Digital Resilience: How Work-From-Home Feasibility Affects Firm Performance
Digital Resilience: How Work-From-Home Feasibility Affects Firm Performance
Digital Resilience: How Work-From-Home Feasibility Affects Firm Performance
Wang Jin
MIT Initiative on the Digital
Economy
March, 2021
Working Paper No. 21-016
NBER WORKING PAPER SERIES
DIGITAL RESILIENCE:
HOW WORK-FROM-HOME FEASIBILITY AFFECTS FIRM PERFORMANCE
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 They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies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.
© 2021 by John (Jianqiu) Bai, Erik Brynjolfsson, Wang Jin, Sebastian Steffen, and Chi Wan. 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.
Digital Resilience: How Work-From-Home Feasibility Affects Firm Performance
John (Jianqiu) Bai, Erik Brynjolfsson, Wang Jin, Sebastian Steffen, and Chi Wan
NBER Working Paper No. 28588
March 2021
JEL No. D23,J21,L0,L25,O0,O33
ABSTRACT
Digital technologies may make some tasks, jobs and firms more resilient to unanticipated shocks.
We extract data from over 200 million U.S. job postings to construct an index for firms' resilience
to the Covid-19 pandemic by assessing the work-from-home (WFH) feasibility of their labor
demand. Using a difference-in-differences framework, we find that public firms with high pre-
pandemic WFH index values had significantly higher sales, net incomes, and stock returns than
their peers during the pandemic. Our results indicate that firms with higher digital resilience, as
measured through our pre-pandemic WFH index, performed significantly better in general, and in
non-essential industries in particular, where WFH feasibility was necessary to continue operation.
The ability to use digital technologies to work remotely also mattered more in non-high-tech
industries than in high-tech ones. Lastly, we find evidence that firms with lower pre-pandemic
WFH feasibility attempted to catch up to their more resilient competitors via greater software
investment. This is consistent with a complementarity between digital technologies and WFH
practices. Our study's results are robust to a variety of empirical specifications and provide a first
look at how WFH practices improved resilience to a major, unanticipated social and economic
shock.
Wang Jin
245 First Street, Room E94
MIT Initiative on the Digital Economy
Cambridge, MA 02142
[email protected]
Digital Resilience: How Work-From-Home Feasibility
Affects Firm Performance
Digital technologies may make some tasks, jobs and firms more re-
silient to unanticipated shocks. We extract data from over 200 mil-
lion U.S. job postings to construct an index for firms’ resilience to
the Covid-19 pandemic by assessing the work-from-home (WFH)
feasibility of their labor demand. Using a difference-in-differences
framework, we find that public firms with high pre-pandemic WFH
index values had significantly higher sales, net incomes, and stock
returns than their peers during the pandemic. Our results indicate
that firms with higher digital resilience, as measured through our
pre-pandemic WFH index, performed significantly better in gen-
eral, and in non-essential industries in particular, where WFH
feasibility was necessary to continue operation. The ability to
use digital technologies to work remotely also mattered more in
non-high-tech industries than in high-tech ones. Lastly, we find
evidence that firms with lower pre-pandemic WFH feasibility at-
tempted to catch up to their more resilient competitors via greater
software investment. This is consistent with a complementarity
between digital technologies and WFH practices. Our study’s re-
sults are robust to a variety of empirical specifications and provide
a first look at how WFH practices improved resilience to a major,
unanticipated social and economic shock.
Keywords: Digital Resilience, Work-From-Home, IT Complemen-
tarity, Firm Performance, Covid-19
“We are finding we can reorganize our companies electronically very rapidly
and that’s the only type of organization that can begin to keep pace with the
changing business conditions.”
equal, the feasibility of shifting employees to WFH practices should have helped
firms become more resilient during the pandemic.
To test our hypothesis, we merge the near-universe of U.S. job postings from
Burning Glass Technologies (BGT) with a recently developed occupational WFH
feasibility indicator by Dingel and Neiman (2020) to construct, for the first time,
a firm-level measure of the percentage of its workforce that has the WFH option
– a firm-level WFH feasibility index. We then estimate a difference-in-differences
(DID) regression model to examine whether and how much a firm’s WFH feasi-
bility prior to the Covid-19 pandemic influences its financial and stock market
performance during this crisis.
Our key finding is that firms with high pre-pandemic WFH feasibility, and
thus higher resilience to the pandemic, fared significantly better – with roughly
15% higher net incomes, 4% higher sales, and better stock market performance
(measured by stock returns and volatility) compared to firms with lower pre-
pandemic WFH indices. In contrast, as a response to the pandemic, firms with
lower pre-Covid-19 WFH indices retained a significantly higher percentage of
software investment, as they attempt to catch up in IT to facilitate remote work
and to continue operation during the pandemic.
We also find that pre-pandemic WFH feasibility is associated with better per-
formance during the pandemic for all firms but especially those in non-essential
industries. These are the ones whose operations were more likely to be adversely
affected by Covid-19 and the associated government restrictions. In addition, we
find that firms in non-essential industries with lower pre-pandemic WFH indices
spent a significantly larger share on IT investments, while essential firms, which
were legally guided to continue operation did not. Moreover, these results are
stronger in non-high-tech industries as high WFH firms in these industries had
the largest relative advantages over their competitors. Conversely, in high-tech in-
dustries, a higher percentage of firms were already able to support remote work,
such that even the low WFH firms in high-tech industries did not fare much
worse than their high WFH peers. Our findings provide further evidence that
WFH practices and complementary digital technologies were critical to continue
operation Rahul De’, Neena Pandey and Abhipsa Pal (2020), especially for those
non-essential businesses that were forced to adjust. These results are robust to
empirical specifications that control for a host of firm-level characteristics and
fixed effects.
Our identifying assumption for causal estimates is that firm’s pre-pandemic
WFH index is orthogonal to the timing of the Covid-19 pandemic. In other words,
firms were unable to anticipate the pandemic and were unable to fully prepare for
it by hiring for more WFH feasible jobs or preemptively altering their business
operation in the short-term (Cosmin Ilut, Matthias Kehrig and Martin Schneider,
2018). We believe this to be a reasonable assumption, given the unanticipated
nature of the public health crisis as well as the delayed governmental and state
responses.
4
Second, our study contributes to the recent literature that studies the shift of
workplace norms towards more flexible work settings, such as working from home.
In particular, Brynjolfsson et al. (2020) survey a nationally-representative sample
of the U.S. population. Their results suggest that over 30 percent of workers
switched to remote work during the pandemic and also report that the switch
to WFH can be predicted by the incidence of Covid-19. This further supports
our use of WFH feasibility as a measure of pandemic resilience. By conditioning
on firms’ ex ante WFH indices we mitigate endogeneity concerns that may arise
from heterogeneity in firms’ shift towards WFH practices during the pandemic.
Thus, by investigating how differences in resilience lead to differential performance
during the pandemic, we also provide further evidence for the value created by
digitally-enabled WFH practices.
5 Several other papers that examine the labor aspects during the Covid-19 crisis include the following:
Seth G Benzell, Avinash Collis and Christos Nicolaides (2020) compare the actual closures of commercial
locations to their recommendation of what should be closed first, and generate implications for the optimal
sequence of re-openings when policymakers revive the economy. Andrew G Atkeson (2020) analyzes the
economic consequences of the Covid-19 pandemic and how they correlate with various assumptions about
the ratio between the susceptible – infected – and recovered groups in the population. Eliza Forsythe,
Lisa B Kahn, Fabian Lange and David Wiczer (2020) and Olivier Coibion, Yuriy Gorodnichenko and
Michael Weber (2020) study the labor market implications of the Covid-19 pandemic.
DIGITAL RESILIENCE: FIRM ESTIMATES 5
I. Hypothesis Development
A large body of prior literature explores the effect of WFH or remote work
practices on work-related outcomes at the individual-level including higher job
satisfaction (Timothy D Golden and John F Veiga, 2005; Timothy D Golden,
2006), prolonged work hours and higher work efficiency (Mary Madden and Syd-
ney Jones, 2008; Bloom et al., 2015), lower worker stress (Ravi S Gajendran and
David A Harrison, 2007), and potentially lower turnover (Phyllis Moen, Erin L
Kelly and Rachelle Hill, 2011; Eleni T Stavrou, 2005). In contrast, relatively few
studies focus on firm and organizational outcomes (Bloom et al. 2015; also see
Tammy D Allen, Timothy D Golden and Kristen M Shockley (2015) for a review)
possibly because large-scale firm-level data on the adoption of WFH practices is
hard to come by. Among limited firm-level studies, the effect of WFH on firm
performance is found to be positive (Christine Siegwarth Meyer, Swati Mukerjee
and Ann Sestero, 2001), but the magnitude tends to be small (Martin and Mac-
Donnell, 2012) and contingent on complementary human resource management
practices (Sánchez et al., 2007; Martı́nez-Sánchez et al., 2008).
much of their operations were disrupted by the pandemic. For instance, stay-
at-home orders affected some firms and industries more than others. A natural
way to think about the level of constraints that the government order has put on
different industries is through the definition of essential industries. If an indus-
try was deemed essential, such as the food retail or health care industries, it was
allowed to continue operation on premise during the pandemic. Thus, firms in
these industries are less likely to depend on the WFH feasibility of their employ-
ees. In contrast, many firms in non-essential industries were forced to cease or
modify their operation due to the risk of Covid-19. Therefore, firms’ feasibility
to have their workers work remotely becomes crucial in these industries. Digital
resilience, as measured by ability to work remotely, is likely to be a greater deter-
minant of success or even survival during the pandemic in non-essential compared
to essential industries.
A. Data
posting portals beginning in 2010 and with increasing popularity in recent eco-
nomic research (Hershbein and Kahn, 2018; David Deming and Lisa B Kahn,
2018; José Azar, Ioana Marinescu, Marshall Steinbaum and Bledi Taska, 2020;
Subhro Das, Sebastian Steffen, Wyatt Clarke, Prabhat Reddy, Erik Brynjolfsson
and Martin Fleming, 2020; Daron Acemoglu, David Autor, Jonathon Hazell and
Pascual Restrepo, 2020).6 BGT annotates each job posting with an occupational
title (6-digit SOC code), industry code, employer information, job posting date,
and more. We aggregate this data to the employer-SOC (6-digit) level to obtain
a measure of how many job postings the employer posts in each month, quarter,
and year for each SOC occupation.
Second, we merge this data with the WFH feasibility data from Jonathan Din-
gel and Brent Neiman (2020). They provide a binary index at the 8-digit Oc-
cupational Information Network (O*NET) code level,7 as well as a continuous,
aggregated index at the 6-digit Standard Occupational Classification (SOC) level,
which takes employment shares into account.8
Finally, we take the weighted average over each firm’s occupational WFH in-
dices, weighted by its number of job postings for each occupation. This procedure
enables us to construct, for each firm, the percentage of its labor demand that
has WFH feasibility. Ideally, one would like to have such a measure constructed
over all existing employees, which would capture the actual prevalence of WFH
feasibility within an organization. Since such data is not available, we take the
next best alternative and use the average value of each firm’s quarterly WFH
index over the 2010-2019 period.
Notably, our measure is a forward-looking demand measure and we do not
observe whether firms manage to fill these positions. In principle, it could bias
our results if employers (job seekers) were able to anticipate the pandemic and
adjust the job postings (offers) and hire (accept) towards occupations with higher
WFH feasibility. However, due to the unanticipated nature of the pandemic, this
seems highly unlikely. In fact, while we observed significant changes in average
quarterly WFH indices during the pandemic (relative to the average indices in the
first quarter of 2019), there are no significant differences among the quarters prior
to the pandemic in 2019 regardless of whether firms had high or low pre-pandemic
WFH indices as can be seen in Figure 1.
A more serious concern might be that we do not observe the large number of
6 BGT provides partial coverage for 2007, but does not provide any coverage of job postings in 2008
or 2009. We therefore use BGT data starting from 2010.
7 O*Net is a free online database that contains hundreds of occupational definitions to help students,
job seekers, businesses and workforce development professionals to understand today’s world of work in
the United States. It was developed under the sponsorship of the US Department of Labor/Employment
and Training Administration (USDOL/ETA) through a grant to the North Carolina Employment Secu-
rity Commission (now part of the NC Commerce Department) during the 1990s.
8 Technically, their measure is at the 6-digit SOC hybrid level defined by the Bureau of Labor Statistics
for the Occupational Employment Survey (OES). The BLS implemented this slightly altered taxonomy as
an interim solution between the switch from the 2010 SOC system to the 2018 SOC system, which will be
fully used for the 2020 OES. The BGT data uses the 6-digit 2010 SOC system, for which a crosswalk to the
6-digit hybrid system is readily available on the BLS website: https://www.bls.gov/oes/soc_2018.htm.
8
layoffs and furloughs during the pandemic. Again, for our main question in which
we are interested in the effect of pre-pandemic WFH feasibility (firm resilience),
this is not a concern.9 Therefore, in the subsequent sections, high WFH index
always refers to a high pre-pandemic WFH index, unless otherwise specified.
B. Summary Statistics
Following the procedure in section II.A, we are able to match over 3,800 unique
firms in Compustat. For instances where a Compustat firm has multiple sub-
9 However, this does matter for studying how firms dynamically adjust their hiring during the pan-
demic. Firms with low pre-pandemic WFH indices, particularly in non-essential industries, likely had
stronger pressure to fire workers in occupations that were not WFH feasible or hire more workers in oc-
cupations that permit remote work to continue operation. Many firms were forced to alter their business
operations entirely towards higher work-from-home feasibility as can be seen in schools or management
consulting. Since we do not observe the number of layoffs or reductions in hours, our WFH index during
the pandemic is a noisy signal of resilience. A decline in a firm’s WFH index during the pandemic may
even be a sign of recovery as the firm begins to rehire previously fired workers in non-WFH occupations.
Indeed, we do observe these trends and further analyze the recovery dynamics in a parallel study. De-
spite the potential endogeneity of the WFH indices during the pandemic, the pre-pandemic WFH indices
are still an exogenous shifter of firms’ pandemic performance, because the pandemic was unanticipated
and significant change in WFH feasibility for large public firms through hiring is highly unlikely in the
short-term (Ilut, Kehrig and Schneider, 2018).
10 We refer interested readers to Campello, Gao and Xu (2019) for a more detailed description of this
part of the matching exercise.
DIGITAL RESILIENCE: FIRM ESTIMATES 9
sidiaries, our firm-level WFH index uses a weighted average of all the subsidiaries’
WFH indices, where the weight is each subsidiary’s number of job postings. After
the cleaning process,11 our final analytical sample includes 9,550 observations cor-
responding to 2,176 unique public firms. The average WFH index by sectors based
on firms in our sample is presented in Figure 2. Information, finance, education,
and professional services are among the sectors with the highest WFH indices
while retail, health care, and accommodation sectors have the lowest average
WFH indices for both 2019 and 2020, consistent with the general presumption.
Table 1 Panel A shows that the mean value of the WFH index is 0.575 with a
standard deviation of 0.286, which indicates a sizeable amount of variation in the
cross section. Panel B further breaks down the key variables into pre- and post-
Covid-19 periods and contrasts their values between subsamples of firms in the
top quarter pre-Covid-19 WFH index (HighWFH=1) and the rest (HighWFH=0).
We observe a differential impact of Covid-19 on high- and low-WFH firms. For
instance, while the two groups of firms have similar average quarterly stock returns
before the pandemic (7.0% vs. 6.9%), the high-WFH firms experience a relatively
higher cumulative return during the period after the breakout of the disease than
low-WFH firms (4.3% vs. -1.5%). Similarly, while the average net income of
high-WFH firms increased modestly from $281.7 million in 2019 to $316.8 million
in 2020, the measure for low-WFH firms has only increased from $199.3 million to
$212.5 million over the same period. This indicates a significantly higher growth
of net income (12.5% vs. 6.6%) between the high-WFH firms and the rest of the
firms during the same period. Although large in magnitude, these differences in
means could be driven by the distribution of WFH indices within and/or across
industry and therefore demand multivariate analyses, which we turn to in the
next section.
A. Empirical Methodology
where subscript i and t index firm and time (i.e., quarter), respectively.12 Yit
specifies the firm-level outcome variables for both financial and stock market per-
formance. In particular, we examine sales, net income, total capital investment,
11 We require all observations to have non-missing information for key variables for our empirical
exercise and hence have less unique firms in our analytic sample than in the matched sample.
12 In particular, by including firm and time fixed effects, we already control for individual firms’
indicators for belonging to the top WFH quartile (HighW F Hi ) as well as for the indicators for time
periods during the pandemic (Covidt ).
10
B. Empirical Results
and consistent using only firms with a history of reporting their capital investment in software. Results
are available upon request.
15 Return volatility and idiosyncratic volatility are calculated as the standard deviation of daily returns
over a quarter and that of residuals obtained from fitting a CAPM using daily returns in a quarter. Our
construction of idiosyncratic volatility is in line with Andrew Ang, Robert J Hodrick, Yuhang Xing and
Xiaoyan Zhang (2006). Table 1 Panel A details the variable definitions.
12
somewhat worse performance prior to the pandemic with a sharp turning point
between 2019 Q4 and 2020 Q1, which coincides with the timing of the pandemic
and thus confirms the non-existence of pre-trends.
Second, our identification strategy exploits the fact that the pandemic was
unanticipated and that some firms happened to be more resilient than their peers
due to their pre-Covid hiring. There is still a possibility that performance dif-
ferences during the pandemic are driven by other, unobserved firm-level char-
acteristics that our resilience measure does not capture. We thus perform both
propensity score matching and coarsened exact matching to construct samples
that consist of similar firms (using pre-pandemic observables) with high and low
WFH indices. More specifically, we matched the firms using return on asset
and size prior to the pandemic within 2-digit NAICS level industries based on
each outcome variable.16 Within this matched subsample, we find consistent and
robust results that are similar to our main findings.17
Lastly, we consider an alternative WFH proxy (WFH index) by using a con-
tinuous WFH index instead of an indicator as our key explanatory variable. We
also test an alternative specification augmenting the time fixed effect with state-
quarter fixed effects in Appendix E.
In all aforementioned robustness tests, we find that our baseline results are fully
retained. These results can be found in the appendix.
C. Industry Heterogeneity
So far, we reported the average effects of high WFH feasibility on firm perfor-
mance using the entire analytic sample. In this section we explore specific groups
of industries in more detail.
volatility than their peers, the effect of high WFH feasibility is more moderate
in essential industries. This is consistent with the argument that pre-pandemic
WFH suitability within non-essential industries played a crucial role for firms re-
silience and their ability to effectively continue their operation. In addition, our
results indicate that the significant differences in capital investments on software
primarily load on firms in non-essential industries where the continuation of oper-
ation hinges on such investments more strongly. This finding further corroborates
the notion of complementarity between work-from-home practices and IT.19
IV. Conclusion
The digital tools and workforce capability to work from home have the potential
to enable firms to maintain enterprise value despite the challenges of the Covid-
19 We further explore the heterogeneous effects of the WFH on firm resilience during the Covid-19
pandemic by breaking down our sample by sectors. Using the baseline specification similar to those in
Tables 2 and 3, we estimate the marginal effects of the interaction term between the high WFH indicator
and the pandemic indicator and plot them in Appendix F Figure F1. We find that the identified effect
of WFH loads primarily on manufacturing, finance, insurance and real estate, and service sectors. The
definitions of sectors are reported in Appendix H1.
20 The high-tech industry is defined as in Decker et al. (2017) and can be found in Appendix Table H2.
21 The mean and median of WFH indices for high-tech industries are 0.71 and 0.78 respectively with a
standard error of 0.25 while the mean and median of WFH indices for other industries are 0.55 and 0.56
with a larger standard error of 0.29.
14
19 pandemic. Firms with greater WFH opportunities can provide more safety
for their workers, and thereby increase the resilience of the firm’s operations.
However, large-scale evidence on the actual effectiveness of WFH practices, is
scarce.
In this paper, we exploit the near-universe of U.S. job postings from the BGT
database to construct a novel firm-level WFH index to give us insight into the re-
silience this capability creates. We find that firms with high pre-pandemic WFH
index values before the unanticipated crisis performed significantly better during
the crisis compared to their peers on several dimensions ranging from financial
performance, such as sales and net income, to stock returns and return volatility.
The magnitude and significance of our results are robust to a range of robustness
checks including tests for pre-trends, propensity score and coarsened exact match-
ings, continuous and dummy measures of WFH index, and the inclusion of state
and quarter fixed effects. Moreover, we find evidence that non-essential industries,
which were more vulnerable to government mandated restrictions on traditional
work arrangements, benefited significantly more compared to essential industries.
Non-high-tech industries also benefited more from high pre-Covid-19 WFH feasi-
bility compared to high-tech industries during the pandemic. Our study provides
some of the first evidence on how WFH practices helped firms cope with major
adverse social and economic shocks and quantifies the magnitudes and significance
of such effects within and across industries.
Our results imply that the pandemic further increased within-industry inequal-
ity. This is because larger firms (including ”digital superstars”) tend to be more
IT-intensive (e.g. Amazon, Microsoft, Apple, IBM, Google, and Facebook to
name a few) and are more likely to be in the high pre-pandemic WFH group.
They therefore suffered significantly less than smaller firms. It thus seems likely
that the pandemic further exacerbated the rising issue of market concentration.
Further investigation into the exact mechanisms through which WFH practices
are linked to performance is a fruitful area for future research. A deep understand-
ing of these underlying issues is particularly valuable to ensure a more efficient
and smooth adoption of and transition into WFH as employers and employees
alike embrace the new reality in the aftermath of the crisis. Striking the right
balance between cost efficiency and supporting the firms and individuals who suf-
fered the most from the pandemic will be critical to ensure a faster and equitable
recovery.
REFERENCES
Acharya, Viral V, and Sascha Steffen. 2020. “The Risk of Being a Fallen
Angel and the Corporate Dash for Cash in the Midst of COVID.” The Review
DIGITAL RESILIENCE: FIRM ESTIMATES 15
Bloom, Nicholas, James Liang, John Roberts, and Zhichun Jenny Ying.
2015. “Does Working from Home Work? Evidence from a Chinese Experi-
ment*.” Quarterly Journal of Economics, 130(1): 165–218.
Brynjolfsson, Erik, and Andrew McAfee. 2014. The second machine age:
Work, progress, and prosperity in a time of brilliant technologies. WW Norton
& Company.
Campello, M., J. Gao, and Q. Xu. 2019. “Personal Income Taxes and La-
bor Downskilling: Evidence from 27 Million Job Postings.” Kelley School of
Business Research Paper.
Collins, Michael. 2005. “The (not so simple) case for teleworking: A study at
Lloyd’s of London.” New Technology, Work and Employment, 20(2): 115–132.
Deming, David, and Lisa B Kahn. 2018. “Skill Requirements across Firms
and Labor Markets: Evidence from Job Postings for Professionals.” Journal of
Labor Economics, 36(S1): S337–S369.
De’, Rahul, Neena Pandey, and Abhipsa Pal. 2020. “Impact of digital surge
during Covid-19 pandemic: A viewpoint on research and practice.” Interna-
tional Journal of Information Management, 55(Impact of COVID-19 Pandemic
on Information Management Research and Practice: Editorial Perspectives).
Dingel, Jonathan, and Brent Neiman. 2020. “How Many Jobs Can be Done
at Home?” Becker Friedman Institute White Paper, , (March): 16–24.
Ding, Wenzhi, Ross Levine, Chen Lin, and Wensi Xie. 2020. “Corpo-
rate Immunity to the COVID-19 Pandemic.” National Bureau of Economic
Research.
Forsythe, Eliza, Lisa B Kahn, Fabian Lange, and David Wiczer. 2020.
“Labor demand in the time of COVID-19: Evidence from vacancy postings and
UI claims.” Journal of public economics, 189: 104238.
Gajendran, Ravi S, and David A Harrison. 2007. “The Good, the Bad, and
the Unknown About Telecommuting: Meta-Analysis of Psychological Media-
tors and Individual Consequences.” Journal of Applied Psychology, 92(6): 1524–
1541.
Notes: This figure plots the average quarterly WFH index for firms with high vs. low pre-Covid-19
WFH feasibility index from 2019 Q2 to 2020 Q3. The values in 2019Q1 are used as the baseline group.
Reported results are based on the specifications using average quarterly WFH index controlling for time-
invariant firm unobservable and other firm controls. The WFH index is calculated based on our analysis
sample. The source data comes from the Burning Glass Technologies (BGT) job vacancies data and the
occupation level WFH feasibility indicator by Dingel and Neiman (2020).
20
Notes: This figure plots the average WFH index by sector in both 2019 Q1-Q4 and 2020 Q1-Q3. The
average WFH index is calculated based on our analysis sample. The source data comes from the Burning
Glass Technologies (BGT) job vacancies data and the occupation level WFH feasibility indicator by
Dingel and Neiman (2020).
DIGITAL RESILIENCE: FIRM ESTIMATES 21
Notes: Panel A reports variable definitions, data sources, and sample means. The sample period is
2019Q1-2020Q3. Panel B contrasts the means of key variables in pre- and post- Covid-19 subsamples
(i.e., 2019 Q1-Q4 and 2020 Q1-Q3, respectively) and in the high and low pre-Covid-19 WFH subsamples.
All continuous variables are winsorized at the 1st and 99th percentiles.
22
Notes: This table implements a difference-in-differences (DID) research design to examine the differential
impact of Covid-19 on various performance metrics between firms with high and low WFH feasibility.
The dependent variables across columns 1 to 4 are the logarithm of net income, logarithm of sales,
total capital expenditure over total asset, and software expenditure over total asset, respectively. The
regression model is specified in Equation 1, and all variables are defined in Table 1 Panel A. Omitting
Tobin’s q and ROE in the specification provide similar results (available upon request). Firm and time
fixed effects are also included in the estimation. Standard errors are clustered at the firm level.
DIGITAL RESILIENCE: FIRM ESTIMATES 23
Notes: This table implements a difference-in-differences (DID) research design to examine the differential
impact of Covid-19 on stock market reactions between firms with high and low WFH feasibility. The
dependent variables are total return and abnormal return in Columns (1) and (2); and return volatility
and idiosyncratic volatility in Columns (3) and (4), respectively. The regression specification is specified
in Equation 1, and all variables are defined in Table 1 Panel A. We further include Tobin’s Q and
return on equity (ROE) in the regression analysis. Firm and time fixed effects are also included in the
estimation. Standard errors are clustered at the firm level.
24
Appendix A
Notes: This table re-assesses the results reported in Tables 2 and 3 in two subsamples: essential vs. non-
essential industries. The definitions of essential and non-essential industries are based on (Papanikolaou
and Schmidt, 2020) and can be found in Appendix Table H3. Panels A and B report the results using
financial performance proxies as the dependent variables. Panels C and D repeat the exercise with
return and volatility as the dependent variables. The specifications in these panels are identical to those
in Tables 2 and 3. Firm and time fixed effects are also included in the estimation. Standard errors are
clustered at the firm level.
DIGITAL RESILIENCE: FIRM ESTIMATES 25
Notes: This table re-assesses the results reported in Tables 2 and 3 in two subsamples: high tech vs.
other industries. The specifications in these panels are identical to those in Tables 2 and 3. The high-tech
industries are defined as in Decker et al. (2017) and can be reviewed in Appendix Table H2. Panels A
and B report the results with financial performance proxies as the dependent variables. Panels C and D
repeat the exercise with return and volatility as the dependent variables. Firm and time fixed effects are
also included in the estimation. Standard errors are clustered at the firm level.
26
Appendix B
(A) Dynamics of the treatment effects in the DID setting on stock returns and
volatility
Notes: Panels A and B plot the impact of HighWFH on firm outcome variables (A: financial perfor-
mance; B: stock returns and volatility) in the period of 2019Q2-2020Q3 with the 90% confidence interval
attached. Results are based on specifications similar to the baseline models: we regress each outcome
variable on the interaction of HighWFH and the quarter dummy alone with other firm controls. The
values in 2019Q1 are used as the baseline for comparison.
DIGITAL RESILIENCE: FIRM ESTIMATES 27
Appendix C
(C) Stock market performance: matched sample analysis using PSM procedure
Models (1) (2) (3) (4)
Dependent Variable Return Abn Return Volatility Idio. Volatility
HighWFH × COVID 0.063 0.051 -0.006 -0.004
(4.59) (3.73) (-2.68) (-2.47)
Observations 5197 5197 5279 5172
Adj. R2 0.447 0.132 0.702 0.739
(D) Stock market performance: matched sample analysis using CEM procedure
Models (1) (2) (3) (4)
Dependent Variable Return Abn Return Volatility Idio. Volatility
HighWFH × COVID 0.075 0.065 -0.005 -0.003
(5.51) (4.87) (-2.61) (-2.29)
Observations 6863 6863 6862 6837
Adj. R2 0.475 0.155 0.708 0.717
Notes: Panels A and B report the matched sample results for financial performance proxies where
the matching procedure is the propensity score matching (PSM) and coarsened exact matching (CEM),
respectively. For results in Panel A, we estimate a probit model where the dependent variable is HighWFH
and the controls include pre-Covid-19 return on asset (operating income / total asset), employment, and
industry (2-digit NAICS) fixed effects using the pre-Covid-19 sample. We further implement a K2K
nearest neighbor matching with no replacement and common support for each of the outcome variables.
The results in Panel B are based on the coarsened exact matching approach using the same set of pre–19
variables. Panels C and D conduct the same matched sample analysis for stock performance. Standard
errors are clustered at the firm level.
28
Appendix D
Notes: Our baseline analysis is based on HighWFH, which takes the value of one if a firm’s average
WFH index calculated based on its annual job posting data during the pre-Covid-19 period (2010-2019)
falls into the top quartile of the sample distribution and zero otherwise. In Panels A and B, we repeat
the analysis using the raw, continuous WFH index. Firm and time fixed effects are also included in the
estimation. Standard errors are clustered at the firm level.
DIGITAL RESILIENCE: FIRM ESTIMATES 29
Appendix E
Notes: In Panels A and B, we repeat the analysis in Tables 2 and 3 using similar specifications but
control for state-quarter fixed effects instead of quarter fixed effects to address the potential concern
of geographic differences. Firm fixed effects are also included in the estimation. Standard errors are
clustered at the firm level.
30
Appendix F
Notes: This figure plots the impact of HighWFH on firm outcome variables including financial perfor-
mance and stock returns by sector with the 90% confidence interval attached. Reported results are based
on the specifications similar to the baseline models: we regress each outcome variable on the interaction
of HighWFH and the COVID dummy alone with other firm controls controlling for firm and quarter
fixed-effects. We aggregate industry sectors following (David Autor, David Dorn, Lawrence F Katz,
Christina Patterson and John Van Reenen, 2020), which can also be found in Appendix H4. Please see
the U.S. Census for the definition of 2-digit NAICS industries in Appendix H1.
DIGITAL RESILIENCE: FIRM ESTIMATES 31
Appendix G
Notes: In Panels A and B, we repeat the analyses from Tables 2 and 3 to run a placebo test, in which
instead of using the WFH index, we construct a firm index based on the firm names in the alphabetical
order prior to Covid-19 pandemic. All specifications here are identical to the baseline specifications in
Tables 2 and 3. Robust standard errors are clustered at the firm level.
32
Appendix H