View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

DISTRICTDIGEST

Economic Trends Across the Region

Learning About the Labor Market from High-Frequency Data
B y J o s ep h M e n ge d o t h

E

mployment in the United States experienced
the sharpest decline on record in April as the
negative economic effects of the COVID-19 pandemic and social distancing measures caused employers
to cut almost 21 million jobs, on net. (The next largest
single-month decline was almost three-quarters of a
century earlier, in September 1945, when almost 2 million jobs were lost.) Yet the full severity of the job loss
was not known for quite a while: More than seven weeks
passed from when the first state, California, issued a
stay-at-home order on March 19 to when the Bureau
of Labor Statistics (BLS) released the first national
employment report fully reflecting the onset of the crisis, the report for April released on May 8.
Traditional sources of employment data are lagged,
sometimes by a lot. At the national level, the employment
report for a given month is typically released on the first
Friday of the following month. And those data are based
on a survey of firms that takes place around the middle of
the month. This is why the jobs report for March had yet
to show the full effect of the widespread social distancing
measures, since many of those were put into place in late
March and early April.
The BLS releases employment data for state and lower
levels of geography at even greater lags. For example,
the state-level data are typically lagged by another two
weeks, coming out in the middle to the end of the month.
County and metro employment and unemployment data
are released a few weeks after that. And the most comprehensive source of data on local employment comes from
the Quarterly Census of Employment and Wages, which
is released between five and six months after the quarterly
period ends. (For more on state and local labor market
data, see “State Labor Markets: What Can Data Tell (or
Not Tell) Us?” Econ Focus, First Quarter 2015.)
These lags are not new, or unknown, but in times
of rapidly changing circumstances, the data are not
sufficiently able to keep up with economic conditions.
Knowing that the official employment counts would
not be available for some time, economists, policymakers, and analysts looked during the COVID-19 crisis to
other sources that could shed light on how the virus and
the shutdown of economic activity were affecting the
labor market. This includes the Federal Open Market
Committee (FOMC), which, according to the minutes
from meetings held in March, April, and June, found that
traditional economic data could not capture the rapidly
evolving situation; instead, the committee referenced
high-frequency data.
Share this article: https://bit.ly/high-freq-data

Unemployment Insurance Claims
One source that directly shows changes in labor markets
on an early basis, which the FOMC relied on in March,
April, and June, is weekly unemployment insurance claims.
Unemployment insurance programs are administered by
individual states. Every state is required to report the
number of initial and continued claims to the Department
of Labor, which in turn releases that data to the public
on the Thursday of the following week. As their names
imply, initial claims are the number of new claims filed in
the reference week, and continued claims are the number
of workers who were already collecting unemployment
benefits and remained unemployed in the reference week.
Because these data are timelier than payroll employment data from the BLS, they can serve as an early
indicator of an economic downturn. In normal times,
there is some variation in these data week to week as
people move from employment to unemployment and
back to employment or as some people decide to leave
the labor force rather than continue to look for a new job.
There are also seasonal patterns in the data, but those can
be removed by applying a statistical procedure known as
seasonal adjustment. Hindsight shows that in the weeks
leading up to the starts of the last several recessions, the
claims data tended to rise steadily and sometimes rapidly.
Take the Great Recession, for example. Data from the
payroll survey began showing the decline in employment
in February 2008, which was the first of 21 consecutive
months of job losses. If we look at the six months prior
to that, from August 2007 through January 2008, the payroll data were not alarming, with a slight increase in total
employment in the United States (0.3 percent or 388,000
jobs). At the same time, though, initial claims (after being
adjusted for seasonal trends) began to steadily increase,
and seasonally adjusted continued claims rose 12.4 percent
or by 314,000 jobs.
Likewise, evidence of an effect on employment from
the COVID-19 pandemic appeared in the initial and continued claims data several weeks before the payroll data
were available — but this time at rates never seen before.
The first increase in initial claims in the United States
came in the week ending March 14, when the number of
claims rose 33.3 percent or by around 70,000. In the next
week, initial claims rose more than tenfold from around
280,000 claims to almost 3.3 million and then more than
doubled the week after to almost 6.9 million. The same
data for Fifth District jurisdictions show similar trends
except for West Virginia, where the initial claims data
didn’t peak for another couple of weeks.
E c o n F o c u s | s e c o n d / T h i r d Q u a r te r | 2 0 2 0

27

Continued Unemployment Insurance Claims

demand for workers. Fortunately, there are some other
high-frequency data sources that can give a glimpse into
the staffing needs of employers.

Source: Author’s calculations using data from U.S. Department of Labor via
Haver Analytics		

A similar story evolved with continued claims, which
began to rise one week after the first spike in initial claims
and continued to increase sharply week over week for the
next several weeks. Claims rose nearly simultaneously
across jurisdictions at the start of the pandemic, but there
were variations in trends after that. Most notably, the
number of people filing continued claims began leveling
off and, in some cases, decreasing by the end of April or
the start of May — except for the District of Columbia,
where claims continued to rise and remained relatively flat
in May and June. (See chart.)
In addition to providing the data to the Department of
Labor, some state agencies release more detailed reports
of the initial claims data on their own websites. Virginia
is one of those states; its weekly reports include breakouts by gender, age, race, ethnicity, education level, and
occupation. These breakouts offer a view into disparate
impacts on different groups of people. The occupational
data, for example, showed that in the week of April 4, the
top two most affected occupations were food preparation
and serving related occupations and personal care and service operations. In contrast, just prior to the start of the
pandemic, the occupations with the largest numbers of
claimants were administrative support and construction.
This gave an early indication of which workers and industries might see the largest effects, which was confirmed in
the payroll employment data several weeks later.
But what about tracking the recovery in real time? One
of the limitations of these data is that we do not know the
characteristics of those who stop filing a continued claim
or the reason why they stopped. A drop-off in continued
claims could indicate that people are going back to work,
but it could also mean that people gave up looking for a
job or exhausted their benefits. So a drop-off doesn’t tell
us much about the types of people who stopped filing
versus those who remain on unemployment or the current
28

E c o n F o c u s | s e c o n d / T h i r d Q u a r te r | 2 0 2 0

New Online Job Postings
Indexed to the week ending March 7, 2020
20
10
0
-10
-20
-30
-40
-50
-60
-70

District of Columbia
South Carolina

Maryland
Virginia

July 11

North Carolina
West Virginia

June 27

July 11

June 27

June 13

May 30

May 16

Maryland
Virginia

June 13

District of Columbia
South Carolina

May 2

April 18

April 4

March 21

March 7

0

May 30

2

May 16

4

May 2

6

April 18

8

Online Job Postings
One way to measure the current demand for workers is to
look at the job advertisements that employers are posting online. To do that, one could simply peruse sites like
LinkedIn or Indeed, but there are companies that offer
aggregated data from across multiple websites. One such
company is Chmura Economics & Analytics, a Richmondbased consulting service and data provider. Among the
company’s offerings is a database of online job postings
called Real-Time Intelligence (RTI).
To create the RTI database, Chmura’s computers
scrape information from over 30,000 websites every day,
including job sites like Indeed and individual company
websites. When the data are processed each night, any
duplicate postings that are identified are removed. One
of the many pieces of information that Chmura gets
from these websites is the date when the job opening
was first posted, if available. If no such date is available,
Chmura assigns one based on the first day on which their
scraping process found the post. This date can be used
as a filter and therefore allows a user to see how many
job advertisements were posted online over a particular
time frame.
Looking at the data by week for the Fifth District
shows the dramatic decline in new job postings starting in
mid-March — around the time when mandatory business
closings and social distancing measures were being put in
place. It’s no surprise that with many businesses essentially
shut down, there was little need to hire new employees, but
these data show the severity with which those job postings
declined. At the lowest point, in the week ending April 18,
new job postings across Fifth District jurisdictions were

April 4

PERCENT

10

March 21

12

March 7

14

PERCENT

Share of state noninstitutionalized population aged 16 and older

North Carolina
West Virginia

Source: Author’s calculations using data from Chmura Economics & Analytics

July 11

June 27

June 13

May 30

May 16

May 2

April 18

April 4

March 21

March 7

Percent

New Online Job Postings for Selected Occupations
down between 36.2 percent (in
Indexed to the week ending March 7, 2020
West Virginia) and 57.9 per40
cent (in South Carolina) when
compared to the number of
20
Transportation and Material Moving
new postings in the first week
0
Office and Administrative Support
of March. (See chart on previ-20
Sales and Related
ous page.)
Food Preparation and Serving
But what can these data
-40
Health Care Practitioners and Technical
tell us about the job recov-60
Computer and Mathematical
ery? For one, they show that
Business and Financial Operations
-80
West Virginia experienced the
Management
strongest and quickest bounce
-100
back in online job postings.
In fact, the number of new
postings in the week ending
Source: Author’s calculations using data from Chmura Economics & Analytics
July 18 exceeded the number
of postings in the first week of
March. One potential reason for West Virginia’s quicker
As with unemployment claims data, online job postrecovery in job postings is that the state was the first in the
ing data do not tell the whole story. For one, given the
Fifth District to ease restrictions on businesses and social
number of jobs that were lost in March and April, if the
gatherings. In fact, the Mountain State entered the second
number of new job postings matches the pre-pandemic
phase of its reopening on May 4, which was the same day
level, that doesn’t mean the labor market has returned to
that South Carolina entered its first phase and before any
the same level of demand. And one might expect to see
other Fifth District jurisdiction began easing restrictions.
the number of new job postings exceed the pre-pandemic
The RTI database includes many other variables that
level for some time in order to fully recover the jobs that
allow users to dig deeper into the data to see what types
have been lost.
of jobs were hit hardest and have recovered the most. For
Additionally, while the data do show some trends
example, Chmura’s web scraping tool examines job titles
in the types of jobs that are being advertised for, they
and job descriptions to assign each job posting an occudo not show how many of those jobs were filled. And
pation code based on the BLS’s Standard Occupational
with part-time jobs, in particular, they do not show how
Classification System. This allows users to examine trends
many hours a week employers needed workers. There is
in job postings for specific professions or to see what types
another high-frequency data source, however, that sheds
some light on the demand for hourly workers.
of occupations were in the highest demand in a particular
time period, which gives insight into the hiring trends in
Homebase
some of the hardest-hit industries.
Homebase is a company that provides free scheduling,
Among the eight occupation groups that accounted
time keeping, and communication products to local
for the largest shares of new job postings in the first week
businesses with hourly employees. These are primarily
of March, postings for food preparation and serving
restaurant, food and beverage, and retail businesses
related occupations declined the furthest in late March
that are individually owned, which were some of the
and early April, followed by office and administrative
hardest-hit industries. In response to the pandemic,
support, sales and related jobs, and transportation and
the company made some of its data free to the public
material moving occupations. (See chart.) This was an
so researchers and community members could track the
early indication that the effects on the labor market
number of hours worked by hourly and shift employees,
would be felt quite differently across different types of
the number of businesses that were currently closed, and
jobs, which was confirmed by the official payroll employthe employees who were not working. All told, these
ment data — several weeks after the online job posting
daily data are based on more than 60,000 businesses
data was available.
employing 1 million hourly employees. Data start in
The same data shed light on the recovery in employJanuary 2020 and are available to the public in more real
ment. Online postings for health care practitioners and
time upon request.
technical workers and transportation and material moving
Because the data are daily, and many businesses are
occupations surged in the Fifth District in the week endnot open seven days a week, the data exhibit some coning July 18. Postings for sales and related jobs also picked
sistent patterns due to normal closures on certain days
up in the first few weeks of July. This could be a sign that
every week, like weekends. To correct for this, the data
business conditions were improving at establishments
can be indexed to a prior period. Data used for this article
that employ these workers, such as doctor’s offices, shiphave been indexed to the median value for the same day
ping companies, and retail shops.
E c o n F o c u s | s e c o n d / T h i r d Q u a r te r | 2 0 2 0

29

Hours Worked by Hourly Employees

District of Columbia
South Carolina

Maryland
Virginia

July 6

June 22

June 8

May 25

May 11

April 27

April 13

March 30

March 16

20
10
0
-10
-20
-30
-40
-50
-60
-70
-80
March 2

PERCENT

Percent change relative to the median same day of the week in January 2020,
trailing seven-day moving average

North Carolina
West Virginia

note: The three pronounced dips in the data coincide with Easter Sunday, Memorial Day,
and Independence Day.
Source: Author’s calculations using data from Homebase

of the week for the period Jan. 4 to Jan. 31. This means,
for example, that the hours worked on Wednesday, July
1 would be indexed to the median hours worked over the
five Wednesdays in January. Looking at the data this
way allows comparison over time relative to a particular
period and across geographies.
Across Fifth District jurisdictions, the trends in these
data broadly coincide with where and when places began
to reopen. For example, hours worked by hourly employees in West Virginia and South Carolina have bounced
back quicker and are closer to their January levels
than in other states — perhaps reflecting that West
Virginia and South Carolina began their phased reopenings much sooner than other jurisdictions. The District
of Columbia, which was the last in the Fifth District to
reopen, remains the furthest from its pre-pandemic level.
(See chart.)
Homebase data are also available broken out by
industry. This means we can observe trends in the hours
worked at just food and drink establishments or the
number of businesses open in the personal care industry. In the Fifth District as a whole, these data show
trends that one might expect, namely, a steep decline
in employees working, hours worked, and locations
open (all the way to zero, in some cases) starting in
mid-March. The series then bottomed out and began to
rise around mid-April when businesses began to resume
operations on a limited basis, reflecting the phased
approach to reopening that was occurring across much
of the nation.
Hours worked leveled off or showed a slight declining
trend toward the end of July. This may be a signal that
the demand for hourly workers is slowing and may remain
30

E c o n F o c u s | s e c o n d / T h i r d Q u a r te r | 2 0 2 0

below prior levels for some time. Of course, hourly workers are only one segment of the labor force, but this pattern anticipated a similar one in July payroll data, which
was released several weeks later and showed a slowdown
in the pace of hiring.
Richmond Fed Surveys
In addition to the high-frequency data sources that
have been discussed so far, the Richmond Fed has been
using its own surveys of business conditions to gain
further insights related to the pandemic. For example,
in the March surveys of manufacturing and service sectors, which were fielded between Feb. 26 and March
18, respondents were asked additional questions about
the impacts to their company so far due to COVID-19
and their expectations for the near term. Although the
Richmond Fed publicly releases the results only after
surveys have closed, staff often view responses as they
come in on a daily or weekly basis.
In general, over the survey period, firms were reporting
only minor negative effects on their operations, and most
of the comments indicated those were due to supply chain
disruptions from China and travel restrictions. By the
third week of the survey, however, responses indicated
that those negative impacts were escalating and outlooks
for the U.S. economy were deteriorating.
The April survey, which ran from March 26 to April
22, was broadened further to include labor market
specific questions. Specifically, that survey asked participants to indicate if they were reducing staff or the
hours worked by staff. Results from those questions
generally showed that the majority of responding firms
were not reducing staff or the hours worked by employees; however, similar to the March survey, the results
deteriorated as the survey continued. For example, in
the first week of the survey, only about 15 percent of
responding firms said they reduced staff, while in the
final week of the survey, approximately 40 percent said
they were cutting staff.
Then, in the May survey, the Richmond Fed collaborated with several chambers of commerce across the Fifth
District to reach even more participants with a set of
COVID-19 related questions. Overall, results from that
survey showed how the labor market responses of firms
varied by size and industry, with the most adverse effects
being felt in the accommodation and food services, retail
industries, and by small businesses. In contrast to earlier
surveys, the results were generally consistent over the
three weeks of the survey period.
The results of these surveys gave the Richmond Fed
timely information about firms’ experiences and the actions
they took while the COVID-19 situation was unfolding.
What’s more, they gave evidence that the changing nature
of the data over time means that one monthly indicator
alone may hide some underlying dynamics or, at the very
least, doesn’t tell the whole story.

Emerging Sources
A few newer sources have become available. The first is
the Real-Time Population Survey (RPS), which is a joint
effort between academic economists and the Dallas Fed.
The goal of the RPS is to provide a survey similar to the
BLS’ household survey of employment and unemployment
(the Current Population Survey), but it differs in that the
RPS is conducted online twice a month, and the results
are made available with a shorter lag. The results of the
RPS are plotted with the official BLS survey measures of
employment and unemployment in reports available on
the Dallas Fed’s website.
The U.S. Census Bureau also began conducting two
new high-frequency surveys to better understand the
effects of COVID-19 on the economy. The first was the
Household Pulse Survey, which was a weekly survey that
began on April 23 and concluded on July 21. The results of
the survey were posted one week after the survey period
closed and gave insights into issues such as childhood education (including availability of computers and internet),
employment, household spending and food sufficiency
and insufficiency, health, and housing. The data, which are
available at a national, state, and metropolitan level (for
the 15 largest metro areas), are still available on the U.S.
Census Bureau’s website at the time of writing this article.
The second new survey from the U.S. Census Bureau
is the Small Business Pulse Survey, which began on May
14 and is still ongoing. It is designed to provide information on small-business operations and finances, including
any government support they have received and their
outlook for the near future. These data are available at

the national and state levels and for the 50 most populous metro areas. An interactive dashboard shows which
industries and areas of the country have a relatively
higher share of small businesses being negatively or positively affected by the pandemic and where firms are the
most optimistic or pessimistic about the near future.
Conclusion
Although none are without limitations, each of these
high-frequency data sources offers a glimpse into the labor
market in nearer to real-time. The initial unemployment
insurance claims data were particularly useful in understanding how many and, in some cases, the characteristics
of workers who were being hurt during the crisis when
many businesses were scaling back or shutting down
operations.
The continued claims data were (and will continue
to be) a useful indicator to track the number of people
who are collecting unemployment each week. In terms
of labor demand, online job posting data offer a glimpse
into the types of jobs that employers are recruiting
for, and the Homebase data show trends in the hours
worked by hourly employees in some of the hardest-hit
industries. Lastly, the Richmond Fed has used and will
continue to use the ability to add special, topical questions to its surveys of business conditions to understand
the effects of the pandemic.
EF
The Richmond Fed has created Pandemic Pulse, an area on its
website that features interactive charts of various high-frequency
indicators.

u

E c o n F o c u s | s e c o n d / T h i r d Q u a r te r | 2 0 2 0

31