Job Flows, Demographics and the Financial Crisis
Eva Sierminska
CEPS/INSTEAD and DIW Berlin
Yelena Takhtamanova
Federal Reserve Bank of San Francisco
January 2010
First draft please do not quote without permission
Abstract
The recession the United States economy entered in December of
2007 is considered to be the most severe downturn the country has experienced since the Great Depression. The unemployment rate reached
as high as 10.1 percent in October 2009 - the highest we have seen
since the 1982 recession. In this paper we decompose the changes in
the unemployment rate by examining worker flows into and out of unemployment during the current recession in the United States with a
special focus on the most vulnerable groups.
1
1
Introduction
In August 2007, the U.S. and global economy were hit by a financial crisis.
Many argued that it is the worst financial crisis in the post-war period, and
some went as far as suggesting it might be the worst in modern history.1
The colossal losses faced by financial institutions led to a credit crunch.
At the same time, the extremely poor performance of housing and stock
market led to an enormous wealth loss by households(over 25 percent of
U.S. households’ net worth was destroyed in the crisis). With weakening
demand, the labor market tumbled, as businesses laid off workers. The U.S.
economy entered the recession December of 2007. Early on, job losses were
low in comparison to previous recessions and the downturn appeared to be
mild (in fact, some questioned if a recession was imminent). As financial
panic intensified in the fall of 2008, massive job losses followed, and it was
clear that not only the country entered a recession, but that this was going
to be a deep one.
As we write this paper, the National Bureau of Economic Analysis
(NBER) has yet to announce the end of the recession. Yet, as early as
September 2009, many analysts and policymakers suggested that the recession might be over. Even if this is so, job losses continue to this day. Overall,
the U.S. economy lost more than 7.2 million jobs since the beginning of the
recession. The unemployment rate reached as high as 10.1 percent. While
the pace of the job losses subsided and the unemployment rate came down
to 10 percent, the multitude of public and private forecasts suggest that it
1
Bernanke 2010 – http://www.federalreserve.gov/newsevents/speech/bernanke20100103a.htm
2
would take years for the U.S. labor market to recover. And, thus, while the
recovery has begun, there is little doubt that it will be slow and painful.
The extreme weakness of the labor market became the focus of attention
of many U.S. policymakers. Policy response was comprehensive and involved
measures aimed at the stabilization of the financial system, improvements
in credit and liquidity and the American Recovery and Reinvestment Act
(ARRA) - an aggressive fiscal expansion. One of the goals of ARRA was to
create and save jobs.
How does this recession compare to the other ones? Were some demographic groups affected more than others and what was the main driving
force behind the rising unemployment? Was it fueled by higher worker
inflows into unemployment or decreasing worker outflows compared to previous recessions? Is ARRA helping the most vulnerable? These are the
questions we attempt to answer with this paper. We look into labor market
experiences of several demographic groups. In particular, we look at gender,
age and race -an important variable in the United States.
The paper is organized as follows. Section 2 discusses reasons to expect
heterogeneity in employment experiences during economic downturns and
briefly summarizes the relevant literature. We then proceed to a discussion
of our data in Section 3. We present our empirical methodology in section
4. Section 5 documents the current state of the U.S. labor market for different demographic groups and compares it to previous recessions. Section
6 discusses policy response and section 7 concludes.
3
2
Heterogeneity in Employment Experiences and
Background Literature
There are reasons to expect employment losses - unemployment inflows to weigh more heavily on women and/or people of color. With regard to
gender, job segregation, gender differences in labor market attachment and
job tenure, and gender employment discrimination all could serve as mechanisms by which women and men experience different effects on employment. With regard to race and age, one motivation is theoretical work by
Blanchard (1995), which argues that economic downturns have ”ladder effects” adversely affecting lower-income individuals. In this section, we outline the potential reasons for expecting differential employment responses
during changing economic conditions in the labor market. In the first instance our focus is on gender differences as research especially with regard
to gender on this issue is sparse.
2.1
Employment segregation
Empirical evidence in OECD countries indicates that women tend to work
in a different and narrower range of occupations than men, leaving the possibility of unevenly distributed employment effects during times of economic
change.2 Traditionally, men are more likely to be employed in manufactur2
Both demand- and supply-side explanations for employment segregation have been
advanced. On the demand side, employer discrimination against women, including the
perception that women are on average less qualified, could result in a greater willingness
to hire men and a greater willingness to lay off women first during economic downturns.
On the supply side, one explanation is that women self-select into occupations that require
smaller human capital investment, due to lower penalties for career breaks . This could be
attributed to ”societal discrimination” whereby women are expected to bear the burden
4
ing and agricultural professions while women tend to concentrate in administrative, public, and service sector occupations in a more restricted range
of professions. In OECD countries, recent shifts have occurred with both
women and men increasing their employment in managerial and professional
occupations. In terms of industries, in the US in the 1970s, for example,
28 percent of men were employed in manufacturing and 17 percent in services versus 21 percent and 42 percent respectively of women. In the 1990s,
this number changed to 21 percent of men in manufacturing and 25 percent
in services versus 11 percent and 47 respectively for women (Blau et al.
(1998)). In the 1990s, OECD countries saw the beginning of a greater demand for women in the labor market due to: technological change that
allowed substitution of men and women workers, the rise of the service sector and the decline of the production sector, increased education levels of
women, and effective anti-discrimination policy measures. As discussed in
the next section, while women’s labor market attachment increased, occupation and industry segregation, although declining, has remained an issue in
these countries (Dolado et al. (2002)). Given the existence of occupational
and industry segregation, a differential employment effect by gender due to
the onset of a recession can be expected, if these sectors have for example,
differing degrees of interest rates sensitivity. Cyclical properties of certain
industries and occupations could also result in a gendered employment effect. For example, in European Union (EU) countries, women’s relative
lower unemployment rates in the past have been attributed to female labor
shifts from manufacturing to the service sector, the latter less affected by the
of raising children, thus requiring more flexible jobs.
5
business cycle. Hence, women, by concentrating in industries less sensitive
to business cycle swings, shelter themselves from both negative and positive
business cycle effects (Buddelmeyer et al. (2004b)). More recently, the influences of changes in occupational distribution, rather than distributions by
industry, have been highlighted as having a greater effect on employment.
Using United Kingdom data, Rives and Sosin (2002) show that although at
times of recession, unemployment rises for both genders, the occupational
distribution favors women’s employment. More specifically, within occupations, women’s unemployment rates are consistently higher than men’s
rates. But the distribution of occupations favors women because low unemployment occupations have relatively higher proportions of women. This
evidence suggests the possibility of gender-specific employment effects, although the direction of that effect is ambiguous.
2.2
Labor market attachment
Men’s and women’s employment effect due to changing economic conditions
may also take place due to gender differences in the division of part-time
and full-time work and labor market attachment (resulting from men’s and
women’s different roles in the care economy) and its correlation with occupational segregation. In both Europe and the US, women have a considerably lower presence in full-time work compared to men (for example,
Blank (1998); Buddelmeyer et al. (2004a); Bardasi and Gornick (2008)) and
concentrate in temporary and part-time jobs, which are more sensitive to
economic downturns and upswings. Occupational segregation is also positively correlated with the share of part-time jobs, as these jobs tend to be
6
in occupations traditionally held by women.
2.3
Job tenure
A third reason we could expect differential employment responses is the
difference between genders in job tenure (Munasinghe and Reif (2008)). Researchers have found that women have shorter tenure (one reason is that they
leave work to start families) and consequently may be laid off faster than
men (see Booth et al. (1999) for the case of the UK). As a result, in times of
economic downturns women may suffer more in terms of employment. However, workers with substantial tenure may also be disproportionately hurt in
terms of employment during economic downturns. Ruhm (1987) finds that
although the inverse relationship between job duration and turnover rates
holds in the US, workers with substantial tenure in recently held jobs are
more vulnerable during cyclical fluctuations. This effect is strengthened in
sectors that are hit particularly hard by recessions. Overall, it is clear that
there are differences in job tenure between men and women, and that tenure
affects employment responses to economic conditions.
2.4
Gender discrimination
Employer gender discrimination can also result in employment segregation
and cause a gendered employment effect during recession. Employers may
perceive the productivity of men and women differently and prefer to hire
one over the other, either in hiring/firing the more productive or hiring/firing
the seemingly less productive and offering a lower wage. This type of behavior may not be evident when the economy is operating close to full em7
ployment but can certainly be in effect in times of economic downturns.
Although the argument of employer discrimination is difficult to maintain
with the existence of widespread occupational segregation, there is empirical evidence for the US showing that in male-dominated occupations and
industries, the unemployment rate for women has in the past increased more
at the cycle troughs (see the literature review in Rives and Sosin (2002) and
Azmat et al. (2006)). More recently, Singh and Zammit (2002) found that
women in developing countries were fired at substantially higher rates than
men after the Asian financial crisis. Another study also found that employers in developing countries may prefer to hire men as a means of reducing
costs in recessionary times given that women are more likely to go on leave
due to maternity or illness despite the fact that they are perceived as reliable
employees (Seguino (2003)).
3
Data
For our project, we use current publicly available data from the U.S. Current
Population Survey (CPS) that is continually updated. Our latest data comes
from November 2009. The unemployment data is collected by the U.S.
Department of Labor, Bureau of Labor Statistics beginning in 1948. For
this project we use three series for each demographic group: the number of
unemployed, unemployment rate and the number of short-term unemployed
(those unemployed for less than 5 weeks). While the unemployment rate
and the number of unemployed is typically available for the whole sample
(i.e. starting in 1948), the data for short-term unemployed is often available
8
from mid-1976 only. Thus, much of our analysis begins in 1976 (rather than
in 1948).
4
Empirical Methodology
Looking at unemployment rates gives us an idea of the share of people
not working in a given period of time or the probability that a randomly
chosen person will be unemployed. A more dynamic approach is to estimate
the underlying movements of workers into and out of unemployment. This
typically refers to the inflow rate, which is the pace at which workers move
into unemployment and the outflow rate, the pace at which workers move out
of unemployment. During recessions, generally, we see more people losing
jobs and becoming unemployed, hence we expect the inflow rate to increase.
At the same time, it is harder for people to find jobs, hence we expect the
outflow rates to decrease. In this paper, we examine both by computing job
finding and separation probabilities with a particular focus on differences
across demographic groups.
We use methodology developed by Shimer (Shimer (2007)). We calculate the job finding (inflow) rate -𝑓𝑡 and job separation (outflow) rate-𝑠𝑡 by
assuming that during period 𝑡, unemployed workers find or lose their job
according to a Poisson process with arrival rate 𝑓𝑡 ≡ −𝑙𝑜𝑔(1 − 𝐹𝑡 ) ≥ 0 or
𝑠𝑡 ≡ −𝑙𝑜𝑔(1 − 𝑆𝑡 ) ≥ 0, where 𝐹𝑡 and 𝑆𝑡 are finding and separation probabilities, respectively and by solving a differential equation for unemployment
and short term unemployment further described in the paper. The relationship then between unemployed workers at time 𝑡 and 𝑡 + 1 (𝑢𝑡 , 𝑢𝑡+1 ) and
9
short-term unemployed workers (𝑢𝑠𝑡+1 ) leads to the job finding probability
𝑢𝑡+1 − 𝑢𝑠𝑡+1
𝑓𝑡 ≡ − log(1 − 𝐹𝑡 ) = − log
𝑢𝑡
[
]
(1)
The implicit expression for the separation probability is
𝑢𝑡+1 =
(1 − exp−𝑓𝑡 −𝑠𝑡 )𝑠𝑡
𝑙𝑡 + exp−𝑓𝑡 −𝑠𝑡 𝑢𝑡
𝑓𝑡 + 𝑠𝑡
(2)
where 𝑙𝑡 ≡𝑡 +𝑒𝑡 is the size of the labor force during period 𝑡
This approach allows us to avoid time aggregation bias, as we work in
a continuous time model in which data are available at discrete intervals
(Shimer (2007)).
5
The Current State of the U.S. Labor Market
As mentioned in the introduction, during the most recent recession so far
more than 7.2 million jobs have been lost since December 2007. A look
into the demographic composition of employment and job losses suggests
that some demographic groups have been hit harder than others(see Table
1). For example, for the youngest group, the share of job losses exceeds
the group’s share in employment. For women, the oldest group (55+) also
suffered by this definition. With respect to race, blacks are affected more
than whites.
Turning to unemployment, the national unemployment rate reached a
high of 10.1 percent (October, 2009), bringing back the memories of unemployment rates as high as 10.8 percent reached during the recession of the
10
Table 1: Demographic composition of employment and job losses (percent)
Men
Women
Emp.comp. Job loss. Emp.comp. Job loss.
lt 25
13
20
14
19
25-54
69
64
68
61
55+
19
16
18
20
100
100
100
100
White
83
80
81
77
Black
10
15
12
16
Other
7
5
7
7
100
100
100
100
Source: Authors’ calculations and Bureau of Labor Statistics.
Note: Employment composition in 12/07. Job losses as of 11/09.
early 1980. To gain additional insight into which forces lead to high unemployment rates during recessions we examine job finding and separation
probabilities. The average job finding probability (Ft) during the whole
sample period (January 1948 - October 2009) is rather high at 43 percentage points, while the average separation probability (St) is rather low at
3.3 percentage points (See Figure 1). We find that in the recessions of the
1970s and 1980s, we observed considerable declines in job finding probabilities which were also accompanied by comparable increases in job separation
probabilities. This resulted in large increases in the unemployment rate
during recessions. However, the large recessionary increases in unemployment were also accompanied by strong unemployment rate declines after the
downturn. The recessions of the early 1990s and early 2000s were different:
as those are characterized by large declines in job finding probability which
were not accompanied by increases in job separation probabilities. Thus,
unemployment rate increases during those two recessions were driven by the
11
lack of hiring rather than firing of workers.
During the most recent economic downturn, the job finding probability fell from the pre-recession peak of just above 40 percentage points to
a low 17 percentage points. This level is the lowest observed since 1948.
The decline in job finding probability from pre-recession peak to trough is
57 percent. This is the largest peak-to-trough decline observed since data
collection began (the next largest decline observed is equal to 45 percent
(in the 1950s)). The separation probability increased from a pre-recession
low of slightly below 2 percent to a peak of just above 3 percent over the
course of the most recent recession. At 3 percent, the separation probability
is not extraordinarily high, as similar levels were observed during the previous recession and higher levels were observed in recessions prior to that.
Shimer (2007) points out the secular decline in separation probability since
the early 1980s. Recent data does not contradict this conclusion, although
the increase in the job separation probability over the course of the most
recent recession allows for a possibility of a reversal of this trend. During the
most recent recession, the job separation probability increased by just over
58 percent. This is the third largest increase in job separation probability
observed during the period (larger increases in job separation probability
were observed in the recessions of 1948 and 1953).
To gain insight into the cyclicality of the unemployment rate, we examine
the contributions of job finding and separation probabilities to unemployment rate fluctuations. We exploit the fact that
𝑠𝑡
𝑠𝑡 +𝑓𝑡
is a good approxi-
mation of the end-of-month unemployment rate. Let 𝑓¯ and 𝑠¯ denote the
average values of 𝑓𝑡 and 𝑠𝑡 over the sample period. We compute
12
𝑠¯
𝑠¯+𝑓𝑡
and
𝑓¯
𝑓¯+𝑠𝑡
as measures of contributions of fluctuations in job finding and separa-
tion rates to overall fluctuations in the unemployment rate. The results are
shown in Figure 2.
Each panel in Figure 2 shows the actual unemployment rate and the
hypothetical one. The hypothetical rate is computed either holding the job
finding rate or job separation rate constant at the sample average. Thus, the
top panel shows the hypothetical unemployment rate computed holding job
separation rate constant and, therefore, lends insight into the contributions
of job finding rate to unemployment rate fluctuations. The bottom panel
shows the hypothetical unemployment rate computed holding the job finding rate constant (and, therefore, gives us an idea about the contributions
of job separation rate to unemployment rate fluctuations). We find that
much of the aggregate unemployment rate fluctuations can be explained by
job finding rate movements, while movements in job separation probabilities
explain much less of the fluctuations in unemployment (as the hypothetical
unemployment rate depicted in the top panel of Figure 2 trails the actual unemployment rate closer than the hypothetical unemployment rate presented
in the bottom panel).
One possible explanation for the decline in the aggregate job finding
probability is the change in the composition of the labor force. For example,
with the aging of the baby boomers and increase in educational attainment
of the population, the share of prime age adults and those with higher
education increased. As these groups, on average, have a lower job finding
probability (see Table 2 later in this paper), the increase in their share in
total labor force would drive the aggregate job finding probability down. As
13
a result, to get a better understanding of the forces driving the changes in the
unemployment rate we proceed by examining differences in unemployment
rates by various demographic groups. To be specific, we look at gender, age,
and race (an important demographic variable in the United States). Our
findings indicate that this recession in many ways is different from those in
the past (in terms of degree of severity) and as a result will have different
implications for the well-being of households and individuals.
5.1
Age and Gender
In the United States, the unemployment rate for males tended to be below
that for females until the early 1980s. After, the situation reversed particularly during recessions (see Figure 3 for gender-specific unemployment
rates). In August 2009, the unemployment rate for males was 2.7 percentage points higher than that for females - the largest unemployment rate
gap observed in the history of the series. We further examine the gender
unemployment gaps by age groups. We distinguish six: 16-19, 20-24, 25-34,
35-44, 45-54, 55+. The results are shown in Figure 4. We show the difference between male and female unemployment rates (i.e. a positive gap
means that the unemployment rate for men is higher than that for women).
During the most recent recession the gap is the largest observed during the
sample period for all age groups. Interestingly, for prime age workers, the
male and female unemployment rate tended to converge since the 1980s (unemployment rate gap is close to zero), but during the recent recession the
gap increased dramatically.
This unusually large gap between male and female unemployment is
14
driven by historically high unemployment rates for males. At 11.4 percent
in October 2009, the unemployment rate for males stands at its highest
level since 1948. Last time male unemployment rate reached the teens was
during the recessions of the 1980s (the peak back then was 11.2 percent in
December of 1982). For females, unemployment rate stands at 8.8 percent.
While this is the highest unemployment rate we have observed for females in
more than two decades, it is not an unprecedented high, as unemployment
level for females reached 10.4 percent in December 1982.
The size of the unemployment rate increase also underscores the severity of this recession. Unemployment rate jumped by 6.3 and 4.1 percentage
points for males and females, respectively. These magnitudes are quite large
in comparison to previous recessions. For instance, the largest unemployment rate increase previously observed was 4.2 percentage points for males
(during the recession of 1981) and 3.6 percentage points for females (during the recession of 1973). Looking into the age break-down, we find that
for men the unemployment rates reached historic highs for all age groups,
whereas for women this is only the case for the youngest (16-19) and the
oldest (45-54) groups.
To learn more about gender and age unemployment rate differentials, we
look into job finding and separation probabilities(see Figure 5) since 1976.
At the beginning of the sample period, the job finding probability for males
tends to be lower than for females. The two rates start converging in the
early 1990s and move closely together during the most recent downturn.
The decline in job finding probability for men was 64.8 percent for men, and
58.5 percent for women. Both groups experienced the largest decline in the
15
job finding probability during the sample period.
The job separation probability for men also tends to be below that for
women over the sample period, but both seem to be systematically decreasing over time (this secular decline in job separation probability was also
pointed out by Shimer (2007)). The current downturn is a noticeable exception, as the job separation probability for men increased and became higher
than that for women. These results suggest that the gender gap differential
observed (higher unemployment rate for men) in the current downturn can
be explained by differences in job separation probabilities (with job separation probability for men exceeding that for women) and not job finding
probabilities. As we have shown this phenomena has not been observed
during previous recessions and is driving the current results.
Examining age-specific gender differentials in job finding and separation
probabilities we find that, on average, job finding probabilities are smaller
at older ages and they are statistically significantly higher for women than
for men (see Table 2). The gender gap in job finding probabilities has
been narrowing over time for all age groups, with the gender-specific job
finding probabilities converging since the mid-1990s. During the most recent
downturn, the job finding rate for females appears to have fared out better
than that for males for the younger group (those under 24), and there are
no noticeable differences for age groups above 24 (results available from the
authors).
The calculated job separation probabilities are also smaller for the older
workers (see Table 2). While gender specific job separation probabilities have
converged over time, in the recent recession job separation probabilities for
16
Table 2: Average Job Finding (F) and Separation (S) Probabilities (Standard Errors in Parenthesis)
Men
Women
F
S
F
S
16-19
0.49*
0.14*
0.53*
0.14*
(0.004 ) (0.001 ) (0.004 ) (0.002 )
20-24
0.40*
(0.004 )
0.06*
(0.000 )
0.47*
(0.004 )
0.06*
(0.001 )
25-34
0.35*
(0.003 )
0.03*
(0.000 )
0.41*
(0.004 )
0.04*
(0.001 )
35-44
0.31*
(0.003 )
0.02*
(0.000 )
0.37*
(0.004 )
0.02*
(0.000 )
45-54
0.28*
(0.014 )
0.01*
(0.001 )
0.34*
(0.017 )
0.02*
(0.001 )
55+
0.29
(0.003 )
0.01
(0.000 )
N/A
N/A
Source: Authors’ calculations.
Note: * indicates results are statistically significantly different at 5%.
females dropped noticeably in comparison to males for all age groups (results
available from the authors).
5.2
Race and Gender
We now take a look at the labor market indicators by race and gender.
The unemployment rate of the whites stands below that of the blacks (see
Figure 6). The available data show that the race gap has been growing since
1976 until early 1980s and then reversed course until the last recession. The
17
trend has been for a decreasing race gap although in 4 out of 5 recessions
the gap increased (the recession of the early 1990s is an exception). As a
result the increase observed during the most recent downturn is not unusual
although it is rather large in magnitude. The peak of 6.4 percent reached in
September of 2009 is about half of what was observed during the recession
of the 1980s (for instance, the gap reached 12.1 percent in January of 1983).
Examining the unemployment rate by race and gender (Figure 7) reveals
that the increase in the race unemployment gap during the current recession
is driven by the increase in the unemployment rate gap for males, as the
unemployment rate gap for females actually declined.
Turning to job finding and separation probabilities (Figure 8), we find
that for white women and men job finding probabilities are for the most
part higher than for blacks. During the current economic downturn, peakto-trough decline in job finding probability was higher for blacks. Job separation probabilities have been steadily declining since 1976 for women and
men and since the mid-1990s there is about a 1 percentage point difference
between the two race groups although those of whites remain lower than
those of blacks. Interestingly, for many demographic groups surveyed, we
observe that relatively speaking groups with lower job finding rates also have
lower job separation rates (for instance, different age groups). In this case,
though, we observe that blacks have lower job finding rates than whites, but
do not enjoy lower job separation rates although the differences are small.
It seems that for males, the observed increase in race unemployment
gap is driven by differences in job separation probabilities, as job separation
probability for blacks jumped noticeably above that for whites during the
18
recent recession (see Figure 8). This is not the case for females and we do
not observe an increase in their race unemployment gap. Thus, once again,
we see that the job separation rate is playing an important role in explaining
the differences between unemployment rates across demographic groups.
6
Policy Response
After documenting the current state of the U.S. labor market, we turn to
policy response. In particular, we look into the American Recovery and
Reinvestment Act (ARRA) of 2009, whose purpose (among others) is to
save and create jobs. The Council of Economic Advisers (an agency within
the Executive Office of the U.S. President charged with offering the President objective economic advice on the formulation of both domestic and
international economic policy) estimates that ARRA would increase employment by 3.5 million by the end of 2010 and 6.8 million by the end of
2012 (Council of Economic Advisers (2009)).
The employment and unemployment experiences during economic downturns, however, vary by demographic groups. The Obama administration
recognizes this and one of ARRA’s aims is to protect the most vulnerable from the deep recession. The administration estimates that roughly 42
percent of jobs created will go to women, which as of December 2007 held
about 48% of jobs and initially (until the end of November, 2008) accounted
for about 27% of the job losses during the current recession (Romer and
Bernstein (2009)).3 In order to assess whether this recovery package favors
3
Our most recent calculations based on Dec 2007-Nov 2009 data indicate women lost
about 35% of the jobs (see Table5).
19
Table 3: Change in Payroll Employment 2007-2009
Q1
Q2
Q3
Q4
2007 133
82
2
167
2008 -113 -153 -208 -553
2009 -691 -428 -199 -69
Source:Department of Labor (Bureau of Labor Statistics)
one demographic groups over another (for example, women over men) we
would need to understand the reasons lower shares of, for example, women
are employed in certain industries in the first place (due to discrimination
or individual preferences). As a result assessing the equity of the stimulus
package based on raw data alone is not fully satisfactory. Other evidence
on the demographic split of jobs created by the ARRA forecasts that less
jobs will go to whites compared to their initial employment share before the
recession, while nonwhites will not gain significantly. The highest job losses
not addressed by ARRA will be for those with low education levels (high
school or less)(Zacharias et al. (2009)).
Compared to the above studies, which forecast the likely path of recovery,
the most recent estimates of the impact of the ARRA published by the
Council of Economic Advisers (Council of Economic Advisers (2009),Council
of Economic Advisers (2010)) examine the effect of the stimulus plan relative
to a baseline scenario. Using past data of GDP and employment and actual
data from 2009 these estimates indicate that employment would be about
2 million jobs lower without the ARRA. In Table 3, BLS data indicate the
extent to which there has been a systematic decrease in the number of jobs
lost since the onset of the recession.
20
21
Jobs
Women
(000s)
CEA
1
34
103
197
42
36
228
35
87
22
34
955
Fraction
Female
13
13
29
43
42
59
44.7
77
53
52
57
Notes: Items may not add to total due to rounding. In bold if estimate of share of jobs created is larger than the share in
employment. Employment composition in 12/07. Levy estimates is ARRA employment estimated as in Zacharias et al. (2009)
considering two scenarios (government and private). See text.
Sources: Authors’ calculations; Bureau of Labor Statistics; Council of Economic Advisers (2009); Zacharias et al. (2009)
Table 4: Employment Effects of the Recovery Act by Sector, 2009:Q4
Factor
Empl.
Share of total jobs created
Total
Share
jobs
(000s)
Levy
CEA
CEA
Gov. Private
Mining and Logging
1%
2%
2%
0%
8
Construction
5%
5%
10 %
13 %
262
Manufacturing
9%
8%
9%
17 %
354
Trade, Transportation and Utilities 19%
15 % 16 %
22 %
459
Information
2%
2%
2%
5%
101
Financial Activities (FIRE)
6%
5%
5%
3%
61
Professional and Business Services
13%
12 % 13 %
25 %
510
Education and Health Services
15%
10 % 13 %
2%
46
Leisure and Hospitality
10%
8%
7%
8%
165
Other Services
4%
4%
4%
2%
43
Government
17%
29 % 19 %
3%
60
Total Nonfarm Employment
100%
100 % 100 %
100 %
2068
Using the employment effects calculated by the CEA we estimate the
possible job effects by gender by industries given the share of groups employed in each of the industries (see Table 4). We see that for some industries
the net gain of total jobs considering the baseline scenario is larger than their
share in total employment (in bold: construction, manufacturing, trade and
to the largest extent professional and business services) as compared to the
other sectors (education and health services, leisure and government). Taking into account the equity effects of the ARRA one should note that in
the former industries the majority of employees are men as compared to the
latter group. In Table 4 besides the estimates of the CEA, for comparison
purposes we also include two types of estimates of jobs created performed
by the Levy Institute based on different assumptions.4 These matched well
with CEA estimated considering the total number of jobs created in 20092011 (about 6.2 million), but there is some variation when comparing the
results by industry (particularly for manufacturing, professional and business services and government).
Finally, we compare the impact of the fiscal stimulus on employment by
demographic groups with the employment composition and job losses until
late 2009 (see Table 5). We find that men and the young have suffered in
terms of job loss relatively more then their share in employment would suggest. Job creation estimates suggest that the nonwhite will benefit relatively
4
In both of these the midpoint of ‘high’ and ‘low’ multipliers for transfers, taxes and
subsidies provided by the Congressional Budget Office is used. The difference lies in
the further assumption regarding the industrial distribution of final demand generated
by government purchases. The ‘government’ scenario assumes it is distributed among
government industries and the ‘private’ scenario assumes most of the final demand increase
is captured by private industries.
22
Table 5: Demographic composition of employment, job losses and ARRA
employment
Emp.comp. Job loss. ARRA emp.
Gov. Priv.
Gender
Men
54
65
60
63
Women
46
35
40
37
Race
White
Nonwhite
81
19
79
21
61
40
61
39
Age
lt 25
25+
13
87
18
80
10
90
12
88
Source: Authors’ calculations. Bureau of Labor Statistics;Zacharias et al. (2009)
Note: Employment composition in 12/07. Job losses as of 11/09. ARRA employment
estimated as in Zacharias et al. (2009) considering two scenarios (government and
private). See text.
more than the white from ARRA job creation and the young relatively less
than prime-age adults.
6.1
ARRA and the income distribution
Our results indicate that men, nonwhite and particularly the young have
been affected relatively more (in terms of percentages) by unemployment
during the current recession than their employment share would suggest.
To some extent this seems to be addressed by ARRA thus affecting the
distribution of earnings, although it still leaves the most vulnerable- vulnerable. Zacharias et al. (2009) estimate that jobs created by ARRA will
23
provide higher average earnings than the earnings of earners in non-ARRA
jobs by 3%. Particularly affected will be those in the bottom quintile of the
earnings distribution compared to the rest of the distribution. There will
be some gain for those with high school diploma, nonwhites and to women
compared to men although these will not be sufficient to close the respective earnings gaps. These authors also find that the gain in average income
resulting from the ARRA stimulus package will benefit those in the lower
quintiles relatively more than those in the higher quintiles, but the pro-poor
pattern of income growth will only have a negligible effect on the shares of
aggregate income enterning each quintile hence, suggesting that the overall
effect of ARRA on income inequality will be negligible.
7
Conclusions
This paper measures worker inflows and outflows into unemployment in the
United States between 1948 and 2009 and between 1976 and 2009 for several
demographic groups. The focus of the paper are the experiences of the most
vulnerable groups during the last recession and a comparison with previous
recessions.
We find that during the most recent recession the job finding probability
exhibited its biggest drop from peak to trough since official measurement
began (57%). In addition the job separation probability also exhibited one
of the largest increases in the post-war period. The decline in the job finding
probability seems to be explaining the majority of the fluctuations in the
unemployment rate, which to a certain extent can be explained by the chang-
24
ing composition of the labor force with older workers exhibiting smaller job
finding probabilities than younger workers (and at the same time smaller
separation probabilities).
This recession has also been accompanied by a large gender gap in unemployment with men driving the unemployment rate upwards (particularly
at older ages). Further insight shows that men currently have one of the
highest unemployment rates in history due to very low job finding probability rates. The increase in separation probabilities has not been so dramatic.
Gender differences though seem to be driven by the higher separation probabilities for men compared to women and not by the historically low finding
probabilities for men and women.
We find that the race gap has also increased being driven by the gap for
males as the differences in unemployment rates for black and white females
has actually decreased. In terms of job finding probabilities, historically they
have been higher for whites, and during this recession both white women and
men have exhibited less of a decline in these probabilities than their black
counterparts. Overall, the increase in the race unemployment gap for males
seems to be driven by differences in job separation probabilities, as job separation probability for blacks jumped noticeably above that for whites during
the recent recession. Yet again, the job separation rate seems to be playing
an important role in explaining the differences between unemployment rates
across demographic groups.
In terms of the ARRA stimulus package and its effect on job creation
the research has only began. For the moment, we find that industries that
have been hit the hardest (trade and professional and manufacturing) and
25
employ a majority of men will benefit the most. Those suffering the most
will be the low educated and the young.
The effect of the income distribution needs to be further examined, but
helping the poorest (through job creation or extension of unemployment
benefits) will have a negligible effect on income inequality although it should
be pointed out that not helping would lead to its further increase. Falling
stock prices and housing prices resulting in vanishing retirement accounts
and retirement wealth could potentially pose a big problem for the future
if this results in having large numbers of baby boomers in poverty. This
could potentially have an effect on the income distribution. Inequality is
mostly driven by very high earnings at the top end of the distribution, which
may decline temporarily as a result of the current recession thus reducing
income inequality. At the same time, disappearing wealth for the rich will
have a negative effect on private business and job creation. One interesting
direction for further research would be to focus on examining the effects of
the recession at the top end of the distribution.
References
Azmat, G., Guell, M., and Manning, A. (2006). Gender Gaps in Unemployment Rates in OECD Countries. Journal of Labor Economics, 24(1):1–37.
Bardasi, E. and Gornick, J. (2008). Working for Less? Women’s Part-Time
Wage Penalties across Countries. Feminist Economics, 14(1):37–72.
Blank, R. (1998). Research in Labor Economics, volume 17, chapter Labor
26
Market Dynamics and Part-Time Work, pages 57–93. Greenwich, CT:
JAI Pres.
Blau, F., Ferber, M., and Winkler, A., editors (1998). The Economics of
Women, Men and Work. Prentice Hall, Upper Saddle Brook, NJ.
Booth, A. L., Francesconi, M., and Garcia-Serrano, C. (1999). Job Tenure
and Job Mobility in Britain.
Industrial and Labor Relations Review,
53(1):43–70.
Buddelmeyer, H., Mourre, G., and Ward, M. (2004a). Recent Developments
in Part-Time Work in EU-15 Countries: Trends and Policy. IZA Discussion Paper 1415, Institute for the Study of Labor.
Buddelmeyer, H., Mourre, G., and Ward, M. (2004b). The Determinants of
Part-Time Work in EU Countries: Empirical Investigations with MacroPanel Data. IZA Discussion Paper 1361, Institute for the Study of Labor.
Council of Economic Advisers (2009). The Economic Impact of the American
Recovery and Reinvestment Act of 2009. First Quarterly Report.
Council of Economic Advisers (2010). The Economic Impact of the American
Recovery and Reinvestment Act of 2009. Second Quarterly Report.
Dolado, J., Felgueroso, F., and Jimeno, J. E. (2002). Recent Trends in
Occupational Segregation by Gender: A Look across the Atlantic. IZA
Discussion Paper 524, Institute for the Study of Labor.
Munasinghe, L. and Reif, T. (2008). The Gender Gap in Wage Returns on
Job Tenure and Experience. Labour Economics, 15(6):1296–316.
27
Rives, J. and Sosin, K. (2002). Occupations and the Cyclical Behavior of
Gender Unemployment Rates. The Journal of Socio-Economics, 31:287–
99.
Romer, C. and Bernstein, J. (2009). The Job Impact of the American
Recovery and Reinvestment Plan.
Ruhm, C. J. (1987). Job Tenure and Cyclical Changes in the Labor Market.
Review of Economics and Statistics, 69(2):372–78.
Seguino, S. (2003). Why Are Women in the Caribbean So Much More Likely
than Men to Be Unemployed? Social and Economic Studies, 52(4):83–120.
Shimer, R. (2007). Reassessing the ins and outs of unemployment. NBER
Working Paper 13421, NBER.
Singh, A. and Zammit, A. (2002). Gender Effects of the Financial Crisis in
South Korea. Paper presented at New Directions in Research on GenderAware Macroeconomics and International Economies: An International
Symposium. Levy Economics Institute of Bard College.
Zacharias, A., Masterson, T., and Kim, K. (2009). Distributional impact
of the american Recovery and Reinvestment Act: A microsimulation approach. Working Paper 568, The Levy Economics Institute of Bard College.
28
Figure 1: Aggregate Job Finding and Separation Probabilities
70%
6.0%
60%
5.0%
50%
4.0%
40%
3.0%
30%
2.0%
20%
1.0%
___ Job finding probability (left axis)
----- Job separation probability (right axis)
10%
29
Apr-06
Sep-08
Jun-01
Nov-03
Jul-96
Jan-99
Feb-94
Apr-89
Sep-91
Jun-84
Nov-86
Jul-79
Jan-82
Feb-77
Apr-72
Sep-74
Jun-67
Nov-69
Jul-62
Jan-65
Feb-60
Apr-55
Sep-57
Nov-52
Jan-48
0.0%
Jun-50
0%
Jan-48
Aug-50
Mar-53
Oct-55
May-58
Jan-61
Aug-63
Mar-66
Oct-68
May-71
Jan-74
Aug-76
Mar-79
Oct-81
May-84
Jan-87
Aug-89
Mar-92
Oct-94
May-97
Jan-00
Aug-02
Mar-05
Oct-07
16%
2%
0%
16%
2%
Jan-48
Feb-50
Mar-52
Apr-54
May-56
Jun-58
Jul-60
Aug-62
Sep-64
Oct-66
Nov-68
Dec-70
Jan-73
Feb-75
Mar-77
Apr-79
May-81
Jun-83
Jul-85
Aug-87
Sep-89
Oct-91
Nov-93
Dec-95
Jan-98
Feb-00
Mar-02
Apr-04
May-06
Jun-08
Figure 2: Contributions of Job Finding and Separation Rates to Fluctuations in the Aggregate Unemployment Rate
14%
Job Finding Rate
12%
10%
8%
6%
4%
--- Hypothetical Unemployment Rate
___ Actual Unemployment Rate
14%
Job Separation Rate
12%
10%
8%
6%
4%
--- Hypothetical Unemployment Rate
___ Actual Unemployment Rate
0%
30
31
Jan-08
Jul-05
Jan-03
Jul-00
Jan-98
Jul-95
Jan-93
Jul-90
Jan-88
Jul-85
Jan-83
Jul-80
Jan-78
Jul-75
Jan-73
Jul-70
Jan-68
Jul-65
Jan-63
Jul-60
Jan-58
Jul-55
Jan-53
Jul-50
Jan-48
Figure 3: Male and Female Unemployment Rate
12.0%
10.0%
8.0%
6.0%
4.0%
--- Male
___ Female
2.0%
0.0%
Figure 4: Unemployment Rate Gap for Men and Women by Age Groups
(3-months moving average).
10.0%
The Youngest (under 25) and Oldest (55+) Workers
8.0%
6.0%
4.0%
2.0%
0.0%
-2.0%
--- --
20-24
___
55 +
16-19
___
Jan-48
Feb-50
Mar-52
Apr-54
May-56
Jun-58
Jul-60
Aug-62
Sep-64
Oct-66
Nov-68
Dec-70
Jan-73
Feb-75
Mar-77
Apr-79
May-81
Jun-83
Jul-85
Aug-87
Sep-89
Oct-91
Nov-93
Dec-95
Jan-98
Feb-00
Mar-02
Apr-04
May-06
Jun-08
-4.0%
4.0%
Prime Age Workers (25-54)
3.0%
2.0%
1.0%
0.0%
-1.0%
-2.0%
-3.0%
---
25-34
___
35-44
45-54
___
Jan-48
Feb-50
Mar-52
Apr-54
May-56
Jun-58
Jul-60
Aug-62
Sep-64
Oct-66
Nov-68
Dec-70
Jan-73
Feb-75
Mar-77
Apr-79
May-81
Jun-83
Jul-85
Aug-87
Sep-89
Oct-91
Nov-93
Dec-95
Jan-98
Feb-00
Mar-02
Apr-04
May-06
Jun-08
-4.0%
Note: Gap shown is the difference between male and female unemployment rates
32
Jun-76
Sep-77
Jan-79
Apr-80
Jul-81
Oct-82
Jan-84
Apr-85
Jul-86
Oct-87
Jan-89
Apr-90
Jul-91
Oct-92
Jan-94
Apr-95
Jul-96
Oct-97
Jan-99
Apr-00
Jul-01
Oct-02
Jan-04
Apr-05
Jul-06
Oct-07
Jan-09
Jun-76
2.0%
1.0%
33
Jun-00
7.0%
Job Separation Probability
6.0%
5.0%
4.0%
3.0%
--- Male
___ Female
0.0%
Oct-09
Jun-08
Feb-07
Oct-05
Jun-04
Feb-03
Oct-01
10.0%
Feb-99
Oct-97
Jun-96
Feb-95
Oct-93
Jun-92
Feb-91
Oct-89
Jun-88
Feb-87
60.0%
Oct-85
Jun-84
Feb-83
Oct-81
Jun-80
Feb-79
Oct-77
Figure 5: Job Finding and Separation Probabilities by Gender.
Job Finding Probability
50.0%
40.0%
30.0%
20.0%
--- Male
___ Female
0.0%
Figure 6: Unemployment Rate and Unemployment Rate Gap by Race (3month moving average).
25%
20%
--- -15%
black
___
white
___
gap
10%
5%
Apr-09
Jun-06
Nov-07
Jul-03
Jan-05
Feb-02
Apr-99
Sep-00
Jun-96
Nov-97
Jul-93
Jan-95
Feb-92
Apr-89
Sep-90
Jun-86
Nov-87
Jul-83
Jan-85
Feb-82
Apr-79
Sep-80
Jun-76
Nov-77
0%
Note: Gap shown is the difference between black and white unemployment rates
34
Jun-76
35
Apr-99
--- --
Female
___
black
white
___
gap
15%
10%
5%
0%
Jun-06
Nov-07
Apr-09
Jun-06
Nov-07
Apr-09
Jul-03
Jan-05
Jan-05
gap
Feb-02
white
___
Jul-03
___
Feb-02
Sep-00
Apr-99
Nov-97
Jun-96
Jan-95
Jul-93
Feb-92
Sep-90
Apr-89
Nov-87
Jun-86
Jan-85
Jul-83
Feb-82
--- --
Sep-00
20%
Jun-96
Apr-79
Sep-80
20%
Nov-97
25%
Jan-95
Jun-76
Nov-77
25%
Jul-93
Feb-92
Sep-90
Apr-89
Nov-87
Jun-86
Jan-85
Jul-83
Feb-82
Sep-80
Apr-79
Nov-77
Figure 7: Unemployment Rate and Unemployment Rate Gap by Race and
Gender
Male
black
15%
10%
5%
0%
Note: Gap shown is the difference between black and white unemployment rates
View publication stats
Jun-76
-10%
Apr-79
Apr-99
0%
recession
36
Jul-91
black
Jul-88
12%
--- -black
50%
10%
Females
8%
___
6%
white
4%
2%
0%
white
Jul-06
Jan-08
Jul-09
Jan-08
Jul-09
Jul-03
Jan-05
Jul-06
Jan-05
Jul-00
60%
Jul-03
white
Jan-02
black
___
Jul-00
Jan-99
Jul-97
Jan-96
Jul-94
Jan-93
Jul-91
Jan-90
--- --
Jan-02
Jan-99
Jul-97
Jan-96
Jul-94
Females
Jan-93
white
Jul-85
Males
Jan-90
black
Jul-88
30%
Jan-87
40%
Jul-82
2%
Jan-87
4%
white
Jul-85
black
___
Jul-79
--- --
Jan-84
6%
Jan-81
30%
Jul-82
8%
Jan-84
40%
Jan-78
10%
Jan-81
0%
Jun-76
Apr-09
Nov-07
Jun-06
Jan-05
Jul-03
Feb-02
Sep-00
Apr-99
Nov-97
12%
50%
Jul-79
Jul-93
Jan-95
Jun-96
60%
Jan-78
Jun-76
Apr-09
Nov-07
Jun-06
Jan-05
Jul-03
Feb-02
Sep-00
___
Nov-97
--- -
Jan-95
10%
Feb-92
Sep-90
Apr-89
Nov-87
Jun-86
Jan-85
Jul-83
Feb-82
Sep-80
10%
Jun-96
Jun-76
Apr-79
Nov-77
20%
Jul-93
20%
Feb-92
Sep-90
Apr-89
Nov-87
Jun-86
Jan-85
Jul-83
Feb-82
Sep-80
-10%
Nov-77
Figure 8: Contributions of Job Finding (left) and Separation Rates (right)
to Fluctuations in the Aggregate Unemployment Rate
Males
0%