This paper is forthcoming in the NBER Macroeconomics. Annual. I am indebted to Jess Benhabib,. Jordi Gali and Julio Rote
NBER
WORKING
INEQUALITY
AND
Ronald
NBER
NATIONAL
PAPER
GROWTH
B6nabou
Working
BUREAU
SERIES
Paper 5658
OF ECONOMIC
1050 Massachusetts Cambridge,
RESEARCH
Avenue
MA 02138
July 1996
This paper is forthcoming
in the NBER Macroeconomics
Jordi Gali and Julio Rotemberg
grateful for their detailed comments as well as to Alberto Raquel
Femandez,
Alesina, Herschel
for providing
assistance
from the C.V,
research program
Grossman,
1996
by Roland
Fluctuations
B6nabou.
Torsten
Persson
acknowledged.
and Growth.
All rights reserved. permission
Bertola,
I am also Rios-Rtill,
Patrick Bolton,
and Thomas
Piketty.
me with their data set, and Tim
the Luxembourg
Bureau of Economic
paragraphs, may be quoted without explicit is given to the source.
series from
to Jess Benhabib,
Perotti and Jos6-Victor
Bernard, Giuseppe
Newman,
Starr Center is gratefully
in Economic
Roberto
Andrew
Andrew
I am indebted
at early stages of this project.
and Lynn Squire for providing
me with specific
author and not those of the National O
to my discussants,
Abhijit Banerjee,
I also wish to thank Klaus Deininger Smeeding
Annual.
for helpful conversations
Income
Study.
Research
This paper is part of NBER’s
Any opinions
expressed
are those of the
Research.
Short sections provided
of text, not to exceed
that full credit, including
two
O notice,
NBER
Working
Paper 5658 July 1996
INEQUALITY
AND
GROWTH
ABSTRACT
Using
two unifying
main theories
linking
The first model for explicit
models
income
integrates
and an empirical
distribution
the political
of
simultaneously investment
redistribution depress
captures the essence
The second
accumulation.
groups’
rent-seeking
abilities,
political
power.
conflict,
the empirical Finally,
empirical
issue:
is surveyed
and discussed
the possibility
of multiple
steady-states
are cross-country
is there convergence
B6nabou of Economics
New York University and NBER
NY
10003
differences
in inequality
not only in first moments
the growth costs policies
the wealth
constraints
which
impede
of the prisoner’s
dilemma
which
reduces the security of property
rights,
rate is shown
to fall with interest
rich and poor.
in the relative
of political
It allows
such
growth
evidence
distribution?
New York,
but inequality
evidence.
schooling:
version
conflict
between
the
or public
as well as with the gap between
For each of the three channels
analysis.
Department
is a growth
The economy’s
per se that matters, however,
Equivalently,
and ameliorate
empirical
capital markets theories.
the tradeoff
land reform
of theories where sociopolitical
discouraging
as well as the relevant
and embodies
model
this paper presents and extends
and imperfect
taxes,
incentives
thereby
Roland
economy
through
savings
by the poor.
inequality
and growth,
departures from perfect democracy
and benefits
exercise,
It is not income
distribution
economy,
of earnings
capital markets
in conjunction
and
and social
with the theoretical
leads me to raise and take up a new permanent,
or gradually
(GDP per capita),
narrowing?
but convergence
in
1
Introduction
1.1
Korea
To introduw beginning respect
the theme
of his article.
to all major
In the early 1960’s,
-nomic
manufacturing,
but in exports
and manufactures, antury
the Philippines
The
enrollment.
Given
Korea
nearly
If one looks beyond
identical
population,
had a somewhat
starting
points,
growth
urbanization,
of primary
about
at the
primary
share of GDP
how can it be that
averaging
(1993)
were similar with
higher
had the same proportions
“miraculous”
at about
in
commodities over the next
6% per annum,
while
2%?
first moments,
shown by Table 1, the distribution lGini(%)ll
per capita,
Philippines
both muntries
experienced
stagnated
GDP
All
raised by Lucas
South Korea and the Philippines
aggregates:
school
Not So Similar After
I shall revisit the puzzle
of this paper
and secondary
quarter
circa 1960:
and the Philippines,
however,
initial conditions
of income was considerably
Q1 [
Q2
I Q3
I
Q4
I
Q5
were in fact quite different.
more unequal I Q3+Q4
As
in the Philippines:
I Q5/Ql
I Q5/(Ql+Q2)
1965
Korea
34.34
5.80
13.54
15.53
23.32
41.81
38.85
7.21
2.16
Philippines
51.32
3.50
12.50
8.00
20.00
56.00
20.50
16.00
3.50
Korea
33.64
7.39
12.29
16.27
21.81
42.24
38.08
5.72
2.15
Philippin=
45.73
5.20
9.10
13.30
19.90
52.50
33.20
10.10
3,67
1988
Table 1: Korea and the Philippines Sours:
Deininger
The Philippines’ seventeen
Ginis, or 2.5 among
points
higher,
East-Asian
top 2070 to the bottom
data set; Qi denotes
about
characterized
1The year
regime.
Ferdinand
1965 is the earliest
based on the same survey method
land ownership:
is for gross personal
income,
quintile.
in the world distribution
the ratio of the income
the Gini coefficient
be blamed
was first elected
one for which data household
for
was of
share of the
for farmland
was 38.7 for
in 1961 and 1960 respectively. 2 This greater concentration
Marcos
(gross
of the i–th
40% was about tloice m la~e in the Philippines,
and wealth, which persists to this day, cannot
the Marms
deviations
Most strikingly,
20Y0, or even to the bottom
the share
below that of Korea. ] The Gini coefficient
1.8 standard
countries.
Korea and 53.4 for the Philippines income
(1995a)
Lorentz curve lay everywhere
percentage
Similar disparities
and Squire
on the “kleptocratic”
president
of
nature of
in 1965, and declared
a state
both countries are available. Moreover, these figures were
income),
making
them comparable.
The 1988 Philippine
but very close to the 1985 figure which still [lsed household
number
income.
‘The sourm for land Ginis is Taylor and Hudson (1972). The twelvepoint gap corrwponds to about one standard deviation. While it is common knowledge that Korea implemented a land reform following World War II, it may be worth mentioning that the Philippin- al= had ib own around the end of World War I, during the American holdings
occupation
were purch~ed
was not as egalitarian became
r~ncentrated
period.
Following
the di=tablishernent
and redistributed. as in Korea, during
where
The relatively individual
the following
of the Catholic
Church,
high land Gini for 1960 suggests
holdings
forty years
1
were limited
to 3 hectarm,
a large part that either or that
of its land land reform
land ownership
of emergency period
ordy in 1972. Preceding
were two presidents
background
whose
him in power at the end of the US–assisted main policies
notis as seeking to “expand
reform programs, Of couw
and develop
Table
Philippine
and diversify
1 does not mnstitute
cert airily not the sole reason–
are described
in the US State
ties to its Asian neighbors,
Department’s
implement
domestic
the economy”. proof
that greater
why the “miracle”
But the facts which it documents
ranstruction
omurred
initial
equality
in one country
was the reason – and not the other.
do suggest that the answer to the puzzle may lie outside
the
represent ative agent framework.
1.2
Empirical
%gularit
ies
South Korea and the other Eas-Asian pines, but to Latin American ventional
wisdom
a significant
to formal empirical
evidence,
(1994),
Table 2 summarizes
typically
(twenty
include:
by mlumn
raises the annual growth
gap of about do=
not mme
and Korea, non–linear
rate of GDP
awith
to eontribute
negatively
(e.g.
is detrimental
for the growth
However,
result
(1996),
observed
several models
of equilibria.
that when male and female stock who also rliseusses
2
possible
(b) the
points.
regressions, distribution,
The magnitude
decrease
On one hand it amounts
differential
These
growth,
deviation
which is far from inconsequential,
These
ratios; as shown
measures of income
to long–run
a one standard
leading to a multiplicity
Barro
(1) mrresponds
and significant;
dummies.
per capita by .5 to .8 percentage
or East Asia and Latin America.
of the puzz]ing
and by
and several controls.
of growth rates found in most samples.
close to accounting
the ~xcePtion
(c) regional
with many different
initial inequality
25% after 30 years,
effect, possibly
Column
is always negative
positive;3
across most studies:
deviation
regressions
or proxies in the form of initial school enrollment
large or small may be in the eye of the beholder. 45% of the standard
growth
the paper.
on initial inequality
whose coefficient
run over a variety of data sets and periods
of this effect is mnsistent
(1994) ancl Persson
where the average growth rate of per capita GDP over some
(8) this effect is systematically
message:
from anecdotal
recent studies of the links from in-
I shall refer to it throughout
(a) initial inmme,
deliver a mnsistent
in the later
testing of the main theories.
years or so) is regressed
initial stock of human capital,
has moved
in cross-country
the main results from twent y–three
to the basic, reduced form regression,
con-
and land in the first group played
by the findings of Alesina and Rodrik
(1996) methodical
economists’
the high levels of wealth concentration
effect of inequality
equality to growth or investment.
long period
of income
In recent years the literature
impulsed
not to the nearby Philip-
It has long been part of development
whereas
to growth.
(1994) of a negative
(1992),
are usually contrasted
that the very equal distribution
were a serious impediment
Perotti’s
countries.
role in their take-off,
and Tabellini
“dragons”
in inequality
Whether
to between
this is
30% and
It also implies an income On the other between
hand,
the Philippines
in the literature
predict
In that case small differences are distinguished,
explanations).
this
the latter
a in
aPPears
initial inequality regressions,
can have marginal
and significantly
The inclusion Sometimes capital
is included,
sub-Saharan presenm
insignificant,
are added.
of these controls, negative
link between
long–inn
(c) typically
as in Benhabib and Squire
with inequality,
income distribution
in Africa
need to be explained, distributions
1.3
when dummies
on inequality,
for Latin America
remains
si~ificant
in and stocks of secondary
and in some of the theories through
education
discussed
have a
below
the
investment.
comes from the high levels
These persistent
in the direction
and
even in the
human capital
variation in inequality
Latin America.
and in fact point
mentioned
(1995b)
and growth arises precisely
and, particularly,
the coefficient
and Spiegel (1993) when the stock of human
In any case, enrollments
Note finally that a large part of the cross-country observed
redums
In most other studies inequality
correlation
by linear
outcomes.
(b) and especially
or in Deininger
Africa
substantial
influen~
of mntrols
it becomes
effects well in excess of the average slope estimated
inter–regional
of the models
differences
with multiple
long–run
above,
Overview
This paper presents well the evidence the empirical mnomy
and extends
on their relevance.
literature,
capital
the security of property
arise solely through Rodrik
can affect output
to inequality,
I start in Section
(1994),
theories.
the balance
Persson
The first model
and Tabellini
integrates
the political
of income
system
(Bertola
(1993),
The idea is that by lowering
the income
investment.
here, which combines mechanism
features of Persson and Tabellini
but also formalizes
there is a role for productive
departures public
with the degree of pr~wealth wit h a general aveat, differentially,
Compared
on redistribution
impact
in left–wing
4 See Fields
~owth
curve
democracy,
(1994) hypothesis.
The
one vote, ideal.
I also provide
affects democracies
income
disparities
model
(1995a),
in this class of models
system.
and growth only in right–wing
or populist
and Jakuhson
the Kuzneb
investment,
bias in the political
to a perfect
(1994) and B6nabou
from the one person,
on the issue of whether inequality
impact
support
discourages
effects
Alesina
the pressure
in turn,
to the
(whether
and distributional
average, greater inequality
This,
and
(1955) hypothesis.4
median voter or pivotal middle class relative to the national for rdstribution.
of
rather than on reverse effects from the level
in the political
(1994)).
as
a survey
are converging
on how the distribution
where asset markets are complete
of power
models,
section asks whether countries
growth,
and growth,
one deals with social conflict
It thus barely touches on the Kuznets
2 with theories
distribution
two unifying
The second
The paper focuses primarily
or endogenous)
of development
markets
income
through
exercise.
rights. The empirical
same level of inequality.
linking
This is done
and an econometric
and imperfect
exogenous
the main theories
and
of the
increases developed
displays this
Whether
or not
always increases
new results,
together
and non–democracies
are shown to have a lesser
or wealth–biased
regimes,
and a greater
ones. for a recent
review
of this issue w well as some new evicleuce,
which do=
not
In S~tion nomic
3 I turn to another
implications
but simply
set of theories,
due to imperfections
“turn off”
With decreasing
increase total output of land reform, to depressed
or growth.
public
Indeed, the main idea in this literature
incentives
the decentralized
equilibrium
Ades
(1993),
for which uninsurable
and Bolton
with the politics
paper
(1996),
(1994),
nomic effects of distribution
Fernandez
(Galor
immunities
and Rogerson
common
principle
equilibrium Section of property and poor
(1994), Maskin and Kremer
predatory
discouraging
at the expense of Grossman
(1994),
of the prisoner’s
dilemma
which captures
the essence
sustainable
seeking abilities,
per se, but inequality
CIWS, mntrary
disparities
to common
degTee of democracy
conflict
in con-
equilibria,
to explain
the
a variety of general
reduces the security
when the gap between to engage
of property
(1995),
Tornell
and Rustichini I propose
them.
and Velaco
(1996).
gTowth version The economy’s
to interest
groups’
It may then be profitable
and political
is not part of the story
interpretation,
As in the case of taxation,
(1992),
Only the lmt two
land reform, eclucat ion subsidies,
of earning
or
rights is the main
of this class of models. related
rich
in rent-seeking
here a simple
or uncertainty
4
I
above,
In particular,
between
distribution
in this
as well.
the analysis reveals that what really matters is not income
in the relative
instability
(1996b),
of new propositions
arise, through
to be negatively
transfer wealth to the poor through
More generally,
out that actual
rate is shown
as well as to in~me
rich to collectively protection.
growth
(1996a),
I then use this example
and Kim (1996) and Benhabib and growth.
Durlauf
long–run
Acemoglu
link inequality
sorting of agents into
can lead to multiple
The security
explicitly
and
the macroeco-
The results obtained
have a greater temptation
(1991),
Verdier
of credit constraints
accumulation.
of the former.
B6nabou
how the mmbination
mentioned
however,
maximum
(1994)).
based on the idea that sociopolitical
the latter presumably
Tornell (1994), Grossman
(1996b),
steady–st at e distributions
in several of the models
rights, thereby
activities
system
(1993),
class of models
a number
rates and levels of inequality.
4 turns to theories
focus of the models
papers,
political
by which multiple
feedbacks,
and Verdier
growth requi i-es some (interior)
democratic
(1993),
market incompleteness
by, the enciogenous (1993),
markets case. I also demonstrate
in growth
widens
(B6nabou
is
Some papers focus on
and Newman
In a related
of this entire set, and include
and a less than perfectly both
(1995a)).
costs due
issue in the literature
while others combine Saint-Paul
can
these growth benefits
and their traditional
Banerjee
arise from, or are magnified
explain for instance why maximizing
differing
(1996)),
or other “clubs”
between
The other central
(Perott i (1993), B6nabou
(1996),
are representative
trast to the complete
transfers,
(1993),
which starts with
the efficient amount of
shocks are of the essence.
and Z.ira Piketty
of redistribution
Saint-Paul
homogeneous
or other progressive
as before
is higher, so that redistributions
the tradeoff
for savings or labor supply.
mobility,
Aghion
prevent the poor from undertaking
My model embodies
schooling
of wealth has macroeco-
I use the very same model
returns their marginal product
intergenerational
(1996a),
in asset markets,
the loan market.
Loury (1981) is that credit constraints investment.
where the distribution
power.
told by most
rent– for the
or trade inequality
I also point models
growth is reduced
in this through
a decline in the expected Finally,
return on investment,
the possibility
of multiple
due to a higher threat of expropriation.
long–run
earlier leads me to raise in Section 5 an empirical are count riea converging Equivalently,
distributions question
model
once
new data set put together convergence
some interesting
puzzles,
The general strategy are thus stripped
it is enriched
by Deininger
in distribution.
Although
including adopted
2
Economy
to be developed
omy and imperfect generations
capital
the main ideas.
-
shocks?
Using
a
pass at this issue of uncovers
in Ginis.5 and unification.
The models
to the richness and realism of the literature, These are briefly discussd
for further research.
with Complete
markets theories.
distributions
to resolve the issue this exercise
in the concluding
All proofs are gathered
in the appendix.
Markets
here and in the next section
families, indexed
I make a fist
differences?
by most versions
idiosyncratic
of mean–reversion
which contribute
before turning to directions
The model
and Squire (1995a)
some evidence
of many elements
Political
as would be predicted
with uninsurable
it is insufficient
permanent
of their income
in this paper is one of simplification
but are not essential to conveying section,
or are there
not only in the first moments
GDP per capita– but also in the second one (and higher), of the neoclassical
discussed
which has not been taken up previously:
to the same level of inequality,
is there convergence
and history~ependenm
will bring together
There is a mntinuurn
by z E [0, 1]. The utility of a member
the political
of non–altruistic i in generation
econ-
overlapping t is
(1)
where ~ and d; denote consumption with resources
W;, distributed
invest in human or physical
when young and old respectively.
independently
capital according
This person is born endowed
across agents with mean wt = E[wj].
Agents
can
to the technolo~ (2)
where r is mnstant income.
and O < ~ S 1; k; is the amount
Note that individuals
can interpret through
face decreasing
investment.
The higher the general level of basic
quality.
sometimes income
Another
work on the general
used as an imperfect
which
technolo~
pretax
is linear.
One
which can be supplemented
skills, the easier it is to develop within a given generation,
or in
can be omitted.
5The paucity and sometim= poor quality of international of data
capital,
I shall focus for a while on what happens
which case time subscripts
as in all empirical
and y: is second-period
returns but the aggregate
w: as agent i’s basic level of skill or human
acquire new knowledge.
here,
invested
isues;
problem proxy),
see Deininger
data on income
is the lack of almost even
though
in mmt
is determinant.
5
distribution
and Scluire (1995a) any data
theories
remain
and Perotti
on the distribution
it is this distribution
binding constraints
(1 W)
for discussions
of wealth rather
than
(land
is
that
of
2.1
Savings and Intragenerational
There is a frictionless other,
credit market where agents in each generation
at some endogenoua
there is a government
inter~t
rate ?. The amount borrowed
who redistributes
case of linear taxes solutions
Growth
and focus
smnd-period
instead
by i is denoted
income.
on a simple
log-linear
borrow from and lend to each
I depart scheme
bi ~ O. Finally,
here from the standard which
will yield
explicit
and missing capital markets. G Let post–t ax and transfer income be
with both perfect
(3)
~i = (Yi)l-T(i)T! where the break-ven
income level ~ is defined by the balanced
~l(yi)l-T(j)7dz
=
I Note that the scheme the maximum
with free disposal
imperfectly
---
Z. Given an expected
everyone
democratic
regimes,
= p“ = p~(l
has the same second-period
GThis scheme
– ~)(w/k)l-D(w’
PP(l – ~)w
from the standard 70ne
markets.
scheme
but none of the essential
(lW5a)
one implies that –
+ bi – k).
condition
J: bi dz = O yields:
(6)
of aggregate
resources.
Consequently
take place
in equilibrium.
markets is the tltmat of expropriation
in a context
and lumpsum
levels would results
is
= y = j, and F = r~(l
to actual transfers of wealth.7
some multiplicative
invatment
problem
(5)
The second
so no actual transfers
here to complete
taxes
it to 7 >
= 5(T)W
s = s(~)
under complete
in Bdnabou Applied
of linear
could easily introduce
lated with wI. Optimal positive,
–7)
income,
w opposed
was introduced
and insurance
0;
then bmmes
thus invests the same fraction
of the return to investment,
I restrict
– Fbi]},
k* = k, hence ii = yi = rkDwl-d
_ fba — – pr~(l
The critical feature of redistribution
reasons
– ~) (ji/ki).
up over agents and using the loan market-clearing
individual
when 7
0.
the tax rate is the one
eficiency
requires more pro-wealth
bias in, a more uneqtLal count~:
A/2. of these results may lead one to think
taxation
They do not.
models
relative
With complete
are homothetic (1993)
~set
wealth,
markets.
(Persson
(Bertola
(1994)).
model
system.
with an elasticity
of political
I show there A, the pivotal
11
elites, ageut
but also
(1993))
or to fi-
that
intuitions
has no efficiency To equalize
and Bdnabou
in particular
not only
(1994))
The underlying
markets wealth heterogeneity
as-
features of most
and Tabellini
an externality
and Rodrik
reflect special
This includes
it does not even affect aggregates.
for a (lynamic
of the political
to correct
(Alesina
that they
These are in fact robust
with mmplete
are required
public investment
are indeed straightforward.
rather
of initial
(15)
be called
where there is no useful role for the government
12See Verdier
that the distribution
power
w.
intertemporal
nanm some productive
absolute
normal.
democracies
= AA – A2/2
rather than general insights,
of alternative
or the bargaining
to the wealth level w“ defined by @((w*
values of J muld
is maximized
those where taxes or subsidies
quences;
either the ideology
the more bimed against the poor is the political system:
very stark nature
sumptions
for the
2
(1) Growth h higher,
W“
(1991)
A2) .
and simply speak of positive
12 Substituting
Proposition
of a standard
But since even Western
“popuhst”.
power,
is log–normal:
voter or pressure group then corresponds
@-1(p).
to a system
lobbying
or Conway
bias reflecting
assume moreover
let it be the
have lower participation
of the proletariat”)
lnwi N N(m,
The pivotal
unequal
individuals
(e. g., Edsall (1984)
(“dictatorship
To get simple expressions,
in emh generation
franchise,
and less educated
p < 1/2 corresponds regime
voting
countries
being decisive,
The case p > 1/2 mrresponds
due to a wealth–restricted
even in industrialized
of a non–democratic
some more general points.
at the pth percentile.
or simply the fact that poor
rates in elections,
I shall
from the “one person, one vote” ideal, in a manner which
but allows me to demonstrate
agent or pressure group located
are and how they can be tested.
(199,5a)
if political
conse-
the social
for a discusion influence
has rauk AA and log–wealth
reflects
n~ + AA*.
marginal
costs and benefits
adopting
the point
markets
also imply
of taxation
of view of the individual that aggregate
state, the same applies to output return on capital, to other souras or regulations
of income.
rather
impact
problems
the perspective
on redistribution (1994)
his valuation
rate; in steady– the rate of
with highest wealth relative
of any complementarities,
public
shied away from the most direct
and taken up by Perotti
discussed
“non+emocracies”.
While
in Proposition it does
mndtion
or model.
the cmss–patiid
and regime.
in this instance
second
second-order
effect:
regimes.
(1992),
inputs
the differential
ancl others,
be stronger
to some
testable
The claim, first made
(1996)
1 should
relate
is that the
in “democracies”
valid intuitions,
cannot
criticism
involves
there
are two
on the direction
to political
inequality
restrictive
and Rodrik
whether
on these remarks,
and inequality,
order rendition:
(1994)
through
union.
are decisive,
= AA — A2/2,
Its crossfunction,
14
distribution
across political
who pointed
out that
What
should
regimes A similar
even diet ators matter
is how
of their constituency.
A and A, on redistribution
ln(l — p(r))
assumptions.
but
by the first-rder
elite or powerful
riots or bribes.
I shall now formally
at the pti wealth permntile
to inmme
of that agent or gToup’s objective
of income
Such
can be signed unambiguously,
lanclowning
functional
impact
affects the desire for realist ribution
Elaborating
with respect
in which one moves away from pure democracy.
by Alesina pressure,
rate,
no accident.
the tax rate) is determined
the third derivative
be signed without
was formulated
are subject
of the growth
a dictator,
its sign, any differential
must surely depend
this is probably
Not only is it rare when cross–partials
the gTowth rate (or equivalently,
thus inevitably
(b) Whatever
Upon reflection,
derivative
of the decisive agent –whether
and therefore
located
13 Complete
with this ar~ment.
distribution
stitution
of the individual
and growth across political
of inequality
concern
partial
rises with the interest
instead on a more indirect,
(a) I know of no formal proof claims
endowment.
growth thus means maximizing
seems to have somewhat
(1), focusing
and Tabellini
than
Maximizing
This includes
literature
effects of inequality
negative
growth
which means
which might affect the private return.
namely
by Persson
growth.
only look at averages,
with the ‘Representative”
consumption
thereby adopting
The empirical implication,
one need therefore
examine
the cross~ffects
and growth.
the equilibrium
When
of political
in-
voters or constituents
tax rate is given by their first–
where p was defined in (14).
This implies
(16)
ls~his
~e=oni~g
the same result
irnpli~itly
BUmes
even with progr=ive
matters is the threat of taxation). 14sakt_paul and Verdier (1993) show how r=tricting neither
paper
growth
(&~/~~~A
the political
provides and
r=ults
linear
‘=ation
taxation
because
and perwn rights
I
in all the literature
in equilibrium
and Tabellini
of different
on how the political
~2g/8~~A in my model),
u
in the working
class=
affects
regime
alters
under
everyone
growth
paPer
the same version
(an effect similar
the impact
of inequality
which is the issue of contention
12
discussion.
invests
here.
My model amount
of their to ~g/~A
yields
(all that
1994 here).
on reclistrihution
article But and
henw ==(l;,$;))(-1+’A-’’A((’)2+5:;~;w(-))))))) The first term in the smnd
bracket
ln(l – p).15 This is a complicated
is negative,
expression,
it numerically.
Alternatively,
first. Together
with similar derivations
version of the “democracy
’17) but the semnd
one depends
on the curvature
of
even for A close to zero. One could of course evaluate
observe that when A is small, the second term is dominated
by the
for the growth rate, this allows me to prove an amended
and inequalit y“ conjecture:
Proposition 3 As long as inequality on growth am weaker, 82g/8A8A
is not too la~e,
the less favomble
its positive
effect
to the poor is the political
to democracies,
in left–wing,
populist
inequality
shotid
have mom impact
regimes and less impact ordy in right-wing
also from (16) that in this latter case taxes first decline, A ~ ~. Distributional skeness
the poor.
~nflict
The regressive
which ensured link between
initially
of the income
~2r/8A8A
to
positive.
effect
0
redistribution
As seen in (10), the missing credit market does not affect him at all because
poorer
the agent
as everyone
the very rich (differential
the social losses from liquidity
or lend.
costs through
Now there are two opposite
savings but did not materialize
liquidity
now maximizes
The first is the progmssivii?y of the tax scheme,
in “soaking”
The opposite
policy
is the third result.
ti > w gain at the expense
With
level if
to A 2. The second one has a familiar ring, although
proportional
make him “un–repre=ntative”.
same amount.
tax
level.
now has not only efficiency
always chose the efficient policy.
threat which discouraged
is above this eflcient
wealth is below the eficient
given that redistribution
EWi [Ui (~)]. Most striking
all those below
W would set a positive
+ p), and above in the reverse case.
it should be noted that the median voter’s
especially
favor
in A.
5(7) but also efficiency
to borrow
with own income
results.
tax rate of the agent with avemge
to internalize
ideal
+ (1 + PB(l – T))(ln wi – n~).
(3) The prejemd
result).
each agent’s
to the social one, let us next rewrite:
tax mte of t}te agent with median wdth
individual
who prefer this
with the average endowment
(2) The prefemd
welfare,
– 1 was
5
(1) A social planner
The fist
> 2A2),
112
(pP)–l V’(Z)
in w; and increases
Proposition
+ (1 + P8(1 – ~))(lnwi
over all ~ 2 Z, where z = – (1 + l/p~)
tax rate ri is then given by the first-order
progressivity.
i’s welfare is equal to:
of all agents,
a similar
is his failure to (1 – ~)A2.
he does not want as both
richer and
(1 – ~) A2 is large enough or simply when ~ < 1/2 this
bias and the average agent chooses an inefficiently
20
low tax rate,
even less efficient than that of the median voter. 24 Democracy
~sibly
To explore in more detail the effects of political the pivotal agent or group correspond
institutions
and inequality
once again to the pt~ percentile,
in Wi = m + AA into (11) and setting U’i(7)
has its virtues,
after all.
on redistribution,
let
with p = O(A). Substituting
= O leads to the following
results:
Proposition 6 (1) Growth and intertempod in the political system, (2,) If the pivotal incmes
a~ hill-shaped with respwt
A. They are rntimized
(3) If the pivotal to inequality.
at Ac > A w >0. with inequality;
such
agent is of higher mnk than the median (A > O), taxes are U-shaped
with
and reduce growth.
Tax cuts induced
by rising inquality
duce
growth if ~ is not too close to 1. Tax incwes
eflmts
up to some
beomes
negative.
point,
a potential
am always inefficient
explanation
to intertemporal
for Barre’s
at some interrnediate
monotonic
relationship
me~uring
IAI rather than A. Note that asset market incompleteness
for this non–linear
Alesina and Rodrik
level of democracy,
(1996)
maximized
between
the two.
system
not change the fact that rest rict ing the political plained following
Proposition
non–distortionary
private expenditures. fore with –A,
markets
and potentially
political
taxes and we~am
~se. 25 The difference
import ant explanation
(1995a)
that model
of any
are both
In models
such as
or spillovers
as ex-
(1993) model where
supplementing
both
does
wealth0.
5.
in (25), in which case cr~it
cases accord efiect
8r/8A
with the claim in B6nabou >0
can be reversed.
In
Direct
3.4
and Indirect
Effects of Inequality
We saw earlier that for a given policy Does
this remain
the envelope
theorem
by an efficient
to (23)
planner,
arises through U’i(~)
true when policy
= O of the pivotal
responds
shows
process.
redu~.s
growth
endogenously?
and intertemporal
A straightforward
that such is the case when
r = 7P(A)
i.e.
the political
r, inequality
-
kt
arg max{W}.
The
application
redistribution
of
is determined
more interesting
r“ (A, A) denote this solution
efficiency.
case is when it
to the first-order
rendition
group:
(p~)-lV’(r)
+ (1 –~)A2
– AA = O.
(26)
We ean write :=%+(~).=..x
(~)
and similarly for W. The first term, which holds policy fixed, is always negative, were analyzed
in Proposition
with inequality;
moreover,
6. In a pure democracy the equilibrium
with respect to inequality.
a detrimental
to which agents fexcept
decreasing
taxes w g does.
Whether
effect on savings incentives
growth rises or falls thus depends
returns and have unequal resources.
that since it puts positive In particular,
(~ S O) taxes increases
With positive wealth bias, on the other hand, taxes are
Lower taxes have a positive
one on credit constraints,
system
rate is always above 7P, hence in the range where W
and a fortiori g decline with redistribution. U–shaped
or a populist
while the l~t two
weight on first-period
Proposition
on the extent
The effects on W are similar,
mnsumption
6 established
but
it does not “overvalue”
that where rising inequality
letis
low
to tax
cuts, these are never efficient. Proposition
7
(1) If the pivotal interternpod
agent is at or below the median (~ s O), inequality
mduc~s both growth and
eficiency.
(2) If the pivotal agent is of higher rank than the median (A > O), inequality long os ~ is not too large, thti can ody
inequality
co.se, inequality
incmes
with the indirect
between these variables. less be possible, calls attention wealth
intertempord
eficiency,
but
reduces growth if A is not too small and ~ not to close to 1. In the growth through a decmse
While the direct effect of inequality its interaction
may increase
occur through an incm~~e in mdisttibution.
(3) For A >0 opposite
For B close to 1 inequality
still lowers W m
on growth
one (policy
(through
response)
wealth constraints)
can generate
Sorting out the two channels through
and even appears
desirable.
to the fact that if political
and if capital
in tax rates.
market
imperfections
Particularly
a structural
correlated
are not too severe,
22
a fairly mmplex
interesting
power is sufficiently
often dominates,
greater
relationship
model should nonethe-
is the third result, with financial inequality
which
or human
may actually
increase growth through
a decline in redistribution.
social conflict
in Section
growth)
dscussed
is not income
of political
power.
former translate actually
3.5
inequality
Where
institutions
or the political
stability
(which promotes
relative
to the inequality
of income
technology
are such that increases
(or here, taxation)
Income
2: even though
pletely exogenous,
Distributions
distribution
ket outmme
need not rise, and may
Ravikumar
(1993),
Saint–Paul
itself from one
of the wealth distribution
policies
and Rogerson (1993),
(Loury
(1994))
Saint–Paul
(1995a)).
Aghion
Making
affects persistence,
in this paper
that intergenerational
mechanism
capital
spillover
is at the
and Bolton (1981),
mar(1996),
Glomm
and
and those which encioge-
and Verdier
(1993),
these distributional
The intergenerational
Verdier
dynamics persistence
as the level of inequality
both in theory and in practice,
largely
is easily amended
to deal with these issues.
linkages operated or through
of the form:
operates
are fairly homogeneous
onl y at the aggregate
universal
w~+l = c~+l y~. ht
public
education,
me now recognize
is therefore
level.
lJntil now Whether
they resulted
in socioeconomic
status:
school
districts,
in a
that the transmission
within the family or at the level of small communities
first the pure human capital spillover interpretation,
simply replaced
neighborhoods,
which
social networks,
under which wj+l = c:+ 1 yt is
by: W;+l = E;+l y:.
Under perfect
the
with the absence
will reproduce
is at least as relevant a social concern
of how policy
an economy-wide
etc. Consider
B6nabou
(1994)
those which focus on the decentralized
redistributive
(Perotti
is mm-
after period,
and Rodrik
of factor prices combines
Galor and Zeira (1993),
Fernandez
(1994),
of endowments
and Alesina
to address the issue of social mobility.
developed
it was assumed
of human
distribution
interest.
The model
transmission
(1993),
outmme
and wealth disparities
per se. The question
(1993)
evolution
This includes
(1996a),
as a political
is also necessary
through
the endogenous
and Newman
B6nabou
nize redistribution
of mnsiderable
In Bertola
the constancy
models.
markets models discussed
(1994) as here it simply reflects, period
those which analyze potential
(1992),
Mobility
to ensure that any initial distribution
markets
(Banerjee
(1996)),
of income
abilities.
By contrast,
core of incomplete
and Ades
of agents’
uncertainty
to the next.
Social
it plays a critical role, the pretax
of wealth is indeterminate:
of idiosyncratic
and
a feature inherited from the complete
In Persson and Tabellini
the fixed distribution
in~me
in the
fall.
in Section
explicit
for sociopolitical
per se, but the inequality
Up to this point I have maintained
Piketty
4 : what matters
into increases in the latter, instability
Endogenous
period
A similar insight will arise from the model of
capital markets this changes nothing,
because y~ = r (5L)P wt for all agents.
remains i.i.d., with variance A2 = Var[ln c] = s2. With imperfect
other hand, equation
(8”)
capital
markets,
(20) now implies that family income follows the law of motion:
23
Family on the
where ~t is the tax rate chosen in period distributed,
and its intergenerational
follows the autoregressive
t and tit is given by (19).
persistence
Income
remains log-normally
is given by ~(1 – ~t). As a result,
inequality
prouss:
(28)
The alternative
mechanism
vately or locally
funded.
1 – ~ of smnd the constant
period
of intergenerational With
the preferences
rwources
~mplete
but these retain the property
human capital.
Together
distortion
long–run
s’(~t)
that redistribution
reduces
and borrowing
constraints
also in dl future perioh: growth is maximized
now pri-
w~+l = ~~+1~,
different
the persistence
for aggregate
where
d: = y: so (27)
individual
dynamics,
of w;. From here on I
gTowth whereas the second one
that
benefit of redistribution
as before,
Therefore
markets lead to somewhat
with (28) this makes clear that once the dynamics
an additional,
investment,
to one. In the absence of credit markets,
shall focus on the first case, since it has implications does not. R,ecall in particular
is educational
( 1‘ ), agents all want to invest the same fraction
in their child’s
has again been normalized
and (28) remain unchanged.
t ranstission
of income distribution
appears.
A marginal
rise in ~t causes the same
but now the growth 1OSS- due to the combination are reduced not only in the current period A~f~ is lower, for all k. This homogenizing
at a higher level of redistribution
a lower deg-ree of wealth bias A in the political
system,
are endogenized,
of wealth inequality
((1 – ~t)2A~/2
is lower), but
effect means that long-run
than previously. ‘G This, in turn, requires which must therefore
be closer to pure
democracy. This brings me to the question bution,
which I will discuss both theoretically
in each generation Together, Proposition
by the preferred
~t and At then determine
ZGAwn~J
policy
behavior
myopic
tax rate,
infinite lives both
feedback
~referenc=
just
(I)
of the pivotal
next period’s
As before,
group:
distribution,
and its income
let taxes be determined
~t = ~“ (At, A), defined
according
distri-
to (28).
in At, so that endogenous
by (26).
Recall now from redistribution
on inequality,
or (I’)
imply
w they do not internalize
are internalized
of the economy
and empirically.
7 that when A s O, ~“ (At, A) is increasing
acts as a stabilizing
current
of the long-run
to an extent
that
they C1Onot internalize
the mean
which reflects
24
this variance
effect of yl on WL+I. With the [Iiseount
factor;
effect when setting
forwar[l-looking
see B&nabou
(1996a),
altruism
the or
Proposition 8 If the median voter
or some
per-iod (A < O), the ewnomy ‘r
~——r*(A,
agent located at a lower pementile
wnve~es
to a unique steudy-state
2 — Am —s 2/( 1 – ~2(1 – ~a)2)
Am),
chooses
the tax rate in eue~
growth path (Tm, Am, gm) when
and tile gmwtl~ rate is
(29)
As one would expect, tnbution
a stronger
Multiple
With
any positive
bng
Run
influena
so on. The feet-bzk Proposition
of wealth on the political
states.
from policy
to intergenerational
steady-state
dynamics
hence also ~~+1, and
is now potentially
through
markets:
a negative
which multiple
wage,
A more unequal
depressing
Conversely,
social mobility generates
wealthy
very unequally
distributions
feedback
lower.
(1993)
entrepreneurs
dynast ies benefit high inequality.
form of “mobility
if the economy’s
in turn, aggravate poor agents’ borrowing low total wealth and high dispersion and wealth vary negatively
this feedback
of labor,
the bequest
through
which
the interest
and make mobility
persist into the next period.
enough
with positive
idiosyncratic
This
rate.
probability;
the wealth
pro-
however,
a poor families
is then ergo[lic
25
makes for low (1996),
When
borrowers. low.
Across stewly–states,
and the long-run
wealth is
This implies
more difficult;
escapes
Piketty
High interest rates,
with the interest rate. In the present model,
uncertainty,
the real
they leave to their
and Bolton
total wealth is relatively
constraints
through
Instead they bemme
lower wages.
Builcling on Aghion
trap”
operates
of
who do not have the mllateral
or employers
from paying
capital
via a worsening
thus both aggregate
as in Saint–Paul
“Multiple equilibria can also arise from a non-convexity at tile level of iu(lividual families, With
true
the common
with imperfect
on mobility,
there are few lenders and a lot of would-be
high interest rates, especially
This is a fortiori
more generally
arise in models
of inequality
the wage and therefore
and persistently a related
higher and Am
of wealth means more people
either self–employed
mllectively
is
it is based illustrate
and Newman
distribution
steady-
be true for the growth rate gm.
long–run
general equilibrium
constraints .27 In Banerjee
required to bwme
Tm
(27) on whi&
credit
Zeira (1993).
destabilizing,
W incre~~es m one moves from a less redistnbutive,
to one where
9 and equation
ptinciple
distributed
eficiency
welfare Ewi [U=], and may &o
Proposition
children.
again. By (28) this tends to increase At+l,
interternporal
more inegditarian
workers,
on the other hand, taxes initially
agent i.s of higher rank than the median (~ > O) there can be multiple
In that cue,
for m-ante
process,
9
If the pivotal
thrmhold
greater redis-
Distributions
decrease with At, before increasing
income
towards
bias shifts this steady-state
and less inequality,
3.6
(1996)
populist
as in Galor
the poverty distribution
and
(no inv~tment) uni~lue.
(1994) and Wnabou of education
(1995a),
the feedback
serve M imperfect
leads to lower redistribution by worsening positively
the credit
mrrelated
substitutes
constraints
form of mtitiplicity
omurs through
In Durlauf
for instance,
of poor
more heterogeneous richer, more educated
neighbors
taxes and transfers or public funding
agents,
When greater inequality
and makes disparities
This is why transfers
in spite of savings distortions,
sufficient
disparities
more persistent,
and growth
as stated above,
in models of endogenous
communities ~t
population,
mobility
segregation
families to form small homogeneous
policy;
for the missing credit market.
this again reduces
across steady–states
(1996b)
is through
in wealth and/or
can be
A related
community
imposition,
education
cause well-off
rather than share the fixed rests with a larger,
off from the tax base and positive
would have provided,
local
spillovers
poor families again experience
which
no or little
upward mobility, 2s Are countries
all headed towards
8, or are there permanently Proposition broader
different
trajectories
9? I shall argue in Section
interest
direction.
4
the same long-run
distribution
in terms of both
empirically,
and Property
votes are aunted,
are models generally
impeding
praumed
that this becomes
of related
(1995),
productive,
to the majority
of property
or -nomic
and Kim
and defensive classes.
strategies,
(1996)
activities
occurs through rule.
an orderly
At the other extreme else’s wealth.
rich and poor
rights discourages
or coalitions.
Using “dynamic
It is
widens,
investment,
extract
and Rustichini
rents from others;
level of development,
also B6nabou effects,
(1993),
as well as Maskin
when segregation
focus on the allocation
commons
of one–shot
problem”
wnflict
thereby
occurs
of human
(1996 b),
interactions
efficient.
multiplicity
long-lasting
(lW6a)
and Fern andez
(1994)
on segregation
But
if one combined
COUIC1occur
once again.
26
(1994), between
between
in-
and Rustichini interest groups
groups’
deal with inequality. constrains
as well as on a form of inequality
Durlauf
and Kremer
it is always capital,
of resources
the role and sources of interest
the y do not directly
(1991),
games with either Markovian
between
(1996) show how the extent to which social conflict
on the economy’s
Grossman
Tornell and Velasco (1992), Tornell (1994) and Benhabib
The first two papers emphasize
opportunistically
in the literature.
in the context
(1996) study how growth is affected by distributional
accumulation
expropriation
more likely when the gap between
ideas have been formalized
and Grossman
predatory
or trigger–type
2sSee
and growth.
Rights
submits
decline in the security
is of even
growth,
A number Acemoglu
then everyone
as in
then take the first few steps in that
where agents or interest groups can simply grab part c)f someone
and that the resulting
sorting
in distribution
But first I turn to the last of the three main theories linking distribution
In the median voter model (at least when taken literally),
dividuals
as in Proposition
level and inequality,
5 that this issue of convergence
and merits to be investigated
Social Conflict
process:
of income,
Benhabib
growth may depend
which relates to equilib-
ancl Rogerson
(1996),
by skills in tile labor market. this ]]roduction
ability to
model
(1994)
for related
In tile latter
case,
with credit-constrained
rium selection
rather than initial conditions:
of their model,
of a very simple model,
ingredients
Dilemma
an mnomy
to the economic
constituted
with Capital
of the problem
(“deviate”).
The consumption
and Rustichini
In period
groups of agents, labelled
or try and extract
a disproportionate
of groups
61(1 –5), al(l
–5)
When both sides moperate,
its claims
at the expense
capital stock at the start of the
1 and 2 are given by the following
shares of kt :
D az(l
az(l
+p15,
1 and 2; allowing
amount,
c
D
(1996) but
t each can choose to either moderate
Formally, let kt be the economy’s
l\2
c
to a familiar object:
Accumulation
by two homogeneous
pie (“cooperate”)
of the other group
structure
from several of the others.
for n groups would be straightforward.
period.
equilibria
and make some new points by means
The model has close links to that of Benhabib
A Prisoner’s
Consider
the essence of this literature
which relates the underlying
dilemma.
also inmrporat=
4.1
of subgam~perfect
the one with fastest growth involves equal utility for all players.
I shall seek here to convey
the prisoner’s
among the continuum
–5)
–5)
al(l
–5)
– (~z ‘62)5’,
– (D1 – 61)5
~z(l
–5)
+~z5
717 ?2
a total of ct = (1 –~ ) k~ is consumed,
while
5k~
is reinvested
according
to the linear technolog: kt+~ = ‘r (k~ – Ct) . The parameters
CYl < CY2=
outcome,
can be thought
which
competitive
shares of capital
a Nash equilibrium: a fraction
1 – al
this opportunistic
s
gain, a fr=tion
opposite 29 or
side.
That
alternatively,
(1992).
~einvat~
fli5k~ could
and the reduction
in investment,
resource
allocation
and Gr=man technolo~
than
be strictly 6i~kl.
are of course
problems
and Kim (1996). and the 6i’s (more
between
pro(luctive,
[Jsing th= generally,
pal]ers’
from
and rlefeusive
terminology,
(net)
.X
Tornell
in expropriation,
by making spent
loss, say piskt, morlelling
as in Groxman
the private
or aggregate
and ‘ela=o
~i = pi + 6i. All costs
them WOUIC1re[luire
activities,
does not
gain (net of resources consumption
thus imposing
the ~i’s reflect
the bi + pi – pi ‘s) its social
27
side’s
can only
be seized back by the
abroad;
i’s private
Of
stock and ~i – 6i
than investment
costs of engaging
to endogenize
reinvested,
the rest of society
such as del]osits
the sum of the other
offensive
the
most of the literature,
they could
6i5kt. Thus deviator
exogenous]y;
reflect
(C, C) is never
have been
to expropriation
I shall not do m for simplicity, specified
however,
Following
extracted
with a lower return,
less than
otherwise
is that otherwise
(1 — s)k~ +
for instance
from raiding the capital
into the model direct or opportunity
on expropriation)
game
-t
could game,
consumption.
is less vulnerable
in ~me
payoffs sum to 1-
payoffs in the stage
1 comes
motivation
consumption
One could also introduce
non-cooperative
O < 6; s
in the non-conflictual
an extra share of the pie, equal to
s kt which would
or rents unilaterally
The underlying
They
solution.
i can appropriate
of the other group’s
that transfers
be mnsumed.29
group
1 of the resources
of income
In the singl~period
and labor.
~ O is at the direct expense it is resumed
the distribution
as the market
by deviating,
O < pi
capture
(30)
efficacy
“destructiven~”.
(1991),
and
agents’ (1994)
of the predation
mean that it is entirely
The possibility
safe, however.
of indirect
levies or outright
theft implies that not all of the deviator’s
is often assumed
in the literature;
have different
import
gain need come from investment,
pi and 6i are conceptually
distinct,
as
and will be seen below to
implications.
On net, a unilateral In a situation
deviation
of open mnflict
by some group z reduces the capital
where both
they will be lead to overconsume is consumed, represent
taxes and subsidies,
groups
try to appropriate
even more: for simplicity
~1 + 72 = 1, but this is not essential
each side’s strength
in the political
stock by a factor resources
from each other
I assume that under (D, D) all capital
for the nature of the results.
struggle,
1 – 8;.
that is, its power.
The shares 7;
They could be related
to the ~i’s and pi ‘s, as dscussed
later on, but in general need not be.30 Let me assume from here
on that ~i(l
< ~i for all i, which implies that the unique equilibrium
one-shot
– S) — (@_i — 6_i)5
game is (D, D), Consider,
however,
agents who are infini tely–lived,
of the
with preferences
m g=() where I/a
> 1 is the intertemporal
play the game described
of substitution
by the above payoffs.
which makes this a dynamic the “first–best”
elasticity
The capital
game, rather than simply
case where there is only one group,
carry out lump–sum
transfers.
intertemporal
U = u(5) kl–a/(l
utility
Starting
and p rl ‘a < 1. In each period stock kt constitutes
a repeated
or equivalently
with k. = k, playing
one.
a state variable
I begin
a central
they
by focusing
planner
(C~,(~) in every period
on
who can yields the
– a), where (1-~)1-a U(5) =
The growth mte is constant entire capital
and equal to r5.
stock is consumed
immediately,
case r = 3, p = .4, a = .25, the function
(31)
1 – p(7”5)l-a” This strategy
is preferable
to (D, D),
if u(s) > u(O) = 1. As depicted
u(.) is strictly concave
and maximized
where the
on Figure 1 for the at s’ ~ pll” rl/”-l.
This is the value which the planner would choose if he could select the savings rate continuously, One could let s = 5*, but more generally
I will simply require that s belong
to the interval (Z, 1)
over which u(s) > 1, so that T5 is indeed the best of the two growth rates achievable mnomy.
The main question
is whether
it can be sustained
when groups have the ability to behave opportunistically share from the mmmon
30For instanm, 31were agenb
as a subgame
and extract
in our simple
perfect
equilibrium
more than their cooperative
pool of resources. 31
Benhabib and Rustichini (1996) impose a symmetry restriction similar to 7, = 1/2. alm allow~ to ~on~ulne l=S t]]au ~i (1 – 5), SUCII deviations could be shown to be unprofitable
long ass
is not too far below the optimal
s(s/s*)”)
is a decreasing
function
savings
with j(s”)
rate 5* (specifically,
= 1).
28
aij(s)
~ 1 where
f(5)
-(1
–s)(5/5’)”
~
/ (1 –
1.6:
as~iiai
Vi(s),
oryi/ai
increases
1.4;
1.2: 1.
I
I
I I I I
0.8:, 0.6:
0.4: 0.2 00
0.2
0.6
0.4
I I I
I
1
: I i
: I
/“
0.8 5’
Figurel:
Income
4.2 Playing
Inequality,
behavior.
These
This high-mnflict,
to which players
trigger strategies
> (~i(l
(Qi)l-aU(5)
on growth
and Growth
is always an equilibrium.
serve as the punishment
from cooperative
wnstraints
Power Inequality
(D, D) in every period
will naturally
incentive
1 3
revert whenever
equal to a fr~tion
in the following condition
period
- 5) + ~i5)1-a
has deviated
+ ~(T5 (1 - 6i)~i)’-a.
~i~ of the capital stock.
where both
someone
outcome
will sustain the first best if, for all i,
The first term on the right hand side is group i’s current consumption extra rmour-
low–growth
when it deviates
The second term measures
groups play D, liquidating
the capital
stock,
and grabs its payoff
Rewriting
this
leads to:
Proposition
10
If the incentiue
compatibility
U(5)
> (1 – 5 + (~i/0!i)5)1-a
b satisfied jor both groups, equilibrium.
If it is tiolated,
Several interesting
comtraint
continued
+ P(T5 (1 – 6;)(~i/CYi))l-a
cooperation
the only equilibrium
and growth at the mte rs is sustail~able 0s an ti the non-woperatiue
results can now be obtained
trated on Figure 1.
29
(32)
~ ~i(5)
by examining
one with no growth.
condition
(32),
which is illus-
(1)
Incentive
compatibility
typically
holds
when s is low enough,
enough values, This implies that there is a maitnu?n by confict
over the distribution
growth rate rs~= (2)
Growth
other through economic s~~
of income,
falls as maxi {~i/a;
becomes
stainable
as in Benhabib
} or maxa{~i/ai}
opportunistic
deviations
at high
growth rate which is constrained
and Rustichini
(1996).
This maximum
rise. 32
harder to sustain when some (or both)
unilateral,
and is violated
group’s
is dispmpotiio?~ate
pie: for any given 5, (32) ceases to hold as ~i/ai
ability to expropriate
to its “normal”
becomes
large enough.
the
share of the Gnsequently,
declines. (3)
Inwrne
inquality
and expropriating mnstraint
ability.
is binding
the poor,
also limits the emnomy’s For instanm
transfer
al.
group
This scenario,
here from the desirability (4)
However,
the incentive
compatibility
2 is better
formalized
be the poor
~1 /al
< T2/a2,
for instance,
model
thus makes clear that what
land reform,
education
subsidies,
a minimum
off if its own share a2 can be permanently
by Grossman
(1994),
who are relatively
reduced
(1995) in a static context,
earning and political power:
more vulnerable
versus large. landowners.
it is the rich’s threat to deviate
of income per se, as often claimed sin~
is 1. 33 Note
payoffs is a potential
come from
arises
of growth.
it could
growth,
1/2,
of power
they are. In this case it may be in the interest of the rich
one can think of small peasants
exploitation;
maximizes
if @l = ~z and ~ =
wealth to the poor through
wage, or trade protection: to increase
rate, for a given allocation
for the group with the smaller income share, al < a2. Deviations
and are more likely the poorer
to voluntarily
growth
problem
matters
If PI/al
which reduces
is not inequality
in the literature,
to expropriation = ~2/a2
but inequality
in the relative
that the important
for the empirical
role played by off-the
implementation
The
distribution distribution
al + a2 = ~1 + 72 = 1, the minimal value of maxi {~i/ai}, also
but
accumulation,
in the (equilibrium)
or
of
which
equilibrium–path
of such gametheoretic
models.
I
shall return to it below. (5) The more of the expropriation
comes from capital,
tion, the easier it is to sustain cooperation:
32For instance 6i)(~i/~) described
=
– 6~)(~i/~~))’
1 < ~i(i)
here there are only two sustainable growth rates, rs and zero.
continuum
of such games,
indexed
by s, which
can be played;
consump-
in d;.
1 + p(rs(l
and u@)
to other people’s
all other
But
‘a
whenever suppose
parameters
x
vi(s)
fli/~i
>
now that
are invariant.
as long as (1 – 1. In the game there
is in fact
Among
a
values
which cooperation can be sustained, both groups will prefer the one with the constrained-optimal growth rate, 5“’ S min{s” ,5ma. }. They will also be unanimous in preferring the s’” game to any other with a single,
of 6 for
uncooperative both
the s’*
33 Deviations with stochastic
equilibrium
if u(s’”)
> m~i{(~;/~i)
and the zero-growth
by the rich also occur endowments
enter
and fight over the appropriation
1‘“}.
If this condition
is not satisfied
the Pareto
frontier
includes
outcomes. in Barbosa,
into an ex–ante
of total
r-urces.
Jovanovic
and Spiegel (1996),
efficient Their
income
decisions
30
sharing
but the context
agreement.
are based
on current
Ex–post,
is different.
Agents
they may renege
payoffs only.
Empirical
4.3
The findings
Evidence
from most studies which have examined
property
rights,
m-ures
of “instability”
sociopolitiml turnover,
political
constructed
violence,
and Poole
(1996)).
are summarized
or “insecurity”
instability
Londregan Perotti
and growth
Because
Alesina
fires
investors,
risk, rule of law, enforceability Keefer and Knack of security
deteriorate
that both
greater
and growth, property
(1995),
of contracts,
turmoil
(1993) or Barro
to be clearly
identified,
hand, greater political
equation,
I also wrote support the latter and intuitive
through
formalized
thus appears
regressions,
(1995)
instability
report
that
becomes
or empirical
notions
In turn,
rights decrease
“friction” model
or resource
with symmetric
sassinations,
etc.
information
about
effect).
information,
the extent of agents’
GDP
in the
is still somewhat
of the story.
The leads,
spread in the return to investment and thus leads, ve~
equilibrium
muclt
or the threat thereof. is achieved
without
in Alesina
struggles along the equilibrium of income
any
in a bargaining
for strikes, riots, coups,
as for instance
over what this distribution
ten-
First, note that contrary
is not one where greater inequality
It has no up-side
information,
31
on the other
because the link between
above.
there is no role in these models
incomplete
remain
and property
for initial
instability
will reflect not so much the distribution
uncertainty
however,
instability
is really not pati
ancl payoffs,
(1991), will generate delays and resource-dissipating extent to which these occur
by the evidenm;
Just as there is no reason for delay to occllr
dissipation,
Incorporating
actions
investment
which in turn makes
well supported
m in the voting modeZs, to a lower mean return due increased expropriation, with perfect
(10) shows
rights, as does a more equal income
model presented
to a mean–preserving
(and even this is known to have an ambiguous
Second,
column
types
insignificant.
here and in most of the literature conflict,
and so on; see
(9) shows that both
when one controls
unurt aint y or unpredictability
more acute political
a
or nationalization
only the latter are significant;
of social or political
uous. This can be seen from the representative
mechanism
Column
for “the general idea” rather than “the models”
interpretation,
(1996),
corruption,
social conflict,
implies more secure property
and Knack political
and Roubini
risk” sold by specialized
(1993) finds that when both political
in investment
stability
Keefer
growth
of
is more directly linked to the
of property
exacerbates
of indices
(1986),
risk, expropriation
inequality.
broad
and Gupta
The specific channels through which this occurs,
Svensson
rights variables are included
to mmmon
income
Two
most studies estimate
of “country
(1996).
and lower protection
and reduces
plausible.
default
Olzer
quality of the bureaucracy,
The general idea that inequality
indeed it is eminently
investment
such as sovereign
as a result of increased
rights less secure
distribution.
by various indicators
2.
over
strikes, government
Venieris
problem
The other form of “instability”
Svensson
political
of protests,
Alesina,
of the potential] y serious simultaneity
rights. It is measured
to international
(1996),
mnflict
The first consists
and the like (e.g.,
and Perotti
inequality,
(9) and (10) of Table
have been considered.
revolutions,
joint model of growth and instability. issue of property
in columns
from various mmbinations
mups,
(1990),
the links between
as-
and Drazen path; but the
(and power)
as
is. Why the two should be related
is not obvious, politiml
Given imperfect
economy
models:
information,
resources
would be dissipated
the voter who turns out to be pivotal
have his way, but only after costly political
campaigns,
lobbying
as well in standard
(say, median)
and legislative
will eventually maneuvers.
will make the two types of model a little more similar yet, as would the incorporation dynamic
mncerns
strategies
5
into the voting model:
reputation
on the part of the private sector,
Convergence:
Are muntries
muntries
to the same level of inequality?
on-.
Similarly,
than European
and minimum
wages.
may seem unusual,
Yet it is important,
It is common
knowledge
ones, themselves
the specter of U, S.–style inequality
of redistribution
(and Higher)
This question
literature.
the facts is in itself of interest.
tend to be more unequal
trigger
and so on.
indeed it has not been taken up by the empirical First, ascertaining
of strategic,
of the pivotal class for fiscal restraint,
On to Second Moments
mnverging
This
and
on several counts,
that Latin American
less equal than East–Asian
is often used in Europe
to justify
But are all these gaps really permanent,
high levels
or inexorably
narrowing?34 %nd,
it can shed light on the relevance of models
dependence
of the inmme
distribution,
mechanisms
of credit market inmmpleteness
bility which underlie these theories, Third,
tion.
conve~ence
i~l distn”bution:
for equity
gree of inequality
at OECD
~oun~ri=
and chang=
had the highest
initial
incre=
low inequality
from inequality
international
steady=tate.
during
by Proposition
should
persistent
rationing wealth
in Piketty (agents
distribution
exogenously, Departing probably
can
(1996) borrow
and Smeeding
the 80’s.
If anything,
9. The largest
(1995)
occurred
may be converging and Denmark;
this exercise
provides
valuable
find no relation
there is some evidence
incre~s
Portugal
to credit
less than constraints
or as the economy from homothetic
grows
preferences
lead to history dependence,
of this ergodicity
what
they
discussed it becomes
muld
in the dereturn or
in Section
with
Some
could
it suffers
too weak to generate
by introducing
a minimum
but the stochastic
version
of these
absent
multiplicity
consumption of that
one)
model
but
because
being
without
still remains
credit
from
the
are specified
and Bolton
level w in Cbatterjee
on a
credit–
even include
the latter
to a
based
the feedback
(Aghion
at
steadyatate.
be converging
from
models
initial
and the [JS, which
to a high inequality
probability
3.6 is either
between
of polarization
in the UI{
thus Denmark
information,
property.
repay
32
differences
of some form of increasing
France,
for examples
tend towards
differences
small number of eountri~, several of which are observed over a very short interval of time. qssee for instanm Banerjee and Newman (1991), Ayagari (1994), Bertola (1995) or the c= constraints
growth model
structure.
that these countri=
in Italy,
While
the presence
l~gO’S, Gottscllalk
levels, su~-ting occurred
of innate abilities,
in per
income distribu-
income. 35 Barring unexplained
or politioeconomic
in the
to social mo-
convergence
with the same fundamentals
indicate
in inequality
of the type suggested
The small~t
test of the joint
shocks, most versions of the neoclassical
countries
mnversely
and history–
3.6.
examining
of wealth and pretax
in the mnornic
levels of inequality the extremm,
an indirect
feedback
in Section
literature
or in the distribution
would
mmplementarity 34~k~g
and negative
as explained
with idiosyncratic
the same invariant distribution in “tmtes”
it provicles
steady–states
This variable is after all only the first mo?nent of each collntry’s
Once augmented
imply
In particular,
there is now a vast empirical
capita incomes.
with multiple
(19M)).
(1994)
to be studied.
could
Having raised the question
of mnvergence
in higher moments,
I shall make here a first pass
at trying to answer it. Ideally, one would apply to an international the same tests which are now standard regressing
rates of change
over time (Barro
(1991),
on initial Barro
by Deininger
(1995).
period.
missing wlls.
I start with a bird’s for which Deininger
Middle
High Income,
and Eastern Europe.
these numbers
I computed
each inequality
measure,
summary
East and North Africa,
by averaging
I
countries
eye view of the issue, by comparing
Africa,
I Gini (%)
the coverage
but considerable
and Squire (1995a) provide
measures,
tests of Quah
is data:
no such panel
is a new data base recently
expands
of previous
falls
put together
data sets, both
from the Luxembourg
effort has been devoted
Income
to making
across countries.
Sub-Saharan
inequality
dispersion
to the extent that one can think of it as a panel with (still many)
far fewer observations
comparable
cross–sectional
however,
measures
of first moments:
or the more complex
constraint,
I shall also use data for a sample of OECD
Study (LIS). It mntains the numbers
(1992)),
The binding
It significantly
on the mnvergence whether
The closest substitute
and Squire (1995a).
over time and over countries,
values, examining
and Sala–i-Martin
(1993) and Durlauf and Johnson exists over a long enough
in the literature
panel of inequality
statistics
on inequality:
East Asia/Pacific,
For each region they constructed all available observations
both
the seven main regions of the world Latin America,
South Asia, OECD
and
decaclal averages of various
from the 1960’s to the 1990’s.
the world average and the cross-regional
standard
From
deviation
of
These are report ed in Table 3.
Aclj. Gini (%) I Bottom 20% I 3d+4’h Quiutilfi
Top 20%
Top 20 / Bottom20
Mean 1960!s
39.99
41.52
6.18
35.65
47.84
9.03
1970’s
39.34
40.92
6.03
36.49
47.24
8.89
1980’s
37.47
39.92
6.52
37.56
44.73
8.31
1990’s
38.28
41.63
6.43
37.24
45.47
8.46
Standard
Deviation
1960’s
10.03
10.77
2.17
4.29
8.91
4.52
1970’s
10.54
11.23
2.77
3.74
8.76
4.29
1980’s
8.33
9.43
1.50
2.97
5.97
3.73
1990’s
7.79
9.46
1.58
3,52
6.66
3.82
Table 3: inequality Regions: South
Latin
Asia,
America,
OECD
and
Sub-Saharan
computed
While
these are somewhat
36 One would want column
from Deininger
level of inequality
and Squire
in the world
to weigh regions
than income
seven main regions East
surveys.
(1995a),
crude statistics,
and North
Tables
Africa
, East
Asia / Pacific,
typically
to some interesting
is one of relative
or income,
made by Deiuinger
The former
5 to 7.
they point
the picture
by population
of Table 3 refers to an adjustment
ture rather
acrm
Middle
High Income, Eastern Europe.
Source:
mean
Africa,
for instance.
and Squire (1995a)
bias the Gini coefficient
33
The
stability; “adjusted
to observations downward.
facts.36 some Gini” bawd
For the
small and in the second on expendi-
The adjustment
consists
gradual
improvement
Cross-regional of variation
is visible until the 1980’s,
variations
in inequality,
after which the situation
starts to deteriorate.
on the other hand, show a clear decline.
falls by 1970 for the Gini mefficients,
The coefficient
30% for the shares of the bottom
quintile,
for the shares of the third and fourth quintiles and 37% for the share the top quintile. that almost all of this “convergent”
occurred
between
shar=
to the extent permitted
or share ratios because
directly ewh
countries,
to the transition
muntry
I then ran a regression
I Sample
1 (N=69)
Gini
of this variable on a constant
The first sample
II Sample
for which
muntries;
(4.49)
(1.48)
-.017
-.036
-.060
(-3.04)
(-2.54)
(-4.75)
(-1.34)
.061
.262
.480
,050
tesh
using using Ginis’ average
To redua of adding
and Squire data set (1995a);
surveys;
Sample
3: subset
of the 69 countries
one is perhaps
Sample
of OECD
2: subset
countrim;
the negative
having
it restricts
been
obtained
and generally
significant
in Table
observations
in eac]l c=,
3 are, unsurprisingly,
set.
and Squire
tbe differences
not statistically
the rate of
to the 25 countries
from similar countries
types
of surveys
in the full sample.
the same regression
The results are consistent
signs indicate
size of the bias, which Deininger
are onlY ~ven
documented
using gro~
4: LIS data
attention
of the 19 OECD
that these results were due to measurement
to the Gini the average there
of countri-
Saml)le
for which I was able to compute
the most reliable:
The third one consists
rate of change
in parenthesis)
by the LIS data set, where I was able to perform
the chanm
data set. 3TGiven that of variation
t=tatistics
the typical time interval is much shorter, however.
four columns:
4 (N=16)
(2.65)
income),
It is complemented
Sample
3 (N=19)
-.015
the data are most comparable,
(gross household
I Sample
2 (N=25)
(3.07)
consists
change, 39 The second
and the initial value
2.56
1: Deininger income
the first and last
1.22
4: convergence
for
in Table 4, for four samples.
(heteroskedasticity-consistent
household
between
.755
R2
Table
it relates more
.641
c
Sample
and because
model, 3s In the first test I computed
(28) in the formal
of the Gini. The results are reported
across
by the data. I chose the Gini rather than particular
the average yearly rate of change in the Gini coefficient
available observations.
initial
tests for the Gini coefficients
it was available for more observations,
equation
Note also
the 1970’s and 1980’s.37
I next went to the full sample and ran simple convergence individual
22%
mean–reversion
on 16
across the
in inequality.
error on the first observation compute in standard
significant.
to be about deviations
6.6 Yo in their ‘r coefficients
In my view this does not make
them any less intriguing, especially in the light of tbe other eviclence discussed below. 38For a log_normal, the Gini is a monotonic function of the variance of the 10gs A?. I also ran manY of ‘lle regressions with the logarithms of the Ginis instead of the levels. This led to similar results. 39~tncting the sample to thcountri~ where at le~t ten years sel]arate the first and last oblsemations to throwing
out nine data
points,
with essentially
unchanged
34
rcsu]ts.
leads
(which bias-
the wefficient
right–hand–side the s~ond virtually
towards
negative
variables but replacing
the left–hand
and last available observations. unchanged
positive
and insignificant;
points
(with the trend often mmputed
betw~n
18 and 62 for the Deininger
standard
cross–sectional
panel it is nassary observations
to interpolate
mnvergence
the basic regression is negative
in the regression
data and net inmme
smnd
mlumn
(“dginires”,
muntries
(Table 5c, 25 countries).
gross inmme,
for observations
income
for both dates.
Some numbers income,
come others
from expenditure
First, I include data,
personal
the results are reported
then run the regression
in the
on the residuals
on the subset of countries
These results, given in the “Sample ones.
Once again there. is significant
3“ column evidence
of
from 1970 to 1980, much less so from 1980 to 1990 and none at all for the
with and without
or by interpolation).
dummies,
The results are presented
are all negative,
to the semnd,
and Squire
(1995a)
in both d~u=
when moving these problems,
between
the first two results and
subsets of countries,
on yet another
subsample
consisting
“Sample
but not in magnitude to the full twenty–years
2“.
of the 24
Once again the
from the first decade period.
of wllicll tile first one is tile most serious.
35
I therefore
were available (either from the original data
under the heading
with a drop in significant
and a drop
discrepancy
are for different
for which 1970, 1980 and 1990 observations
40Deininger
of concern
but some are for net (after tax) figures.40
and netdum);
years),
which might explain the puzzling
the regressions,
estimates
comparable.
Finally, I run the regression
the last one is that the data for ewh subperiod
muntries
One sourw
only at
1970 to 1990.
One problem
reran
The coefficient
I first regTess both the initial and final Gini coefficients
at the relevant
of each table).
198&
The first mlurnn presents
for 1970–1 990.
generated
perdum
are very similar to the previous
mnvergence
a usable
5a, 33 countries),
ways, all of which lead to similar results.
data (expdum,
(evaluated
which used gross family
whole period
1970–1 980 (Table
may not be directly
of Table 5. Alternatively,
third column
of each table,
To generate
for 1970, 1980 and 1990. I was then able to
and small and insignificant
variables
inmme
on these dummies
over different
This is what I did, using neighboring
data,
for the periods
finally, most concern
dummy
to
of the data set does not permit one to run the
some missing
I deal with this issue in three different
inequality
and i ts rate of change are computed
surveys, others from income surveys; some pert ain to household
income;
12 data
at the 5% level for 197&l 980, similar in size but significant
is that the data from different
to personal
only
by just a few years) as opposed
over a fixed, ten or twenty year period.
and 1970-1990
the 107o level for 1980-1990,
from expenditure
contained
of the change in the Gini on the initial value and a constant.
and significant
remained
and Squire data set.
regressions
1990 (Table 5b, 38 muntries)
that regression
over dates separated
when required to obtain Gini coefficients
run inequality
the inefficient
(t–stat ist ics ranged from 1,9 to 3, O). For the LIS
The incompleteness
regression
with the exact same
side by the average rate of change between
however,
with this test is that inequality
periods for different countries.
I ran the regression
For the first three subsamples
and st artistically significant
data set, it bmme
One problem
values),
A possible
explanation
would for some countries
to go from above to below the mean (say) during the first
decade,
then move back above during the second one, In the fourth,
OECD
countries,
magnitude
and stability
insistently periods
by wntrast, across
periods.
of mean-reversion
across major
regions
koking
between
1970 and 1980, but the picture
no virtually
in Tables
countries,
most of its decline
however,
deviation
(sdev),
between
of inequality
the 1970’s and the
one sees no evidenw
reported
5a to 5C
for the other
to recall that in Table 3, the dispersion
of the world experienced
the range of Ginis. The standard
reported
of
in terms of both
the regressions
It is interesting
across individual
seems stronger
Overall,
point to some mean–reversion
is much less clear.
1980’s.
the evidence
much smaller subsample
of any narrowing
in
in the last two lines of Table 5, shows
no change during any of the two decades under considerate ion.
The general picture general stability experiencing
which emerges
at the end of this empirical
in the world distribution
some non–negligible
actual convergence awaits a definite
answer.
of Ginis, within which countries
amount
or whether countries
of relative
oscillate
This is perhaps
for empirical
research.
data (looking
is thus a mixed one:
and regions seem to be
The question
around dist inet long–run
of whether
given the rather preliminary
as an important
across states or regions as in Barro and Sala-i-Martin
nature of
however,
and essentially
future studies with more sophisticated
there is
levels of inequalit y still
of the data.41 My main purpose,
in distribution
Hopefully,
mobility.
not surprising
the analysis and, above all, the limitations forward the issue of convergence
exercise
was to put
unexplored
econometrics
(1992))
topic
and better
will help resolve the
issue. Two related ticular
empirical
by Proposition
countries,
questions
9, is whether
or even industrialized
or are there permanent by Lindert
(1996)
differences
suggests
broad
xtually
increase
mobility,
a
capital
markets.
Recent
of education answer.
41In particular,
‘*See
Piketty
Are
Evidence
formal
on OECD
countries
tests on larger samples
(27) and suggested (1996)
studying
provided
are needed.42 redistribution
by most
models
with
the effect of state financing
of families in poor communities
suggests
that borrowing
a positive constraints
of inequality.
hypothesis)
of mean income (1995)
in par-
in redistribution”.
fiscal, or other forms of progressive
work by Cooper mobility
One, motivated
to the same relative size of the welfare state,
in equation
I have only looked at unconditional
the Kuznets
the dynamics
formalized
“convergence
(1995), on the other hand, finds only mixed evidence
affect the transmission
ity (e.g.,
but more
educational,
on the intergenerational
Mulligan
converging
in the social contr~t?
the latter,
issue is whether
serious investigation.
there is so to speak
countries,
Another
imperfect
also deserve
and the reverse
and income
and B6nabou
inequality
(l995A)
convergence.
The l]otential
links from inequality should
for models
be examined
studied
on inequal-
in this pal)er sugg=t
jointly.
which seek to explain
36
effects of develo]]ment
to growth
such persistent
differences.
that
6
Conclusion
Non representativ~agent
growth
is necessarily
First, I focused
growth,
incomplete.
theory
rather than the reverse.
response
to credit mnstraints
technology
and policy
abstracted
from the Kuznets
This choim
reflects both
model
The first model’s
and policy
precluded
(1955)
sp=
income
outmmes,
hypothesis,
Another
except
when discussing
whether
formalized domestic
by Murphy, market
growth regressions
distribution
in their data.
Keefer
and interacted
An empirically
distribution,
of the middle
fertility,
profitable,
mnsidered
these mrrelations,
out theory
Perotti
on fertility,
Among
the models
3 often play a critical
a number
of simplifications.
altruism, occupational as from spillovers the literature, Where
do we go from here?
43 See for instance and B6nabou
impede
consumption=rno
rather
(1995b)
Quadrini
to trade,
for the theory
the links between share
and because
to formalize
investment
some of these
which were extensively
role (e.g., Galor and Zang (1993)). in some depth, the analysis necessarily
I abstracted
from i nheritecl ability, dynastic as well
have been incorporated
in
at various points in the paper.
The political
economy
and imperfect stage. 43
the quantitative
and Rfos–Rfill
in the WOnd.
in
and grcjwth which can robustly
ties of any kind. All these elements
and have reached
Krusell,
(1994)
by including
and this in turn has a significant
which have begin
For ancision
and their role mentioned
of
even with
choice and effort decisions of workers, ent reprenellrs or rent–seekers,
and non~onvexi
now well understood
concerns
large
is the absolute
finds that a greater income
fertility
Even for the three main theories which were examined involved
however,
due to space constraints
on educational
theory
concentration
total GDP, and openness
(1996)
rent-
and Rustichini
mechanism
of research
of distribution,
the kind of wealth constraints
in Section
direction
effect
in the
They find no support
I did not discuss this mechanism
there is yet no well worked
links, moreover,
negative
properties
like Inclia is considerable
with income distribution.
and development.
arising
and extensive
excessive
test the “big push”
(1995)
curve,
requires a sufficiently
The relevant constraint,
more successful
CISSShas a strong
impact on growth.
generate
and Knack
conflict
If industrialization
technologies
to growth.
the role of fixed rests.
arises in the ‘(big push”
measures of market size such as population,
both by themselves
positive
an obstacle
returns
I thus entirely
or rich ones (see Benhabib
weight of the middle class, which in a large country
very high inequality.
income
Shleifer and Vishny (1989).
to make increasing
wealth may represent mnomic
effect of income
in
of preferences,
in the Kuznets
sociopolitical
on
endogenously
Similar homogeneity
in most remnt studies,
level-dependent
evolved
level of development.
and a waning interest
me from examining
or wealth inequality
but due to the homotheticity
seeking are diseases more likely to afflict poor countries (1996)).
distribution
by the economy’s
constraints
support
so rapiclly in recent years that this tour
only on the effects of income
it WM not affected
from its lack of empirical s-rid
has developed
There
othing but not invmtment
(l W), also
(1995)
many
(e. g., Ayagari
different.
37
capital While
in the first
quantitative (1994),
Bertola
further
case,
models
market models
(1995)),
developments
Fern andez
where
are
ancl Rogerson
liquidity
constraints
hut their ]]roperties
are
(perhaps swpe
in relation
to fertility
of the theoretical
work a high priority: put it mildly),
or technolog)
literature
will surely prove valuable,
and the scarcity
of direct
evidence
signs of the adverse effects of redistribution
and inferences
about
the gap between
makes further
on growth
the role of credit constraints
empirical
remain elusive
far too indirect.
With
(to
respect
to sociopolitiml
conflict,
there remains
room for theoretical
to one another.
Existing
models provide
many valuable insights but few robust results about the
effects of income inequality path),
fully predictable
on mfiict,
whether mnflict
or uncertain.
rights variables should be included
between
mnomic
open ~tict
in the regressions,
considered
and the Philippines?
pure political
economy
through
average
it
mnsumed
and
broadly deed
1.97%
more: with
in secondary
education
Philippines.
Similarly,
Philippines’
between factors ity
and
207.
by
very
government
the ~nmdotal
elit=.
GDP,
from
42%
versus
(1989).
1992
roles
puzzle
in 1960
were
is the the
extent
with
now
to theories
of social
on the phi]i~Pinm
ducation
and
changes.
(1W5))
and low growth,
and Wolf (1989)
are from the World
that
su~~t
Bank’s
38
1986
to
rate and
much
less,
numbers
are
50%
28%
to
65%
was
political
instabil-
high
Korea more
went
the tyl]e of l)ath analyzed
rights and to the richest
in Proposition
(1 993)
model
over 1-1985;
Development
3.6070
similarly
Their indices of political 40% of hea] th subsidies
An-
to cultural
assassinations,
here are averages
the
behind
versus
experienced
and Ades’
rate
respectively.
(5.55%
rights,
in-
for
far
as opposed
property
and 1995 World
and
enrollment
of 6%
rate
economic
or even Verdier
data mentioned
the
GDP shares
hypothesis,
as opposed
fertility
of
with
These
Philippines:
and political
and the fact
is cited by Alesina
to the wealthy
Most of the Barro
conflict
1.46%
pattern,
Korea.
of 42%
it reflects
had more revolutions
crises and constitutional
(this number
rates
from the poor,
redistributed
in this inst ante, as the two countries
relevant
evidence
same
enrollment
higher
which
Once again, the
versus
accumulation
of the
1965
much
to
South
in 1985,
Korea’s
Philippines’
although
that
to 95%
reversed,
in
capital
past
3.71% the
to clearly
transfers in Korea were
government
1.58%
human
education
show
Korea
(see for example
pressure emanating
Philippines’
skyrocketed
tertiary
by redistribution
given below for tertiary
of
the
creditwnstrained
The Philippines
of the population
renta=king
for
1985),
appear
riots, coups,
characterized
of
linking distri-
of South
csse studies
by Barro and Wolf
expenditures
While
investment
Turning
levels of turmoil.
44Both
the occurrence
the most relevant mechanisms
redistributive
~ucation
44
1.66%
the
of the
is unclear. not
of excessive
1980.
went
19Yo;
piece
1965
does
and
educational
likely
one would expect
such as those compiled
respectively.
wnsistent
other
for detailed
half times higher than in the Philippines:
1970
a little
Korea’s
the
the interaction
which of the theories
the ballot box or the street, does not fare well. Public
between
5,13%
and property it is instability
Examining
power in determining
by asking:
this is not a substitute
hypothesis
low but still two-and-a
of
whether
in this paper best shed light on the experiences
While
show up in basic indicators
on
that matters.
of political
to conclude
(1995) on Korea and Taiwan),
whether
of policy
so as to determine
redistribution
and inequality
full circle, it is tempting
and growth
Rodrik
side, the vatiances
also seems desirable.
Coming bution
inequality
work to move closer
will be open or latent (off the equilibrium
On the empirical
or just the average level of extra–legal
and empirical
the
9,
of purely the figures
Reports.
civil liberties protection
were also virtually
similar. 45 The more discriminating
discussed earlier (rule of law, enforceability
instrumental
in inducing banking
system,
government,
45 Barro are Sttie
not only in creating
Korean households
without
and Wolf’s
excessive
(1989)
= (0.00, 0.00),
data
expropriation
a favorable
climate
for business
to entrust much of their considerable
which in turn channeled
(2.1,
enforceability
2.3, 2.6)
last ten years.
of contracts
and the Philippines The averages
2.2 and 1.7 respectively.
number
of th=
Rwt = (2.12, 1.00),
and ~.3,
and nationalization
potential.
(1 .5, 1.9, 2.2).
]]attern
of these three BERI
The
average
on a scale of zero to 50.
ICRG
See Keefer
rights was
investment,
deemed
events
l)er year
for Korea
Assass= (0.12, 0.73),
= (0.38,0.08), Constih = (0.22, 0.09), Polr-ight = (4.80, 4.7), CiVlib = (5.0, 4.5). dGThe earliat available data are BERI’s (BusinH Environmental Risk Intelligence) delays,
risk, mrruption,
but also
strategic
by the
along the way.
on the average
Revel= (0.31, 0.46),
rights
savings to a stat~controlled
credit towards the industries
dissipation
of property
ahead. 46 This general security of property
etc. ), on the other hand, put Korea distinctly probably
of contracts,
indicators
This xores
computed
(International and Knack
by Keefer
Country (1995)
39
1972 scor=
On a scale of O (worst) is confirme(l
and Knack
for sourc=
scores
(1995) over
and definitions.
data
Crisis
for bureaucratic
to 4 (best),
by the additional
Risk Guirle)
and the Phili])pines
CoILp= (0.09, 0.00),
Korea
available
over 198&l 1986–1995
scored for the
995 were were 36.7
Appendix Proof of Proposition
1
Let us first study the properties
of the function
that it has a unique, global maximum
‘—= PP
1 +pp(l (1 -T)2(1
is positive
if and only if 72 < 1 + l/p~.
where z -
–~~.
+pp(l
~early,
(13) implies
(Al)
-T))2 concave
i’s utility function
also strictly concave
claims which follow equation
by (12).
– 72)
Thus V is strictly
By (11), individual and therefore
defined
at ~ = O. Moreover,
v“(T)
p) ln(A + Bi~)
V(7)
on the policy
U;(7)
(z, 1),
is of the form V(~)
on (z, 1). The envelope
(11 ), as well as Proposition
domain
theorem
+ (1 +
then yields the
1.
Proof of Proposition 3 The result for taxation
For A small enough,
was proved in the text.
~2r/~A~A
to A makes the second apply to intertemporal
Next, note from (9) that
was seen to be negative,
term again dominated efficiency
and the fact that
dT/dA
by the first; hence the result,
is proportional
Similar derivations
V.
Proof of Proposition 4 The first claim follows impli-
directly
from
(22).
Differentiating
that
this equation
and using
(6) also
. (A.2)
Since the left side is increasing This maximum
is interior
in ~, g increases with r up to some 79(A)
(and increasing
in A) if and only if
(1- P)A2> (1 As z < –1,
it suffices for instance
rate is positive
E [z, 1), then decresses.
-Z)2;1
(A,3)
+PP)’
that (1 – @)A2
> 1/(4 + 8p~).
>0~(1–~)A2
1 > ~+pP.
The growth-maximizing
tax
if 79(A)
(A.4)
Proof of Proposition 5 Because W(7)
is the sum of V(T)
on (z, 1). The same statements social and (interior) the planner’s
and a concave,
hold for individual
private optima
quadratic preferences
function
of 7, it is strictly concave
ZJi(T), now given by (24).
are thus respective] y defined by W’(~)
case, (23) implies that V’(~p)
= –p@(l
40
– D)(1 – 7P)A2.
The
= O and ZJ’i(7) = O. In
The fact that V is strictly
mncave
and maximized
in the first-order
at zero then yields Claim 1, while Claim 2 follows from setting in w; = m
condition
U’i (~) = O derived
the agent with average wealth, which by (25) is equivalent ~tivalently,
W’(I
7P >1
– (2p)-1
+
in w = m + A2/2.
to W’(rP)
- (29)-1)
from
>0,
Clearly,
Next,
let ra be the preferred
tax of
~“ < Tp if and only if U’a(~p)
+ [p~2(l – 7P) – p~/2]A2
‘
CZaim 3. Finally, let us show that when ~ < 1/4 there exists an interval where the tax rate
Tm set by the median voter is less inefficient zero the median
mnverges
to the mean, and both Tm and To converge
which is 7P(O) = O. Therefore, and W’(rm)/p~ W’(~”) W(r”)
= –~(1
than that set by the average agent, ~a. As A goes to
from (25):
– 7~)A2
W’(~a)/p~
= –@A2
= (1/2 –Q(l
+ 0(A2).
to the planers’
–T”))A2
optimum,
= (1/2 –@)A2
+O(A2)
Thus if ~ < 1/4 and A is small enough,
> –W’ (~m), which implies (using Taylor expansions)
that
Tp – r“
>
rm
– Tp
and therefore
< w(~m).
Proof of Proposition 6 Strictmncavity of individual preferences U’(7) implies that the equilibrium defined by the fist~rder
rendition
as long as there is an interior strictly
increasing
(26), or
solution
~“ (A, J) > Z. Since WA < 0 < —Vr it is clear that ~“ is
in A. Proposi tion 5 showed that maximizing
Aw < A because for J > A, V(O, A, J) Aw is obvious;
W requires
implying conversely,
value which makes 7* (A, ~) equal to Z, T* (A, A) falls below ~g (A), ~G is interior,
as long as (A. 3) is satisfied.
the effects of A on equilibrium
tax rate is uniquely
taxes r*(A,
This concludes
W’(~*)
J w > O; conversely >0
since 7P(A)
>0.
as ~ increases towards the
implying
g’ (~”) > 0. Therefore
the proof of Claim 1, which concerns
A).
I now turn to the effects of A on ~“, and the resulting contributions to (A,5) is interior for all A if and only if maxA{(P@)-lV’(l)
to W and g. The solution
+ (1 – Z)A2
– AA}
>0,
or
(A.6)
which will be assumed ~uations
(A.5)
from here on (since z < – 1, a sufficient
condition
is ~ 2 0 and W’(7*)
< 0, hence
g’(~’)
0 if and only if ~“ S 1 – J/(2 A),
(13) into (A.5),
or
this means
‘%-’A~(’::?;:4)
that p~A2/4
=*m
< 1, hence
~
from r’ (O, A) = O to a minimum
the ~ = O axis at A = A. Finally,
to one with 1 – ~“ = O/A,
> 0. Therefore 7“ 0, ~7*/~A
+ pO~/2A)
(A.6)
0 for each A. For further
O~O–
reference
l/0=
A,
(All)
note also that this limit is
reached from below: V(A, ~) E n:, ~t
us now turn to the efficiency
and only if ~’ 21
– A/~A.
consequences
Equivalently,
–T*(A,A))A @2/(1
– ~) then W’(~*)
where taxes decre=e
with inequality,
taxes
inequality,
increases
limA+w[A(l A2 > p2/(1 Cue
with
– 7P(A))]
=
> 0 for all A. Therefore
these reductions
although
(1 – ~)-112,
never which
on the interval
in ~“ are always inefficient.
sufficiently is less than
fast.
Indeed,
6 defined
[0, A]
On [A, +aI)
one can show
in (All)
that
if and only
if
–p).
(ii):
if A2 < ~2/(1
– ~) then W’(r*)
>0
on [0, A) and W’(T*)
A > A. This implies that W’(r*)
T“ decreases subsidized
with inequality,
>0
over the whole interval
[0, A] where
and even over a strict superset of the interval [0, A] where capital is
(~ < O). This is intuitive,
as the planner
42
always wants to set ~P(A)
> 0. Only once
inequality
has reached
This mncludm
taxes ~“ (A, A) > 0 to excessively
the proof of Claim 3, u far as intertemporal
It only remains (A.2)
A does it drive equilibrium
and (A.5)
to mnsider
AA
-
< (1
7“(1 – 7*)
–@)A2
T*[A/A – (1
@
sinm the losses from heterogeneity
tax effects on savings are of first order.
A))
>
0 e
20 e
the growth
efficiency
high levels.
>0.
(A.14)
is never sat isfiecl.
order in A2, whereas the induced
the condition
in the neighborhood
of A
1, > 1. This inequality
holds if and only if ~ is not too close to
one, in which case there exists a range of A’s satisfying parameter
configuration,
tax cuts induced
(A.6)
by rising inequality
such that g’(~*(A,
A)) >0.
In this
just below A reduce growth,
while
tax increases which occur as A rises just above A tend to improve it. To show that this last result is not specific to the region regressive of tipi tal subsidies A = A. Sin~
7“(A, A) = O, (A.2)
A)) >0 e A2>
it can be shown that the right-hand
some fied
upper bound.
in the neighborhood
is compatible
of A = A+ generate
asymptotically: of maximizing increwe
(1-B):l+PP)”
side is greater than of (A.6)
tax increases
(A. 14) as A ~
1) is equivalent
with (A.6)
at
as long as ~ is below
In that case there exists a range of A’s such that increases in inequality
growth rate, Finally, let us mnsider A ~ ~19, which by (Al
g’ (7”)
shows that:
g’(7*(A,
Again
(~” < O), let us next examine
to A2 2 ~2/(1
to positive
levels which still improve
m and 1 – ~“ s O/A,
– ~). When this condition
if and only if ~ is s not too close to one) holds,
equilibrium
taxes increase with inequality,
not only W but even g. men
too fast with rising inequality,
The condition
the
becomes
(which once more
g’(~*(A,
A)) > 0 even
but too slowly from the point of view
A2 < ~2/ (1 – ~), on the other hand, taxes eventually
whether from the point of view of welfare or from that of
growth,
43
Proof of Proposition 7 CZaim 1 immediately A >0,
results from Propositions
4 and 6, or directly
from
(A.7)-(A.8),
For
let us write out:
(pPA)-’
-(1
=
_
~
-Q)(I
(~ -PA(1
-T”)’+
of the second
D)(l – ~*) A]2/4.
(1-9)
(1-W
Therefore, V“(T*)
(2 - B)2 4 +
) by (Al).
term is a qutiatic (dW/dA).=T*
polynomial
greater than 6-2 (1+ p~(l – Z) ) -2. Provided
-~”)
-A)
+ A2
(A.16)
“
in A whose maximal
value is [(2 –
as long as: –v//(T*)
~’ 4(1 –~)
Now, recall from (A. 12) that A(l
(2A(1
–V’’(7*)/p~
T—T’ The numerator
-T*))
A and ~g/~A
for A large enough in this c~e,
~
and W’(~*)
>0,
-p)(l
which
3) (in which ~
the proof of Claim 2.
to make clear the role played by @ 0
First, we showed earlier that if ~2 < ~2/( 1 – ~), g’ (~”) becomes
large enough.
A, for which polynomial
in ~ whose maximal
T*
>0.
The numerator
value is (1 –7’)2(1
of the second
+ (1 –~)~*)2A2/4.
term is a quadratic
Therefore,
(dg/dA),=T.
A as long as: (1-
V“(T*)
B)(l-W
> (1 +(1 )
As A tends to infinity, 1 + 6-2>
(1 – @/2)2/(1
1 – T’ H 13/A and –V’’(~*)/(p~A2)
complete
>0.
markets we,
= 6-2,
– ~). This holds if and only if ~ is below
hence the first part of Claim 3. Conversely, A2Q limA+o[–A/~*]
-p)T*)2 47.
Indeed as A -0 where ~g/ar
so the rendition
some critical
bmmes
value in (O, 1),
one can show from (A. 18) that lim~+O[(dg/dA)T=T. (or also as ~ -
~j+l,
By Proposition
at some 7P (A ~) > 0. Since W(7, Aj)
–z)2)).
and let W(~j,
5, each W(-, Aj)
Aj,)
is strictly
is decreasing
in its second argument,
~(~j) Aj,),
(A.19)
this allows us to write:
W(Tj+I,
Aj+I,)
hence the result that intertemporal state.
The same holds a fortiori
efficiency
~(~j+l,
Aj,)
declines
from Tj to ~j+l
A ranking similar to (A. 19) obtains for gowth
such is the case in particular all j, implying
A;
> s2/(1