.45 -0.50
(0.00) (0.00) (0.00)
Modern
205 205 1.15 86.0 1.00 43.3 3.22 99.6
1.00
(0.00) (0.00) (0.00)
consumption
205 14.3 5.6 27.9
1.00
(0.00) (0.00) (0.00) 0.11 -0.38 -0.35
expropriation Ethnolinguistic fractionalization Government
163 0.25 0.00 0.86
205 7.3 3.3 10
-0.21
0.54
0.68
-0.21
-0.03
(0.00) (0.00)
value added/GDP (0.00) (0.00) (0.00)
1.00
(0.00)
0.67
(0.72) -0.68
(0.00) (0.00)
1.00
(0.00) 0.47
(0.00)
-0.06
1.00
(0.00) (0.43)
Gini coefficient from Deininger and Squire (1996) and Lundberg and Squire (2003). Gini, measurement-adjusted GDP per capita, real per capita GDP; Source: Loayza et al. (1999). Private Credit, claims on the private sector by financial institutions divided by GDP. Source: Beck, Demirgii?-Kunt, and Levine (2000). Bank Assets, claims on domestic nonfinancial sector by deposit money banks divided by GDP. Source: Demirgii?-Kunt and Levine (2000). Risk of Expropriation, index indicating risk of expropriation through confiscation or forced nationalization. Higher values indicate that risk is lower. Source: PRS Group (2003). Fractionalization, average value of five indices of ethnolinguistic fractionalization, with values ranging Ethnolinguistic from 0 to 1, with higher values indicating greater fractionalization. Source: Levine, Loayza, and Beck (2000). Government Consumption, government consumption as share of GDP. Source: World Bank (2004). Inflation Rate, log difference of Consumer Price Index. Source: International Monetary Fund (2002). Modern Sector Value Added/GDP, value added of service and industrial sectors as share of GDP. Source: World Bank (2004).
use private credit rather than the ratio of money
We a measure
used
commonly
to measure
financial
sector
and quasimoney
development
(M2) to GDP
and Levine
(King
1993;
(M2), Levine
1998), for several reasons. First, the ratio of M2 to GDP includes the liabilities of central in addition to banks and other financial intermediaries. Second, it includes credit to
and Zervos banks
governments
and
development
than private
Our
sample
state-owned
shows
enterprises.
Because
of
this,
it is a less
clean
measure
of financial
sector
credit. a
large
variation
in financial
intermediary
development.
Private
credit
ranges
to over 200% in Japan (1990-1995). The indicators of (1990-1995) financial intermediary development are positively and significantly correlated (see Table 1). The pairwise correlations indicate that income inequality is lower in countries with deeper financial markets; financial sector development is significantly and negatively correlated with theGini coefficient. Plotting from 2% of GDP
in Uganda
the logarithm of the Gini coefficient and its fitted value (from the regression of the logarithm of the
Clarke, Xu, and Zou
584
o log(Gini)
A fitted
value
4.5
3.5 0oo0o%0?0?
o
3H "i-r~
6
4
8 per
log(GDP
10
capita)
and log(GDPper 1. Log(Gini) Figure capita) in a panel of 91 countries. The fitted line is from a regression of log(Gini) on the log of real per capita GDP and its square. All data are averaged over seven 5-year periods between 1960 and 1995.
Gini coefficient on the logarithm of private credit) against the logarithm of private credit, Figure 2 sug gests
a negative,
and possibly
4. Empirical To
we
explore estimate
the
between
relationship
the following
financial
intermediary
discussed
claims
previously,
on
sector
the private
by
(private credit) and claims on the nonfinancial domestic GDP
(bank
f(Financeit)
are
assets)
based
which,
the measures on
earlier
and
development
income
regression:
In (Gini Coef.it) = ^o + /(Financeit) As
two.
the
between
Framework
further
inequality,
relation
nonlinear,
of
sector
financial
discussions,
we
QL\iFinanceit
+
(i)
+ a2CV;, + e,-,.
financial
as percentage
institutions
sector by deposit money The
development.
assume
has
focus
the following
of GDP
banks divided by the
of
functional
is
analysis form:
a^Financel.
The inequality-narrowing hypothesis predicts an < 0 and oti2= 0, the inequality-widening hypothesis = 0, and the inverted predicts an > 0 and ai2 U-shape hypothesis predicts an > 0 and ai2 < 0. In addition
to the financial
sector
we
variables,
include
several
to control
variables
for other
factors
thatmight affect inequality. Specifically, we include linear and squared terms of the log of (initial) real per
capita
GDP
to control
for a direct
that is independent of financial f(Financeit) steady-state convergence.
captures
the effects
situations,
initial
However,
because
"Kuznets
per
would capita
on
steady-state
Once
capture GDP
is highly
has
correlated
controlling If the
inequality.
whatever
on
development
intermediary development.
of finance GDP
of economic
effect"
been with
real
achieved financial
income
inequality
for initial GDP, data by sector
do
not
the
reflect
force
development,
of
and Income Inequality
Finance o log(Gini)
a
fitted
585
value
4.5
Qd
?&A1
^
o O
?o
? O
oOdpo O,? _ fi
M|ftA
3.5
3H "i-r
4
2 log(PRIVATE CREDIT)
log(Private Credit) in a panel of 91 countries. The fitted line is from a regression of log(Gini) 1960 and 1995. and its square. All data are averaged over seven 5-year periods between
against Figure 2. Log(Gini) on the log of Private Credit
that our
sure
to make GDP
per
capita
to these
In addition inflation
rate,
than
the rich
their
exposure
latter
to inflation.6
We,
a measure
and
income
inequality
across
available
It is less
time,
income
decrease against
include
For by
inflation
of
government
consumption,
ethnic
of
ethnic
diversity
to the
same
although
it could
also have
rights
fractionaliza We
might
effect,
expect
if, for example,
This
was
variable
not
all periods. protection
of property
the opposite
to hedge
them
is greater
for
more
relatively
ethnolinguistic expropriation).
value
the protection
class
the
coefficient.
is greater.7
and property
include
We
that allow
fractionalization
where
consumption
the middle
a positive
to have
it is set equal
example,
the poor,
and
between variables.
these
variables.
instruments
(the risk
rights
where
in countries
government
inequality.
expropriation
expect
in countries
therefore,
and,
whether
to financial
of property
the protection
to redistribution
clear
therefore,
the poor
access
omitting
control
additional
hurts
to the multicollinearity
respect the model
estimate
several
instability better
measures
to be higher
are averse
people
rich
of
have
also
include
that monetary the
we
we
with
is robust
we
development,
measures,
conjecturing because
Additionally, tion,
test of the three hypotheses
and financial
rights
will
increase protect
might
that is, protecting
or the
the poor
against exploitation by the rich. Similarly, ifmost redistribution through the tax and transfer system is toward
low-income
groups,
government
consumption
might
result
in greater
equality.
However,
it
could also have the opposite effect if rich households use their political power to exploit the poor. 6 7
See, for example, Easterly and Fischer (2001). Consistent with this, Alesina, Baqir, and Easterly ( 1999) find that spending on productive public goods (e.g., on schools) is lower in fractionalization was unavailable for many of the countries in our U.S. cities where ethnic diversity is greater. Ethnolinguistic on the other regressors for countries with missing data. Results we values based To excessive avoid loss, imputed sample sample. were robust to using other imputation techniques, including hotdeck imputation (Mander and Clayton 1999) and multiple impu tation (Royston 2004). Although the hotdeck approach was used for all regressions, themultiple imputation approach could be used only for OLS regressions. Results were similar in terms of size and statistical significance for the coefficients on the finance variables. Results were also similar for those coefficients when we simply dropped ethnolinguistic
fractionalization
from the estimation.
Clarke, Xu, and Zou
586
Kuznets
income
a variable
include
inequality
The
might
correlation
on
depend
the share
representing
to agriculture).
(as opposed
industry
that
suggests
we
Thus,
economy. and
(1955)
of value
the
added
the modern,
of
sectoral
of
for by
services
accounted
that
an
structure
is, nonagricultural
sector,
share of GDP and GDP per capita indicates that richer countries have largermodern sectors. Although the simple this
correlation
After
entire
conduct
the
long-term
contrast, ity and,
the long-term
of most
the convention
Following
analysis
using and
cross-country
empirical
rather
shorter
than
for most
basis
time
because
spans in our
countries
financial
although
sample,
be
they might
The
panels.
between
finance U-shaped the panel
to business
of testing
and
inequal
analysis
splits
use 5-year periods on
a
yearly
that are
fluctuations
cycle
In
hypothesis.
are available
data
intermediary
the
hypotheses.
5-year periods. We
subject
a way
offering
studies,
over
cross-sectional
the inverted
panel
the sample period 1960 to 1995 into seven nonoverlapping
significant.
averaged
inequality-widening
to test
in which
setup
sectors.
agricultural
data
inequality,
is negative,
and
using
comovement
of
larger
five-year
and
coefficient
positive
analysis,
finance
the process
the Gini and
becomes
cross-sectional
between
appropriate
and
inequality
correlation
inequality-narrowing
examine
might
be a more
might
of GDP
greater
data
relationship in the
featured
analysis
therefore,
a pure
and a panel
1995,
share
have
the partial
in two ways:
and
relationship
the panel
sector's
countries income,
capita
1960
capture
might
the modern poorer
the analysis
between
period
analysis
because for per
controlling We
between
to be
appears
controlled for by averaging over longer time periods. All panel regressions include time dummies to account we
for
structural results
present
from
random
bias
OLS
because
To
periods.
effects
account
take
of
does
not
allow
for
the possibility
reverse
of
who
affects
income
financial
we
Because we
inequality, similar
development
to join financial
is able
sector.
the financial
use
an
are
variables
in Levine
development
approach,
(1997a,
on economic
and,
in the effect
interested
primarily
used
coalitions
intermediary
instrumental
to the ones
intermediary
of
the data,
affect
might sector
the size of on
development
for financial
the exogenous
sector of
impact
are a set of dummy
instruments
the
suggested in some of the the initial distribution of
instruments assesses
which
The
growth.
therefore,
of financial
adopting
1999),
is, for
causality?that
possibility that inequality affects the provision of financial services?something theoretical models. For example, inGreenwood and Jovanovic's (1990) model, wealth
structure
the panel
estimation.
1 using ordinary least squares (OLS) (or random effects) estimation might
Estimating Equation introduce
across
differences
variables
proposed by La Porta et al. (1998) that identify the origin of the country's legal system.8We use the legal because
(1998), differences
rather
variables,
dummy
origin
are
they
in legal
the measures
than
available are
origin
for
a wider
significantly
of
creditor
sample
of
to financial
related
also
rights,
proposed
Several
countries. sector
papers
development,
by La
Porta
have
et al.
shown
perhaps
that
because
different legal traditions put different levels of emphasis on the rights of property owners or because some
systems
examine
8
are more
the validity
of
adaptable the
to exogenous
instruments
using
than
changes Hansen's
J-test
others.9
to test
In the empirical
the overidentifying
analysis,
we
restrictions.10
The measures
of legal origin were taken from the Global Development Network Growth Database produced by William Sewadeh (see Easterly 2001). Easterly 9 and Levine (2001) provide an excellent summary of much of the empirical and theoretical literature on Beck, Demirgii?-Kunt, this topic. La Porta et al. (1998) show that protection for corporate shareholders and creditors are strongest in common law and Mirvat
in French Civil Law countries. La Porta et al. (1997) relate these variables to some measures of capital (external market capitalization over GDP, number of listed firms per capita, initial public offerings), development showing that they are generally lower in civil law (especially French Civil Law) countries than in common law countries. and Levine (2001) show that private credit is lower in French Civil law countries than inGerman Civil Beck, Demirgii?-Kunt, countries and weakest
market
Law and common law countries. 10 In similar regressions of financial sector development on economic restrictions are valid. hypothesis that the overidentifying
growth, Levine
(1997a,
1999) fails to reject the null
Finance and Income Inequality Results
5. Empirical
from Cross-Sectional
Relationship
Long-Term test
To
587
the
and
inequality-widening
the
Samples
inequality-narrowing
we
hypotheses,
the natural
regress
log of the Gini coefficient on linear terms for the measure of financial sector development (private credit) and the additional control variables. Before we control for the possible endogeneity of the measures
sector
of financial
statistically
significant,
on private
the coefficient
development,
that
indicating
in countries
is lower
inequality
credit
but
is negative
statistically
on banks assets is also negative but is
1 of Table 2). The coefficient
insignificant (see column
where
are greater
assets
bank
as
a share of GDP (see column 3 of Table 2). Results for both measures are qualitatively similar when we omit per capita GDP and per capita GDP squared from the regression (see Table 3). These results, as we
discussed
not
do
earlier,
take
account
into
the
issue
of
the endogeneity
the finance
of
variables.
After controlling for endogeneity using the indicators of legal origin as instruments, the coefficient on
credit
private
remains
are
5). Results
similar
but
negative when
in size
increases
and becomes
statistically
as the measure
are used
assets
bank
(see column
significant sector
of financial
(see
development
column 7). This suggests that financial sector development reduces income inequality, supporting the inequality-narrowing hypothesis and rejecting the inequality-widening hypothesis of financial development. favoring
as
development dummies
origin that
tests
Hypothesis the results
are
they
from
the null
reject
the 2SLS
regressions,
endogenous.11
In addition,
are uncorrelated
with
we
the error
instruments
appropriate
that
hypothesis consistent
with
are unable term
(see Hansen
after
the financial
to reject
for in the
J-Statistics
variables,
relevant
financial
that
hypothesis
the other
exogenous,
that view
papers
the null
controlling
are
variables
the theoretical
tables).
the
legal
suggesting on
Based
the
coefficient estimates in column 5, a 1% increase in private credit decreases the Gini coefficient by 0.31%. Results are similar when per capita GDP is omitted (see Table 3), with the point estimate of the parameter
the
the measures
at -0.27.
smaller
slightly
test
To
inverted
U-shape
of financial
sector
of financial
hypothesis
(see
development
we
development,
columns
6 and
include
8). Because
squared
terms
for
on
the
the coefficient
squared term is statistically insignificant in all model specifications, the results do not support this hypothesis. Although the coefficient on the linear term becomes statistically insignificant in both model the
when
specifications linear
and
development
squared is treated
development
does
the squared terms
are
jointly
as endogenous.
affect
term
inequality,
is included,
significant
to note
it is important at a
1%
level
Thus,
these
regressions
it appears
to do
so in a roughly
that
the coefficients
when
financial
that although
financial
or higher
suggest linear
fashion.
data might provide a better way of testing the inverted U-shape hypothesis or medium-run
short-
After
controlling
variations for
in the comovements
the endogeneity
of
of financial
the financial
sector
sector the panel
However,
if panel data better capture
development variables,
on sector
many
and of
inequality. the coefficients
on
the other control variables are statistically insignificant (see column 5 of the relevant tables). Although the coefficients on the linear and squared terms for initial GDP per capita are statistically insignificant, they are jointly significant inmost model and
the negative
11 When
coefficient
on
the
squared
specifications.12 The positive coefficient on the linear term term
suggest
an
inverted
U-shape,
with
income
inequality
we perform a Durbin-Wu-Hausman test using an auxiliary regression (see Davidson and MacKinnon 1993), the null that credit" is is rejected at a 1% significance level (p-value = 0.001). For "bank assets," the exogenous hypothesis "private null hypothesis that it is exogenous is rejected at a 5% significance level (/7-value = 0.043). 12 are a at 1% level or higher when bank assets are included, jointly significant at a 10% level when They jointly significant private credit is included linearly, and statistically insignificant when private credit is included linearly and in squared terms.
588
Clarke,
Xu,
and
Zou
o ^
r^ (N (N^ rj- ^ r- on m i- O 0\
dodo
(N
OO Tt (N^ r- ^ 00 tj- vo O oo ^o in 'o -h' O -H d' w I
00
00 00
O
in ^o es e^ 00 (N (N O -h' d -h' o
?n
d I
, m ^H (N O
r- (N ^h (N O O (N
odd
d I
8
I
cd C O "
Oh ? O
Ph Q O
o
.5 g
1/3^^ 2
g3
CD^-( fi ?t h C W >m r-1 r*
'S
S i 5I i h D 43 > ^^ O
C3 +j cd fi c "^ Jd ^H V>-'
Finance
T-H ON OO /?s 00 ^ ^h en ^h on tj T-H m O m o ? odd -^ ? ?
-Mon ^o ^ Hinm O ^ O ^
?
and
c ON
-H in
in m r-H^H O O
in O
00 -ni es oo ^ i?i O
^ en ^ 00 ^h i-? i-i
en
d
-M-
ON o ^H O
O ON
odd
*
-ni r^ ^ tj- ^ i? oo m en in es o oo m o on p "si;p T-H en es d
-MH? in on es ^ o ^ 1-Hen On oo es o ^H ^H m o O C^ d m* es es' d d
.s O
o ON /-*so ^ m ^h in o r^ 1-H en r^ p en es es o
es o in o
d d
O in o O ^ ON^ ^H r-H VO VO O O Ti" r-H es es o o m
-mON on ^ es o 00 p es d
-h 00 in ^h 00 es
o
/-s oo m r ^t Tt es es d d
CD c 3 c
c? CD cd
'S? .s s
CD /-s CD
o O o ? o
C O U