FEDERAL RESERVE BANK OF SAN FRANCISCO
WORKING PAPER SERIES
Pandemic Layoffs and the Role of Stay-At-Home Orders
Marianna Kudlyak
Federal Reserve Bank of San Francisco
Hoover Institution
CEPR and IZA
Erin L. Wolcott
Middlebury College
July 2024
Working Paper 2024-20
https://doi.org/10.24148/wp2024-20
Suggested citation:
Kudlyak, Marianna and Erin L. Wolcott. 2024. “Pandemic Layoffs and the Role of Stay-
At-Home Orders.” Federal Reserve Bank of San Francisco Working Paper 2024-20.
https://doi.org/10.24148/wp2024-20
The views in this paper are solely the responsibility of the authors and should not be interpreted
as reflecting the views of the Federal Reserve Bank of San Francisco or the Board of Governors
of the Federal Reserve System.
Pandemic Layoffs and the Role of Stay-At-Home Orders
∗
Marianna Kudlyak
†
Erin L. Wolcott
‡
July 10, 2024
Abstract
We compile a novel high-frequency, detailed geographic dataset on mass layoffs
from U.S. state labor departments. Using recent advances in difference-in-difference
estimation with staggered treatment, we find that locally-mandated stay-at-home or-
ders issued March 16–22, 2020 triggered mass layoffs equal to half a percent of the
population in just one week. Our findings contribute to explanations for why job loss
in 2020 was synchronous and catastrophic, yet temporary.
JEL: E32, J63, J64.
Keywords: Pandemic. Mass Layoffs. Unemployment. Recovery.
∗
The authors thank Anjana Giridhar, Noah A. Kouchekinia, Celeste Liu, Claire Moy, Lachlan Pinney,
and Samuel R. Tarasewicz for outstanding research assistance. The authors thank Robert E. Hall, Fernando
Rios-Avila, Pedro H. C. SantAnna, Tanya Byker, and David Wiczer (discussant) for useful comments. Any
opinions expressed are those of the authors and do not reflect those of the Federal Reserve Bank of San
Francisco or the Federal Reserve System.
†
‡
Middlebury College; wolcott@middlebury.edu
1 Introduction
The beginning of the 2020 pandemic in the U.S. was accompanied by layoffs of catastrophic
proportions. Between February and April, the unemployment rate shot up from 3.5% to
14.8%. Not only was the level of unemployment unprecedented, so was the speed of its
increase. In March 2020, local governments issued stay-at-home or shelter-in-place orders
(which we refer to as SAH orders), advising residents, with the exception of those employed
in essential occupations, to remain home. The orders triggered shutdowns of in-person
production facilities, regardless of whether firms experienced less demand.
We study whether the timing of stay-at-home orders, caused by the epidemiological sit-
uation, triggered mass layoffs during the pandemic. Given the local and abrupt nature of
SAH orders, measuring their effect on the labor market requires data that is both geographi-
cally granular and high-frequency. First, the orders were local: some localities (counties and
cities) imposed SAH orders before state-wide orders. Ignoring local SAH orders might bias
results toward concluding layoffs occurred prior to orders. Second, the timing of the orders
across localities was staggered: analyzing their effect on layoffs requires proper selection of a
control group. Ignoring the staggered nature of SAH orders might bias results towards zero.
We compile a novel dataset made possible by the 1988 Federal Worker Adjustment and
Retraining Notification Act (WARN) containing the date and location of mass layoff events.
Following recent advances in difference-in-difference estimation with time-varying treatment
(Goodman-Bacon and Marcus (2020) and Callaway and Sant’Anna (2021)), we compare
localities that recently issued a SAH order to localities that had not yet issued a SAH order.
We find that, at the onset of the pandemic, locally mandated stay-at-home orders triggered
mass layoffs. While we find that some layoffs preceded local SAH orders, their magnitudes
were small in comparison. We also present evidence that employers expected layoffs to be
temporary.
Our findings provide evidence that the pandemic shock differed from a typical recessionary
shock. In a typical recession, layoffs occur to help cut costs in response to declining demand;
these layoffs are predominantly permanent, which destroys firm-worker relationships and the
associated match capital. In contrast, during the March–April 2020 recession, layoffs were
mostly triggered by social distancing policies caused by the epidemiological situation and
firms explicitly preserved the recall option at the time of layoff. This helps explain why the
labor market recovered relatively quickly from the recession despite the unemployment rate
reaching historic highs.
1
Our work also contributes to the growing literature that highlights the trade-offs between
public health measures and economic outcomes, allowing us to better assess the effectiveness
of the policy and how best to design similar policies to prepare for future events. Despite
increasing unemployment, the literature finds containment measures were a cost-effective
policy
1
2 Data
To comply with WARN, state labor departments collect data on mass layoff events and post
this data to varying levels of detail on their websites. We scrape data for 21 states that
publish the number of workers laid off, the date the layoff event took effect, and the location
(either address, county, or city) of the plant or establishment where the layoff occurred.
2
WARN layoffs events are where: (1) the plant closure or job loss is large enough to trigger
WARN and (2) the firm doing the layoffs is large enough to be subject to WARN.
3
Although
WARN layoffs are not representative of the average employment loss event, they do capture
the severity and abruptness of the pandemic recession.
4,5
To account for the geographic unit for which a SAH order binds, we collect data from The
New York Times (and corroborated with government websites and local news sources) on the
date of all SAH orders at the state, county, and city level that occurred within the 21 states.
6
We define a locality as an geographic area with a binding SAH order. For example, the state
of Alaska imposed a SAH order on March 28, but the city of Anchorage imposed a SAH
order a week earlier on March 22. There were no other SAH orders in Alaska relevant for
our layoff data. We therefore define Anchorage as one locality and Alaska minus Anchorage
as another locality. We have 207 unique localities.
We aggregate the data by calendar week, where the first day of the week is Monday and
the last day is Sunday. Figure 1 plots the average layoff rate for each locality and calendar
week in our sample. Mass layoffs in February and the first week of March were close to
1
Beland, Brodeur and Wright (2020), Gupta, Montenovo, Nguyen, Lozano-Rojas, Schmutte, Simon,
Weinberg and Wing (2020), Baek, McCrory, Messer and Mui (2021) find that state SAH policies increased
unemployment. Beland et al. (2020), Farboodi, Jarosch and Shimer (2021), and Eichenbaum, Rebelo and
Trabandt (2021) find the health benefits of SAH policies were substantial.
2
The U.S. states in the study are AK, AL, CA, CO, FL, GA, ID, IL, MI, MO, MS, NC, NM, NY, OR,
PA, RI, SC, TN, TX, WA.
3
Typically, this includes employers with at least 100 employees. See Appendix A for more detail.
4
See Appendix B.
5
In a comprehensive analysis, Krolikowski and Lunsford (2024) scrape and aggregate monthly WARN
data from 33 states. We require a higher level of detail, namely, the exact date and location.
6
https://www.nytimes.com/interactive/2020/us/coronavirus-stay-at-home-order.html
2
Figure 1: Weekly WARN Layoffs during Onset of 2020 Pandemic, As Share of Population %
zero. This was an expansionary period for the U.S. labor market. However, starting the
week of March 9, mass layoffs spiked to a tenth of a percent even though no SAH order had
been imposed. The first SAH orders went into effect Tuesday, March 17. Seven counties in
California imposed a SAH order two days before the state order on March 19. Many more
cities, counties, and states followed in the days after.
For controls, we include the number of current covid cases and deaths as a share of the
population. Health data are from John Hopkins Center for Systems Science and Engineering
and population data is from the Census Bureau. Appendix C provides summary statistics
for the variables in our sample.
2.1 Empirical Strategy
We estimate the following dynamic two-way fixed effects model:
Y
i,t
= α
i
+ ϕ
t
+
X
j̸=−3
β
j
[SAH
i,t
= j] + γX
i,t
+ ϵ
i,t
, (1)
3
where Y
i,t
is the share of the population experiencing a WARN layoff at locality i during
calendar week t. Parameter α
i
represents locality fixed effects and ϕ
t
represents calendar
week fixed effects. SAH
i,t
= t − SAHweek
i
is the week relative to treatment. To allow for
anticipation up to two weeks before treatment, three weeks before the SAH order is left out
of the summation. The coefficients of interest, β
j
, indicate how the layoff rate in any given
week differs from three weeks before the SAH order. X
i,t
is a vector of controls including
covid case rates and death rates during calendar week t, and ϵ
i,t
is the residual.
Following Callaway and Sant’Anna (2021), our control group only consists of not-yet and
never treated localities. This is in contrast with standard event study techniques where the
control group consists of observations that are not treated in that period but might have
already been treated in the past. The problem with the latter approach for our analysis is
that it would group localities that already had a SAH order in the control group. This is
not the group we are interested in comparing the treated localities to because these localities
might still be experiencing layoffs related to their SAH order. Instead, we want to compare
the treated (localities with a SAH order this week) to the not-yet and never treated (localities
without a SAH order to date).
While communities in the early SAH localities may, in fact, have been surprised, it is
difficult to argue that communities in the late SAH localities did not anticipate one coming.
Therefore, we split the sample by the calendar week of their SAH order. From March 16–22,
27 localities imposed a SAH order; from March 23–29, 147 localities imposed a SAH order;
and from March 30–April 5, 34 localities imposed a SAH order. The rest of the sample
includes Missouri and South Carolina, which both imposed a SAH order on April 6, and
Iowa, which never imposed a SAH order.
7
3 Results and Discussion
Figure 2 plots the estimated effects. Panel A shows that for early adopting localities, mass
layoffs increased the week before the SAH order. Firms laid off workers either because they
anticipated a SAH order or because they were concerned about health or economic conditions
related to the virus. That said, the magnitude of the increase one week before is dwarfed by
layoffs spiking to half a percent of the population during the week of the order, suggesting
7
The Missouri locality excludes Cole County, Crawford County, and the city of Hannibal, and the South
Carolina locality excludes the the city of Charleston and the city of Columbia, all of which had SAH orders
before their state-issued order.
4
Figure 2: Estimated Local WARN Layoffs Since the Local Stay-at-Home Order
Note: The figure shows estimates of the mass layoffs in percent of the local population. Early adopters are
the localities that adopted SAH orders in March 16–22. Middle adopters are the localities that adopted SAH
orders in March 23–29. Late adopters are the localities that adopted SAH orders in March 30–April 5.
5
local SAH orders triggered mass layoffs. Layoffs remained elevated one and two weeks after
the SAH order before returning to previous levels.
Panel B shows that for middle adopting localities, mass layoffs may have also increased
before the SAH order. According to point estimates, mass layoffs increased a tenth of
a percent of the population one and two weeks before the SAH order, but the 95 percent
confidence bands are large and include zero. Panel B also shows that mass layoffs experienced
a statistically significant increase the week of the order and the week after the order, yet,
relative to Panel A, effects are small.
Panel C shows that for late adopting localities, layoffs may have slightly increased the
week before the SAH order, but the relationship is statistically insignificant.
Similar to Baek et al. (2021), we find that some of the increase in unemployment during
the pandemic can be attributed to SAH orders. Our approach differs because our more
granular data allow us to capture the effect of city- and county-issued SAH orders predating
the state order. Baek et al. (2021) use unemployment claims to measure labor market slack
which is only available at the state-level and therefore cannot fully capture the effect of a
local SAH order. We also differentiate between localities that were the first to adopt SAH
orders from localities that adopted SAH orders a week or two later. The first group of
adaptors experienced an arguably unexpected shock. Late adopters likely anticipated their
SAH order which would have influenced when they laid off workers.
We only have data for 21 states for which WARN layoffs and their associated location
and layoff date could be scrapped from government websites. Our sample of 207 localities is
not large enough to estimate equation (1) using daily data so we aggregate to the calendar
week. Apart from Iowa, our sample does not include states that never imposed a SAH order
because these states do not post detailed WARN data. Lastly, our layoff data is not the
universe of layoffs but rather layoff events and employers large enough to trigger WARN
legislation. Despite its limitations, the WARN dataset we compile, to our knowledge, is the
only daily, city-level, and direct measure of layoffs during the pandemic.
Another feature of the pandemic recession was the temporary nature of the layoffs. In a
typical recession, most unemployment is from reasons other than temporary layoffs. How-
ever, during the pandemic, temporary layoffs were widespread. Appendix B shows that for
four states that distinguish between temporary and permanent layoffs in their WARN data,
the massive uptick in 2020 was from layoffs firms expected to be temporary. Pandemic lay-
offs were temporary, allowing firms and workers to retain their match-specific capital and
6
avoid the costly process of building new firm-worker relationships.
8
Our finding that SAH
orders (and anticipation of SAH orders) contributed to mass layoffs together with the tem-
porary nature of the layoffs help explain why the labor market recovered so quickly from the
pandemic recession.
4 Conclusion
After job loss of catastrophic proportions, the U.S. labor market recovered remarkably
quickly in 2020. We find that locally-mandated SAH orders in early-adopting localities
triggered WARN layoffs equal to half a percent of the population in just one week. While
we find evidence that some layoffs preceded local SAH orders, their magnitudes were small
in comparison.
8
For example, Bartik, Bertrand, Lin, Rothstein and Unrath (2020), Cajner, Crane, Decker, Grigsby,
Hamins-Puertolas, Hurst, Kurz and Yildirmaz (2020), Wolcott, Ochse, Kudlyak and Kouchekinia (2020),
Kudlyak and Wolcott (2020), Hall and Kudlyak (2022), and Forsythe, Kahn, Lange and Wiczer (2022) study
implications of temporary versus permanent layoffs for recovery from the pandemic.
7
References
Baek, ChaeWon, Peter B. McCrory, Todd Messer, and Preston Mui, “Unemployment Effects
of Stay-at-Home Orders: Evidence from High-Frequency Claims Data,” The Review of
Economics and Statistics, 2021, 103 (5), 979–993.
Bartik, Alexander W., Marianne Bertrand, Feng Lin, Jesse Rothstein, and Matt Unrath,
“Measuring the Labor Market at the Onset of the COVID-19 Crisis,” Brookings Papers
on Economic Activity, 2020, Summer, 239–326.
Beland, Louis-Philippe, Abel Brodeur, and Taylor Wright, “Covid-19, Stay-at-Home Orders
and Employment: Evidence from CPS Data,” Discussion Paper 13282, IZA 2020.
Cajner, Tomaz, Leland D. Crane, Ryan A. Decker, John Grigsby, Adrian Hamins-Puertolas,
Erik Hurst, Christopher Kurz, and Ahu Yildirmaz, “The US Labor Market during the
Beginning of the Pandemic Recession,” Brookings Papers on Economic Activity, 2020,
Summer 2020 (Special Edition), 3–33.
Callaway, Brantly and Pedro H.C. Sant’Anna, “Difference-in-Differences with Multiple Time
Periods,” Journal of Econometrics, 2021, 225 (2), 200–230.
Eichenbaum, Martin S., Sergio Rebelo, and Mathias Trabandt, “The Macroeconomics of
Epidemics,” The Review of Financial Studies, 2021, 34, 5149–5187.
Farboodi, Maryam, Gregor Jarosch, and Robert Shimer, “Internal and External Effects of
Social Distancing in a Pandemic,” Journal of Economic Theory, 2021, 196 (105293).
Forsythe, Eliza, Lisa Kahn, Fabian Lange, and David Wiczer, “Where Have All the Work-
ers Gone? Recalls, Retirements, and Reallocation in the COVID Recovery,” Labour
Economics, 2022, 78 (102251).
Goodman-Bacon, Andrew and Jan Marcus, “Using Difference-in-Differences to Identify
Causal Effects of COVID-19 Policies,” Survey Research Methods, 2020, 14 (2), 153–158.
Gupta, Sumedha, Laura Montenovo, Thuy Nguyen, Felipe Lozano-Rojas, Ian Schmutte,
Kosali Simon, Bruce A. Weinberg, and Coady Wing, “Effects of Social Distancing Policy
on Labor Market Outcomes,” Contemporary Economic Policy, 2020, 41 (1), 166–193.
Hall, Robert E. and Marianna Kudlyak, “The Unemployed with Jobs and without Jobs,”
Labour Economics, 2022, 79.
8
Krolikowski, Pawel M. and Kurt G. Lunsford, “Advance Layoff Notices and Aggregate Job
Loss,” Journal of Applied Econometrics, 2024.
Kudlyak, Marianna and Erin Wolcott, “Pandemic Layoffs,” Working Paper, Middlebury
College 2020.
Wolcott, Erin, Mitchell G. Ochse, Marianna Kudlyak, and Noah A. Kouchekinia, “Tempo-
rary Layoffs and Unemployment in the Pandemic,” Economic Letter, FRB San Fran-
cisco, 2020, (2020-34).
9
A WARN Data Appendix
The Federal Worker Adjustment and Retraining Notification Act (WARN) requires employ-
ers to notify workers and state and local governments 60 days before a plant closing or mass
lay off. According to the US Department of Labor “Advance notice provides workers and
their families some transition time to adjust to the prospective loss of employment, to seek
and obtain alternative jobs and, if necessary, to enter skill training or retraining...” Advance
notice to government officials allows state dislocated worker units time to provide assistance.
9
Employers are covered by the Federal WARN Act if they have 100 or more employees,
not counting employees who have worked less than 6 months in the last 12 months and
not counting employees who work an average of less than 20 hours a week. The term
employment loss means (1) an employment termination, other than a discharge for cause,
voluntary departure, or retirement; (2) a layoff exceeding 6 months; or (3) a reduction in an
employee’s hours of work of more than 50 percent in each month of any 6-month period. A
plant closing occurs if an employment site will be shut down, and the shutdown will result
in an employment loss for 50 or more employees during any 30-day period. A mass layoff
occurs without a plant closing if the layoff results in an employment loss at the employment
site during any 30-day period for 500 or more employees, or for 50-499 employees if they
make up at least 33 percent of the employer’s active workforce.
Some states have their own, more restrictive WARN laws. For example, the New York
WARN Act applies to establishments with 50 or more full-time workers and covers plant
closings and layoffs of 25 or more full-time workers constituting at least 33 percent of all the
workers at a site. Layoffs involving 250 or more full-time workers are covered regardless of
percentage. The California WARN Act applies to establishments with at least 75 full- and
part-time employees in California laying off 50 or more employees regardless of percentage
of the workforce.
The Federal WARN Act exempts firms from filling in advance if a closure is due to
“unforeseeable business circumstances.”
10
States may modify this exemption. For example,
the California WARN Act typically does not include this exemption and so on March 17,
2020, because of the pandemic, California suspended its advance notice requirement.
11
As
a result, firms laid-off workers immediately without advanced notice and because of this
stipulation, were not fined for doing so.
9
https://www.govinfo.gov/content/pkg/CFR-1998-title20-vol3/pdf/
CFR-1998-title20-vol3-part639.pdf
10
https://www.edd.ca.gov/Jobs_and_Training/Layoff_Services_WARN.htm
11
https://www.dir.ca.gov/dlse/WARN-FAQs.html
10
B Temporary vs. Permanent Layoffs
Mass Layoffs from Worker Adjustment and Retraining Notification Data
0 100 200 300 400 500
Layoffs, thousands
2015 2016 2017 2018 2019 2020
Temporary WARN layoffs Type not recorded
Permanent WARN layoffs
Data available through Oct 27, 2021
CA
Data available through
Dec 21
0
50
100
150
200
Layoffs, thousands
2002
2004
2006
2008
2010
2012
2014
2016
2018
2020
Temporary WARN layoffs Type not recorded
Permanent WARN layoffs
NY
0 5 10 15 20
Layoffs, thousands
2010 2012 2014 2016 2018 2020
Temporary WARN layoffs Type not recorded
Permanent WARN layoffs
Data available through Oct 27, 2021
OR
0 10 20 30 40 50
Layoffs, thousands
2005 2010 2015 2020
Temporary WARN layoffs Type not recorded
Permanent WARN layoffs
Data available through Oct 27, 2021
WA
Note: Data from the state labor departments.
11
C Summary Statistics
Variable Mean Standard Deviation Minimum Maximum Observations
Layoff Rate 0.05 0.33 0 10.87 2,070
SAH Order Week 8.01 0.80 0 10 2,070
Covid Case Rate 0.01 0.03 0 0.54 2,070
Covid Death Rate 0.0003 0.002 0 0.05 2,070
Notes: Author’s calculations using data from state labor departments, the Census Bureau, The New York
Times corroborated with local sources, and the John Hopkins Center for Systems Science and Engineering.
An observation is a locality in a given calendar week. There are 207 localities and 10 calendar weeks. Rates
are in units of percentage points. SAH Order Week is the calendar week—where 1 indicates February 3–9,
2020 and 10 indicates April 6–12, 2020—that localities imposed a stay-at-home or shelter-in-place orders.
12
