Trends in global income
inequality and their political
implications
Branko Milanovic
LIS Center; Graduate School City University of New York
Autumn 2014
Branko Milanovic
A. National inequalities mostly
increased
Branko Milanovic
Ginis in the late 1980s and around now
1985-90
After
2008
Change
Average Gini
36.3
38.8
+2.5
Pop-weighted
Gini
GDP-weighted
Gini
33.9
37.3
+3.4
32.2
36.4
+4.2
Countries with
higher Ginis
32.0
36.2
+4.5
Countries with
lower Ginis
42.8
39.5
-3.3
Branko Milanovic
From final-complete3.dta and key_variables_calcul2.do (lines 2 and 3; rest from AlltheGinis)
60
70
Ginis in the late 1980s and around now
50
40
30
GTM
HND
PAN
CHL
CHN
CRI
ECU DOM
BOLSLV
USA
PER
NGA
MYS
SGP
URY ARG
CIV
UGA
MKD
ISR
GEO
TUR
IRN
MRT
RUS
KOR
THA VEN
PHL
IDN
GBR
LVA BGR
PRT
JOR
LKA
KGZ
IND
CAN
ITA
LTUPOL
FRA GRC MLI
MDAROU
ESP BGD
JPN
TWN
EST
IRL
DEU AZE
TJK
HRV
PAK AUS
BEL
NLD
FIN
AUTKAZ
HUN
NOR ARM
SVK
DNK
SWE
CZE UKR
BLR
SVN
BRA
MEX
Branko Milanovic
20
Gini after 2008
COL
20
30
40
50
Gini between 1985 and 1990
60
twoway (scatter bbb aaa if year==2000, mlabel(contcod) msize(vlarge)) (function y=x, range(20 60) legend(off) xtitle(Gini between 1985 and
1990) ytitle(Gini after 2008)) using allginis.dta
60
Ginis in 1988 and 2008 (population-weighted countries)
BRA
MEX
USA
40
RUS
CHN-R
CHN-U
IND-U
30
Gini in 2008
50
NGA
IND-R
20
Branko Milanovic
20
From twenty_years/… key_variables_calcul3.do
30
40
Gini in 1988
50
60
20
Convergence of countries’ Ginis: an empirical
observation without theoretical explanation
GTM
10
ECU
ARG
BGR
-10
0
HUN
POL
CZE
CHN
GBR NZL
CHL
USA
SYC
JAM DOM
HKG
SGP
PAN
VEN
IND
PRI
ISR
COL
IDN
SDN
IRN
ZMB
LKA
BELTWN
FJI
CAN
KOR
THA
SLV
BRA
AUS
GRC
CRI
BOL
NLDESP
SWE
PAK IRL
MEX
BHS
JPN
PRT
BGD
DEUITA
NOR
FIN
BRBMYS
DNK EGY
HND
PHL
TUN
PER
TTO
FRA
TZA
TUR SLE
NPL
-20
GAB
20
30
40
50
average country Giniall before 1980
60
twoway (scatter change_gini gini_pre1980 if nvals==1, mlabel(contcod)) (lfit change_gini gini_pre1980, yline(0, lpattern(dash)) ytitle(change in Gini after 1980)
legend(off))
Using Allthe Ginis.dta
Branko Milanovic
Market, gross and disposable income
Ginis in the US and Germany
Germany
.25
.25
.3
.3
.35
.35
.4
.4
.45
.45
.5
.5
USA
1970
1980
1990
year
Define_variables.do using data_voter_checked.dta
2000
2010
1970
Branko Milanovic
1980
1990
year
2000
2010
Issues raised by growing national
inequalities
• Social separatism of the rich
• Hollowing out of the middle classes
• Inequality as one of the causes of the global
financial crisis
• Perception of inequality outstrips real
increase because of globalization, role of
social media and political (crony) capitalism
(example of Egypt)
• Hidden assets of the rich
Branko Milanovic
Some long-term examples set in the
Kuznets framework
Branko Milanovic
Inequality (Gini) in the USA 1929-2009
(gross income across households)
50.0
48.0
46.0
44.0
42.0
40.0
38.0
1929
From ydisrt/us_and_uk.xls
1939
1949
1959
1969
1979
1989
1999
2009
Kuznets and Piketty “frames”
70
Ginis for England/UK and the United States in a very long run
60
50
USA
40
30
England/UK
20
10
0
1600
1650
1700
1750
1800
1850
1900
1950
2000
2050
From uk_and_usa.xls
11
Contemporary examples of Brazil and China:
moving on the descending portion of the Kuznets
curve
China, 1967-2007
Gini
40
40
50
Gini
50
60
60
Brazil 1960-2010
7.5
8
8.5
ln GDP per capita
updated Giniall
9
Fitted values
twoway (scatter Giniall lngdpppp if contcod=="BRA", connect(l) ylabel(40(10)60) xtitle(2000
6000 12000) ytitle(Gini) xtitle(ln GDP per capita)) (qfit Giniall lngdpppp if contcod=="BRA",
lwidth(thick))
From gdppppreg4.dta
9.5
5
6
7
ln GDP per capita
updated Giniall
8
9
lowess Giniall lngdpppp
twoway (scatter Giniall lngdpppp if contcod=="CHN" & year>1960, connect(l) ylabel(40(10)60)
xtitle(2000 6000 12000) ytitle(Gini) xtitle(ln GDP per capita)) (qfit Giniall lngdpppp if
contcod=="CHN" & year>1960, lwidth(thick))
From gdppppreg4.dta
12
B. Between national inequalities
remained very high even if
decreasing
Branko Milanovic
30
Distribution of people by income of the country where they
live: emptiness in the middle (year 2013; 2011 PPPs)
India, Indonesia
10
Percent
20
China
W.Europe, Japan
USA
0
Brazil, Mexico, Russia
0
From defines.do in interyd
10000
20000
30000
GDP per capita in 2005 PPP
40000
50000
percentile of world income distribution
10 20 30 40 50 60 70 80 90 100
Different countries and income classes in global income distribution in
2008
USA
Brazil
Russia
China
India
1
Branko Milanovic
1
From calcu08.dta
20
40
60
country percentile
80
100
100
90
80
Denmark
50
60
70
Uganda
30
40
Mali
10
20
Tanzania
1
Mozambique
1
5
10
country ventile
15
20
Countries with more than 1% of their population in top global percentile
(above $PPP 72,000 per capita in 2008 prices)
14
12
12
10
9
9
CHE
SGP
8
7
7
CAN
LUX
6
6
5
4
3
2
2
2
2
2
2
2
CYP
DEU
IRL
KOR
NLD
TWN
3
0
From summary_data.xls
FRA
NOR
GBR
Branko Milanovic
JPN
USA
C. Global inequality is the product of
within- and between-county
inequalities
How did it change in the last 60 years?
Branko Milanovic
Essentially, global inequality is
determined by three forces
• What happens to within-country income
distributions?
• Is there a catching up of poor countries?
• Are mean incomes of populous & large
countries (China, India) growing faster or
slower that the rich world?
Branko Milanovic
.75
Global and international inequality
after World War II
.65
Concept 3
.55
Concept 2
.45
Concept 1
1950
1960
1970
1980
year
1990
2000
Concept2: 1960-1980 from Bourguignon & Morrisson
Defines.do using gdppppreg5.dta
Branko Milanovic
2010
.65
Concept 2 inequality with 2011 PPPs
and without China and India
.55
.6
all countries
.5
Without China
Without India and China
.45
47
1940
Defines.do using gdppppreg5.dta
1960
1980
year
Branko Milanovic
2000
2020
Population coverage
1988 1993 1998 2002
2005
2008
2011
Africa
48
76
67
77
78
78
71
Asia
93
95
94
96
94
98
89
E.Europe
99
95
100
97
93
92
87
LAC
87
92
93
96
96
97
97
WENAO
92
95
97
99
99
97
95
World
87
92
92
94
93
94
88
Branko Milanovic
Non-triviality of the omitted countries (Maddison vs. WDI)
Three important technical issues in the
measurement of global inequality
• The ever-changing PPPs in particular for
populous countries like China and India
• The increasing discrepancy between GDP per
capita and HS means, or more importantly
consumption per capita and HS means
• Inadequate coverage of top 1% (related also
to the previous point0
Branko Milanovic
The issue of PPPs
Branko Milanovic
100
150
The effect of the new PPPs on countries’ GDP
per capita (compared to the US level)
ZMB SDN
GHA
0
50
PAK
SAU
JOR
IDN
SUR
MNG
OMN
EGY
KWT
FJI
AZE KAZ
QAT
DZA
CPV
THA
MAC
MDG
LKA
GTM
BRN
PHL
VNM
NER
MAR
RUS
MLI
VEN
GNQ
COG
ARE
TCD
HTI
MYS
MDV
IND
MRT
TGO
KEN
LSOKGZ
NGA MDA
NAM BRA
AGO
CHN
SLE
UGA
SWZ
LVA
SGP
BDI
CHL
NOR
TUR
CMR
PRY
GEO
BTN
UKR
BIHCOLMNE
CHE
LUX
GIN
URY
KHM
HUN
BGR
MEX
SEN
DNK
ARM
ESTMLT
LTU
TTO
BLR
DOM
ITA
MKD
CAF
TUN
NZL
ETH
BOL
ZAF
MWI
BEN
BLZ
HRV
PER
ECU
AUS
HND
SLV
NIC
POL
GNB
SRB
FRA
BEL
TJK
FIN
MUS
SVK
JAM CRIPAN
PRT
GRC
ESP
TWN
SWE
GAB
DEU
AUT
RWA
IRL
USA
BFA
TZA
NLD
CAN
ISL
SVN
ISR
HKG
CZE
DJI
ALB
JPN
MOZ
GBR
KOR
LBR
BWA
CYP
GMB
NPLBGD YEM
CIV LAO
-50
BHS
COM
50000 100000
150000
gdppc in 2011ppp
C:\Branko\worldyd\ppp\2011_icp\define
Branko Milanovic
The effect of new PPPs
Country
GDP per capita
increase (in %)
GDP per capita
increase populationweighted (in %)
Indonesia
90
---
Pakistan
66
---
Russia
35
---
India
26
---
China
17
---
Africa
23
32
Asia
48
33
Latin America
13
17
Eastern Europe
16
24
WENAO
3
2
.85
Global income inequality using
nominal dollars
.75
.8
Concept 3
.65
.7
Concept 2
Concept 1
.55
.6
63
1970
1980
From two_concepts_exrate.do using Global_new5.dta
1990
Year
2000
2010
The gap between national accounts
and household surveys
Branko Milanovic
.65
Both the level and change: Use of GDP per capita gives
a lower lever and a faster decrease of global inequality
.6
HS means--countries in HS sample
.55
GDPs pc countries in HS sample
.45
.5
Gini
usual Concept 2
1990
Defines.do based on gdppppreg5.dta
1995
2000
year
Branko Milanovic
2005
2010
2015
How global inequality changes with
different definitions of income
72
71
Step 2
70
69
68
Step 1
GDP ppp
67
Consumption
66
Survey mean
65
64
63
62
Global inequality
Branko Milanovic
Step 1 driven by low consumption shares in China and India
(although on an unweighted base C/GDP decreases with GDP)
1
1.2
C/GDP from national accounts in year 2008
.6
.8
USA
.4
India
.2
China
1000
10000
GDP per capita in ppp
50000
twoway scatter cons_gdp gdpppp if group==1 & cons_gdp<1.4 [w=totpop], xscale(log) xtitle(GDP per capita in ppp) xlabel(1000 10000 50000)
ytitle(share of consumption in GDP) title(C/GDP from national accounts in year 2008)
Branko Milanovic
using final08,dta
Step 2. No clear (weighted) relationship between
survey capture and NA consumption
1.2
survey mean/consumption from national account in year 2008
.8
1
China
.4
.6
USA
.2
India
1000
10000
GDP per capita in ppp
50000
twoway scatter scale2 gdpppp if group==1 & scale2<1.5 [w=totpop], xscale(log) xtitle(GDP per capita in ppp)
xlabel(1000 10000 50000) ytitle(survey mean over NA consumption) title(survey mean/consumption from national
account in year 2008)
Branko Milanovic
The issue of top underestimation
Branko Milanovic
Rising HS/NA gap and top
underestimation
• If these two problems are really just one & the
same problem.
• Assign the entire positive (NA consumption –
HS mean) gap to national top deciles
• Use Pareto interpolation to “elongate” the
distribution
• No a priori guarantee that global Gini will
increase
Branko Milanovic
Gini: accounting for missing top
incomes
1988
1993
1998
2003
2008
Surveys
only
72.5
71.8
71.9
71.9
69.6
NAC
instead of
survey
mean
71.5
70.5
70.6
70.7
67.6
NAC with
Pareto
71.8
70.8
71.0
71.1
68.0
NAC with
top-heavy
Pareto
76.3
76.1
77.2
78.1
75.9
Branko Milanovic
The results of various adjustments
• Replacing HS survey mean with private
consumption from NA reduces Gini by 1 to 2
points
• Elongating such a distribution (that is, without
changing the consumption mean) adds less than
½ Gini point
• But doing the top-heavy adjustment (NA-HS gap
ascribed to top 10% only) adds between 5 and 7
Gini points
• It also almost eliminates the decrease in global
Gini between 1988 and 2008
Branko Milanovic
How Global Gini in 2008 changes with different
adjustments
Increase in global Gini with each “marginal”adjustment
10
8
6
4
2
Allocate the
gap
proportionally
along each
national income
distributions
Allocate the gap
to top 10% and
add Pareto
“elongation”
Allocate the gap
proportionately
and add a Pareto
“elongation”
0
-2
-4
Branko Milanovic
With full adjustment (allocation to the top 10%
+ Pareto) Gini decline almost fully disappears
80
78
Top-heavy allocation of the
gap + Pareto adjustment
76
74
Survey data only
72
70
68
66
64
1988
1993
1998
Branko Milanovic
2003
2008
D. How has the world changed
between the fall of the Berlin Wall and
the Great Recession
Branko Milanovic
Real income growth at various percentiles of global
income distribution, 1988-2008 (in 2005 PPPs)
Real PPP income change (in percent)
80
X “China’s middle class”
$PPP2
70
60
$PPP4.5
$PPP 110
$PPP12
50
40
30
20
Branko Milanovic
X
10
“US lower middle class”
0
0
20
40
60
80
100
Percentile of global income distribution
From twenty_years\final\summary_data
Estimated at mean-over-mean
Real income gains (in $PPP) at different percentile of global income
distribution 1988-2008
90
World
Real PPP income change (in percent)
80
70
60
50
Without China
40
30
20
10
0
-10
-20
0
10
20
30
40
50
60
70
Percentile of global income distribution
80
90
100
Quasi non-anonymous GIC: Average growth rate 1988-2008 for
different percentiles of the 1988 global income distribution
Branko Milanovic
40
60
80
Growth incidence curve (1988-2008) estimated
at percentiles of the income distribution
0
20
mean growth
2
Using my_graphs.do
10
20
30
40
50
60
70
80
percentile of global income distribution
Branko Milanovic
90 95 100
Mean-on-mean
Distribution of the global absolute gains in income, 1988-2008:
more than ½ of the gains went to the top 5%
30
27
25
Distribution (in percent) of gain
25
20
15
10
10
5
3 4
3
2 2
2
1
1
0 0 1 1 1
5 4
5
1
3
5
10
15
20
25
30
35
40
45
50
55
60
65
70
75
80
85
90
95
99
100
0
Branko Milanovic
ventile/percentile of global income distribution
From summary_data.xls
Annual per capita after-tax income in international dollars
US 2nd decile
5000
Chinese 8th
urban decile
500
1988
From summary_data.xls
1993
1998
2003
2008
2011
1
Global income distributions in
1988 and 2008
1988
Emerging global “middle
class” between $3 and $16
100000
50000
30000
10000
6000
3000
1000
300
0
.2
.4
density
.6
.8
2008
log of annual PPP real income
twoway (kdensity logRRinc [w=pop] if logRRinc>2 & bin_year==2008 & keep==1 & mysample==1) (kdensity logRRinc [w=pop] if logRRinc>2 &
bin_year==1988 & keep==1 & mysample==1, legend(off) xtitle(log of annual PPP real income) ytitle(density) text(0.95 2.5 "1988") text(0.85 3
Branko Milanovic
"2008"))
Or using adding_xlabel.do; always using final_complete7.dta
Increasing gains for the rich with a
widening urban-rural gap
Urban and rural Indonesia
210
urban
2
3
4
5
6
decile
From key_variables_calcul2.do
7
200
190
rural
180
rural
Branko Milanovic
1
urban
8
9
10
170
200
250
300
350
combined real_growth 1 and 2
400
220
450
Urban and rural China
1
2
3
4
5
6
decile
7
8
9
10
E. Issues of justice and politics
1. Citizenship rent
2. Migration
3. Hollowing out of the middle classes
Branko Milanovic
Global inequality of opportunity
• Regressing (log) average incomes of 118
countries’ percentiles (11,800 data points)
against country dummies “explains” 77% of
variability of income percentiles
• Where you live is the most important
determinant of your income; for 97% of
people in the world: birth=citizenship.
• Citizenship rent.
Branko Milanovic
Is citizenship a rent?
• If most of our income is determined by
citizenship, then there is little equality of
opportunity globally and citizenship is a rent
(unrelated to individual desert, effort)
• Key issue: Is global equality of
opportunity something that we ought to
be concerned or not?
• Does national self-determination dispenses
with the need to worry about GEO?
Branko Milanovic
The logic of the argument
• Citizenship is a morally-arbitrary circumstance,
independent of individual effort
• It can be regarded as a rent (shared by all
members of a community)
• Are citizenship rents globally acceptable or
not?
• Political philosophy arguments pro (social
contract; statist theory; self-determination)
and contra (cosmopolitan approach)
Branko Milanovic
The Rawlsian world
• For Rawls, global optimum
distribution of income is simply a
sum of national optimal income
distributions
• Why Rawlsian world will remain
unequal?
Branko Milanovic
Global Ginis in Real World, Rawlsian World, Convergence
World…and Shangri-La World (Theil 0; year 2008)
Mean country
incomes
All equal
Different (as
now)
All equal
0
68
(all country
Ginis=0)
Different (as
now)
30 (all mean
incomes same; all
country Ginis as
now)
Individual incomes
within country
Branko Milanovic
98
Conclusion
• Working on equalization of
within-national inequalities will
not be sufficient to significantly
reduce global inequality
• Faster growth of poorer countries
is key and also…
Branko Milanovic
Migration: a different way to reduce
global inequality and citizenship rent
• A new view of development:
Development is increased income for
poor people regardless of where they
are, in their countries of birth or
elsewhere
• Migration and LDC growth thus become
the two equivalent instruments for
development
Branko Milanovic
A migrant point of view: trade-off between country’s
mean income and its inequality
How much is one Gini point change worth in terms of mean country
income?
14
12
Percent of income
10
8
6
4
Increase
in Gini
Decrease in Gini
2
0
1
2
3
4
5
6
7
8
9
10
11
12
Ventile
Branko Milanovic
From interyd..\ventil_vs_country.xls
13
14
15
16
17
18
19
20
Political issue: Global vs. national level
• Our income and employment is increasingly
determined by global forces
• But political decision-making still takes place at
the level of the nation-state
• If stagnation of income of rich countries’ middle
classes continues, will they continue to support
globalization?
• Two dangers: populism and plutocracy
• To avert both, need for within-national
redistributions: those who lose have to be helped
Branko Milanovic
Final conclusion
• To reduce global inequality: fast
growth of poor countries +
migration
• To preserve good aspects of
globalization: redistribution
within rich countries
Branko Milanovic
Additional slides
Branko Milanovic
H. Global inequality over the long-run
of history
Branko Milanovic
Global income inequality, 1820-2008
100
(Source: Bourguignon-Morrisson and Milanovic; 1990 PPPs )
80
Theil
20
40
60
Gini
0
Branko Milanovic
1820
1860
1900
1940
1980
year
twoway (scatter Gini year, c(l) xlabel(1820(40)2020) ylabel(0(20)100) msize(vlarge) clwidth(thick)) (scatter Theil year, c(l) msize(large)
legend(off) text(90 2010 "Theil") text(70 2010 "Gini"))
2020
Shares of global income received by top 10% and bottom 60% of world population
70
Top 10% (L-M data)
60
Percentage share of global income
Top 10% (B-M data)
50
40
30
20
Bottom 60% (B-M data)
10
Bottom 60% (L-M data)
0
1800
1850
1900
1950
Year
Branko Milanovic
2000
2050
A non-Marxist world
• Over the long run, decreasing importance of
within-country inequalities despite some
reversal in the last quarter century
• Increasing importance of between-country
inequalities (but with some hopeful signs in
the last five years, before the current crisis),
• Global division between countries more than
between classes
Branko Milanovic
Composition of global inequality changed: from being
mostly due to “class” (within-national), today it is mostly
due to “location” (where people live)
100
Theil 0 index (mean log deviation)
80
Location
60
Location
40
20
Class
Class
Branko Milanovic
0
1870
Based on Bourguignon-Morrisson (2002), Maddison data, and Milanovic (2005)
2008
From thepast.xls
Very high but decreasing importance of location in global inequality
90
Share of the between component in global Theil (0)
80
L-M data
70
Between component, in percent
B-M data
60
50
40
30
20
10
0
1800
From thepast.xls under c:\history
1850
1900
Year
Branko Milanovic
1950
2000
2050