Corporate Hiring Under COVID-19 (NBER, May 2020)
Summary
NBER Working Paper No. 27208, Corporate Hiring under COVID-19: Labor Market Concentration, Downskilling, and Income Inequality, by Murillo Campello, Gaurav Kankanhalli and Pradeep Muthukrishnan of Cornell University, dated May 2020. The paper uses job-vacancy postings data from LinkUp to study how U.S. companies changed hiring during the pandemic. It reports that active job postings in the first week of May 2020 were 40% below the 2017–2019 average for that week, and that firms cut high-skill postings more than low-skill postings. It also examines firm size, unionization, local labor market concentration, credit access and firms receiving Paycheck Protection Program (PPP) funding. The document closes with a regression table relating new job postings and the high-to-low-skills postings ratio to county-level COVID exposure.
Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used
Full text
NBER WORKING PAPER SERIES
CORPORATE HIRING UNDER COVID-19:
LABOR MARKET CONCENTRATION, DOWNSKILLING, AND INCOME INEQUALITY
Murillo Campello
Gaurav Kankanhalli
Pradeep Muthukrishnan
Working Paper 27208
http://www.nber.org/papers/w27208
NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
May 2020
We are grateful to Tom Ashenmacher at LinkUp for data access and support. All errors are our
own. The views expressed herein are those of the authors and do not necessarily reflect the views
of the National Bureau of Economic Research.
NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official NBER publications.
© 2020 by Murillo Campello, Gaurav Kankanhalli, and Pradeep Muthukrishnan. All rights
reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit
permission provided that full credit, including © notice, is given to the source.
Corporate Hiring under COVID-19: Labor Market Concentration, Downskilling, and Income
Inequality
Murillo Campello, Gaurav Kankanhalli, and Pradeep Muthukrishnan
NBER Working Paper No. 27208
May 2020
JEL No. E24,G31,J23
ABSTRACT
Big data on job-vacancy postings reveal several dimensions of the impact of COVID-19 on the
U.S. job market. Firms have cut back on postings for high-skill jobs more than for low-skill jobs,
with small firms nearly halting their new hiring altogether. New-hiring cuts and downskilling are
most pronounced in local labor markets lacking depth (where employment is concentrated within
a few firms), in low-income areas, and in areas with greater income inequality. Cuts are deeper in
industries where workers are more unionized and in the non-tradable sector. Access to finance
modulates corporate hiring, with credit-constrained firms curtailing their job postings the most.
Our study shows how the early-2020 global pandemic is shaping the dynamics of hiring,
identifying the firms, jobs, places, industries, and labor markets most affected by it. Our results
point to important challenges to the scale and speed of a recovery.
Murillo Campello Pradeep Muthukrishnan
Johnson Graduate School of Management Johnson Graduate School of Management
Cornell University Cornell University
114 East Avenue 241 Sage Hall, 114 East Avenue
369 Sage Hall Ithaca, NY 14853
Ithaca, NY 14853-6201 USA
and NBER pm532@cornell.edu
campello@cornell.edu
Gaurav Kankanhalli
Johnson Graduate School of Management
Cornell University
241 Sage Hall, 114 East Avenue
Ithaca, NY 14853
gak79@cornell.edu
1 Introduction
The COVID-19 pandemic has brought about the largest economic dislocation since the Great
Depression. The ongoing crisis hit the corporate sector in a unique fashion. While prior shocks
to business activity came through channels such as the supply of capital (the Financial Crisis)
or technology (digitalization and automation), COVID-19 hit the human capital component of
the production process. No machines were destroyed or became obsolete, no banks failed, and
access to credit did not collapse — instead, human capital was threatened by a global health
crisis. The consequences to labor markets are likely deep and long lasting. While aggregate
data point to unprecedented labor market disruptions, understanding and addressing the forces
driving aggregate movements will require identifying the firms, jobs, places, industries, and the
skill level of workers most directly affected by the pandemic.
We present an assessment of the impact of the COVID-19 pandemic on the hiring decisions
of U.S. companies. Hiring represents a costly, forward-looking investment in human capital
and the decision to accelerate or scale back hiring reflects managers’ expectations about their
companies’ future. We are able to track these decisions during the COVID-19 pandemic using
big data on firms’ job postings from LinkUp, a leading labor market research firm. The LinkUp
data comprise job postings sourced from company hiring boards and websites of over 50,000
employers, encompassing public and private firms across all industries and regions of the U.S.
The data are continuously updated and provide information on the employer, position sought,
desired worker skill-level, and location of each job posting. These detailed records enable us to
match the job postings data with a number of firm-specific information, allowing us to gauge
how various margins of corporate hiring respond to the 2020 pandemic in real time.
Job postings by American companies have been dramatically altered by COVID-19; see
Figure 1. An abnormal drop begins in the first week of March 2020, coinciding with the very
beginning of the pandemic spread in the U.S.1 The drop in hiring leads the unprecedented
spike in initial jobless claims by almost two weeks. It also predates the first local stay-at-home
1
At the time, the first U.S. death officially attributed to COVID-19 by the Centers for Disease Control
and Prevention (CDC) was recorded on February 29th , 2020. Subsequent reports suggest that deaths that
occurred earlier in February in Santa Clara County, California were potentially attributable to COVID-19.
1
50 60 70 80 90 100 110 120
6
Initial Jobless Claims (Millions)
5
Job Postings (Indexed)
4
3
2
1
0
Week 1
Week 2
Week 3
Week 4
Week 5
Week 6
Week 7
Week 8
Week 9
Week 10
Week 11
Week 12
Week 13
Week 14
Week 15
Week 16
Week 17
Week 18
Average 2017-19 Postings 2020 Postings
2020 Initial Jobless Claims
Figure 1. Job Posting Dynamics. This figure plots the 7-day rolling average of total active job postings
across the first 18 weeks of 2017–2019, contrasting it to the level of total active job postings over the same
period of 2020. The series are indexed such that they take the value of 100 for the first week of January (on
the left y-axis). On the right y-axis, we show the weekly level of initial jobless claims (in millions).
orders, as well as state and federal emergency declarations by one week. The magnitude of the
decline in new hiring ads is striking, with the level of active job postings as of the first week of
May 2020 dropping 40% below the average level of postings as of the same week in 2017–2019.
A key benefit of our data is that they contain critical firm-specific information, allowing us to
track corporate hiring patterns at a great level of detail. Doing so reveals significant disparities
in both the types of firms curtailing their hiring and hiring cuts across job skill levels. Figure
2 depicts these dynamics as the pandemic unfolds. The decline in active job postings by small
firms substantially exceeds that of large firms, with small firms reducing their hiring by over
50% and large firms reducing hiring by some 30–40% (see Panel A). Panel B shows that hiring
cuts among high-skill jobs (e.g., CEOs, lawyers, post-secondary teachers, statisticians, and
physicians) exceed those among low-skill jobs (agricultural, food services, landscaping, garment,
and timber logging workers). The differential effects of COVID-19 on firms and jobs may
determine which policy responses are likely to be effective in re-establishing a well-functioning
labor market. They can inform us about the extent and speed of a potential recovery.
2
50 60 70 80 90 100 110 120 50 60 70 80 90 100 110 120
Job Postings (Indexed) Job Postings (Indexed)
Week 1 Week 2 Week 3 Week 4 Week 5 Week 6 Week 7 Week 8 Week 9 Week 10 Week 11 Week 12 Week 13 Week 14 Week 15 Week 16 Week 17 Week 18 Week 1 Week 2 Week 3 Week 4 Week 5 Week 6 Week 7 Week 8 Week 9 Week 10 Week 11 Week 12 Week 13 Week 14 Week 15 Week 16 Week 17 Week 18
Average 2017-19 Large Firms 2020 Large Firms Average 2017-19 High Skill 2020 High Skill
Average 2017-19 Small Firms 2020 Small Firms Average 2017-19 Low Skill 2020 Low Skill
(A) Postings by Firm Size (B) Postings by Job Skill Levels
Figure 2. Weekly Job Postings by Firm Size and Job Skill Levels. This figure plots the 7-day rolling
average of active job postings across the first 18 weeks of 2017–2019, contrasting it to the job postings over
the same period in 2020. Data are partitioned by firm size (Panel A) and job skills (Panel B). Large firms
are public firms in the top tercile of the asset distribution in the preceding quarter and small firms are those
in the bottom tercile. High-skill jobs are those whose O*NET codes map to Job Zone 5 and low-skill jobs are
those whose O*NET codes map to Job Zone 1 (1–5 Job Zone scale). The series are indexed such that they
take the value of 100 for the first week of January (on the left y-axis).
Our analyses go deeper into the granular nature of hiring cuts by way of assembling a firm-
ZIP-week panel of job posting activity. As we detail below, our empirical testing simultaneously
accounts for unobserved heterogeneity at the firm, ZIP, and week levels, also accounting for
various local labor market characteristics. Under this setup, we contrast and compare hiring
changes within firms and geographical areas over time. Our baseline tests show that firms
cut their new job postings in local labor markets on a weekly basis by 9% of the 2017–2019
average level with the onset of the pandemic. This estimate translates to a 57% (=1–(1–0.09)9 )
cumulative 9-week decline. Panel A of Figure 2 points to firm size playing an important role
in modulating firms’ reactions to the pandemic. Our regression analyses confirm this, with
results showing that small firms cut job postings significantly more than large firms. In effect,
small firms reduced their weekly new postings by a striking 59% of the 2017–2019 average level
more than their larger counterparts.
We further characterize the nature of hiring cuts by categorizing job postings according to
skill levels. Within-firm analyses show that hiring cuts are more pronounced at the high end of
the worker-skill spectrum. In the first nine weeks of the pandemic, the ratio of high-to-low-skills
3
new job postings declined by 5 percentage points; one-sixth of the ratio for the same time win-
dow in 2017–2019. This result is in contrast to literature pointing to new-hire upskilling in the
aftermath of the Financial Crisis (e.g., Hershbein and Kahn (2018)). It is, however, consistent
with reports of accelerated hiring into low-skill occupations during the pandemic and research
showing association between unemployment and downskilling (Modestino et al. (2016)).2
We verify that COVID-19 is dynamically driving our results by conditioning our tests on
local-area exposure to the spread of the coronavirus. Location–time-specific estimations show
that declines in new job postings are progressively more pronounced in labor markets that
became more affected by the COVID-19 contagion. As the virus spreads, firms cut weekly job
postings by 10% of the 2017–2019 average in counties at the top of the weekly distribution of
COVID-19 cases relative to those at the bottom. Geographical analyses, too, reveal a dramatic,
ongoing curtailment in new hiring in areas that became most exposed to COVID-19.
Our study also shows how various labor market characteristics modulate firms’ responses to
the pandemic. We consider the role played by unionization and find that declines in job post-
ings are more acute for highly unionized jobs. This is consistent with unionization translating
into higher ex ante labor adjustment costs. We also show that firms in the non-tradable sector
cut their job postings the most. This reflects the fact that this sector has been most affected
by restrictions on in-person economic activity imposed since COVID-19 began spreading.
The next feature we study is local labor market depth, or the extent to which hiring is
concentrated in the hands of a small number of local employers. Employers in a competitive
labor market may not reduce their hiring as much as those in more concentrated markets do.
This is because employers in a competitive labor market face the risk of being unable to re-
hire their workers when conditions improve. In line with this argument, within-firm estimates
show that firms cut their job postings more aggressively in less competitive labor markets.
Subsequently, we examine the role of firm geographic concentration. Controlling for size, firms
whose operations are widely diversified across different areas of the country reduce their hiring
less pronouncedly than firms whose operations are concentrated. Geographical diversification
2
New York Times, March 22nd , 2020, “Help Wanted: Grocery Stores, Pizza Chains and Amazon Are Hiring.”
4
appears to enhance firms’ ability to withstand negative shocks to their workforce (see also
Giroud and Mueller (2019)).
Our subsequent set of analyses investigates the role played by credit access. Credit con-
straints shaped firms’ hiring decisions in the aftermath of the Financial Crisis (Campello et al.
(2010) and Chodorow-Reich (2014)). While the COVID-19 crisis originated outside the finan-
cial system, the availability of financing has been viewed as an important buffer for firms hit by
the pandemic (Granja et al. (2020) and Fahlenbrach et al. (2020)). Accordingly, economic pol-
icy responses such as the Paycheck Protection Program (PPP) and other elements of the Coro-
navirus Aid, Relief, and Economic Security (CARES) Act have targeted firms with infusions of
capital. We consider a number of metrics capturing firms’ ex ante access to financing, such as
their private or public status, credit ratings, access to outstanding credit lines, and cash hold-
ings. Across all proxies, we find that credit constraints intensify the cuts in job postings. For
instance, firms without bank credit lines cut their weekly job postings by 13% of the 2017–2019
average, relative to firms with at least one credit line available. Firms in the latter group have
between 2 and 3 credit lines outstanding, on average, representing up to $780 million in avail-
able facilities (24% of their total assets). In a final set of tests, we evaluate the effectiveness of
the PPP by comparing the hiring decisions of public firms receiving funding under the program
with those of a matched group of control firms. We show that PPP recipient firms cut their job
postings by more than other firms in the days after receiving funding. Our findings highlight
side effects of stimulus policies designed to preserve existing employment on firms’ new hiring.
A granular examination of the location of job postings indicates that the decline in high-skill
postings is particularly pronounced in (often-depressed) rural and exurban areas of the U.S. In
addition, we find that firms cut back on job postings the most in low-income areas and areas
with greater income inequality. The patterns we observe raise concerns about whether jobs
lost to the pandemic are likely to return even when overall economic conditions improve. One
of the likely consequences of COVID-19 is that of aggravating regional economic inequalities.
Our study contributes to a growing and important body of work on the economic impact
of COVID-19 by providing a granular analysis of firm-specific recruitment activities during the
5
pandemic.3 It also contributes to the understanding of how the pandemic may affect recent
developments in labor markets, including upskilling and downskilling (Autor and Dorn (2013),
Modestino et al. (2016), Hershbein and Kahn (2018), and Campello et al. (2020)), job polariza-
tion (Autor (2014) and Jaimovich and Siu (2020)), as well as increasing market concentration
(Azar et al. (2017) and Benmelech et al. (2018)). Our results carry policy implications. They
suggest that economic recovery may be hindered by the fact that hiring cuts have been par-
ticularly severe in concentrated local labor markets, among high-skill jobs, and across smaller
firms with limited access to capital. The pandemic also brought about particularly deleterious
effects to the hiring of workers in poorer areas and places where income inequality was already
high. Economic stimuli focusing on ameliorating the impact of COVID-19 should consider
these labor market dynamics.
2 Data and Empirical Methodology
2.1 Job Postings Data
The core of our data is obtained from LinkUp, a leading provider of job market data and
analytics. LinkUp assembles a comprehensive database of job openings sourced directly from
over 50,000 employers, starting from 2007. These data are continuously updated by crawling
company websites, capturing information on, among other things, job posting creation, modifi-
cation, and deletion dates. The data for each posting also contain information on the job title,
firm identifier, and geographical tracking to the ZIP code level. LinkUp attributes an O*NET
occupation code to each posting based on a natural language processing algorithm. LinkUp has
made available to us their entire database consisting of raw records and other processed fields.
3
Examples include Alfaro et al. (2020), Ding et al. (2020), and Ramelli and Wagner (2020) on stock
market reactions to COVID-19, Coibion et al. (2020) who survey household labor force participation, Baker
et al. (2020a) on uncertainty surrounding the pandemic, Hassan et al. (2020) on measuring firm exposures to
COVID-19, Bartik et al. (2020) and Granja et al. (2020) on small business responses, Baker et al. (2020b) on
household spending reactions to the pandemic, Kahn et al. (2020) on the aggregate decline in job vacancies
and spike in unemployment insurance claims, and Cajner et al.’s (2020) evidence on employment contraction
based on firm-anonymized payroll records.
6
Our sampling runs from January 1st , 2017 through May 5th , 2020. We restrict our attention
to American firms and job postings. We use LinkUp’s linking tables to map each firm’s inter-
nal identifier to its ticker, NAICS industry code, and then to its Compustat GVKEY. We link
these data to other firm-level data sources using tickers and GVKEYs. To gauge the required
skill level of a job posting, we map the posting’s O*NET code to a Job Skill Zone (1 to 5 scale)
based on the O*NET Skill Zone linking table.4
2.2 Other Data Sources
Our analysis uses additional data on firm fundamentals and operations, labor markets, credit
conditions, and various geography-level information. We obtain firm financial data from Com-
pustat’s Quarterly and Annual files. For information on the geographical location of firms’
operations, we use the Your-economy Time-Series (YTS) database, maintained by the Busi-
ness Dynamics Research Consortium at the University of Wisconsin. The YTS database is
compiled from Infogroup’s historical business files and are linked longitudinally to track loca-
tion, employment, and sales information at the establishment-year level for public and private
firms. Information on firms’ credit ratings come from Compustat’s Ratings files. Data on
outstanding credit lines are from WRDS-Reuters DealScan. Data on unionization are from the
Bureau of Labor Statistics (BLS). County-level data on income inequality are obtained from
the U.S. Census Bureau’s American Community Survey (ACS). Monthly state-level unemploy-
ment and labor force figures are obtained from the BLS. We obtain statistics on daily recorded
COVID-19 cases in each county from the New York Times.
2.3 Variable Construction and Measurement
2.3.1 Job Postings and Skills
Our base data come from 26,414 firms (both public and private). We collapse the job posting-
level data into a firm–week–ZIP code panel, consisting of 49,385,544 observations. We compute
4
The O*NET classification of Job Skill Zones is based upon the Specific Vocational Preparation required
for an occupation as per the Dictionary of Occupational Titles (see Autor et al. (2003) and Donangelo (2014)).
7
our tests’ dependent variables using this panel as follows. N ew Job P ostings is the logarithm
of one plus the total number of new job postings created by a given firm in a given week in a
given 3-digit ZIP code.5 As an alternative metric of corporate hiring, we also track the number
of active job postings maintained by a firm in a week in a ZIP code. Using the number of
active job postings, we compute ∆Active Job P ostings as the percentage change in job post-
ings (relative to the same firm, same ZIP, and same week in the previous year). The benefit
of ∆Active Job P ostings is that it incorporates both the creation of new job postings and the
deletion of postings.
Our next set of dependent variables gauges heterogeneity in the skill level of job postings.
∆Low Skill P ostings is the percentage change in job postings (relative to the same firm, same
ZIP, and same week in the previous year) for occupations with O*NET codes corresponding
to Job Zone 1. ∆High Skill P ostings is measured analogously for occupations with O*NET
codes corresponding to Job Zone 5. The total number of low-skill job postings created in our
sample is 2,355,279, while the number of high-skill postings is 3,127,187.6 Additionally, we
compute the High-to-Low-Skills Postings Ratio as the number of job postings created in Job
Zone 5 divided by the number of job postings created in Job Zone 1 for a given firm–week–
ZIP triple. Through this ratio, we can measure whether hiring activity within a firm and labor
market is skewed towards low-skill (if the ratio is <1) or high-skill (if the ratio is >1) positions.
2.3.2 Conditioning Variables
Several dynamics may modulate firm responses to the COVID-19 pandemic, and we proxy
for these forces using a number of conditioning variables. The first variable captures the
intensity of the pandemic at a local-area level; that is, the locality where a firm seeks to
hire. High COV ID Exposure is an indicator that takes the value of 1 for each county-week
5
We follow Chetty et al. (2013) in defining the relevant area boundaries of our geographical analysis. There
are 899 3-digit ZIP codes in the U.S. and they provide for more granular mapping than commuting zones (709)
or MSAs (392), yet allow for more precise estimations than 5-digit ZIP codes (oftentimes arbitrarily assigned
to large buildings or university campuses). The average (median) population of a 3-Digit ZIP code is 349,490
(212,964) based on the 2010 U.S. Census.
6
See Appendix A for the complete listing of occupations included under the low- and high-skill categories.
Our results are robust to alternate definitions of low- and high-skill jobs, including defining low-skill (high-skill)
jobs as those in as Job Zones 1 and 2 (4 and 5).
8
(A) Week 8 (February 25th , 2020) (B) Week 10 (March 10th , 2020)
(C) Week 12 (March 24th , 2020) (D) Week 14 (April 7th , 2020)
(E) Week 16 (April 21st , 2020) (F) Week 18 (May 5th , 2020)
Figure 3. COVID-19 Exposure. This figure illustrates the distribution of cumulative COVID-19 cases per
capita at the ZIP code level at two-week intervals, beginning on Week 8 (Panel A, February 25th ) and ending
on Week 18 (Panel F, May 5th ) of 2020. ZIP codes belonging to the highest tercile of cumulative confirmed
COVID-19 cases per capita are colored in the darkest shade, ZIP codes belonging to the second tercile are
colored in the intermediate shade, and ZIP codes in the lowest tercile are colored in the lightest shade.
belonging to the highest tercile of the number of confirmed COVID-19 cases per capita in
the United States and 0 for the lowest tercile.7 Figure 3 depicts the time-series evolution of
the exposure of counties based on total recorded number of COVID-19 cases per capita since
7
We map ZIP codes to counties using the HUD-USPS ZIP Code Crosswalk. While we partition and rank
areas into terciles for convenience, we demonstrate our results are robust to conditioning on various alternative
cutoff points along the COVID-19 case distribution, including quartiles, quintiles, and deciles.
9
February 25th , 2020. In Week 8 of 2020 (Panel A) there was relatively little human exposure to
the coronavirus, with only a handful of cases reported across the country. By Week 10 (Panel
B), there were noticeable outbreaks on the West Coast (Seattle and San Francisco Bay Area)
and evidence of community contagion in cities throughout the U.S. By Week 12 (Panel C), the
coronavirus outbreak appeared to have spread across the country, notably in the Northeast
(New York, New Jersey, and Massachusetts). Exposure as of Week 18 (Panel F) shows the
widespread presence of COVID-19 cases across the U.S., with urban centers being most affected.
The next set of conditioning variables relates to firm size. They are computed using data
on firm assets and employees for public firms in our sample. Small F irm (Assets) is an
indicator variable that takes the value of 1 for firms in the bottom tercile of the total assets
distribution (measured in December 2019) and 0 for firms in the top tercile. In analogous
fashion, Small F irm (Employees) is an indicator that takes the value of 1 for firms in the
bottom tercile of total employees and 0 for firms in the top tercile.
We also condition our tests on variables capturing local labor market characteristics and
firm geographical diversification. High U nionization is an indicator variable that takes the
value of 1 for firms in the top tercile of the labor unionization rate in 2019 (at the 4-digit SIC
industry level) and 0 for firms in the bottom tercile.8 Non-Tradables is an indicator that takes
the value of 1 for firms in the non-tradable sector and 0 for firms in the tradable sector (cf. Mian
and Sufi (2014)).9 Low Local Labor M arket Depth takes the value of 1 for ZIP codes in the
top tercile of the Herfindahl–Hirschman Index (HHI) of active job postings and 0 for ZIP codes
in the bottom tercile. The HHI is calculated within a ZIP code across all employers with active
job postings in that ZIP code for the year 2019. High F irm Geographic Concentration takes
the value of 1 for firms in the top tercile of the HHI of their operations and 0 for firms in the
bottom tercile. The HHI of a firm’s operations is calculated by taking the sum of the squared
employment shares across all ZIP codes a firm operates in, based on 2018 establishment-level
data from YTS. Low Local Household Income is an indicator that takes the value of 1 for
8
Examples of highly unionized industries include airlines, shipping, telecommunications, healthcare, steel,
coal, and motion pictures.
9
Examples of the non-tradable sector businesses include supermarkets, restaurants, office supplies, car
dealerships, food service, and clothing.
10
counties in the bottom tercile of the median household income distribution as of 2018 and 0
for counties in the top tercile. High Local Income Inequality is an indicator that takes the
value of 1 for counties in the top tercile of the income inequality distribution (measured by the
2018 5-year Gini coefficient) and 0 for counties in the bottom tercile of that distribution.
The final set of conditioning variables captures a firm’s ability to raise financing. Private
Firm is an indicator variable that takes the value of 1 for private firms and 0 for public firms.
Speculative Grade is an indicator that takes the value of 1 for firms with an S&P issuer rating
of less than BBB– (or unrated) as of 2019 and 0 otherwise. N o Credit Lines is an indicator
that takes the value of 1 for firms with no active lines of credit and 0 for firms with at least
one line of credit as of the end of 2019. Low Cash Holdings is an indicator variable that takes
the value of 1 for firms in the lowest tercile of the corporate cash-to-asset distribution and 0
for firms in the highest tercile of cash (measured in December 2019).
2.3.3 Control Variables
We account for several additional variables that are likely to influence firm hiring. At the state-
month level, we control for the logarithm of the total labor force and the unemployment rate.
At the firm-quarter level, we control for the logarithm of total assets, profitability (net income
divided by lagged assets), cash (divided by lagged assets), financial leverage (total short- and
long-term debt divided by lagged assets), Q (market-to-book ratio of assets), and investment
(capital expenditures divided by lagged assets); all measured using pre-COVID-19 data.
2.4 Data Coverage and Validation
It is important to verify the quality of our job postings data both in terms of geographical
representativeness and correspondence with overall job creation in the economy. Figure 4
showcases the geographical coverage of the LinkUp data. It does so using pre-COVID-19 data
from years 2017 through 2019. Panels A and B depict job posting activity by large and small
firms. Based on these panels, it appears that hiring across small and large firms were similarly
11
(A) Postings by Small Firms (B) Postings by Large Firms
(C) Low-Skill Postings (D) High-Skill Postings
Figure 4. Geographical Distribution of Job Postings (2017–2019). This figure depicts the distribution
of the number of active job postings at the ZIP code level, averaged over 2017–2019. Panel A (B) shows total
active postings by small (large) firms defined as firms in the bottom (top) tercile of total assets. Panel C (D)
shows total active postings for low (high) skill jobs defined as jobs whose O*NET codes map to Job Zone 1 (5).
geographically distributed prior to the 2020 pandemic. High-skill job postings (Panel D) seemed
more prevalent in urban centers and on the coasts, compared to low-skill job postings (Panel C).
We compare our job postings data with administrative data on employment in Figure 5.
Panel A shows that the total number of job postings in LinkUp consistently captures around
50% of total private-sector hires in the BLS Job Openings and Labor Turnover Survey (JOLTS).
Panel B shows the relation between total new postings calculated from LinkUp data and firm
job gains from the U.S. Census Bureau’s Quarterly Workforce Indicators (QWI) data. The
plot suggests a close link between job posting activity and job gains recorded at firms. The
LinkUp data appear to provide a reasonable representation of corporate hiring.
12
11
Postings and Hires (Millions)
New Postings (LinkUp)
1 2 3 4 5 6
8 9 10
7
0 aaaaa
9 10 11 12 13
Jan 2017
Feb 2017
Mar 2017
Apr 2017
May 2017
Jun 2017
Jul 2017
Aug 2017
Sep 2017
Oct 2017
Nov 2017
Dec 2017
Jan 2018
Feb 2018
Firm Job Gains (QWI)
Mar 2018
Apr 2018
May 2018
Jun 2018
Jul 2018
Aug 2018
Sep 2018
Oct 2018
Nov 2018
Dec 2018
Jan 2019
Feb 2019
Mar 2019
Apr 2019
May 2019
Jun 2019
Jul 2019
Aug 2019
Sep 2019
Oct 2019
Nov 2019
Dec 2019
Total Job Postings (LinkUp) Total Hires (JOLTS)
(A) JOLTS (B) QWI
Figure 5. Data Validation. Panel A depicts total job postings (from LinkUp data) and total private-sector
hires (seasonally adjusted) from the BLS Job Openings and Labor Turnover Survey (JOLTS). Panel B plots
the average state-level relation between total new postings (from LinkUp data) and firm job gains from the
U.S. Census Bureau’s Quarterly Workforce Indicators (QWI) data. Data in Panel B are in logs and represented
in the form of 20 equal-sized bins based on the cross-sectional distribution of the depicted variables.
2.5 Summary Statistics
Table 1 reports descriptive statistics for the key variables used in our analysis. The average
number of new postings by a firm in a given ZIP per week over the 2017–2020 period is 1.36 (or,
expressed in log terms, 0.31). The average for the pre-COVID-19 period (1.48 postings, or 0.39
in log terms) is higher than the 2020 average of 1.34 (or 0.30 in log terms). The average percent-
age change in active postings by a firm–ZIP–week is 12% over the entire 2017–2020 period. As
with new postings, the average percentage change in active postings is 14% over 2017–2019, con-
sistent with the expansion of economic activity and hiring in the last few years. The average per-
centage change in active postings is a much lower 3.5% in 2020 (this includes the pre-pandemic
months of January and February). Firms in our sample tend to post 2.8 high-skill jobs for every
10 low-skill jobs, reflected in an average High-to-Low-Skills Postings Ratio of 0.28. The reported
summary statistics for firm-level control variables suggest that the public firms in our sample
are representative of the Compustat universe of firms (see, e.g., Barrot and Sauvagnat (2016)).
Table 1 about here.
13
Year-on-Year Change in Job Postings
0
-20
-40
(2020 versus 2017-19)
-60
-80
ica Acc
l E om
q
Re Min uipmmod
pa in e ati
ir & g, nt on
Tr an sp
Fu rni tur MaOil &Mfg
int G .
e a
W ast Te rta eta PrReal nanc s
o M
e M lec tio ls od Est ee
an om n Eq & M uct ate
ag & ui ac Mf
em In pm hi g.
en form en ner
t
Ma& Re ationMfg t y
Am Trus ans
em po na me Sv .
ge dia c.
en rta
t, G tio m ti
C ent on
Co mp am n & Fina hem Svc.
bli W nci ica
ng are al S ls
Pr ofe u
Be ter & ho vc
ssi on al, ve & TRecr using.
r ag Ele ext eati
e & ctr ile on
Sci e
Pe ntific To oni Mfg
ba c M .
rso & na Te Uo Mfg. cc f
Ele Ad l & ch ti g.
n
Pu Pri Launical ties
bli nti dr Sv l i
ctr mi nis sh ng y c.
tra Eding I & P Svc.
tiv uc nd ap
e & ati us er
Am Su ona tries
pp l S
o v
bu lat ory W Ho rt Svc.
ho spit c.
He Ret lesa als
al ai le
Nu rsi W Sociath C l Tra rs
oo l A are de
d
ng & Pr ssis Svc
od ta .
uc nc
Re sid tM e
en Fofg.
Ag tial C od
Co ricu are
ns ltu
tru re
cti on
Figure 6. Industry Distribution of Post-COVID Declines in Job Postings. This figure plots
the industry distribution of the cumulative percentage change in the number of active job postings for the
post-COVID period (relative to the average number of active job postings in the same weeks of 2017–2019).
Firms are assigned to industries based on their 3-digit NAICS codes (see Appendix B).
We showcase the existence of business-sector heterogeneity in the way COVID-19 affects
the economy in Figure 6. Firms in the accommodation and electrical equipment manufactur-
ing industries posted the greatest decline in hiring activity, nearly 90%. This is almost five
times as large as that of the least affected industries. Industries in the latter category include
construction, agriculture, and nursing & residential care facilities, whose services and goods
have been deemed essential, and consequently have been in high demand since the onset of
the pandemic. This variation suggests the inclusion of firm-by-time–fixed effects (subsuming
industry-by-time–fixed effects) in our analysis, thereby alleviating concerns that our results
may be driven by industry dynamics.
2.6 Empirical Specification
As a baseline, we empirically estimate a model that relates firms’ job postings with a time
indicator variable that captures the onset of the pandemic. We additionally interact that
time indicator with several conditioning variables, while controlling for other drivers of firms’
14
postings. Our specification takes the following form:
Yi,j,t = β1 P ost COV IDt + β2 Xi,j,t + β3 P ost COV IDt × Xi,j,t (1)
+ θControls i,j,t + F Es + i,j,t ,
where Yi,j,t ∈ {New Postings, ∆Active Postings, ∆Low-Skill Postings, ∆High-Skill Postings,
High-to-Low-Skills Postings Ratio} for firm i in ZIP code j in week t. P ost COV ID is a
dichotomous variable that takes the value of 1 for each week after February 29th , 2020 and 0
otherwise. Xi,j,t ∈ {High COVID Exposure, Small Firm (Assets), Small Firm (Employees),
High Unionization, Non-Tradables, Low Local Labor Market Depth, High Firm Geographic
Concentration, Low Local Household Income, High Local Income Inequality, Private Firm,
Speculative Grade, No Credit Lines, Low Cash Holdings} refers to a relevant conditioning
variable. Controls is a vector of variables described in Section 2.3.3. Our baseline specifi-
cation accounts for unobserved heterogeneity with the inclusion of dynamic firm–, ZIP–, and
week–fixed effects. Standard errors are triple–clustered by firm, ZIP, and week.10
3 Base Results
3.1 Job Posting Activity
We first estimate Eq. (1) without conditioning variables to gauge the baseline impact of
COVID-19 on job postings. We then assess whether firm responses are heightened in areas
with more severe exposure to COVID-19, as the pandemic spreads. The results are reported
in Table 2.
Table 2 about here.
10
In light of the large number of observations in our sample, we adopt a stringent, multi-dimensional
clustering scheme. It simultaneously allows for arbitrary cross-correlation of standard errors within the firm,
geographical, and time dimensions. Our estimated t-statistics would be larger by one order of magnitude
otherwise (see Cameron et al. (2011)).
15
The estimate in column (1) points to the pandemic imposing a negative and highly sig-
nificant toll on firm hiring. The economic magnitude is striking. The coefficient of –0.035
implies that firms cut their average weekly postings in a ZIP code area by 9% (= 0.035 / 0.39)
of the 2017–2019 average, following the start of the pandemic. The magnitude of this effect
increases as we condition on the local level of exposure to the coronavirus contagion. In par-
ticular, the estimate in column (2) implies that firms have been curtailing their weekly new
job posting activity within areas with highest levels of confirmed COVID-19 cases by 1.14
(= −0.040 / − 0.035) times the unconditional effect, relative to those areas with fewer cases.11
The next metric of corporate hiring activity that we consider is the change in active job
postings. In particular, the percentage change in active postings maintained by a firm in a ZIP
code, relative to the same week in the prior year (this reflects both the addition and deletion of
postings). The results in columns (3) and (4) are statistically significant and consistent in sign
with those reported in columns (1) and (2). Recall, the average change in active postings over
the last 3 years was 14%. The coefficient in column (3) indicates that in the weeks following
February 29th , 2020, this rate has declined by 14.7 percentage points. In effect, COVID-19
reversed the growth in active job postings observed over the 2017–2019 period.
3.2 Worker Skill
Next, we assess whether the quality of human capital that firms seek to hire has been affected
by the pandemic. We do so by comparing changes in active postings for high-skill positions
relative to low-skill positions by the same firm in the same locality over time. The results
obtained from estimating Eq. (1) using ∆Low-Skill Postings, ∆High-Skill Postings, and High-
to-Low-Skills Postings Ratio as dependent variables are reported in Table 3.
Table 3 about here.
The estimates in columns (1) through (4) show a decline in both low- and high-skill job
postings. Critically, the decline is more pronounced among high-skill postings. In columns (5)
11
The uninteracted P ost COV ID and High COV ID Exposure regression terms are subsumed by dynamic
fixed effects.
16
and (6) we report the effects on the High-to-Low-Skills Postings Ratio. That ratio declines sig-
nificantly as the pandemic spreads. The number of high-skill postings made by a firm in a given
ZIP–week declines disproportionately relative to the number of low-skill postings previously
made by the same firm in that same ZIP–week. The drop of –0.051 (column (5)) in this ratio
represents 17% (= 0.051 / 0.30) of the 2017–2019 average High-to-Low-Skills Postings Ratio.12
Likewise, in the regions most exposed to COVID-19, this ratio declines by 21.3% of the pre-
COVID-19 average. Our tests account for the local unemployment rate, and imply that the pan-
demic drives a spike in downskilling over and above the previously reported association between
the two in the U.S. (see Modestino et al. (2016)). These results provide unique insights into the
emergence of downskilling dynamics in local labor markets under COVID-19. They signify a re-
versal of the upskilling trend observed since the Financial Crisis (Hershbein and Kahn (2018)).
3.3 Firm Size
Our analysis considers several characteristics that modulate firms’ hiring responses. The first
dimension we study is firm size. The stimulus programs enacted in response to the pandemic,
such as the Paycheck Protection Program (PPP), have focused on providing support to small
businesses. These policies are predicated on the notion that small businesses have been the
hardest hit. We verify that this is the case by re-running the baseline specification in Eq.
(1) conditioning on firm size. Our classification of small firms is based on balance sheet data
reported by public firms, which tend to be larger than private firms. Our results would likely
form the upper bound for the universe of small private firms, which are likely to be cutting
back on their hiring even more than our estimates imply. The results are reported in Table 4.
Table 4 about here.
Regardless of whether firms are classified based on assets or employees, small firms cut their
job posting activity substantially (relative to large firms) following the onset of the pandemic.
The decline in new postings of –0.231 (see column (1)) is 59.2% of the 2017–2019 average
12
See Table C.1 for evidence that our inferences are robust to alternative definitions of high- and low-skill jobs.
Table C.2 shows that they are robust to considering alternative cutoffs along the COVID-19 case distribution.
17
weekly job posting rate of 0.39. Notably, small firms appear to skew their job postings away
from high-skill jobs relative to large firms under the pandemic (columns (3) and (4)). Our
findings show that small firms are bearing the brunt of COVID-19, justifying the targeting of
economic stimulus towards such firms as a way to minimize the loss of jobs in the economy.
3.4 Mapping Firm Hiring Responses
We provide geographical context for the results in Tables 3 and 4 by plotting the post-COVID
cuts in job postings by location in Figure 7. Across all figure panels, we statistically reject
the null of spatial randomization (p < 0.01) based on Moran’s I test statistics. A comparison
between Panels A and B highlights that small firms have more acutely reduced their job post-
ings relative to large firms.13 The disproportionate cuts in high-skill job postings relative to
low-skill job postings is evident from Panels C and D.
Our geography-based approach is particularly useful in illustrating the interactive effects
of firm size and job skills on the extent of COVID-19-induced hiring cuts. Panel E shows rel-
atively mild declines in low-skill postings by large firms across the country, with some regions
experiencing growth.14 Panel F paints a dramatically different picture in showing that small
firms have made deep cuts to their hiring for high-skill positions. Notably, cuts in high-skill
postings appear to be most severe in exurban and rural areas. This heightens concerns about
the economic impact of COVID-19 on these particular regions, and their potential for recov-
ery. These labor markets were unlikely to have been particularly active in the pre-pandemic
period. With the spread of COVID-19, they run the risk of experiencing a large contraction
in high-skill hiring as small local firms are disproportionately impacted by the pandemic.
13
This happens despite the geographical distribution of job postings being virtually identical across small
and large firms before the pandemic (see Panels A and B of Figure 4).
14
This is consistent with widespread reports of certain large firms continuing to hire into low-skill positions.
18
(A) Postings by Large Firms (B) Postings by Small Firms
(C) Low-Skill Postings (D) High-Skill Postings
(E) Low-Skill Postings by Large Firms (F) High-Skill Postings by Small Firms
Figure 7. Geographical Distribution of Post-COVID Declines in Job Postings. This figure depicts
the distribution of the change in the number of job postings at the ZIP code level for the post-COVID period
(relative to the average number of job postings in the same weeks of 2017–2019). Panel A (B) shows the
percentage change in postings by large (small) firms defined as firms in the top (bottom) tercile of total assets.
Panel C (D) shows the percentage change in postings for low (high) skill jobs defined as jobs whose O*NET
codes map to Job Zone 1 (5). Panel E shows the percentage change in postings for low-skill jobs by large
firms, while Panel F shows the percentage change in postings for high-skill jobs by small firms.
19
4 Labor Markets, Credit Access, and Income Distribution
4.1 Labor Market and Geographical Characteristics
In our next set of tests, we estimate the baseline specification in Eq. (1) conditioning on labor
market and geographical characteristics that are likely to shape hiring. The characteristics we
consider are unionization, tradability, labor market depth, and geographic concentration. The
results are reported in Table 5.
Table 5 about here.
Firms in highly unionized industries curtail their new postings more than those in less
unionized industries under the pandemic (see column (1)). This is consistent with such firms
facing higher labor adjustment costs. Accordingly, they may prefer adopting a “wait-and-see”
approach before entering into rigid labor contracts during uncertain economic conditions. The
coefficient estimate of –0.022 implies that firms in highly unionized industries cut their job
postings by 5.6% of the 2017–2019 new posting creation rate more than firms in less unionized
industries. Similarly, firms operating in the non-tradable sector cut their job posting activity
more than firms operating in the tradable sector (column (2)). Non-tradable firms are highly
exposed to local economic conditions as they rely more heavily on foot traffic, which has de-
clined precariously with the pandemic. These firms also disproportionately reduce their hiring
into high-skill positions relative to low-skill positions (see columns (5) and (6)).
We next consider labor market depth, measured as the concentration of employers in a
local labor market. The logic underlying this test is that labor markets dominated by a small
set of employers may experience more substantial cuts in hiring in bad times. This is because
dominant local employers run a lower risk of having to compete with other employers when
conditions improve.15 Results in columns (3) and (7) of Table 5 show that concentrated labor
markets experience greater declines in new job postings, particularly high-skill postings, with
COVID-19. These results are particularly striking as they are estimated within firms, that is,
15
For theoretical reviews on labor market concentration, see Boal and Ransom (1997), Bhaskar et al. (2002),
and Manning (2011), and for empirical evidence see Azar et al. (2017) and Benmelech et al. (2018).
20
they compare the new job posting intensity by the same firm across local labor markets that
are more versus less concentrated. This phenomenon is likely to be of concern to policymakers.
Such markets may become even more concentrated in the future as marginal employers may
not survive the downturn, leading to a potentially slower labor market recovery.
Finally, we look at firm geographic concentration. The estimates in columns (4) and (8) of
Table 5 suggest that (controlling for firm size) firms whose footprints are concentrated in fewer
regions cut back on their hiring more than firms with geographically dispersed operations.
Geographical dispersion appears to confer an advantage to firms in withstanding the current
pandemic-led crisis.
We provide economic interpretation and dynamics for the results in Table 5 by plotting
various indicators of job posting activity in Figure 8. Panel A shows that firms in highly
unionized industries cut their active job postings substantially more, and earlier, than firms in
low unionization industries. Panel B shows similar patterns of hiring cuts among firms in the
non-tradable sector relative to the tradable sector. The disproportionate decline in job postings
in the non-tradable sector points to this sector being relatively hard hit as the pandemic spreads
and social distancing orders are put in place (hiring cuts are higher starting from mid-March).
Panel C shows that local labor markets with “lower depth” (more concentrated) have experi-
enced cuts in job postings that are 5 percentage points higher than labor markets that are less
concentrated. Panel D stresses that firms with more concentrated operations reduce their active
job postings by almost twice as much as do firms with a more diversified geographic footprint.
4.2 Access to Finance and Credit
Our subsequent set of analyses concerns firms’ access to financing. We condition our baseline
specification in Eq. (1) on four proxies of financing constraints: a firm’s private or public status,
whether it is rated as speculative or investment grade, whether it has an outstanding credit
line to tap into, and whether it holds a large cash buffer. The results are presented in Table 6.
Table 6 about here.
21
10 10
Year-on-Year Change in Job Postings Year-on-Year Change in Job Postings
0 0
-60 -50 -40 -30 -20 -10
-10
-20
-30
(2020 versus 2017-19) (2020 versus 2017-19)
-40
-50
-60
Week 1 Week 2 Week 3 Week 4 Week 5 Week 6 Week 7 Week 8 Week 9 Week 10 Week 11 Week 12 Week 13 Week 14 Week 15 Week 16 Week 17 Week 18 Week 1 Week 2 Week 3 Week 4 Week 5 Week 6 Week 7 Week 8 Week 9 Week 10 Week 11 Week 12 Week 13 Week 14 Week 15 Week 16 Week 17 Week 18
Low Unionization High Unionization Tradables Non-Tradables
(A) Postings by Low versus High Unionization (B) Postings by Tradables versus Non-Tradables
10 10
Year-on-Year Change in Job Postings Year-on-Year Change in Job Postings
0 0
-60 -50 -40 -30 -20 -10
-10
-20
-30
(2020 versus 2017-19) (2020 versus 2017-19)
-40
-50
-60
Week 1 Week 2 Week 3 Week 4 Week 5 Week 6 Week 7 Week 8 Week 9 Week 10 Week 11 Week 12 Week 13 Week 14 Week 15 Week 16 Week 17 Week 18 Week 1 Week 2 Week 3 Week 4 Week 5 Week 6 Week 7 Week 8 Week 9 Week 10 Week 11 Week 12 Week 13 Week 14 Week 15 Week 16 Week 17 Week 18
High Depth Low Depth Low Concentration High Concentration
(C) Postings by Local Labor Market Depth (D) Postings by Firm Geographic Concentration
Figure 8. Job Posting Dynamics by Labor Market and Geographical Characteristics. This figure
depicts the prior 7-day rolling average percentage change in total active postings on each day of 2020 relative
to the same day averaged across 2017–2019. Panel A shows the changes in postings for firms in low versus
high unionization industries (based upon BLS data). Panel B shows the changes in postings for firms in the
tradable and non-tradable sectors as per the classification in Mian and Sufi (2014). Panel C shows the change
in postings for ZIP codes in the lowest versus highest tercile of local labor market depth. Panel D shows the
change in postings for firms in the highest versus lowest tercile of geographical concentration of operations.
Across all four metrics, financially constrained firms cut back on job postings — particularly
postings for high-skill jobs — by more than other firms in the wake of the pandemic. The asso-
ciated economic magnitudes are large. The estimate in column (3) implies that firms without
access to liquidity in the form of credit lines reduced their weekly job postings by 13.1% of the
2017–2019 average as compared to firms with at least one credit line outstanding. The latter set
of firms, on average, had 2.4 facilities outstanding amounting to $780 million of available liquid-
ity. Our results suggest that bank credit lines work as a buffer during the pandemic, preventing
22
more severe declines in hiring. The coefficient in column (4) implies that cash savings perform a
similar function. Firms in the bottom tercile of the cash distribution cut their weekly postings
by 12.6% of the pre-COVID-19 average more than firms with large cash buffers. The cash-
constrained firms had an average of $562 million in cash holdings (1.3% of assets) while firms in
the highest tercile of cash holdings had $5.2 billion in cash reserves (30.5% of assets) at the end
of 2019. Unlike the Financial Crisis of 2008-09, the COVID-19 crisis did not originate in the fi-
nancial system, yet lack of access to financing has substantially hampered firms’ hiring activity.
Among the many policies aimed at mitigating the economic impact of COVID-19, the
Paycheck Protection Program (PPP) administered by the SBA as part of the CARES Act has
garnered substantial attention. The PPP provides loans to small businesses and repayment is
forgiven if these businesses utilize the funds to retain their existing employees. The effectiveness
of the program has been questioned, with reports highlighting that funds have been allocated
towards firms with pre-existing access to external financing (e.g., public firms).16 We conduct a
test of the labor market consequences of the PPP by examining its impact on recipient firms’ job
posting activity. We do so focusing on the subset of public firms receiving funds under the PPP.
Figure 9 tracks job postings by PPP recipient firms around the date on which they sub-
mitted relevant regulatory filings.17 We contrast the job posting activity of these firms with
a control group of public firms in the same industry selected by propensity-score matching
(based on firm characteristics such as size, profitability, cash holdings, and financial leverage).
PPP recipient firms and control firms display statistically indistinguishable hiring trends prior
to filing. Recipient firms show a declining trend in job postings up to ten days prior to filing,
nonetheless. This is indicative of the time lag between these firms receiving funding and having
to file the required regulatory disclosures to notify investors. The differential decline in postings
becomes statistically significant five days post-filing, and persists for at least ten days thereafter.
Notably, loan forgiveness under the PPP is conditional upon recipient firms not laying off their
16
See Wall Street Journal, April 26th , 2020, “At Least 13 Public Companies Give Back $170 Million in
Small-Business Stimulus Money. Others Say They’ll Keep It.” Granja et al. (2020) find that funds from the
PPP have not been disbursed to areas most adversely affected by the pandemic.
17
A total of 140 public firms in our sample submitted regulatory filings (8-Ks) to the SEC as of May
th
5 , 2020 stating they had received PPP funding. Among these, 80 firms have complete data for the days
surrounding the filing date, did not report returning the funds, and were matched to at least one control firm.
23
105
Job Postings (Indexed)
100
95
90
85
-20 -15 -10 -5 0 5 10 15
Days From Filing
Matched-Control Firms PPP Recipient Firms
Figure 9. Paycheck Protection Program and Job Posting Dynamics. This figure illustrates the
7-day rolling average of total active job postings around the dates public firms reported receiving PPP funding
in 8-K filings submitted to the U.S. Securities and Exchange Commission (SEC). Control firms are selected
among public firms in the same industry by estimating propensity scores as a function of firm size, profit, cash
holdings, financial leverage, Q, and investment. For each PPP recipient firm, we identify the nearest-neighbor
non-recipient firm (with replacement) and include it in the control group. Confidence intervals are calculated
as ±1.5 standard deviations from the mean active postings for each day. Postings are indexed to 100 at the
20th day prior to firms reporting receiving PPP funding.
existing workforce. This regulatory-induced reduction in workforce turnover is associated with
an ostensibly unintended decline in new hiring among recipient firms. Our results highlight
the challenges faced by policy makers in promoting job creation in a post-pandemic recovery.
4.3 Income Levels and Inequality
Our final set of tests considers the role played by regional economic characteristics in modu-
lating firms’ hiring responses to the pandemic. We condition our baseline specification in Eq.
(1) on two relevant metrics: household income (county-level median) and income inequality
(county-level 5-year Gini coefficient). The results are reported in Table 7.
Table 7 about here.
Firms appear to cut back on new job postings — particularly high-skill postings — sub-
stantially more in low income regions (columns (1) and (3)) and regions with greater income
disparity (columns (2) and (4)). Figure 10 provides geographical context for these results by
24
(A) Median Household Income Terciles (B) Gini Coefficient Terciles
Figure 10. Distribution of Income Levels and Inequality. This figure illustrates the distribution of
median household income (Panel A) and Gini coefficient (Panel B) as of 2018, based on data from the U.S.
Census Bureau American Community Survey (ACS).
mapping the distribution of income levels and income inequality across the country. Our re-
sults imply that local job markets most negatively affected by COVID-19 lie in the intersection
of low household income and high income inequality. Examples include rural areas of Maine,
Mississippi, New Mexico, Texas, and Northern California, as well as certain urban centers
and exurban areas surrounding New Orleans, Detroit, and St. Louis. The COVID-19-induced
weakening of labor markets in such regions is likely to exacerbate within- and across-regional
inequalities. This could render the eventual recovery even more precarious due to the aggra-
vation of pre-existing disparities.
5 Conclusion
This study provides an early account of the labor market effects of the COVID-19 pandemic
based on real-time, granular, firm-level job posting data. We report sharp declines in corporate
hiring across the board, but with meaningful heterogeneity along the lines of firm size, various
labor market characteristics, and credit access. A particularly concerning trend is that firms are
disproportionately cutting back on high-skill hiring (“downskilling”) and on hiring in areas out-
side the major cities. This is likely to be detrimental to local government revenues in the near
term as reduced hiring into higher wage jobs spells lower income-based tax collections. In the
longer term, this may harm local recovery prospects as the re-hiring costs for high-skill positions
25
will likely be high. This suggests that additional stimulus may be warranted targeting firms
who hire for high-skill positions in rural and exurban areas. Regional disparities may also be
amplified by the abnormal hiring cuts across low-income, high-inequality areas of the country.
While our analyses provide timely insights into the initial impact of the COVID-19 pan-
demic, conditions continue to evolve and may shape some of our findings. Uncertainty still
looms large over the disease’s spread and its impact on the U.S. economy. We plan to con-
tinue tracking the effects of the COVID-19 pandemic on corporate hiring, helping with the
assessment of current and proposed policy interventions and measuring signs of a recovery.
26
References
Alfaro, Laura, Anusha Chari, Andrew Greenland, and Peter Schott, 2020, Aggregate and firm-
level stock returns during pandemics, in real time, Technical report, National Bureau of
Economic Research.
Autor, David, 2014, Polanyi’s paradox and the shape of employment growth, Technical report,
National Bureau of Economic Research.
Autor, David, and David Dorn, 2013, The growth of low-skill service jobs and the polarization
of the US labor market, American Economic Review 103, 1553–97.
Autor, David, Frank Levy, and Richard Murnane, 2003, The skill content of recent technological
change: An empirical exploration, The Quarterly Journal of Economics 118, 1279–1333.
Azar, José, Ioana Marinescu, and Marshall Steinbaum, 2017, Labor market concentration,
Technical report, National Bureau of Economic Research.
Baker, Scott, Nicholas Bloom, Steven Davis, and Stephen Terry, 2020a, Covid-induced eco-
nomic uncertainty, Technical report, National Bureau of Economic Research.
Baker, Scott, Robert Farrokhnia, Steffen Meyer, Michaela Pagel, and Constantine Yannelis,
2020b, How does household spending respond to an epidemic? Consumption during the 2020
Covid-19 pandemic, Technical report, National Bureau of Economic Research.
Barrot, Jean-Noël, and Julien Sauvagnat, 2016, Input specificity and the propagation of id-
iosyncratic shocks in production networks, The Quarterly Journal of Economics 131, 1543–
1592.
Bartik, Alexander, Marianne Bertrand, Zoë Cullen, Edward Glaeser, Michael Luca, and
Christopher Stanton, 2020, How are small businesses adjusting to COVID-19? Early ev-
idence from a survey, Technical report, National Bureau of Economic Research.
Benmelech, Efraim, Nittai Bergman, and Hyunseob Kim, 2018, Strong employers and weak em-
ployees: How does employer concentration affect wages?, Technical report, National Bureau
of Economic Research.
Bhaskar, Venkataraman, Alan Manning, and Ted To, 2002, Oligopsony and monopsonistic
competition in labor markets, Journal of Economic Perspectives 16, 155–174.
Boal, William, and Michael Ransom, 1997, Monopsony in the labor market, Journal of Eco-
nomic Literature 35, 86–112.
Cajner, Tomaz, Leland Dod Crane, Ryan Decker, John Grigsby, Adrian Hamins-Puertolas, Erik
Hurst, Christopher Johann Kurz, and Ahu Yildirmaz, 2020, The US labor market during
the beginning of the pandemic recession, University of Chicago, Becker Friedman Institute
for Economics Working Paper .
Cameron, Colin, Jonah Gelbach, and Douglas Miller, 2011, Robust inference with multiway
clustering, Journal of Business & Economic Statistics 29, 238–249.
Campello, Murillo, Janet Gao, and Qiping Xu, 2020, Personal income taxes and labor down-
skilling: Evidence from 27 million job postings, Working Paper.
27
Campello, Murillo, John Graham, and Campbell Harvey, 2010, The real effects of financial
constraints: Evidence from a financial crisis, Journal of Financial Economics 97, 470–487.
Chetty, Raj, John Friedman, and Emmanuel Saez, 2013, Using differences in knowledge across
neighborhoods to uncover the impacts of the EITC on earnings, American Economic Review
103, 2683–2721.
Chodorow-Reich, Gabriel, 2014, The employment effects of credit market disruptions: Firm-
level evidence from the 2008–9 financial crisis, The Quarterly Journal of Economics 129,
1–59.
Coibion, Olivier, Yuriy Gorodnichenko, and Michael Weber, 2020, Labor markets during the
COVID-19 crisis: A preliminary view, Technical report, National Bureau of Economic Re-
search.
Ding, Wenzhi, Ross Levine, Chen Lin, and Wensi Xie, 2020, Corporate immunity to the
COVID-19 pandemic, Technical report, National Bureau of Economic Research.
Donangelo, Andres, 2014, Labor mobility: Implications for asset pricing, The Journal of Fi-
nance 69, 1321–1346.
Fahlenbrach, Rüdiger, Kevin Rageth, and René Stulz, 2020, How valuable is financial flexibility
when revenue stops? Evidence from the Covid-19 crisis, Fisher College of Business Working
Paper 007.
Giroud, Xavier, and Holger Mueller, 2019, Firms’ internal networks and local economic shocks,
American Economic Review 109, 3617–49.
Granja, Joao, Christos Makridis, Constantine Yannelis, and Eric Zwick, 2020, Did the Pay-
check Protection Program hit the target?, Technical report, National Bureau of Economic
Research.
Hassan, Tarek Alexander, Stephan Hollander, Laurence van Lent, and Ahmed Tahoun, 2020,
Firm-level exposure to epidemic diseases: Covid-19, SARS, and H1N1, Technical report,
National Bureau of Economic Research.
Hershbein, Brad, and Lisa Kahn, 2018, Do recessions accelerate routine-biased technological
change? Evidence from vacancy postings, American Economic Review 108, 1737–72.
Jaimovich, Nir, and Henry Siu, 2020, Job polarization and jobless recoveries, Review of Eco-
nomics and Statistics 102, 129–147.
Kahn, Lisa, Fabian Lange, and David Wiczer, 2020, Labor demand in the time of COVID-
19: Evidence from vacancy postings and UI claims, Technical report, National Bureau of
Economic Research.
Manning, Alan, 2011, Imperfect competition in the labor market, in Handbook of Labor Eco-
nomics, volume 4, 973–1041 (Elsevier).
Mian, Atif, and Amir Sufi, 2014, What explains the 2007–2009 drop in employment?, Econo-
metrica 82, 2197–2223.
Modestino, Alicia, Daniel Shoag, and Joshua Ballance, 2016, Downskilling: Changes in em-
ployer skill requirements over the business cycle, Labour Economics 41, 333–347.
Ramelli, Stefano, and Alexander Wagner, 2020, Feverish stock price reactions to Covid-19,
Working Paper.
28
Table 1. Descriptive Statistics
This table presents descriptive statistics for the main variables used in our empirical analyses over the 2017–
2020 period. The dependent variables are N ew Job P ostings, ∆Active Job P ostings, ∆Low-Skill P ostings,
∆High-Skill P ostings, and High-to-Low-Skills Postings Ratio. The unit of observation is a firm–ZIP–week,
where ZIP is the three-digit ZIP-code of a job posting. N ew Job P ostings is the logarithm of one plus the total
number of job postings created. ∆Active Job P ostings is the percentage change (relative to the same week
in the previous year) in total active job postings. ∆Low-Skill (High-Skill ) P ostings postings is the percentage
change, relative to the same week in the previous year, in job postings for occupations with O*NET occupation
codes corresponding to Job Zone 1 (5). High-to-Low-Skills Postings Ratio is the total number of job postings
created for occupations with O*NET occupation codes corresponding to Job Zone 5 divided by the total
number of job postings created for occupations with O*NET occupation codes corresponding to Job Zone 1.
P ost COV ID is an indicator variable that takes the value of 1 for each week after February 29th , 2020 and
zero otherwise. High COV ID Exposure is an indicator variable that takes the value of 1 for each county-week
belonging to the highest tercile of the number of confirmed COVID cases per capita and 0 for each county-week
belonging to the lowest tercile of the number of confirmed COVID cases per capita. Small F irm (Assets) is
an indicator variable that takes the value of 1 for firms in the bottom tercile of total assets (measured in the
last available quarter) and 0 for firms in the top tercile. Small F irm (Employees) is an indicator variable that
takes the value of 1 for firms in the bottom tercile of total employees (measured in the last available year) and
0 for firms in the top tercile. High U nionization is an indicator variable that takes the value of 1 for firms
in the top tercile of the labor unionization rate (at the industry level) and 0 for firms in the bottom tercile.
Non-Tradables is an indicator variable that takes the value of 1 for firms in the non-tradable sector and 0 for
firms in the tradable sector. Low Local Labor M arket Depth is an indicator variable that takes the value of 1
for ZIP codes in the top tercile of the Herfindahl–Hirschman Index of active job postings and 0 for ZIP codes
in the bottom tercile of the HHI of active job postings. The HHI is calculated within a ZIP across all employers
with active job postings in that ZIP. High F irm Geographic Concentration is an indicator variable that takes
the value of 1 for firms in the top tercile of the HHI of their operations and 0 for firms in the bottom tercile
of the HHI of their operations. The HHI of a firm’s operations is calculated by taking the sum of the squared
employment shares across all ZIP codes a firm operates in. Low Local Household Income is an indicator
variable that takes the value of 1 for counties in the bottom tercile of median household income as of 2018 and
0 for counties in the top tercile. High Local Income Inequality is an indicator variable that takes the value
of 1 for counties in the top tercile of Gini coefficient and 0 for firms in the bottom tercile. P rivate F irm is
an indicator variable that takes the value of 1 for private firms and 0 for public firms. Speculative Grade is
an indicator variable that takes the value of 1 for firms with a speculative grade rating and 0 for firms with
an investment grade rating. N o Credit Lines is an indicator variable that takes the value of 1 for firms with
no outstanding lines of credit and 0 for firms with at least one outstanding line of credit. Low Cash Holdings
is an indicator variable that takes the value of 1 for firms in the lowest tercile of cash (scaled by lagged assets)
and 0 for firms in the highest tercile of cash. State controls are the unemployment rate and the logarithm of
the labor force. Firm controls are the quarterly logarithm of total assets, profitability (net income divided by
lagged assets), cash (divided by lagged assets), leverage (total short- and long-term debt divided by lagged
assets), Q (market-to-book ratio), and investment (capital expenditures divided by lagged assets).
29
Variable N Mean SD Median IQR
Dependent Variables
N ew Job P ostings (Log) 49,385,544 0.31 0.60 0.00 0.69
∆Active Job P ostings 26,928,778 0.12 3.80 –0.24 1.17
∆Low-Skill P ostings 2,692,847 –0.06 2.02 0.00 1.00
∆High-Skill P ostings 3,652,180 –0.20 2.27 –0.67 1.00
High-to-Low-Skills Postings Ratio 567,440 0.28 2.19 0.00 0.00
COVID-19 Exposure
High COV ID Exposure 32,994,411 0.79 0.41 1.00 0.00
Firm Size
Small F irm (Assets) 14,501,393 0.04 0.19 0.00 0.00
Small F irm (Employees) 15,676,115 0.01 0.10 0.00 0.00
Labor Market Characteristics
High U nionization 10,942,649 0.37 0.48 0.00 1.00
Non-Tradables 13,267,100 0.50 0.50 0.00 1.00
Low Local Labor M arket Depth 35,198,194 0.21 0.41 0.00 0.00
High F irm Geographic Concentration 9,603,235 0.50 0.50 0.00 1.00
Low Local Household Income 32,570,048 0.28 0.45 0.00 1.00
High Local Income Inequality 33,817,979 0.67 0.47 1.00 1.00
Financial Constraints Measures
P rivate F irm 49,385,544 0.52 0.50 1.00 1.00
Speculative Grade 20,175,698 0.55 0.50 1.00 1.00
N o Credit Lines 49,385,544 0.64 0.48 1.00 1.00
Low Cash Holdings 10,656,009 0.79 0.41 1.00 0.00
State Controls
State U nemployment Rate (%) 11,404,003 3.93 0.70 4.00 1.00
State Labor F orce (Log) 11,404,003 15.38 0.90 15.40 1.19
Firm Controls
Size 19,853,663 9.11 1.83 9.07 2.54
Cash 19,560,335 0.09 0.13 0.05 0.09
Leverage 19,816,521 0.39 1.08 0.33 0.31
P rof itability 19,702,080 0.02 0.04 0.01 0.02
Q 19,578,570 1.99 1.48 1.78 1.10
Investment 19,552,052 0.03 0.04 0.02 0.03
30
Table 2. The Impact of COVID-19 on Job Postings: Baseline
This table reports output from Eq. (1). The dependent variables are N ew Job P ostings and
∆Active Job P ostings. The unit of observation is a firm–ZIP–week, where ZIP is the three-digit ZIP-
code of a job posting. N ew Job P ostings is the logarithm of one plus the total number of job postings created.
∆Active Job P ostings is the percentage change (relative to the same week in the previous year) in total active
job postings. P ost COV ID is an indicator variable that takes the value of 1 for each week after February 29th ,
2020 and zero otherwise. High COV ID Exposure is an indicator variable that takes the value of 1 for each
county-week belonging to the highest tercile of the number of confirmed COVID cases per capita and 0 for each
county-week belonging to the lowest tercile of the number of confirmed COVID cases per capita. State controls
are the unemployment rate and the logarithm of the labor force. Firm controls are the quarterly logarithm
of total assets, profitability (net income divided by lagged assets), cash (divided by lagged assets), leverage
(total short- and long-term debt divided by lagged assets), Q (market-to-book ratio), and investment (capital
expenditures divided by lagged assets). Firm×quarter–, ZIP–, and week–fixed effects are included as indicated.
All regressions are estimated over a sample of private and public firms over the January 1st , 2017 to May 5th ,
2020 period. Robust standard errors, reported in parentheses, are triple–clustered by firm, ZIP, and week.
N ew Job P ostings ∆Active Job P ostings
(1) (2) (3) (4)
P ost COV ID –0.035*** –0.147***
(0.011) (0.034)
P ost COV ID × High COV ID Exposure –0.040*** –0.155***
(0.013) (0.045)
Controls Yes Yes Yes Yes
Fixed Effects
Firm × Quarter Yes Yes Yes Yes
ZIP Yes Yes Yes Yes
Week No Yes No Yes
Observations 11,387,645 7,841,929 25,989,424 17,458,249
R-squared 0.385 0.386 0.267 0.287
Statistical significance is indicated as follows: *** p <0.01, ** p <0.05, * p <0.1.
31
Table 3. The Impact of COVID-19 on Job Postings: Worker Skills
This table reports output from Eq. (1). The dependent variables are ∆Low-Skill P ostings, ∆High-Skill P ostings, and High-to-Low-Skills Postings Ratio.
The unit of observation is a firm–ZIP–week, where ZIP is the three-digit ZIP-code of a job posting. ∆Low-Skill (High-Skill ) P ostings postings is the
percentage change, relative to the same week in the previous year, in job postings for occupations with O*NET occupation codes corresponding to Job
Zone 1 (5). High-to-Low-Skills Postings Ratio is the total number of job postings created for occupations with O*NET occupation codes corresponding
to Job Zone 5 divided by the total number of job postings created for occupations with O*NET occupation codes corresponding to Job Zone 1.
P ost COV ID is an indicator variable that takes the value of 1 for each week after February 29th , 2020 and zero otherwise. High COV ID Exposure is
an indicator variable that takes the value of 1 for each county-week belonging to the highest tercile of the number of confirmed COVID cases per capita
and 0 for each county-week belonging to the lowest tercile of the number of confirmed COVID cases per capita. State and firm controls are as defined in
Table 2. Firm×quarter–, ZIP–, and week–fixed effects are included as indicated. All regressions are estimated over a sample of private and public firms
over the January 1st , 2017 to May 5th , 2020 period. Robust standard errors, reported in parentheses, are triple–clustered by firm, ZIP, and week.
∆Low-Skill P ostings ∆High-Skill P ostings High-to-Low-Skills
Postings Ratio
32 (1) (2) (3) (4) (5) (6)
P ost COV ID –0.088** –0.112*** –0.051**
(0.033) (0.038) (0.020)
P ost COV ID × High COV ID Exposure –0.091*** –0.117*** –0.064***
(0.029) (0.029) (0.017)
Controls Yes Yes Yes Yes Yes Yes
Fixed Effects
Firm × Quarter Yes Yes Yes Yes Yes Yes
ZIP Yes Yes Yes Yes Yes Yes
Week No Yes No Yes No Yes
Observations 2,622,347 1,736,534 3,499,555 2,458,596 544,138 397,951
R-squared 0.164 0.162 0.261 0.248 0.621 0.593
Statistical significance is indicated as follows: *** p <0.01, ** p <0.05, * p <0.1.
Table 4. The Impact of COVID-19 on Job Postings: Firm Size
This table reports output from Eq. (1). The dependent variables are N ew Job P ostings and High-to-Low-Skills
Postings Ratio. The unit of observation is a firm–ZIP–week, where ZIP is the three-digit ZIP-code of a job
posting. N ew Job P ostings is the logarithm of one plus the total number of job postings created. High-to-Low-
Skills Postings Ratio is the total number of job postings created for occupations with O*NET occupation codes
corresponding to Job Zone 5 divided by the total number of job postings created for occupations with O*NET
occupation codes corresponding to Job Zone 1. P ost COV ID is an indicator variable that takes the value of 1 for
each week after February 29th , 2020 and zero otherwise. Small F irm (Assets) is an indicator variable that takes
the value of 1 for firms in the bottom tercile of total assets (measured in the last available quarter) and 0 for firms
in the top tercile. Small F irm (Employees) is an indicator variable that takes the value of 1 for firms in the
bottom tercile of total employees (measured in the last available year) and 0 for firms in the top tercile. State and
firm controls are as defined in Table 2. Firm×quarter–, industry×quarter–, ZIP–, and week–fixed effects are in-
cluded as indicated. All regressions are estimated over a sample of public firms over the January 1st , 2017 to May
5th , 2020 period. Robust standard errors, reported in parentheses, are triple–clustered by firm, ZIP, and week.
N ew Job P ostings High-to-Low-Skills
Postings Ratio
(1) (2) (3) (4)
P ost COV ID × Small F irm (Assets) –0.231*** –0.183***
(0.042) (0.048)
P ost COV ID × Small F irm (Employees) –0.234*** –0.162***
(0.066) (0.037)
Controls Yes Yes Yes Yes
Fixed Effects
Industry × Quarter Yes Yes Yes Yes
ZIP Yes Yes Yes Yes
Week Yes Yes Yes Yes
Observations 3,570,023 3,911,149 137,621 176,172
R-squared 0.175 0.164 0.113 0.110
Statistical significance is indicated as follows: *** p <0.01, ** p <0.05, * p <0.1.
33
Table 5. The Impact of COVID-19 on Job Postings: Labor Market and Geographical Characteristics
This table reports output from Eq. (1). The dependent variables are N ew Job P ostings and High-to-Low-Skills
Postings Ratio. The unit of observation is a firm–ZIP–week, where ZIP is the three-digit ZIP-code of a job
posting. N ew Job P ostings is the logarithm of one plus the total number of job postings created. High-to-Low-
Skills Postings Ratio is the total number of job postings created for occupations with O*NET occupation codes
corresponding to Job Zone 5 divided by the total number of job postings created for occupations with O*NET
occupation codes corresponding to Job Zone 1. P ost COV ID is an indicator variable that takes the value
of 1 for each week after February 29th , 2020 and zero otherwise. High U nionization is an indicator variable
that takes the value of 1 for firms in the top tercile of the labor unionization rate (at the industry level) and
0 for firms in the bottom tercile. Non-Tradables is an indicator variable that takes the value of 1 for firms in
the non-tradable sector and 0 for firms in the tradable sector. Low Local Labor M arket Depth is an indicator
variable that takes the value of 1 for ZIP codes in the top tercile of the Herfindahl–Hirschman Index of active
job postings and 0 for ZIP codes in the bottom tercile of the HHI of active job postings. The HHI is calculated
within a ZIP across all employers with active job postings in that ZIP. High F irm Geographic Concentration
is an indicator variable that takes the value of 1 for firms in the top tercile of the HHI of their operations and
0 for firms in the bottom tercile of the HHI of their operations. The HHI of a firm’s operations is calculated by
taking the sum of the squared employment shares across all ZIP codes a firm operates. State and firm controls
are as defined in Table 2. Firm×quarter–, industry×quarter–, ZIP–, and week–fixed effects are included as
indicated. All regressions are estimated over a sample of private and public firms (except column (4) which
is estimated over public firms) over the January 1st , 2017 to May 5th , 2020 period. Robust standard errors,
reported in parentheses, are triple–clustered by firm, ZIP, and week.
N ew Job P ostings
(1) (2) (3) (4)
P ost COV ID × High U nionization –0.022**
(0.010)
P ost COV ID × Non-Tradables –0.007**
(0.003)
P ost COV ID × Low Local Labor M arket Depth –0.073**
(0.029)
P ost COV ID × High F irm Geographic Concentration –0.067***
(0.015)
Controls Yes Yes Yes Yes
Fixed Effects
Firm × Quarter Yes Yes Yes No
Industry × Quarter No No No Yes
ZIP Yes Yes Yes Yes
Week Yes Yes Yes Yes
Observations 2,592,686 2,997,243 8,324,025 2,309,688
R-squared 0.388 0.255 0.331 0.211
34
High-to-Low-Skills Postings Ratio
(5) (6) (7) (8)
P ost COV ID × High U nionization –0.015***
(0.002)
P ost COV ID × Non-Tradables –0.009**
(0.004)
P ost COV ID × Low Local Labor M arket Depth –0.016**
(0.007)
P ost COV ID × High F irm Geographic Concentration –0.029*
(0.016)
Controls Yes Yes Yes Yes
Fixed Effects
Firm × Quarter Yes Yes Yes No
Industry × Quarter No No No Yes
ZIP Yes Yes Yes Yes
Week Yes Yes Yes Yes
Observations 128,638 122,037 344,106 160,312
R-squared 0.440 0.482 0.614 0.110
Statistical significance is indicated as follows: *** p <0.01, ** p <0.05, * p <0.1.
35
Table 6. The Impact of COVID-19 on Job Postings: Access to Finance and Credit
This table reports output from Eq. (1). The dependent variables are N ew Job P ostings and High-to-Low-Skills
Postings Ratio. The unit of observation is a firm–ZIP–week, where ZIP is the three-digit ZIP-code of a job
posting. N ew Job P ostings is the logarithm of one plus the total number of job postings created. High-to-
Low-Skills Postings Ratio is the total number of job postings created for occupations with O*NET occupation
codes corresponding to Job Zone 5 divided by the total number of job postings created for occupations with
O*NET occupation codes corresponding to Job Zone 1. P ost COV ID is an indicator variable that takes the
value of 1 for each week after February 29th , 2020 and zero otherwise. P rivate F irm is an indicator variable
that takes the value of 1 for private firms and 0 for public firms. Speculative Grade is an indicator variable that
takes the value of 1 for firms with a speculative grade rating and 0 for firms with an investment grade rating.
N o Credit Lines is an indicator variable that takes the value of 1 for firms with no outstanding lines of credit
and 0 for firms with at least one outstanding line of credit. Low Cash Holdings is an indicator variable that
takes the value of 1 for firms in the lowest tercile of cash (scaled by lagged assets) and 0 for firms in the highest
tercile of cash. State and firm controls are as defined in Table 2. Firm×quarter–, industry×quarter–, ZIP–, and
week–fixed effects are included as indicated. All regressions are estimated over a sample of private and public
firms (except columns (2) through (4) which are estimated over public firms) over the January 1st , 2017 to May
5th , 2020 period. Robust standard errors, reported in parentheses, are triple–clustered by firm, ZIP, and week.
N ew Job P ostings
(1) (2) (3) (4)
P ost COV ID × P rivate F irm –0.027**
(0.013)
P ost COV ID × Speculative Grade –0.038**
(0.014)
P ost COV ID × N o Credit Lines –0.051***
(0.011)
P ost COV ID × Low Cash Holdings –0.049***
(0.015)
Controls Yes Yes Yes Yes
Fixed Effects
Firm × Quarter Yes No No No
Industry × Quarter No Yes Yes Yes
ZIP Yes Yes Yes Yes
Week Yes Yes Yes Yes
Observations 11,387,645 4,649,078 10,486,090 2,454,872
R-squared 0.385 0.144 0.080 0.118
36
High-to-Low-Skills Postings Ratio
(5) (6) (7) (8)
P ost COV ID × P rivate F irm –0.080**
(0.036)
P ost COV ID × Speculative Grade –0.024**
(0.010)
P ost COV ID × N o Credit Lines –0.054***
(0.009)
P ost COV ID × Low Cash Holdings –0.069**
(0.034)
Controls Yes Yes Yes Yes
Fixed Effects
Firm × Quarter Yes No No No
Industry × Quarter No Yes Yes Yes
ZIP Yes Yes Yes Yes
Week Yes Yes Yes Yes
Observations 544,138 185,686 538,114 70,405
R-squared 0.603 0.101 0.122 0.224
Statistical significance is indicated as follows: *** p <0.01, ** p <0.05, * p <0.1.
37
Table 7. The Impact of COVID-19 on Job Postings: Income Levels and Inequality
This table reports output from Eq. (1). The dependent variables are N ew Job P ostings and High-to-Low-Skills
Postings Ratio. The unit of observation is a firm–ZIP–week, where ZIP is the three-digit ZIP-code of a job
posting. N ew Job P ostings is the logarithm of one plus the total number of job postings created. High-to-
Low-Skills Postings Ratio is the total number of job postings created for occupations with O*NET occupation
codes corresponding to Job Zone 5 divided by the total number of job postings created for occupations with
O*NET occupation codes corresponding to Job Zone 1. P ost COV ID is an indicator variable that takes the
value of 1 for each week after February 29th , 2020 and zero otherwise. Low Local Household Income is an
indicator variable that takes the value of 1 for counties in the bottom tercile of median household income as
of 2018 and 0 for counties in the top tercile. High Local Income Inequality is an indicator variable that takes
the value of 1 for counties in the top tercile of Gini coefficient and 0 for firms in the bottom tercile. State and
firm controls are as defined in Table 2. Firm×quarter–, industry×quarter–, ZIP–, and week–fixed effects are
included as indicated. All regressions are estimated over a sample of private and public firms (except columns
(2) and (3) which are estimated over public firms) over the January 1st , 2017 to May 5th , 2020 period. Robust
standard errors, reported in parentheses, are triple–clustered by firm, ZIP, and week.
N ew Job P ostings High-to-Low-Skills
Postings Ratio
(1) (2) (3) (4)
P ost COV ID × Low Local Household Income –0.025** –0.008**
(0.010) (0.003)
P ost COV ID × High Local Income Inequality –0.033*** –0.019**
(0.011) (0.007)
Controls Yes Yes Yes Yes
Fixed Effects
Firm × Quarter Yes Yes Yes Yes
ZIP Yes Yes Yes Yes
Week Yes Yes Yes Yes
Observations 7,596,947 7,774,418 329,495 346,391
R-squared 0.344 0.377 0.579 0.553
38
Appendix A O*NET Job Zone Classification
Table A.1. List of Low Skill Jobs
This table reports the list of low skill jobs with O*NET occupation codes corresponding to Job Zone 1.
Low Skill Occupations
Cooks, Fast Food
Food Preparation Workers
Combined Food Preparation and Serving Workers, Including Fast Food
Counter Attendants, Cafeteria, Food Concession, and Coffee Shop
Baristas
Food Servers, Nonrestaurant
Dining Room and Cafeteria Attendants and Bartender Helpers
Dishwashers
Landscaping and Groundskeeping Workers
Amusement and Recreation Attendants
Models
Door-To-Door Sales Workers, News and Street Vendors, and Related Workers
Graders and Sorters, Agricultural Products
Agricultural Equipment Operators
Farmworkers and Laborers, Crop
Fishers and Related Fishing Workers
Hunters and Trappers
Fallers
Logging Equipment Operators
Cement Masons and Concrete Finishers
Plasterers and Stucco Masons
Helpers–Painters, Paperhangers, Plasterers, and Stucco Masons
Septic Tank Servicers and Sewer Pipe Cleaners
Derrick Operators, Oil and Gas
Rock Splitters, Quarry
Roustabouts, Oil and Gas
Fabric Menders, Except Garment
Meat, Poultry, and Fish Cutters and Trimmers
Laundry and Dry-Cleaning Workers
Pressers, Textile, Garment, and Related Materials
Sewing Machine Operators
Grinding and Polishing Workers, Hand
Cutters and Trimmers, Hand
Painting, Coating, and Decorating Workers
Bridge and Lock Tenders
Conveyor Operators and Tenders
39
Table A.2. List of High Skill Jobs
This table reports the list of high skill jobs with O*NET occupation codes corresponding to Job Zone 5.
High Skill Jobs
Chief Executives Sociologists
Chief Sustainability Officers Urban and Regional Planners
Treasurers and Controllers Anthropologists
Education Administrators, Elementary and Secondary School Archeologists
Education Administrators, Postsecondary Geographers
Distance Learning Coordinators Historians
Architectural and Engineering Managers Political Scientists
Medical and Health Services Managers Substance Abuse and Behavioral Disorder Counselors
Natural Sciences Managers Educational, Guidance, School, and Vocational Counselors
Investment Fund Managers Marriage and Family Therapists
Management Analysts Mental Health Counselors
Financial Quantitative Analysts Rehabilitation Counselors
Computer and Information Research Scientists Healthcare Social Workers
Mathematicians Mental Health and Substance Abuse Social Workers
Operations Research Analysts Clergy
Statisticians Lawyers
Biostatisticians Judicial Law Clerks
Environmental Engineers Administrative Law Judges, Adjudicators, and Hearing Offi-
cers
Human Factors Engineers and Ergonomists Arbitrators, Mediators, and Conciliators
Fuel Cell Engineers Judges, Magistrate Judges, and Magistrates
Microsystems Engineers Business Teachers, Postsecondary
Nanosystems Engineers Computer Science Teachers, Postsecondary
Animal Scientists Mathematical Science Teachers, Postsecondary
Soil and Plant Scientists Architecture Teachers, Postsecondary
Biologists Engineering Teachers, Postsecondary
Biochemists and Biophysicists Agricultural Sciences Teachers, Postsecondary
Microbiologists Biological Science Teachers, Postsecondary
Zoologists and Wildlife Biologists Forestry and Conservation Science Teachers, Postsecondary
Bioinformatics Scientists Atmospheric, Earth, Marine, and Space Sciences Teachers,
Postsecondary
Molecular and Cellular Biologists Chemistry Teachers, Postsecondary
Geneticists Environmental Science Teachers, Postsecondary
Epidemiologists Physics Teachers, Postsecondary
Medical Scientists, Except Epidemiologists Anthropology and Archeology Teachers, Postsecondary
Astronomers Area, Ethnic, and Cultural Studies Teachers, Postsecondary
Physicists Economics Teachers, Postsecondary
Materials Scientists Geography Teachers, Postsecondary
Climate Change Analysts Political Science Teachers, Postsecondary
Environmental Restoration Planners Psychology Teachers, Postsecondary
Industrial Ecologists Sociology Teachers, Postsecondary
Hydrologists Health Specialties Teachers, Postsecondary
Remote Sensing Scientists and Technologists Nursing Instructors and Teachers, Postsecondary
Economists Education Teachers, Postsecondary
Environmental Economists Library Science Teachers, Postsecondary
Survey Researchers Criminal Justice and Law Enforcement Teachers, Postsec-
ondary
School Psychologists Law Teachers, Postsecondary
Clinical Psychologists Social Work Teachers, Postsecondary
Counseling Psychologists Art, Drama, and Music Teachers, Postsecondary
Industrial-Organizational Psychologists Communications Teachers, Postsecondary
Neuropsychologists and Clinical Neuropsychologists English Language and Literature Teachers, Postsecondary
40
High Skill Jobs
History Teachers, Postsecondary Veterinarians
Philosophy and Religion Teachers, Postsecondary Advanced Practice Psychiatric Nurses
Graduate Teaching Assistants Clinical Nurse Specialists
Home Economics Teachers, Postsecondary Nurse Anesthetists
Recreation and Fitness Studies Teachers, Postsecondary Nurse Midwives
Special Education Teachers, Preschool Nurse Practitioners
Archivists Audiologists
Curators Acupuncturists
Librarians Naturopathic Physicians
Farm and Home Management Advisors Orthoptists
Instructional Coordinators Cytotechnologists
Instructional Designers and Technologists Orthotists and Prosthetists
Set and Exhibit Designers Athletic Trainers
Chiropractors Genetic Counselors
Dentists, General Foreign Language and Literature Teachers, Postsecondary
Oral and Maxillofacial Surgeons
Orthodontists
Prosthodontists
Dietitians and Nutritionists
Optometrists
Pharmacists
Anesthesiologists
Family and General Practitioners
Internists, General
Obstetricians and Gynecologists
Pediatricians, General
Psychiatrists
Surgeons
Allergists and Immunologists
Dermatologists
Hospitalists
Neurologists
Nuclear Medicine Physicians
Ophthalmologists
Pathologists
Physical Medicine and Rehabilitation Physicians
Preventive Medicine Physicians
Radiologists
Sports Medicine Physicians
Urologists
Physician Assistants
Anesthesiologist Assistants
Podiatrists
Occupational Therapists
Low Vision Therapists, Orientation and Mobility Specialists,
and Vision Rehabilitation Therapists
Physical Therapists
Art Therapists
Speech-Language Pathologists
Exercise Physiologists
41
Appendix B NAICS Industry Classification
Table B.1. List of 3-Digit NAICS Industries
This table reports the list of 3-digit NAICS codes belonging to each industry.
Industry Category 3-Digit NAICS Codes
Agriculture 111,112,115
Mining, Oil & Gas 211,212,213
Utilities 221
Construction 236,237,238
Food 311,722
Beverage & Tobacco Manufacturing 312
Textile Manufacturing 313,314,315,316
Wood Product Manufacturing 321
Printing & Paper 322,323
Chemicals 324,325,326
Metals & Machinery 327,331,332,333
Computer & Electronic Manufacturing 334
Electrical Equipment Manufacturing 335
Transportation Equipment Manufacturing 336
Furniture Product Manufacturing 337
Wholesalers 423,424,425
Retail Trade 441,442,443,444,445,446,447,448,451,452,453,454
Transportation & Warehousing 481,482,483,484,485,486,487,488,492,493
Publishing Industries 511
Telecom & Information Services 512,515,517,518,519
Financial Services 522,523,524,525
Real Estate 531,532,533
Professional, Scientific & Technical Services 541
Management Services 551
Administrative & Support Services 561
Waste Management & Remediation 562
Educational Services 611
Ambulatory Health Care Services 621
Hospitals 622
Nursing & Residential Care Facilities 623
Social Assistance 624
Amusement, Gambling & Recreation 713
Accommodation 721
Repair & Maintenance 811
Personal & Laundry Service 812
42
Table C.1. The Impact of COVID-19 on Job Postings: Worker Skills Robustness
This table reports output from Eq. (1). The dependent variables are High (4-5)-to-Low (1-2)-Skills Postings Ratio, High (5)-to-Low (1-2)-Skills Postings
Ratio, and High (4-5)-to-Low (1)-Skills Postings Ratio. The unit of observation is a firm–ZIP–week, where ZIP is the three-digit ZIP-code of a job posting.
High (4-5)-to-Low (1-2)-Skills Postings Ratio is the total number of job postings created for occupations with O*NET occupation codes corresponding to
Job Zones 4 and 5 divided by the total number of job postings created for occupations with O*NET occupation codes corresponding to Job Zones 1 and
2. High (5)-to-Low (1-2)-Skills Postings Ratio is the total number of job postings created for occupations with O*NET occupation codes corresponding
to Job Zone 5 divided by the total number of job postings created for occupations with O*NET occupation codes corresponding to Job Zones 1 and 2.
High (4-5)-to-Low (1)-Skills Postings Ratio is the total number of job postings created for occupations with O*NET occupation codes corresponding
to Job Zones 4 and 5 divided by the total number of job postings created for occupations with O*NET occupation codes corresponding to Job Zone 1.
P ost COV ID is an indicator variable that takes the value of 1 for each week after February 29th , 2020 and zero otherwise. High COV ID Exposure is
an indicator variable that takes the value of 1 for each county-week belonging to the highest tercile of the number of confirmed COVID cases per capita
and 0 for each county-week belonging to the lowest tercile of the number of confirmed COVID cases per capita. State and firm controls are as defined in
Table 2. Firm×quarter–, ZIP–, and week–fixed effects are included as indicated. All regressions are estimated over a sample of private and public firms
over the January 1st , 2017 to May 5th , 2020 period. Robust standard errors, reported in parentheses, are triple–clustered by firm, ZIP, and week.
High (4-5)-to-Low High (5)-to-Low High (4-5)-to-Low
43 (1-2)-Skills Postings (1-2)-Skills Postings (1)-Skills Postings
Ratio Ratio Ratio
(1) (2) (3) (4) (5) (6)
P ost COV ID –0.015*** –0.036*** –0.082**
(0.005) (0.003) (0.038)
P ost COV ID × High COV ID Exposure –0.034*** –0.051*** –0.092***
(0.09) (0.005) (0.016)
Controls Yes Yes Yes Yes Yes Yes
Fixed Effects
Firm × Quarter Yes Yes Yes Yes Yes Yes
ZIP Yes Yes Yes Yes Yes Yes
Week Yes Yes Yes Yes Yes Yes
Observations 544,138 397,951 544,138 397,951 544,138 397,951
R-squared 0.202 0.202 0.204 0.209 0.541 0.539
Statistical significance is indicated as follows: *** p <0.01, ** p <0.05, * p <0.1.
Table C.2. The Impact of COVID-19 on Job Postings: COVID-19 Exposure Robustness
This table reports output from Eq. (1). The dependent variables are N ew Job P ostings and High-to-Low-Skills Postings Ratio. The unit of observation is
a firm–ZIP–week, where ZIP is the three-digit ZIP-code of a job posting. N ew Job P ostings is the logarithm of one plus the total number of job postings
created. High-to-Low-Skills Postings Ratio is the total number of job postings created for occupations with O*NET occupation codes corresponding to Job
Zone 5 divided by the total number of job postings created for occupations with O*NET occupation codes corresponding to Job Zone 1. P ost COV ID
is an indicator variable that takes the value of 1 for each week after February 29th , 2020 and zero otherwise. High COV ID Exposure (Quartiles) is an
indicator variable that takes the value of 1 for each county-week belonging to the highest quartile of the number of confirmed COVID cases per capita
and 0 for each county-week belonging to the lowest quartile of the number of confirmed COVID cases per capita. High COV ID Exposure (Quintiles)
and High COV ID Exposure (Deciles) are analogously defined. State and firm controls are as defined in Table 2. Firm×quarter–, ZIP–, and week–fixed
effects are included as indicated. All regressions are estimated over a sample of private and public firms over the January 1st , 2017 to May 5th , 2020
period. Robust standard errors, reported in parentheses, are triple–clustered by firm, ZIP, and week.
N ew Job P ostings High-to-Low-Skills Postings Ratio
(1) (2) (3) (4) (5) (6)
44 P ost COV ID × High COV ID Exposure (Quartiles) –0.042*** –0.066**
(0.013) (0.028)
P ost COV ID × High COV ID Exposure (Quintiles) –0.047*** –0.092**
(0.013) (0.041)
P ost COV ID × High COV ID Exposure (Deciles) –0.051*** –0.098**
(0.015) (0.044)
Controls Yes Yes Yes Yes Yes Yes
Fixed Effects
Firm × Quarter Yes Yes Yes Yes Yes Yes
ZIP Yes Yes Yes Yes Yes Yes
Week Yes Yes Yes Yes Yes Yes
Observations 6,030,895 4,804,020 1,941,381 224,017 175,644 69,949
R-squared 0.387 0.394 0.414 0.600 0.604 0.580
Statistical significance is indicated as follows: *** p <0.01, ** p <0.05, * p <0.1.
File and source
- File
- corporate-hiring-under-covid-19-labor-market-concentration-downskilling-and-income-inequality.pdf
- Size
- 1,410,095 bytes
- SHA-256
- 1e9635dba23c4c8fcd8cca57a16d4de5fe362d41363d4c99a2d18042d3c77e60
- Our copy
- corporate-hiring-under-covid-19-labor-market-concentration-downskilling-and-income-inequality.pdf
- Original
- www.nber.org