Treasury Job Preservation PPP 2020 01b

Summary

Office of Economic Policy Working Paper 2020-01B, The Job-Preservation Effects of Paycheck Protection Program Loans, by Michael Faulkender, Robert Jackman and Stephen Miran, dated July 2024 with an original version from December 2020. The paper states that its views are the authors' and not official Treasury positions. Using county-level weekly unemployment insurance data and variation in local banking markets and loan approval speed, it estimates that a 10 percentage point increase in PPP payroll coverage at sub-100 employee businesses suppressed initial claims by 1.0 percentage points. The authors estimate that PPP loans saved 10.9 million jobs at sub-100 employee businesses and 14.0 million overall, at about $33,200 to $37,600 per job saved. The paper closes with regression tables and notes on its covariates.

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

Office of Economic Policy
Working Paper 2020-01B
July 2024
Original Version: December 2020

The Job-Preservation Effects of Paycheck
Protection Program Loans
Michael Faulkender, Robert Jackman, and Stephen Miran

This Office of Economic Policy Working Paper presents original research by the staff of the Office of Economic
Policy. It is intended to generate discussion and critical comment while informing and improving the quality
of the analysis conducted by the Office. The paper is a work in progress and subject to revision. Views and
opinions expressed are those of the authors and do not necessarily represent official Treasury positions or
policy. Comments are welcome, as are suggestions for improvements, and should be directed to the authors.
This Working Paper may be quoted without additional permission.


The Job-Preservation Effects of Paycheck Protection
Program Loans*
Michael Faulkender„
Robert Jackman
Stephen Miran§
July 25, 2024

Abstract
The Paycheck Protection Program (PPP) supported over 60 million jobs through August 2020.
How many of those jobs would have otherwise been lost? We estimate the number of jobs saved
by leveraging the relationship between local banking markets and the average speed to loan
approval. With county-level weekly unemployment insurance (UI) data, we estimate that a 10
percentage point increase in PPP payroll coverage at sub-100 employee businesses led to a 1.0
percentage points suppression of initial UI claims. That same increase suppressed the insured
unemployment rate (IUR) by 2.5 percentage points. In aggregate, we estimate that PPP loans
saved 10.9 million jobs at sub-100 employee businesses and 14.0 million overall.

JEL Codes: E0, G0, H0, J0.
*
We are especially grateful to Samuel Brown, Steven Johnson, Matthijs Schendstok, Antoinette Schoar, and two
anonymous referees for many helpful comments and assistance with analysis. We are also grateful to Tanner Black,
John Friedman, Jonathan Greenstein, Andrew Martinez, Mitchell Petersen, Jim Poterba, Josh Rauh, and Jason
Sockin for helpful feedback. We thank Leah Damelin and Cameron Greene for excellent research assistance during
their time as Treasury interns. We thank the Bureau of Labor Statistics for generating customized statistics and
providing expert advice in support of this work. Special thanks are given to Economist Jeremy Oreper and Associate
Commissioner Julie Hatch Maxfield, as well as their partners at the State Labor Market Information offices. The
findings and views are those of the authors and do not necessarily reflect the position of the United States Department
of the Treasury.
„
University of Maryland, Robert H. Smith School of Business. mfaulken@umd.edu
United States Treasury. Corresponding author: robert.jackman@treasury.gov
§
Hudson Bay Capital Management, LP and Manhattan Institute. smiran@hudsonbaycapital.com


Disclosure Statements
Michael Faulkender: I led Treasury’s efforts to support the Small Business Administration’s
implementation of PPP, in my capacity as Treasury’s Assistant Secretary for Economic Policy.
Robert Jackman: I have nothing to disclose.
Stephen Miran: I contributed to Treasury’s efforts to support the Small Business Administration’s implementation of PPP, in my capacity as Senior Advisor at Treasury’s Office of Economic
Policy.


I. Introduction
The Paycheck Protection Program (PPP) was intended to sustain U.S. small businesses and their
employees through an economic crisis.1 With the onset of Covid-19 in spring 2020, most U.S. small
businesses faced mandatory closures and drastically reduced revenue. The U.S. lost 20.5 million jobs
(13.6%) in a single month,2 while single-week initial unemployment insurance (UI) claims peaked
at 6.1 million – over 9 times more than the worst week of the global financial crisis. In the second
quarter of 2020, the economy contracted at an annualized 28.0% rate. In April 2020, the press,
markets, and academics expected conditions to deteriorate further. Widespread and permanent
small business closures seemed likely, which would result in continuing mass layoffs (Bartik et al.
(2020a), Barrero, Bloom, and Davis (2020), Humphries, Neilson, and Ulyssea (2020)).3 In response,
the Federal Government enacted PPP as part of the CARES Act on March 27, 2020.4 PPP coverage
was broad: it covered over 60 million total jobs when the second tranche closed on August 8th,
2020. While calculating the number of jobs supported by the PPP is relatively simple, calculating
the number of jobs preserved poses a harder task. This paper addresses that thornier question:
how many more workers would have been on UI, in the absence of PPP? In other words, how many
paychecks did the Paycheck Protection Program protect?
This question is empirically challenging to answer, due primarily to the absence of an obvious
observed counterfactual. The program intentionally set minimal eligibility restrictions, and nearly
all small businesses qualified for a PPP loan. With extraordinarily generous financing terms, takeup reflected the near-universal eligibility: PPP loans ultimately supported 85% of jobs at businesses
with fewer than 100 employees. Likewise, by the time of program closure in August 2020, geographic
coverage was relatively uniform across the nation. Any effort to compare some PPP “treatment”
and “control” faces the obstacle of finding businesses who weren’t truly treated, even before dealing
with selection into that treatment.
One proposed solution has been to use the PPP’s 500-employee eligibility cutoff. In most industries,
only firms with 500 or fewer employees were eligible for a PPP loan. Prominent examples include
Chetty et al. (Forthcoming, 2024), Autor et al. (2022a), and Hubbard and Strain (2020), which
estimate PPP’s employment effects by comparing employment changes at firms falling just below
the eligibility cutoff to firms just above it. When extrapolated to PPP recipients of all sizes, these
1

Throughout this paper, our discussion of PPP is restricted to the loans approved through August 8, 2020. In
our calculations and considerations, we do not account for the “second round” of PPP loans, which was authorized
in the Consolidated Appropriations Act signed into law on December 21, 2020.
2
Though commonly attributed to April 2020, this number reflects the change in total nonfarm payrolls from the
week of March 12, 2020 to the week of April 12, 2020.
3
Similar sentiment was widespread in the popular press (e.g., Cohen (2020) and Irwin (2020)) and among professional forecasters. The Bloomberg median forecast for May 2020 payrolls was a loss of 7.5 million further jobs, with
the most optimistic forecast showing a loss of 2.5 million.
4
CARES is an acronym for Coronavirus Aid, Relief, and Economic Security.

1


estimates imply at 3.6 million or fewer jobs were saved by PPP. Considering the program’s overall
cost—$525 billion—this would be a relatively inefficient use of government funds.
Such an approach assumes that PPP loans were equally effective in saving jobs, regardless of firm
size. We challenge that assumption. Decades of finance and economics literature has documented
that smaller firms, like lower-income individuals, have weaker access to capital, face stricter credit
constraints, maintain smaller cash reserves and credit lines, and are more vulnerable to economic
shocks than their larger counterparts (e.g., Duygan-Bump, Levkov, and Montoriol-Garriga (2015);
Siemer (2019); Petersen and Rajan (1997)).
The difference in credit access via banks, long-established, was particularly acute during the Covid
crisis (Chodorow-Reich et al. (2022)). Further, evidence from the 2008-2009 financial crisis suggests
that credit frictions are directly linked to employment outcomes, particularly for the smallest firms.
For example, Chodorow-Reich (2014) finds that credit loss accounted for between one-third and
one-half of job loss at small firms through that crisis - while finding no significant evidence of an
impact at larger firms. According to this literature’s results, the effects of the pandemic recession
should have been much greater on firms with fewer than 100 employees—which accounted for 68.5%
of the jobs covered by PPP loans—than those just below 500 employees. However, if that is not the
case—if firms of all sizes would have suffered relatively moderate losses in the absence of PPP—it
is a great challenge to this literature.5
A complete study of PPP must estimate the job effects across the entire distribution of firm sizes.
Therefore, we propose a different source of variation to identify the employment effects of PPP loans.
Specifically, we use delays in loan approval to evaluate the difference between firms and regions
that enjoyed earlier PPP financing to those that experienced delays. However, an identification
problem remains: differences in timing alone are insufficient, since small business finances and local
employment are not independent. For example, stronger businesses are more likely to have wellestablished relationships with banks, which would increase the likelihood of rapid loan approval
and reduce the likelihood of immediate layoffs.
To address the endogeneity arising from the financing process, we add geographic variation stemming from heterogeneity in local banking markets. Differences in banking market structure help
to isolate an exogenous component of loan timing: due to differing perceptions of regulatory risk,
community banks were markedly quicker to approve and disburse first tranche PPP funds than
national banks and non-bank lenders. Financing delays during the first tranche of funding (from
April 3 – 16, 2020) were exacerbated by the exhaustion of PPP’s initial appropriation of roughly
$350 billion. Firms without an approved loan on April 16 would have to wait until at least April

27, when a second tranche of funding re-opened the program.6
5

The suggested loss of 3.6 million jobs at firms sized 0-499 would be roughly half the loss suffered during the Great
Recession.
6
See Doniger and Kay (2023) and Kurmann et al (2024) for an evaluation of PPP’s effects using this gap as an

2


Under the condition that the composition of the local banking market has no relationship with
early pandemic changes in local employment—save through PPP loans—this variation can identify
the employment effects of PPP loans. This empirical approach—leveraging geographic variation in
banking markets which generate differences in time to loan approval—is shared by Granja et al.
(2022) and Bartik et al. (2020c). Though the approaches in these papers differ on crucial details,
at a high-level their identification strategies are related.
This paper’s core contribution is the estimation of PPP’s employment effects across firms of different sizes. We leverage novel data on UI claims to disambiguate PPP’s effects on the “smallest”
businesses (1 to 99 employees) from the effects on larger businesses (100 to 499 employees). Not
only do we observe UI claims at the county-week level, but we can split these claims into three bins,
based on the number of firm-level employees at each claimant’s former employer.7 We are thus able
to precisely observe increases in unemployment stemming from job cuts at PPP’s target firms, as
opposed to the noisier measure of all unemployment claims. Further, we observe both initial and
continuing UI claims, giving a measure of both job loss (initial claims), and the persistence of job
loss (continuing claims). With uncensored data on the universe of PPP loans, we are also able to
match these UI data to precise measures of the program’s rollout.
This paper’s core result is that PPP saved substantially more jobs than has been previously suggested in the literature. Further, those savings were concentrated at the smallest firms, consistent
with and contributing to the literature on firm size and resilience. When restricting our focus to
firms with fewer than 100 employees, we find that a 10 percentage point increase in early-April
PPP coverage suppressed the jump in their employees’ initial claims rate by 1.03 percentage points
during a single week. The same increase in payroll coverage suppressed the insured unemployment
rate (IUR, or “continuing claims rate”) among small business employees by 2.52 percentage points.
Expanding that focus to all firms with fewer than 500 employees, a 10 percentage point increase
in early-April PPP coverage suppressed the jump in their employees’ initial claims rate by 1.14
percentage points during a single week. With a lag in time, the same increase in PPP coverage
suppressed the increase in continuing claims rate by 2.38 percentage points, also among workers at
sub-500 employee firms.
Though the difference between the two estimated continuing claims rates – one for sub-100 employee
firms, the other for all sub-500 employee firms – is seemingly small and insignificant, it implies a
meaningful gap in PPP’s effectiveness. As a fraction of total pre-pandemic employment, this gap
indicates that PPP preserved jobs at the smallest firms (0-99) at a rate between 5.5 and 7.1
percentage points greater than it did at mid-sized firms (100-499). The difference between the two
continuing claims rate estimates increases through April and May 2020, reinforcing the importance
exogenous source of funding delays.
7
Based on pre-pandemic employment. Those bins are for firms with 1 to 99 employees, those with 100 to 499
employees, and those with 500 or more.

3


of firm-size heterogeneity.
Extrapolating these estimates, our results suggest that PPP saved 10.9 million jobs at sub-100
employee businesses and 14.0 million jobs overall, at an average cost of approximately $33,200 to
$37,600 per job saved.

The rest of this paper is organized as follows: Section II details the context and background of
PPP loans. Section III engages further with PPP literature. Section IV offers back-of-the-envelope
calculations, with the intent of establishing a reasonable expectation of potential estimates. Section
V reviews the data used, while Section VI discusses the empirical model, exclusion restriction, and
results. Section VII concludes.

II. The Paycheck Protection Program
In January 2020, the CDC announced the first case of COVID-19 had been diagnosed in the United
States. In early March, the White House National Economic Council convened an inter-agency
working group to evaluate the economic impacts of COVID-19 and generate policy proposals to
ease the economic hardship that would result. Michael Faulkender, in his capacity as Assistant
Secretary for Economic Policy at the time, represented Treasury on this working group and led the
PPP implementation team at the US Treasury. Stephen Miran was part of that implementation
team. Therefore, the narrative in this section provides a unique primary source explanation for
why the program was structured and implemented as it was.
Given the pace with which the public health crisis was building, Congress and the Administration
prioritized speed over precision in their economic support programs. Delays, whether due to time
spent on refining targeting or fraud mitigation, might have produced a better program in those
respects; but as we document in this paper, it also would have resulted in significantly higher
unemployment levels.
PPP was structured as a modification of the Small Business Administration’s (SBA) existing 7(a)
loan program and provided small businesses with forgivable loans to eligible small businesses, sole
proprietors, and small non-profits equivalent to 2.5 months of average monthly payroll. Borrowers
were able to apply for these 100% SBA guaranteed loans from eligible lenders (banks, credit unions,
farm credit institutions, and non-bank lenders, ultimately totaling more than 5,400 institutions)
by filling out a two-page application on which they self-certified that they met most of the eligibility criteria. Their lenders then ensured that Bank Secrecy Act and Anti Money Laundering
requirements were met and verified the amount of eligible monthly payroll.
If borrowers used all the money on payroll, utilities, mortgage interest, and rent within the covered
period (originally eight weeks but later extended to up to 24 weeks), with at least 75% going

4


towards eligible payroll (later reduced by Congress to 60%), the full amount of the loan could be
forgiven. Congress originally appropriated $350 billion to fund PPP loans. Within two weeks of
the program opening, these funds were fully allocated. Congress appropriated another $310 billion
of funding and the program reopened on April 27, 2020.
The program was generous for borrowers, provided that they used the funds for forgivable expenses.
For the program to work, lenders also needed to voluntarily participate. This was accomplished
through a statutorily defined fee structure on the loans that paid five percent of the loan amount
on loans up to $350,000, three percent on loans from $350,000 to $2 million, and one percent on
loans above $2 million. While the original statute did not stipulate the interest rate on the loan
or the specific maturity, Treasury and SBA issued rulings that all PPP loans would share common
terms. For that reason, borrowers gained nothing from shopping around for their loan, knowing
instead they were getting the same terms from all lenders.
Policymakers set interest rates to cover the lenders’ cost of funds and any ongoing expenses associated with billing and collecting from the borrower. A loan fee would cover the cost of underwriting
on the front end and processing of forgiveness on the back end. Both interest and the loan fee were
covered by SBA as a forgivable expense, so any money spent on interest took away from the finite
appropriation that was available to fund payroll. Therefore, Treasury and SBA wanted to set the
interest rate as low as possible, but high enough to ensure lender participation. For large lenders
with access to the Fed, the Federal Funds rate was 0.0 to 0.25 percent at the time. For smaller
lenders, Treasury worked with the Federal Reserve to create a facility that would allow the banks
to use their PPP loans as collateral to obtain additional funds. The Fed facility would charge 35
basis points. Given that the highest cost of funds for lenders should only be 35 basis points, the
uniform interest rate on PPP loans was set at one percent.
Because the program launched so quickly, the rules and list of frequently asked questions (FAQs)
were incomplete. Within days of passage, a term sheet and initial borrower and lender applications
were published by the SBA in cooperation with the Department of the Treasury. The first interim
final rule (IFR) was released on April 2, 2020. It explained the eligibility requirements of borrowers,
the obligations of lenders, and described the forgiveness process. The forgiveness application and
accompanying IFR were issued in May 2020. As borrowers, lenders, and the media started interacting with the program, additional issues arose requiring SBA and Treasury to regularly publish
updates and additional rules as well as answer questions regarding situations lenders or borrowers
may find themselves confronting.
The result of the ongoing updating of the rules and requirements was that there was variation across
lenders in the speed with which they participated in the program. Some lenders were more willing
than others to begin accepting and processing loans in early April as they needed to establish
their own policies and systems for providing these loans and to train their personnel on these

5


policies and systems. As we document in section VI.A.ii, community banks were the most likely to
participate in the program earlier, while the largest banks mostly sat out the program’s first week.
These empirical observations of program uptake are supported by the authors’ conversations with
lenders8 , which featured a reticence on the part of many of the largest banks to start issuing loans
until there was greater clarity on the rules and a better understanding of the legal standards to
which the lenders would be subject. In contrast, smaller lenders felt that they were less likely to
be targeted by regulators for minor processing errors and they commenced issuing PPP loans to
their customers as soon as the program opened. These subjective differences in regulatory risk led
to substantial differences in initial program participation, and play a critical role in our empirical
strategy. We argue that this dynamic created a source of variation in the supply of PPP funds that
should be orthogonal to demand for these loans from borrowers.

III. Literature
Three significant studies, Chetty et al. (Forthcoming, 2024), Autor et al. (2022a), and Hubbard
and Strain (2020) use innovative high frequency, large datasets to study the effects of PPP. In most
industries, only firms with fewer than 500 employees were eligible to receive PPP money, and these
studies leverage this eligibility cutoff, comparing employment at firms just above and below that
threshold. Autor et al. (2022a) estimate that PPP preserved 3.6 million jobs at its peak, while
Chetty et al. (Forthcoming, 2024) and Hubbard and Strain (2020) estimate smaller effects.
In contrast, some market participants have argued that the PPP was pivotal in mitigating the pademic’s potential economic devastation. Jamie Dimon of JPMorgan Chase stated that he estimates
the program to have saved “30 to 35 million jobs” (Ruhle, Miranda, and Capetta (2020)) while the
chief economist at Standard & Poors (Fox et al. (2020)) publicly stated that 13.6 million jobs were
saved. Goldman Sachs’ chief US economist stated that PPP was a prime factor preventing what
“really seemed like it had the potential to be a huge collapse...[the lack of bankruptcies] has come
as a pleasant surprise”. How do we reconcile the significant differences in the estimated impact of
the program between recent academic studies and market participants?
One explanation is that the program’s impact was not uniform across firms of different sizes.
While the empirical design of Autor et al. (2022a) and Hubbard and Strain (2020) is credible in
the neighborhood of the eligibility cutoff, projecting those estimates onto smaller firms raises the
question of external validity. Estimates based on larger firms would apply to smaller firms only if we
believe large and small firms were equally vulnerable to the pandemic’s economic shock. However,
decades of economic literature document that smaller firms are more sensitive to economic shocks
than larger ones. Larger firms tend to have more sophisticated managers, larger cash buffers, more
8

Occurring in the context of policymakers speaking to banks, not as researchers speaking to banks.

6


ready access to established lines of credit, fixed contracts with customers, and more bargaining
power with suppliers. This puts larger firms in a stronger position to weather economic shocks
than smaller businesses. Because of their superior ability to access lending and financial markets,
larger firms were able to benefit from other treatments not available to smaller firms, like the Federal
Reserve’s interventions. If the equal vulnerability assumption is violated, then these studies can
only tell us what happened in the neighborhood of those large firms.
This is particularly an issue for PPP because smaller firms (fewer than 100 employees) comprise
the vast majority of program participants. Based on comprehensive PPP loan microdata, Figure
I displays the density of PPP loan recipients based on firm size, as measured by self-reported
employment on loan applications. We make some limited adjustments to these data, as detailed
in appendix A. Table I shows the fraction of PPP loans that went to firms of various sizes, by
number of loans, number of workers covered, and total PPP dollars. The firm-size distribution of
PPP loans is roughly proportional to the firm-size distribution itself – meaning that smaller firms
took up most of the program’s resources. One of this paper’s empirical estimates highlights PPP’s
effects on firms with fewer than 100 employees, and that group of firms received 69.2% of all dollars
loaned. Similarly, PPP loans covered 41.5 million workers at sub-100 employee firms (68.5% of
the total). Since PPP dollars skewed towards smaller firms, we must know PPP’s effects on small
firms, if we are to accurately estimate program effectiveness.
TABLE I
PPP Loans and Firm Sizes, by Three Measures

Firm Size
≥ 500
≥ 450
≥ 400
≥ 350
≥ 250

No. of Loans
Total Dollars
Panel A. Larger Firms
3,016 (0.1%)
$15.7 (3.0%)
5,494 (0.1%)
$25.3 (4.8%)
8,108 (0.2%)
$33.9 (6.5%)
11,356 (0.2%)
$43.3 (8.3%)
22,786 (0.4%)
$70.8 (13.6%)

< 250
< 150
< 100
< 50
< 25
< 10
<5

Panel B. Smaller Firms
5,113,576 (99.6%)
$451.1 (86.4%)
5,083,959 (99.0%)
$403.6 (77.3%)
5,043,726 (98.2%)
$361.0 (69.2%)
4,914,410 (95.7%)
$282.3 (54.1%)
4,630,653 (90.2%)
$198.7 (38.1%)
3,858,764 (75.1%)
$103.4 (19.8%)
2,904,529 (56.5%)
$52.2 (10.0%)

Total Workers
1.9 (3.2%)
3.1 (5.1%)
4.2 (6.9%)
5.4 (8.9%)
8.7 (14.4%)
52.0 (85.6%)
46.4 (76.4%)
41.5 (68.5%)
32.7 (54.0%)
23.1 (38.0%)
11.6 (19.1%)
5.3 (8.8%)

Source: SBA and author’s calculations. Data as of 8 August, 2020. Firm Size is as reported by the
SBA, with adjustments as reported in appendix A.

Mindful of the differences between small and mid-sized firms, we need an empirical strategy to
7


Figure I
Who Received PPP Loans? Employees and Dollars. Source: SBA and author’s calculations. Data as of 8
August, 2020. Firm Size is as reported by the SBA, with adjustments as reported in appendix A.

identify the effect across the spectrum of firm sizes. This is particularly important if we are to
evaluate PPP’s aggregate employment effect. As we have argued, the assumption that small and
mid-sized firms are equally vulnerability is unsupported by either theory or prior research. If
that assumption is violated, then Chetty et al. (Forthcoming, 2024) and Autor et al. (2022a) only
accurately tell us what happened in the neighborhood of the 500-employee cutoff. Those results
would be uninformative about the outcomes at smaller firms, who received the vast majority of
loans and dollar amount of PPP funding. As such, evaluations using the employee size cutoff likely
underestimate the true impact of PPP by a significant margin.
Autor et al. (2022b) addresses the differences between small and mid-sized firms via a differencesin-differences framework, estimated only firms of fewer than 50 employees. That paper uses a Sun
and Abraham (2020) estimator, meaning that firms are grouped by the week they took a loan,
then the coefficient for each group is separately estimated, then those estimates are averaged.9
9

In each of those individual regressions, the control group is all firms (of fewer than 50 employees) who took a
loan in the week ending June 27, 2020 through the end of the program on August 8, 2020. The critical parallel trends
assumption: firms which took a PPP loan sometime between April 3, 2020 and mid-June would have, in the absence
of PPP, followed an employment trajectory in April, May, and June 2020 equal those firms that chose to take loans

8


The estimates suggest that PPP loans temporarily increased employment by 12% at firms of fewer
than 50 employees, relative to similarly-sized firms that did not take PPP loans over the initial 2.5
months of the program.
A fourth study, conducted by Granja et al. (2022), finds precisely estimated and small effects of
PPP on hours worked, business shutdowns and UI claims. With respect to hours worked and
business shutdowns, we note that the goal of PPP was to facilitate employees being paid while not
necessarily physically showing up to workplaces, often working zero hours, and for businesses to
shut down temporarily without laying off workers, closing permanently or filing for bankruptcy. In
normal recessions, hours worked and business shutdowns are a good indicator of severity. However,
a public-health recession is unusual in many ways. Unique to these circumstances, policy was
explicitly aimed at reducing business activity and face-to-face interaction, and provided liquidity
to facilitate compliance with stay-at-home orders.
Further, Granja et al. (2022) study employment outcomes as a function of ultimate PPP penetration. However, since final PPP penetration is very high everywhere—in excess of 80% of eligible
employment is covered, nationally—there might not be enough variation in the underlying independent variable of interest to identify the key effects. Granja et al. (2022) instead use an alternative
independent variable, which is a function of bank market shares like ours below, and which likely
induces more variation than ultimate PPP penetration itself. Our approach is distinct, precisely
because it exploits the dynamic variation in PPP receipt.
Using survey data, Bartik et al. (2020c) estimates a larger employment effect for the PPP. However,
their core employment estimates are not statistically significant. Bartik et al. (2020c) uses preexisting relationships with banks to instrument for loan receipt, marking a similarity with this
paper’s use of banking market structure. The two banking-relationship instruments yield point
estimates of 3.31 and 4.99 jobs saved per loan, though the associated standard errors are also large.
If we multiply the point estimates by the roughly five million loans, this would imply between 16.6
and 25 million jobs preserved by the PPP, somewhat larger than our main results.
The work which comes closest to our own approach is that in Doniger and Kay (2023), which
exploits the discontinuity in loan receipt around a ten-day window in which the Program had
run out of funds from the first Congressional appropriation until the Program received the second
Congressional appropriation. From April 17th through 26th, no new loans were approved. Doniger
and Kay (2023) compares geographies which had large portions of loans approved between April
14th to 16th to geographies which had large portions of loans approved on April 27th and 28th,
and, like our own work, finds large and significant job preservation effects of early PPP receipt. On
the margin, Doniger and Kay (2023) find that an extra $35 billion of funding in the first tranche
in late-June through early-August 2020. This assumption stands in contrast to our paper, which argues that loan
timing is endogenous, and uses geographic differences in banking structure to instrument for loan timing.

9


would have increased employment in June by 2.8 million. That implies a marginal flow-cost of
jobs at $12,500. Doniger and Kay (2023) estimates that, on this same margin, 1.4 million jobs
were saved in each month through November, implying a marginal flow-cost of $25,000 per job.
However, these estimates are not directly comparable with this paper’s, which are an estimate of
average cost per job, not marginal cost. Further, Doniger and Kay (2023) conduct an exercise on
firm-size heterogeneity, finding that the effect was heavily concentrated in firms with fewer than 10
employees, which is consistent with our own estimates.
Using the empirical strategy from Doniger and Kay (2023), Kurmann et al (2024) find that the tenday delay of funding reduced employment in four key sectors by 3 million as of July 2020. Kurmann
et al (2024) makes a crucial point that most of the employment effect was on the extensive margin of
firm closure, a point also emphasized in Dalton (2023) and Autor et al. (2022b). Therefore, failing
to distinguish sample churn from firm closures in microdata would lead researchers to miss the bulk
of PPP’s effect. Kurmann et al (2023) also find that the ten-day delay still had a significant effect
on employment as of the end of their sample in January 2021. Using a related timing strategy,
Denes, Lagaras, and Tsoutsoura (2021) estimates that firms receiving delayed PPP loans experience
a significant increase in the likelihood of facing financial distress and of permanently shutting down.
Of note, these papers study outcomes relative to a counterfactual with no-delays implementation
of PPP, not one with no PPP at all.
Finally, work by Cole (2024) also focuses on smaller firms (median 5 employees), and finds in
data from a private payroll processor that PPP recipients increased their employment relative
to nonparticipants by 7.5% five months after loan receipt. Cole (2024) finds that PPP works
primarily through job preservation rather than via incentivizing hiring, and further finds significant
heterogeneity in PPP take-up and effectiveness. In this sample more job preservation occurs in
firms that have fewer hourly workers, are better at incorporating remote work, and are essential
businesses. By corroborating that PPP effects are much greater for smaller firms than for larger
firms, Cole (2024) helps reconcile our results to the rest of the literature.

IV. Back-of-the-Envelope Calculations
At the onset of the pandemic recession, economists anticipated high job losses for extended periods
of time. According to the BLS establishment survey (CES), 20.5 million jobs were lost in April
2020, following March’s loss of 1.5 million jobs.10 Markets expected that job losses would continue
into the summer, with concerns that the unemployment rate would exceed 20%.11 For May 2020,
10

Losses were deeper in the household survey (CPS), where employment fell by 3.3 million in March, followed by
the loss of an additional 22.3 million in April. The household survey counts the unincorporated self-employed, while
the establishment survey does not.
11
The Bloomberg median forecast for the May 2020 U3 unemployment rate was 19.7%. Instead, U3 fell 1.5
percentage points to 13.2% that month.

10


the Bloomberg median forecast projected 7.5 million additional jobs lost. The most optimistic
forecast in Bloomberg projected a loss of over two million jobs that month. The forecast error
was pervasive and historic: the economy added more than 2.6 million jobs in May, a surprise of
more than 10 million jobs. While PPP was just one part of the CARES Act, it was the largest
component by dollars appropriated and was fully implemented by the May 12 reference week
of the May employment surveys. Given the astonishing reversal in the employment situation –
immediately following the hundreds of billions of dollars of PPP lending – the claim that PPP had
minimal impact on the unexpected job rebound would require an alternate explanation with strong
evidence.
To further provide context for the more rigorous results documented below, we propose two simple
back-of-the-envelope calculations to guide thinking about job preservation through the PPP. The
first takes the threat of large-scale small business closure seriously, and estimates the job losses
that might arise from those closures. The second uses the small business survey responses from
Bartik et al. (2020b) and estimates potential job losses based on those data. Though not rigorous
empirical work, these back-of-the-envelope calculations offer a ballpark estimate of PPP’s effects –
a kind of check on empirical studies on PPP, including our own.
At the outset of the pandemic, small businesses suddenly found themselves facing significant revenue
declines that would likely cause insolvency. Smaller businesses are less likely to have access to an
established credit line; and indeed if all small businesses sought credit simultaneously, it would
strain the banking system’s capacity to respond, and at minimum would lead to a substantial
increase in the relevant interest rates.12 A study by Farrell and Wheat (2016) found that the
median small business had cash buffers to last only 27 days without revenue, and only 25% of small
businesses could last more than 62 days without income. Their sample comprised roughly 600,000
small businesses; of these, 70% had five or fewer employees, which closely matches the universe of
PPP loans wherein 70% of recipients had seven or fewer employees.
Against this backdrop, economic theory and evidence from past recessions would indicate a surge
in small business bankruptcies, particularly given the magnitude of Covid’s economic shock. Nevertheless, small business bankruptcies increased relatively modestly, despite the largest economic
shock in nearly a century, according to data from the Justice Department and analysis by the Council of Economic Advisers (2020). Indeed, contrary to expectations that small business bankruptcies
should surge in this economic environment, The Council of Economic Advisors found that after
spiking in February and March due to regulatory changes around Chapter 11 filings,13 increases
12

Although the Federal Reserve eased conditions in financial markets, it did not provide regulatory relief that would
have explicitly eased the ability or willingness of banks to directly make loans to the number of small businesses under
stress.
13
For a discussion of the rules changes, see Ekvall and Evanston (2020). Small businesses took advantage of the
easing criteria for Chapter 11 reorganizations, in the weeks before the pandemic hit. The Small Business Reorganization Act was signed in August 2019, and thus forward-looking firms had plenty of time to plan for the implementation

11


in small business bankruptcies in April through June were lower than they were before pandemic
struck. In aggregate, business exits were little affected by the pandemic. Despite a notable increase
in 2020:Q2, the total number of exits in the first year of the pandemic (2020:Q2 – 2021:Q1) were
only 4.4% higher than the total number of exits in the year leading up to the pandemic.14
Indeed, industry economists expected economy-wide bankruptcies throughout the recession. According to David Mericle, head of US economics at Goldman Sachs, “This really seemed like it
had the potential to be a huge collapse. For most people, and I would include myself, [the lack of
bankruptcies] has come as a pleasant surprise.” For Mericle, the Paycheck Protection Program was
the top item explaining the lower-than-expected number of bankruptcy filings (Coy (2020)).
First, we consider the consequences of large-scale small business closure. Although we cannot
directly observe the cash flows of small firms, we draw on findings in Farrell and Wheat (2016) that
fewer than 75% of small firms hold cash buffers to cover more than 62 days of expenses. The PPP
gave these financially vulnerable small businesses vital cash flow to replace the drastic reduction
in revenues caused by the pandemic and economic shutdown, thereby facilitating their survival.
The fragility of small businesses found in Farrell and Wheat (2016) is corroborated in Bartik et al.
(2020a), which finds in an independent survey that fewer than 30% of firms had cash on hand to
cover more than two months’ expenses. Additionally, the Census Bureau’s Small Business Pulse
Survey found in the week ending May 2 that only 16.7% of small businesses had enough cash on
hand to cover three or more months of business operations.
If, following Farrell and Wheat (2016), 75% of small businesses covered by PPP would have shut
down, at least temporarily, then their workers would in the Program’s absence have been laid
off. This assumption is strong, but the evidence cited on small business fragility provides some
justification. While it is possible small businesses could secure liquidity from private sources to
help them manage shocks, evidence from the Joint Small Business Credit Survey in Federal Reserve
Banks of New York, Atlanta, Cleveland, and Philadelphia (2014) suggests otherwise. According to
these Fed Banks’ report, only 32% of firms with 1-9 employees (which correspond to 75% of PPP
recipients) received any credit in 2014, and a majority of firms with less than $1 million in revenue
did not secure any credit whatsoever. Moreover, 40% of firms seeking credit said the primary
purpose was for expansion, suggesting that fewer than 13% of firms with 1-9 employees had a line
of credit which could be used to buffet revenue shocks.
The Fed report further finds that the primary means of financing of firms with less than $250,000
in revenues is personal savings; that the average time it takes a small business to fill out a credit
application is 24 hours; and that typical wait times for approval are on the order of months. Credit
became harder to get after the onset of the pandemic: according to the Federal Reserve’s Senior
in February.
14
BLS’ Business Employment Dynamics, Private Sector Establishment Deaths.

12


Loan Officer Survey, the net percentage of banks tightening lending conditions for commercial and
industrial loans to small firms reached 70% in the third quarter of 2020, only a few percentage
points away from the previous peak in the series in the fourth quarter of 2008.
In recognizing that some firms will have access to alternate lines of credit, and that revenues did
not fall to zero for many businesses, we make a more conservative assumption regarding which
firms close (at least temporarily). We assume that only the smallest firms who received PPP
stop operations. Further, insofar as this particular assumption misses the mark, it leads us to
understate the true number of jobs preserved. Therefore, this back-of-the-envelope calculation
assumes that the smallest 75% of PPP recipients would have had to lay off their workers without
PPP. Implementation of this assumption suggests that 13.4 million workers had their jobs preserved
due to PPP - almost precisely the upper bound of our estimate range.
An alternative back-of-the-envelope calculation can be derived from surveys in Bartik et al. (2020b).
These surveys indicated that small firms in early April 2020 expected to have employment levels
relative to January fall 40% by year-end. However, when told about the forgiveness provisions in
the CARES Act loans, the survey results reduced that forecast to a 6% reduction (firms told about
loans but not forgiveness expected reductions of 14%). Extrapolated over the 59 million small
business employees supported by PPP, this implicit reduction in unemployment is equal to 20.1
million workers.
One final point of reference is results in Barlett and Morse (2021), who find in a survey of 278
small businesses in Oakland, CA, that PPP receipt reduced the subjective risk of medium-term
small business closure by an average 20.5 percentage points; if 20.5% of firms with fewer than 500
employees were forced to shut down due to the recession, that would destroy approximately 12.4
million jobs. Because the sample is small and localized, the results from this survey may generalize
less readily than those of our other calculations above.
While these back-of-the-envelope calculations are useful for providing context, they are not careful
empirical work. We now turn to an empirical strategy to identify and estimate the employment
effects of PPP.

V. Data
Courtesy of the Small Business Administration (SBA), we observe the universe of approved SBA
loans through August 8th, when the Program closed to new applicants. These data include loan
recipient, address, exact loan amount and date, a self-reported number of jobs covered made at
application, plus a follow-up report of jobs covered.15 In contrast to the publicly available data,
15

We exclude loans that were subsequently canceled or never disbursed. Further modifications to the raw data are
detailed in appendix A

13


our jobs covered data is not winsorized at 500 employees. Further, we observe the lender for each
loan, which we match up to FDIC data in order to identify community banks.
County-level Unemployment Insurance (UI) claims weekly data are furnished by the Bureau of
Labor Statistics (BLS). Initial claims data reflect the number of initial claims approved by the state.
Importantly, the approved claims measure is distinct from the number of claims filed, which report
claims whether or not they are ultimately accepted.16 While this may seem like a relatively minor
distinction, it is critical in the context of March and April 2020. Not only does this distinction
explain a key timing issue in our results, but it also serves as an imperfect filter for problems
stemming from fraudulent claims. Finally, note that each observed claim is recorded for the week
that was claimed, which is distinct from the week filed and the week approved.
Critical to our empirical approach – and unique in this literature - these BLS data allows us to
observe the size of the employer linked to each claim, split into three size buckets ({1 – 99, 100 –
499, 500+}). To be clear, these data feature three observations for every county-week pair. This
allows us to precisely measure the effects of loans to small firms on the UI claims that originated
from employees of small firms. Absent this disambiguation, we would be inferring the effect of
small-firm loans on UI claims originating from any employer. At best, that would attenuate our
estimates. At worst, it would introduce significant bias.
Continuing UI claims data are also furnished by the BLS, and similarly reflect approved continuing claims, not filed continuing claims. Unlike the data on initial UI claims, we observe these
data monthly, not weekly. We impute the data for the interim weeks using both approved initial
claims and the attrition rate implied by the monthly difference in continuing claims. We observe
continuing claims with employer-size buckets as well. These data reliably cover 45 states and DC.
The remaining 5 states either do not report county-level statistics to the BLS, or their county-level
data suffer from inconsistent reporting.17
We construct data for community bank penetration from the Federal Deposit Insurance Company’s
(FDIC) Summary of Deposits data, and we define community banks following the FDIC’s Institution Directory. Data on county population size and density are from the US Census Bureau.
Additionally, measures of eligible county-level payroll for firms with fewer than 500 employees are
estimated based on Census’ 2018 Statistics of U.S. Business (SUSB). The reported payrolls have
been adjusted for two years of estimated growth. These estimates are further adjusted to account
for the fact that larger firms in the Accommodation and Food Services industry were also permitted
16

The widely-used weekly state-level UI release reports the number of claims filed. Given that our data include
only claims which are accepted, they avoid some of the double-counting issues involved in applications for ordinary
state programs and Pandemic Unemployment Assistance funds, discussed in Cajner et al. (2020) and elsewhere. Our
metric of initial UI claims is arguably cleaner than the state-level releases made available to the public. This measure
will cause a timing issue in the data that will further facilitate precise identification, something we will return to in
section VI.B.iv.
17
The omitted states are California, Florida, Hawaii, Michigan, and Minnesota.

14


loans. With the exception of Nevada, this adjustment was relatively small. We also use SUSB to
measure industry shares by firm size in each county.
County-level data on Covid-19 cases and deaths are from the New York Times, based on reports
from state and local health agencies. Additional data on small business revenues by county-week,
and job losses by industry-week come from Chetty et al. (Forthcoming, 2024) and Opportunity
Insights. Data by industry-month come from the BLS’ Current Employment Statistics. Finally, we
use the metric calculated in Dingel and Neiman (2020) to measure the fraction of jobs that can by
done from home, by NAICS 2-digit industry.

VI. Empirics and Results
VI.A.

Empirical Strategy

VI.A.i.

Overview

Did PPP loans preserve jobs? If so, to what extent? To credibly estimate answers to those two
questions, we first address two crucial complications: endogeneity and how to distinguish treatment
from control. Before turning to our results, we discuss how we address those complications, and
how the approach differs from other studies on PPP loans.
First, endogeneity: the factors influencing early PPP loan approval also drive business outcomes
like employment and firm survival. On one hand, factors like managerial competence, attentiveness,
and a pre-existing banking relationship may increase the likelihood of early loan approval and of
salutary business outcomes. On the other hand, firms in financial distress face worse business
outcomes, while also having a strong incentive to apply for a PPP loan urgently. Regardless of
which effect is stronger, a regression of business outcomes of PPP loan receipt would not recover
the causal effects of the PPP loans themselves. The core problem is that firms are on the demand
side for both PPP loans and for labor. Any firm-level shocks can potentially affect both loan and
labor demand, ruling out a direct OLS regression of employment on PPP loans.
To estimate the causal effect of PPP loans on employment, we propose an instrumental variable
approach. Our strategy leverages local variation that temporarily differentiates the supply of loans
(i.e. bank lending) from firms’ demand for loans. Specifically, we use county-level Community
Bank market shares as an instrument for the fraction of county-level small business payroll covered
by a PPP loan on April 11, 2020 (eight days after the program’s debut).18 Community Banks
submitted PPP loans to the SBA more quickly than bigger banks did, for reasons we discuss in the
following section.
18

Market share is measured by pre-pandemic value of deposits, by branch location.

15


To address endogeneity concerns, our instrument isolates a supply shifter from demand-side determinants. In contrast to firms, banks typically do not influence the supply and demand of labor,
save through firm financing.19
The second complication lies in distinguishing treatment from control. The near universal takeup of PPP loans presents an empirical challenge: if (nearly) every firm is ‘treated’ with a loan,
then comparisons to observed counterfactual outcomes are impossible. However, many firms were
unable to secure a loan at the program’s outset. Delays in loan approval—at least those otherwise
orthogonal to business outcomes—open a brief window for the study of PPP’s effects. The brevity
of this opening has important, and precise, implications for employment dynamics, which we will
test. PPP loans had nearly saturated the market by early June, so there is no meaningful control to
be observed at that point, at least in the sense of loan receipt. Once both ‘treatment’ and ‘control’
firms had received loans, earlier access to loans is manifested in the likelihood of early- vs. late-loan
approval, a less informative division for our study.
To emphasize the point, this paper’s identification strategy has two fulcra, first and foremost being
geography. Specifically, geography yields unequal access to early-moving lenders. The likelihood of
early loan approval, in turn, is a function of this access. However, geography alone is insufficient,
since ultimate loan receipt is nearly universal. We therefore lean on our second fulcrum, differences
in timing, to separate treatment from control.

VI.A.ii.

Community Bank Share as an Instrument

What distinguished Community banks from other lenders during PPP’s initial roll-out? Community
banks were markedly quicker than larger banks in submitting PPP loans to the SBA. Larger banks,
with the institutional memory of the post-TARP fallout, were reticent to issue loans absent clear
guidance from SBA and Treasury.20 Several large banks told Michael Faulkender directly, in his
capacity as the Treasury principal responsible for implementation of PPP, that they were hesitant
to extend PPP loans until SBA and Treasury issued further regulatory clarification via Interim
Final Rules (IFRs) and frequently asked questions.21 These banks had suffered years of legal
and regulatory disputes following TARP, when they had acted quickly and without comprehensive
guidance. The consequence of these disputes was an aversion to regulatory risk – larger banks
19

Theoretically, banks can influence labor supply through personal loans. However, personal loans did not increase
during the early pandemic – instead, household balance sheets became much healthier due to government stimulus.
Also, banks themselves were not eligible for PPP loans.
20
TARP stands for the Troubled Assets Relief Program, enacted in October 2008 to help stabilize the U.S. financial
system.
21
Michael Faulkender, an author of this paper, was the Treasury principal responsible for policy implementation of
PPP. The first IFR for PPP was issued on April 2, but was frequently updated in response to concerns expressed by
a variety of stakeholders, including banks. The IFR was updated on April 3, then a further five times in April and
four times in May. Additionally, Treasury and SBA issued 39 FAQs in April, including 17 on April 6 alone.

16


wanted explicit official guidance to serve as legal protection from regulators.22 Empirically, we
observe significant delays from the largest banks, illustrated in Figure IIb. Over the program’s
first five days, the largest five banks issued a total of 568 loans, worth $172 million. In that same
timeframe, community banks issued 218,984 loans, worth $46.3 billion.
In contrast to the largest banks, community and other small banks did not believe they would be
an attractive political target, and did not have the same memory of recent regulatory trauma: no
Administration of either party would be eager to sue them. Smaller banks therefore demonstrated a
willingness to act quickly to fill this market opening, taking regulatory risk before all the program’s
details were entered into the Federal Register. This difference in perceived regulatory risk exposure
was the key contributor to speed of PPP loan disbursement. Crucially, geographic variation in
small businesses’ banks’ perceived regulatory risk exposure, conditioned on controls, should be
orthogonal to the spread of the virus or firms’ demand for PPP loans.
The primary source testimonial is supplemented by contemporaneous press articles and statements
from Congressional leaders. On May 6, 2020, Merker (2020) wrote in American Banker, “Given the
uncertainties surrounding the PPP, banks need both better guidance and more reassurance from
the financial regulators. Otherwise, the ghosts of 2008 could continue to haunt bankers as they try
to carry out government policy at a perilous moment in the nation’s economic history.” However,
these concerns were not uniformly felt by all banks. Also in American Banker, Haggerty (2020)
quotes Ian Katz, a director at Capital Alpha Partners, “If I were the big banks on the second goaround, I would want more clarity from the administration on what’s permitted and what’s not.”
Notably, these regulatory risk warnings concentrated on big banks – the perception was that other
banks were less exposed to this threat.
Statements from Congressional leaders of both parties raised the specter of regulatory risk. They
expressed particular concern that big banks reportedly gave preference to pre-existing customers, as
opposed to issuing loans on a first-come, first-served basis. Republican Senator Marco Rubio – then
Chair of the Senate Small Business Committee – wrote to the CEOs of twelve large banks, “I, as well
as other members of the Senate, have received reports of priority being given to certain applicants
over others. While I recognize the challenges of setting up a program of this size, processes to
handle applications, and appropriate guidance to administer the program, it is important for small
businesses and nonprofits of various sizes, regional locations, and missions to have equal access
to PPP assistance.”23 Senator Rubio’s letter went on to request responses to a series of questions
about prioritizing some borrowers over others. Democrats Maxine Waters (then Chair of the
House Financial Services Committee) and Nydia Velazquez (then Chair of the House Small Business
22

For example, U.S. Bank did not have a single PPP loan approved during the programs’ first five days.
Rubio (2020), April 23, 2020. The letters were sent to the Chief Executive Officers of Bank of America Corporation, JPMorgan Chase, Wells Fargo, PNC Financial Services Group, Inc., KeyBank, NA, M&T Bank, Huntington
Bancshares, Inc., TD Bank, Truist Bank, Zions Bank, Regions Bank, and US Bancorp.
23

17


Committee), wrote the CEOs of four large banks, “...we are troubled by concerns expressed by small
business owners that megabanks are favoring certain customers and shutting out others. We expect
each of you to make a commitment that your institutions will do all you can to help consumers
and small businesses... As your participation in the [Paycheck Protection Program] moves forward,
we would appreciate periodic updates on... PPP implementation and other pandemic recovery
efforts.”24
These differences in perceived regulatory risk generated substantial differences across markets in
early PPP penetration, based in part on significant variation in community bank market share. Liu
and Volker (2020) documented that community banks were among the fastest to issue PPP loans.
We also document this dynamic: Figure IIa illustrates the strength of community banks in issuing
early loans. Community banks facilitated more dollars of PPP loans during the program’s first six
days than all other lenders combined. When funding for the first tranche of loans was exhausted
on April 16, 46.8% of all loans approved had originated at community banks. At that same point,
community banks had originated 47.9% of their final total PPP loans. For all other lenders, that
number was 24.0%. Put simply, it was much easier to secure an early PPP loan with a community
bank. Given the clear evidence that community banks were quicker to disburse PPP loans, we use
geographic variation in community bank market share to instrument for early loan receipt (for a
map, see Figure III). We also illustrate the strong relationship between county-level community
bank shares and early PPP loans with a first-stage binscatter (Figure VIb).
Before describing our empirical model, we pause to contrast our approach with Granja et al. (2022),
which also leverages local banking conditions to assess the employment effects of PPP loans. We
highlight a number of differences and note their importance. First, this paper’s instrument is
a supply-shifter, in the spirit of Wright (1928). In contrast, Granja et al. (2022) use a Bartik
approach, measuring the shift in shares from the pre-pandemic small business loan market to the
PPP loan market.
In our view, there are several drawbacks to the Bartik approach in this context. First, it treats a
standard small business loan and a PPP loan as the same (or very similar) products. We argue
that these are very different financial products. Most obviously, while structured as a forgivable
loan, PPP loans were, economically, effectively a grant, and the Federal government was very clear
on that matter from the outset. PPP “loans” were loans in name only, and both banks and firms
knew that. Second, a standard business loan is typically meant to finance capital expenditures,
not operating costs. PPP loans were intended to support operating costs, and the condition of
forgiveness was that loaned money directly financed operating costs, particularly labor.
Moreover, since the goal of the instrument is to isolate exogenous variation in PPP loan coverage,
24

Velazquez and Waters (2020), April 10, 2020. The letters were sent to the Chief Executive Officers of Citigroup,
JPMorgan Chase, Bank of America, and Wells Fargo.

18


the difference between pre-pandemic small business loan market share and PPP loan market share is
not necessarily a relevant metric. For example, suppose a bank exclusively covers some geographic
area, both before and during the pandemic. That bank may end up with the same market share of
pre-pandemic loans and PPP loans, leading to a measured shift share of 0, though that bank might
have been very effective in delivering PPP loans during the program’s first week. The central issue
is that the total share of PPP loans is the relevant metric for studying the employment effects of
PPP loans – not the change in share, relative to pre-pandemic small business loans.
A second major distinction between this paper and Granja et al. (2022) is the effective baseline. In
this paper, the baseline is February 2020, prior to the pandemic’s American economic effects. In
Granja et al. (2022), the baseline is the final weeks of March. Those weeks followed the onset of
the pandemic, but preceded PPP loans.
Those weeks were among the most tumultuous in U.S. economic history, with 8.9 million initial
claims for unemployment insurance filed in the weeks ending March 21 and March 28, 2020 combined. The following week (ending April 4) saw an additional 6.1 million claims alone. Using a
late-March baseline means that any measure is exquisitely sensitive to the exact timing of the Covid
shock across region. For example, if the worst of the economic shock hits one area in the week
ending March 28, then metrics using the late-March baseline will show recovery. Contrast that with
a region that is hit just one week later: metrics with the late-March baseline will immediately show
deterioration, regardless of which region sustains worse job loss as a percentage of pre-pandemic
employment.

19


(a) Community Bank Share of Cumulative PPP Loans, by Approval Date

(b) Top-Five Largest Bank Share of Cumulative PPP Loans, by Approval Date

Figure II
Bank Types and PPP Loans Issuance Over Time. Source: SBA, Treasury, and author’s calculations. All
loans which were eventually canceled are excluded from calculations.

20


Figure III
Community Bank Share of County-Level Banking Market, by Deposits
Figure IV
Community Bank Share of County-Level Banking Market, by Deposits. Source: FDIC and author’s
calculations.

VI.A.iii.

Empirical Model

We instrument early PPP loan receipt (measured at the county x firm-size level) with community
bank shares, thereby exploiting variation in local banking markets to predict early PPP receipt.
For the first stage, we model the endogenous variable at time t′ as

′
P P Pcjt′ = α0,s(c)jt + α1,jt CB Sharec + Xcjt
α2,jt + ηcjt

(1)

The exclusion restriction is that CB Sharec does not enter 2, save through 1. The second stage,
21


which we refer to as our primary specification, is repeated cross-sections of

′
ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt

(2)

where ycjt is an unemployment outcome in county c from firms of size j, during week t, s(c) is
the state of county c, and Xcjt is a vector of controls. Summarizing the covariates in our ‘primary
specification’: we include β0,s(c)jt - a state-week fixed effect, ycj,F eb. - the February 2020 average of
the outcome variable, county-level median income and poverty rate, a battery of Covid-19 controls
covering the rates of new cases and deaths over the prior week and month, the log of county
population density, and finally a county-level work from home (WFH) index.
Our key variable of interest, P P Pcjt′ , is the cumulative percentage of small business payroll covered
(at firms of size j in county c) by a PPP loan, as of t′ = April 11, 2020. In all cases, the percentage of
UI claims is measured as (a) the number of approved UI claims originating from employees in county
c during week t (employed by firms of size j), divided by (b) the number of pre-pandemic employees
eligible for UI in county c, who were employed by firms of size j. We estimate this regression in
repeated cross-sections, separately for each week, in order to illustrate the dynamic effects of early
loan receipt. Note that t′ is held constant in each regression while t varies. The regression studies
the dynamic relationship of employment with early PPP receipt; the key independent variable
is fixed in time while the dependent variable evolves. The treatment is not the receipt of PPP
money, but the early receipt of PPP money. The core dynamic hypothesis, which we test, is the
rapid convergence of outcomes across counties, as loans saturate the market. As (t − t′ ) gets large,
βP P P,jt should converge to 0.
Why April 11 in particular? In terms of cumulative small business payroll covered by PPP loans,
we observe the greatest county-level variation early in the program’s first round. By a variety of
measures, the highest degree of dispersion in PPP coverage occurs during the week ending April
11, shortly after the program begins and when the first tranche of funds was nearly exhausted. On
April 16, the first tranche of $349 billion dollars had been fully depleted, and the program closed for
11 days. The resulting delay until the SBA resumed accepting loans bolsters our timing strategy by
opening a substantial temporal gap between first- and second-tranche loan recipients. In Figure V,
we show dispersion in PPP loan coverage across counties. Though dispersion is high on April 11,
it had mostly disappeared by August 8, more-or-less eliminating the variation needed to evaluate
the Program’s effects.
We feature two primary outcome variables. The first is the insured unemployment rate (IUR) for
regular state programs (i.e. approved continuing claims as a share of covered employment25 ) at
25

“Covered employment” refers to the pre-pandemic number of UI-eligible workers at the county-firm size level,

22


the county-firm size level. Using the weekly IUR avoids the double counting problems introduced
by the Pandemic Unemployment Assistance program and captures high-frequency reentry into the
workplace. The second outcome variable is approved initial claims at the county-firm-size level,
also expressed as a share of covered employment. As robustness checks, we will also consider several
other variables as outcomes.

(a) Early Variation in Loan Coverage

(b) Late Variation in Loan Coverage

Figure V
Early in the program, there was significant geographical heterogeneity in PPP receipt. Later in the
program, there was very little, limiting the ability of ultimate PPP receipt to identify the Program’s
effects. Source: SBA, Treasury, and author’s calculations. Coverage calculated as loan dollars, as fraction
of total estimated eligible payrolls. Eligible payroll estimated by projecting payrolls from the 2018 Census
SUSB forward for two years growth.

VI.A.iv.

The Exclusion Restriction

Though it may be exceedingly unlikely that community bank shares are causally related to business
outcomes – financing channel aside – this alone does not satisfy the exclusion restriction. If community bank market shares are correlated with variables that determine business outcomes, that
correlation would likewise violate the exclusion restriction. Throughout this section, it is important
to keep in mind that our hypothesis is about the timing of business outcomes. This means that any
potential confounder would not merely have to correlate with community bank market shares, but
also correlate with employment outcomes in a rapidly changing manner as the weeks progressed.
In other words, confounding variable(s) would have to introduce bias in April and May, but not in
July through October. Any alternative explanation of the estimated effects would have to account
for why these community bank shares are relevant in some weeks but not others.
The spread of the virus itself is one obvious threat to our identification strategy. As Granja et al.
(2022) notes, first-round PPP loans tended to flow to areas that were not as hard hit by the
not PPP coverage.

23


pandemic. Since community banks tend to have stronger market shares in rural areas, it’s possible
that our instrument is negatively correlated with early virus prevalence—which is in turn negatively
correlated with unemployment. We address this concern via two types of control variables. First,
we include new Covid-19 cases reported in county c in week t, as well as the preceding 4 weeks.
Additional controls for Covid-19 deaths are included for the same timeframe, all measured as a
fraction of the county’s population. Second, to account for the differences between urban and rural
areas, we control for the log of county population density. Finally, we include controls for the
insured unemployment rate for the week of February 15th, 2020, to avoid mistaking pre-existing
level differences for early effects.
Directly related to the spread of Covid is the government’s reaction. States controlled many of
the decisions regarding lockdowns and economic restrictions during March and April, and these
states took a variety of approaches to combating the virus, both in terms of severity and timing.
To account for this, we include state-week fixed effects in all our specifications. This is crucial: we
exploit within-state and within-week variation to identify the parameter of interest, βP P P,jt . Our
results don’t capture the difference between New York and Nebraska, but differences within each
of those states, in a given week. In many cases, state laws and other measures had a direct impact
on employment, for example when restaurants were prohibited from offering dine-in service, or
when non-essential businesses were ordered to close. Insofar as individual attitudes towards Covid
varied across states, these fixed effects would also control for that. States also determine the major
sub-national regulations that affect the local banking markets, as well as the relative capacity of
localities to cope with economic shocks. Cognizant of these heterogeneous economic restrictions,
we focus on within-state-week variation at the county-level.
To address possible differences across counties that pre-date the pandemic, we control for the
2019 county-level median income and poverty rates. We might be concerned that the change in
economic activity varies with income. For example, a lower-income area might see a smaller decline
in economic activity if pre-pandemic spending was more concentrated on necessities like groceries.
Additionally, we control for the February 2020 average of the relevant IUR (measured at the country
x firm-size level to match the outcome variable). This control effectively establishes a pre-pandemic
baseline for UI claims.
We also need to be aware of any differential actions taken by community banks themselves - aside
from the provision of PPP loans. These actions would need to have taken place at the outset of
the pandemic, and not have been contemporaneously replicated by other banks. For example, if
community banks extended small businesses credit in greater volume than large banks prior to
the debut of PPP, this may have supported small business employment more strongly in counties
with high community bank shares. However, there is no evidence that community banks were
extending small businesses credit at relatively high volumes prior to PPP. As documented in Lopez

24


and Spiegel (2023), non-PPP small business lending declined at small and medium sized banks in
the first half of 2020, but not at large banks. Such an outcome is the opposite of what would be
required for our estimates to be confounded by differential lending activity unrelated to PPP.
Finally, we construct a metric to measure how easily a county’s employment base could shift to a
work from home environment. The concern here is that some areas may be better suited to handle
the pandemic shock, simply because they had more jobs that could be done remotely. We use data
from Dingel and Neiman (2020) to measure the share of jobs that can be done at home, by 2-digit
NAICS code. We then take the inner product of those shares with the share of employment at
firms of 1-99 workers in each industry for each county, based on the 2018 Census SUSB. We call
this product the “Work From Home (WFH) Index”.
To reiterate, we estimate our primary specification in a repeated cross-section. Therefore, each of
these controls can dynamically covary with the outcomes of interest. We consider this crucial in
such a rapidly changing economic environment. Even if a covariate itself remains constant from
week-to-week, its correlation with outcomes can and does change over time.
Near the end of subsection VI.B, we consider alternate outcomes and covariates as robustness checks.
In particular, we include the decline in county-level small business revenue from pre-pandemic to
an average of mid to late-March – i.e. following the outset of the pandemic, but before PPP. While
this considerably reduces our sample size due to the limited data coverage, the point estimates in
our main results are little changed with this inclusion.
Finally, note that our study cannot fully address general equilibrium concerns: if there are spillovers
because layoffs in one location or firm can cause layoffs in another location, then no county-level or
firm-level analysis will truly isolate treatment effects. If the “control group” is affected by outcomes
at a treatment group, any empirical strategy will be problematic. However, the other PPP studies
cited face the same challenge.

VI.B.

Results

We present our main results sequentially. In our primary results, firm-size j represents all firms
with fewer than 100 pre-pandemic employees. In the secondary results, we expand our analysis to
all small businesses with fewer than 500 pre-pandemic employees.

VI.B.i.

First Stage: Firms Sized 0-99

Figure VIb shows (a) a county-level binned scatterplot of community bank shares against early
PPP coverage, and (b) a binned scatterplot of residuals from the first-stage regression.26 With
26

Using data from the week ending April 11, 2020, with Initial Claims as the control for ‘average February IUR’.

25


the full battery of controls and standard errors clustered at the state level, the initial claims
regression’s standard first stage F-Stat is 73.7 (Kleibergen-Paap: 9.2) during the week of April 11,
for example. We report first-stage statistics and second-stage tests following the recommendations
given in Andrews, Stock, and Sun (2019).
Since we do not assume our errors are i.i.d. – we cluster at the state-level – we also report tests
developed in Kleibergen and Paap (2006). In Appendix F and in all regression tables, we report
the robust Kleibergen-Paap F-Statistics alongside the Anderson-Rubin confidence intervals and
p-values. Since our specification is just-identified with one endogenous regressor, the KleibergenPaap F-Stat is identical to the effective F-Statistics proposed by Olea and Pflueger (2013), which is
itself robust to heteroskedasticity and clustering. The Anderson-Rubin confidence intervals are also
robust to heteroskedasticity and clustering. More importantly, they assure the correct coverage in
the case of weak instruments (Moreira (2009)).
While the traditionally reported F-Statistics for our primary estimates are very strong, KleibergenPaap F-Statistics typically fall just below the ‘rule-of-thumb’ cutoff statistic of 10. While it is
generally good practice to report Anderson-Rubin confidence intervals with any application of IV,
they carry particular importance in our case, in order to address any concerns of weak instrument
bias. However, our figures present traditional confidence intervals based on clustered standard
errors. We do this to keep the figures legible – Anderson-Rubin confidence intervals are asymmetric
and can often have one long tail. Further, there is no assurance that they are finite, nor are they
always continuous. We note that our primary results for firms sized 0-99 hold up with AndersonRubin confidence intervals, and in some cases the p-values shrink under Anderson-Rubin (compared
to the standard Wald test). This is made possible by the asymmetry of Anderson-Rubin: a longer
confidence interval can be consistent with a smaller upper bound.

26


(a) Binned Scatterplot: Early PPP loan coverage for firms sized 0-99 and community bank
deposit shares. Bins calculated with small firm employment weights.

(b) First Stage Binscatter: residuals from regressions of (i) the instrument & (ii) the endogenous
variable, on the full set of controls. Regressions weighted by small business employment.

Figure VI
First stage correlation, with and without controls. Full set of first-stage F-Stats presented in appendix
Appendix F. “Full set of controls” as described in Table III, column (4) and the associated footnote.

27


VI.B.ii.

Second Stage: Firms Sized 0-99

Figure VII as well as Tables A.5 and A.6 present the results for the second stage coefficient
(βP P P,0−99,t from Equation 2). Additionally, we report traditional regression tables with alternate specifications for all weeks in the online appendix.27 As an example, we include results for the
week ending March 21, 2020 for approved initial claims (Table II) and the week ending April 11,
2020 for approved continuing claims (Table III). Results for our primary specification are in column
(4), with columns (5) and (6) showing results from robustness checks (discussed in Section VI.B.v
and Appendix Appendix B). To reiterate, we estimate these results separately for each week. For
both initial and continuing claims, we estimate precise zeros through mid-March, and once again
as take-up nears saturation. This is a baseline sanity check: we should not observe an effect for
April’s PPP loans in February. Further, we should not observe large effects for early loans receipt
after the vast majority of firms secured PPP funding, unless delays proved fatal to a meaningful
number of firms.28
The estimates follow different dynamics for initial and continuing claims, reflecting their stock/flow
distinction. The estimates βP P P,0−99,t are precise zeros until the week ending March 21, when it
plummets to its minimum estimate of -0.099. In plain English: increasing early PPP coverage of
small-firm payrolls by 1 percentage point resulted in a 0.099 percentage points smaller increase in
the approved initial UI claims rate (among small-firm employees), during the week ending March
21, 2020, alone. The initial claims estimates ease slightly over the weeks ending March 28 and April
4, then continue to ease before returning to statistical insignificance – permanently - in the week
ending May 23. This pattern matches our hypothesis that stronger early penetration of PPP loans
led to substantially reduced approved initial UI claims.
Naturally, estimates of the effects on continuing claims lag the effects on initial claims. Our point
estimate is first statistically significant during the week ending March 28.29 The estimated effects
increase quickly over the next two weeks, reaching a minimum of -0.252 during the week ending
April 11. In plain English: increasing early PPP coverage of small-firm payrolls by 1 percentage
point resulted in a 0.252 percentage points smaller increase in the approved continuing UI claims
rate. The estimated effects level off from there, then gradually decline until they are no longer
statistically significant during the week ending May 30.
We note that our initial claims estimates reach their minimum values before the CARES Act is even
27

In the online appendix, we also report regressions where the cross-sections have been pooled into five groups of
weeks: Pre-Covid, Covid-Onset, First PPP Tranche, Second PPP Tranche, and Full PPP Rollout
28
Our estimate are precise zeros, starting in July, suggesting that the delays correlated with our instrument did
not lead to lasting damage. This stands in contrast with papers such as Doniger and Kay (2023), Kurmann et al
(2024), and Cole (2024), which find statistically significant effects persisting into late 2020.
29
This is true for both traditional clustered standard errors and the weak-IV robust Anderson-Rubin standard
errors.

28


passed. We discuss this issue in detail in subsection VI.B.iv, showing why this occurs and how it is
consistent with our causal claims. In short, the observation of an approved claim is conditional on
the approval itself, which necessarily follows a process where claims are filed, challenged, withdrawn,
and finally approved or rejected. The lag between filing and approval was particularly important
during the explosion of UI claims during the early pandemic, when state UI offices were subject to
an overwhelming volume of initial claims.
In all of the papers in the PPP literature, translating point estimates into an aggregate number
of jobs saved (more precisely in our case, number of approved initial UI claims averted) requires
further assumptions and nuance. There are good reasons to be cautious of these extrapolation
exercises. However, given their prevalence in the literature and in comparable papers, we offer the
aggregate numbers based on our own estimates.
Multiplying the pre-pandemic observed total national covered employment (144.50 million, from
the Quarterly Census of Employment and Wages (QCEW)) by the fraction of national employment
in firms sized 0-99 (34.3%, based on the 2018 Census SUSB), we get covered employment at firms
sized 0-99 of 49.53 million. Our calculation of the final percentage of small firm payroll covered by
PPP loans is 88.1%. Multiplying these by our minimum weekly estimate (-0.252), we arrive at 10.9
million jobs saved at firms sized 0-99 (alternatively, approved continuing UI claims averted in each
week).
One potential objection here is that we have inflated this aggregate number by selecting our minimum estimate. In response, we note that our estimates compare the effect of earlier treatment to
the effect of later treatment – a weaker comparison than treatment and control. Consequently, we
can never truly recover the full effects of PPP loans – each estimate is attenuated by our imperfect
and vanishing control, and our aggregation exercise is attenuated along with it. We therefore likely
underestimate the true treatment effect, however we believe the underestimation is significantly
less than in other studies.

VI.B.iii.

Results for Firms Sized 0-499

We estimate PPP’s employment effects for all firms sized 0-499 using the same specification, with
variables adjusted to reflect the larger set of firms. The specific adjustments are to county-week
unemployment claims (which now reflect claims from employees of all firms 0-499), the county-level
February average of those claims, and PPP loan coverage as of April 11. Regression weights are
also also adjusted to reflect pre-pandemic county-level covered employment at all firms of fewer
than 500 workers.
At first glance (see Figure VIII), the results for firms sized 0-499 look very similar to the those at
firms sized 0-99. This is a consequence of composition, at least in part: firms sized 0-99 employ

29


(a) Initial UI Claims, Firms Sized 0-99

(b) Continuing UI Claims, Firms Sized 0-99

Figure VII
IV estimates of the effect of PPP penetration as of April 11th on employment outcomes at firms sized 0-99. Full
results with multiple specifications and an extended footnote are available in the online appendix. Example tables
for 2020-03-21 (for Initial Claims) and 2020-04-11 (for Continuing Claims) are available in Tables A.5 and A.6,
respectively.
30


TABLE II
Initial Claims, Firms Size 0-99 (Week Ending 2020-03-21)
(1)

(2)

(3)

(4)

(5)

(6)

-0.005
(0.016)

-0.079∗∗∗
(0.019)

-0.105∗
(0.044)

-0.103∗∗
(0.039)

-0.044∗∗
(0.017)

-0.086∗∗
(0.029)

February IUR

3.366∗
(1.627)

3.431∗
(1.559)

4.026∗∗∗
(0.940)

3.671∗
(1.572)

Log(Med. Income)

-0.019∗
(0.009)

-0.019
(0.010)

-0.016∗∗
(0.006)

-0.019∗
(0.008)

Poverty Rate

-0.001∗
(0.000)

-0.001∗
(0.000)

-0.001∗∗
(0.000)

-0.001∗∗
(0.000)

Log(Pop. Density)

0.000
(0.001)

0.000
(0.002)

0.001
(0.001)

-0.001
(0.001)

Covid Cases, 1w

-47.049
(32.431)

-46.439
(24.271)

-37.961
(27.573)

Covid Cases, 4w

39.550
(29.004)

36.156
(21.136)

29.540
(23.960)

Covid Deaths, 1w

571.098
(633.069)

499.412
(524.802)

606.625
(623.377)

Covid Deaths, 4w

-197.636
(404.064)

-232.353
(325.509)

-221.566
(390.597)

WFH Index

-0.005
(0.046)

0.083∗∗∗
(0.023)

0.005
(0.036)

Early PPP Coverage

0.305∗∗∗
(0.040)

Industry Index

-0.026∗∗∗
(0.006)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,588
0.1
86.5
[ -0.044, 0.027]
0.749

Yes
2,588
0.4
51.2
[ -0.152, -0.049]
0.000

Yes
2,588
47.0
9.8
[ -0.336, -0.039]
0.003

Yes
2,586
13.6
12.2
[ -0.289, -0.045]
0.001

Yes
2,586
399.2
18.3
[ -0.102, -0.013]
0.011

Yes
1,584
90.3
15.6
[ -0.210, -0.040]
0.001

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-03-21.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

31


TABLE III
Continuing Claims, Firms Size 0-99 (Week Ending 2020-04-11)
(1)

(2)

(3)

(4)

(5)

(6)

-0.111∗∗
(0.034)

-0.164∗∗∗
(0.035)

-0.274∗
(0.112)

-0.252∗
(0.107)

-0.135∗∗
(0.051)

-0.185∗
(0.078)

February IUR

1.580∗∗∗
(0.256)

1.507∗∗∗
(0.285)

1.391∗∗∗
(0.191)

1.458∗∗∗
(0.238)

Log(Med. Income)

-0.053∗
(0.023)

-0.043
(0.029)

-0.044∗
(0.018)

-0.039
(0.025)

Poverty Rate

-0.003∗
(0.001)

-0.002
(0.001)

-0.002∗∗
(0.001)

-0.002
(0.001)

Log(Pop. Density)

0.000
(0.003)

0.001
(0.005)

0.002
(0.002)

-0.001
(0.004)

Covid Cases, 1w

-12.599
(7.567)

-8.205
(5.727)

-10.194
(7.303)

Covid Cases, 4w

2.983
(3.260)

1.937
(2.458)

2.253
(2.980)

Covid Deaths, 1w

-228.778∗∗∗
(54.139)

-172.036∗∗
(55.083)

-240.967∗∗∗
(50.009)

Covid Deaths, 4w

158.201∗∗∗
(35.273)

111.486∗∗∗
(33.758)

162.884∗∗∗
(30.422)

0.027
(0.132)

0.191∗
(0.091)

0.040
(0.108)

Early PPP Coverage

WFH Index

0.594∗∗∗
(0.101)

Industry Index

-0.050∗∗∗
(0.014)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,352
10.1
72.9
[ -0.181, -0.029]
0.017

Yes
2,352
0.5
55.9
[ -0.257, -0.081]
0.007

Yes
2,339
31.1
8.2
[ -0.927, -0.083]
0.013

Yes
2,337
80.6
8.9
[ -0.840, -0.068]
0.015

Yes
2,337
142.3
11.6
[ -0.309, -0.012]
0.040

Yes
1,486
217.8
10.5
[ -0.499, -0.020]
0.036

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-04-11.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

32


70% of the workers at all firms sized 0-499. Despite the apparent similarity in estimates, we
want to differentiate the effects at small firms (0-99 employees) from the effects at mid-sized firms
(100-499 employees). By distinguishing PPP’s employment effects across the firm-size distribution,
we can start to reconcile the difference between this paper’s estimates and those in Chetty et al.
(Forthcoming, 2024) and Autor et al. (2022a).
Our primary approach to differentiate the effect size is to leverage the just-introduced aggregation
exercise (Section VI.B.ii) to estimate the number of jobs preserved at mid-sized firms. For the sake
of direct comparison, we repeat the aggregation exercise for all firms sized 0-499 using estimates
from the same week. In both regressions, the minimum estimates (i.e., largest estimated effects)
are observed during the week of 11 April 2020. In that week, β̂P P P,0−499,t is -0.240, little different
than that week’s β̂P P P,0−99,t . However, the relatively small difference in estimates significantly
understates the difference in underlying effects. To illustrate this, consider the same aggregation
exercise with the covered employment and final PPP loan payroll coverage measured for all firms
sized 0-499 (70.1 million and 82.7%, respectively). Multiplying these together yields 14.0 million
jobs preserved at firms sized 0-499. Then, subtracting the estimated 10.9 million jobs saved at
small firms yields 3.3 million jobs saved at firms with between 100 and 499 employees. As a
fraction of pre-pandemic employment, that aggregate number implies a ‘preservation rate’ of 15.8%
- indicating the fraction of all jobs saved at mid-sized firms. This rate is 5.5 percentage points
below the small-firm rate.
This direct method of calculating ‘preservation rates’ likely understates the true difference between
mid- and small-sized firms. The understatement comes from two sources: first, the choice of April
11 as the reference week. The gap in that week is relatively small in that week, as it grows through
the PPP’s initial weeks. If we use the week of April 18 instead, the rate difference would be 6.4
percentage points. From the week ending April 18 to the final week of June, the average difference
in point estimates is over twice that of April 11’s, and never falls back to April 11’s level. If we use
that average difference as a basis for calculations, then the gap between preservation rates grows
to 7.1%, implying that loans were one-third less effective at preserving jobs at mid-sized firms than
at small firms.
Instrument relevance is the second source of understatement. Specific to firms sized 100-499, the
correlation between community bank shares and early PPP loan penetration is relatively weak,
once conditioned on controls. For this reason, we do not separately estimate coefficients for these
mid-sized firms. While the instrument remains relevant for all firms sized 0-499, the correlation is
primarily driven by loans to smaller firms.30 Hence the 2SLS estimates for all firms 0-499 reflect
variation that is disproportionately driven by firms sized 0-99. This is also why we caution that
30

For simple regressions of residualized community bank shares on residualized PPP loan payroll coverage, the
coefficient estimate is 0.10 (p = 0.000) for firms sized 0-99. For firms sized 100-499, it is 0.04 (p = 0.055). For all
firms sized 0-499, the same estimate is 0.08 (p = 0.000). Figure XIX shows the first-stage for all firms 0-499.

33


our identification is cleaner for small firms than for mid-sized firms.

VI.B.iv.

Discussion of Pre-Effects

We estimate that PPP loans have statistically significant and meaningful effects on UI claims –
before the first PPP loan is approved. This observation casts fundamental doubt on our results.
Post hoc ergo propter hoc is a fallacy; pre hoc ergo propter hoc is nonsense.
What could explain these estimated pre-effects? If workers exhibited anticipatory behavior, then
we could satisfy our hypothesis while observing the ‘inverted’ causal timeline. However, we view
that as exceedingly unlikely. It would require workers to accurately forecast the probability that
their firm would receive an early PPP loan, then refrain from applying for UI benefits as a result.
Alternately, firms could forecast this likelihood and refrain from laying off their workers, though
we also view this as unlikely. More realistically, we should be worried that pre-effects are evidence
of bias. If counties with high and low community bank shares differ in ways correlated with earlypandemic UI claims (and not accounted for in our specifications), then our estimated effects would
be spurious.
However, there is a mechanism that brings our core results back into alignment with the principles
of temporal cause and effect. That mechanism lies within the UI claims process, and we show how
post-PPP actions affect pre-PPP approved UI claims data. Further, this mechanism was prevalent
in the UI claims process in March and April 2020.
Crucially, approved UI claims are the dependent variable in all of section VI.B’s estimated regressions. Approved UI claims are distinct from filed UI claims. Filed claims are relatively simple: if
a worker (or employer) files a claim, then we observe a filed claim.31 In contrast, an approved UI
claim is observed only when all the following are satisfied: (a) a claim is filed, (b) the state UI office
determines the claim is valid, (c) the employer does not appeal the claim (or that appeal fails), and
(d) the worker does not rescind the claim prior to final approval. The observation of an approved
UI claim is conditional on subsequent events.
We illustrate the sequence of events involved in an initial UI claim in Figure IX. Note: the date
attached to an approved claim reflects the actual week of unemployment being claimed – not when
the claim was filed, nor when the claim was finally approved. As a result, an observed approved
claim which compensates a worker for the week of March 21, 2020 depends on actions and decisions
that come later – potentially much later. In the example, the firm lays off a worker on Tuesday,
March 17. The worker files an initial UI claim a week later, on March 24. The state then reviews
the claim, eventually approving it on April 8. Since the worker filed the claim on March 24,
we observe a filed UI claim with an observed date of March 28 (UI claims are always dated to
31

Employers may file claims on behalf of their workers in certain cases.

34


(a) Initial UI Claims, Firms Sized 0-499

(b) Continuing UI Claims, Firms Sized 0-499

Figure VIII
IV estimates of the effect of PPP penetration as of April 11th on employment outcomes at firms sized 0-499. Full
results reported in Appendix Appendix F and the online appendix.

35


Saturdays). We also observe an approved UI claim dated March 21 - though this is only observed
because the state approved the claim weeks later. Had the state denied the claim (or if it were
challenged/withdrawn), only the filed claim would have been observed.

Figure IX
Example Timeline of a UI Claim.

We can directly test whether our proposed mechanism appears in the data, since we observe both
filed and approved initial claims for a subset of counties. Keep in mind that filed claims and
approved claims are not directly comparable in a given week: a claim may be ‘approved’ for some
week, but filed two weeks later. Therefore, we calculate ϕc as the ratio of approved:filed initial
claims in county c over a longer time period than just the weeks in question. The metric is the
sum of all approved initial claims over a given period (for workers from firms of all sizes, a data
limitation) divided by the sum of all initial claims over a given period, i.e.

P
T (ApprovedClaims)c
P
T (F iledClaims)c
Where we consider a number of different time periods, T . If our proposed mechanism is in the
data, it would manifest itself as a negative correlation between our instrument and this metric.32 In
other words, the stronger the early PPP exposure, the lower the fraction of filed initial claims that
end up approved. In our primary regression, where T encompasses the weeks ending February 29
32

In practice, we also include state fixed effects in the regression, so as not to compare results from different state
UI offices.

36


through May 2, the estimated coefficient is -0.258 (p = 0.007), suggesting that a 10 percentage point
increase in community bank share was associated with a 2.6 percentage point decrease in the ratio
of filed:approved claims.33 As a placebo, we repeat the exercise during a period when we expect no
effect, T encompassing the weeks ending August 8 (when the program ended) through October 31.
There, the estimated coefficient is -0.015 (p = 0.102), meaning we detect no statistically significant
effect post-PPP, as expected.34
We further use filed and approved initial claims to revisit the challenge to identification. Recall
that the SBA approved the first PPP loan on April 3, 2020. Meanwhile, our estimates imply that
early PPP loans first suppressed initial UI claims during the week ending March 21 – reflecting
unemployment dating back to the preceding Monday (the 16th). Moreover, the weeks ending
on March 21, March 28, and April 4, 2020 feature the largest point estimates for initial claims.
Estimates for these three weeks are the focus of this section.
While approved UI claims are conditional on subsequent actions, filed UI claims are not. If an early
PPP recipient re-hires its employees quickly after they are laid off, it is significantly more likely
that the initial UI claim will either be canceled by the employee or appealed by the employer. This
would reduce the number of approved claims, while leaving filed claims unaffected. Therefore, we
test our claim by replicating our core regressions, using filed initial claims as a dependent variable.
The filed initial UI claims data come from Chetty et al. (Forthcoming, 2024), who collected these
data and kindly shared them for public use. There are two important differences to highlight
between our approved claims data and these filed claims data. First, the approved claims data
is observed at the (county x week x establishment size bin) level, while the filed claims data is
observed at the (county x week) level. Second, the approved initial claims data has a consistent set
of 2,643 counties in its panel, while the filed initial claims data reaches 1,134 counties per week at
most.
Of these two differences, the level of observation requires more consideration. To compare like
regressions, we aggregate the approved claims data up to the (county x week) level. However,
aggregation means the outcome variables now reflect population bases that are partly ineligible
for PPP support, requiring an adjustment to make these coefficients comparable to our primary
estimates. To make that adjustment, consider this simple model of an economy with small and
large firms (indexed by j ∈ {ω, Ω}).
Total unemployment claims in each county, c, at time, t, are simply the sum of unemployment
claims from small and large firms:
33

If we consider the tight timeframe of February 29 - April 4, then the coefficient is -0.423 (p = 0.016).
The results are nearly identical if we consider earlier timeframes. For example, when T encompasses the weeks
ending July 4 through October 31, the estimate is also -0.015 (p = 0.094).
34

37


Yct = Yωct + YΩct

(3)

Define αωc as the small firms’ share of covered employment in county c, and αΩc = (1 − αωc ). These
shares are not dependent on t, since covered employment is deliberately fixed to pre-pandemic levels.
Then, use yct to denote the overall county insured unemployment rate (IUR), with

yct = αωc yωct + (1 − αωc ) yΩct

(4)

For simplicity, consider a simple linear model for the claims rate at small and large firms:

′
yωct = β0,ωs(c)t + β1,ωt P P Pct′ + Xct
β2,t + ϵωct

(5)

′
yΩct = β0,Ωs(c)t + Xct
β2,t + ϵΩct

(6)

We assume that PPP loans have no effect on larger firms, thereby ruling out general equilibrium
effects. Though we rule them out, we argue that PPP loans functioned as aggregate demand
stimulus, which would lead general equilibrium effects to be strongly positive for employment.
The additional labor income, continuing rent payments, reduction in canceled orders, and other
small business expenditures should strongly outweigh any crowding out in employment. We have
also assumed that covariates have equal effects on small and large-firm IURs, which is a practical
limitation of using all-claims data.
Returning from the aside on general equilibrium, we then plug the above linear models into Equation
4, we get


′
yct = β0,Ωs(c)t + αωc β0,ωs(c)t − β0,Ωs(c)t + β1,t αωc P P Pct′ + Xct
β2,t + (αωc ϵωct + αΩc ϵΩct ) .

(7)

We instrument for PPP using county-level community bank deposit share, in the same way as our
primary specifications. In practice, this is the primary specification’s second stage regression with
two modifications. It adds αωc as a covariate and as an interaction term for P P Pct′ . The results
for this regression are shown in Figure X, with filed and approved claims estimated separately. We
highlight the weeks ending March 21, March 28, and April 4, which are the weeks when pre-effects
are a primary concern.

38


The standard errors for filed claims are relatively large, in part due to a smaller sample size.
However, when we pool the claims from these three weeks, we get more precise estimates (see Table
IV). Nonetheless, it is apparent that the disconcerting pre-trends are entirely absent from filed
claims data, as we require. The filed initial UI claims estimates reach their nadir during the week
ending March 21, with a coefficient estimate of -0.074 (p = 0.34). The two following weeks show
positive estimates: 0.119 (p = 0.49) and 0.0.067 (p = 0.65) for March 28 and April 4, respectively. In
contrast, the approved initial claims regression shows consistently negative effects. Chronologically,
the three weeks of concern have estimates of {-0.117, -0.058, -0.056}, with corresponding p-values
of {0.005, 0.404, 0.277}. When pooled, filed initial claims in these weeks carry a point estimate of
0.014 (p = 0.842) and approved initial claims carry a point estimate of -0.077 (p = 0.036).
We also note that filed initial claims seem to spike in the weeks following April 4. Though the
standard errors are rather large, this still might be disconcerting - nothing in our theory predicts a
correlation between filed initial claims and community bank shares, nor do we have an institutional
explanation for why this might be the case. To investigate the effects over those weeks of spiking
filed claims, we pool claims in the three weeks ending April 11, April 18, and April 25. Somewhat
reassuringly, we find that the coefficient estimate remains statistically insignificant in the pooled
regression (estimate = 0.104, p = 0.390). Tables A.3 and A.4 offer full results and details on the
pooled regressions for filed initial UI claims.
TABLE IV
Regression Coefficients for Pooled Initial UI Claims

Weeks ending 2020-03-21 - 2020-04-04
Point Estimate
Standard Error

Filed Claims
0.014
(0.074)

Approved Claims
-0.077∗∗
(0.035)

Weeks ending 2020-04-11 - 2020-04-25
Point Estimate
Standard Error

Filed Claims
0.104
(0.074)

∗

Approved Claims
-0.033∗
(0.019)

, ∗∗ represent significance at the 10% at 5% level, respectively. Both
regressions involve pooling three weeks of data together, and include stateby-week FE. Filed claims regressions are fully reported in Tables A.3 and
A.4. See footnote to those tables for details on the regression and its
covariates. Regressions for approved claims have similar adjustments,
and are estimated to match concepts as closely as possible.

39


Figure X
Large and statistically significant effects between March 21 – April 4 are present in regressions with
approved claims. Those weeks have no significant effects in regressions with filed claims.

VI.B.v.

Small Business Revenue and Industry Index as Controls

As a robustness check, we repeat our primary exercise with an additional regressor measuring small
business revenues. The county-level revenue data are from Womply, collected and publicly shared
by Chetty et al. (Forthcoming, 2024). Covid brought a profound, dynamic, and heterogeneous
economic shock, which naturally raises concerns about a host of unobservables. This exercise is
intended to address a broad range of potential endogeneity concerns. While revenues should be a
function of the loans following approval, that relationship should not hold prior to loan approval.
For example, if April 11 PPP loans have a positive and significant effects of on March 21 revenues,
then we may be concerned that our estimates reflect the effects of relatively weaker Covid rather
than the loans themselves.
Specifically, we include the average change in revenue over the three weeks ending April 4, 2020.
We do not include contemporaneous revenues, since those are an outcome of PPP loan approval
40


once the program debuted. The additional regressor should help to control for differences in the
early (pre-PPP) Covid economic shocks. This would be particularly important if the impact and
persistence of the initial shock was time-dependent.
Shown in Figure XI, the estimates from these regressions - with both initial and continuing claims
as outcome variables - illustrate that the main result is robust to controlling for pre-PPP revenue
changes. The estimates follow the same pattern, with a small decrease in magnitudes. For example,
the small firms continuing claims estimate for the week ending April 11 falls from 0.25 to 0.19.
Though the Anderson-Rubin p-values drift up somewhat, this is primarily due to the roughly
one-third reduction in sample size necessitated by the inclusion of Womply data. In our primary
exercise, that week’s sample has 2,338 counties, compared to 1,486 counties when including Womply
data.
In a second robustness check, we also include an county-level industry index as a control. However,
the inclusion of this covariate demands careful consideration, which we offer in appendix Appendix
B. In the appendix, we show heavy job losses in Accomodation and Food Services during the first
months of the pandemic, leading us to develop an industry-based model of job-losses. We show that
both excluding and including the industry index in our primary regression can introduce bias. We
conduct a regression based on the industry-based model of job losses, and the results are consistent
with our main conclusions regarding the jobs saved by PPP. The considerations around industry
controls are crucial to interpreting the estimates in column (5) of our regression tables (eg. table
III), which on their face show meaningful differences relative to our primary specifications in column
(4).

VI.B.vi.

Social Welfare

From a social welfare maximization perspective, we should consider three points when evaluating
the no-PPP counterfactual. First, as we and others document, state-based unemployment insurance
systems were not prepared to handle the realized spike in claims, let alone the volume of claims
that would have resulted absent PPP. This would have resulted in further extending the substantial
delays in making payments that occurred. Second, PPP helped to stem business closures that
would have lengthened UI spells and would have made it significantly harder for workers to return
to employment.35 Had the employee-employer relationships been broken, the costly, time-intensive,
and uncertain matching process would have significantly slowed the economic recovery. Clearly,
this is an important effect but given the nearly uniform coverage of PPP, evaluating these long-term
benefits is beyond the scope of this paper. Third, putting workers on UI rather than PPP would
likely have led to losses of health insurance coverage during a fast-moving public health crisis, since
35

The literature on the effects of long-run unemployment that comes with such closures is vast (for instance,
Jacobson, LaLonde, and Sullivan (1993), Davis and von Wachter (2012)).

41


(a) Initial UI Claims, Firms Sized 0-99

(b) Continuing UI Claims, Firms Sized 0-99, with added Revenue Control

Figure XI
IV estimates of the effect of PPP penetration as of April 11th on employment outcomes at firms sized 0-499. 95%
confidence bands shown. Regressions include control variables as described in Equation 2, in addition to the change
in county-level small business revenue, January 2019 - (three-week average of March 21 - April 4, 2020). Data on
small business revenues from Womply, as calculated, seasonally adjusted, and furnished by Chetty et al.
(Forthcoming, 2024).

42


most Americans still receive health insurance via employer-sponsored plans.36 Accordingly, the
objective of the CARES Act was to economically facilitate compliance with temporary stay-athome orders, as the public health establishment indicated that temporary work stoppages would
“flatten the curve.”
As Government Accountibility Office (2022) discusses, the state-based unemployment system struggled to handle the volume of claims submitted during the pandemic. Navarrete (2023) documents
that in states where the UI system is COBOL based, the delays were significantly longer than those
with a more modern computer language. Estimating the potential welfare losses that would have
occurred absent PPP requires knowing how much longer delays in receiving UI payments would
have been if the jobs PPP saved would have instead resulted in UI claims. We therefore gathered
data from 2007 to the present on the payments for initial UI claims that were delayed. Prior to
the pandemic, including during the Great Financial Crisis, in only one month did the percentage
of initial claims that were delayed five weeks or more exceed ten percent (March 2014 had 11% of
the payments be for claims filed in excess of five weeks earlier). Of the initial claims first paid in
May 2020, 21% of them represented claims that were at least five weeks old. For June 2020, 38%
of the paid initial claims were at least five weeks old. This delay percentage did not dip below 30%
until November 2020.
We run univariate regressions of delayed UI payments on initial UI claims. Specifically, delayed UI
payments are defined as payments more than five weeks since the first compensable week. They
are measured by state, expressed as a fraction of pre-pandemic covered employment, and cover all
payments initiated between May 2020 and October 2020. Initial UI claims are also measured by
state, also expressed as a fraction of pre-pandemic covered employment, and cover all claims filed
between March 2020 and May 2020. We estimate that for every ten percentage point increase in
initial UI claims, there is a 1.7 percentage point increase in the fraction of pre-pandemic employed
people receiving a late initial UI payment. Such a result suggests that states facing abnormally
higher UI claims saw significantly greater delays in the processing of new UI claims. This evidence
supports the contention that absent PPP, an even larger fraction of households would have suffered
a multi-month loss of income. There are likely important non-linearities in delays and when we
estimate this same regression using a 4th order polynomial, we find minimal difference going from
UI claims of ten percent to twenty percent but when going from twenty to thirty percent, delays
increase by a factor of almost 2.5. Projected across the US population, that represents an additional
four million initial UI claim filings that would be delayed at least five weeks.
Putting aside the potential for further delays in UI processing and payments, UI was also an
expensive proposition for the federal and state governments in 2020. With an additional $600/week
36

While many job losers would have sought out Affordable Care Act plans while on UI or paid high COBRA
premiums, there is no guarantee that all would have done so, and there would have been gaps between when some
lost employer-sponsored coverage and when they gained ACA coverage.

43


Federal Pandemic Unemployment Compensation (FPUC) payment on top of the Q2 2020 average
base UI payment of $31937 , a worker on UI for the 17 weeks of FPUC (April 4 - July 31) would
have collected $15,617. If the estimated 14.0 million jobs saved by PPP loans had instead been
14.0 million additional workers collecting 17 weeks of UI payments, this would have cost federal
and state governments an additional $218 billion dollars. Given the long length of UI spells over
this period, we consider 17 weeks to be a relatively conservative assumption. The total number
of weeks compensated in Q3 2020 was 166.7 million, down from Q2 2020’s 207.9 million. Workers
were not coming off the rolls en masse at the end of July 2020, and therefore would have likely
had to have been compensated for more weeks of unemployment. Given the potential cost of the
alternative-both in dollars and in additional UI wait time-PPP appears to be a relatively strong
option from a social welfare perspective.

VII. Conclusion
We estimate the employment effects of PPP loans, finding significantly more job preservation than
the extant literature. To do so, we leverage heterogeneity in local banking markets to identify
exogenous differences between early loan recipients and later loan recipients. We find strong evidence that early PPP loan receipt led to superior labor market outcomes. As PPP loans achieved
near-total saturation across the country, those advantages faded, a dynamic consistent with our
hypothesis.
Our unique UI data, with observations at the (county x week x firm-size) level permits a finer
firm-size disambiguation of PPP’s effects than has been achieved to this point in the literature.
We establish that coarser measures of outcomes will attenuate estimates, which emphasizes the
importance of these more granular data.
PPP’s effects were strongest for the smallest firms, who were the most vulnerable to the financial
shock of COVID closures. In total, we estimate that PPP loans saved 10.9 million jobs at firms
sized 0-99, thereby alleviating pressure on aging unemployment insurance systems that were not
designed to handle the scale of layoffs the pandemic would have brought absent PPP. The empirical
focus on the smallest firms is of first-order importance, since nearly 70% of all PPP dollars flowed
to those companies, and preserving employment at these smaller companies was the primary goal
of PPP. These findings are consistent with previous literature documenting that smaller firms are
more likely to face financial constraints and be less resilient during economic shocks.
We also estimate the total number of jobs saved at all firms employing 0-499 workers. We estimate
that PPP saved 14.0 million jobs in total – yielding an average cost between $33,200 and $37,600
per job.
37

Source: Employment and Training Administration Unemployment Insurance Data

44


References
Andrews, Isaiah, James H. Stock, and Liyang Sun. 2019. “Weak Instruments in IV Regression:
Theory and Practice.” Annual Review of Economics .
Autor, David, David Cho, Leland Crane, Mita Goldar, Byron Lutz, Joshua Montes, William Peterman, David Ratner, Daniel Villar, and Ahu Yildirmaz. 2022a. “An Evaluation of the Paycheck
Protection Program Using Administrative Payroll Microdata.” Journal of Public Economics
211:104664.
Autor, David, David Cho, Leland D. Crane, Mita Goldar, Byron Lutz, Joshua Montes, William B.
Peterman, David Ratner, Daniel Villar, and Ahu Yildirmaz. 2022b. “The $800 Billion Paycheck
Protection Program: Where Did the Money Go and Why Did It Go There?” Journal of Economic
Perspectives 36 (2):55–80.
Barlett, Robert P. and Adair Morse. 2021. “Small Business Survival Capabilities and Policy Effectiveness: Evidence from Oakland.” Journal of Financial and Quantitative Analysis 56 (7):2500–
2544.
Barraza, Santiago, Martı́n Rossi, and Timothy J. Yeager. 2020. “The Short-Term Effect of the
Paycheck Protection Program on Unemployment.” Working Paper .
Barrero, Jose Maria, Nicholas Bloom, and Steven J. Davis. 2020. “COVID-19 Is Also a Reallocation
Shock.” Brookings Papers on Economic Activity :329 – 371.
Bartik, Alexander, Marianne Bertrand, Zoë B. Cullen, Edward L. Glaeser, Michael Luca, and
Christopher Stanton. 2020a. “How Are Small Businesses Adjusting to COVID-19? Early Evidence From a Survey.” NBER Working Paper No. 26989 .
———. 2020b. “The Impact of COVID-19 on Small Business Outcomes and Expectations.” Proceedings of the National Academy of Sciences 117:202006991.
Bartik, Alexander W., Zoë B. Cullen, Edward L. Glaeser, Michael Luca, Christopher Stanton, and
Adi Sunderam. 2020c. “The Targeting and Impact of Paycheck Protection Program Loans to
Small Businesses.” NBER Working Paper 27623 .
Cajner, Tomaz, Andrew Figura, Brendan M. Price, David Ratner, and Alison Weingarden. 2020.
“Reconciling Unemployment Claims with Job Losses in the First Months of the COVID-19 Crisis.” Finance and Economics Discussion Series 2020-055. Washington: Board of Governors of
the Federal Reserve System .

45


Chetty, Raj, John N. Friedman, Nathaniel Hendren, Michael Stepner, and The Opportunity Insights
Team. Forthcoming, 2024. “The Economic Impacts of COVID-19: Evidence from a New Public
Database Built Using Private Sector Data.” Quarterly Journal of Economics .
Chodorow-Reich, Gabriel. 2014. “The Employment Effects of Credit Market Disruptions: Firmlevel Evidence from the 2008-09 Financial Crisis.” Quarterly Journal of Economics 129 (1):1–59.
Chodorow-Reich, Gabriel, Olivier Darmouni, Stephan Luck, and Matthew Plosser. 2022. “Bank Liquidity Provision Across the Firm Size Distribution.” Journal of Financial Economics 144 (3):908–
932.
Cohen, Patricia. 2020. “Another 2.4 Million Jobs Vanish, And Many May Be Gone Forever.” New
York Times, Print Edition URL https://www.nytimes.com/2020/05/21/business/economy/
coronavirus-unemployment-claims.html.
Cole, Allison. 2024. “The Impact of the Paycheck Protection Program on Really Small Businesses.”
Working Paper .
Council of Economic Advisers. 2020. “Tracking Chapter 7 and Chapter 11 Bankruptcies.” Economic
Issue Brief .
Coy, Peter. 2020. “Bankruptcy Decline Surprises Goldman Chief Economist Mericle.” Bloomberg
Businessweek .
Dalton, Michael. 2023. “Putting the Paycheck Protection Program into Perspective: An Analysis
Using Administrative and Survey Data.” National Tax Journal 76 (2):393–437.
Davis, Steven J. and Till von Wachter. 2012. “Recessions and the Costs of Job Loss.” Brookings
Papers on Economic Activity .
Denes, Matthew, Spyridon Lagaras, and Margarita Tsoutsoura. 2021. “First Come, First Served:
The Timing of Government Support and Its Impact on Firms.” Working Paper .
Doniger, Cynthia and Benjamin Kay. 2023. “Long-Lived Employment Effects of Delays in Emergency Financing for Small Business.” Journal of Monetary Economics 140.
Duygan-Bump, Burcu, Alexey Levkov, and Judit Montoriol-Garriga. 2015. “Financing Constraints
and Unemployment: Evidence from the Great Recession.” Journal of Monetary Economics
75:89–105.
Ekvall, Lei Lei Wang and Timothy Evanston. 2020. “The Small Business Reorganization Act: Big
Changes for Small Businesses.” ABA Business Law Today URL https://www.americanbar.
org/groups/business_law/publications/blt/2020/02/small-business-reorg/.
46


Farrell, Diana and Chris Wheat. 2016. “Cash is King: Flows, Balances, and Buffer Days. Evidence from 600,000 Small Businesses.” JP Morgan Chase Research Institute Discussion Paper
URL https://institute.jpmorganchase.com/content/dam/jpmc/jpmorgan-chase-and-co/
institute/pdf/jpmc-institute-small-business-report.pdf.
Federal Reserve Banks of New York, Atlanta, Cleveland, and Philadelphia. 2014. “Joint Small
Business Credit Survey Report.” URL https://www.newyorkfed.org/medialibrary/media/
smallbusiness/SBCS-2014-Report.pdf.
Fox, Zach, Benjamin Yung, Ali Shayan Sikander, and Brian Scheid. 2020. “As virus crisis persists,
PPP recipients lay off thousands.”

Standard and Poors Market Intelligence URL https:

//www.spglobal.com/marketintelligence/en/news-insights/latest-news-headlines/
as-virus-crisis-persists-ppp-recipients-lay-off-thousands-59602815.
Government Accountibility Office. 2022. “Unemployment Insurance: Transformation Needed to
Address Program Design, Infrastructure, and Integrity Risks.” GAO-22-105162 .
Granja, João, Christos Makridis, Constantine Yannelis, and Eric Zwick. 2022. “Did the Paycheck
Protection Program Hit the Target?” Journal of Financial Economics 145 (3):725–761.
Haggerty, Neil. 2020. “Did Banks Play Favorites in PPP or Were They Just Being Prudent?”
American Banker .
Hubbard, Glenn and Michael R. Strain. 2020. “Has the Paycheck Protection Program Succeeded?”
Brookings Papers on Economic Activity 51 (3):335–390.
Humphries, John Eric, Christopher A. Neilson, and Gabriel Ulyssea. 2020. “Information Frictions
and Access to the Paycheck Protection Program.” Journal of Public Economics 190:104244.
Irwin,
York

Neil. 2020.
Times,

“Despite Loans,
Print

Edition

Many Small Businesses Won’t Survive.”

URL

New

https://www.nytimes.com/2020/05/21/upshot/

virus-small-business-closures.html.
Jacobson, Louis S., Robert J. LaLonde, and Daniel G. Sullivan. 1993. “Earnings Losses of Displaced
Workers.” The American Economic Review 83 (4):685–709.
Joaquim, Gustavo and Felipe Netto. 2021. “Bank Incentives and the Effect of the Paycheck Protection Program.” FRB of Boston Working Paper .
———. 2024. “Optimal Allocation of Relief Funds: The Case of the Paycheck Protection Program.”
Working Paper .
Joaquim, Gustavo and J. Christina Wang. 2023. “What Do 25 Million Records of Small Businesses
Say About the Effects of the PPP?” FRB of Boston Working Paper .
47


Kleibergen, Frank and Richard Paap. 2006. “Generalized Reduced Rank Tests Using the Singular
Value Decomposition.” Journal of Econometrics 133 (1):97–126.
Li, Lei and Philip E. Strahan. 2021. “Who Supplies PPP Loans and Does It Matter? Banks, Relationships, and the COVID Crisis.” Journal of Financial and Quantitative Analysis 56 (7):2411–
2438.
Liu, Haoyang and Desi Volker. 2020. “Where Have the Paycheck Protection Loans Gone So Far?.”
New York Federal Reserve URL https://libertystreeteconomics.newyorkfed.org/2020/
05/where-have-the-paycheck-protection-loans-gone-so-far.html.
Lopez, Jose and Mark Spiegel. 2023. “Small Business Lending under the PPP and PPPLF Programs.” Journal of Financial Intermediation 53:101017.
Merker, Vivian. 2020. “PPP is a Compliance Minefield for Banks.” American Banker .
Moreira, Marcelo J. 2009. “Tests with Correct Size When Instruments Can Be Arbitrarily Weak.”
Journal of Econometrics 152 (2):131–140.
Olea, José Luis Montiel and Carolin Pflueger. 2013. “A Robust Test for Weak Instruments.” Journal
of Business & Economic Statistics 31 (3):358 – 369.
Petersen, Mitchell A. and Raghuram G. Rajan. 1997. “Trade Credit: Theories and Evidence.” The
Review of Financial Studies 10 (3):661–691.
Rubio, Marco. 2020. “Rubio Sends Letters to Large Banks Participating in the Paycheck Protection
Program.” Press Release URL https://www.rubio.senate.gov/public/index.cfm/2020/4/
rubio-sends-letters-to-large-banks-participating-in-the-paycheck-protection-program.
Ruhle, Stephanie, Leticia Miranda, and Michael Capetta. 2020. “PPP likely saved 35 million jobs,
says JP Morgan CEO Jamie Dimon.” NBC News URL https://www.nbcnews.com/business/
economy/ppp-likely-saved-35-million-jobs-says-jpmorgan-chase-ceo-n1236341.
Siemer, Michael. 2019. “Employment Effects of Financial Constraints during the Great Recession.”
The Review of Economics and Statistics 101 (1):16–29.
Splinter, David, Eric Heiser, Michael Love, and Jacob Mortenson. 2024. “The Paycheck Protection
Program: Progressivity and Tax Effects.” Working Paper .
Staples, Aaron J. and Thomas P. Krumel. 2023. “The Paycheck Protection Program and Small
Business Performance: Evidence from Craft Breweries.” Small Business Economics 61 (3):931–
956.

48


Summers, Larry H. 2008. “Fiscal Stimulus Issues.” Testimony Before the Joint Economic Commitee, January 16, 2008 .
Velazquez, Nydia and Maxine Waters. 2020.

“Waters and Velazquez Write to Megabank

CEOs to Express Strong Concerns About Small Business Access to Paycheck Protection Program.” Press Release URL https://financialservices.house.gov/news/documentsingle.
aspx?DocumentID=406496.
Wright, Philip G. 1928. The Tariff on Animal and Vegetable Oils. New York, NY: MacMillan.

49


Appendix A: Additional Figures

Figure XII
Filed initial UI claims per week, not seasonally adjusted.

Figure XIII
Distribution of the time lapsed between initial claim filing and claim payment, by month of payment.

50


Appendix B: Industry Shares as a Covariate
Industry strongly affected the probability of job loss in March and April 2020 (see figure XIV).
The Leisure and Hospitality industry stood at one end of the spectrum, losing 48% of all payroll
jobs from February through April 2020, according to BLS’ Current Employment Statistics. On the
other end was the Utilities industry, which only shed 1% of its jobs. Based on this heterogeneity,
it would seem important to control for industry shares in this paper’s primary regressions - though
there are conceptual complications with its inclusion.

Figure XIV
Nationwide Job Loss by Industry, from February - April 2020, source: Bureau of Labor Statistics’ Current
Establishment Statistics.

As a robustness check, we include a Bartik-like control for county-level predicted job loss based on
pre-pandemic industry shares. Specifically, the control Γcj is defined as

Γcj ≡

X

λi scji

i

Where scji is the share of industry i in county c’s small-firm employment at firms in size-bucket
j, based on the 2019 Census SUSB. We define λi as the nationwide fraction of jobs lost between

51


February and April 2020 in the 2-digit NAICS industry i.38 In regression tables, we call Γcj
“Industry Share”. More precisely, it is a sort of predicted county-level job loss, on the basis of
pre-pandemic industry shares.
The ‘predicted county-level job loss’ industry index differs from other papers in the literature. In
contrast, papers with firm-level microdata typically include industry fixed effects or industry-week
fixed effects (e.g. Autor et al. (2022a) or Cole (2024)). Chetty et al. (Forthcoming, 2024), which
uses aggregated data similar to this paper, pools all industries but weights its data by industry-level
pre-pandemic shares (see that paper’s Online Appendix Figure XXVIII). Doniger and Kay (2023)
uses pre-pandemic employment shares (by county) in three select industries. Our single measure
allows us to account for time-varying employment effects of industry shares across 2-digit NAICS
codes without adding 19 extra covariates to each regression.
Including this covariate typically has a substantial impact on the estimated coefficients in our
primary regressions. However, the case for including industry shares as a covariate is not as clearcut as it might seem. To illustrate the relevant considerations, we write a simple model of PPP
and heterogeneous job loss probabilities, based on industry. Suppose the data-generating process
(DGP) for unemployment is given by

!
ycjt = 1 − ξP P P,t P P Pcjt′

 X

λi scji

′
+ Xcjt
ξ1,jt + µcjt

i
′
= Γcj − ξP P P,t P P Pcjt′ Γcj + Xcjt
ξ1,jt + µcjt

(8)

where ycjt is the insured unemployment rate in county c at time t for firms sized j, and P P Pcjt′ is
the fraction of jobs covered by PPP loans at time t′ . In words, baseline county-level unemployment
is determined by industry shares, but PPP loans reduce unemployment at rate ξP P P,t . Note that
the coefficient on PPP loans, ξP P P,t , differs from the coefficient in the main body of this paper
(βP P P,t ). They have different interpretations: the main paper coefficient is simply the effect of
PPP loans on unemployment. The coefficient ξP P P,t is the efficacy of PPP loans in reducing the
counterfactual baseline unemployment rate, Γcj .39
If this simple model describes the true data generating process, then - when Γcj is omitted - the
38

In practice, we measure λi using the employment data from Opportunity Insights for the week ending April
3. This includes four aggregated industries (Leisure & Hospitality, Retail & Transportation, Education & Health
Services, and Professional & Business Services) and their residual. We also conduct the regressions below using true
2-digit NAICS code data from BLS’ Current Employment Statistics, measuring the decline in payrolls from the week
of February 12 - the week of April 12. The differences in results were negligible.
39
We do not make this our primary specification for a number of reasons, including the endogeneity inherent in
the baseline rate, Γcj .

52


estimate in our primary regression is, in expectation,



˜ P cjt′ , ỹcjt
Cov P P


β̂P P P,t =
˜ P cjt′
V ar P P




˜ P cjt′ , Γ̃cj
˜ P cjt′ , P P Pcjt
˜ ′ ∗ Γcj
Cov P P
Cov P P

 + ξP P P,t


=
˜ P cjt′
˜ P cjt′
V ar P P
V ar P P
where the tilde operator represents the residuals from a regression of each given variable on the
′ . The first term in this expression is classic omitted variable bias. The
vector of controls, Xcjt

second term is an attenuated measurement of βP P P,t , further attenuated if Γcj is included as a
covariate. We deem this an “opportunity effect”. In words, if PPP was poorly targeted at first,
then this term underestimates the eventual effect of PPP loans once the industry index is included
as a regressor. By analogy, suppose that a medicine is effective for sick patients, but in a trial the
assignment of treatment is biased towards a healthier population. Then, controlling for pre-trial
health appropriately corrects for the assignment bias, however the estimates do not reflect the true
effect that the medicine would have in the sickest patients.
Since we observe or estimate all the terms in equation Appendix B, we can therefore estimate this
regression directly. Like in the main paper, we first estimate each week separately and plot the
coefficients ξP P P,t in Figure XV. We also pool key weeks and present regression tables from that
pooled cross-section below. For initial claims, we pool the three weeks ending on March 21, March
28, and April 4, 2020, while for continuing claims we pool the three weeks ending April 11, April
18, and April 25, 2020. These estimates are shown in Tables A.1 and A.2.
For the model in equation (8) (column 5), the estimate of ξP P P,t in the initial claims regression is 0.248 (Anderson-Rubin p-value of 0.006). The estimate of ξP P P,t in the continuing claims regression
is -0.620 (Anderson-Rubin p-value of 0.012). This latter estimate would aggregate to 6.8 million
jobs saved at firms sized 0 - 99 upon full roll-out of PPP loans.40 This result differs from those in
the main paper, and we still prefer the headline estimates to the results delivered by this stylized
industry index-based model. The industry index-based estimate comes with an implicit assumption
that job losses in the counterfactual ‘no-PPP’ scenario would have reached their maximum extent in
the week ending April 3.41 However, net job losses continued for two weeks after April 3, according
to Opportunity Insights data. Further, we consider it likely that job losses would have continued
40


P
′
The specific calculation is 0.620 *
c P P Pcjt ∗ Γcj ∗ Employmentcj,P re−P andemic with j representing firms of
0 - 99 employees.
41
The week ending April 3 is the week for which we measure industry-level job-loss shares (λi ), as reported by
Opportunity Insights. We choose this week since it is the last week before the shares are directly affected by PPP
loans.

53


into later weeks without PPP, and that some industries which were lesser-hit initially would have
increased layoffs once it became clear that the pandemic – and the shelter-in-place orders – would
continue longer than initially expected.

54


(a) Initial UI Claims, Firms Sized 0-99

(b) Continuing UI Claims, Firms Sized 0-99

Figure XV
IV estimates of the effect of PPP penetration as of April 11th on employment outcomes at firms sized 0-499.

55


TABLE A.1
(Continuing Claims * Industry Index), Firms Size 0-99 (Pooled Regression of First Tranche Weeks)
(1)

(2)

(3)

(4)

(5)

(6)

-0.665∗∗∗
(0.171)

-0.713∗∗∗
(0.149)

-0.500∗
(0.226)

-0.590∗
(0.263)

-0.620∗∗
(0.192)

-0.466∗∗
(0.180)

February IUR

1.582∗∗∗
(0.187)

1.555∗∗∗
(0.216)

1.402∗∗∗
(0.164)

1.377∗∗∗
(0.165)

Log(Med. Income)

-0.027
(0.014)

-0.018
(0.020)

-0.049∗∗
(0.015)

-0.047∗∗
(0.014)

Poverty Rate

-0.002∗
(0.001)

-0.001
(0.001)

-0.003∗∗∗
(0.001)

-0.003∗∗∗
(0.001)

Log(Pop. Density)

0.006∗∗∗
(0.002)

0.006∗
(0.003)

0.004∗∗
(0.002)

0.004∗
(0.002)

Covid Cases, 1w

-6.386
(3.495)

-5.110∗
(2.315)

-5.418∗
(2.585)

Covid Cases, 4w

1.423
(1.563)

1.302
(0.917)

1.435
(0.956)

Covid Deaths, 1w

1.575
(7.619)

-6.791
(5.801)

-5.691
(7.124)

Covid Deaths, 4w

1.655
(7.907)

-0.698
(4.384)

-1.601
(4.523)

WFH Index

-0.093
(0.074)

0.141∗
(0.061)

0.133∗
(0.052)

0.853∗∗∗
(0.130)

0.764∗∗∗
(0.112)

(Early PPP Coverage x Industry Index)

Industry Index

-0.027∗∗∗
(0.008)

March Small-Firm Rev.

State-by-Week FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
7,081
14.8
88.4
[ -1.008, -0.262]
0.007

Yes
7,081
0.2
75.4
[ -1.062, -0.339]
0.008

Yes
7,014
19.7
16.2
[ -1.119, 0.020]
0.057

Yes
7,011
53.9
14.4
[ -1.458, -0.057]
0.036

Yes
7,011
127.1
15.1
[ -1.234, -0.220]
0.012

Yes
4,465
146.6
14.0
[ -1.034, -0.058]
0.034

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt (P P Pcjt′ ∗ Γcj ) + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an
approved UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage * Industry Index) is the product
of (a) the fraction of jobs at firms sized 0 - 99 which are covered by PPP loans as of April 11, 2020 and (b) the predicted number of jobs lost, based on county-level industry
composition of small firms. This variable is instrumented by the product of (a) the county-level share of deposit funds in community banks and (b) the same industry index.
Details on measurements are covered in the Data section in the main paper and in this section of the appendix. Principal covariates: state, week, and state-by-week fixed
effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income),
the poverty rate, and log(population density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior
four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the
inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county
c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j nationwide, as of April 3, 2020
(from Opportunity Insights composite employment measure) and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March
Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. In this regression,
the weeks ending 2020-04-11 through 2020-04-25 are pooled, with the remaining weeks excluded from the data.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

56


TABLE A.2
Initial Claims, Firms Size 0-99 (Pooled Regression of Covid-Onset Weeks)
(1)

(2)

(3)

(4)

(5)

(6)

-0.035
(0.073)

-0.272∗∗∗
(0.061)

-0.234∗
(0.115)

-0.245∗
(0.106)

-0.248∗∗
(0.076)

-0.193∗∗
(0.066)

February IUR

3.985∗∗∗
(1.112)

3.936∗∗∗
(1.029)

4.366∗∗∗
(0.743)

4.788∗∗∗
(0.684)

Log(Med. Income)

-0.012
(0.006)

-0.006
(0.007)

-0.017∗∗∗
(0.005)

-0.016∗∗∗
(0.004)

Poverty Rate

-0.001∗∗
(0.000)

-0.001
(0.000)

-0.001∗∗∗
(0.000)

-0.001∗∗∗
(0.000)

Log(Pop. Density)

0.001
(0.001)

0.001
(0.001)

0.001
(0.001)

0.000
(0.001)

Covid Cases, 1w

-0.852
(0.706)

-1.946∗∗∗
(0.530)

-1.702∗∗
(0.571)

Covid Cases, 4w

0.551
(0.600)

1.189∗∗
(0.397)

1.093∗
(0.429)

Covid Deaths, 1w

-69.099∗
(28.751)

-39.322
(46.333)

-42.068
(50.133)

Covid Deaths, 4w

43.659∗
(19.617)

12.885
(33.407)

14.856
(36.020)

WFH Index

-0.042
(0.032)

0.059∗
(0.025)

0.060∗∗
(0.021)

0.356∗∗∗
(0.055)

0.322∗∗∗
(0.048)

(Early PPP Coverage x Industry Index)

Industry Index

-0.009∗∗∗
(0.003)

March Small-Firm Rev.

State-by-Week FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
7,761
0.2
102.8
[ -0.209, 0.107]
0.623

Yes
7,761
0.1
69.6
[ -0.454, -0.148]
0.003

Yes
7,761
60.3
17.9
[ -0.559, 0.001]
0.051

Yes
7,758
15.6
18.3
[ -0.586, -0.047]
0.020

Yes
7,758
128.2
21.2
[ -0.475, -0.098]
0.006

Yes
4,743
146.6
21.3
[ -0.388, -0.058]
0.014

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ ∗ Γcj + X′cjt β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an
approved UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage * Industry Index) is the product of
(a) the fraction of jobs at firms sized 0 - 99 which are covered by PPP loans as of April 11, 2020 and (b) the predicted number of jobs lost, based on county-level industry
composition of small firms. This variable is instrumented by the product of (a) the county-level share of deposit funds in community banks and (b) the same industry index.
Details on measurements are covered in the Data section in the main paper and in this section of the appendix. Principal covariates: state, week, and state-by-week fixed
effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic),
county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid cases and deaths (separately) over the
prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From
Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the
employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in
industry j nationwide, as of April 3, 2020 (from Opportunity Insights composite employment measure) and (b) the employment-share of industry j in county c (for firms size 0 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment
at firms sized 0-99. In this regression, the weeks ending 2020-03-21 through 2020-04-04 are pooled, with the remaining weeks excluded from the data.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

57


Appendix C: Pooled Regression Results for Filed Initial Claims

58


TABLE A.3
Initial Claims, Firms Size 0-99 (Pooled Regression of Covid-Onset Weeks)
(1)

(2)

(3)

(4)

(5)

(6)

0.003
(0.022)

-0.000
(0.019)

0.018
(0.077)

0.014
(0.074)

0.059
(0.081)

0.023
(0.074)

February IUR

3.196∗∗∗
(0.812)

2.969∗∗∗
(0.836)

2.972∗∗∗
(0.832)

3.040∗∗∗
(0.884)

Log(Med. Income)

-0.011∗
(0.005)

-0.007
(0.007)

-0.005
(0.007)

-0.007
(0.007)

Poverty Rate

-0.001∗
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Log(Pop. Density)

0.001
(0.001)

0.002
(0.001)

0.002∗
(0.001)

0.002
(0.001)

Eligible Firm Employment Share

-0.008
(0.023)

-0.017
(0.026)

-0.032
(0.029)

-0.017
(0.027)

Covid Cases, 1w

1.381∗
(0.653)

0.737
(0.702)

1.227
(0.673)

Covid Cases, 4w

-0.475
(0.631)

0.103
(0.728)

-0.315
(0.661)

Covid Deaths, 1w

1.435
(19.312)

-13.654
(21.764)

-4.697
(19.515)

Covid Deaths, 4w

-10.226
(13.960)

-5.845
(15.800)

-8.106
(15.928)

WFH Index

-0.038
(0.022)

-0.027
(0.026)

-0.041
(0.025)

Share-Adjusted Early PPP Coverage

0.132∗
(0.062)

Industry Index

March Small-Firm Rev.

State-by-Week FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.009
(0.005)
No
3,982
0.0
71.1
[ -0.048, 0.054]
0.899

Yes
3,982
0.0
98.1
[ -0.038, 0.057]
0.997

Yes
3,233
17.5
6.4
[ -0.128, 0.763]
0.805

Yes
3,233
343.0
6.2
[ -0.146, 0.582]
0.851

Yes
3,233
469.0
6.3
[ -0.105, 0.718]
0.430

Yes
2,163
527.3
5.9
[ -0.144, 0.740]
0.747

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for Early PPP Coverage. The
Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with a filed UI initial claim in the given
week. The primary variable of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99 which are covered by PPP loans as of April 11, 2020, times the small firm share of employment in
county c. The adjustment of the primary regressor (relative to the baseline model) is described in section VI.B.iv. The instrument remains the county-level share of deposit funds in community banks.
Details on measurements are covered in the Data section in the main paper. Principal covariates: state, week, and state-by-week fixed effects, February 2020 IUR (IUR in terms of initial claims in this
case) for workers from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density),
from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights.
The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs that can be
done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in
industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level.
March Small-Firm Revenue comes from Womply, via Opportunity Insights. We also include county c’s small firm employment share as a covariate, for reasons described in section VI.B.iv. Regressions
weighted by the pre-pandemic county-level employment at firms sized 0-99. In this regression, the weeks ending 2020-03-21 through 2020-04-04 are pooled, with the remaining weeks excluded from the
data.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

59


TABLE A.4
Initial Claims, Firms Size 0-99 (Pooled Regression of First Tranche Weeks)
(1)

(2)

(3)

(4)

(5)

(6)

-0.001
(0.031)

-0.003
(0.028)

0.089
(0.117)

0.104
(0.116)

0.107
(0.113)

0.124
(0.112)

February IUR

1.822∗
(0.796)

1.288
(0.992)

1.286
(0.995)

1.074
(0.967)

Log(Med. Income)

-0.009
(0.010)

0.003
(0.010)

0.003
(0.010)

0.004
(0.011)

Poverty Rate

-0.000
(0.000)

0.000
(0.000)

0.000
(0.001)

0.000
(0.001)

Log(Pop. Density)

0.002
(0.001)

0.004∗
(0.002)

0.004∗
(0.002)

0.004∗
(0.002)

Eligible Firm Employment Share

-0.027
(0.032)

-0.047
(0.037)

-0.048
(0.036)

-0.053
(0.036)

Covid Cases, 1w

-0.183
(0.460)

-0.181
(0.453)

-0.108
(0.456)

Covid Cases, 4w

0.255
(0.244)

0.261
(0.249)

0.275
(0.232)

Covid Deaths, 1w

-6.626∗∗
(2.484)

-6.736∗∗
(2.335)

-7.332∗∗
(2.308)

Covid Deaths, 4w

-1.093
(1.643)

-1.165
(1.677)

-1.397
(1.534)

WFH Index

-0.066
(0.038)

-0.065
(0.040)

-0.073∗
(0.037)

Share-Adjusted Early PPP Coverage

Industry Index

0.009
(0.039)

March Small-Firm Rev.

State-by-Week FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.001
(0.006)
No
3,982
0.0
71.2
[ -0.064, 0.081]
0.971

Yes
3,982
0.0
98.3
[ -0.063, 0.081]
0.912

Yes
3,233
367.5
6.4
[ -0.105, 1.373]
0.377

Yes
3,233
8047.9
5.7
[ -0.087, 1.258]
0.290

Yes
3,233
9018.3
5.8
[ -0.085, 1.061]
0.267

Yes
2,163
17660.8
5.5
[ -0.064, 1.438]
0.185

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with a filed UI
initial claim in the given week. The primary variable of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99 which are covered by PPP loans as of April
11, 2020, times the small firm share of employment in county c. The adjustment of the primary regressor (relative to the baseline model) is described in section VI.B.iv. The
instrument remains the county-level share of deposit funds in community banks. Details on measurements are covered in the Data section in the main paper. Principal covariates:
state, week, and state-by-week fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the inner
product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of industry
j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. We also include county c’s small
firm employment share as a covariate, for reasons described in section VI.B.iv. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. In this
regression, the weeks ending 2020-04-11 through 2020-04-25 are pooled, with the remaining weeks excluded from the data.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

60


Appendix D: Online Appendix

1


Appendix E: Data
A.

SBA (PPP Loans)

The data on PPP loans is courtesy of the Small Business Administration (SBA). It is an augmented
version of the publicly available data on PPP loans.42 The version used for this paper was furnished
directly by the SBA following the conclusion of the first round of loans (the data cut was made in
November 2020). While our data and the publicly available data are closely aligned, there are a
handful of differences. The publicly available data has some censored information, which is not the
case in our data. We also observe additional variables, though the crucial variables are observed in
both versions of the data. We modify the raw data to improve data coverage and accuracy. In the
end, we observe 5,136,362 first round loans. The publicly available SBA data has 5,136,386 loans
for the first round.

•

We drop loans which were approved, but later fully cancelled. We also drop loans that had
an “Active Un-Disbursed” status in November 2020, but were not observed in later publicly
available data (75,766 loans, or 1.5

•

We start with the SBA’s most recent report of firm-level jobs covered, instead of the initial
reported number of jobs. The most recent number has better coverage – 1.3% observations
do not include those data, whereas the initial jobs reported variable is missing in 6.5% observations. In cases where we have both an initial reported number of jobs and an updated one,
12.1% diverge.

•

Where an initial jobs number is reported, but an updated one is not, we use the initial number
(0.3%). When either (a) both counts are missing or (b) both counts exceed 1,000, then we
replace with the most recent publicly reported number (0.9%). When jobs are not observed
in any of the three variables (26 observations, total), then we use simple imputation based
on dollars per employee.

•

We set an upper bound on firm size at 1,000 employees. While the 500-employee cutoff became
a popularly-known feature of PPP loans, the cutoff was not strict. Beyond the “NAICS 72”
exemption for restaurants and accommodations, firms could also establish eligibility based
on the SBA’s definition of a “small business interest”.43 Therefore, when the updated count
exceeds 1,000 while the initial count does not, then we use the initial count (0.3% of all loans).

42

Publicly available data can be accessed at https://data.sba.gov/dataset/ppp-foia
For a complete list of standards, see https://www.sba.gov/document/support-table-size-standards. In terms of
6-digit NAICS codes, 306 industries have explicit employee-size limits in excess of 500. A further 299 industries had
revenue-based cutoffs of $16.5 million per year or greater.
43

2


– For firms with ≥ 500 employees, we replace the number of jobs with a simple imputation
if loans are less than $735,000 (affecting 952 loans). This brings the jobs count among
these firms down from 567,569 to 22,139. We choose this specific cutoff so that a firm
with only 500 part-time employees (working 20 hours per week), all earning the federal
minimum wage, would not be affected – assuming they applied for full salary coverage.
•

At the bottom, we winsorize jobs per loan to one, assuming that each loan covers at minimum
one job. This is consistent with the SBA’s methodology, and accounts for loan recipients who
neglected to count themselves.

There are 46,347 observations where our final jobs number differs from the publicly available data.
Most differences are small – see plot below. An exception is with some firms marked as having 500
employees in the public data. Based on loan sizes, we believe that many of these are misclassified.
See comparative distributions (below) for the difference.

Figure XVI
Histogram of the difference between individual firm sizes observed in the publicly available data and this
paper’s data. Histogram is censored at a difference of 50 employees.

3


(a) Based on this Paper’s Data

(b) Based on Publicly Available Data

Figure XVII
Histograms of loan amounts, for loans to firms with 500 or more employees in the two data sets. The distribution of
loan amounts in the publicly available data have a substantial mass of small loans going to the largest firms.

Figure XVIII
Histograms of observed/imputed firm sizes and observed loan amounts, for firms that appear to have more than 500
employees. Appendix section A describes the rule for determining whether or not a firms appears to have more than
500 employees. “Exception” and “No Exception” refer to whether or not a firm’s NAICS code allowed for an
exception to the 500-employee cutoff rule, based on SBA guidelines for “small business interests”.

4


Appendix F: First Stage

Figure XIX
First-stage binscatter for firms with 0-499 employees.

5


TABLE A.5
Estimates: Approved Initial Claims, Firms Sized 0-99
Week Ending

Estimate

Anderson-Rubin

Wald

F-Statistics
FN

N

p-val.

95% CI

p-val.

95% CI

F KP

-0.000
0.001∗∗
-0.001
-0.000

0.584
0.047
0.193
0.674

[-0.002, 0.001]
[0.000, 0.003]
[-0.003, 0.000]
[-0.003, 0.002]

0.589
0.056
0.201
0.674

[-0.001, 0.001]
[0.000, 0.002]
[-0.002, 0.000]
[-0.002, 0.002]

14.9
16.2
15.8
14.6

91.0
98.4
92.9
85.6

2,586
2,586
2,586
2,586

-0.098∗∗∗
-0.094∗∗∗
-0.076∗∗

0.001
0.007
0.037

[-0.246, -0.046]
[-0.220, -0.036]
[-0.190, -0.007]

0.004
0.004
0.025

[-0.165, -0.031]
[-0.159, -0.030]
[-0.142, -0.010]

14.6
14.1
14.5

85.4
84.0
83.4

2,586
2,586
2,586

-0.043∗∗
-0.043∗∗
-0.031∗∗

0.041
0.012
0.027

[-0.100, -0.003]
[-0.093, -0.014]
[-0.078, -0.005]

0.019
0.003
0.020

[-0.079, -0.007]
[-0.071, -0.015]
[-0.057, -0.005]

14.1
14.0
14.2

81.6
81.9
82.5

2,586
2,586
2,586

-0.032∗∗
-0.023∗∗
-0.015∗∗
-0.013∗
-0.008∗∗

0.014
0.046
0.041
0.052
0.049

[-0.083, -0.008]
[-0.064, -0.001]
[-0.038, -0.001]
[-0.035, 0.000]
[-0.023, 0.000]

0.015
0.043
0.032
0.048
0.053

[-0.057, -0.006]
[-0.046, -0.001]
[-0.028, -0.001]
[-0.026, 0.000]
[-0.016, 0.000]

13.6
14.4
13.9
14.1
14.7

82.2
84.4
85.2
83.8
88.2

2,586
2,586
2,586
2,586
2,586

-0.006
-0.006∗
-0.005∗
-0.005∗
-0.002
-0.001
-0.002
-0.001
-0.001
-0.001
-0.000
-0.000
0.000
-0.000
-0.001∗
-0.001∗
-0.001
-0.001∗∗
-0.001
-0.000
-0.000
-0.000
-0.001

0.139
0.071
0.087
0.064
0.249
0.321
0.208
0.297
0.391
0.220
0.693
0.723
0.901
0.586
0.078
0.097
0.280
0.047
0.162
0.575
0.570
0.627
0.173

[-0.021, 0.002]
[-0.021, 0.001]
[-0.018, 0.001]
[-0.014, 0.000]
[-0.005, 0.004]
[-0.005, 0.002]
[-0.005, 0.002]
[-0.004, 0.002]
[-0.002, 0.001]
[-0.003, 0.001]
[-0.001, 0.001]
[-0.002, 0.002]
[-0.001, 0.002]
[-0.002, 0.001]
[-0.004, 0.000]
[-0.004, 0.000]
[-0.003, 0.000]
[-0.003, 0.000]
[-0.004, 0.000]
[-0.003, 0.001]
[-0.003, 0.001]
[-0.002, 0.002]
[-0.002, 0.000]

0.155
0.097
0.114
0.058
0.120
0.294
0.151
0.246
0.381
0.203
0.685
0.722
0.902
0.589
0.110
0.139
0.342
0.064
0.207
0.598
0.588
0.623
0.177

[-0.014, 0.002]
[-0.013, 0.001]
[-0.011, 0.001]
[-0.010, 0.000]
[-0.005, 0.001]
[-0.004, 0.001]
[-0.004, 0.001]
[-0.003, 0.001]
[-0.002, 0.001]
[-0.002, 0.000]
[-0.001, 0.001]
[-0.002, 0.001]
[-0.001, 0.001]
[-0.002, 0.001]
[-0.003, 0.000]
[-0.002, 0.000]
[-0.002, 0.001]
[-0.002, 0.000]
[-0.003, 0.001]
[-0.002, 0.001]
[-0.002, 0.001]
[-0.002, 0.001]
[-0.002, 0.000]

15.2
14.4
14.5
15.2
15.2
17.1
17.2
18.8
19.7
18.1
17.9
17.3
15.2
15.1
14.6
14.3
14.7
16.1
15.6
15.5
15.6
15.8
16.0

86.6
86.5
86.6
91.9
90.4
96.4
96.1
101.0
96.2
94.7
96.4
90.8
87.0
89.0
88.0
87.4
87.6
78.9
73.7
70.1
71.5
72.7
72.4

2,586
2,586
2,586
2,586
2,586
2,586
2,586
2,586
2,586
2,586
2,586
2,586
2,586
2,586
2,586
2,586
2,586
2,586
2,586
2,586
2,586
2,586
2,586

Pre-Pandemic
February 22
February 29
March 7
March 14
Onset of COVID
March 21
March 28
April 4
PPP: 1st Tranche
April 11
April 18
April 25
PPP: 2nd Tranche
May 2
May 9
May 16
May 23
May 30
PPP Rolled-Out
June 6
June 13
June 20
June 27
July 4
July 11
July 18
July 25
August 1
August 8
August 15
August 22
August 29
September 5
September 12
September 19
September 26
October 3
October 10
October 17
October 24
October 31
November 7

Estimates include indicators of significance ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01, corresponding to the Anderson-Rubin
Confidence Sets. Standard errors clustered at the state-level. Coefficients in each week are estimated separately.

6


TABLE A.6
Estimates: Approved Continuing Claims, Firms Sized 0-99
Week Ending

Estimate

Anderson-Rubin

Wald

F-Statistics
FN

N

p-val.

95% CI

p-val.

95% CI

F KP

-0.001
-0.002
-0.004
-0.007

0.354
0.388
0.203
0.128

[-0.007, 0.002]
[-0.010, 0.003]
[-0.015, 0.003]
[-0.024, 0.003]

0.369
0.391
0.198
0.120

[-0.004, 0.002]
[-0.006, 0.002]
[-0.010, 0.002]
[-0.016, 0.002]

12.2
13.1
12.8
12.1

70.7
76.2
71.4
65.5

2,337
2,337
2,337
2,337

-0.012
-0.090∗∗∗
-0.174∗∗∗

0.137
0.002
0.006

[-0.041, 0.005]
[-0.253, -0.040]
[-0.453, -0.071]

0.134
0.005
0.003

[-0.028, 0.004]
[-0.153, -0.028]
[-0.290, -0.058]

11.7
11.3
11.5

63.1
61.9
61.2

2,337
2,337
2,337

-0.233∗∗
-0.210∗∗∗
-0.207∗∗∗

0.013
0.008
0.006

[-0.622, -0.072]
[-0.491, -0.087]
[-0.473, -0.093]

0.008
0.001
0.001

[-0.404, -0.061]
[-0.338, -0.082]
[-0.327, -0.087]

11.3
11.4
11.5

60.4
60.3
61.0

2,337
2,337
2,337

-0.201∗∗∗
-0.185∗∗∗
-0.164∗∗∗
-0.140∗∗
-0.092∗∗

0.004
0.005
0.007
0.011
0.039

[-0.459, -0.099]
[-0.410, -0.092]
[-0.354, -0.076]
[-0.305, -0.056]
[-0.236, -0.009]

0.000
0.000
0.000
0.001
0.017

[-0.313, -0.089]
[-0.284, -0.086]
[-0.253, -0.074]
[-0.222, -0.058]
[-0.167, -0.017]

11.2
11.0
11.3
11.4
11.8

61.3
61.8
63.3
63.0
66.0

2,337
2,337
2,337
2,337
2,337

-0.054
-0.012
-0.005
-0.004
-0.020
-0.010
-0.007
-0.007
-0.001
-0.003
-0.004
0.013
0.004
0.009
0.008
0.006
0.007
0.003
-0.001
-0.008
-0.007
-0.007
-0.007

0.158
0.744
0.891
0.921
0.457
0.715
0.792
0.783
0.981
0.912
0.882
0.581
0.834
0.644
0.682
0.700
0.631
0.790
0.928
0.137
0.157
0.156
0.114

[-0.174, 0.032]
[-0.116, 0.090]
[-0.118, 0.108]
[-0.107, 0.101]
[-0.078, 0.063]
[-0.077, 0.075]
[-0.071, 0.077]
[-0.061, 0.068]
[-0.096, 0.085]
[-0.085, 0.076]
[-0.078, 0.065]
[-0.042, 0.093]
[-0.048, 0.077]
[-0.041, 0.079]
[-0.039, 0.069]
[-0.035, 0.062]
[-0.031, 0.060]
[-0.026, 0.036]
[-0.021, 0.023]
[-0.022, 0.004]
[-0.020, 0.004]
[-0.020, 0.004]
[-0.019, 0.003]

0.132
0.743
0.891
0.921
0.418
0.708
0.788
0.779
0.981
0.912
0.882
0.587
0.835
0.648
0.685
0.702
0.634
0.790
0.928
0.094
0.122
0.121
0.080

[-0.124, 0.016]
[-0.081, 0.058]
[-0.080, 0.070]
[-0.074, 0.067]
[-0.067, 0.028]
[-0.063, 0.043]
[-0.058, 0.044]
[-0.052, 0.039]
[-0.065, 0.064]
[-0.060, 0.054]
[-0.054, 0.046]
[-0.033, 0.059]
[-0.037, 0.046]
[-0.031, 0.049]
[-0.029, 0.044]
[-0.026, 0.038]
[-0.023, 0.038]
[-0.020, 0.026]
[-0.016, 0.015]
[-0.017, 0.001]
[-0.016, 0.002]
[-0.015, 0.002]
[-0.014, 0.001]

12.4
11.7
11.7
12.3
12.1
13.2
13.0
14.1
14.8
13.9
14.0
13.9
12.5
12.6
12.8
12.2
12.5
14.1
13.0
13.2
13.2
13.0
12.8

65.8
65.7
66.1
70.5
68.2
72.5
71.7
76.0
72.5
71.1
73.7
69.6
66.8
68.9
69.7
68.7
68.8
60.7
55.9
53.0
54.1
54.7
54.1

2,337
2,337
2,336
2,336
2,336
2,336
2,336
2,335
2,335
2,335
2,335
2,335
2,335
2,335
2,335
2,335
2,335
2,335
2,335
2,335
2,332
2,332
2,332

Pre-Pandemic
February 22
February 29
March 7
March 14
Onset of COVID
March 21
March 28
April 4
PPP: 1st Tranche
April 11
April 18
April 25
PPP: 2nd Tranche
May 2
May 9
May 16
May 23
May 30
PPP Rolled-Out
June 6
June 13
June 20
June 27
July 4
July 11
July 18
July 25
August 1
August 8
August 15
August 22
August 29
September 5
September 12
September 19
September 26
October 3
October 10
October 17
October 24
October 31
November 7

Estimates include indicators of significance ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01, corresponding to the Anderson-Rubin
Confidence Sets. Standard errors clustered at the state-level. Coefficients in each week are estimated separately.

7


TABLE A.7
Estimates: Approved Initial Claims, Firms Sized 0-499
Week Ending

Estimate

Anderson-Rubin

Wald

F-Statistics
FN

N

p-val.

95% CI

p-val.

95% CI

F KP

-0.000
0.001
-0.000
-0.001

0.850
0.226
0.585
0.583

[-0.002, 0.002]
[-0.001, 0.004]
[-0.004, 0.001]
[-0.008, 0.003]

0.851
0.245
0.606
0.592

[-0.001, 0.001]
[-0.001, 0.002]
[-0.002, 0.001]
[-0.004, 0.002]

7.4
8.5
8.0
7.2

39.2
44.8
41.7
37.3

2,579
2,579
2,579
2,579

-0.114∗∗∗
-0.102∗∗
-0.070∗

0.001
0.023
0.090

[-0.470, -0.045]
[-0.372, -0.020]
[-0.255, 0.016]

0.024
0.031
0.084

[-0.213, -0.015]
[-0.195, -0.009]
[-0.150, 0.009]

8.1
8.2
8.6

40.5
40.3
41.1

2,579
2,579
2,579

-0.029
-0.040∗∗
-0.027∗

0.164
0.034
0.081

[-0.108, 0.019]
[-0.128, -0.005]
[-0.095, 0.005]

0.144
0.023
0.074

[-0.067, 0.010]
[-0.075, -0.005]
[-0.056, 0.003]

8.5
8.3
8.4

40.3
40.4
40.7

2,579
2,579
2,579

-0.028∗∗
-0.021∗∗
-0.012∗
-0.011∗
-0.004

0.018
0.047
0.052
0.066
0.275

[-0.106, -0.006]
[-0.082, 0.000]
[-0.043, 0.000]
[-0.045, 0.001]
[-0.023, 0.005]

0.026
0.055
0.051
0.078
0.283

[-0.053, -0.003]
[-0.042, 0.000]
[-0.024, 0.000]
[-0.024, 0.001]
[-0.013, 0.004]

7.8
8.0
8.0
7.9
7.8

39.8
40.0
41.2
41.0
41.1

2,579
2,579
2,579
2,579
2,579

-0.003
-0.004
-0.002
-0.003
0.000
0.001
-0.001
-0.000
-0.001
-0.001
0.000
0.000
0.000
0.000
-0.001
-0.001
-0.000
-0.001
-0.001
-0.000
-0.000
-0.000
-0.001

0.353
0.208
0.339
0.149
0.949
0.777
0.574
0.957
0.559
0.460
0.686
0.752
0.522
0.956
0.215
0.219
0.611
0.145
0.373
0.693
0.893
0.688
0.277

[-0.021, 0.006]
[-0.024, 0.002]
[-0.019, 0.003]
[-0.015, 0.002]
[-0.004, 0.015]
[-0.003, 0.008]
[-0.005, 0.005]
[-0.002, 0.004]
[-0.003, 0.002]
[-0.003, 0.001]
[-0.001, 0.002]
[-0.002, 0.003]
[-0.001, 0.003]
[-0.003, 0.002]
[-0.006, 0.000]
[-0.006, 0.001]
[-0.004, 0.001]
[-0.004, 0.000]
[-0.007, 0.001]
[-0.006, 0.002]
[-0.003, 0.002]
[-0.002, 0.002]
[-0.003, 0.001]

0.364
0.258
0.373
0.145
0.950
0.783
0.550
0.957
0.553
0.458
0.694
0.751
0.535
0.956
0.289
0.276
0.640
0.184
0.422
0.709
0.894
0.682
0.294

[-0.011, 0.004]
[-0.010, 0.003]
[-0.008, 0.003]
[-0.008, 0.001]
[-0.005, 0.005]
[-0.003, 0.004]
[-0.004, 0.002]
[-0.002, 0.002]
[-0.002, 0.001]
[-0.002, 0.001]
[-0.001, 0.001]
[-0.001, 0.002]
[-0.001, 0.002]
[-0.001, 0.001]
[-0.002, 0.001]
[-0.002, 0.001]
[-0.001, 0.001]
[-0.002, 0.000]
[-0.003, 0.001]
[-0.002, 0.002]
[-0.001, 0.001]
[-0.001, 0.001]
[-0.002, 0.001]

7.8
7.5
7.2
7.8
9.4
9.7
10.0
11.5
10.0
10.3
10.1
9.0
8.2
7.4
7.3
7.2
7.5
7.7
7.1
6.6
7.0
6.9
6.8

39.2
38.4
36.9
39.9
44.1
44.5
44.4
47.6
42.1
43.7
43.0
39.6
38.8
38.4
38.1
37.7
37.7
32.7
29.6
27.2
28.6
28.9
27.7

2,579
2,579
2,579
2,579
2,579
2,579
2,579
2,579
2,579
2,579
2,579
2,579
2,579
2,579
2,579
2,579
2,579
2,579
2,579
2,579
2,579
2,579
2,579

Pre-Pandemic
February 22
February 29
March 7
March 14
Onset of COVID
March 21
March 28
April 4
PPP: 1st Tranche
April 11
April 18
April 25
PPP: 2nd Tranche
May 2
May 9
May 16
May 23
May 30
PPP Rolled-Out
June 6
June 13
June 20
June 27
July 4
July 11
July 18
July 25
August 1
August 8
August 15
August 22
August 29
September 5
September 12
September 19
September 26
October 3
October 10
October 17
October 24
October 31
November 7

Estimates include indicators of significance ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01, corresponding to the Anderson-Rubin
Confidence Sets. Standard errors clustered at the state-level. Coefficients in each week are estimated separately.

8


TABLE A.8
Estimates: Approved Continuing Claims, Firms Sized 0-499
Week Ending

Estimate

Anderson-Rubin

Wald

F-Statistics
FN

N

p-val.

95% CI

p-val.

95% CI

F KP

-0.001
-0.001
-0.003
-0.006

0.793
0.826
0.414
0.355

[-0.008, 0.011]
[-0.009, 0.015]
[-0.019, 0.012]
[-0.039, 0.017]

0.789
0.822
0.380
0.326

[-0.004, 0.003]
[-0.007, 0.005]
[-0.011, 0.004]
[-0.017, 0.006]

5.9
6.8
6.4
5.7

27.8
32.3
29.5
26.0

2,294
2,294
2,294
2,294

-0.013
-0.106∗∗∗
-0.190∗∗

0.223
0.008
0.020

[-0.099, 0.012]
[-0.806, -0.032]
[-1.297, -0.041]

0.246
0.045
0.042

[-0.034, 0.009]
[-0.210, -0.002]
[-0.374, -0.007]

6.0
6.0
6.0

27.0
27.0
27.2

2,294
2,294
2,294

-0.241∗
-0.217∗∗
-0.213∗∗

0.050
0.038
0.028

[-1.538, -0.001]
[-1.390, -0.024]
[-1.270, -0.035]

0.072
0.053
0.039

[-0.503, 0.022]
[-0.437, 0.003]
[-0.415, -0.011]

6.1
6.0
6.1

27.6
27.0
27.4

2,294
2,294
2,294

-0.203∗∗
-0.187∗∗
-0.159∗∗
-0.134∗∗
-0.085∗

0.020
0.015
0.015
0.021
0.073

[-1.255, -0.048]
[-1.199, -0.062]
[-0.731, -0.055]
[-0.575, -0.039]
[-0.399, 0.014]

0.029
0.017
0.008
0.009
0.044

[-0.386, -0.020]
[-0.340, -0.034]
[-0.277, -0.041]
[-0.234, -0.034]
[-0.168, -0.002]

5.9
5.6
6.0
6.2
6.1

27.3
26.9
28.5
29.1
29.2

2,294
2,294
2,294
2,293
2,293

-0.045
0.000
0.007
0.008
-0.012
-0.000
0.003
0.002
0.010
0.004
0.002
0.019
0.011
0.018
0.016
0.014
0.014
0.009
0.003
-0.007
-0.006
-0.006
-0.007

0.283
0.999
0.868
0.839
0.656
0.998
0.912
0.929
0.801
0.910
0.946
0.483
0.643
0.491
0.496
0.496
0.462
0.532
0.738
0.322
0.328
0.296
0.204

[-0.228, 0.091]
[-0.117, 0.262]
[-0.123, 0.380]
[-0.107, 0.301]
[-0.082, 0.121]
[-0.072, 0.170]
[-0.066, 0.182]
[-0.057, 0.130]
[-0.095, 0.188]
[-0.085, 0.142]
[-0.083, 0.125]
[-0.041, 0.222]
[-0.048, 0.205]
[-0.041, 0.283]
[-0.039, 0.218]
[-0.034, 0.223]
[-0.030, 0.181]
[-0.024, 0.105]
[-0.020, 0.084]
[-0.041, 0.016]
[-0.038, 0.013]
[-0.041, 0.012]
[-0.048, 0.007]

0.242
0.999
0.870
0.842
0.637
0.998
0.914
0.930
0.804
0.910
0.946
0.515
0.659
0.525
0.526
0.527
0.495
0.557
0.748
0.295
0.308
0.273
0.194

[-0.120, 0.030]
[-0.078, 0.078]
[-0.082, 0.097]
[-0.074, 0.091]
[-0.062, 0.038]
[-0.059, 0.059]
[-0.055, 0.061]
[-0.049, 0.053]
[-0.067, 0.087]
[-0.060, 0.067]
[-0.055, 0.059]
[-0.038, 0.076]
[-0.039, 0.062]
[-0.037, 0.072]
[-0.034, 0.066]
[-0.029, 0.058]
[-0.027, 0.056]
[-0.022, 0.041]
[-0.018, 0.024]
[-0.019, 0.006]
[-0.018, 0.006]
[-0.017, 0.005]
[-0.018, 0.004]

6.3
6.1
5.9
6.3
7.0
7.2
7.1
8.2
7.5
7.8
7.7
7.0
6.5
6.0
6.3
6.0
6.3
6.4
5.7
5.2
5.4
5.3
5.2

28.1
27.7
26.1
28.6
31.2
31.3
30.5
33.4
29.4
30.4
30.3
27.6
27.0
27.2
27.9
27.2
27.7
22.6
20.0
18.1
18.9
19.1
18.4

2,293
2,293
2,293
2,293
2,293
2,293
2,293
2,293
2,293
2,293
2,293
2,293
2,293
2,293
2,293
2,293
2,293
2,293
2,293
2,293
2,290
2,290
2,290

Pre-Pandemic
February 22
February 29
March 7
March 14
Onset of COVID
March 21
March 28
April 4
PPP: 1st Tranche
April 11
April 18
April 25
PPP: 2nd Tranche
May 2
May 9
May 16
May 23
May 30
PPP Rolled-Out
June 6
June 13
June 20
June 27
July 4
July 11
July 18
July 25
August 1
August 8
August 15
August 22
August 29
September 5
September 12
September 19
September 26
October 3
October 10
October 17
October 24
October 31
November 7

Estimates include indicators of significance ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01, corresponding to the Anderson-Rubin
Confidence Sets. Standard errors clustered at the state-level. Coefficients in each week are estimated separately.

9


TABLE A.9
Estimates: Approved Initial Claims, Firms Sized 0-99 (Includes control for average %(change) in small
business revenue, weeks ending March 21 - April 4)
Week Ending

Estimate

Anderson-Rubin

Wald

F-Statistics
FN

N

p-val.

95% CI

p-val.

95% CI

F KP

-0.000
0.001
-0.001
-0.000

0.472
0.146
0.126
0.733

[-0.002, 0.001]
[0.000, 0.002]
[-0.003, 0.000]
[-0.003, 0.002]

0.487
0.164
0.137
0.732

[-0.001, 0.001]
[0.000, 0.001]
[-0.002, 0.000]
[-0.002, 0.002]

18.6
21.0
20.2
18.6

73.9
81.3
75.7
69.3

1,585
1,585
1,584
1,584

-0.081∗∗∗
-0.058∗∗
-0.045∗

0.001
0.019
0.066

[-0.186, -0.040]
[-0.117, -0.015]
[-0.103, 0.004]

0.002
0.003
0.030

[-0.132, -0.031]
[-0.097, -0.019]
[-0.086, -0.004]

18.4
18.0
17.9

67.5
66.2
64.9

1,584
1,579
1,580

-0.025∗
-0.028∗∗
-0.020∗∗

0.093
0.014
0.026

[-0.053, 0.007]
[-0.049, -0.010]
[-0.044, -0.004]

0.033
0.000
0.010

[-0.047, -0.002]
[-0.043, -0.013]
[-0.036, -0.005]

17.9
17.8
17.8

64.5
64.8
64.7

1,578
1,579
1,583

-0.026∗∗∗
-0.018∗∗
-0.012∗∗
-0.010∗∗
-0.006∗∗

0.008
0.021
0.018
0.029
0.044

[-0.061, -0.009]
[-0.043, -0.004]
[-0.027, -0.003]
[-0.024, -0.001]
[-0.016, 0.000]

0.006
0.017
0.011
0.027
0.042

[-0.044, -0.007]
[-0.032, -0.003]
[-0.020, -0.003]
[-0.018, -0.001]
[-0.011, 0.000]

16.7
17.3
17.3
17.8
18.4

64.7
65.5
64.8
66.6
70.6

1,584
1,583
1,584
1,583
1,566

-0.004
-0.006∗
-0.005∗
-0.004∗
-0.002
-0.000
-0.001
-0.001
-0.000
-0.001
0.000
0.000
0.000
0.000
-0.001
-0.001
-0.000
-0.001
-0.001
-0.000
-0.000
-0.000
-0.000

0.192
0.079
0.098
0.084
0.412
0.727
0.226
0.449
0.623
0.280
0.880
0.952
0.719
0.837
0.154
0.200
0.512
0.127
0.364
0.744
0.851
0.699
0.434

[-0.016, 0.003]
[-0.019, 0.001]
[-0.017, 0.001]
[-0.013, 0.001]
[-0.004, 0.005]
[-0.003, 0.003]
[-0.004, 0.001]
[-0.002, 0.002]
[-0.001, 0.001]
[-0.002, 0.001]
[-0.001, 0.002]
[-0.002, 0.002]
[-0.001, 0.002]
[-0.001, 0.002]
[-0.003, 0.000]
[-0.003, 0.000]
[-0.002, 0.001]
[-0.002, 0.000]
[-0.003, 0.001]
[-0.002, 0.001]
[-0.003, 0.002]
[-0.002, 0.002]
[-0.002, 0.001]

0.201
0.106
0.121
0.078
0.315
0.719
0.173
0.404
0.613
0.257
0.881
0.952
0.730
0.837
0.170
0.223
0.540
0.143
0.379
0.750
0.853
0.697
0.433

[-0.011, 0.002]
[-0.012, 0.001]
[-0.011, 0.001]
[-0.009, 0.001]
[-0.005, 0.002]
[-0.003, 0.002]
[-0.004, 0.001]
[-0.002, 0.001]
[-0.001, 0.001]
[-0.002, 0.000]
[-0.001, 0.001]
[-0.001, 0.001]
[-0.001, 0.001]
[-0.001, 0.001]
[-0.002, 0.000]
[-0.002, 0.000]
[-0.001, 0.001]
[-0.002, 0.000]
[-0.002, 0.001]
[-0.002, 0.001]
[-0.002, 0.001]
[-0.002, 0.001]
[-0.002, 0.001]

18.9
17.9
18.5
19.4
19.0
21.7
21.1
23.4
24.8
23.2
21.8
22.3
19.5
20.0
19.2
18.2
18.7
20.5
19.3
19.8
19.7
19.1
19.6

69.8
69.8
70.1
73.9
68.5
76.6
75.7
79.8
77.5
76.3
78.1
73.7
69.9
73.2
72.8
70.2
70.4
63.2
58.4
56.2
57.8
57.4
56.2

1,583
1,582
1,584
1,585
1,556
1,583
1,585
1,585
1,583
1,583
1,584
1,585
1,574
1,582
1,547
1,583
1,582
1,582
1,578
1,582
1,582
1,574
1,581

Pre-Pandemic
February 22
February 29
March 7
March 14
Onset of COVID
March 21
March 28
April 4
PPP: 1st Tranche
April 11
April 18
April 25
PPP: 2nd Tranche
May 2
May 9
May 16
May 23
May 30
PPP Rolled-Out
June 6
June 13
June 20
June 27
July 4
July 11
July 18
July 25
August 1
August 8
August 15
August 22
August 29
September 5
September 12
September 19
September 26
October 3
October 10
October 17
October 24
October 31
November 7

Estimates include indicators of significance ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01, corresponding to the Anderson-Rubin
Confidence Sets. Standard errors clustered at the state-level. Coefficients in each week are estimated separately.

10


TABLE A.10
Estimates: Approved Continuing Claims, Firms Sized 0-99 (Includes control for average %(change) in small
business revenue, weeks ending March 21 - April 4)
Week Ending

Estimate

Anderson-Rubin

Wald

F-Statistics
FN

N

p-val.

95% CI

p-val.

95% CI

F KP

-0.001
-0.001
-0.002
-0.004

0.673
0.704
0.402
0.266

[-0.005, 0.002]
[-0.008, 0.004]
[-0.012, 0.004]
[-0.017, 0.005]

0.681
0.708
0.402
0.257

[-0.003, 0.002]
[-0.005, 0.003]
[-0.008, 0.003]
[-0.012, 0.003]

13.6
15.2
14.7
13.7

57.1
62.6
57.9
52.8

1,493
1,493
1,492
1,492

-0.009
-0.076∗∗∗
-0.132∗∗

0.228
0.005
0.016

[-0.031, 0.009]
[-0.187, -0.032]
[-0.303, -0.040]

0.207
0.003
0.003

[-0.023, 0.005]
[-0.126, -0.026]
[-0.219, -0.044]

13.2
12.8
12.6

49.5
48.2
47.1

1,492
1,488
1,488

-0.170∗∗
-0.133∗∗
-0.134∗∗

0.034
0.027
0.020

[-0.393, -0.022]
[-0.280, -0.026]
[-0.277, -0.038]

0.009
0.003
0.002

[-0.298, -0.042]
[-0.222, -0.044]
[-0.217, -0.051]

12.8
12.9
12.9

47.3
47.4
47.3

1,486
1,488
1,491

-0.135∗∗
-0.127∗∗
-0.115∗∗
-0.099∗∗
-0.060∗

0.012
0.012
0.016
0.020
0.070

[-0.284, -0.051]
[-0.264, -0.049]
[-0.230, -0.038]
[-0.208, -0.027]
[-0.166, 0.007]

0.001
0.001
0.001
0.002
0.045

[-0.215, -0.056]
[-0.199, -0.054]
[-0.183, -0.047]
[-0.161, -0.036]
[-0.119, -0.001]

12.7
12.2
13.0
13.0
13.3

48.2
47.9
48.1
49.6
52.5

1,492
1,491
1,492
1,491
1,475

-0.028
0.006
0.010
0.010
-0.006
0.002
0.004
0.003
0.006
0.003
0.001
0.019
0.010
0.013
0.012
0.010
0.010
0.006
0.001
-0.006
-0.005
-0.005
-0.005

0.325
0.835
0.756
0.741
0.800
0.946
0.855
0.900
0.834
0.907
0.962
0.363
0.593
0.449
0.480
0.508
0.471
0.595
0.878
0.245
0.275
0.258
0.213

[-0.120, 0.042]
[-0.079, 0.091]
[-0.084, 0.102]
[-0.077, 0.094]
[-0.058, 0.071]
[-0.058, 0.074]
[-0.054, 0.079]
[-0.047, 0.068]
[-0.087, 0.076]
[-0.081, 0.068]
[-0.075, 0.060]
[-0.031, 0.084]
[-0.039, 0.068]
[-0.032, 0.065]
[-0.031, 0.059]
[-0.029, 0.053]
[-0.026, 0.051]
[-0.022, 0.032]
[-0.019, 0.022]
[-0.019, 0.006]
[-0.018, 0.006]
[-0.019, 0.006]
[-0.018, 0.004]

0.314
0.835
0.754
0.739
0.795
0.946
0.856
0.901
0.832
0.906
0.962
0.357
0.592
0.437
0.470
0.501
0.462
0.587
0.878
0.214
0.258
0.239
0.194

[-0.084, 0.027]
[-0.051, 0.063]
[-0.052, 0.072]
[-0.048, 0.068]
[-0.051, 0.039]
[-0.045, 0.049]
[-0.042, 0.051]
[-0.039, 0.044]
[-0.052, 0.064]
[-0.050, 0.056]
[-0.045, 0.048]
[-0.021, 0.059]
[-0.027, 0.047]
[-0.020, 0.047]
[-0.020, 0.043]
[-0.018, 0.037]
[-0.017, 0.037]
[-0.014, 0.025]
[-0.013, 0.016]
[-0.015, 0.003]
[-0.014, 0.004]
[-0.014, 0.003]
[-0.013, 0.003]

13.7
12.8
13.2
14.1
13.8
15.0
14.4
15.7
16.5
15.8
15.2
15.9
14.3
15.0
15.0
14.1
14.4
15.9
14.2
14.9
14.8
14.1
14.1

52.7
52.2
52.8
56.2
51.8
57.2
55.9
59.5
57.8
56.7
59.1
56.3
53.3
56.8
57.2
55.2
55.3
48.5
44.1
42.1
43.7
43.5
42.3

1,491
1,491
1,492
1,493
1,468
1,491
1,493
1,493
1,491
1,491
1,492
1,493
1,482
1,490
1,459
1,491
1,491
1,491
1,486
1,490
1,490
1,484
1,491

Pre-Pandemic
February 22
February 29
March 7
March 14
Onset of COVID
March 21
March 28
April 4
PPP: 1st Tranche
April 11
April 18
April 25
PPP: 2nd Tranche
May 2
May 9
May 16
May 23
May 30
PPP Rolled-Out
June 6
June 13
June 20
June 27
July 4
July 11
July 18
July 25
August 1
August 8
August 15
August 22
August 29
September 5
September 12
September 19
September 26
October 3
October 10
October 17
October 24
October 31
November 7

Estimates include indicators of significance ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01, corresponding to the Anderson-Rubin
Confidence Sets. Standard errors clustered at the state-level. Coefficients in each week are estimated separately.

11


TABLE A.11
Estimates: Approved Initial Claims, Firms Sized 0-499 (Includes control for average %(change) in small
business revenue, weeks ending March 21 - April 4)
Week Ending

Estimate

Anderson-Rubin

Wald

F-Statistics
FN

N

p-val.

95% CI

p-val.

95% CI

F KP

0.000
0.000
-0.000
-0.001

0.984
0.461
0.677
0.577

[-0.002, 0.002]
[-0.001, 0.002]
[-0.003, 0.001]
[-0.006, 0.003]

0.984
0.474
0.692
0.577

[-0.001, 0.001]
[-0.001, 0.001]
[-0.002, 0.001]
[-0.004, 0.002]

10.6
12.6
11.8
10.3

33.8
39.3
36.2
32.0

1,585
1,585
1,584
1,584

-0.091∗∗∗
-0.057∗
-0.033

0.001
0.095
0.223

[-0.282, -0.038]
[-0.165, 0.016]
[-0.108, 0.033]

0.011
0.068
0.187

[-0.161, -0.021]
[-0.118, 0.004]
[-0.081, 0.016]

11.0
11.2
11.6

33.8
33.4
33.9

1,584
1,579
1,580

-0.007
-0.023∗
-0.013

0.633
0.081
0.164

[-0.040, 0.042]
[-0.054, 0.005]
[-0.038, 0.009]

0.616
0.032
0.119

[-0.036, 0.021]
[-0.043, -0.002]
[-0.029, 0.003]

11.6
11.3
11.4

33.4
33.8
33.4

1,578
1,579
1,583

-0.022∗∗∗
-0.016∗∗
-0.008∗∗
-0.008∗∗
-0.003

0.009
0.016
0.017
0.027
0.336

[-0.065, -0.007]
[-0.049, -0.004]
[-0.022, -0.002]
[-0.026, -0.001]
[-0.012, 0.004]

0.010
0.021
0.019
0.041
0.338

[-0.039, -0.005]
[-0.029, -0.002]
[-0.014, -0.001]
[-0.016, 0.000]
[-0.008, 0.003]

10.4
10.6
10.7
11.0
11.2

32.8
33.0
32.5
34.4
35.0

1,584
1,583
1,584
1,583
1,566

-0.002
-0.003
-0.003
-0.003
0.001
0.001
-0.001
0.000
-0.000
-0.000
0.000
0.000
0.000
0.000
-0.001
-0.000
-0.000
-0.001
-0.001
-0.000
-0.000
-0.000
-0.000

0.469
0.230
0.266
0.203
0.795
0.642
0.488
0.664
0.545
0.430
0.526
0.522
0.435
0.434
0.279
0.424
0.811
0.353
0.516
0.709
0.973
0.620
0.397

[-0.012, 0.005]
[-0.016, 0.002]
[-0.015, 0.003]
[-0.011, 0.002]
[-0.003, 0.013]
[-0.002, 0.007]
[-0.004, 0.003]
[-0.001, 0.004]
[-0.002, 0.001]
[-0.002, 0.001]
[-0.001, 0.002]
[-0.001, 0.003]
[-0.001, 0.003]
[-0.001, 0.002]
[-0.003, 0.001]
[-0.003, 0.001]
[-0.002, 0.001]
[-0.002, 0.001]
[-0.004, 0.001]
[-0.004, 0.001]
[-0.002, 0.002]
[-0.002, 0.001]
[-0.002, 0.001]

0.471
0.275
0.302
0.194
0.805
0.658
0.455
0.684
0.531
0.427
0.542
0.520
0.455
0.426
0.318
0.441
0.816
0.369
0.533
0.721
0.973
0.614
0.400

[-0.008, 0.004]
[-0.009, 0.002]
[-0.008, 0.003]
[-0.007, 0.001]
[-0.004, 0.005]
[-0.002, 0.004]
[-0.003, 0.001]
[-0.001, 0.002]
[-0.002, 0.001]
[-0.002, 0.001]
[-0.001, 0.001]
[-0.001, 0.002]
[-0.001, 0.002]
[-0.001, 0.002]
[-0.002, 0.001]
[-0.002, 0.001]
[-0.001, 0.001]
[-0.002, 0.001]
[-0.003, 0.001]
[-0.002, 0.002]
[-0.001, 0.001]
[-0.002, 0.001]
[-0.001, 0.001]

11.0
10.6
10.6
11.5
12.8
13.7
13.7
15.9
15.0
15.5
14.8
13.6
12.3
11.8
11.5
11.0
11.5
11.6
10.7
10.0
10.5
9.8
10.0

34.0
32.7
31.7
34.0
35.9
37.4
36.6
39.4
35.9
37.3
37.4
34.4
33.1
34.6
35.0
32.9
33.4
28.2
25.4
23.6
24.8
24.3
23.0

1,583
1,582
1,584
1,585
1,556
1,583
1,585
1,585
1,583
1,583
1,584
1,585
1,574
1,582
1,547
1,583
1,582
1,582
1,578
1,582
1,582
1,574
1,581

Pre-Pandemic
February 22
February 29
March 7
March 14
Onset of COVID
March 21
March 28
April 4
PPP: 1st Tranche
April 11
April 18
April 25
PPP: 2nd Tranche
May 2
May 9
May 16
May 23
May 30
PPP Rolled-Out
June 6
June 13
June 20
June 27
July 4
July 11
July 18
July 25
August 1
August 8
August 15
August 22
August 29
September 5
September 12
September 19
September 26
October 3
October 10
October 17
October 24
October 31
November 7

Estimates include indicators of significance ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01, corresponding to the Anderson-Rubin
Confidence Sets. Standard errors clustered at the state-level. Coefficients in each week are estimated separately.

12


TABLE A.12
Estimates: Approved Continuing Claims, Firms Sized 0-499 (Includes control for average %(change) in
small business revenue, weeks ending March 21 - April 4)
Week Ending

Estimate

Anderson-Rubin

Wald

F-Statistics
FN

N

p-val.

95% CI

p-val.

95% CI

F KP

0.000
-0.000
-0.003
-0.005

0.984
0.968
0.479
0.404

[-0.006, 0.008]
[-0.008, 0.011]
[-0.015, 0.010]
[-0.025, 0.014]

0.984
0.968
0.458
0.379

[-0.004, 0.004]
[-0.006, 0.005]
[-0.010, 0.005]
[-0.015, 0.006]

7.5
8.9
8.4
7.3

23.9
28.2
25.6
22.2

1,492
1,492
1,491
1,491

-0.012
-0.090∗∗
-0.141∗∗

0.238
0.012
0.049

[-0.061, 0.012]
[-0.422, -0.027]
[-0.623, -0.005]

0.241
0.029
0.047

[-0.031, 0.008]
[-0.171, -0.009]
[-0.280, -0.002]

7.2
7.2
7.0

22.4
22.2
22.2

1,491
1,488
1,488

-0.170
-0.131
-0.131∗

0.114
0.125
0.093

[-0.722, 0.068]
[-0.590, 0.055]
[-0.561, 0.033]

0.102
0.118
0.083

[-0.374, 0.034]
[-0.295, 0.033]
[-0.279, 0.017]

7.4
7.2
7.3

22.7
22.4
22.4

1,486
1,488
1,491

-0.127∗
-0.118∗∗
-0.101∗∗
-0.085∗
-0.048

0.068
0.048
0.046
0.053
0.156

[-0.541, 0.015]
[-0.494, -0.004]
[-0.327, -0.006]
[-0.264, 0.002]
[-0.186, 0.034]

0.060
0.035
0.021
0.021
0.118

[-0.260, 0.006]
[-0.227, -0.008]
[-0.188, -0.015]
[-0.157, -0.013]
[-0.109, 0.012]

7.2
6.7
7.2
7.7
7.8

22.6
22.1
22.4
24.3
24.6

1,492
1,491
1,491
1,490
1,474

-0.015
0.020
0.024
0.023
0.003
0.012
0.015
0.011
0.017
0.010
0.007
0.024
0.016
0.021
0.019
0.017
0.016
0.011
0.005
-0.005
-0.004
-0.005
-0.006

0.619
0.536
0.506
0.501
0.883
0.635
0.568
0.623
0.626
0.733
0.791
0.300
0.446
0.345
0.352
0.358
0.349
0.392
0.588
0.460
0.457
0.387
0.288

[-0.113, 0.097]
[-0.062, 0.203]
[-0.071, 0.242]
[-0.064, 0.202]
[-0.051, 0.116]
[-0.047, 0.139]
[-0.044, 0.157]
[-0.040, 0.113]
[-0.079, 0.137]
[-0.076, 0.105]
[-0.074, 0.092]
[-0.028, 0.147]
[-0.036, 0.127]
[-0.031, 0.133]
[-0.028, 0.113]
[-0.026, 0.111]
[-0.024, 0.099]
[-0.020, 0.065]
[-0.018, 0.050]
[-0.025, 0.014]
[-0.025, 0.012]
[-0.028, 0.011]
[-0.029, 0.008]

0.607
0.549
0.519
0.513
0.885
0.648
0.586
0.636
0.626
0.732
0.790
0.325
0.461
0.362
0.365
0.374
0.365
0.402
0.598
0.442
0.447
0.374
0.280

[-0.073, 0.043]
[-0.045, 0.084]
[-0.049, 0.098]
[-0.046, 0.091]
[-0.043, 0.050]
[-0.040, 0.064]
[-0.039, 0.068]
[-0.035, 0.057]
[-0.051, 0.084]
[-0.048, 0.068]
[-0.044, 0.058]
[-0.024, 0.073]
[-0.027, 0.060]
[-0.024, 0.065]
[-0.022, 0.059]
[-0.020, 0.053]
[-0.019, 0.051]
[-0.015, 0.038]
[-0.014, 0.024]
[-0.017, 0.007]
[-0.016, 0.007]
[-0.016, 0.006]
[-0.016, 0.005]

7.9
7.5
7.4
8.0
8.4
8.8
8.5
9.8
9.7
10.0
9.6
9.2
8.3
8.5
8.8
8.1
8.6
8.5
7.4
7.1
7.3
6.9
7.0

24.1
23.2
22.1
24.0
25.1
25.9
24.7
27.3
24.7
25.6
26.1
23.9
22.7
24.6
25.3
23.7
24.4
19.5
17.2
15.9
16.7
16.4
15.7

1,490
1,490
1,491
1,492
1,467
1,490
1,492
1,492
1,490
1,490
1,491
1,492
1,481
1,489
1,458
1,490
1,490
1,490
1,485
1,489
1,489
1,483
1,491

Pre-Pandemic
February 22
February 29
March 7
March 14
Onset of COVID
March 21
March 28
April 4
PPP: 1st Tranche
April 11
April 18
April 25
PPP: 2nd Tranche
May 2
May 9
May 16
May 23
May 30
PPP Rolled-Out
June 6
June 13
June 20
June 27
July 4
July 11
July 18
July 25
August 1
August 8
August 15
August 22
August 29
September 5
September 12
September 19
September 26
October 3
October 10
October 17
October 24
October 31
November 7

Estimates include indicators of significance ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01, corresponding to the Anderson-Rubin
Confidence Sets. Standard errors clustered at the state-level. Coefficients in each week are estimated separately.

13


Appendix G: Online Appendix Continued: Continuing Claims Tables

14


TABLE A.14
Continuing Claims, Firms Size 0-99 (Pooled Regression of Pre-Covid Weeks)
(1)

(2)

(3)

(4)

(5)

(6)

0.014
(0.009)

0.029∗∗
(0.011)

-0.005
(0.003)

-0.004
(0.003)

-0.003
(0.003)

-0.002
(0.003)

February IUR

0.909∗∗∗
(0.020)

0.911∗∗∗
(0.020)

0.910∗∗∗
(0.020)

0.913∗∗∗
(0.022)

Log(Med. Income)

0.001
(0.001)

-0.000
(0.001)

-0.000
(0.001)

0.000
(0.001)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

12.827
(12.026)

11.952
(11.391)

11.290
(11.516)

Covid Cases, 4w

-8.419
(9.080)

-7.957
(8.582)

-7.377
(8.679)

Covid Deaths, 1w

33.275
(129.171)

23.676
(136.110)

15.071
(148.205)

Covid Deaths, 4w

299.640∗∗∗
(75.370)

306.303∗∗∗
(79.429)

316.881∗∗∗
(85.905)

0.005
(0.003)

0.006
(0.004)

0.005
(0.003)

Early PPP Coverage

WFH Index

Industry Index

0.004
(0.006)

March Small-Firm Rev.

State-by-Week FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.001
(0.001)
No
9,369
2.2
72.9
[ -0.004, 0.037]
0.126

Yes
9,369
0.0
55.8
[ 0.011, 0.073]
0.002

Yes
9,356
19.8
8.3
[ -0.021, 0.003]
0.178

Yes
9,348
28.9
10.6
[ -0.016, 0.003]
0.230

Yes
9,348
46.0
13.0
[ -0.013, 0.003]
0.269

Yes
5,970
35.3
12.5
[ -0.011, 0.004]
0.445

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state, week, and state-by-week fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e.
the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data.
Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity
Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the
share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry
Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b)
the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights.
Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. In this regression, the weeks ending 2020-02-22 through 2020-03-14 are pooled, with the
remaining weeks excluded from the data.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

15


TABLE A.15
Continuing Claims, Firms Size 0-99 (Pooled Regression of Covid-Onset Weeks)
(1)

(2)

(3)

(4)

(5)

(6)

-0.025
(0.013)

-0.040∗
(0.016)

-0.090∗∗
(0.034)

-0.095∗∗
(0.035)

-0.053∗∗
(0.018)

-0.074∗∗
(0.025)

February IUR

1.096∗∗∗
(0.086)

1.096∗∗∗
(0.088)

1.040∗∗∗
(0.059)

1.080∗∗∗
(0.072)

Log(Med. Income)

-0.015∗
(0.008)

-0.018
(0.010)

-0.018∗∗
(0.007)

-0.017∗
(0.008)

Poverty Rate

-0.001∗
(0.000)

-0.001
(0.000)

-0.001∗∗
(0.000)

-0.001∗
(0.000)

Log(Pop. Density)

0.000
(0.001)

-0.000
(0.001)

0.000
(0.001)

-0.001
(0.001)

Covid Cases, 1w

2.970
(1.820)

1.983
(1.406)

2.853
(1.556)

Covid Cases, 4w

-2.298
(1.506)

-1.669
(0.913)

-2.267
(1.182)

Covid Deaths, 1w

-257.353∗
(123.727)

-243.874∗
(100.637)

-255.617∗
(123.984)

Covid Deaths, 4w

217.110∗∗
(81.334)

198.370∗∗
(68.908)

214.056∗∗
(83.024)

WFH Index

0.013
(0.043)

0.079∗∗
(0.027)

0.017
(0.034)

Early PPP Coverage

0.236∗∗∗
(0.037)

Industry Index

-0.019∗∗∗
(0.005)

March Small-Firm Rev.

State-by-Week FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
7,056
3.3
72.9
[ -0.052, 0.006]
0.102

Yes
7,056
0.0
55.9
[ -0.077, 0.003]
0.061

Yes
7,017
1.7
8.2
[ -0.310, -0.037]
0.004

Yes
7,011
26.9
9.8
[ -0.288, -0.039]
0.004

Yes
7,011
71.2
12.3
[ -0.121, -0.015]
0.019

Yes
4,468
39.6
11.4
[ -0.187, -0.027]
0.010

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state, week, and state-by-week fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e.
the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data.
Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity
Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the
share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry
Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b)
the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights.
Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. In this regression, the weeks ending 2020-03-21 through 2020-04-04 are pooled, with the
remaining weeks excluded from the data.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

16


TABLE A.16
Continuing Claims, Firms Size 0-99 (Pooled Regression of First Tranche Weeks)
(1)

(2)

(3)

(4)

(5)

(6)

-0.169∗∗∗
(0.039)

-0.200∗∗∗
(0.036)

-0.227∗∗
(0.076)

-0.235∗∗
(0.083)

-0.111∗
(0.043)

-0.157∗∗
(0.054)

February IUR

1.566∗∗∗
(0.210)

1.536∗∗∗
(0.229)

1.397∗∗∗
(0.156)

1.472∗∗∗
(0.189)

Log(Med. Income)

-0.045∗
(0.018)

-0.043
(0.026)

-0.043∗∗
(0.016)

-0.038
(0.021)

Poverty Rate

-0.002∗∗
(0.001)

-0.002
(0.001)

-0.002∗∗
(0.001)

-0.002∗
(0.001)

Log(Pop. Density)

0.003
(0.002)

0.003
(0.003)

0.005∗∗
(0.001)

0.002
(0.003)

Covid Cases, 1w

-6.102
(3.326)

-4.989∗
(2.457)

-5.807
(3.372)

Covid Cases, 4w

1.107
(1.389)

1.246
(1.051)

1.226
(1.301)

Covid Deaths, 1w

3.468
(7.014)

-6.957
(6.088)

-0.227
(7.746)

Covid Deaths, 4w

3.857
(7.018)

-0.161
(5.072)

1.886
(6.390)

WFH Index

-0.013
(0.102)

0.156∗∗
(0.059)

-0.007
(0.073)

Early PPP Coverage

0.616∗∗∗
(0.089)

Industry Index

-0.057∗∗∗
(0.014)

March Small-Firm Rev.

State-by-Week FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
7,084
18.4
72.8
[ -0.250, -0.079]
0.004

Yes
7,084
0.2
55.7
[ -0.297, -0.117]
0.004

Yes
7,017
14.1
8.2
[ -0.666, -0.095]
0.007

Yes
7,011
52.5
9.1
[ -0.687, -0.095]
0.007

Yes
7,011
162.6
11.7
[ -0.237, 0.009]
0.060

Yes
4,465
85.2
10.6
[ -0.373, -0.045]
0.018

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state, week, and state-by-week fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e.
the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data.
Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity
Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the
share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry
Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b)
the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights.
Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. In this regression, the weeks ending 2020-04-11 through 2020-04-25 are pooled, with the
remaining weeks excluded from the data.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

17


TABLE A.17
Continuing Claims, Firms Size 0-99 (Pooled Regression of Second Tranche Weeks)
(1)

(2)

(3)

(4)

(5)

(6)

-0.180∗∗∗
(0.036)

-0.210∗∗∗
(0.027)

-0.127∗
(0.057)

-0.159∗∗
(0.049)

-0.075
(0.038)

-0.110∗∗
(0.037)

February IUR

0.752∗∗∗
(0.111)

0.742∗∗∗
(0.124)

0.632∗∗∗
(0.094)

0.683∗∗∗
(0.104)

Log(Med. Income)

-0.022
(0.012)

-0.018
(0.013)

-0.019
(0.011)

-0.016
(0.012)

Poverty Rate

-0.001∗
(0.000)

-0.001
(0.001)

-0.001∗
(0.000)

-0.001
(0.001)

Log(Pop. Density)

0.005∗∗
(0.002)

0.005∗∗∗
(0.001)

0.006∗∗∗
(0.002)

0.005∗∗∗
(0.001)

Covid Cases, 1w

-1.587
(1.043)

-1.496∗
(0.760)

-2.761∗
(1.360)

Covid Cases, 4w

-0.836∗∗
(0.304)

-0.588∗
(0.250)

-0.485
(0.336)

Covid Deaths, 1w

-17.892
(14.284)

-2.831
(11.321)

-10.867
(15.004)

Covid Deaths, 4w

11.457∗∗∗
(3.022)

5.075
(2.744)

7.703∗
(3.077)

WFH Index

-0.072
(0.041)

0.056
(0.050)

-0.073∗
(0.034)

Early PPP Coverage

0.461∗∗∗
(0.079)

Industry Index

-0.047∗∗∗
(0.011)

March Small-Firm Rev.

State-by-Week FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
11,830
24.7
72.8
[ -0.255, -0.098]
0.002

Yes
11,830
0.3
55.6
[ -0.294, -0.155]
0.001

Yes
11,695
5.5
8.3
[ -0.341, 0.027]
0.077

Yes
11,685
6.9
9.4
[ -0.408, -0.069]
0.008

Yes
11,685
14.0
11.8
[ -0.176, 0.047]
0.130

Yes
7,441
7.4
10.8
[ -0.267, -0.034]
0.018

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state, week, and state-by-week fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e.
the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data.
Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity
Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the
share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry
Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b)
the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights.
Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. In this regression, the weeks ending 2020-05-02 through 2020-05-30 are pooled, with the
remaining weeks excluded from the data.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

18


TABLE A.18
Continuing Claims, Firms Size 0-99 (Pooled Regression of Post-PPP Rollout Weeks)
(1)

(2)

(3)

(4)

(5)

(6)

-0.042∗∗∗
(0.008)

-0.061∗∗∗
(0.007)

0.006
(0.033)

-0.004
(0.021)

0.005
(0.020)

0.003
(0.018)

February IUR

0.111∗∗∗
(0.033)

0.110∗∗∗
(0.031)

0.096∗∗
(0.032)

0.102∗∗
(0.032)

Log(Med. Income)

0.006
(0.005)

0.007
(0.005)

0.006
(0.005)

0.007
(0.004)

Poverty Rate

0.000∗
(0.000)

0.000∗∗
(0.000)

0.000∗
(0.000)

0.000∗∗
(0.000)

Log(Pop. Density)

0.002∗
(0.001)

0.003∗∗
(0.001)

0.003∗
(0.001)

0.003∗∗
(0.001)

Covid Cases, 1w

0.277∗∗∗
(0.060)

0.265∗∗∗
(0.065)

0.266∗∗∗
(0.079)

Covid Cases, 4w

-0.089
(0.110)

-0.118
(0.107)

-0.119
(0.096)

Covid Deaths, 1w

-1.290
(2.052)

-0.630
(1.941)

-0.335
(2.362)

Covid Deaths, 4w

13.059∗
(5.700)

13.470∗∗
(4.570)

14.365∗∗
(5.130)

WFH Index

-0.024
(0.018)

-0.006
(0.029)

-0.026
(0.022)

Early PPP Coverage

0.061∗
(0.026)

Industry Index

-0.007∗∗∗
(0.002)

March Small-Firm Rev.

State-by-Week FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
54,732
28.3
75.2
[ -0.057, -0.022]
0.003

Yes
54,732
0.1
55.8
[ -0.084, -0.048]
0.001

Yes
53,751
0.1
8.3
[ -0.070, 0.151]
0.863

Yes
53,705
0.2
10.5
[ -0.064, 0.066]
0.843

Yes
53,705
0.3
12.9
[ -0.043, 0.075]
0.806

Yes
34,224
0.2
12.4
[ -0.048, 0.057]
0.883

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state, week, and state-by-week fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e.
the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data.
Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity
Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the
share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry
Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b)
the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights.
Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. In this regression, the weeks ending 2020-06-06 through 2020-11-07 are pooled, with the
remaining weeks excluded from the data.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

19


TABLE A.19
Continuing Claims, Firms Size 0-99 (Week Ending 2020-02-22)
(1)

(2)

(3)

(4)

(5)

(6)

0.016
(0.009)

0.033∗∗
(0.012)

-0.001
(0.001)

-0.001
(0.002)

-0.001
(0.002)

-0.000
(0.001)

February IUR

0.964∗∗∗
(0.009)

0.965∗∗∗
(0.009)

0.964∗∗∗
(0.009)

0.965∗∗∗
(0.010)

Log(Med. Income)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

111.499
(93.795)

132.102
(98.893)

118.423
(81.755)

Covid Cases, 4w

-116.717
(92.273)

-138.574
(98.076)

-122.935
(80.135)

Covid Deaths, 1w

0.000
(.)

0.000
(.)

0.000
(.)

Covid Deaths, 4w

0.000
(.)

0.000
(.)

0.000
(.)

WFH Index

0.002
(0.001)

0.003
(0.002)

0.002
(0.001)

Early PPP Coverage

Industry Index

0.003
(0.003)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,339
3.0
72.9
[ -0.002, 0.039]
0.078

Yes
2,339
0.2
55.8
[ 0.015, 0.081]
0.001

Yes
2,339
10896.7
8.2
[ -0.008, 0.002]
0.302

Yes
2,337
17185.0
10.3
[ -0.008, 0.002]
0.406

Yes
2,337
997.7
12.5
[ -0.006, 0.003]
0.562

Yes
1,493
17126.0
11.9
[ -0.006, 0.003]
0.722

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-02-22.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

20


TABLE A.20
Continuing Claims, Firms Size 0-99 (Week Ending 2020-02-29)
(1)

(2)

(3)

(4)

(5)

(6)

0.015
(0.009)

0.030∗∗
(0.011)

-0.002
(0.003)

-0.002
(0.002)

-0.001
(0.002)

-0.001
(0.002)

February IUR

0.929∗∗∗
(0.017)

0.931∗∗∗
(0.017)

0.930∗∗∗
(0.017)

0.933∗∗∗
(0.019)

Log(Med. Income)

0.001
(0.001)

0.000
(0.001)

0.000
(0.001)

0.000
(0.001)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

0.000
(0.000)

Covid Cases, 1w

-127.346
(85.485)

-136.811
(88.444)

-149.471
(92.001)

Covid Cases, 4w

10.160
(15.708)

9.729
(16.145)

14.276
(18.024)

Covid Deaths, 1w

4801.676∗∗∗
(542.750)

4796.020∗∗∗
(546.586)

4926.635∗∗∗
(533.095)

Covid Deaths, 4w

0.000
(.)

0.000
(.)

0.000
(.)

WFH Index

0.004
(0.003)

0.005
(0.003)

0.004
(0.003)

Early PPP Coverage

Industry Index

0.003
(0.005)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.001
(0.000)
No
2,339
2.4
72.9
[ -0.003, 0.038]
0.109

Yes
2,339
0.2
55.8
[ 0.012, 0.076]
0.002

Yes
2,339
2722.8
8.2
[ -0.013, 0.004]
0.335

Yes
2,337
300.1
11.0
[ -0.011, 0.004]
0.420

Yes
2,337
417.3
13.8
[ -0.010, 0.004]
0.541

Yes
1,493
328.7
13.1
[ -0.008, 0.004]
0.734

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-02-29.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

21


TABLE A.21
Continuing Claims, Firms Size 0-99 (Week Ending 2020-03-07)
(1)

(2)

(3)

(4)

(5)

(6)

0.014
(0.009)

0.028∗
(0.011)

-0.005
(0.004)

-0.004
(0.003)

-0.004
(0.003)

-0.002
(0.003)

February IUR

0.892∗∗∗
(0.024)

0.894∗∗∗
(0.024)

0.893∗∗∗
(0.024)

0.897∗∗∗
(0.026)

Log(Med. Income)

0.001
(0.001)

-0.000
(0.001)

-0.000
(0.001)

0.000
(0.001)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

28.780
(37.502)

32.622
(38.092)

35.333
(41.070)

Covid Cases, 4w

-28.646
(31.658)

-32.232
(33.130)

-34.412
(35.502)

Covid Deaths, 1w

458.485∗∗∗
(53.627)

456.360∗∗∗
(53.585)

467.216∗∗∗
(52.336)

Covid Deaths, 4w

0.000
(.)

0.000
(.)

0.000
(.)

WFH Index

0.005
(0.004)

0.007
(0.005)

0.005
(0.004)

Early PPP Coverage

Industry Index

0.004
(0.007)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.001
(0.001)
No
2,339
2.0
72.9
[ -0.005, 0.037]
0.144

Yes
2,339
0.1
55.8
[ 0.010, 0.071]
0.003

Yes
2,339
50.7
8.2
[ -0.025, 0.003]
0.187

Yes
2,337
62.0
10.7
[ -0.017, 0.004]
0.226

Yes
2,337
85.2
13.3
[ -0.015, 0.004]
0.260

Yes
1,492
73.9
12.7
[ -0.013, 0.005]
0.427

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-03-07.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

22


TABLE A.22
Continuing Claims, Firms Size 0-99 (Week Ending 2020-03-14)
(1)

(2)

(3)

(4)

(5)

(6)

0.012
(0.010)

0.024∗
(0.010)

-0.009
(0.006)

-0.007
(0.005)

-0.006
(0.004)

-0.005
(0.004)

February IUR

0.851∗∗∗
(0.032)

0.854∗∗∗
(0.032)

0.853∗∗∗
(0.031)

0.858∗∗∗
(0.034)

Log(Med. Income)

0.000
(0.001)

-0.001
(0.001)

-0.001
(0.001)

-0.000
(0.001)

Poverty Rate

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

5.592
(15.919)

3.491
(14.243)

2.705
(14.768)

Covid Cases, 4w

-4.195
(12.888)

-2.843
(11.687)

-2.042
(12.177)

Covid Deaths, 1w

-235.851
(169.895)

-260.684
(169.768)

-321.356∗
(159.174)

Covid Deaths, 4w

409.629∗∗∗
(105.101)

424.662∗∗∗
(105.521)

463.352∗∗∗
(99.462)

0.009
(0.006)

0.010
(0.007)

0.009
(0.006)

Early PPP Coverage

WFH Index

Industry Index

0.005
(0.010)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.001
(0.001)
No
2,352
1.5
72.9
[ -0.007, 0.035]
0.200

Yes
2,352
0.1
55.9
[ 0.007, 0.064]
0.006

Yes
2,339
19.9
8.2
[ -0.041, 0.003]
0.126

Yes
2,337
28.3
10.0
[ -0.029, 0.004]
0.140

Yes
2,337
39.3
12.4
[ -0.023, 0.003]
0.132

Yes
1,492
30.4
11.7
[ -0.020, 0.005]
0.278

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-03-14.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

23


TABLE A.23
Continuing Claims, Firms Size 0-99 (Week Ending 2020-03-21)
(1)

(2)

(3)

(4)

(5)

(6)

0.009
(0.009)

0.023∗
(0.010)

-0.015
(0.012)

-0.013
(0.009)

-0.011
(0.007)

-0.010
(0.008)

February IUR

0.859∗∗∗
(0.038)

0.862∗∗∗
(0.037)

0.860∗∗∗
(0.036)

0.863∗∗∗
(0.039)

Log(Med. Income)

-0.001
(0.003)

-0.002
(0.003)

-0.002
(0.003)

-0.002
(0.003)

Poverty Rate

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

-4.186
(3.350)

-4.046
(3.353)

-4.624
(3.527)

Covid Cases, 4w

2.480
(3.175)

2.276
(3.172)

2.598
(3.240)

Covid Deaths, 1w

-255.092
(157.771)

-260.541
(149.532)

-228.799
(172.292)

Covid Deaths, 4w

356.195∗∗∗
(77.149)

358.015∗∗∗
(74.180)

343.280∗∗∗
(85.790)

0.011
(0.008)

0.013
(0.011)

0.012
(0.008)

Early PPP Coverage

WFH Index

Industry Index

0.008
(0.014)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.002
(0.001)
No
2,352
1.1
72.9
[ -0.008, 0.031]
0.280

Yes
2,352
0.1
55.9
[ 0.006, 0.061]
0.010

Yes
2,339
12.3
8.2
[ -0.075, 0.008]
0.170

Yes
2,337
1074.7
9.6
[ -0.052, 0.006]
0.155

Yes
2,337
1108.8
12.3
[ -0.036, 0.004]
0.120

Yes
1,492
115.6
11.1
[ -0.038, 0.010]
0.246

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-03-21.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

24


TABLE A.24
Continuing Claims, Firms Size 0-99 (Week Ending 2020-03-28)
(1)

(2)

(3)

(4)

(5)

(6)

-0.018
(0.015)

-0.040∗
(0.016)

-0.078∗∗
(0.029)

-0.098∗
(0.039)

-0.055∗∗
(0.020)

-0.082∗∗
(0.030)

February IUR

1.060∗∗∗
(0.068)

1.066∗∗∗
(0.075)

1.017∗∗∗
(0.049)

1.052∗∗∗
(0.064)

Log(Med. Income)

-0.011
(0.007)

-0.017
(0.009)

-0.016∗
(0.006)

-0.016∗
(0.008)

Poverty Rate

-0.000
(0.000)

-0.001
(0.000)

-0.001∗
(0.000)

-0.001
(0.000)

Log(Pop. Density)

0.001
(0.001)

-0.000
(0.002)

0.000
(0.001)

-0.001
(0.001)

Covid Cases, 1w

-28.708∗
(12.851)

-20.117
(11.791)

-24.629∗
(11.645)

Covid Cases, 4w

21.759∗
(10.485)

14.863
(9.523)

18.446
(9.419)

Covid Deaths, 1w

-212.683
(127.485)

-244.143∗
(118.196)

-237.995
(135.850)

Covid Deaths, 4w

259.579∗
(119.486)

268.482∗
(109.186)

279.727∗
(125.625)

WFH Index

0.010
(0.037)

0.066∗∗
(0.021)

0.015
(0.029)

Early PPP Coverage

0.207∗∗∗
(0.033)

Industry Index

-0.015∗∗
(0.005)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,352
1.4
72.9
[ -0.050, 0.015]
0.245

Yes
2,352
0.1
55.9
[ -0.079, 0.002]
0.057

Yes
2,339
43.4
8.2
[ -0.277, -0.036]
0.001

Yes
2,337
64.9
9.1
[ -0.336, -0.040]
0.003

Yes
2,337
123.0
11.9
[ -0.143, -0.020]
0.009

Yes
1,488
96.8
10.6
[ -0.232, -0.033]
0.005

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-03-28.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

25


TABLE A.25
Continuing Claims, Firms Size 0-99 (Week Ending 2020-04-04)
(1)

(2)

(3)

(4)

(5)

(6)

-0.065∗∗
(0.023)

-0.104∗∗∗
(0.026)

-0.178∗∗
(0.066)

-0.190∗∗
(0.073)

-0.096∗∗
(0.033)

-0.144∗∗
(0.054)

February IUR

1.370∗∗∗
(0.176)

1.342∗∗∗
(0.182)

1.241∗∗∗
(0.113)

1.326∗∗∗
(0.152)

Log(Med. Income)

-0.034∗
(0.015)

-0.036
(0.020)

-0.035∗∗
(0.013)

-0.033
(0.017)

Poverty Rate

-0.002∗
(0.001)

-0.002
(0.001)

-0.002∗∗
(0.001)

-0.002∗
(0.001)

Log(Pop. Density)

0.001
(0.002)

-0.000
(0.003)

0.001
(0.002)

-0.001
(0.003)

Covid Cases, 1w

-2.162
(1.840)

0.654
(2.188)

-0.882
(1.745)

Covid Cases, 4w

0.856
(1.089)

-0.645
(0.848)

0.172
(0.783)

Covid Deaths, 1w

-432.578∗∗∗
(105.044)

-299.903∗∗∗
(70.592)

-396.044∗∗∗
(94.031)

Covid Deaths, 4w

330.223∗∗∗
(74.117)

221.310∗∗∗
(48.437)

300.529∗∗∗
(65.500)

0.012
(0.089)

0.143∗∗
(0.052)

0.020
(0.071)

Early PPP Coverage

WFH Index

0.469∗∗∗
(0.068)

Industry Index

-0.037∗∗∗
(0.010)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,352
8.1
72.9
[ -0.113, -0.013]
0.022

Yes
2,352
0.4
55.9
[ -0.172, -0.042]
0.011

Yes
2,339
24.9
8.2
[ -0.592, -0.071]
0.006

Yes
2,337
45.1
9.2
[ -0.612, -0.071]
0.006

Yes
2,337
54.2
11.7
[ -0.217, -0.021]
0.026

Yes
1,488
39.9
10.4
[ -0.386, -0.041]
0.016

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-04-04.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

26


TABLE A.26
Continuing Claims, Firms Size 0-99 (Week Ending 2020-04-11)
(1)

(2)

(3)

(4)

(5)

(6)

-0.111∗∗
(0.034)

-0.164∗∗∗
(0.035)

-0.274∗
(0.112)

-0.252∗
(0.107)

-0.135∗∗
(0.051)

-0.185∗
(0.078)

February IUR

1.580∗∗∗
(0.256)

1.507∗∗∗
(0.285)

1.391∗∗∗
(0.191)

1.458∗∗∗
(0.238)

Log(Med. Income)

-0.053∗
(0.023)

-0.043
(0.029)

-0.044∗
(0.018)

-0.039
(0.025)

Poverty Rate

-0.003∗
(0.001)

-0.002
(0.001)

-0.002∗∗
(0.001)

-0.002
(0.001)

Log(Pop. Density)

0.000
(0.003)

0.001
(0.005)

0.002
(0.002)

-0.001
(0.004)

Covid Cases, 1w

-12.599
(7.567)

-8.205
(5.727)

-10.194
(7.303)

Covid Cases, 4w

2.983
(3.260)

1.937
(2.458)

2.253
(2.980)

Covid Deaths, 1w

-228.778∗∗∗
(54.139)

-172.036∗∗
(55.083)

-240.967∗∗∗
(50.009)

Covid Deaths, 4w

158.201∗∗∗
(35.273)

111.486∗∗∗
(33.758)

162.884∗∗∗
(30.422)

0.027
(0.132)

0.191∗
(0.091)

0.040
(0.108)

Early PPP Coverage

WFH Index

0.594∗∗∗
(0.101)

Industry Index

-0.050∗∗∗
(0.014)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,352
10.1
72.9
[ -0.181, -0.029]
0.017

Yes
2,352
0.5
55.9
[ -0.257, -0.081]
0.007

Yes
2,339
31.1
8.2
[ -0.927, -0.083]
0.013

Yes
2,337
80.6
8.9
[ -0.840, -0.068]
0.015

Yes
2,337
142.3
11.6
[ -0.309, -0.012]
0.040

Yes
1,486
217.8
10.5
[ -0.499, -0.020]
0.036

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-04-11.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

27


TABLE A.27
Continuing Claims, Firms Size 0-99 (Week Ending 2020-04-18)
(1)

(2)

(3)

(4)

(5)

(6)

-0.200∗∗∗
(0.045)

-0.218∗∗∗
(0.038)

-0.207∗∗
(0.071)

-0.228∗∗
(0.079)

-0.103∗
(0.049)

-0.145∗∗
(0.052)

February IUR

1.658∗∗∗
(0.205)

1.587∗∗∗
(0.238)

1.467∗∗∗
(0.169)

1.538∗∗∗
(0.203)

Log(Med. Income)

-0.042∗∗
(0.016)

-0.041
(0.025)

-0.041∗∗
(0.015)

-0.035
(0.020)

Poverty Rate

-0.002∗∗
(0.001)

-0.002
(0.001)

-0.002∗∗
(0.001)

-0.002∗
(0.001)

Log(Pop. Density)

0.005∗∗
(0.002)

0.004
(0.003)

0.006∗∗∗
(0.002)

0.004
(0.002)

Covid Cases, 1w

-8.763
(5.030)

-6.243
(4.147)

-7.316
(4.814)

Covid Cases, 4w

1.463
(1.506)

1.333
(1.230)

1.356
(1.395)

Covid Deaths, 1w

-53.260
(36.244)

-29.370
(36.922)

-39.298
(41.377)

Covid Deaths, 4w

26.037∗∗
(9.892)

9.927
(11.638)

17.582
(12.276)

WFH Index

-0.035
(0.094)

0.134∗
(0.059)

-0.032
(0.064)

Early PPP Coverage

0.615∗∗∗
(0.093)

Industry Index

-0.058∗∗∗
(0.014)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,366
18.8
72.8
[ -0.296, -0.096]
0.003

Yes
2,366
0.8
55.6
[ -0.322, -0.131]
0.004

Yes
2,339
52.7
8.2
[ -0.587, -0.072]
0.013

Yes
2,337
248.3
8.9
[ -0.653, -0.092]
0.007

Yes
2,337
682.6
11.5
[ -0.240, 0.040]
0.101

Yes
1,488
346.1
10.5
[ -0.344, -0.034]
0.024

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-04-18.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

28


TABLE A.28
Continuing Claims, Firms Size 0-99 (Week Ending 2020-04-25)
(1)

(2)

(3)

(4)

(5)

(6)

-0.197∗∗∗
(0.043)

-0.218∗∗∗
(0.035)

-0.199∗∗
(0.066)

-0.224∗∗
(0.073)

-0.100∗
(0.044)

-0.147∗∗
(0.048)

February IUR

1.460∗∗∗
(0.190)

1.426∗∗∗
(0.217)

1.298∗∗∗
(0.155)

1.361∗∗∗
(0.185)

Log(Med. Income)

-0.040∗∗
(0.015)

-0.039
(0.024)

-0.039∗∗
(0.014)

-0.035
(0.019)

Poverty Rate

-0.002∗∗
(0.001)

-0.002
(0.001)

-0.002∗∗
(0.001)

-0.002∗
(0.001)

Log(Pop. Density)

0.004∗∗
(0.002)

0.004
(0.003)

0.006∗∗∗
(0.001)

0.003
(0.002)

Covid Cases, 1w

-5.086∗
(2.378)

-4.441∗
(2.046)

-5.446
(2.971)

Covid Cases, 4w

0.986
(0.950)

1.187
(0.845)

1.303
(0.973)

Covid Deaths, 1w

-60.886
(52.756)

-25.743
(49.717)

-57.530
(56.809)

Covid Deaths, 4w

15.852
(9.412)

1.973
(8.783)

11.837
(10.645)

WFH Index

-0.031
(0.087)

0.133∗
(0.052)

-0.027
(0.060)

Early PPP Coverage

0.605∗∗∗
(0.088)

Industry Index

-0.058∗∗∗
(0.014)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,366
20.1
72.8
[ -0.288, -0.097]
0.003

Yes
2,366
0.9
55.6
[ -0.317, -0.138]
0.003

Yes
2,339
40.3
8.2
[ -0.545, -0.074]
0.013

Yes
2,337
142.1
9.1
[ -0.619, -0.098]
0.006

Yes
2,337
398.1
11.8
[ -0.219, 0.033]
0.092

Yes
1,491
271.1
10.6
[ -0.339, -0.045]
0.017

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-04-25.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

29


TABLE A.29
Continuing Claims, Firms Size 0-99 (Week Ending 2020-05-02)
(1)

(2)

(3)

(4)

(5)

(6)

-0.194∗∗∗
(0.041)

-0.220∗∗∗
(0.033)

-0.180∗∗
(0.060)

-0.214∗∗
(0.066)

-0.099∗
(0.042)

-0.145∗∗
(0.045)

February IUR

1.213∗∗∗
(0.158)

1.172∗∗∗
(0.189)

1.060∗∗∗
(0.127)

1.107∗∗∗
(0.157)

Log(Med. Income)

-0.035∗
(0.014)

-0.032
(0.021)

-0.033∗
(0.013)

-0.030
(0.018)

Poverty Rate

-0.002∗∗
(0.001)

-0.002
(0.001)

-0.002∗∗
(0.001)

-0.002
(0.001)

Log(Pop. Density)

0.005∗∗
(0.001)

0.004
(0.002)

0.006∗∗∗
(0.001)

0.004∗
(0.002)

Covid Cases, 1w

-1.071
(1.180)

-1.516
(1.060)

-2.861
(1.740)

Covid Cases, 4w

-0.740
(0.392)

-0.254
(0.383)

-0.075
(0.472)

Covid Deaths, 1w

-43.178
(25.879)

0.192
(22.950)

-38.974
(25.299)

Covid Deaths, 4w

13.213∗∗∗
(3.092)

2.091
(2.874)

8.624∗∗
(3.296)

WFH Index

-0.048
(0.075)

0.108∗
(0.048)

-0.044
(0.052)

Early PPP Coverage

0.575∗∗∗
(0.085)

Industry Index

-0.056∗∗∗
(0.013)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,366
22.1
72.8
[ -0.280, -0.100]
0.002

Yes
2,366
1.1
55.6
[ -0.314, -0.149]
0.002

Yes
2,339
23.6
8.2
[ -0.475, -0.058]
0.018

Yes
2,337
87.4
9.0
[ -0.578, -0.103]
0.004

Yes
2,337
58.5
11.7
[ -0.212, 0.025]
0.080

Yes
1,492
122.8
10.6
[ -0.335, -0.056]
0.011

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-05-02.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

30


TABLE A.30
Continuing Claims, Firms Size 0-99 (Week Ending 2020-05-09)
(1)

(2)

(3)

(4)

(5)

(6)

-0.191∗∗∗
(0.039)

-0.223∗∗∗
(0.030)

-0.158∗∗
(0.058)

-0.196∗∗∗
(0.057)

-0.092∗
(0.040)

-0.135∗∗∗
(0.041)

February IUR

0.927∗∗∗
(0.131)

0.914∗∗∗
(0.150)

0.790∗∗∗
(0.108)

0.844∗∗∗
(0.125)

Log(Med. Income)

-0.029∗
(0.013)

-0.027
(0.018)

-0.027∗
(0.012)

-0.024
(0.015)

Poverty Rate

-0.001∗
(0.001)

-0.001
(0.001)

-0.002∗∗
(0.001)

-0.001
(0.001)

Log(Pop. Density)

0.005∗∗
(0.002)

0.005∗∗
(0.002)

0.006∗∗∗
(0.002)

0.004∗∗
(0.001)

Covid Cases, 1w

-1.643
(2.257)

-2.331
(1.821)

-2.280
(2.313)

Covid Cases, 4w

-1.026∗
(0.399)

-0.514
(0.269)

-0.824
(0.486)

Covid Deaths, 1w

-0.042
(19.449)

15.348
(15.962)

15.642
(17.071)

Covid Deaths, 4w

11.702∗∗∗
(3.394)

3.255
(2.748)

7.218∗
(3.531)

WFH Index

-0.065
(0.055)

0.078
(0.049)

-0.067
(0.039)

Early PPP Coverage

0.525∗∗∗
(0.083)

Industry Index

-0.052∗∗∗
(0.013)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,366
24.0
72.8
[ -0.273, -0.103]
0.002

Yes
2,366
1.3
55.6
[ -0.314, -0.160]
0.002

Yes
2,339
5.1
8.2
[ -0.406, -0.020]
0.036

Yes
2,337
10.2
9.0
[ -0.500, -0.098]
0.005

Yes
2,337
23.9
11.5
[ -0.198, 0.030]
0.090

Yes
1,491
19.0
10.3
[ -0.306, -0.052]
0.012

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-05-09.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

31


TABLE A.31
Continuing Claims, Firms Size 0-99 (Week Ending 2020-05-16)
(1)

(2)

(3)

(4)

(5)

(6)

-0.189∗∗∗
(0.037)

-0.226∗∗∗
(0.028)

-0.135∗
(0.062)

-0.171∗∗∗
(0.050)

-0.082∗
(0.040)

-0.120∗∗
(0.038)

February IUR

0.632∗∗∗
(0.117)

0.625∗∗∗
(0.131)

0.510∗∗∗
(0.105)

0.553∗∗∗
(0.115)

Log(Med. Income)

-0.023
(0.013)

-0.020
(0.015)

-0.020
(0.012)

-0.017
(0.013)

Poverty Rate

-0.001
(0.001)

-0.001
(0.001)

-0.001∗
(0.001)

-0.001
(0.001)

Log(Pop. Density)

0.005∗∗
(0.002)

0.005∗∗∗
(0.001)

0.007∗∗∗
(0.002)

0.005∗∗∗
(0.001)

Covid Cases, 1w

-0.045
(2.292)

-0.913
(1.192)

0.438
(2.776)

Covid Cases, 4w

-1.175∗
(0.483)

-0.808
(0.413)

-1.173
(0.639)

Covid Deaths, 1w

-1.225
(23.816)

24.378
(19.872)

1.110
(27.675)

Covid Deaths, 4w

11.974∗∗
(4.507)

3.765
(3.316)

9.068
(4.747)

WFH Index

-0.082∗
(0.040)

0.048
(0.056)

-0.086∗
(0.035)

Early PPP Coverage

0.473∗∗∗
(0.087)

Industry Index

-0.048∗∗∗
(0.012)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,366
25.9
72.8
[ -0.269, -0.106]
0.001

Yes
2,366
1.5
55.6
[ -0.315, -0.170]
0.001

Yes
2,339
2.8
8.2
[ -0.360, 0.037]
0.084

Yes
2,337
5.4
9.3
[ -0.418, -0.080]
0.007

Yes
2,337
18.1
11.9
[ -0.183, 0.041]
0.115

Yes
1,492
5.9
10.8
[ -0.260, -0.039]
0.016

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-05-16.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

32


TABLE A.32
Continuing Claims, Firms Size 0-99 (Week Ending 2020-05-23)
(1)

(2)

(3)

(4)

(5)

(6)

-0.175∗∗∗
(0.034)

-0.205∗∗∗
(0.025)

-0.097
(0.062)

-0.146∗∗
(0.045)

-0.070
(0.037)

-0.103∗∗
(0.034)

February IUR

0.543∗∗∗
(0.098)

0.559∗∗∗
(0.109)

0.455∗∗∗
(0.088)

0.501∗∗∗
(0.093)

Log(Med. Income)

-0.014
(0.012)

-0.016
(0.013)

-0.017
(0.010)

-0.014
(0.011)

Poverty Rate

-0.001
(0.000)

-0.001
(0.001)

-0.001
(0.000)

-0.001
(0.001)

Log(Pop. Density)

0.006∗∗
(0.002)

0.005∗∗∗
(0.001)

0.006∗∗∗
(0.002)

0.005∗∗∗
(0.001)

Covid Cases, 1w

-5.693∗∗
(2.187)

-3.459∗
(1.609)

-6.500∗∗
(2.431)

Covid Cases, 4w

-0.268
(0.364)

-0.400
(0.278)

0.100
(0.492)

Covid Deaths, 1w

-25.198
(32.343)

-31.270
(26.801)

-20.873
(34.150)

Covid Deaths, 4w

17.282∗∗
(6.528)

12.909∗∗
(4.524)

11.679∗
(5.928)

WFH Index

-0.077∗
(0.035)

0.036
(0.053)

-0.079∗
(0.034)

Early PPP Coverage

0.408∗∗∗
(0.076)

Industry Index

-0.042∗∗∗
(0.010)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,366
25.9
72.8
[ -0.246, -0.097]
0.002

Yes
2,366
1.6
55.6
[ -0.284, -0.156]
0.001

Yes
2,339
2.7
8.2
[ -0.292, 0.102]
0.188

Yes
2,337
6.4
9.5
[ -0.351, -0.059]
0.010

Yes
2,337
11.5
11.9
[ -0.167, 0.045]
0.135

Yes
1,491
6.8
11.0
[ -0.231, -0.028]
0.021

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-05-23.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

33


TABLE A.33
Continuing Claims, Firms Size 0-99 (Week Ending 2020-05-30)
(1)

(2)

(3)

(4)

(5)

(6)

-0.150∗∗∗
(0.030)

-0.177∗∗∗
(0.022)

-0.065
(0.060)

-0.094∗
(0.040)

-0.039
(0.035)

-0.061
(0.032)

February IUR

0.445∗∗∗
(0.082)

0.448∗∗∗
(0.082)

0.368∗∗∗
(0.073)

0.405∗∗∗
(0.075)

Log(Med. Income)

-0.008
(0.010)

-0.004
(0.009)

-0.007
(0.009)

-0.004
(0.009)

Poverty Rate

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Log(Pop. Density)

0.005∗∗
(0.002)

0.006∗∗∗
(0.001)

0.006∗∗∗
(0.002)

0.005∗∗∗
(0.002)

Covid Cases, 1w

-2.101
(1.339)

-0.740
(1.136)

-2.068
(1.654)

Covid Cases, 4w

-1.009∗
(0.477)

-1.019∗∗
(0.393)

-0.859
(0.569)

Covid Deaths, 1w

11.805
(28.849)

9.757
(25.305)

7.443
(32.387)

Covid Deaths, 4w

16.006
(8.532)

10.833
(5.657)

12.990
(7.980)

WFH Index

-0.083∗∗
(0.027)

0.011
(0.057)

-0.087∗
(0.035)

Early PPP Coverage

0.327∗∗∗
(0.071)

Industry Index

-0.035∗∗∗
(0.008)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,366
24.8
72.8
[ -0.212, -0.081]
0.002

Yes
2,366
1.6
55.6
[ -0.247, -0.135]
0.001

Yes
2,339
2.4
8.2
[ -0.237, 0.145]
0.334

Yes
2,337
3.8
9.8
[ -0.261, -0.005]
0.043

Yes
2,337
4.9
12.2
[ -0.128, 0.075]
0.321

Yes
1,475
4.2
11.4
[ -0.181, 0.012]
0.082

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-05-30.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

34


TABLE A.34
Continuing Claims, Firms Size 0-99 (Week Ending 2020-06-06)
(1)

(2)

(3)

(4)

(5)

(6)

-0.121∗∗∗
(0.024)

-0.146∗∗∗
(0.018)

-0.027
(0.059)

-0.053
(0.038)

-0.012
(0.034)

-0.027
(0.030)

February IUR

0.344∗∗∗
(0.067)

0.346∗∗∗
(0.062)

0.287∗∗∗
(0.058)

0.313∗∗∗
(0.056)

Log(Med. Income)

0.001
(0.010)

0.004
(0.008)

0.002
(0.009)

0.005
(0.008)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.005∗
(0.002)

0.005∗∗∗
(0.001)

0.006∗∗
(0.002)

0.005∗∗
(0.002)

Covid Cases, 1w

0.982
(1.434)

0.366
(1.188)

-0.083
(1.509)

Covid Cases, 4w

-1.239∗
(0.573)

-0.886∗
(0.417)

-0.903
(0.537)

Covid Deaths, 1w

-29.806
(30.389)

-23.026
(26.812)

-30.638
(39.066)

Covid Deaths, 4w

22.072∗∗
(8.336)

15.781∗∗∗
(4.613)

18.611∗∗
(6.981)

WFH Index

-0.076∗∗
(0.028)

-0.006
(0.056)

-0.081∗
(0.037)

Early PPP Coverage

0.244∗∗∗
(0.061)

Industry Index

-0.026∗∗∗
(0.006)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,366
24.7
72.8
[ -0.171, -0.066]
0.002

Yes
2,366
1.6
55.6
[ -0.206, -0.113]
0.001

Yes
2,339
2.1
8.2
[ -0.180, 0.205]
0.660

Yes
2,337
3.6
10.3
[ -0.187, 0.044]
0.193

Yes
2,337
4.0
12.8
[ -0.094, 0.104]
0.723

Yes
1,491
3.2
11.8
[ -0.127, 0.053]
0.391

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-06-06.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

35


TABLE A.35
Continuing Claims, Firms Size 0-99 (Week Ending 2020-06-13)
(1)

(2)

(3)

(4)

(5)

(6)

-0.093∗∗∗
(0.019)

-0.114∗∗∗
(0.015)

0.011
(0.060)

-0.007
(0.039)

0.019
(0.036)

0.011
(0.032)

February IUR

0.238∗∗∗
(0.063)

0.232∗∗∗
(0.053)

0.195∗∗∗
(0.055)

0.212∗∗∗
(0.051)

Log(Med. Income)

0.009
(0.010)

0.013
(0.008)

0.012
(0.010)

0.014
(0.008)

Poverty Rate

0.001
(0.000)

0.001∗
(0.000)

0.001
(0.000)

0.001∗∗
(0.000)

Log(Pop. Density)

0.005∗
(0.002)

0.005∗∗
(0.002)

0.005∗∗
(0.002)

0.005∗∗
(0.002)

Covid Cases, 1w

1.621
(0.921)

1.253
(0.853)

1.514
(1.280)

Covid Cases, 4w

-1.081∗∗∗
(0.304)

-0.842∗∗
(0.271)

-1.056∗∗
(0.371)

Covid Deaths, 1w

28.315
(17.598)

31.880
(19.046)

40.674
(22.122)

Covid Deaths, 4w

11.633
(6.269)

7.274
(4.767)

7.299
(5.841)

WFH Index

-0.069∗
(0.034)

-0.023
(0.056)

-0.076
(0.041)

Early PPP Coverage

0.160∗∗
(0.051)

Industry Index

-0.018∗∗∗
(0.004)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,366
24.4
72.8
[ -0.132, -0.050]
0.002

Yes
2,366
1.3
55.6
[ -0.167, -0.087]
0.001

Yes
2,339
1.9
8.2
[ -0.130, 0.273]
0.847

Yes
2,337
2.8
9.6
[ -0.121, 0.120]
0.859

Yes
2,337
3.6
11.8
[ -0.064, 0.157]
0.593

Yes
1,491
2.8
11.0
[ -0.082, 0.114]
0.738

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-06-13.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

36


TABLE A.36
Continuing Claims, Firms Size 0-99 (Week Ending 2020-06-20)
(1)

(2)

(3)

(4)

(5)

(6)

-0.088∗∗∗
(0.017)

-0.109∗∗∗
(0.014)

0.013
(0.061)

-0.000
(0.042)

0.021
(0.039)

0.015
(0.035)

February IUR

0.220∗∗∗
(0.062)

0.212∗∗∗
(0.055)

0.180∗∗
(0.057)

0.194∗∗∗
(0.053)

Log(Med. Income)

0.009
(0.009)

0.014
(0.009)

0.012
(0.010)

0.014
(0.009)

Poverty Rate

0.001
(0.000)

0.001∗
(0.000)

0.001
(0.000)

0.001∗
(0.000)

Log(Pop. Density)

0.005∗
(0.002)

0.005∗∗
(0.002)

0.005∗
(0.002)

0.005∗∗
(0.002)

Covid Cases, 1w

1.800
(1.270)

0.963
(1.125)

1.450
(1.237)

Covid Cases, 4w

-0.850∗∗
(0.307)

-0.596∗
(0.301)

-0.855∗
(0.381)

Covid Deaths, 1w

16.797
(22.151)

19.142
(21.701)

37.457
(27.497)

Covid Deaths, 4w

13.259∗
(5.398)

9.991∗
(4.900)

9.244
(5.493)

WFH Index

-0.066
(0.036)

-0.026
(0.058)

-0.073
(0.044)

Early PPP Coverage

0.140∗∗
(0.051)

Industry Index

-0.016∗∗∗
(0.004)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,365
25.6
72.8
[ -0.125, -0.049]
0.002

Yes
2,365
1.4
55.6
[ -0.157, -0.083]
0.001

Yes
2,338
2.3
8.2
[ -0.126, 0.279]
0.821

Yes
2,336
3.1
9.7
[ -0.124, 0.141]
0.991

Yes
2,336
3.7
11.8
[ -0.070, 0.169]
0.576

Yes
1,492
3.0
11.3
[ -0.089, 0.125]
0.679

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-06-20.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

37


TABLE A.37
Continuing Claims, Firms Size 0-99 (Week Ending 2020-06-27)
(1)

(2)

(3)

(4)

(5)

(6)

-0.085∗∗∗
(0.016)

-0.103∗∗∗
(0.013)

0.012
(0.057)

0.001
(0.039)

0.020
(0.036)

0.014
(0.032)

February IUR

0.203∗∗∗
(0.059)

0.196∗∗∗
(0.047)

0.174∗∗∗
(0.051)

0.183∗∗∗
(0.048)

Log(Med. Income)

0.009
(0.009)

0.013
(0.009)

0.011
(0.010)

0.013
(0.008)

Poverty Rate

0.001
(0.000)

0.001∗
(0.000)

0.001
(0.000)

0.001∗
(0.000)

Log(Pop. Density)

0.004∗
(0.002)

0.005∗∗
(0.002)

0.005∗
(0.002)

0.005∗
(0.002)

Covid Cases, 1w

2.088
(1.272)

1.490
(1.350)

1.887
(1.297)

Covid Cases, 4w

-0.729∗∗
(0.277)

-0.626∗
(0.296)

-0.784∗
(0.374)

Covid Deaths, 1w

-11.636
(6.735)

1.713
(6.444)

-10.594
(7.356)

Covid Deaths, 4w

16.140∗
(7.479)

12.882∗∗
(4.962)

16.047∗∗
(5.830)

WFH Index

-0.060
(0.034)

-0.022
(0.054)

-0.066
(0.041)

Early PPP Coverage

0.129∗∗
(0.046)

Industry Index

-0.015∗∗∗
(0.004)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,365
27.6
72.8
[ -0.118, -0.048]
0.001

Yes
2,365
1.5
55.6
[ -0.146, -0.080]
0.001

Yes
2,338
2.9
8.2
[ -0.118, 0.265]
0.834

Yes
2,336
3.6
10.4
[ -0.113, 0.127]
0.986

Yes
2,336
3.6
12.4
[ -0.067, 0.152]
0.582

Yes
1,493
3.2
12.2
[ -0.080, 0.112]
0.670

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-06-27.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

38


TABLE A.38
Continuing Claims, Firms Size 0-99 (Week Ending 2020-07-04)
(1)

(2)

(3)

(4)

(5)

(6)

-0.080∗∗∗
(0.014)

-0.097∗∗∗
(0.011)

0.010
(0.054)

-0.017
(0.027)

0.002
(0.027)

-0.003
(0.025)

February IUR

0.184∗∗
(0.056)

0.171∗∗∗
(0.047)

0.153∗∗
(0.048)

0.157∗∗
(0.048)

Log(Med. Income)

0.009
(0.008)

0.007
(0.006)

0.007
(0.007)

0.008
(0.007)

Poverty Rate

0.001
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.004∗
(0.002)

0.004∗∗
(0.001)

0.004∗∗
(0.001)

0.004∗∗
(0.001)

Covid Cases, 1w

2.687∗∗∗
(0.811)

1.865∗
(0.757)

2.421∗∗
(0.798)

Covid Cases, 4w

-0.695∗∗∗
(0.201)

-0.553∗∗
(0.193)

-0.713∗∗
(0.225)

Covid Deaths, 1w

50.856∗∗
(17.038)

34.376
(17.680)

40.920∗
(18.540)

Covid Deaths, 4w

-1.060
(7.055)

3.615
(7.504)

1.395
(7.592)

WFH Index

-0.048
(0.026)

-0.020
(0.045)

-0.053
(0.033)

Early PPP Coverage

0.101∗
(0.044)

Industry Index

-0.012∗∗∗
(0.003)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,365
31.1
72.8
[ -0.110, -0.047]
0.001

Yes
2,365
1.7
55.6
[ -0.135, -0.075]
0.001

Yes
2,338
3.7
8.2
[ -0.112, 0.255]
0.849

Yes
2,336
16.9
10.0
[ -0.081, 0.086]
0.559

Yes
2,336
33.8
12.0
[ -0.051, 0.116]
0.950

Yes
1,468
30.1
11.7
[ -0.060, 0.089]
0.910

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-07-04.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

39


TABLE A.39
Continuing Claims, Firms Size 0-99 (Week Ending 2020-07-11)
(1)

(2)

(3)

(4)

(5)

(6)

-0.075∗∗∗
(0.012)

-0.091∗∗∗
(0.010)

0.008
(0.051)

-0.007
(0.029)

0.007
(0.028)

0.005
(0.026)

February IUR

0.165∗∗
(0.055)

0.165∗∗∗
(0.046)

0.145∗∗
(0.049)

0.153∗∗
(0.048)

Log(Med. Income)

0.009
(0.008)

0.009
(0.006)

0.008
(0.007)

0.010
(0.006)

Poverty Rate

0.001∗
(0.000)

0.001∗
(0.000)

0.000
(0.000)

0.001∗
(0.000)

Log(Pop. Density)

0.004∗
(0.002)

0.004∗∗
(0.001)

0.004∗
(0.001)

0.004∗
(0.001)

Covid Cases, 1w

-0.106
(0.470)

-0.159
(0.545)

-0.130
(0.484)

Covid Cases, 4w

0.214
(0.273)

0.095
(0.231)

0.156
(0.232)

Covid Deaths, 1w

-20.384
(18.765)

-8.940
(17.155)

-18.671
(23.020)

Covid Deaths, 4w

17.535∗
(8.178)

15.862∗∗
(5.349)

15.627∗∗
(5.318)

WFH Index

-0.044
(0.027)

-0.016
(0.044)

-0.048
(0.033)

Early PPP Coverage

0.094∗
(0.040)

Industry Index

-0.012∗∗∗
(0.003)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,365
35.2
72.8
[ -0.101, -0.047]
0.001

Yes
2,365
1.9
55.6
[ -0.123, -0.071]
0.001

Yes
2,338
4.7
8.2
[ -0.105, 0.243]
0.866

Yes
2,336
7.7
11.1
[ -0.079, 0.093]
0.809

Yes
2,336
17.1
13.4
[ -0.054, 0.110]
0.792

Yes
1,491
12.4
13.0
[ -0.060, 0.088]
0.861

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-07-11.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

40


TABLE A.40
Continuing Claims, Firms Size 0-99 (Week Ending 2020-07-18)
(1)

(2)

(3)

(4)

(5)

(6)

-0.070∗∗∗
(0.011)

-0.084∗∗∗
(0.009)

0.009
(0.050)

-0.004
(0.028)

0.008
(0.027)

0.007
(0.026)

February IUR

0.149∗∗
(0.054)

0.149∗∗
(0.046)

0.131∗∗
(0.048)

0.138∗∗
(0.049)

Log(Med. Income)

0.009
(0.007)

0.009
(0.006)

0.008
(0.007)

0.010
(0.006)

Poverty Rate

0.001∗
(0.000)

0.001∗
(0.000)

0.000
(0.000)

0.001∗
(0.000)

Log(Pop. Density)

0.003∗
(0.002)

0.003∗∗
(0.001)

0.004∗
(0.001)

0.004∗
(0.001)

Covid Cases, 1w

0.632
(0.537)

0.724
(0.477)

0.951
(0.541)

Covid Cases, 4w

-0.036
(0.354)

-0.180
(0.310)

-0.204
(0.320)

Covid Deaths, 1w

-0.155
(17.148)

12.577
(17.688)

17.112
(24.685)

Covid Deaths, 4w

16.298∗
(7.792)

14.956∗∗
(5.617)

13.039∗
(5.877)

WFH Index

-0.037
(0.026)

-0.014
(0.041)

-0.043
(0.032)

Early PPP Coverage

0.080∗
(0.037)

Industry Index

-0.011∗∗∗
(0.003)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,365
39.8
72.8
[ -0.094, -0.045]
0.001

Yes
2,365
2.0
55.6
[ -0.113, -0.066]
0.001

Yes
2,338
6.2
8.2
[ -0.099, 0.240]
0.859

Yes
2,336
14.0
11.0
[ -0.074, 0.096]
0.884

Yes
2,336
24.3
13.1
[ -0.051, 0.110]
0.761

Yes
1,493
19.1
12.6
[ -0.056, 0.094]
0.778

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-07-18.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

41


TABLE A.41
Continuing Claims, Firms Size 0-99 (Week Ending 2020-07-25)
(1)

(2)

(3)

(4)

(5)

(6)

-0.065∗∗∗
(0.010)

-0.079∗∗∗
(0.009)

0.012
(0.049)

-0.004
(0.025)

0.007
(0.024)

0.005
(0.023)

February IUR

0.157∗
(0.062)

0.149∗∗
(0.057)

0.134∗
(0.057)

0.142∗
(0.060)

Log(Med. Income)

0.009
(0.007)

0.008
(0.005)

0.007
(0.006)

0.008
(0.005)

Poverty Rate

0.001∗
(0.000)

0.000∗
(0.000)

0.000
(0.000)

0.000∗
(0.000)

Log(Pop. Density)

0.003∗
(0.002)

0.003∗∗
(0.001)

0.003∗∗
(0.001)

0.003∗
(0.001)

Covid Cases, 1w

0.208
(0.630)

0.429
(0.615)

0.420
(0.657)

Covid Cases, 4w

0.035
(0.260)

-0.101
(0.225)

-0.042
(0.244)

Covid Deaths, 1w

-23.545∗
(9.236)

-15.568∗
(6.378)

-19.361∗
(7.901)

Covid Deaths, 4w

23.667∗∗
(9.173)

22.076∗∗
(6.918)

21.862∗∗∗
(6.524)

WFH Index

-0.035
(0.024)

-0.013
(0.038)

-0.039
(0.028)

Early PPP Coverage

0.073∗
(0.037)

Industry Index

-0.010∗∗∗
(0.003)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,364
41.1
72.8
[ -0.087, -0.043]
0.001

Yes
2,364
1.9
55.6
[ -0.106, -0.061]
0.001

Yes
2,337
4.9
8.2
[ -0.093, 0.240]
0.808

Yes
2,335
8.9
11.9
[ -0.063, 0.083]
0.879

Yes
2,335
11.9
14.1
[ -0.044, 0.094]
0.759

Yes
1,493
10.5
13.7
[ -0.048, 0.080]
0.819

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-07-25.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

42


TABLE A.42
Continuing Claims, Firms Size 0-99 (Week Ending 2020-08-01)
(1)

(2)

(3)

(4)

(5)

(6)

-0.060∗∗∗
(0.009)

-0.072∗∗∗
(0.008)

0.006
(0.049)

0.002
(0.036)

0.010
(0.033)

0.009
(0.032)

February IUR

0.179∗
(0.089)

0.174∗
(0.085)

0.161
(0.084)

0.176
(0.091)

Log(Med. Income)

0.006
(0.009)

0.008
(0.009)

0.007
(0.009)

0.008
(0.009)

Poverty Rate

0.001
(0.000)

0.001
(0.000)

0.000
(0.000)

0.001
(0.000)

Log(Pop. Density)

0.003
(0.002)

0.003∗
(0.002)

0.003∗
(0.002)

0.003∗
(0.002)

Covid Cases, 1w

-0.448
(0.626)

-0.262
(0.566)

-0.568
(0.773)

Covid Cases, 4w

0.096
(0.436)

-0.027
(0.395)

0.075
(0.419)

Covid Deaths, 1w

-4.468
(5.787)

-4.267
(5.382)

-0.118
(6.175)

Covid Deaths, 4w

12.836
(12.026)

15.420
(10.011)

13.384
(11.338)

WFH Index

-0.033
(0.028)

-0.015
(0.039)

-0.038
(0.032)

Early PPP Coverage

0.061∗
(0.030)

Industry Index

-0.007∗∗
(0.002)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,364
45.8
72.8
[ -0.078, -0.039]
0.001

Yes
2,364
1.9
55.6
[ -0.097, -0.056]
0.001

Yes
2,337
3.3
8.2
[ -0.109, 0.215]
0.898

Yes
2,335
4.2
12.7
[ -0.104, 0.099]
0.964

Yes
2,335
4.0
14.8
[ -0.083, 0.100]
0.752

Yes
1,491
4.1
14.4
[ -0.095, 0.086]
0.792

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-08-01.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

43


TABLE A.43
Continuing Claims, Firms Size 0-99 (Week Ending 2020-08-08)
(1)

(2)

(3)

(4)

(5)

(6)

-0.054∗∗∗
(0.008)

-0.064∗∗∗
(0.007)

0.005
(0.044)

-0.001
(0.033)

0.007
(0.030)

0.005
(0.030)

February IUR

0.166
(0.085)

0.160
(0.083)

0.149
(0.082)

0.162
(0.089)

Log(Med. Income)

0.005
(0.008)

0.006
(0.009)

0.006
(0.009)

0.007
(0.008)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.003
(0.001)

0.003∗
(0.001)

0.003∗
(0.001)

0.003
(0.002)

Covid Cases, 1w

-0.302
(0.627)

-0.049
(0.551)

-0.185
(0.732)

Covid Cases, 4w

0.066
(0.373)

-0.035
(0.357)

0.010
(0.409)

Covid Deaths, 1w

-2.606
(11.288)

-6.990
(10.508)

-7.913
(14.397)

Covid Deaths, 4w

9.628
(9.274)

12.285
(8.097)

12.889
(11.061)

WFH Index

-0.025
(0.025)

-0.011
(0.035)

-0.030
(0.029)

Early PPP Coverage

Industry Index

0.050
(0.028)
-0.006∗∗
(0.002)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,364
49.3
72.8
[ -0.070, -0.037]
0.000

Yes
2,364
1.8
55.6
[ -0.087, -0.049]
0.001

Yes
2,337
2.2
8.2
[ -0.100, 0.187]
0.914

Yes
2,335
34.4
11.6
[ -0.094, 0.094]
0.979

Yes
2,335
36.1
13.8
[ -0.076, 0.094]
0.818

Yes
1,491
47.4
13.7
[ -0.090, 0.080]
0.856

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-08-08.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

44


TABLE A.44
Continuing Claims, Firms Size 0-99 (Week Ending 2020-08-15)
(1)

(2)

(3)

(4)

(5)

(6)

-0.049∗∗∗
(0.007)

-0.056∗∗∗
(0.007)

0.004
(0.039)

-0.002
(0.028)

0.004
(0.026)

0.003
(0.026)

February IUR

0.152
(0.083)

0.147
(0.082)

0.138
(0.080)

0.148
(0.088)

Log(Med. Income)

0.005
(0.008)

0.005
(0.008)

0.005
(0.009)

0.006
(0.008)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.002
(0.001)

0.002∗
(0.001)

0.003∗
(0.001)

0.003∗
(0.001)

Covid Cases, 1w

0.303
(0.672)

0.396
(0.620)

0.506
(0.711)

Covid Cases, 4w

-0.059
(0.255)

-0.085
(0.248)

-0.103
(0.249)

Covid Deaths, 1w

-1.802
(7.018)

1.065
(7.178)

4.038
(9.848)

Covid Deaths, 4w

7.409∗∗∗
(1.808)

7.077∗∗∗
(1.549)

7.074∗∗∗
(1.601)

WFH Index

-0.019
(0.021)

-0.007
(0.030)

-0.022
(0.025)

Early PPP Coverage

Industry Index

0.040
(0.025)
-0.004∗
(0.002)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,364
53.0
72.8
[ -0.063, -0.034]
0.000

Yes
2,364
1.7
55.6
[ -0.077, -0.043]
0.001

Yes
2,337
1.0
8.2
[ -0.092, 0.165]
0.919

Yes
2,335
11.9
11.5
[ -0.087, 0.080]
0.942

Yes
2,335
11.1
14.1
[ -0.070, 0.078]
0.883

Yes
1,492
12.5
13.0
[ -0.083, 0.071]
0.917

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-08-15.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

45


TABLE A.45
Continuing Claims, Firms Size 0-99 (Week Ending 2020-08-22)
(1)

(2)

(3)

(4)

(5)

(6)

-0.039∗∗∗
(0.008)

-0.051∗∗∗
(0.007)

0.021
(0.037)

0.016
(0.027)

0.019
(0.025)

0.022
(0.023)

February IUR

0.100∗
(0.051)

0.095
(0.050)

0.090
(0.049)

0.092
(0.051)

Log(Med. Income)

0.009
(0.005)

0.009
(0.006)

0.009
(0.006)

0.010
(0.006)

Poverty Rate

0.001∗∗
(0.000)

0.001∗∗
(0.000)

0.001∗
(0.000)

0.001∗∗
(0.000)

Log(Pop. Density)

0.003∗
(0.001)

0.003∗
(0.001)

0.003∗
(0.001)

0.003∗
(0.001)

Covid Cases, 1w

-0.627
(0.465)

-0.587
(0.431)

-0.724
(0.439)

Covid Cases, 4w

0.057
(0.169)

0.053
(0.171)

0.095
(0.174)

Covid Deaths, 1w

-8.316
(7.727)

-7.017
(6.838)

-5.769
(11.407)

Covid Deaths, 4w

7.994∗∗∗
(1.493)

7.860∗∗∗
(1.525)

8.167∗∗∗
(1.566)

WFH Index

-0.016
(0.024)

-0.011
(0.031)

-0.019
(0.027)

Early PPP Coverage

Industry Index

0.018
(0.025)
-0.005∗
(0.002)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,364
26.0
72.8
[ -0.054, -0.020]
0.003

Yes
2,364
1.2
55.6
[ -0.072, -0.036]
0.002

Yes
2,337
0.8
8.2
[ -0.053, 0.209]
0.538

Yes
2,335
2.8
11.4
[ -0.044, 0.119]
0.530

Yes
2,335
3.0
14.1
[ -0.037, 0.107]
0.433

Yes
1,493
4.7
13.5
[ -0.033, 0.103]
0.338

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-08-22.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

46


TABLE A.46
Continuing Claims, Firms Size 0-99 (Week Ending 2020-08-29)
(1)

(2)

(3)

(4)

(5)

(6)

-0.035∗∗∗
(0.009)

-0.053∗∗∗
(0.007)

0.012
(0.031)

0.008
(0.024)

0.012
(0.022)

0.013
(0.021)

February IUR

0.044
(0.032)

0.038
(0.032)

0.031
(0.031)

0.028
(0.033)

Log(Med. Income)

0.009∗
(0.004)

0.010
(0.005)

0.009
(0.005)

0.010∗
(0.005)

Poverty Rate

0.001∗∗∗
(0.000)

0.001∗∗∗
(0.000)

0.001∗∗∗
(0.000)

0.001∗∗∗
(0.000)

Log(Pop. Density)

0.002∗
(0.001)

0.002∗
(0.001)

0.002∗
(0.001)

0.002∗
(0.001)

Covid Cases, 1w

-0.795∗∗
(0.283)

-0.841∗∗
(0.296)

-0.992∗∗
(0.345)

Covid Cases, 4w

0.093
(0.108)

0.116
(0.104)

0.150
(0.126)

Covid Deaths, 1w

-7.772
(6.550)

-6.243
(6.712)

-5.325
(9.689)

Covid Deaths, 4w

6.972∗
(2.727)

6.627∗
(3.052)

7.039∗
(3.591)

WFH Index

-0.006
(0.023)

0.002
(0.030)

-0.009
(0.026)

Early PPP Coverage

Industry Index

0.028
(0.023)
-0.005∗∗
(0.002)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,364
16.2
72.8
[ -0.051, -0.013]
0.009

Yes
2,364
1.4
55.6
[ -0.077, -0.041]
0.001

Yes
2,337
0.8
8.2
[ -0.057, 0.148]
0.692

Yes
2,335
1.5
10.2
[ -0.050, 0.101]
0.738

Yes
2,335
1.8
12.7
[ -0.040, 0.094]
0.580

Yes
1,482
3.0
12.0
[ -0.040, 0.084]
0.524

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-08-29.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

47


TABLE A.47
Continuing Claims, Firms Size 0-99 (Week Ending 2020-09-05)
(1)

(2)

(3)

(4)

(5)

(6)

-0.028∗∗
(0.011)

-0.049∗∗∗
(0.007)

0.016
(0.028)

0.013
(0.022)

0.016
(0.021)

0.016
(0.018)

February IUR

0.028
(0.035)

0.022
(0.034)

0.017
(0.034)

0.012
(0.036)

Log(Med. Income)

0.010∗
(0.004)

0.010∗
(0.005)

0.010∗
(0.005)

0.011∗∗
(0.004)

Poverty Rate

0.001∗∗∗
(0.000)

0.001∗∗∗
(0.000)

0.001∗∗∗
(0.000)

0.001∗∗∗
(0.000)

Log(Pop. Density)

0.002∗
(0.001)

0.002∗
(0.001)

0.002∗
(0.001)

0.002∗
(0.001)

Covid Cases, 1w

-0.759
(0.392)

-0.820∗
(0.393)

-1.474∗∗
(0.571)

Covid Cases, 4w

0.058
(0.138)

0.074
(0.126)

0.206
(0.208)

Covid Deaths, 1w

-3.889
(4.187)

-4.552
(4.242)

-6.602
(4.724)

Covid Deaths, 4w

4.698
(3.649)

5.142
(3.412)

6.068
(3.960)

WFH Index

-0.010
(0.020)

-0.004
(0.025)

-0.013
(0.022)

Early PPP Coverage

Industry Index

0.021
(0.018)
-0.005∗∗
(0.002)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,364
6.5
72.8
[ -0.049, -0.001]
0.048

Yes
2,364
1.3
55.6
[ -0.071, -0.037]
0.001

Yes
2,337
0.7
8.2
[ -0.043, 0.150]
0.547

Yes
2,335
1.0
10.5
[ -0.041, 0.098]
0.564

Yes
2,335
1.0
12.8
[ -0.035, 0.092]
0.455

Yes
1,490
1.5
12.9
[ -0.033, 0.076]
0.393

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-09-05.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

48


TABLE A.48
Continuing Claims, Firms Size 0-99 (Week Ending 2020-09-12)
(1)

(2)

(3)

(4)

(5)

(6)

-0.019
(0.014)

-0.045∗∗∗
(0.007)

0.015
(0.026)

0.010
(0.020)

0.013
(0.020)

0.014
(0.017)

February IUR

0.008
(0.041)

0.001
(0.039)

-0.004
(0.040)

-0.009
(0.043)

Log(Med. Income)

0.008∗
(0.003)

0.009∗
(0.004)

0.009∗
(0.004)

0.010∗∗
(0.003)

Poverty Rate

0.001∗∗∗
(0.000)

0.001∗∗∗
(0.000)

0.001∗∗∗
(0.000)

0.001∗∗∗
(0.000)

Log(Pop. Density)

0.002∗
(0.001)

0.002∗
(0.001)

0.002∗
(0.001)

0.002∗
(0.001)

Covid Cases, 1w

-0.756
(0.490)

-0.878
(0.521)

-0.725
(0.540)

Covid Cases, 4w

-0.149
(0.084)

-0.138
(0.078)

-0.199
(0.132)

Covid Deaths, 1w

7.574
(4.648)

8.730
(4.710)

9.896
(7.314)

Covid Deaths, 4w

3.586
(3.211)

3.854
(3.119)

5.088
(3.274)

WFH Index

-0.012
(0.016)

-0.006
(0.020)

-0.014
(0.017)

Early PPP Coverage

Industry Index

0.021
(0.017)
-0.004∗
(0.002)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,376
1.9
77.3
[ -0.045, 0.016]
0.225

Yes
2,376
1.0
56.0
[ -0.068, -0.033]
0.001

Yes
2,337
0.7
8.2
[ -0.042, 0.134]
0.551

Yes
2,335
1.9
10.7
[ -0.040, 0.084]
0.601

Yes
2,335
1.9
12.6
[ -0.034, 0.083]
0.493

Yes
1,459
1.7
13.0
[ -0.032, 0.067]
0.422

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-09-12.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

49


TABLE A.49
Continuing Claims, Firms Size 0-99 (Week Ending 2020-09-19)
(1)

(2)

(3)

(4)

(5)

(6)

-0.016
(0.013)

-0.040∗∗∗
(0.006)

0.013
(0.023)

0.009
(0.017)

0.011
(0.017)

0.012
(0.015)

February IUR

0.010
(0.037)

0.004
(0.036)

0.000
(0.036)

-0.005
(0.039)

Log(Med. Income)

0.007∗
(0.003)

0.009∗
(0.004)

0.008∗
(0.004)

0.009∗∗
(0.003)

Poverty Rate

0.001∗∗∗
(0.000)

0.001∗∗∗
(0.000)

0.001∗∗∗
(0.000)

0.001∗∗∗
(0.000)

Log(Pop. Density)

0.002∗
(0.001)

0.002∗
(0.001)

0.002∗
(0.001)

0.002∗
(0.001)

Covid Cases, 1w

0.235
(0.201)

0.251
(0.201)

0.379
(0.240)

Covid Cases, 4w

-0.338∗∗∗
(0.085)

-0.355∗∗∗
(0.092)

-0.426∗∗∗
(0.093)

Covid Deaths, 1w

-1.523
(3.617)

-1.401
(3.716)

-2.231
(4.314)

Covid Deaths, 4w

5.533
(3.389)

5.927
(3.192)

7.558∗
(3.728)

WFH Index

-0.011
(0.014)

-0.006
(0.018)

-0.013
(0.016)

Early PPP Coverage

Industry Index

0.015
(0.015)
-0.003∗
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,375
1.6
77.3
[ -0.040, 0.016]
0.255

Yes
2,375
1.1
55.9
[ -0.060, -0.029]
0.001

Yes
2,337
0.6
8.2
[ -0.038, 0.118]
0.568

Yes
2,335
1.3
10.2
[ -0.035, 0.076]
0.612

Yes
2,335
1.2
12.1
[ -0.030, 0.072]
0.514

Yes
1,491
1.5
12.2
[ -0.029, 0.061]
0.446

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-09-19.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

50


TABLE A.50
Continuing Claims, Firms Size 0-99 (Week Ending 2020-09-26)
(1)

(2)

(3)

(4)

(5)

(6)

-0.014
(0.012)

-0.035∗∗∗
(0.005)

0.013
(0.022)

0.010
(0.017)

0.011
(0.016)

0.012
(0.014)

February IUR

0.012
(0.035)

0.007
(0.033)

0.004
(0.033)

-0.001
(0.036)

Log(Med. Income)

0.007∗
(0.003)

0.008∗
(0.004)

0.008∗
(0.004)

0.008∗
(0.003)

Poverty Rate

0.000∗∗∗
(0.000)

0.001∗∗∗
(0.000)

0.001∗∗∗
(0.000)

0.001∗∗∗
(0.000)

Log(Pop. Density)

0.002∗
(0.001)

0.002∗
(0.001)

0.002∗
(0.001)

0.002∗
(0.001)

Covid Cases, 1w

0.599∗∗∗
(0.162)

0.625∗∗∗
(0.158)

0.736∗∗∗
(0.161)

Covid Cases, 4w

-0.370∗∗∗
(0.098)

-0.385∗∗∗
(0.101)

-0.444∗∗∗
(0.100)

Covid Deaths, 1w

-3.034
(3.680)

-3.308
(3.744)

-5.480
(4.840)

Covid Deaths, 4w

4.383
(2.389)

4.739∗
(2.094)

6.045∗
(2.754)

WFH Index

-0.009
(0.014)

-0.006
(0.017)

-0.011
(0.015)

Early PPP Coverage

Industry Index

0.011
(0.014)
-0.003∗
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,375
1.3
77.3
[ -0.036, 0.017]
0.303

Yes
2,375
1.0
55.9
[ -0.052, -0.025]
0.001

Yes
2,337
0.6
8.2
[ -0.034, 0.113]
0.534

Yes
2,335
1.7
10.4
[ -0.031, 0.074]
0.555

Yes
2,335
1.5
12.4
[ -0.028, 0.068]
0.479

Yes
1,491
2.1
12.3
[ -0.027, 0.059]
0.418

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-09-26.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

51


TABLE A.51
Continuing Claims, Firms Size 0-99 (Week Ending 2020-10-03)
(1)

(2)

(3)

(4)

(5)

(6)

-0.010
(0.011)

-0.029∗∗∗
(0.004)

0.006
(0.015)

0.005
(0.012)

0.006
(0.012)

0.007
(0.011)

February IUR

0.017
(0.029)

0.015
(0.028)

0.013
(0.029)

0.009
(0.031)

Log(Med. Income)

0.005∗
(0.002)

0.005∗
(0.003)

0.005
(0.003)

0.005∗
(0.002)

Poverty Rate

0.000∗∗∗
(0.000)

0.000∗∗∗
(0.000)

0.000∗∗∗
(0.000)

0.000∗∗∗
(0.000)

Log(Pop. Density)

0.001∗
(0.001)

0.001∗
(0.001)

0.001∗
(0.001)

0.001∗
(0.001)

Covid Cases, 1w

0.063
(0.246)

0.056
(0.257)

0.037
(0.287)

Covid Cases, 4w

-0.129
(0.090)

-0.132
(0.098)

-0.131
(0.108)

Covid Deaths, 1w

3.536
(4.826)

3.329
(4.679)

3.872
(6.660)

Covid Deaths, 4w

1.813
(2.534)

2.040
(2.263)

2.629
(3.243)

WFH Index

-0.004
(0.010)

-0.001
(0.013)

-0.005
(0.011)

Early PPP Coverage

Industry Index

0.007
(0.011)
-0.002∗
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,375
0.8
77.3
[ -0.031, 0.018]
0.401

Yes
2,375
1.0
55.9
[ -0.043, -0.020]
0.001

Yes
2,337
0.6
8.2
[ -0.028, 0.073]
0.671

Yes
2,335
1.4
11.4
[ -0.026, 0.045]
0.703

Yes
2,335
1.5
13.9
[ -0.023, 0.041]
0.622

Yes
1,491
1.7
13.4
[ -0.023, 0.037]
0.535

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-10-03.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

52


TABLE A.52
Continuing Claims, Firms Size 0-99 (Week Ending 2020-10-10)
(1)

(2)

(3)

(4)

(5)

(6)

-0.006
(0.010)

-0.023∗∗∗
(0.004)

-0.000
(0.009)

0.000
(0.009)

0.001
(0.008)

0.002
(0.008)

February IUR

0.023
(0.025)

0.021
(0.024)

0.020
(0.024)

0.015
(0.027)

Log(Med. Income)

0.003∗
(0.001)

0.003
(0.002)

0.003
(0.002)

0.003∗
(0.002)

Poverty Rate

0.000∗∗∗
(0.000)

0.000∗∗∗
(0.000)

0.000∗∗∗
(0.000)

0.000∗∗∗
(0.000)

Log(Pop. Density)

0.001∗∗
(0.000)

0.001∗
(0.000)

0.001∗
(0.000)

0.001∗
(0.000)

Covid Cases, 1w

-0.048
(0.131)

-0.053
(0.131)

-0.102
(0.161)

Covid Cases, 4w

-0.046
(0.035)

-0.048
(0.038)

-0.043
(0.043)

Covid Deaths, 1w

1.405
(2.769)

1.410
(2.763)

1.927
(4.473)

Covid Deaths, 4w

1.306
(1.570)

1.387
(1.441)

2.043
(1.911)

WFH Index

0.001
(0.008)

0.002
(0.010)

0.000
(0.009)

Early PPP Coverage

Industry Index

0.004
(0.009)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.001
(0.001)
No
2,375
0.4
77.3
[ -0.027, 0.019]
0.552

Yes
2,375
0.9
55.9
[ -0.035, -0.016]
0.001

Yes
2,337
0.8
8.2
[ -0.023, 0.037]
0.989

Yes
2,335
1.9
10.6
[ -0.021, 0.029]
0.967

Yes
2,335
2.0
13.1
[ -0.019, 0.026]
0.903

Yes
1,486
1.5
12.0
[ -0.020, 0.025]
0.799

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-10-10.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

53


TABLE A.53
Continuing Claims, Firms Size 0-99 (Week Ending 2020-10-17)
(1)

(2)

(3)

(4)

(5)

(6)

-0.001
(0.010)

-0.017∗∗∗
(0.003)

-0.009
(0.005)

-0.008
(0.005)

-0.007
(0.005)

-0.006
(0.005)

February IUR

0.031
(0.022)

0.030
(0.020)

0.030
(0.021)

0.024
(0.023)

Log(Med. Income)

0.001
(0.001)

0.001
(0.001)

0.001
(0.001)

0.001
(0.001)

Poverty Rate

0.000∗∗∗
(0.000)

0.000∗∗
(0.000)

0.000∗∗
(0.000)

0.000∗∗
(0.000)

Log(Pop. Density)

0.000∗∗
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Covid Cases, 1w

0.131
(0.139)

0.125
(0.132)

0.097
(0.182)

Covid Cases, 4w

-0.025
(0.041)

-0.026
(0.043)

-0.024
(0.052)

Covid Deaths, 1w

-4.133
(2.704)

-4.032
(2.694)

-6.584
(5.203)

Covid Deaths, 4w

1.455
(1.086)

1.501
(1.039)

2.226
(1.393)

WFH Index

0.007
(0.006)

0.009
(0.008)

0.008
(0.007)

Early PPP Coverage

Industry Index

0.004
(0.007)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.001
(0.001)
No
2,440
0.0
80.6
[ -0.022, 0.023]
0.891

Yes
2,440
0.6
56.2
[ -0.028, -0.011]
0.001

Yes
2,337
3.4
8.2
[ -0.032, 0.004]
0.126

Yes
2,335
5.3
10.6
[ -0.024, 0.005]
0.159

Yes
2,335
5.4
13.1
[ -0.020, 0.005]
0.167

Yes
1,490
5.7
12.5
[ -0.021, 0.007]
0.268

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-10-17.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

54


TABLE A.54
Continuing Claims, Firms Size 0-99 (Week Ending 2020-10-24)
(1)

(2)

(3)

(4)

(5)

(6)

-0.001
(0.010)

-0.015∗∗∗
(0.003)

-0.007
(0.005)

-0.007
(0.005)

-0.006
(0.004)

-0.005
(0.005)

February IUR

0.036
(0.019)

0.036∗
(0.018)

0.035
(0.018)

0.035∗
(0.018)

Log(Med. Income)

0.001
(0.001)

0.001
(0.001)

0.001
(0.001)

0.001
(0.001)

Poverty Rate

0.000∗∗∗
(0.000)

0.000∗∗∗
(0.000)

0.000∗∗∗
(0.000)

0.000∗∗∗
(0.000)

Log(Pop. Density)

0.000∗∗
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Covid Cases, 1w

0.399∗∗
(0.148)

0.406∗∗
(0.148)

0.500∗∗
(0.173)

Covid Cases, 4w

-0.103∗
(0.052)

-0.108∗
(0.052)

-0.144∗
(0.066)

Covid Deaths, 1w

1.103
(1.478)

1.148
(1.511)

0.771
(2.372)

Covid Deaths, 4w

0.375
(0.882)

0.455
(0.788)

0.912
(1.314)

WFH Index

0.007
(0.006)

0.008
(0.007)

0.008
(0.006)

Early PPP Coverage

Industry Index

0.004
(0.006)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.001
(0.001)
No
2,437
0.0
80.6
[ -0.021, 0.023]
0.957

Yes
2,437
0.6
56.4
[ -0.025, -0.009]
0.002

Yes
2,334
4.1
8.3
[ -0.028, 0.004]
0.153

Yes
2,332
2.8
10.6
[ -0.022, 0.005]
0.186

Yes
2,332
2.8
13.1
[ -0.019, 0.005]
0.204

Yes
1,490
3.3
12.4
[ -0.020, 0.007]
0.308

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-10-24.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

55


TABLE A.55
Continuing Claims, Firms Size 0-99 (Week Ending 2020-10-31)
(1)

(2)

(3)

(4)

(5)

(6)

0.001
(0.010)

-0.013∗∗∗
(0.003)

-0.007
(0.005)

-0.007
(0.005)

-0.006
(0.004)

-0.005
(0.005)

February IUR

0.042∗
(0.016)

0.042∗∗
(0.015)

0.041∗∗
(0.015)

0.039∗
(0.016)

Log(Med. Income)

0.001
(0.001)

0.001
(0.001)

0.000
(0.001)

0.001
(0.001)

Poverty Rate

0.000∗∗∗
(0.000)

0.000∗∗
(0.000)

0.000∗∗
(0.000)

0.000∗∗
(0.000)

Log(Pop. Density)

0.000∗∗
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Covid Cases, 1w

0.125
(0.087)

0.127
(0.087)

0.160
(0.111)

Covid Cases, 4w

0.002
(0.040)

-0.000
(0.040)

-0.017
(0.049)

Covid Deaths, 1w

-0.677
(1.224)

-0.692
(1.226)

0.809
(2.500)

Covid Deaths, 4w

-0.053
(0.754)

0.014
(0.687)

-0.191
(1.338)

WFH Index

0.006
(0.005)

0.007
(0.006)

0.007
(0.006)

Early PPP Coverage

Industry Index

0.003
(0.006)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.001
(0.001)
No
2,437
0.0
80.6
[ -0.019, 0.024]
0.955

Yes
2,437
0.6
56.4
[ -0.021, -0.008]
0.002

Yes
2,334
5.5
8.3
[ -0.026, 0.004]
0.164

Yes
2,332
3.3
10.5
[ -0.022, 0.005]
0.180

Yes
2,332
3.3
12.9
[ -0.019, 0.005]
0.192

Yes
1,484
3.4
11.9
[ -0.021, 0.006]
0.282

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-10-31.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

56


TABLE A.56
Continuing Claims, Firms Size 0-99 (Week Ending 2020-11-07)
(1)

(2)

(3)

(4)

(5)

(6)

0.002
(0.010)

-0.011∗∗∗
(0.002)

-0.007
(0.004)

-0.007
(0.004)

-0.006
(0.004)

-0.005
(0.004)

February IUR

0.049∗∗∗
(0.014)

0.049∗∗∗
(0.013)

0.048∗∗∗
(0.013)

0.045∗∗
(0.014)

Log(Med. Income)

0.001
(0.001)

0.000
(0.001)

0.000
(0.001)

0.000
(0.001)

Poverty Rate

0.000∗∗∗
(0.000)

0.000∗∗
(0.000)

0.000∗∗
(0.000)

0.000∗∗
(0.000)

Log(Pop. Density)

0.000∗
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Covid Cases, 1w

-0.054
(0.042)

-0.055
(0.041)

-0.026
(0.059)

Covid Cases, 4w

0.067∗∗
(0.026)

0.066∗∗
(0.025)

0.054
(0.030)

Covid Deaths, 1w

-0.399
(1.582)

-0.448
(1.555)

-0.848
(2.846)

Covid Deaths, 4w

-0.640
(0.765)

-0.591
(0.719)

-0.489
(1.200)

WFH Index

0.007
(0.005)

0.007
(0.006)

0.007
(0.005)

Early PPP Coverage

Industry Index

0.002
(0.005)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.001
(0.001)
No
2,437
0.0
80.6
[ -0.018, 0.025]
0.851

Yes
2,437
0.5
56.4
[ -0.018, -0.006]
0.003

Yes
2,334
6.3
8.3
[ -0.026, 0.002]
0.108

Yes
2,332
3.5
10.3
[ -0.021, 0.003]
0.129

Yes
2,332
3.6
12.7
[ -0.018, 0.003]
0.137

Yes
1,491
4.6
11.8
[ -0.020, 0.005]
0.232

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI continuing claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements are
covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR for workers from firms sized 0-99 (i.e. the dependent variable as
measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population density), from Census data. Also included are Covid
cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New York Times, via Opportunity Insights. The WFH Index
for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel & Neiman (2020) measure of the share of industry-level jobs
that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. Industry Index is calculated as the
inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current Establishment Survey and (b) the employment-share of
industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply, via Opportunity Insights. Regressions weighted by
the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this table presenting the week ending 2020-11-07.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

57


Appendix H: Online Appendix Continued: Initial Claims Tables

58


TABLE A.57
Initial Claims, Firms Size 0-99 (Pooled Regression of Pre-Covid Weeks)
(1)

(2)

(3)

(4)

(5)

(6)

0.001
(0.001)

0.001
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

February IUR

0.863∗∗∗
(0.030)

0.856∗∗∗
(0.029)

0.855∗∗∗
(0.029)

0.867∗∗∗
(0.029)

Log(Med. Income)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Poverty Rate

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Covid Cases, 1w

3.303∗∗
(1.243)

3.483∗∗
(1.307)

3.651∗∗
(1.173)

Covid Cases, 4w

-0.745
(1.263)

-0.840
(1.338)

-1.038
(1.137)

Covid Deaths, 1w

-387.876∗∗∗
(21.797)

-385.651∗∗∗
(22.482)

-388.391∗∗∗
(22.425)

Covid Deaths, 4w

373.671∗∗∗
(14.330)

372.124∗∗∗
(14.727)

373.322∗∗∗
(14.445)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Early PPP Coverage

WFH Index

Industry Index

-0.001
(0.001)

March Small-Firm Rev.

State-by-Week FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

0.000
(0.000)
No
10,352
0.9
86.5
[ -0.001, 0.002]
0.363

Yes
10,352
0.0
51.2
[ -0.000, 0.002]
0.197

Yes
10,352
157.4
9.8
[ -0.002, 0.001]
0.751

Yes
10,344
1441.5
13.0
[ -0.001, 0.001]
0.772

Yes
10,344
1417.3
17.8
[ -0.001, 0.001]
0.485

Yes
6,338
1433.3
17.3
[ -0.001, 0.001]
0.580

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state, week, and state-by-week fixed effects, February 2020 IUR (IUR in terms of initial claims in this
case) for workers from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and
log(population density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected
by the New York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the
Dingel & Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99),
at the 2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’
Current Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes
from Womply, via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. In this regression, the weeks ending 2020-02-22
through 2020-03-14 are pooled, with the remaining weeks excluded from the data.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

59


TABLE A.58
Initial Claims, Firms Size 0-99 (Pooled Regression of Covid-Onset Weeks)
(1)

(2)

(3)

(4)

(5)

(6)

-0.012
(0.018)

-0.076∗∗∗
(0.016)

-0.098∗
(0.039)

-0.094∗∗
(0.034)

-0.045∗∗
(0.016)

-0.066∗∗
(0.021)

February IUR

3.813∗∗
(1.307)

3.902∗∗
(1.207)

4.331∗∗∗
(0.668)

4.200∗∗∗
(1.064)

Log(Med. Income)

-0.018∗
(0.008)

-0.015
(0.009)

-0.016∗∗
(0.005)

-0.014∗
(0.007)

Poverty Rate

-0.001∗∗
(0.000)

-0.001∗
(0.000)

-0.001∗∗∗
(0.000)

-0.001∗∗
(0.000)

Log(Pop. Density)

0.000
(0.001)

0.000
(0.002)

0.001
(0.001)

-0.000
(0.001)

Covid Cases, 1w

-0.637
(0.722)

-1.669∗∗
(0.617)

-0.763
(0.429)

Covid Cases, 4w

0.221
(0.627)

0.921∗
(0.414)

0.311
(0.352)

Covid Deaths, 1w

-53.846
(38.001)

-36.921
(39.663)

-46.067
(39.604)

Covid Deaths, 4w

36.702
(27.105)

13.799
(28.747)

28.312
(29.114)

WFH Index

-0.008
(0.044)

0.072∗∗
(0.024)

-0.000
(0.032)

Early PPP Coverage

0.274∗∗∗
(0.036)

Industry Index

-0.024∗∗∗
(0.005)

March Small-Firm Rev.

State-by-Week FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
7,764
0.4
86.5
[ -0.054, 0.022]
0.497

Yes
7,764
0.2
51.2
[ -0.132, -0.046]
0.001

Yes
7,764
41.1
9.8
[ -0.285, -0.034]
0.007

Yes
7,758
12.3
12.3
[ -0.242, -0.038]
0.004

Yes
7,758
156.0
17.6
[ -0.090, -0.007]
0.030

Yes
4,743
17.8
16.0
[ -0.144, -0.027]
0.006

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state, week, and state-by-week fixed effects, February 2020 IUR (IUR in terms of initial claims in this
case) for workers from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and
log(population density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected
by the New York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the
Dingel & Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99),
at the 2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’
Current Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes
from Womply, via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. In this regression, the weeks ending 2020-03-21
through 2020-04-04 are pooled, with the remaining weeks excluded from the data.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

60


TABLE A.59
Initial Claims, Firms Size 0-99 (Pooled Regression of First Tranche Weeks)
(1)

(2)

(3)

(4)

(5)

(6)

-0.014
(0.013)

-0.037∗∗∗
(0.011)

-0.048∗
(0.022)

-0.041∗
(0.017)

-0.023∗∗
(0.008)

-0.025∗∗
(0.010)

February IUR

1.563∗∗∗
(0.428)

1.667∗∗∗
(0.330)

1.837∗∗∗
(0.206)

1.715∗∗∗
(0.274)

Log(Med. Income)

-0.007∗
(0.004)

-0.005
(0.004)

-0.005∗
(0.002)

-0.004
(0.003)

Poverty Rate

-0.000∗
(0.000)

-0.000
(0.000)

-0.000∗∗
(0.000)

-0.000
(0.000)

Log(Pop. Density)

-0.000
(0.001)

0.000
(0.001)

0.000
(0.000)

-0.000
(0.001)

Covid Cases, 1w

-0.625
(0.405)

-0.461
(0.283)

-0.543
(0.379)

Covid Cases, 4w

0.143
(0.178)

0.178
(0.138)

0.170
(0.159)

Covid Deaths, 1w

0.509
(0.888)

-1.029
(0.787)

0.080
(0.887)

Covid Deaths, 4w

-0.993
(1.014)

-1.656∗
(0.752)

-1.326
(0.892)

WFH Index

0.006
(0.021)

0.031
(0.021)

0.009
(0.017)

Early PPP Coverage

0.088∗∗
(0.030)

Industry Index

-0.007∗∗
(0.003)

March Small-Firm Rev.

State-by-Week FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
7,764
1.2
86.5
[ -0.043, 0.011]
0.258

Yes
7,764
0.1
51.2
[ -0.069, -0.013]
0.012

Yes
7,764
17.9
9.8
[ -0.138, -0.005]
0.035

Yes
7,758
127.4
11.6
[ -0.105, -0.007]
0.025

Yes
7,758
143.0
17.6
[ -0.045, -0.001]
0.043

Yes
4,740
174.2
15.2
[ -0.053, -0.003]
0.035

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state, week, and state-by-week fixed effects, February 2020 IUR (IUR in terms of initial claims in this
case) for workers from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and
log(population density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected
by the New York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the
Dingel & Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99),
at the 2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’
Current Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes
from Womply, via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. In this regression, the weeks ending 2020-04-11
through 2020-04-25 are pooled, with the remaining weeks excluded from the data.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

61


TABLE A.60
Initial Claims, Firms Size 0-99 (Pooled Regression of Second Tranche Weeks)
(1)

(2)

(3)

(4)

(5)

(6)

-0.003
(0.006)

-0.017∗∗
(0.007)

-0.022
(0.012)

-0.019∗
(0.009)

-0.013∗
(0.005)

-0.015∗
(0.006)

February IUR

0.325
(0.199)

0.380∗
(0.151)

0.432∗∗∗
(0.117)

0.365∗
(0.154)

Log(Med. Income)

-0.002
(0.002)

-0.002
(0.002)

-0.002
(0.002)

-0.002
(0.002)

Poverty Rate

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.026
(0.089)

0.030
(0.080)

0.011
(0.124)

Covid Cases, 4w

-0.056
(0.045)

-0.038
(0.031)

-0.037
(0.043)

Covid Deaths, 1w

2.476
(1.532)

3.602∗
(1.589)

1.805
(1.427)

Covid Deaths, 4w

-0.567
(0.440)

-1.004∗
(0.401)

-0.582
(0.376)

WFH Index

0.007
(0.010)

0.016
(0.012)

0.010
(0.010)

Early PPP Coverage

0.030∗
(0.015)

Industry Index

-0.002∗
(0.001)

March Small-Firm Rev.

State-by-Week FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
12,940
0.3
86.5
[ -0.017, 0.010]
0.604

Yes
12,940
0.0
51.2
[ -0.038, -0.004]
0.021

Yes
12,940
11.0
9.8
[ -0.074, -0.000]
0.047

Yes
12,930
8.6
11.8
[ -0.056, -0.002]
0.032

Yes
12,930
9.2
16.8
[ -0.032, -0.003]
0.024

Yes
7,900
12.1
15.0
[ -0.040, -0.003]
0.018

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state, week, and state-by-week fixed effects, February 2020 IUR (IUR in terms of initial claims in this
case) for workers from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and
log(population density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected
by the New York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the
Dingel & Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99),
at the 2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’
Current Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes
from Womply, via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. In this regression, the weeks ending 2020-05-02
through 2020-05-30 are pooled, with the remaining weeks excluded from the data.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

62


TABLE A.61
Initial Claims, Firms Size 0-99 (Pooled Regression of Post-PPP Rollout Weeks)
(1)

(2)

(3)

(4)

(5)

(6)

0.001
(0.001)

-0.002∗
(0.001)

-0.002
(0.002)

-0.002
(0.001)

-0.002
(0.001)

-0.001
(0.001)

February IUR

0.167∗∗∗
(0.039)

0.174∗∗∗
(0.038)

0.176∗∗∗
(0.039)

0.171∗∗∗
(0.042)

Log(Med. Income)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.036∗∗∗
(0.009)

0.036∗∗∗
(0.009)

0.037∗∗∗
(0.007)

Covid Cases, 4w

0.008
(0.006)

0.007
(0.005)

0.007
(0.004)

Covid Deaths, 1w

0.113
(0.155)

0.133
(0.149)

0.192
(0.186)

Covid Deaths, 4w

-0.048
(0.149)

-0.034
(0.171)

-0.038
(0.169)

WFH Index

0.002
(0.002)

0.003
(0.002)

0.003
(0.002)

Early PPP Coverage

Industry Index

0.002
(0.002)

March Small-Firm Rev.

State-by-Week FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
59,524
0.9
86.5
[ -0.001, 0.003]
0.343

Yes
59,524
0.0
51.2
[ -0.005, -0.000]
0.028

Yes
59,524
11.7
9.8
[ -0.009, 0.001]
0.137

Yes
59,478
23.9
13.1
[ -0.006, 0.000]
0.088

Yes
59,478
25.8
17.7
[ -0.004, 0.000]
0.086

Yes
36,325
21.1
17.2
[ -0.004, 0.000]
0.099

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state, week, and state-by-week fixed effects, February 2020 IUR (IUR in terms of initial claims in this
case) for workers from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and
log(population density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected
by the New York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the
Dingel & Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99),
at the 2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’
Current Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes
from Womply, via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. In this regression, the weeks ending 2020-06-06
through 2020-11-07 are pooled, with the remaining weeks excluded from the data.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

63


TABLE A.62
Initial Claims, Firms Size 0-99 (Week Ending 2020-02-22)
(1)

(2)

(3)

(4)

(5)

(6)

0.001
(0.001)

0.001
(0.001)

-0.000
(0.001)

-0.000
(0.000)

-0.000
(0.001)

-0.000
(0.000)

February IUR

1.206∗∗∗
(0.143)

1.206∗∗∗
(0.144)

1.205∗∗∗
(0.144)

1.242∗∗∗
(0.183)

Log(Med. Income)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

0.000
(0.000)

Poverty Rate

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

55.526
(65.374)

51.904
(64.012)

44.892
(67.894)

Covid Cases, 4w

-62.107
(64.723)

-58.166
(63.365)

-54.066
(66.796)

Covid Deaths, 1w

0.000
(.)

0.000
(.)

0.000
(.)

Covid Deaths, 4w

0.000
(.)

0.000
(.)

0.000
(.)

WFH Index

0.000
(0.001)

0.000
(0.001)

0.000
(0.001)

Early PPP Coverage

Industry Index

-0.001
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

0.000
(0.000)
No
2,588
2.6
86.5
[ -0.001, 0.003]
0.145

Yes
2,588
0.1
51.2
[ -0.000, 0.003]
0.071

Yes
2,588
226.2
9.8
[ -0.002, 0.001]
0.678

Yes
2,586
97.4
12.8
[ -0.002, 0.001]
0.568

Yes
2,586
263.1
17.1
[ -0.002, 0.001]
0.457

Yes
1,585
283.3
16.4
[ -0.002, 0.001]
0.454

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-02-22.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

64


TABLE A.63
Initial Claims, Firms Size 0-99 (Week Ending 2020-02-29)
(1)

(2)

(3)

(4)

(5)

(6)

0.001
(0.001)

0.001∗
(0.001)

0.001∗
(0.001)

0.001
(0.000)

0.001∗
(0.000)

0.001
(0.000)

February IUR

0.949∗∗∗
(0.060)

0.944∗∗∗
(0.059)

0.943∗∗∗
(0.058)

0.935∗∗∗
(0.068)

Log(Med. Income)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.000∗
(0.000)

0.000∗
(0.000)

0.000∗
(0.000)

0.000
(0.000)

Covid Cases, 1w

-53.178∗∗∗
(15.819)

-51.430∗∗
(15.646)

-49.904∗∗
(17.108)

Covid Cases, 4w

15.279∗∗∗
(3.709)

15.400∗∗∗
(3.885)

16.767∗∗∗
(3.986)

Covid Deaths, 1w

228.347∗
(90.227)

228.077∗
(89.273)

185.276∗
(88.612)

Covid Deaths, 4w

0.000
(.)

0.000
(.)

0.000
(.)

WFH Index

-0.001
(0.000)

-0.001
(0.001)

-0.001
(0.001)

Early PPP Coverage

Industry Index

-0.001
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

0.000
(0.000)
No
2,588
2.3
86.5
[ -0.001, 0.002]
0.165

Yes
2,588
0.1
51.2
[ 0.000, 0.003]
0.011

Yes
2,588
349.8
9.8
[ 0.000, 0.003]
0.051

Yes
2,586
147.7
13.8
[ 0.000, 0.003]
0.046

Yes
2,586
206.9
19.1
[ 0.000, 0.002]
0.044

Yes
1,585
142.8
18.4
[ -0.000, 0.002]
0.138

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-02-29.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

65


TABLE A.64
Initial Claims, Firms Size 0-99 (Week Ending 2020-03-07)
(1)

(2)

(3)

(4)

(5)

(6)

0.001
(0.001)

0.001
(0.000)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

February IUR

0.646∗∗∗
(0.071)

0.649∗∗∗
(0.071)

0.649∗∗∗
(0.071)

0.655∗∗∗
(0.086)

Log(Med. Income)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Poverty Rate

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

-5.477
(4.129)

-5.363
(3.936)

-2.963
(4.208)

Covid Cases, 4w

6.222
(4.024)

6.117
(3.858)

3.621
(4.192)

Covid Deaths, 1w

-2.197
(4.301)

-2.243
(4.333)

-4.344
(4.239)

Covid Deaths, 4w

0.000
(.)

0.000
(.)

0.000
(.)

WFH Index

0.001
(0.000)

0.001
(0.000)

0.001
(0.000)

Early PPP Coverage

Industry Index

0.000
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
0.8
86.5
[ -0.001, 0.002]
0.392

Yes
2,588
0.1
51.2
[ -0.000, 0.002]
0.091

Yes
2,588
227.8
9.8
[ -0.003, 0.000]
0.157

Yes
2,586
156.0
13.4
[ -0.003, 0.001]
0.208

Yes
2,586
157.0
18.3
[ -0.002, 0.000]
0.190

Yes
1,584
118.7
17.7
[ -0.003, 0.000]
0.136

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-03-07.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

66


TABLE A.65
Initial Claims, Firms Size 0-99 (Week Ending 2020-03-14)
(1)

(2)

(3)

(4)

(5)

(6)

-0.000
(0.001)

-0.001
(0.001)

-0.001
(0.001)

-0.000
(0.001)

-0.001
(0.001)

-0.000
(0.001)

February IUR

0.652∗∗∗
(0.072)

0.625∗∗∗
(0.064)

0.623∗∗∗
(0.062)

0.634∗∗∗
(0.071)

Log(Med. Income)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Poverty Rate

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Covid Cases, 1w

2.989
(1.657)

3.770∗
(1.713)

3.174∗
(1.606)

Covid Cases, 4w

-0.675
(1.488)

-1.184
(1.519)

-0.798
(1.420)

Covid Deaths, 1w

-367.557∗∗∗
(25.261)

-359.981∗∗∗
(25.438)

-369.173∗∗∗
(25.028)

Covid Deaths, 4w

363.417∗∗∗
(15.721)

358.601∗∗∗
(15.663)

363.931∗∗∗
(15.614)

0.001
(0.001)

-0.000
(0.001)

0.001
(0.001)

Early PPP Coverage

WFH Index

Industry Index

-0.002
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
0.0
86.5
[ -0.002, 0.002]
0.942

Yes
2,588
0.0
51.2
[ -0.003, 0.001]
0.431

Yes
2,588
53.3
9.8
[ -0.006, 0.002]
0.690

Yes
2,586
1041.3
12.4
[ -0.004, 0.002]
0.649

Yes
2,586
1143.8
17.3
[ -0.004, 0.001]
0.384

Yes
1,584
1073.6
16.2
[ -0.003, 0.002]
0.710

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-03-14.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

67


TABLE A.66
Initial Claims, Firms Size 0-99 (Week Ending 2020-03-21)
(1)

(2)

(3)

(4)

(5)

(6)

-0.005
(0.016)

-0.079∗∗∗
(0.019)

-0.105∗
(0.044)

-0.103∗∗
(0.039)

-0.044∗∗
(0.017)

-0.086∗∗
(0.029)

February IUR

3.366∗
(1.627)

3.431∗
(1.559)

4.026∗∗∗
(0.940)

3.671∗
(1.572)

Log(Med. Income)

-0.019∗
(0.009)

-0.019
(0.010)

-0.016∗∗
(0.006)

-0.019∗
(0.008)

Poverty Rate

-0.001∗
(0.000)

-0.001∗
(0.000)

-0.001∗∗
(0.000)

-0.001∗∗
(0.000)

Log(Pop. Density)

0.000
(0.001)

0.000
(0.002)

0.001
(0.001)

-0.001
(0.001)

Covid Cases, 1w

-47.049
(32.431)

-46.439
(24.271)

-37.961
(27.573)

Covid Cases, 4w

39.550
(29.004)

36.156
(21.136)

29.540
(23.960)

Covid Deaths, 1w

571.098
(633.069)

499.412
(524.802)

606.625
(623.377)

Covid Deaths, 4w

-197.636
(404.064)

-232.353
(325.509)

-221.566
(390.597)

WFH Index

-0.005
(0.046)

0.083∗∗∗
(0.023)

0.005
(0.036)

Early PPP Coverage

0.305∗∗∗
(0.040)

Industry Index

-0.026∗∗∗
(0.006)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,588
0.1
86.5
[ -0.044, 0.027]
0.749

Yes
2,588
0.4
51.2
[ -0.152, -0.049]
0.000

Yes
2,588
47.0
9.8
[ -0.336, -0.039]
0.003

Yes
2,586
13.6
12.2
[ -0.289, -0.045]
0.001

Yes
2,586
399.2
18.3
[ -0.102, -0.013]
0.011

Yes
1,584
90.3
15.6
[ -0.210, -0.040]
0.001

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-03-21.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

68


TABLE A.67
Initial Claims, Firms Size 0-99 (Week Ending 2020-03-28)
(1)

(2)

(3)

(4)

(5)

(6)

-0.014
(0.021)

-0.077∗∗∗
(0.018)

-0.100∗
(0.040)

-0.100∗∗
(0.038)

-0.034∗
(0.016)

-0.061∗∗
(0.022)

February IUR

4.774∗∗∗
(1.421)

4.831∗∗∗
(1.339)

5.461∗∗∗
(0.670)

5.407∗∗∗
(1.094)

Log(Med. Income)

-0.020∗
(0.009)

-0.017
(0.010)

-0.014∗
(0.006)

-0.013
(0.008)

Poverty Rate

-0.001∗∗
(0.000)

-0.001∗
(0.000)

-0.001∗∗∗
(0.000)

-0.001∗
(0.000)

Log(Pop. Density)

0.000
(0.001)

0.000
(0.002)

0.001
(0.001)

-0.000
(0.001)

Covid Cases, 1w

-20.822
(11.447)

-7.695
(9.356)

-12.911
(9.307)

Covid Cases, 4w

16.295
(9.554)

5.819
(7.744)

9.906
(7.736)

Covid Deaths, 1w

71.329
(207.879)

25.649
(174.597)

58.904
(199.701)

Covid Deaths, 4w

-71.499
(194.274)

-60.996
(163.202)

-67.503
(185.991)

WFH Index

-0.019
(0.050)

0.072∗
(0.030)

-0.013
(0.035)

Early PPP Coverage

0.322∗∗∗
(0.047)

Industry Index

-0.028∗∗∗
(0.006)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,588
0.4
86.5
[ -0.066, 0.025]
0.486

Yes
2,588
0.4
51.2
[ -0.136, -0.040]
0.004

Yes
2,588
15.8
9.8
[ -0.278, -0.028]
0.014

Yes
2,586
33.1
11.6
[ -0.262, -0.036]
0.006

Yes
2,586
80.3
18.1
[ -0.071, 0.011]
0.098

Yes
1,579
52.9
15.2
[ -0.133, -0.016]
0.017

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-03-28.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

69


TABLE A.68
Initial Claims, Firms Size 0-99 (Week Ending 2020-04-04)
(1)

(2)

(3)

(4)

(5)

(6)

-0.016
(0.024)

-0.072∗∗∗
(0.021)

-0.091∗
(0.045)

-0.080∗
(0.037)

-0.040∗
(0.020)

-0.047∗
(0.022)

February IUR

3.299∗∗∗
(0.982)

3.474∗∗∗
(0.828)

3.886∗∗∗
(0.545)

3.736∗∗∗
(0.727)

Log(Med. Income)

-0.016∗
(0.007)

-0.012
(0.007)

-0.011∗
(0.004)

-0.008
(0.005)

Poverty Rate

-0.001∗∗∗
(0.000)

-0.001∗∗
(0.000)

-0.001∗∗∗
(0.000)

-0.001∗∗∗
(0.000)

Log(Pop. Density)

0.000
(0.001)

0.000
(0.001)

0.001
(0.001)

-0.000
(0.001)

Covid Cases, 1w

-0.135
(0.664)

1.321∗
(0.575)

0.719∗
(0.328)

Covid Cases, 4w

0.115
(0.473)

-0.627
(0.375)

-0.275
(0.274)

Covid Deaths, 1w

-88.114∗
(42.440)

-30.166
(23.939)

-66.327
(35.937)

Covid Deaths, 4w

58.179
(31.646)

10.413
(16.650)

39.998
(25.735)

WFH Index

-0.001
(0.039)

0.058
(0.037)

0.006
(0.028)

Early PPP Coverage

0.202∗∗
(0.064)

Industry Index

-0.017∗∗
(0.006)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,588
0.4
86.5
[ -0.072, 0.032]
0.504

Yes
2,588
0.3
51.2
[ -0.130, -0.025]
0.013

Yes
2,588
39.5
9.8
[ -0.271, -0.002]
0.046

Yes
2,586
105.1
11.8
[ -0.220, -0.007]
0.037

Yes
2,586
216.7
17.8
[ -0.090, 0.010]
0.090

Yes
1,580
234.2
15.0
[ -0.115, 0.005]
0.066

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-04-04.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

70


TABLE A.69
Initial Claims, Firms Size 0-99 (Week Ending 2020-04-11)
(1)

(2)

(3)

(4)

(5)

(6)

-0.018
(0.016)

-0.042∗∗
(0.013)

-0.054∗
(0.026)

-0.045∗
(0.020)

-0.024∗
(0.011)

-0.026∗
(0.012)

February IUR

2.172∗∗∗
(0.555)

2.228∗∗∗
(0.417)

2.458∗∗∗
(0.232)

2.417∗∗∗
(0.360)

Log(Med. Income)

-0.009∗
(0.004)

-0.007
(0.004)

-0.007∗∗
(0.003)

-0.005
(0.003)

Poverty Rate

-0.000∗∗
(0.000)

-0.000∗
(0.000)

-0.000∗∗∗
(0.000)

-0.000∗
(0.000)

Log(Pop. Density)

0.000
(0.001)

0.000
(0.001)

0.001
(0.000)

0.000
(0.001)

Covid Cases, 1w

-2.997∗
(1.499)

-1.962
(1.238)

-2.109
(1.347)

Covid Cases, 4w

0.960
(0.555)

0.706
(0.460)

0.724
(0.502)

Covid Deaths, 1w

-38.338∗∗∗
(11.205)

-29.601∗∗
(10.173)

-33.401∗∗
(10.217)

Covid Deaths, 4w

23.973∗∗
(7.723)

16.328∗
(6.510)

19.859∗∗
(6.569)

WFH Index

0.005
(0.023)

0.035
(0.025)

0.009
(0.019)

Early PPP Coverage

0.104∗∗
(0.036)

Industry Index

-0.009∗∗
(0.003)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,588
1.2
86.5
[ -0.056, 0.014]
0.267

Yes
2,588
0.2
51.2
[ -0.078, -0.012]
0.019

Yes
2,588
33.7
9.8
[ -0.154, 0.001]
0.055

Yes
2,586
244.6
11.4
[ -0.117, -0.003]
0.040

Yes
2,586
270.4
17.5
[ -0.050, 0.007]
0.090

Yes
1,578
377.2
14.9
[ -0.058, 0.007]
0.092

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-04-11.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

71


TABLE A.70
Initial Claims, Firms Size 0-99 (Week Ending 2020-04-18)
(1)

(2)

(3)

(4)

(5)

(6)

-0.014
(0.012)

-0.040∗∗∗
(0.010)

-0.053∗
(0.022)

-0.045∗∗
(0.017)

-0.026∗∗
(0.008)

-0.030∗∗∗
(0.009)

February IUR

1.523∗∗∗
(0.443)

1.599∗∗∗
(0.351)

1.812∗∗∗
(0.278)

1.654∗∗∗
(0.302)

Log(Med. Income)

-0.008
(0.004)

-0.006
(0.004)

-0.006∗
(0.002)

-0.005
(0.003)

Poverty Rate

-0.000∗
(0.000)

-0.000
(0.000)

-0.000∗∗
(0.000)

-0.000
(0.000)

Log(Pop. Density)

-0.000
(0.001)

0.000
(0.001)

0.000
(0.000)

-0.000
(0.001)

Covid Cases, 1w

-1.120
(0.572)

-0.774
(0.427)

-1.016
(0.520)

Covid Cases, 4w

0.204
(0.202)

0.204
(0.160)

0.217
(0.176)

Covid Deaths, 1w

1.812
(7.319)

7.965
(6.650)

10.916
(6.851)

Covid Deaths, 4w

-0.613
(2.478)

-3.883
(2.316)

-3.779
(2.069)

WFH Index

0.007
(0.022)

0.033
(0.021)

0.010
(0.018)

Early PPP Coverage

0.094∗∗
(0.029)

Industry Index

-0.008∗∗
(0.003)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,588
1.3
86.5
[ -0.042, 0.011]
0.249

Yes
2,588
0.3
51.2
[ -0.071, -0.018]
0.008

Yes
2,588
14.3
9.8
[ -0.143, -0.012]
0.021

Yes
2,586
162.2
11.2
[ -0.111, -0.014]
0.013

Yes
2,586
186.2
17.1
[ -0.046, -0.005]
0.029

Yes
1,579
228.6
14.6
[ -0.057, -0.010]
0.014

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-04-18.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

72


TABLE A.71
Initial Claims, Firms Size 0-99 (Week Ending 2020-04-25)
(1)

(2)

(3)

(4)

(5)

(6)

-0.010
(0.010)

-0.029∗∗
(0.010)

-0.038∗
(0.019)

-0.032∗
(0.015)

-0.020∗∗
(0.008)

-0.021∗
(0.009)

February IUR

0.995∗∗
(0.337)

1.079∗∗∗
(0.260)

1.198∗∗∗
(0.194)

1.050∗∗∗
(0.232)

Log(Med. Income)

-0.005
(0.003)

-0.003
(0.003)

-0.003
(0.002)

-0.002
(0.002)

Poverty Rate

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

0.000
(0.001)

0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

-0.181
(0.169)

-0.134
(0.140)

-0.109
(0.176)

Covid Cases, 4w

0.060
(0.085)

0.091
(0.083)

0.083
(0.079)

Covid Deaths, 1w

-14.972∗
(6.189)

-11.027∗
(4.314)

-14.458∗∗
(4.934)

Covid Deaths, 4w

2.690
(1.562)

1.173
(0.881)

2.273∗
(1.102)

WFH Index

0.006
(0.017)

0.022
(0.019)

0.008
(0.015)

Early PPP Coverage

0.058∗
(0.024)

Industry Index

-0.005∗∗
(0.002)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,588
1.1
86.5
[ -0.034, 0.009]
0.285

Yes
2,588
0.2
51.2
[ -0.058, -0.009]
0.013

Yes
2,588
4.9
9.8
[ -0.118, -0.002]
0.040

Yes
2,586
457.9
11.4
[ -0.092, -0.004]
0.029

Yes
2,586
424.9
17.3
[ -0.044, -0.003]
0.027

Yes
1,583
408.0
14.8
[ -0.051, -0.003]
0.029

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-04-25.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

73


TABLE A.72
Initial Claims, Firms Size 0-99 (Week Ending 2020-05-02)
(1)

(2)

(3)

(4)

(5)

(6)

-0.006
(0.009)

-0.030∗∗
(0.011)

-0.038∗
(0.019)

-0.033∗
(0.015)

-0.022∗∗
(0.008)

-0.027∗
(0.011)

February IUR

0.481
(0.347)

0.560∗
(0.258)

0.681∗∗∗
(0.164)

0.539∗
(0.266)

Log(Med. Income)

-0.005
(0.004)

-0.004
(0.003)

-0.003
(0.003)

-0.004
(0.003)

Poverty Rate

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Log(Pop. Density)

-0.000
(0.001)

0.000
(0.001)

0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.074
(0.115)

0.026
(0.089)

0.101
(0.217)

Covid Cases, 4w

-0.090
(0.054)

-0.036
(0.044)

-0.036
(0.045)

Covid Deaths, 1w

-5.955
(5.213)

-0.968
(4.879)

-9.485
(6.816)

Covid Deaths, 4w

0.521
(0.714)

-0.684
(0.618)

0.567
(0.748)

WFH Index

0.006
(0.016)

0.021
(0.016)

0.010
(0.015)

Early PPP Coverage

0.055∗∗
(0.021)

Industry Index

-0.004∗∗
(0.002)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,588
0.4
86.5
[ -0.026, 0.013]
0.532

Yes
2,588
0.2
51.2
[ -0.064, -0.009]
0.014

Yes
2,588
18.3
9.8
[ -0.123, -0.006]
0.026

Yes
2,586
24.8
11.1
[ -0.097, -0.008]
0.015

Yes
2,586
31.2
16.9
[ -0.048, -0.007]
0.013

Yes
1,584
46.1
14.2
[ -0.070, -0.008]
0.010

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-05-02.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

74


TABLE A.73
Initial Claims, Firms Size 0-99 (Week Ending 2020-05-09)
(1)

(2)

(3)

(4)

(5)

(6)

-0.008
(0.008)

-0.022∗
(0.009)

-0.030
(0.016)

-0.024
(0.012)

-0.016∗
(0.007)

-0.018∗
(0.008)

February IUR

0.326
(0.255)

0.422∗
(0.187)

0.495∗∗
(0.157)

0.402∗
(0.176)

Log(Med. Income)

-0.003
(0.003)

-0.002
(0.002)

-0.002
(0.002)

-0.002
(0.002)

Poverty Rate

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.108
(0.242)

0.039
(0.209)

0.081
(0.281)

Covid Cases, 4w

-0.081
(0.074)

-0.034
(0.051)

-0.065
(0.077)

Covid Deaths, 1w

0.882
(2.062)

1.778
(1.629)

1.832
(1.597)

Covid Deaths, 4w

-0.438
(0.460)

-1.078∗∗
(0.366)

-0.610
(0.390)

WFH Index

0.011
(0.013)

0.022
(0.016)

0.014
(0.012)

Early PPP Coverage

Industry Index

0.040
(0.022)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.003
(0.002)
No
2,588
1.0
86.5
[ -0.027, 0.008]
0.301

Yes
2,588
0.1
51.2
[ -0.048, -0.003]
0.030

Yes
2,588
18.9
9.8
[ -0.099, 0.002]
0.060

Yes
2,586
21.2
12.0
[ -0.071, -0.000]
0.048

Yes
2,586
22.9
17.7
[ -0.039, -0.002]
0.035

Yes
1,583
29.3
14.9
[ -0.048, -0.003]
0.023

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-05-09.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

75


TABLE A.74
Initial Claims, Firms Size 0-99 (Week Ending 2020-05-16)
(1)

(2)

(3)

(4)

(5)

(6)

-0.004
(0.006)

-0.015∗∗
(0.006)

-0.019
(0.010)

-0.015∗
(0.007)

-0.010∗
(0.004)

-0.012∗
(0.005)

February IUR

0.335∗
(0.164)

0.405∗∗∗
(0.120)

0.452∗∗∗
(0.113)

0.433∗∗∗
(0.129)

Log(Med. Income)

-0.001
(0.002)

-0.000
(0.002)

-0.000
(0.001)

-0.000
(0.001)

Poverty Rate

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.089
(0.184)

0.031
(0.121)

0.093
(0.216)

Covid Cases, 4w

-0.034
(0.037)

-0.010
(0.026)

-0.024
(0.045)

Covid Deaths, 1w

2.511
(2.477)

3.915
(2.111)

1.777
(2.572)

Covid Deaths, 4w

-0.966∗
(0.463)

-1.422∗∗
(0.478)

-0.976∗
(0.419)

WFH Index

0.006
(0.009)

0.014
(0.012)

0.008
(0.009)

Early PPP Coverage

Industry Index

0.026
(0.015)
-0.002∗
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

No
2,588
0.5
86.5
[ -0.017, 0.008]
0.468

Yes
2,588
0.2
51.2
[ -0.032, -0.003]
0.024

Yes
2,588
33.9
9.8
[ -0.062, 0.001]
0.063

Yes
2,586
12.3
11.7
[ -0.043, -0.000]
0.046

Yes
2,586
12.6
17.2
[ -0.022, -0.001]
0.038

Yes
1,584
19.2
14.8
[ -0.029, -0.002]
0.022

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-05-16.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

76


TABLE A.75
Initial Claims, Firms Size 0-99 (Week Ending 2020-05-23)
(1)

(2)

(3)

(4)

(5)

(6)

0.000
(0.006)

-0.012∗
(0.006)

-0.015
(0.009)

-0.013
(0.007)

-0.009∗
(0.004)

-0.010∗
(0.005)

February IUR

0.198
(0.169)

0.236
(0.133)

0.273∗
(0.109)

0.209
(0.135)

Log(Med. Income)

-0.001
(0.002)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

Poverty Rate

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

-0.383
(0.364)

-0.267
(0.284)

-0.431
(0.376)

Covid Cases, 4w

0.020
(0.038)

0.014
(0.031)

0.058
(0.051)

Covid Deaths, 1w

8.629
(5.953)

8.454
(5.687)

7.656
(5.546)

Covid Deaths, 4w

-2.174
(1.351)

-2.401
(1.401)

-2.352
(1.383)

WFH Index

0.009
(0.009)

0.015
(0.012)

0.010
(0.009)

Early PPP Coverage

Industry Index

0.021
(0.013)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.002
(0.001)
No
2,588
0.0
86.5
[ -0.012, 0.012]
0.959

Yes
2,588
0.1
51.2
[ -0.030, -0.000]
0.049

Yes
2,588
10.9
9.8
[ -0.055, 0.002]
0.079

Yes
2,586
21.4
11.8
[ -0.040, 0.000]
0.054

Yes
2,586
25.1
16.8
[ -0.023, -0.001]
0.033

Yes
1,583
34.1
15.4
[ -0.027, -0.001]
0.032

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-05-23.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

77


TABLE A.76
Initial Claims, Firms Size 0-99 (Week Ending 2020-05-30)
(1)

(2)

(3)

(4)

(5)

(6)

0.002
(0.003)

-0.008∗
(0.003)

-0.009
(0.006)

-0.008
(0.004)

-0.006∗
(0.003)

-0.006
(0.003)

February IUR

0.282∗∗
(0.107)

0.309∗∗∗
(0.092)

0.324∗∗∗
(0.093)

0.306∗∗
(0.100)

Log(Med. Income)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

Poverty Rate

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.066
(0.073)

0.091
(0.072)

0.043
(0.073)

Covid Cases, 4w

-0.039
(0.038)

-0.037
(0.032)

-0.022
(0.038)

Covid Deaths, 1w

5.167∗
(2.631)

5.269∗
(2.578)

5.164
(2.699)

Covid Deaths, 4w

-1.339∗
(0.555)

-1.503∗
(0.595)

-1.393∗
(0.603)

WFH Index

0.004
(0.005)

0.007
(0.006)

0.005
(0.004)

Early PPP Coverage

Industry Index

0.009
(0.007)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.001
(0.000)
No
2,588
0.3
86.5
[ -0.006, 0.009]
0.599

Yes
2,588
0.1
51.2
[ -0.018, -0.001]
0.028

Yes
2,588
27.4
9.8
[ -0.034, 0.001]
0.072

Yes
2,586
3.3
12.5
[ -0.026, 0.000]
0.055

Yes
2,586
49.0
17.2
[ -0.017, -0.001]
0.033

Yes
1,566
4.9
16.1
[ -0.018, -0.000]
0.050

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-05-30.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

78


TABLE A.77
Initial Claims, Firms Size 0-99 (Week Ending 2020-06-06)
(1)

(2)

(3)

(4)

(5)

(6)

0.002
(0.004)

-0.007
(0.004)

-0.007
(0.006)

-0.006
(0.005)

-0.005
(0.003)

-0.005
(0.004)

February IUR

0.270∗
(0.126)

0.289∗
(0.116)

0.299∗∗
(0.115)

0.299∗
(0.150)

Log(Med. Income)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

Poverty Rate

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Covid Cases, 1w

0.254
(0.195)

0.238
(0.184)

0.089
(0.127)

Covid Cases, 4w

-0.104
(0.085)

-0.095
(0.077)

-0.048
(0.057)

Covid Deaths, 1w

3.830
(2.278)

4.088
(2.306)

4.669
(2.948)

Covid Deaths, 4w

-0.832
(0.526)

-1.000
(0.713)

-1.225
(0.776)

WFH Index

0.006
(0.005)

0.008
(0.007)

0.007
(0.005)

Early PPP Coverage

Industry Index

0.007
(0.007)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
0.2
86.5
[ -0.007, 0.010]
0.652

Yes
2,588
0.1
51.2
[ -0.020, 0.000]
0.057

Yes
2,588
42.5
9.8
[ -0.032, 0.004]
0.179

Yes
2,586
58.4
12.9
[ -0.024, 0.003]
0.149

Yes
2,586
61.4
17.8
[ -0.017, 0.002]
0.125

Yes
1,583
84.5
16.5
[ -0.018, 0.003]
0.198

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-06-06.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

79


TABLE A.78
Initial Claims, Firms Size 0-99 (Week Ending 2020-06-13)
(1)

(2)

(3)

(4)

(5)

(6)

0.002
(0.004)

-0.008
(0.004)

-0.007
(0.005)

-0.006
(0.004)

-0.005
(0.003)

-0.006
(0.004)

February IUR

0.293∗∗
(0.107)

0.302∗∗
(0.098)

0.314∗∗
(0.096)

0.270∗
(0.111)

Log(Med. Income)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

Poverty Rate

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.221
(0.232)

0.200
(0.219)

0.308
(0.328)

Covid Cases, 4w

-0.078
(0.075)

-0.066
(0.067)

-0.107
(0.094)

Covid Deaths, 1w

0.479
(1.533)

0.611
(1.438)

0.008
(2.479)

Covid Deaths, 4w

-0.249
(0.488)

-0.427
(0.651)

0.035
(0.660)

WFH Index

0.006
(0.005)

0.008
(0.007)

0.008
(0.006)

Early PPP Coverage

Industry Index

0.007
(0.007)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
0.3
86.5
[ -0.007, 0.011]
0.575

Yes
2,588
0.1
51.2
[ -0.021, 0.000]
0.059

Yes
2,588
34.2
9.8
[ -0.031, 0.002]
0.120

Yes
2,586
44.0
12.2
[ -0.025, 0.001]
0.084

Yes
2,586
47.1
16.5
[ -0.017, 0.001]
0.080

Yes
1,582
51.6
15.6
[ -0.022, 0.001]
0.089

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-06-13.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

80


TABLE A.79
Initial Claims, Firms Size 0-99 (Week Ending 2020-06-20)
(1)

(2)

(3)

(4)

(5)

(6)

0.002
(0.002)

-0.005
(0.003)

-0.005
(0.004)

-0.005
(0.004)

-0.004
(0.003)

-0.005
(0.003)

February IUR

0.204∗
(0.091)

0.216∗
(0.085)

0.223∗∗
(0.082)

0.207∗
(0.101)

Log(Med. Income)

-0.000
(0.001)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

Poverty Rate

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

-0.000
(0.000)

0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.287
(0.175)

0.261
(0.156)

0.269
(0.165)

Covid Cases, 4w

-0.040
(0.034)

-0.033
(0.029)

-0.039
(0.035)

Covid Deaths, 1w

-2.170
(1.918)

-1.973
(1.733)

-3.728
(2.570)

Covid Deaths, 4w

-0.157
(0.447)

-0.260
(0.510)

0.118
(0.533)

WFH Index

0.003
(0.003)

0.004
(0.004)

0.004
(0.003)

Early PPP Coverage

Industry Index

0.005
(0.005)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
0.4
86.5
[ -0.004, 0.007]
0.533

Yes
2,588
0.1
51.2
[ -0.014, 0.000]
0.053

Yes
2,588
17.3
9.8
[ -0.024, 0.002]
0.122

Yes
2,586
23.5
12.3
[ -0.021, 0.001]
0.100

Yes
2,586
28.0
16.4
[ -0.015, 0.001]
0.094

Yes
1,584
26.1
16.2
[ -0.019, 0.001]
0.108

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-06-20.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

81


TABLE A.80
Initial Claims, Firms Size 0-99 (Week Ending 2020-06-27)
(1)

(2)

(3)

(4)

(5)

(6)

0.001
(0.002)

-0.004∗
(0.002)

-0.006
(0.004)

-0.005
(0.003)

-0.004∗
(0.002)

-0.005
(0.003)

February IUR

0.216∗∗∗
(0.064)

0.216∗∗∗
(0.059)

0.223∗∗∗
(0.059)

0.209∗∗
(0.068)

Log(Med. Income)

-0.000
(0.001)

-0.000
(0.001)

-0.000
(0.001)

-0.000
(0.001)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.360∗∗
(0.116)

0.338∗∗∗
(0.101)

0.387∗∗
(0.119)

Covid Cases, 4w

-0.073
(0.038)

-0.069∗
(0.033)

-0.091∗
(0.045)

Covid Deaths, 1w

-1.655
(1.396)

-1.126
(1.127)

-1.752
(1.571)

Covid Deaths, 4w

-0.066
(0.400)

-0.167
(0.480)

0.199
(0.491)

WFH Index

0.003
(0.003)

0.005
(0.004)

0.004
(0.003)

Early PPP Coverage

Industry Index

0.005
(0.004)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
0.2
86.5
[ -0.003, 0.006]
0.639

Yes
2,588
0.1
51.2
[ -0.009, -0.000]
0.038

Yes
2,588
0.5
9.8
[ -0.021, 0.001]
0.084

Yes
2,586
34.3
13.2
[ -0.016, 0.001]
0.069

Yes
2,586
33.7
17.1
[ -0.012, 0.000]
0.067

Yes
1,585
43.4
17.3
[ -0.014, 0.001]
0.088

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-06-27.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

82


TABLE A.81
Initial Claims, Firms Size 0-99 (Week Ending 2020-07-04)
(1)

(2)

(3)

(4)

(5)

(6)

0.000
(0.002)

-0.003∗∗
(0.001)

-0.004
(0.002)

-0.002
(0.002)

-0.002
(0.001)

-0.002
(0.002)

February IUR

0.211∗∗
(0.066)

0.237∗∗∗
(0.064)

0.240∗∗∗
(0.066)

0.253∗∗
(0.083)

Log(Med. Income)

0.000
(0.001)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Covid Cases, 1w

0.264∗
(0.113)

0.252∗∗
(0.097)

0.257∗
(0.118)

Covid Cases, 4w

-0.032
(0.031)

-0.030
(0.029)

-0.033
(0.036)

Covid Deaths, 1w

-1.730∗∗
(0.653)

-2.012∗
(0.804)

-2.376∗∗
(0.808)

Covid Deaths, 4w

-0.191
(0.564)

-0.100
(0.527)

0.056
(0.637)

WFH Index

0.001
(0.002)

0.002
(0.003)

0.002
(0.002)

Early PPP Coverage

Industry Index

0.002
(0.003)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
0.0
86.5
[ -0.003, 0.004]
0.911

Yes
2,588
0.2
51.2
[ -0.006, -0.001]
0.029

Yes
2,588
4.5
9.8
[ -0.010, 0.004]
0.201

Yes
2,586
24.4
12.7
[ -0.005, 0.004]
0.261

Yes
2,586
24.4
17.5
[ -0.004, 0.004]
0.272

Yes
1,556
35.2
16.3
[ -0.004, 0.005]
0.430

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-07-04.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

83


TABLE A.82
Initial Claims, Firms Size 0-99 (Week Ending 2020-07-11)
(1)

(2)

(3)

(4)

(5)

(6)

0.001
(0.002)

-0.003∗
(0.001)

-0.003
(0.003)

-0.001
(0.001)

-0.001
(0.001)

-0.000
(0.001)

February IUR

0.285∗∗∗
(0.078)

0.311∗∗∗
(0.084)

0.313∗∗∗
(0.086)

0.308∗∗
(0.094)

Log(Med. Income)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Covid Cases, 1w

-0.054
(0.078)

-0.056
(0.080)

-0.089
(0.098)

Covid Cases, 4w

0.074
(0.046)

0.072
(0.044)

0.081
(0.051)

Covid Deaths, 1w

1.925
(2.385)

2.135
(2.248)

1.846
(3.243)

Covid Deaths, 4w

-1.406∗
(0.662)

-1.426∗
(0.726)

-1.461∗
(0.721)

WFH Index

0.003
(0.003)

0.004
(0.004)

0.004
(0.003)

Early PPP Coverage

Industry Index

0.002
(0.003)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
0.2
86.5
[ -0.004, 0.005]
0.623

Yes
2,588
0.1
51.2
[ -0.007, -0.000]
0.044

Yes
2,588
57.1
9.8
[ -0.012, 0.003]
0.296

Yes
2,586
125.4
14.7
[ -0.005, 0.002]
0.349

Yes
2,586
120.2
19.9
[ -0.004, 0.002]
0.346

Yes
1,583
2.3
19.1
[ -0.003, 0.003]
0.757

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-07-11.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

84


TABLE A.83
Initial Claims, Firms Size 0-99 (Week Ending 2020-07-18)
(1)

(2)

(3)

(4)

(5)

(6)

0.001
(0.001)

-0.002∗
(0.001)

-0.003
(0.002)

-0.002
(0.001)

-0.002
(0.001)

-0.001
(0.001)

February IUR

0.281∗∗∗
(0.060)

0.301∗∗∗
(0.058)

0.302∗∗∗
(0.059)

0.322∗∗∗
(0.066)

Log(Med. Income)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.071
(0.046)

0.072
(0.045)

0.071
(0.055)

Covid Cases, 4w

0.025
(0.030)

0.024
(0.027)

0.023
(0.031)

Covid Deaths, 1w

3.487∗
(1.509)

3.574∗
(1.644)

3.981∗
(1.988)

Covid Deaths, 4w

-1.243∗
(0.561)

-1.249∗
(0.588)

-1.290∗
(0.650)

WFH Index

0.002
(0.002)

0.002
(0.003)

0.003
(0.002)

Early PPP Coverage

Industry Index

0.001
(0.002)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
0.2
86.5
[ -0.003, 0.004]
0.645

Yes
2,588
0.1
51.2
[ -0.006, -0.000]
0.049

Yes
2,588
3.2
9.8
[ -0.011, 0.002]
0.189

Yes
2,586
10.1
14.9
[ -0.005, 0.002]
0.229

Yes
2,586
9.4
19.7
[ -0.004, 0.001]
0.178

Yes
1,585
10.8
18.7
[ -0.004, 0.002]
0.252

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-07-18.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

85


TABLE A.84
Initial Claims, Firms Size 0-99 (Week Ending 2020-07-25)
(1)

(2)

(3)

(4)

(5)

(6)

0.000
(0.001)

-0.001
(0.001)

-0.002
(0.002)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

February IUR

0.200∗∗∗
(0.046)

0.213∗∗∗
(0.048)

0.214∗∗∗
(0.050)

0.231∗∗∗
(0.060)

Log(Med. Income)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.128∗∗∗
(0.034)

0.131∗∗∗
(0.038)

0.136∗∗
(0.046)

Covid Cases, 4w

-0.001
(0.017)

-0.004
(0.014)

-0.002
(0.016)

Covid Deaths, 1w

0.869
(0.730)

1.000
(0.648)

1.227
(0.887)

Covid Deaths, 4w

-0.613
(0.423)

-0.638
(0.481)

-0.763
(0.429)

WFH Index

0.001
(0.001)

0.002
(0.002)

0.002
(0.001)

Early PPP Coverage

Industry Index

0.001
(0.002)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
0.2
86.5
[ -0.002, 0.003]
0.658

Yes
2,588
0.1
51.2
[ -0.003, 0.001]
0.136

Yes
2,588
200.9
9.8
[ -0.007, 0.002]
0.255

Yes
2,586
517.5
16.3
[ -0.004, 0.002]
0.316

Yes
2,586
708.3
21.8
[ -0.003, 0.001]
0.265

Yes
1,585
953.0
20.9
[ -0.002, 0.002]
0.482

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-07-25.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

86


TABLE A.85
Initial Claims, Firms Size 0-99 (Week Ending 2020-08-01)
(1)

(2)

(3)

(4)

(5)

(6)

0.000
(0.001)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

-0.000
(0.001)

February IUR

0.155∗∗∗
(0.045)

0.160∗∗∗
(0.046)

0.160∗∗∗
(0.047)

0.164∗∗
(0.051)

Log(Med. Income)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.035
(0.024)

0.036
(0.024)

0.058
(0.036)

Covid Cases, 4w

0.019
(0.014)

0.018
(0.012)

0.013
(0.014)

Covid Deaths, 1w

0.257
(0.395)

0.257
(0.397)

0.357
(0.506)

Covid Deaths, 4w

-0.050
(0.508)

-0.035
(0.463)

-0.063
(0.565)

WFH Index

0.002
(0.001)

0.002
(0.002)

0.002
(0.002)

Early PPP Coverage

Industry Index

0.000
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

0.000
(0.000)
No
2,588
0.0
86.5
[ -0.002, 0.002]
0.911

Yes
2,588
0.1
51.2
[ -0.003, 0.000]
0.111

Yes
2,588
41.2
9.8
[ -0.005, 0.001]
0.290

Yes
2,586
103.6
17.1
[ -0.003, 0.001]
0.400

Yes
2,586
107.1
22.1
[ -0.002, 0.001]
0.356

Yes
1,583
192.2
22.3
[ -0.002, 0.001]
0.628

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-08-01.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

87


TABLE A.86
Initial Claims, Firms Size 0-99 (Week Ending 2020-08-08)
(1)

(2)

(3)

(4)

(5)

(6)

-0.000
(0.001)

-0.001∗
(0.000)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

February IUR

0.125∗∗
(0.038)

0.129∗∗∗
(0.036)

0.130∗∗∗
(0.036)

0.123∗∗
(0.038)

Log(Med. Income)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

-0.037
(0.027)

-0.034
(0.023)

-0.046
(0.027)

Covid Cases, 4w

0.023∗
(0.010)

0.021∗∗
(0.008)

0.026∗
(0.010)

Covid Deaths, 1w

-0.562∗
(0.266)

-0.620∗∗
(0.235)

-0.581∗
(0.290)

Covid Deaths, 4w

0.281
(0.229)

0.316
(0.178)

0.265
(0.244)

WFH Index

0.002
(0.001)

0.002
(0.001)

0.002
(0.001)

Early PPP Coverage

Industry Index

0.001
(0.002)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
0.0
86.5
[ -0.002, 0.001]
0.857

Yes
2,588
0.1
51.2
[ -0.003, -0.000]
0.037

Yes
2,588
15.4
9.8
[ -0.005, 0.001]
0.151

Yes
2,586
31.8
15.2
[ -0.003, 0.001]
0.238

Yes
2,586
30.1
20.2
[ -0.002, 0.001]
0.192

Yes
1,583
37.5
20.4
[ -0.002, 0.001]
0.297

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-08-08.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

88


TABLE A.87
Initial Claims, Firms Size 0-99 (Week Ending 2020-08-15)
(1)

(2)

(3)

(4)

(5)

(6)

0.000
(0.001)

-0.001
(0.000)

-0.001
(0.001)

-0.000
(0.001)

-0.000
(0.000)

0.000
(0.000)

February IUR

0.165∗∗∗
(0.038)

0.164∗∗∗
(0.039)

0.163∗∗∗
(0.040)

0.168∗∗∗
(0.045)

Log(Med. Income)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Covid Cases, 1w

-0.093
(0.066)

-0.093
(0.063)

-0.103
(0.069)

Covid Cases, 4w

0.047
(0.026)

0.048
(0.025)

0.052
(0.027)

Covid Deaths, 1w

0.847∗
(0.390)

0.842
(0.437)

0.897
(0.558)

Covid Deaths, 4w

-0.290
(0.285)

-0.289
(0.294)

-0.297
(0.310)

WFH Index

0.001
(0.001)

0.001
(0.002)

0.001
(0.001)

Early PPP Coverage

Industry Index

-0.000
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

0.000
(0.000)
No
2,588
0.0
86.5
[ -0.002, 0.001]
0.988

Yes
2,588
0.1
51.2
[ -0.002, 0.000]
0.106

Yes
2,588
1.3
9.8
[ -0.004, 0.001]
0.413

Yes
2,586
4.4
14.9
[ -0.001, 0.001]
0.708

Yes
2,586
6.0
20.2
[ -0.001, 0.001]
0.657

Yes
1,584
4.1
18.8
[ -0.001, 0.002]
0.877

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-08-15.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

89


TABLE A.88
Initial Claims, Firms Size 0-99 (Week Ending 2020-08-22)
(1)

(2)

(3)

(4)

(5)

(6)

0.000
(0.001)

-0.001
(0.000)

-0.000
(0.001)

-0.000
(0.001)

-0.000
(0.001)

0.000
(0.001)

February IUR

0.173∗∗
(0.054)

0.173∗∗
(0.053)

0.173∗∗
(0.053)

0.171∗∗
(0.062)

Log(Med. Income)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Covid Cases, 1w

-0.055∗
(0.025)

-0.054∗
(0.023)

-0.071∗∗
(0.023)

Covid Cases, 4w

0.027∗∗∗
(0.007)

0.027∗∗∗
(0.006)

0.031∗∗∗
(0.007)

Covid Deaths, 1w

-0.228
(0.406)

-0.194
(0.330)

-0.316
(0.439)

Covid Deaths, 4w

-0.076
(0.117)

-0.080
(0.129)

-0.049
(0.132)

WFH Index

0.001
(0.001)

0.001
(0.001)

0.001
(0.001)

Early PPP Coverage

Industry Index

0.000
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

0.000
(0.000)
No
2,588
0.2
86.5
[ -0.001, 0.001]
0.642

Yes
2,588
0.0
51.2
[ -0.002, 0.000]
0.145

Yes
2,588
53.3
9.8
[ -0.004, 0.002]
0.623

Yes
2,586
103.9
14.3
[ -0.002, 0.002]
0.750

Yes
2,586
102.8
19.8
[ -0.002, 0.002]
0.799

Yes
1,585
114.1
19.0
[ -0.002, 0.002]
0.953

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-08-22.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

90


TABLE A.89
Initial Claims, Firms Size 0-99 (Week Ending 2020-08-29)
(1)

(2)

(3)

(4)

(5)

(6)

0.000
(0.001)

-0.000
(0.000)

-0.000
(0.001)

0.000
(0.001)

-0.000
(0.001)

0.000
(0.001)

February IUR

0.145∗∗∗
(0.040)

0.143∗∗∗
(0.040)

0.142∗∗∗
(0.040)

0.147∗∗∗
(0.041)

Log(Med. Income)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Covid Cases, 1w

-0.034
(0.020)

-0.033
(0.021)

-0.047∗
(0.024)

Covid Cases, 4w

0.010
(0.006)

0.010
(0.006)

0.012
(0.008)

Covid Deaths, 1w

-0.128
(0.427)

-0.149
(0.411)

0.004
(0.348)

Covid Deaths, 4w

0.033
(0.158)

0.038
(0.152)

0.029
(0.190)

WFH Index

0.000
(0.001)

0.000
(0.001)

0.001
(0.001)

Early PPP Coverage

Industry Index

-0.000
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

0.000
(0.000)
No
2,588
0.4
86.5
[ -0.001, 0.002]
0.534

Yes
2,588
0.1
51.2
[ -0.001, 0.000]
0.178

Yes
2,588
93.0
9.8
[ -0.002, 0.002]
0.970

Yes
2,586
137.1
12.5
[ -0.001, 0.002]
0.907

Yes
2,586
138.8
17.4
[ -0.001, 0.002]
0.994

Yes
1,574
28.0
16.6
[ -0.001, 0.002]
0.737

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-08-29.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

91


TABLE A.90
Initial Claims, Firms Size 0-99 (Week Ending 2020-09-05)
(1)

(2)

(3)

(4)

(5)

(6)

0.000
(0.001)

-0.001
(0.000)

-0.000
(0.001)

-0.000
(0.001)

-0.000
(0.001)

0.000
(0.001)

February IUR

0.113∗∗
(0.037)

0.113∗∗
(0.037)

0.114∗∗
(0.038)

0.107∗
(0.042)

Log(Med. Income)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Covid Cases, 1w

0.003
(0.016)

-0.001
(0.016)

-0.015
(0.019)

Covid Cases, 4w

-0.010
(0.015)

-0.009
(0.015)

-0.012
(0.017)

Covid Deaths, 1w

0.224
(0.377)

0.190
(0.354)

0.507
(0.370)

Covid Deaths, 4w

0.312
(0.423)

0.335
(0.432)

0.401
(0.585)

WFH Index

0.001
(0.001)

0.001
(0.001)

0.001
(0.001)

Early PPP Coverage

Industry Index

0.001
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
0.2
86.5
[ -0.001, 0.002]
0.630

Yes
2,588
0.1
51.2
[ -0.002, 0.000]
0.171

Yes
2,588
5.2
9.8
[ -0.003, 0.002]
0.682

Yes
2,586
6.2
12.9
[ -0.003, 0.001]
0.622

Yes
2,586
6.5
17.2
[ -0.002, 0.001]
0.796

Yes
1,582
7.1
17.5
[ -0.001, 0.002]
0.850

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-09-05.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

92


TABLE A.91
Initial Claims, Firms Size 0-99 (Week Ending 2020-09-12)
(1)

(2)

(3)

(4)

(5)

(6)

0.001
(0.001)

-0.001∗
(0.000)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

February IUR

0.153∗∗∗
(0.034)

0.157∗∗∗
(0.032)

0.159∗∗∗
(0.032)

0.166∗∗∗
(0.036)

Log(Med. Income)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

-0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.041
(0.025)

0.031
(0.022)

0.051
(0.028)

Covid Cases, 4w

-0.010∗∗
(0.004)

-0.010∗∗
(0.004)

-0.018∗∗
(0.007)

Covid Deaths, 1w

-0.755
(0.590)

-0.637
(0.587)

-1.348
(0.959)

Covid Deaths, 4w

0.253
(0.285)

0.274
(0.288)

0.420
(0.394)

WFH Index

0.002
(0.001)

0.002∗
(0.001)

0.002∗
(0.001)

Early PPP Coverage

0.002∗
(0.001)

Industry Index

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
0.9
86.5
[ -0.001, 0.002]
0.339

Yes
2,588
0.1
51.2
[ -0.002, -0.000]
0.024

Yes
2,588
8.3
9.8
[ -0.006, 0.000]
0.102

Yes
2,586
9.4
12.4
[ -0.005, 0.000]
0.083

Yes
2,586
9.4
16.1
[ -0.004, 0.000]
0.120

Yes
1,547
16.9
16.9
[ -0.003, 0.000]
0.154

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-09-12.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

93


TABLE A.92
Initial Claims, Firms Size 0-99 (Week Ending 2020-09-19)
(1)

(2)

(3)

(4)

(5)

(6)

0.001
(0.001)

-0.001
(0.000)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

February IUR

0.147∗∗∗
(0.038)

0.152∗∗∗
(0.037)

0.153∗∗∗
(0.037)

0.150∗∗
(0.046)

Log(Med. Income)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

-0.008
(0.015)

-0.006
(0.016)

0.005
(0.016)

Covid Cases, 4w

-0.005
(0.008)

-0.007
(0.008)

-0.013
(0.010)

Covid Deaths, 1w

0.139
(0.389)

0.158
(0.376)

0.666
(0.584)

Covid Deaths, 4w

-0.005
(0.180)

0.026
(0.182)

-0.105
(0.229)

WFH Index

0.002
(0.001)

0.002
(0.001)

0.002
(0.001)

Early PPP Coverage

Industry Index

0.001
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
0.9
86.5
[ -0.001, 0.002]
0.341

Yes
2,588
0.1
51.2
[ -0.002, 0.000]
0.081

Yes
2,588
7.7
9.8
[ -0.006, 0.000]
0.121

Yes
2,586
9.7
12.2
[ -0.004, 0.000]
0.106

Yes
2,586
66.1
15.9
[ -0.003, 0.000]
0.145

Yes
1,583
13.2
16.0
[ -0.003, 0.000]
0.200

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-09-19.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

94


TABLE A.93
Initial Claims, Firms Size 0-99 (Week Ending 2020-09-26)
(1)

(2)

(3)

(4)

(5)

(6)

0.001
(0.001)

-0.001
(0.000)

-0.001
(0.001)

-0.001
(0.001)

-0.000
(0.000)

-0.000
(0.000)

February IUR

0.079∗∗
(0.030)

0.083∗∗
(0.029)

0.084∗∗
(0.030)

0.063∗
(0.026)

Log(Med. Income)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.005
(0.008)

0.009
(0.008)

0.013
(0.008)

Covid Cases, 4w

-0.003
(0.005)

-0.005
(0.005)

-0.008
(0.006)

Covid Deaths, 1w

-0.301
(0.419)

-0.333
(0.423)

-0.572
(0.462)

Covid Deaths, 4w

0.175
(0.110)

0.223
(0.117)

0.300∗
(0.130)

WFH Index

0.002
(0.001)

0.002
(0.001)

0.002
(0.001)

Early PPP Coverage

0.002∗
(0.001)

Industry Index

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
1.1
86.5
[ -0.001, 0.002]
0.304

Yes
2,588
0.1
51.2
[ -0.002, 0.000]
0.098

Yes
2,588
36.1
9.8
[ -0.005, 0.001]
0.286

Yes
2,586
32.6
12.4
[ -0.003, 0.000]
0.283

Yes
2,586
31.6
16.3
[ -0.002, 0.001]
0.472

Yes
1,582
51.3
16.3
[ -0.002, 0.001]
0.497

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-09-26.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

95


TABLE A.94
Initial Claims, Firms Size 0-99 (Week Ending 2020-10-03)
(1)

(2)

(3)

(4)

(5)

(6)

0.001
(0.001)

-0.001
(0.000)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.000)

-0.001
(0.001)

February IUR

0.080∗∗
(0.027)

0.084∗∗
(0.026)

0.085∗∗∗
(0.026)

0.068∗∗
(0.021)

Log(Med. Income)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.022∗
(0.011)

0.022
(0.011)

0.022
(0.013)

Covid Cases, 4w

-0.003
(0.004)

-0.003
(0.004)

-0.005
(0.004)

Covid Deaths, 1w

-0.337
(0.353)

-0.372
(0.347)

-0.259
(0.504)

Covid Deaths, 4w

0.056
(0.134)

0.087
(0.129)

0.117
(0.184)

WFH Index

0.001
(0.001)

0.002
(0.001)

0.002
(0.001)

Early PPP Coverage

Industry Index

0.001
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

0.000
(0.000)
No
2,588
1.6
86.5
[ -0.001, 0.002]
0.206

Yes
2,588
0.1
51.2
[ -0.002, 0.000]
0.067

Yes
2,588
116.6
9.8
[ -0.005, 0.000]
0.070

Yes
2,586
76.5
13.3
[ -0.003, 0.000]
0.051

Yes
2,586
77.2
18.0
[ -0.002, 0.000]
0.053

Yes
1,582
160.3
17.4
[ -0.003, 0.000]
0.125

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-10-03.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

96


TABLE A.95
Initial Claims, Firms Size 0-99 (Week Ending 2020-10-10)
(1)

(2)

(3)

(4)

(5)

(6)

0.002
(0.001)

-0.001∗
(0.001)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

February IUR

0.150∗∗∗
(0.042)

0.156∗∗∗
(0.040)

0.158∗∗∗
(0.040)

0.142∗∗∗
(0.040)

Log(Med. Income)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.002
(0.016)

0.000
(0.016)

-0.002
(0.021)

Covid Cases, 4w

0.006
(0.004)

0.005
(0.003)

0.005
(0.004)

Covid Deaths, 1w

-0.019
(0.350)

-0.015
(0.347)

-0.047
(0.516)

Covid Deaths, 4w

-0.044
(0.088)

-0.015
(0.086)

0.026
(0.108)

WFH Index

0.002
(0.001)

0.003∗
(0.001)

0.003∗
(0.001)

Early PPP Coverage

0.002∗
(0.001)

Industry Index

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
2.3
86.5
[ -0.001, 0.004]
0.132

Yes
2,588
0.1
51.2
[ -0.003, -0.000]
0.046

Yes
2,588
67.9
9.8
[ -0.007, 0.001]
0.201

Yes
2,586
100.1
12.8
[ -0.005, 0.000]
0.180

Yes
2,586
25.8
17.6
[ -0.003, 0.001]
0.291

Yes
1,578
151.4
16.5
[ -0.003, 0.001]
0.379

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-10-10.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

97


TABLE A.96
Initial Claims, Firms Size 0-99 (Week Ending 2020-10-17)
(1)

(2)

(3)

(4)

(5)

(6)

0.001
(0.001)

-0.001
(0.001)

-0.001
(0.001)

-0.000
(0.001)

-0.000
(0.001)

-0.000
(0.001)

February IUR

0.119∗∗∗
(0.035)

0.126∗∗∗
(0.032)

0.127∗∗∗
(0.032)

0.116∗∗∗
(0.032)

Log(Med. Income)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.036
(0.020)

0.034
(0.020)

0.051
(0.026)

Covid Cases, 4w

-0.002
(0.006)

-0.002
(0.006)

-0.005
(0.007)

Covid Deaths, 1w

-0.370
(0.359)

-0.339
(0.347)

-0.281
(0.520)

Covid Deaths, 4w

-0.092
(0.126)

-0.084
(0.124)

-0.191
(0.140)

WFH Index

0.002
(0.001)

0.002
(0.001)

0.002
(0.001)

Early PPP Coverage

Industry Index

0.001
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
2.2
86.5
[ -0.000, 0.003]
0.138

Yes
2,588
0.0
51.2
[ -0.003, 0.000]
0.186

Yes
2,588
94.2
9.8
[ -0.006, 0.001]
0.551

Yes
2,586
100.8
12.7
[ -0.003, 0.001]
0.592

Yes
2,586
201.9
17.4
[ -0.003, 0.001]
0.692

Yes
1,582
202.3
16.8
[ -0.003, 0.001]
0.750

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-10-17.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

98


TABLE A.97
Initial Claims, Firms Size 0-99 (Week Ending 2020-10-24)
(1)

(2)

(3)

(4)

(5)

(6)

0.002
(0.002)

-0.000
(0.000)

-0.000
(0.001)

-0.000
(0.001)

-0.000
(0.001)

-0.000
(0.001)

February IUR

0.053
(0.032)

0.056
(0.029)

0.056
(0.029)

0.040
(0.032)

Log(Med. Income)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.032∗∗
(0.012)

0.033∗∗
(0.012)

0.039∗∗
(0.014)

Covid Cases, 4w

0.004
(0.005)

0.003
(0.005)

0.001
(0.005)

Covid Deaths, 1w

0.528
(0.347)

0.528
(0.348)

0.684
(0.542)

Covid Deaths, 4w

-0.287∗
(0.126)

-0.278∗
(0.121)

-0.176
(0.184)

WFH Index

0.001
(0.001)

0.001
(0.001)

0.001
(0.001)

Early PPP Coverage

Industry Index

0.000
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
2.1
86.5
[ -0.001, 0.006]
0.143

Yes
2,588
0.0
51.2
[ -0.002, 0.001]
0.375

Yes
2,588
6.4
9.8
[ -0.004, 0.001]
0.673

Yes
2,586
10.6
12.8
[ -0.003, 0.001]
0.605

Yes
2,586
19.5
17.5
[ -0.003, 0.001]
0.671

Yes
1,582
11.3
16.7
[ -0.003, 0.002]
0.880

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-10-24.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

99


TABLE A.98
Initial Claims, Firms Size 0-99 (Week Ending 2020-10-31)
(1)

(2)

(3)

(4)

(5)

(6)

0.002
(0.001)

-0.000
(0.000)

-0.000
(0.001)

-0.000
(0.001)

-0.000
(0.001)

-0.000
(0.001)

February IUR

0.099
(0.055)

0.102
(0.054)

0.102
(0.054)

0.096
(0.066)

Log(Med. Income)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.020
(0.017)

0.020
(0.017)

0.030
(0.024)

Covid Cases, 4w

0.008
(0.004)

0.008
(0.004)

0.005
(0.006)

Covid Deaths, 1w

0.172
(0.184)

0.172
(0.184)

0.159
(0.369)

Covid Deaths, 4w

-0.259∗∗
(0.098)

-0.260∗∗
(0.099)

-0.134
(0.163)

WFH Index

0.001
(0.001)

0.001
(0.001)

0.001
(0.001)

Early PPP Coverage

Industry Index

-0.000
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
2.6
86.5
[ -0.000, 0.005]
0.105

Yes
2,588
0.0
51.2
[ -0.001, 0.001]
0.747

Yes
2,588
58.5
9.8
[ -0.003, 0.002]
0.823

Yes
2,586
49.3
13.0
[ -0.002, 0.002]
0.683

Yes
2,586
61.6
17.8
[ -0.002, 0.002]
0.667

Yes
1,574
62.3
16.4
[ -0.003, 0.002]
0.749

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-10-31.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

100


TABLE A.99
Initial Claims, Firms Size 0-99 (Week Ending 2020-11-07)
(1)

(2)

(3)

(4)

(5)

(6)

0.002
(0.001)

-0.000
(0.000)

-0.000
(0.001)

-0.001
(0.001)

-0.000
(0.001)

-0.000
(0.001)

February IUR

0.131∗∗∗
(0.032)

0.132∗∗∗
(0.031)

0.134∗∗∗
(0.030)

0.140∗∗∗
(0.036)

Log(Med. Income)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Poverty Rate

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

0.000
(0.000)

Log(Pop. Density)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

-0.000
(0.000)

Covid Cases, 1w

0.023∗∗
(0.008)

0.022∗∗
(0.008)

0.035∗∗
(0.012)

Covid Cases, 4w

0.007
(0.006)

0.007
(0.006)

0.005
(0.005)

Covid Deaths, 1w

-0.059
(0.164)

-0.073
(0.172)

0.202
(0.249)

Covid Deaths, 4w

-0.166
(0.116)

-0.140
(0.114)

-0.324
(0.183)

WFH Index

0.001
(0.001)

0.001
(0.001)

0.001
(0.001)

Early PPP Coverage

Industry Index

0.001
(0.001)

March Small-Firm Rev.

State FE
N
Wald F-Stat
K-P F-Stat
A-R 95% Conf. Set
A-R p-value

-0.000
(0.000)
No
2,588
2.8
86.5
[ -0.000, 0.005]
0.095

Yes
2,588
0.0
51.2
[ -0.001, 0.001]
0.687

Yes
2,588
11.4
9.8
[ -0.003, 0.001]
0.486

Yes
2,586
14.0
13.1
[ -0.003, 0.001]
0.212

Yes
2,586
24.4
17.8
[ -0.002, 0.001]
0.395

Yes
1,581
22.2
16.8
[ -0.002, 0.001]
0.484

Standard errors in parentheses, clustered at the state-level. Anderson-Rubin 95% confidence sets, listed at the bottom of the table, corresponds to the estimated coefficient for
Early PPP Coverage. The Anderson-Rubin p-value also corresponds to the estimated coefficient for Early PPP Coverage. K-P F-Stat stands for Kleibergen-Papp F-Statistic.
′
Table presents coefficients estimated from ycjt = β0,s(c)jt + βP P P,jt P P Pcjt′ + Xcjt
β1,jt + ϵcjt . The dependent variable is the fraction of the covered workforce with an approved
UI initial claim for the given week (only workers from firms sized 0-99). The primary varaible of interest (Early PPP Coverage) is the fraction of jobs at firms sized 0 - 99
which are covered by PPP loans as of April 11, 2020. This variable is instrumented by the county-level share of deposit funds in community banks. Details on measurements
are covered in the Data section in the main paper. Principal covariates: state-level fixed effects, February 2020 IUR (IUR in terms of initial claims in this case) for workers
from firms sized 0-99 (i.e. the dependent variable as measured immediately pre-pandemic), county-level measures of log(median income), the poverty rate, and log(population
density), from Census data. Also included are Covid cases and deaths (separately) over the prior week, and cumulatively over the prior four weeks, as collected by the New
York Times, via Opportunity Insights. The WFH Index for small firms stands for a ’Work From Home’ index. This is calculated as the inner product of (a) the Dingel &
Neiman (2020) measure of the share of industry-level jobs that can be done from home and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the
2-digit NAICS level. Industry Index is calculated as the inner product of (a) fraction of jobs lost in industry j from February 2020 - April 2020, as measured by BLS’ Current
Establishment Survey and (b) the employment-share of industry j in county c (for firms size 0 - 99), at the 2-digit NAICS level. March Small-Firm Revenue comes from Womply,
via Opportunity Insights. Regressions weighted by the pre-pandemic county-level employment at firms sized 0-99. Each week is estimated and presented seperately, with this
table presenting the week ending 2020-11-07.
∗

p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

101

File and source

File

treasury-job-preservation-ppp-2020-01b.pdf

Size

7,195,276 bytes

SHA-256

fdea5f8e3dad96b7455283602a1c121d6bf11cd171b6b0458ee5de63cc6d139a

Our copy

treasury-job-preservation-ppp-2020-01b.pdf

Original

home.treasury.gov