how unequal is europe? · yields undesirable results: the owners of corporation can choose to...
TRANSCRIPT
How Unequal Is Europe?
Evidence From Distributional National Accounts, 1980–2017
Thomas Blanchet1,2 Lucas Chancel2,3 Amory Gethin2
OECD
ELS Seminar
September 10th, 2019
1Paris School of Economics – EHESS
2World Inequality Lab
3IDDRI
WI D.WO R L D
What Do We Know about Inequality in Europe?
Unlike some other parts of the world, there is quite a lot of data available
on the distribution of income in Europe.
So why write a paper on the topic?
• Not because of a lack of data per se. . .
• . . . but because existing data is scattered across a variety of sources,
so it can be hard to make sense of it.
Disparate set of indicators:
• hard to compare
• hard to aggregate
• hard to tell consistent stories
1
WI D.WO R L D
A Variety of Sources
• National Accounts
• Income concept of reference
• Inequality between countries, but not within countries
• Surveys
• Microdata: good geographical coverage, but only in recent years
• Tabulations: better coverage in older years, but inconsistent
concepts and units
• Covers the whole distribution, but with large biases at the top
• Tax data
• Better at covering the top of the distribution
• Covers only the top of the distribution
• Inconsistent concepts and units
2
WI D.WO R L D
Many Questions Lack a Clear Answer
⇒ Literature has struggled to answer simple questions:
• Is Europe as a whole more or less unequal than the United States?
• Is the difference between European and US inequality driven by
pre-tax incomes or redistribution?
• Is European inequality driven by the distribution of income between
or within countries?
• Which parts of the distribution have benefited the most from
European growth?
⇒ Difficulties monitoring of internationally agreed goals
• European institutions’ Pillar of Social Rights (2017)
• Sustainable Development Goals adopted by all European
countries (2015)
3
WI D.WO R L D
This Paper: Distributional National Accounts
This paper is an attempt to address this problem by constructing
distributional national accounts (DINA) for Europe since 1980.
Part of larger ongoing project in which we construct similar estimates for
as many geographical areas as possible, over as long as possible.
This paper contributes to that project in several respects:
• Productions of new estimates of the income distribution within the
DINA framework.
• Methodological advances to build such estimates in spite of data
limitations.
4
Conceptual Framework
WI D.WO R L D
A Short History of DINA
• Early 2000s: Piketty (2003) for France and Piketty & Saez (2003)
for the United States revived an old literature (Kuznets, 1953) that
used tax data to estimate inequality.
• Work extended to large number of countries by many researchers,
and collected in two collective volumes edited by Piketty & Atkinson
(2007, 2010).
• Data collected into the World Top Income Database (WTID), which
then became the World Inequality Database (WID.world),
maintained by the World Inequality Lab.
• Estimated based on raw tax data have many qualities, but also many
drawbacks:
• Inconsistent concepts
• Only covers the top of the distribution
• Does not take redistribution into account
5
WI D.WO R L D
DINA in Broad Strokes
• The DINA project was created to address those criticisms.
• Set of guidelines for measuring inequality:
• consistent concepts.
• consistent methodology.
• consistent with the framework put forth by the System of National
Accounts (SNA).
• Combine existing sources (surveys, tax data, national accounts).
• Be consistent the macro totals and distribute 100% of national
income.
• Describe both the top and the bottom of the distribution.
• Measure several income concepts, including pre-tax and post-tax
inequality.
6
WI D.WO R L D
Comparison With Other Initiatives
• Some papers have used harmonized income statistics (EU-SILC) to
estimate the distribution at the European level, but reying solely on
survey data (Filauro (2018) and Brandolini and Rosolia (2019)).
• Several intiatives to adress gaps between micro estimates and macro
totals:
• EG-DNA since 2011 at the OECD.
• Various “experimental statistics” published by Eurostat (2018) and
other national statistical institutes.
• Mostly rely on survey data.
• Only distribute a fraction of national income.
7
WI D.WO R L D
Comparison With Other Parts of the DINA Project
Several highly detailed country-specific studies:
• United States (Piketty, Saez, & Zucman), France (Garbinti,
Goupille-Lebret & Piketty, 2018), . . .
• Very precise work, very detailed decomposition, but takes a lot of
time. . .
• Will be very long until we can apply such work universally.
Our work seek to provide roughly comparable estimates:
• At a lower cost, using less precise data and simpler assumptions.
• (On the plus side, using more straightforward and transparent
assumptions.)
8
WI D.WO R L D
What We Do In This Paper
We construct DINA Estimates for 38 countries over 1980–2017.
• Consistent statistical unit (equal-split adults).
• Pre-tax and post-tax income.
• From the bottom to the top 0.001%.
We face numerous challenges in constructing but we try to overcome
them in a meaningful and transparent way.
• All the data is public and available on WID.world.
• All the codes are also available on WID.world.
• Detailed appendix describing the sources and specific assumptions
for each country.
9
WI D.WO R L D
DINA concept: net national income income
Distributing all of national income means accounting for income that
may never explicitly appear on the bank account of any household, yet is
still part of “economic growth”.
net national income = GDP - depreciation + net foreign income
We consider income from all sectors of the economy:
• Income of the household sector (incl. imputed rents)
• Income of the government:
• before taxes: production taxes
• after taxes: taxes - transfers + government consumption
• Income of corporations (retained earnings)
10
WI D.WO R L D
Standard Inequality Statistics Miss a Large Part of Income
interests and dividends
capital componentof mixed income
imputed rents
retained earnings
corporate taxabout halfof capital
income notcaptured by
commoninequalitystatistics
0%
5%
10%
15%
20%
25%Sh
are
of n
et n
atio
nal i
ncom
e
1980 1985 1990 1995 2000 2005 2010 2015Year
11
WI D.WO R L D
Yet “missing income” is still income
Not accounting for income outside of the household sector sometimes
yields undesirable results:
• The owners of corporation can choose to distribute income to
themselves arbitrarily.
• Some taxes are accounted for (e.g. income tax) but not others (e.g.
VAT).
• Countries with strong provision of public goods appear to have
poorer households.
We make simple and transparent assumptions to distribute these types of
income. Still a lot of room for improvement.
12
WI D.WO R L D
DINA concepts: pre-tax and post-tax income
• pre-tax income: income after the operation of social insurance
systems (pension and unemployment insurance), but before other
types of transfers. We add pensions and unemployment benefits, but
remove the social contributions that pay for them. Conceptually
similar to “taxable income” in many countries.
• post-tax: income after the operation of all government
redistribution. We remove all taxes, the remaining social
contributions, and then add all transfers and government
consumption.
13
Methodology
WI D.WO R L D
A myriad of income inequality datasets in Europe
Macro data:
• UN, OECD, Eurostat
Survey microdata:
• Eurostat surveys: SILC, ECHP, HBS
• Luxembourg Income Study (LIS)
• Imputation of social contributions
Survey tabulations:
• Diverse sources, compiled in WIID by the UN.
• We turn them into complete distributions using generalized Pareto
interpolation (Blanchet, Fournier & Piketty, 2018)
Tax data:
• WID.world
• Couple of formerly unused sources: Iceland, East Germany.
14
WI D.WO R L D
Data availability: good coverage for most of the population
Low (survey tabulations, low time coverage)
Medium low (survey tabulations, good time coverage)
Medium (survey tabulations + microdata)
Medium high (survey tabulations + microdata + tax data)
High (distributional national accounts)
15
WI D.WO R L D
Different factors explain discrepancies between sources
Three reasons why these data sources yield different results:
• Sampling error: survey sample size too small to properly capture
top income groups
• Conceptual differences: data refers to different concepts:
• income concepts
• statistical unit: individuals, households, per capita, per adults,
equivalence scales
• Non-sampling error: heterogeneous non-response in surveys,
misreporting
16
WI D.WO R L D
Correcting for sampling error with a generalized Pareto
Sampling error makes it difficult to study the distribution within the
top 10%
• extreme value theory provides tools to infer shape of the top tail
with limited samples sizes
• top tail modelled using the generalized Pareto distribution
P{X > x} =
[1− ξ
(x − µσ
)]−1/ξ
• estimated using the method of probability-weighted moments
(Greenwood et al., 1979)
17
WI D.WO R L D
Harmonizing concepts using machine learning
• Use existing survey data to establish correspondences between the
different concepts used in the literature:
• 3 income concepts: pre-tax income, post-tax income, consumption
• 5 statistical units: households, per capita, per adult, OECD
equivalence scale, square root equivalence scale
• 3× 5 = 15 distributions by year and country
• We train a machine learning algorithm to map how all these
different concepts relate to one another, and then use it to correct
for systematic biases.
18
WI D.WO R L D
Harmonization of Concepts: Example
Top 10% income share, Portugal
10%
15%
20%
25%
30%
35%
40%
1980 1985 1990 1995 2000 2005 2010 2015
Pre-tax income per adult(EU-SILC)
19
WI D.WO R L D
Harmonization of Concepts: Example
Top 10% income share, Portugal
10%
15%
20%
25%
30%
35%
40%
1980 1985 1990 1995 2000 2005 2010 2015
Pre-tax income per adult(EU-SILC)Post-tax income per adult(EU-SILC)
19
WI D.WO R L D
Harmonization of Concepts: Example
Top 10% income share, Portugal
10%
15%
20%
25%
30%
35%
40%
1980 1985 1990 1995 2000 2005 2010 2015
Pre-tax income per adult(EU-SILC)Post-tax income per adult(EU-SILC)Post-tax income per adult(ECHP)
19
WI D.WO R L D
Harmonization of Concepts: Example
Top 10% income share, Portugal
10%
15%
20%
25%
30%
35%
40%
1980 1985 1990 1995 2000 2005 2010 2015
Pre-tax income, adults(EU-SILC)Post-tax income, adults(EU-SILC)Post-tax income, adults(ECHP)Post-tax income, square-root scale(Atkinson, Rainwater and Smeeding 1995)Post-tax income, individuals(Gouveia and Tavares 1995)Post-tax income, households(Gouveia and Tavares 1995)
19
WI D.WO R L D
Harmonization of Concepts: Methodology
• Discretize the quantile function of distributions, normalized by the
average: Q1, . . . ,Qn
• Statistical model between distributions A and B:
QAk = f (pk ,Q
B1 , . . . ,Q
Bn ,Z ) + εk
where Z is a set of additional variables to improve prediction
(regional dummy, average income, demographic composition, tax
schedule, etc.)
The problem is hard statistically because:
• Non-parametric functional form.
• Large number of highly correlated predictors.
• Monotonicity constraint on QAk .
⇒ machine learning is very good at dealing with this type of problem: we
use a standard, state-of-the-art algorithm known as gradient tree boosting
20
WI D.WO R L D
Harmonization of Concepts: Example
Top 10% income share, Portugal
10%
15%
20%
25%
30%
35%
40%
1980 1985 1990 1995 2000 2005 2010 2015
Pre-tax income, per adult(harmonized)Post-tax income, per adult(harmonized)Pre-tax income, adults(EU-SILC)Post-tax income, per adult(EU-SILC)Post-tax income, per adult(ECHP)Post-tax income, square-root scale(Atkinson, Rainwater and Smeeding 1995)Post-tax income, per capita(Gouveia and Tavares 1995)Post-tax income, households(Gouveia and Tavares 1995)
21
WI D.WO R L D
Harmonization of Concepts: Example
Top 10% income share, Portugal
10%
15%
20%
25%
30%
35%
40%
1980 1985 1990 1995 2000 2005 2010 2015
Pre-tax income, per adult(harmonized, 95% CI)Post-tax income, per adult(harmonized, 95% CI)Pre-tax income, per adult(harmonized)Post-tax income, per adult(harmonized)Pre-tax income, adults(EU-SILC)Post-tax income, per adult(EU-SILC)Post-tax income, per adult(ECHP)Post-tax income, square-root scale(Atkinson, Rainwater and Smeeding 1995)Post-tax income, per capita(Gouveia and Tavares 1995)Post-tax income, households(Gouveia and Tavares 1995)
21
WI D.WO R L D
Correcting for misreporting and nonresponse using tax data
• Adapt the standard survey calibration framework (age, gender, . . . )
to improve the representation of the rich (Blanchet, Flores &
Morgan, 2018).
• Adjust weights to enforce constraints on the top income shares while
minimizing distortion from the original survey:
minf ∗
∫[f ∗(x)− f (x)]2
f (x)f (x) dx st. income shares from tax data
• Empirically, the reweighting depends on the inequality in the survey.
• When no tax data is available, we impute a reweighting profile using
the same machine learning method as before.
22
WI D.WO R L D
Distribution of incomes not captured by survey or tax data
Imputed rents:
• Total of imputed rents from the national accounts.
• Statistical matching using the distribution of imputed rents recorded
in EU-SILC.
Retained earnings:
• Distribution proportional to stock ownership.
• Using statistical matching with data from HFCS.
23
WI D.WO R L D
Distribution of incomes not captured by survey or tax data
Production taxes:
• Taxes on products distributed proportionally to consumption (from
HBS).
• Other taxes on production proportional to income.
Government consumption:
• Benchmark assumption: proportional to income (distribution
neutral)
• Experiment with other assumption to assess how government
consumption affects inequality.
24
WI D.WO R L D
United Kingdom: meaningful survey correction but little impact
of missing incomes
10%
15%
20%
25%
30%
35%
40%
1980 1985 1990 1995 2000 2005 2010 2015 2020
Harmonized surveyTax-corrected surveyDINA
25
WI D.WO R L D
Sweden: mild survey correction but significant role for missing
incomes
10%
15%
20%
25%
30%
35%
40%
1980 1985 1990 1995 2000 2005 2010 2015 2020
Harmonized surveyTax-corrected surveyDINA
26
Results
WI D.WO R L D
Large inequalities persist between European countries
Average national incomes in Europe, 2017
850€ / month
1500-2000€ / month
2500-4000€ / month
5000€ / month
27
WI D.WO R L D
In the long run, no clear trend towards convergence
Average national incomes of European regions, 1980-2017
.25
.5
.75
1
1.25
1.5
1.75R
egio
nal i
ncom
e pe
r adu
lt /
Euro
pean
inco
me
per a
dult
1980 1985 1990 1995 2000 2005 2010 2015
Northern Europe Western EuropeSouthern Europe Eastern Europe
28
WI D.WO R L D
Inequalities within European countries
Top 10% income shares across European countries: 1980 vs. 2017
Decreasinginequality
Increasinginequality
15
20
25
30
35
40
Top
10%
inco
me
shar
e (2
017)
15 20 25 30 35 40
Top 10% income share (1980)29
WI D.WO R L D
Inequalities increased in nearly all countries
Top 10% income shares across European countries: 1980 vs. 2017
ATBA BE
BG CY
CZFI
FRGB
GR
HRHU
IE
LT
MKNO
PL
RORS
SE
SK
CH
DE
EE ESIT
LULV
ME
DK
IS
NL
PT Decreasinginequality
Increasinginequality
15
20
25
30
35
40
Top
10%
inco
me
shar
e (2
017)
15 20 25 30 35 40Top 10% income share (1980)
29
WI D.WO R L D
Inequality rose more in Eastern and Northern Europe
Share of national income received by top 10%, 1980
20
30
40
30
WI D.WO R L D
Inequality rose more in Eastern and Northern Europe
Share of national income received by top 10%, 1990
20
30
40
30
WI D.WO R L D
Inequality rose more in Eastern and Northern Europe
Share of national income received by top 10%, 2000
20
30
40
30
WI D.WO R L D
Inequality rose more in Eastern and Northern Europe
Share of national income received by top 10%, 2007
20
30
40
30
WI D.WO R L D
Inequality rose more in Eastern and Northern Europe
Share of national income received by top 10%, 2017
20
30
40
30
WI D.WO R L D
Inequalities increased in all regions, but at different speeds
Top 10% income shares in European regions, 1980-2017
20
22.5
25
27.5
30
32.5
35To
p 10
% in
com
e sh
are
(%)
1980 1985 1990 1995 2000 2005 2010 2015
Northern Europe Western EuropeSouthern Europe Eastern Europe
31
Inequalities between European
citizens
WI D.WO R L D
Inequality has risen between European citizens since 1980...
Top 10% vs. Bottom 50% income shares, 1980-2017
15
20
25
30
35
40Sh
are
of n
atio
nal i
ncom
e (%
)
1980 1985 1990 1995 2000 2005 2010 2015
Top 10%Bottom 50%
32
WI D.WO R L D
... especially at the top of the distribution
Total income growth by percentile, 1980-2017
Bottom 50% captured15% of growth
Top 1% captured17% of growth
020406080
100120140160180200220240260
Tota
l inc
ome
grow
th (%
)
10 20 30 40 50 60 70 80 90 99 99.9 99.99 99.999Income group (percentile)
33
WI D.WO R L D
European inequalities remain geographically concentrated...
Geography of European inequality, 2017
Eastern Europe
Southern EuropeOther Western
UK
France
Germany
Scandinavia
0
10
20
30
40
50
60
70
80
90
100
Popu
latio
n sh
are
(%)
0 10 20 30 40 50 60 70 80 90 99 99.9Income group (percentile)
34
WI D.WO R L D
... but within-country inequality remains much more important
Between-country vs. within-country inequality in Europe
10
15
20
25
30
35
40Sh
are
of n
atio
nal i
ncom
e (%
)
1980 1985 1990 1995 2000 2005 2010 2015
Top 10% income share... assuming perfect equality between countries... assuming perfect equality within countries
35
WI D.WO R L D
... but within-country inequality remains much more important
Theil decomposition of European income inequality, 1980-2017
Inequality within countries
Inequality between countries
0
.1
.2
.3
.4
.5
.6
Thei
l ind
ex
1980 1985 1990 1995 2000 2005 2010 2015
36
Redistribution in Europe
WI D.WO R L D
Taxes and transfers reduce inequality less in the East where pretax poverty is higher
Pre-tax vs. post-tax ratio of top 10% to bottom 50% average incomes
- 29%- 15%
- 23%
- 23%
0
1
2
3
4
5
6
7
8
Rat
io o
f top
10
% to
bot
tom
50
% a
vera
ge in
com
e
Western Europe Eastern Europe Southern Europe Northern Europe
Pre-tax income Post-tax income
37
WI D.WO R L D
The role of redistribution remains limited
2017 (before taxes)Top 10% make 8x more
than bottom 50%
2017 (after taxes)Top 10% make 6x more
than bottom 50%
3
4
5
6
7
8
9
10R
atio
of t
op 1
0% to
bot
tom
50%
ave
rage
inco
me
1980 1985 1990 1995 2000 2005 2010 2015
Pre-tax inequalityPost-tax inequality
38
Is Europe more unequal than the
US?
WI D.WO R L D
Spatial inequalities remain stronger in Europe than in the US...
US vs. Europe ratio of top 10% to bottom 50% average state / country
incomes, 1929-2017
Europe (2017)Top 10% countries earn 2.8x
more than bottom 50%
United States (2017)Top 10% states earn 1.5x
more than bottom 50%
.5
1
1.5
2
2.5
3
3.5
4
Rat
io o
f top
10%
to b
otto
m 5
0% c
ount
ries/
stat
es a
vera
ge
1925 1935 1945 1955 1965 1975 1985 1995 2005 2015
39
WI D.WO R L D
... yet overall inequalities are much higher and have increased
much more in the US
US vs. Europe Bottom 50% and Top 1% income shares
5
7.5
10
12.5
15
17.5
20
22.5
25
Shar
e of
nat
iona
l inc
ome
(%)
1980 1985 1990 1995 2000 2005 2010 2015
United States
5
7.5
10
12.5
15
17.5
20
22.5
25
1980 1985 1990 1995 2000 2005 2010 2015
Europe
Bottom 50% Top 1%
40
WI D.WO R L D
... yet overall inequalities are much higher and have increased
much more in the US
0
100
200
300
400
500
600
700
Tota
l inc
ome
grow
th (%
)
10 20 30 40 50 60 70 80 90 99 99.9 99.99 99.999Income group (percentile)
United States Europe
41
WI D.WO R L D
Poor Europeans have benefited much more from growth
US vs. Europe Bottom 50% income growth, 1980-2017
US Bottom 50% growth: +3%
Europe Bottom 50% growth: +40%
.7
.8
.9
1
1.1
1.2
1.3
1.4
1.5
1.6Bo
ttom
50%
ave
rage
inco
me
(rela
tive
to 1
980)
1980 1985 1990 1995 2000 2005 2010 2015
United States Europe
42
WI D.WO R L D
These differences have little to do with redistribution
Europe vs. US ratios of top 10% to bottom 50% average incomes
(a) Pre-tax income
4
6
8
10
12
14
16
18
20
Rat
io o
f top
10%
to b
otto
m 5
0% p
re-ta
x in
com
es
1980 1985 1990 1995 2000 2005 2010 2015
Europe (pre-tax inequality) US (pre-tax inequality)
(b) Post-tax income
4
6
8
10
12
14
Rat
io o
f top
10%
to b
otto
m 5
0% p
ost-t
ax in
com
es
1980 1985 1990 1995 2000 2005 2010 2015
Europe (post-tax inequality) US (post-tax inequality)
43
WI D.WO R L D
Conclusion
• A new public database to study income inequality in Europe fully
consistent with national accounts.
• Inequality increased in nearly all EU countries since 1980, but at
different speeds (North & East vs. South & West).
• Despite rising income disparities, EU still much more equal than the
US, before and after taxes.
• Inequality in Europe is much more about within-country than about
between country inequality.
• Yet EU policies essentially about between-country convergence (with
relatively small impact).
44
WI D.WO R L D
Compared to existing databases
Eurostat(weighted average)DINA
(weighted average)
DINA(aggregated distribution)
2
3
4
5
6
7
8
9
10R
atio
of t
op 2
0% to
bot
tom
20%
sha
re
1980 1985 1990 1995 2000 2005 2010 2015
45
WI D.WO R L D
⇒ https://wid.world/
45