Income Inequality and Income Risk:
Old Myths vs. New Facts1
Fatih Guvenen
University of Minnesota and NBER
JDP Lecture Series on “Dilemmas in Inequality”
at Princeton University, Fall 2013
(Updated: October 2014)
1
This lecture summarizes research conducted jointly with Serdar Ozkan, Fatih Karahan, Greg Kaplan, and Jae Song.
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
1 / 41
Not everything that counts can be counted...
... and not everything that can be counted counts.
Sign on Einstein’s office wall at Princeton
Motivation
Nature of income inequality/risk: critical for many questions in
social sciences.
Survey-based US panel datasets have important limitations:
I
small sample size
I
large measurement (survey-response) error
I
non-random attrition
I
top-coding, etc.
=) myths about income inequality and income risk.
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
3 / 41
Data: SSA Master Earnings File
Population sample: Universe of all individuals with a U.S. Social
Security number
Currently covers 35 years: 1978 to 2012 (soon to be updated with
2013 data)
Basic demographic info: sex, age, race, place of birth, etc.
Earnings data:
I
I
Salary and wage earnings from W-2 form, Box 1
F
No topcoding
F
Unique employer identifier (EIN) for each job held in a given year.
F
4–5 digit SIC codes for each employer
Self-employment earnings from IRS tax forms (Schedule SE)
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
4 / 41
Our Sample
10% Representative panel of US males from 1978 to 2012
Salary and wage workers (from W-2 forms)
I
exclude self-employed (data top coded before 1994)
I
Focus on workers aged 25–60
I
Key Advantages:
F
Very large sample size (200+ million individual-year observations)
F
No survey response error (W-2 forms sent from employer directly to
SSA)
F
No sample attrition
F
No top-coding
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
5 / 41
Six Myths
Six Myths
1
Myth #1: Income risk has been trending up in the past 40 years.
2
Myths #2 and #3: Income risk over the business cycle is...
mostly about countercyclical variance of shocks
3
Myth #4: Top 1% are largely immune to business cycle risk
4
Myths #5 and #6: Income over the life cycle can be modeled as:
(A polynomial in age... + ...a random walk process...) with
Gaussian shocks
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
7 / 41
Risk and Inequality
Over Time
Trends in Income Risk
Myth #1:
The volatility of income shocks...
has increased significantly over the past 40 years.
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
9 / 41
Upward Trend in Income Risk: Background
This conclusion has been reached by virtually all papers that use
PSID data.
Moffitt and Gottschalk (1995) documented it first in a now-famous
paper, and it has been confirmed by a large subsequent literature.
The fact that this finding is robust across various PSID studies
suggests that it is more about the data set rather than the
methodology.
Here is how the basic result looks like (from Moffitt-Gottschalk’s
updated paper: Moffitt and Gottschalk (2012))
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
10 / 41
Myth #1: Upward Trend in Income Risk
Figure 10: Permanent, Transitory, and Total Variances for those 30-39 with Education Greater
than 12
0.45
0.4
permanent
transitory
total
0.35
0.3
0.25
0.2
0.15
0.1
0.05
2004
2003
2002
2001
2000
1999
1998
1997
1996
1995
1994
1993
1992
1991
1990
1989
1988
1987
1986
1985
1984
1983
1982
1981
1980
1979
1978
1977
1976
1975
1974
1973
1972
1971
1970
0
year
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
11 / 41
Fact #1: No Upward Trend in Volatility
When researchers turned to administrative datasets, such as the
one described above, the opposite conclusion emerges robustly
See, e.g., Congressional Budget Office (2007); Sabelhaus and
Song (2010); Guvenen et al. (2014b)
In fact, looking by age, gender, and industry groups, we see the
same pattern of flat or declining volatility in all groups (with the
exception of agriculture, which is very small).
Here is the basic figure from Guvenen et al. (2014b):
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
12 / 41
Fact #1: No Upward Trend in Volatility
0.85
0.8
Standard Deviation
0.75
0.7
0.65
0.6
0.55
0.5
y˜t − y˜t−1
0.45
0.4
y˜t − y˜t−5
1980
1985
1990
1995
Year
2000
2005
2010
Source: Guvenen, Ozkan, Song (JPE, 2014)
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
13 / 41
Risk and Inequality Over the
Business Cycle
Business Cycle Variation in Shocks
Myth #2:
The variance of idiosyncratic income shocks
rises substantially during recessions.
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
15 / 41
Myth #2: Countercyclical Shock Variances
Dens ity
Expansion
Recession
−0.8
−0.6
−0.4
−0.2
0
y
−y
t+k
Fatih Guvenen (Myths vs. Facts)
0.2
0.4
0.6
0.8
t
Myths vs. Facts
16 / 41
Countercyclical Variance
Constantinides and Duffie (1996): countercyclical variance can
generate interesting and plausible asset pricing behavior.
Existing indirect parametric estimates find a tripling of the variance
of persistent innovations during recessions (e.g., Storesletten et al
(2004)).
Our direct and non-parametric estimates show no change in
variance over the cycle. See the next figure.
The following figures on Myths 2 to 4 are from Guvenen et al.
(2014b).
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
17 / 41
Fact #2: No Change in Variance
Disp ersion in Recession/Disp ersion in Expansion
2
1.8
1.6
Storesletten et al (2004)’s
benchmark estimate: 1.75
Std. dev. ratio
L90−10 ratio
1.4
1.2
1
0.8
0
10
20
30
40
50
60
70
80
90
100
Percentiles of 5-Year Average Income Distribution (Y t −1 )
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
18 / 41
Dens ity
Fact #2: Countercyclical Left-Skewness
Expansion
Recession
−0.6
−0.5
−0.4
Fatih Guvenen (Myths vs. Facts)
−0.3
−0.2
−0.1
0.0
yt+k− yt
Myths vs. Facts
0.1
0.2
0.3
0.4
0.5
19 / 41
Kelley’s Skewness Measure of y t +k − y t , k = 1, 5
Fact #2: Countercyclical Skewness
Expansion
Recession
0.1
0
−0.1
−0.2
−0.3
−0.4
0
10
20
30
40
50
60
70
80
90
100
Percentiles of 5-Year Average Income Distribution (Y t −1 )
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
20 / 41
Is Business Cycle Risk Predictable?
Myth #3:
Business cycle risk is mostly ex-post risk
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
21 / 41
Fact #3: Business Cycle Risk is Predictable
Mean Log Income Change During Recession
0.1
0.05
0
−0.05
−0.1
−0.15
−0.2
−0.25
1979-83
1990-92
2000-02
2007-10
−0.3
0
10
20
30
40
50
60
70
80
90
100
Percentiles of 5-Year Average Income Distribution (Y t −1 )
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
22 / 41
Business Cycle Risk for Top 1%
Myth #4:
The top 1% are largely immune
to the pain of business cycles.
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
23 / 41
Fact #4: The “Suffering” of the Top 1%
Mean Log Income Change During Recession
0.1
0.05
0
−0.05
−0.1
−0.15
−0.2
−0.25
1979-83
1990-92
2000-02
2007-10
−0.3
−0.35
0
10
20
30
40
50
60
70
80
90
100
Percentiles of 5-Year Average Income Distribution (Y t −1 )
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
24 / 41
Fact #4: 1-Year Income Growth, Top 1%
Log 1-Year Change in Mean Income Level
0.2
0.1
0
−0.1
−0.2
−0.3
−0.4
Top 0.1%
Top 1%
P50
1980
Fatih Guvenen (Myths vs. Facts)
1985
1990
1995
Year
Myths vs. Facts
2000
2005
2010
25 / 41
Fact #4: 5-Year Income Growth, Top 0.1%
Log 5-Year Change in Mean Income Level
0.5
0.4
0.3
0.2
0.1
0
−0.1
−0.2
−0.3
−0.4
−0.5
Top 0.1%
1980
Fatih Guvenen (Myths vs. Facts)
1985
1990
Year
Myths vs. Facts
1995
2000
2005
26 / 41
Risk and Inequality Over the
Life Cycle
Lifecycle Profile of Income
Myth #5:
A reasonable specification of income over the life cycle consists of:
1
A common polynomial in age... +
2
...a random walk process...
3
with Gaussian shocks
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
28 / 41
Myth #5: Lifecycle Profile of Income
10.6
Log Average Income
10.4
127%
rise
10.2
10
9.8
9.6
25
30
35
40
45
50
55
60
Age
Source for the rest of this section: Guvenen et al. (2014a)
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
29 / 41
Fact #5: Lifecycle Profiles of Income
3
Top 1%:
15−fold increase!
2.5
log(Y 55) – log(Y 25)
2
1.5
Income Growth from Pooled Regression
1
Random Walk Model
0.5
0
HIP (Guvenen (2009))
−0.5
−1
0
10
Fatih Guvenen (Myths vs. Facts)
20
30
40
50
60
70
80
Percentiles of Lifetime Income Distribution
Myths vs. Facts
90
100
30 / 41
Distribution of Income Shocks
Myth #6:
It is OK to model income growth...
...as a lognormal distribution
=) it is OK to assume...
...zero skewness and no excess kurtosis
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
31 / 41
Kurtosis
Myth #6: Lognormal Histogram of yt+1
yt
5
N(0,0.43 2 )
4.5
4
3.5
Density
3
2.5
2
1.5
1
0.5
0
−3
−2
Fatih Guvenen (Myths vs. Facts)
−1
0
y t +1 − y t
Myths vs. Facts
1
2
3
33 / 41
Fact #6: Excess Kurtosis
5
N(0,0.43 2 )
4.5
US Data, Ages 35-54, P90 of Y
4
3.5
Density
3
Kurtosis: 28.5
2.5
2
1.5
1
Kurtosis: 3.0
0.5
0
−3
−2
Fatih Guvenen (Myths vs. Facts)
−1
0
y t +1 − y t
Myths vs. Facts
1
2
3
34 / 41
Fact #6: Excess Kurtosis
Prob(|yt+1
x#
0.05
0.10
0.20
0.50
1.00
Fatih Guvenen (Myths vs. Facts)
yt | < x)
Data
N(0, 0.432 )
0.39
0.57
0.70
0.80
0.93
0.08
0.16
0.30
0.59
0.94
Myths vs. Facts
35 / 41
Fact #6: Excess Kurtosis
32
Kurtosis of (y t+1 − y t )
28
24
20
16
12
Ages
Ages
Ages
Ages
8
4
0
10
Fatih Guvenen (Myths vs. Facts)
25-29
30-34
35-39
40-54
20
30
40
50
60
70
80
90
Percentiles of Past 5-Year Average Income Distribution
Myths vs. Facts
100
36 / 41
Skewness
Fact #6: Skewness of yt+1
yt
0
Age=25-34
Age=35-44
Age=45-49
Age=50-54
Skewness of (y t+1 − y t )
−0.5
−1
−1.5
−2
−2.5
−3
0
10
Fatih Guvenen (Myths vs. Facts)
20
30
40
50
60
70
80
90
Percentiles of Past 5-Year Average Income Distribution
Myths vs. Facts
100
38 / 41
Double Pareto Tails of Earnings Growth
2
US Data
Normal (0.0.482 )
Log Density
0
-2
-4
-6
-8
-3
Fatih Guvenen (Myths vs. Facts)
-2
-1 y 0− y
t+1
t
Myths vs. Facts
1
2
3
39 / 41
Conclusions
For too long, we have played the “blind men and the elephant.”
But there is hope: some fantastic datasets are becoming more
accessible.
Challenges: Data on consumption.. still very limited.
We hope these new (or revised) facts will feed back into theory
and policy work.
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
40 / 41
References
Congressional Budget Office, “Trends in Earnings Variability over the
Past 20 Years,” Technical Report, Congressional Budget Office 2007.
Guvenen, Fatih, Fatih Karahan, Serdar Ozkan, and Jae Song,
“What Do Data on Millions of U.S. Workers Say About Labor Income
Risk?,” Working Paper, University of Minnesota 2014.
, Serdar Ozkan, and Jae Song, “The Nature of Countercyclical
Income Risk,” Journal of Political Economy, 2014, 122 (3), 621–660.
Moffitt, Robert A. and Peter Gottschalk, “Trends in the Variances of
Permanent and Transitory Earnings in the U.S. and Their Relation to
Earnings Mobility,” Boston College Working Papers in Economics
444, Boston College July 1995.
Moffitt, Robert and Peter Gottschalk, “Trends in the Transitory
Variance of Male Earnings: Methods and Evidence,” Winter 2012, 47
(2), 204–236.
Sabelhaus, John and Jae Song, “The Great Moderation in Micro
Labor Earnings,” Journal of Monetary Economics, 2010, 57,
391–403.
Fatih Guvenen (Myths vs. Facts)
Myths vs. Facts
41 / 41