Distributional National Accounts:
Methods and Estimates for the
United States since 1913
Thomas Piketty (PSE)
Emmanuel Saez (UC Berkeley)
Gabriel Zucman (LSE)
April 2015
There is a large disconnect today between
the study of inequality and macro
Macro: use national accounts, with no info on distribution
Inequality: use survey & tax data, inconsistent with macro totals
This gap makes it hard to know how growth is distributed:
How does growth of bottom 90% compare to total growth?
What fraction of growth accrues to top 1%?
⇓
We need answers to these questions to better understand
interplay between growth, inequality, and gov. intervention
Sources of the disconnect between
inequality and macro
Large gap between surveys and macro totals
Conceptual differences: household income vs. national income
Practical pbs: non-response & under-reporting at the top
Some progress using tax data, but insufficient
Atkinson, Alvaredo, Piketty, Saez and 30+ used tax data to
construct top shares in 28+ countries, 1870-2012 → WTID
But: tax data miss tax-exempt income and tax evasion
Ex: in US only capture 60% of US national income
Silent on post-tax and transfer income
Silent on distribution within bottom 90%
DINAs are a new tool to connect
inequality and macro
While today’s national accounts are spreadsheets with aggregates
only, tomorrow they could be micro databases where:
Each observation is a synthetic individual
Distributions of income, wealth, taxes, transfers... are consistent
with what survey/tax data show
Totals match macro aggregates
Creating DINAs involves:
Defining a clear common conceptual framework
Incorporating results from economic theory
Developing statistical techniques to optimally use information
from available micro sources
Following up on King (1696) and
Kuznets, with improved data and method
Global DINAs
WTID will be replaced by a World Wealth and Income
Database (W2ID)
Ultimate goal is to enable governments to publish their own
DINAs all over the world
Gov. agencies able to deploy large teams to refine estimates
Same process as for today’s NA
1930s-40s: prototypes developed by Kuznets, Kendrick, Meade,
Stone → improved by agencies 1950s-2015
This paper
Goal: construct micro database of income, taxes and transfers
consistent with NIPA totals in the United States
By combining tax, survey, and national accounts data in a
consistent manner
Pre-tax & transfers vs. post-tax & transfers (same aggregate)
⇒ First estimates of income inequality covering 100% of
national income
⇒ First decompositions of growth by groups consistent with
growth used in macro
Roadmap
1. The limits of current estimates of US income inequality
2. US DINA methodology: general principles
3. Methodology to capture 100% of pre-tax capital and labor
income
4. The distribution of pre-tax national income since 1913
5. The interplay between income and wealth inequality
The limits of current estimates of
US income inequality
Goal: allocate total US national income
Y = YK + YL to the adult pop.
Composition of national income in the United States since 1913
100%
80%
70%
Labor income
60%
50%
40%
30%
20%
2013
2008
2003
1998
1993
1988
1983
1978
1973
1968
1963
1958
1953
1943
1938
1933
1928
1923
1918
0%
1948
Capital income
10%
1913
% of factor-price national income
90%
Most capital income is missed by tax data
From tax-reported to total capital income
35%
Retained earnings
% of factor-price national income
30%
25%
20%
15%
10%
5%
Non-filers
&
unreported sole Corporate income tax
prop. profits
Income paid to pensions &
insurance
Imputed rents
Didivends, interest, rents & profits reported on tax returns
0%
1920 1925 1930 1935 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
Most capital income is missed by tax data
From tax-reported to total capital income
35%
Retained earnings
% of factor-price national income
30%
25%
20%
15%
10%
5%
2/3 missed
Non-filers
&
unreported sole Corporate income tax
prop. profits
by tax data
Income paid to pensions &
insurance
Imputed rents
Didivends, interest, rents & profits reported on tax returns
0%
1920 1925 1930 1935 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
A growing fraction of labor income is
missed by tax data
From taxable to total employee compensation
% of total NIPA compensation of employees
100%
95%
Other
Employer payroll taxes
90%
85%
Pension contributions
80%
Health benefits
75%
70%
65%
60%
Reported taxable wages
55%
50%
1953 1958 1963 1968 1973 1978 1983 1988 1993 1998 2003 2008 2013
A growing fraction of labor income is
missed by tax data
From taxable to total employee compensation
% of total NIPA compensation of employees
100%
95%
Other
Employer payroll taxes
90%
85%
1/4 missedPension
by tax
data
contributions
80%
Health benefits
75%
70%
65%
60%
Reported taxable wages
55%
50%
1953 1958 1963 1968 1973 1978 1983 1988 1993 1998 2003 2008 2013
There is much less growth in tax data
than in the national accounts
Average real income: national accounts vs. tax data (1946 = 100)
National income per adult
300
260
220
180
IRS market income per family
(Piketty-Saez, CPI)
140
Source: Appendix Tables XX.
2010
2006
2002
1998
1994
1990
1986
1982
1978
1974
1970
1966
1962
1958
1954
1950
1946
100
US DINA methodology:
General principles
Income concept and unit of observation
Income concept: US (pre-tax) national income:
Only adjustment: remove investment income of nonprofits (≈
2% of national income)
Unit of observation:
Adult [individual over 20]
Preferable to family (fewer marriage → less growth) and
individual (higher fertility → less growth)
Deflator:
National income deflator
Preferable to CPI (overstates inf.) and PCE deflator (too narrow)
Reconciling growth in tax data and
national accounts
Average real income: national accounts vs. tax data (1946 = 100)
National income per adult
300
IRS market income per adult
(national income deflator)
260
IRS market income per family
(national income deflator)
220
180
IRS market income per family
(Piketty-Saez, CPI)
140
Source: Appendix Tables XX.
2010
2006
2002
1998
1994
1990
1986
1982
1978
1974
1970
1966
1962
1958
1954
1950
1946
100
How we allocate taxes
We follow standard tax incidence results:
Labor taxes fall on labor; capital taxes on capital (and corporate
tax on all capital assets: Harberger 1962)
Reasonable if Y = F (K , L) has elasticity of substitution σ >>
labor supply elasticity eL and capital supply elasticity eK
Cross-country and time-series evolution of α = rK /Y and
β = K /Y broadly consistent with view that σ > 1 and eL and eK
relatively small in the long run σ eK
But this is uncertain → will revisit if needed
Methodology:
Distributing pre-tax capital and
labor income
How we construct micro-measures of
capital income matching macro totals
1. Construct micro-measures of wealth
Follow Saez and Zucman 2014: category by category, pragmatic
approach
Delivers micro-measures of family wealth matching macro totals
Split capital 50/50 among spouses
2. Derive micro-measures of pre-tax capital income
Compute aggregate pre-tax rates of return for each asset class
that reconcile NIPA income flows with Flow of Funds wealth
Apply these rates of return to individual assets
How we construct distributional estimates
of national labor income
1. Start from reported wages; split income of spouses using W2 forms
2. Employer payroll taxes: apply schedule, e.g., in 2015:
Medicare: 1.45%
Social Security: 7.2% capped at $118,500
Unemployment insurance: vary by state
3. Pension and health insurance contributions:
Available on W2 forms since 1999 for pensions, since 2012 for
health
Before, assume same distribution
The distribution of pre-tax income
since 1913
National income is more concentrated
than tax income
Top 10% pre-tax income shares
55%
45%
National income per adult
(DINA)
40%
35%
30%
2012
2007
2002
1997
1992
1987
1982
1977
1972
1967
1962
1957
1952
1947
1942
1937
1932
1927
1922
25%
IRS family market income
(Piketty-Saez)
1917
% of total income
50%
This figure displays the share of total pre-tax national income earned by top 10% adult income earners and the share of total IRS
market income earned by top 10% family tax units. Source: Appendix Tables XX.
Less true for top 1%
Share of national income earned by top 1% adult income earners
25%
National income per adult
(DINA)
% of total income
20%
15%
10%
This figure displays the share of total pre-tax national income earned by top 1% adult income earners and the share of total IRS
market income earned by top 1% tax units. Source: Appendix Tables XX.
2013
2008
2003
1998
1993
1988
1983
1978
1973
1968
1963
1958
1953
1948
1943
1938
1928
1923
1918
0%
1913
5%
1933
IRS family market income
(Piketty-Saez)
Until late 1990s, rise in the top 1% driven
by an upsurge in top wages
Labor income of top 1% adult income earners
14%
10%
8%
6%
Compensation of employees
4%
2%
2008
2003
1998
1993
1988
1983
1978
1973
1968
1963
1958
1953
1948
1943
1938
1933
1928
1923
1918
Labor income in noncorporate businesses
0%
1913
% of national income
12%
... But since late 1990s, top 1% rises
because of capital income
Capital income of top 1% adult income earners
14%
12%
Profits & interest paid to pensions
8%
6%
Corporate profits
4%
Net interest
2%
2008
2003
1998
1993
1988
1983
1978
1973
1968
1963
1958
1953
1948
1943
1938
1933
1928
1923
1918
0%
Noncorporate profits
Housing rents
1913
% of national income
10%
DINAs make it possible to compute
growth rates consistent with macro totals
Real average national income:
Full adult population vs. bottom 90%
All adults
60,000
1.4%
50,000
40,000
2.0%
30,000
20,000
Bottom 90% adults
0.7%
2.0%
10,000
2014
2010
2006
2002
1998
1994
1990
1986
1982
1978
1974
1970
1966
1962
1958
1954
1950
0
1946
Average income in constant 2012 dollars
70,000
Real values are obtained by using the national income deflator and expressed in 2012 dollars. Source: Appendix Tables
XX.
The top 10% has grown three times
faster than the bottom 90% since 1980
300,000
70,000
60,000
250,000
50,000
2.3%
200,000
40,000
1.9%
150,000
30,000
Top 10%
(left axis)
0.7%
100,000
Bottom 90%
(right axis)
2.0%
50,000
20,000
10,000
2014
2010
2006
2002
1998
1994
1990
1986
1982
1978
1974
1970
1966
1962
1958
1954
0
1950
0
Bottom 90% real average national income
Real average national income of bottom 90% and top 10%
adults
1946
Top 10% real average national income
350,000
Real values are obtained by using the national income deflator and expressed in 2012 dollars. Source: Appendix Tables
XX.
Top 1% vs. bottom 90%: from a factor
of 20 to a factor of 40
Real average national income of bottom 90% and top 1% adults
Top 1%
(left axis)
1,200,000
1,000,000
60,000
50,000
3.5%
800,000
40,000
600,000
30,000
400,000
0.7% Bottom 90%
(right axis)
1.4%
200,000
20,000
10,000
2.0%
2014
2010
2006
2002
1998
1994
1990
1986
1982
1978
1974
1970
1966
1962
1958
1954
0
1950
0
Real values are obtained by using the national income deflator and expressed in 2012 dollars. Source: Appendix Tables XX.
Bottom 90% real average national income
70,000
1946
Top 1% real average national income
1,400,000
The distribution of economic growth in
the US since 1946
All
Bottom
90%
Bottom
99%
Top
10%
Top
1%
Top
0.1%
Top
0.01%
Average yearly growth rates of real national income per adult
1.7%
1.4%
1.6%
2.1%
2.4%
3.0%
3.8%
1946-1980
1946-2012
2.0%
2.0%
2.1%
1.9%
1.4%
1.4%
1.8%
1980-2012
1.4%
0.7%
1.0%
2.3%
3.5%
4.7%
6.1%
2009-2012
2.0%
0.1%
0.8%
3.9%
7.0%
9.1%
9.5%
Fraction of national income growth accruing to each groups
1946-2012
100%
43%
76%
57%
24%
12%
6%
1946-1980
100%
62%
92%
38%
8%
3%
1%
1980-2012
100%
27%
62%
73%
38%
20%
10%
2009-2012
100%
1%
32%
99%
68%
38%
18%
The interplay between income and
wealth inequality
Synthetic saving rates by fractile
How we compute saving rates by fractile:
1. Start from individual i wealth accumulation equation:
i
Wt+1
= (1 + qti ) · (Wti + sti · Yti )
where Wti is wealth, Yti is income, sti is net savings rate, 1 + qti is
pure price effect on assets in year t
2. Consider similar equation for fractile p (e.g., top 1%):
p
Wt+1
= (1 + qtp ) · (Wtp + stp · Ytp )
3. Combining information on composition of p’s wealth and
economy-wide price effects by asset class, we can compute 1 + qtp
4. So we can infer from (2) the synthetic saving rate stp of fractile
p, i.e., the flow of saving that reconciles p’s change in wealth
btw t and t + 1 given change in price of assets held by p
Saving rates rise with wealth
Saving rates by wealth class (decennial averages)
45%
Top 1%
35%
25%
Top 10 to 1%
15%
5%
Bottom 90%
2010-12
2000-09
1990-99
1980-89
1970-79
1960-69
1950-59
1940-49
1930-39
-15%
1920-29
-5%
1917-19
% of each group's total primary income
55%
The average private (household + corporate) saving rate has been 11.4% over 1913-2013, but the rich save more as a fraction of
their income, except in the 1930s when there was large dis-saving through corporations. Source: Appendix Table B33.
Combination of rising income and saving
rate inequality is fueling wealth inequality
140,000
12,000,000
120,000
Top 1% real average welath
14,000,000
10,000,000
100,000
8,000,000
80,000
6,000,000
Bottom 90%
(right axis) 60,000
4,000,000
40,000
Real values are obtained by using the GDP deflator, 2010 dollars. Source: Appendix Tables B3.
bottom90
2014
2010
2006
2002
1998
1994
1990
20,000
1986
1982
1978
1970
1966
1962
1958
1954
1950
1946
0
1974
Top 1%
(left axis)
2,000,000
0
Bottom 90% real average wealth
Real average wealth of bottom 90% and top 1% families
Capital accumulated out of surging labor
income generates a sizable return
Yield and total return on U.S. private wealth
20%
Pure yield = capital income (including
retained earnings) / wealth
15%
10%
-5%
Total return = pure yield + asset price effect
-10%
back
2013
2008
2003
1998
1993
1988
1983
1978
1973
1968
1963
1958
1953
1948
1943
1938
1933
1928
1923
1918
0%
1913
5%
The labor income share of top wealth
holders has grown but is stabilizing
Share of income earned by top 0.1% wealth-holders
8%
% of total pre-tax income
7%
6%
National income
5%
4%
3%
2%
Labor income
1%
0%
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
This figure shows the share of total pre-tax national income and pre-tax labor income earned by top 0.1% wealth-holders. Labor
income includes employee compensation and the labor component of business income. Source: Appendix Tables B25 and B28.
back
Conclusion
The national accounts of the future
The DINA agenda:
Construct new series on the distribution of wealth, income,
saving... fully consistent with macro aggregates
Key tool: a new micro-dataset of synthetic observations coherent
with distributional data and matching macro totals
Preliminary, exploratory and uncertain, but conceptually the right
way to go
First US DINA results:
Large rise in top national income shares
Rise in income inequality since late 1990s due to capital, not
labor
DINAs for policymakers
We will make our prototype DINAs publicly available on dedicated
websites → everybody able to conduct own distributional analysis
Useful for policy-makers:
Simulate reforms of tax and transfer policies
Monitor financial stability (e.g., distribution of saving rates)
Useful for the public debate:
Big demand for decompositions of growth by social groups
Not possible today
Next steps in the US DINA
Adding all taxes and transfers—monetary and in kind—and
imputations for public goods consumption
Will make it possible to provide measures of:
Consumption inequality
Distribution of post-tax and transfer income
How pre-tax and post-tax distributions differ → will give a
comprehensive view of the redistributive effect of government