1) The document analyzes inequality in income, consumption, and wealth in the United States using data from the Survey of Consumer Finances (SCF).
2) It finds that income, consumption, and wealth inequality have all increased since the 1980s, with wealth inequality being the highest. The top 1% are missing from traditional surveys.
3) By imputing consumption data from the Consumer Expenditure Survey to the SCF, it connects all three measures (income, consumption, wealth) for the same households. This shows interactions between the distributions and how being at an endpoint in one distribution correlates to placement in the others.
1. Research | Training | Policy | Practice
“Inequality in 3-D:
Income, Consumption, and
Wealth”
Tim Smeeding
University ofWisconsin
For the HLEG workshop on inequalities,
Berlin, September, 2015
(with Jonathan Fisher, David Johnson and JeffreyThompson)
--thx to Russell Sage Foundation for their support -
2. Overview and Objectives
Y, C and W for same persons/households:
• Why did we do what we did ?
• How did we do it ? – just an outline , no time for details here
• What did we find ?
• Who cares– quick application to intergenerational mobility ?
• Take-aways and bottom lines
• What do we do next ?
Key objectives:
• get the ‘top 1 percent’ into the “regular survey data “ mix (
wealth vs income and consumption)
• pay attention to micro-macro linkages & reduce differences
3. Basic Premise: multiple inequalities in the
USA are rising—how do they interact ?
• Income (Y) inequality is increasing.
• Consumption (C) inequality has risen by 2/3 as much asY
inequality since 1984.
• Wealth ( W or Net Worth) inequality is the highest and has
been increasing as well.
Biggest challenge forY and C : Bring in the top one, two –
three percent which are missing in most popular datasets
Biggest question: How do these interact : W and C ?
W andY ? Y and C ?
Answer: Need to connect the three together to make
sense of flows -- C,Y and, the stock—W, with all three
together in same dataset
4. Why all three?
• Y Flow measure, capacity to consume, preferred
for poverty, inequality work.
• C Flow measure, actual consumption, preferred
measure of well being for theory bound economists.
• W Stock measure, value at a point in time which
provides insurance, direct links to jobs, source for in-
vivos transfers (schooling, housing, start-up
businesses, etc.) and end of life transfers (bequests
and inheritances).
5. Trends in Inequalities (Ginis) forY (income),
(C) consumption and (W)wealth/net worth
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1979 1982 1985 1988 1991 1994 1997 2000 2003 2006 2009 2012
Money Income
After tax and
Transfer Income
(CBO)
Consumption
(PSID –
Attanasio/
Pistaferri)
Consumption
(CE - FJS)
SCF Household
Income
(Wolff)
SCF Household
Wealth
(Wolff)
5
7. Life-cycle changes in mean Y, C, W
0
10
20
30
40
50
60
70
20 25 30 35 40 45 50 55 60 65 70 75 80 85
$1000s
Consumption
Income
0
200
400
600
800
1000
20 25 30 35 40 45 50 55 60 65 70 75 80 85
Net Worth
8. What connects the two flows ?
• Haig (1921 ) and Simon (1938) :
Y = C +/- changeW
• In other words, if Y > C, W rises by saving or loaning
, and ifY< C then W falls by dissaving or borrowing
• Even with this identity W is important as a source of
C (one can consume from W) as well as a source of
Y (Y includes interest, rent , dividends- flow value
of W, as well as capital gains ) and wealth transfer
• Key Point: W itself, the stock, is also important and
a remainder from the flow equation above
9. US Data Series: Range of Choices
Y: CPS (workhorse in many ways—including CBO
base) , but also CEX, SCF, IRS income series
W andY : SCF ( much closer-- SCF has top endW and
Y—no other survey dataset has the top end )
C : CE (usingY/limited W, to impute C to SCF )
• Here for first time:Y,W, from SCF is
matched to C using (Y,C) from CE ,
1989-2013 (same definitions, also same
demography )
10. Starting Point: SCF
• Need a data set with all three measures –
income, wealth and consumption
• Only a few data sets have overall consumption
(CE, PSID)
– And under-reporting in the CE for both income and
consumption
– None of “usual” datasets ( CE, CPS, PSID) have the
top 1 %
• SCF: good income data, good wealth data, good
data on a limited set of expenditure types
• Technical Question : How well can we impute
total spending/consumption into the SCF ?
11. SCF reports more income than CPS, which
reports more income than CE
12. SCF/CE Income Ratio OverTime—
Note growth with rise in top 1 %
0.00
0.50
1.00
1.50
2.00
2.50
p25 p50 p75 p90 p95 p99
1989 1992 1995 1998 2001 2004 2007 2010 2013
13. Our preferred method: predictive
mean matching
• Alternative approaches, depending on inclusion of
food and different covariates –
--Mean of Nearest Neighbor -- 5 imputation methods , -- averaged
• Known and common expenditures in SCF and CEX :
food, rent, owners expenditures, transportation
• Variables: age, race, and educational attainment (5) of head, urban
indicator, housing tenure (3), family-type, and marital status, # under 18, #
over 64, family size, Census Division FE, income, income squared, indicators
for negative income, presence of public transfers, and capital income losses
• All in common across CE, SCF
• Impute both share and level of unknown
expenditures to those already known in the SCF
14. SCF captures more consumption at higher
income percentiles than does the CEX
15. Our C imputation is closer to Comparable PCE
Aggregates. Comparing CE, SCF and PCE trends
16. Bottom Lines on Data Creation, so far
• A great deal ofY flows to the top percentiles in the USA; they
have a low APC, so what happens to the rest of theirY and W ?
• Those with high incomes turn over year to year, but with low
APC’s they build up their wealth, and the top wealth quintiles
build in value with less mobility in or out of top 5 percent of W
Important Question : How do new dataset estimates
match up with BEA/SNA/FF aggregates –W,Y and C ?
Answer : for C alone, not so well and also missing top
end of Y in CE, (and therefore also mainly missing
income from capital) . But SCF W is pretty good actually
(Sabelhaus, et al, 2015) and our imputations improve on
aggregate C relative to that reported in the CE
17. How about interactions?
• Distributions are sticky at the end points,
especially W, but also C andY
• Sticky at bottom – then at top
--top 5 percent extremely advantaged in all ways
-- bottom 20 percent not advantaged in any way
• Elders differ-they have relatively lowY,
but relatively high C and W (not shown
here)
18. Distribution byY and C & Y and W
quintiles (transitions for 2013*)
*Axes show percentage of people in each quintile pair. Hence, about
9 % of people are in bottom income and consumption quintile, and
same in bottom wealth and income quintile. Close to 12 percent in
top wealth and income quintiles and top income and consumption
quintiles
21. Increases in the share for top ventile
( top 5%) for all measures
22. Over 60 percent of the top 5% in W are also in top
5% of Y or C; Over 30% in top 5% for all 3 measures
22
23. Summary so Far
• Bottom quintile/ventile bunching, but top
bunched even more so, with rising inequality
• Y and W done well ( Z&S; P&S)
• Consumption in top wealth and income
ventiles (5%) is very high but needs some
work, ( for 2013 especially)
• Question– does CEX capture all wealth
transfer in vivos ? ( We think not –more
below )
24. So What ?
• These slides suggest that :
-being in the bottom or top of theW distribution
is really important for determining your position
inY&C distribution.
--we would see an even more dramatic pictures if
we restricted to working age adults only
-- or by adult education level where the W
differences are much bigger
• Bottom Line—Wealth really matters
25. QUICKLY Apply to Equality of Opportunity
& Intergenerational Mobility (IGM)
• Usual IGM measures are applied to income
only—but consumption, wealth, family
structure, stability and parenting matter
too
• Fisher/Johnson, 2006, using the PSID for
income and consumption and find similar
IGM overall elasticities
• Pfeffer shows that wealth matters a lot for
mobility, especially wealth transfer in-vivos
26. What else besides the usual index ,Y ?
W and C also matter
• Most children are concentrated in low C and
low W quintiles, even more so than in terms of
Y quintiles
• Means fewer advantaged ( highW and C as
well as highY) kids—more disadvantaged kids,
e.g. see C on enrichment expenditures below
and W for ‘family safety net’ following
• Elders at other end with high C and W, but with
lowerY, though picture is much the same
without them
28. How does W, C andY affect mobility
for vulnerable groups ?
• Whose kids are least likely to be upwardly
mobile?
--blacks
--kids from SPFs
--kids whose parents who are high school drop outs
• Call these theVulnerable
• See howY, C and W lead to triple disadvantages,
more than 70 percent of lone parents and
undereducated adults kids are in bottom two
quintiles of Y, C or W in any year
29. The Demography of Inequality in 3D
TheVulnerable*
0.0
0.1
0.2
0.3
0.4
0.5
Income Cons Wealth Income Cons Wealth Income Cons Wealth
Blacks Children in Single Parent
Families
High School
Dropouts
*Fraction of each type of person in each overall quintile forY, C, W
30. The other/top end : family “safety net” and
inter-vivos “strategic transfers”
• Top 5-10% parents/grandparents have their own
built-in private safety nets for kids:
1. They live in good neighborhood (safety/better schools )
2. Their kids graduate college and often post grad degrees,
with no worry of debt
3. Their kids intern, often for free, in expensive high pay cities
to overcome high end spatial job mismatch
4. They help buy a first home or flat at favorable interest rates
by co-signing the offspring’s mortgage
5. Plus direct lifetime jobs for kids (US,DK and CN evidence on
nepotism for the top 5 percent )
31. Don’t worry about the elderly ?
• Life cycle patterns are clearly visible, but muted
• Rich elders do not consume most wealth when
old, rather they make large inter-vivos transfers
(HRS evidence) and leave inheritances/bequests
which mainly benefit upper income 60
something's (Yellen, 2014)
• Population aging is not as important as one
would think (basicY,C,W pictures look the
same once elders are out, & some, e.g.
education effects onW look stronger)
32. SummaryTake-Aways & Bottom Lines
• Income, Consumption and Wealth are measured
differently on different surveys – need them all
to make sense of economic inequality
• Using SCF corrects some of the under-reporting
in CE and most importantly it captures the top
end of the W andY distributions
• Changes in inequality from 1989 to 2013 are
different than earlier changes with much of the
increase driven by the top of the distribution
32
33. MoreTake-Aways & Bottom Lines
• Joint distributions show higher correlation/
bunching at top and bottom, yet the correlation
is not perfect, so many households with highY,
but not high W (many “off diagonal” in the
middle 60 percent of the distribution )
• In most cases, the measure matters—
-- Especially for the elderly ( High C, W, lowY)
-- Measures are re-enforcing for children, blacks,
SPFs ,lower educated
--Higher educated are much wealthier
34. Next Steps
• Improve estimates for 2013( C andY in SCF)
-- try to get aggregate C closer to PCE in the
next steps (as SCF has good estimates of W
andY relative to SNA aggregates)
• Exploit the longitudinal nature of the PSID
to examine immobility -- being in the top
quintile ofY, C, and especially W, in one
year is highly predictive of being there in a
future year, even if PSID misses top 1-2
percent
35. THE END--
• Thanks for your patience
• Email questions / comments to
David.Johnson@bea.gov
smeeding@lafollette.wisc.edu
• A few bonus slides on micro surveys vs
PCE,NAS aggregates follow
36. Comparing CE, SCF and PCE trends, no health care --
our imputation is close to comparable PCE aggregates
37. Under-reporting is due to missing people
and income at the top of the distribution
37
Sabelhaus, et al (2015)
38. Under-reporting in CE and CPS
(ratio survey to national aggregates)
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
CPS/PI - Katz (Fixler/Johnson)
CE/PCE - McCully