Brookings Papers on Economic Activity

Full text

335
GLENN HUBBARD
Columbia University
MICHAEL R. STRAIN
American Enterprise Institute
Has the Paycheck Protection
Program Succeeded?
ABSTRACT    Enacted March  27, 2020, the Paycheck Protection Program
(PPP) was the most ambitious and creative fiscal policy response to the
pandemic recession in the United States. PPP offers forgivable loans—essentially
grants—to businesses with 500 or fewer employees that meet certain require-
ments. In this paper, we present evidence that PPP has substantially increased
the employment, financial health, and survival of small businesses, using data
from Dun & Bradstreet, Inc. We use event studies and standard difference-
in-differences models to estimate the effect of a small business applying for
larger PPP loans and of a small business being eligible for PPP based on size.
While our findings are informative, we believe it is too early to issue conclusive
judgment on PPP’s success. We offer lessons for the future from the PPP
experience thus far.
T
he Paycheck Protection Program (PPP) was the most ambitious and
creative—and, potentially, the most important—fiscal policy response
to the pandemic recession in the United States. With a $670 billion budget
from April through August 2020, the program was the largest single
component of the nation’s fiscal policy response to the crisis during that
period, and by itself it approaches the total amount spent by Congress on
the American Recovery and Reinvestment Act of 2009 in response to the
Great Recession.
Conflict of Interest Disclosure: The authors did not receive financial support from any firm
or person for this paper or from any firm or person with a financial or political interest in this
paper. They are currently not officers, directors, or board members of any organization with
an interest in this paper. No outside party had the right to review this paper before publication.

336
Brookings Papers on Economic Activity, Fall 2020
PPP was enacted on March 27, 2020, as part of the Coronavirus Aid,
Relief, and Economic Security (CARES) Act, the $1.8 trillion “phase 3”
response to the pandemic crisis. An entirely new program, PPP began
issuing loans seven days later, on April 3. Lending under PPP continued
until August. PPP offers forgivable loans—essentially, grants—to businesses
with 500 or fewer employees that meet certain requirements, including
maintaining employment at prepandemic levels.
Has it succeeded? In this paper, we present evidence that PPP has
substantially increased the employment, financial health, and survival of
small businesses. In addition, we find that the effect of PPP on small busi-
ness outcomes is increasing over time, with larger effects in August than
in April or May. We also find some evidence to suggest that PPP was most
effective for relatively smaller firms. We use data from Dun & Bradstreet
for our analysis, employing standard difference-in-differences models to
estimate the effects of a small business applying for a PPP loan of greater
than $150,000 (we only observe PPP applications for loans of that size) and
of a small business being eligible for PPP based on size and using event
studies to trace the dynamic effects of PPP.
Despite this finding, our ultimate conclusion is that it is too early to issue
any definitive judgment on PPP’s success. The program had important short-
run goals, to be sure. These include supporting employment and replacing
worker wages, maintaining worker-firm attachments, boosting consumer
spending, and ensuring small business continuity during the shutdown. But
the program had important medium-run goals, as well, including preventing
a wave of bankruptcies once the economy partially reopened; increasing
productivity by preserving firm-specific human capital, worker-firm matches,
and networks; and helping the economy recover faster by keeping workers
off the unemployment rolls. Our data run through August, and we cannot
adequately investigate any of these outcomes. The effects of PPP are
unfolding, and it will be particularly important to see what happens to
businesses that received PPP and the workers they employ once they have
exhausted their forgivable loan.
PPP is a novel program, and many standard intuitions about fiscal
policy do not apply to it. It was not a stimulus program in the sense that
its purpose was not to stimulate the economy; that is, it is not a program
calling for a measure of the multiplier. Instead, its purpose was to preserve
the productive capacity of the small business sector and to shorten the tran-
sition to a new, post-pandemic equilibrium by supporting labor demand
over the medium term, allowing for a more rapid economic recovery. It was

HUBBARD and STRAIN
337
not a jobs program in the sense that its goal was not exclusively to preserve
employment. Instead, its goals were to maintain worker-firm attachments,
particularly during the shutdown, and to ensure small business continuity.
It intentionally did not attempt to exclude inframarginal recipients because
the unique circumstances under which it was enacted made this impracti-
cal. In the early days of the shutdown, how could the government have
known which firms were inframarginal? And given the numerous goals of
the program, it’s not clear how marginal would be defined in this context.
These design features affect intuitive measures of the cost per job saved,
as we describe later.
In this paper, we discuss the need for, goals of, and key design features
in a small business revenue replacement program (section II). We then
describe PPP and contrast select features of the program to what we view
as the best design (section III). We discuss the program’s implementa-
tion challenges—extensively covered in the press—and offer qualitative
analysis of PPP (section IV). In section V, we present our empirical analysis
of PPP. In section VI, we offer a retrospective and discuss lessons for
the future.
I.  The Pandemic Recession and Potential Policy Responses
The pandemic recession is remarkable in both its suddenness and depth.
In the week ending March 14, 2020, there were 282,000 initial claims
for unemployment insurance benefits, about one-third higher than the
average number of new claims over the preceding three months. The next
week, there were 3.3 million initial claims, shattering the previous record
of 695,000 new claims set in October 1982. The week after that, ending
March 28, there were 6.9 million initial claims.
The unemployment rate in February 2020 was 3.5 percent. In March,
the first month of the pandemic recession, it stood at 4.4 percent. In April,
it hit its peak of 14.7 percent, the highest rate since the Great Depression.1
In two months, the official unemployment rate increased by a factor of four.
1.  The official unemployment rate reported by the Bureau of Labor Statistics for April
2020 was 14.7 percent. The household survey on which the unemployment rate is calculated
showed a large increase in the number of respondents who were classified as employed but
absent from work. Most of these responses should have been classified as unemployed on
temporary layoff. Incorporating this change, the actual unemployment rate for April was
likely 19.5 percent.

338
Brookings Papers on Economic Activity, Fall 2020
For comparison, during the Great Recession it took nearly two years for the
unemployment rate to double, from 5 percent when the recession began in
December 2007 to its peak of 10 percent in October 2009.2
The pandemic’s economic devastation extended beyond the labor market.
Real GDP contracted at a 31.4 percent annual rate in the second quarter of
2020. Using the same measure, the worst quarter in the Great Recession
saw an 8.4 percent decline, and the only quarter since the Great Depression
to register a double-digit contraction was 1958:Q1, at 10 percent. Relative
to the same quarter one year prior, 2020:Q2 real GDP contracted by
9 percent. The peak contraction using this metric in the Great Recession
was 2009:Q2 at 3.9 percent.
Some of the ways policy needed to respond to this unprecedented eco-
nomic crisis were relatively straightforward. The Federal Reserve needed
to support the economy and to ensure liquidity and smooth functioning in
financial markets. Social insurance and safety net programs needed to be
strengthened, and their gaps needed to be plugged. Large businesses, with
diversified revenue streams and access to capital markets, could be supported
with lending programs.
But policy to support small and midsize businesses was harder to
formulate. The need for a prolonged shutdown made interruption loans
for such businesses inadequate, and even with a more conventional loan
many businesses would likely not be able to survive. Firms needed more
equity to shore up weakening balance sheets and replace lost cash flows
and many businesses would not be interested in adding to debt burdens in
any case. Equity injections were not implementable for many firms of this
size, and operationalizing a program based on them would be extremely
difficult to do in the time needed. The best available option was a revenue
replacement program for small business.
II.  A Small Business Revenue Replacement Program
The pandemic recession created the need for a revenue replacement
program for small businesses. In this section, we discuss that need.3 We
argue that the goals of such a program should be twofold: to ensure small
business continuity and prevent a cascade of small business failures, and to
2.  For research on the labor market effects of the pandemic, see Bartik, Bertrand, and
others (2020), Coibion, Gorodnichenko, and Weber (2020), Goolsbee and Syverson (2020),
and Forsythe and others (2020).
3.  This section draws on Hubbard and Strain (2020) and Strain (2020).

HUBBARD and STRAIN
339
preserve existing employment relationships while shelter-in-place orders
are in effect. We offer our view on some key program design features to
achieve these goals. We also address moral hazard concerns and briefly
review programs enacted by other major economies.
II.A.  The Need to Replace Small Business Revenue
The pandemic itself can be thought of as a large shock to aggregate
supply: businesses could no longer produce goods and services because
workers could not safely go to work. The inability of workers to work
caused downstream supply chain disruptions, as well.
Shelter-in-place orders ameliorated the supply shock by reducing the
spread of the coronavirus. The catch is that these policies led to a precipitous
drop in aggregate demand, including labor demand (Forsythe and others
2020) as businesses were temporarily closed and workers lost jobs, faced
reductions in hours, and experienced nominal wage cuts (Cajner and others
2020). In the private economy, workers faced a large reduction in earned
income and businesses lost revenue.
The sharp and sudden nature of the pandemic recession left smaller
firms in the service sector particularly at risk. Unlike larger businesses,
these firms could not readily access capital markets to shore up their balance
sheets. Capital market imperfections link equity contractions to business
fluctuations, and these firms were particularly vulnerable to a lack of
collateralizable net worth (Gertler and Hubbard 1988). Small and midsize
businesses generally do not have diversified revenue streams, as well. And
they have limited cash holdings. Only half of small businesses hold cash
reserves sufficient to cover fifteen days, and only four in ten have a three-
week cash buffer (Farrell, Wheat, and Grandet 2019).
And unlike manufacturing firms, businesses in the service sector
would not return to partial operations with a backlog of orders following
the lockdowns. Nearly all of the revenue they lost during the lockdowns
was lost forever—for example, diners did not eat twice as many meals out
in May and June because restaurants were shut in March and April.
To summarize, the economy was at risk of a cascade of small business
bankruptcies. Small businesses play a critical role in the economy. In 2019,
firms with fewer than 500 employees accounted for 47 percent of private
sector employees and 41 percent of private sector payroll. There were
30.7 million such businesses, 19  percent of which had paid employees
(US Small Business Administration 2019). A wave of small business failures
could have created an aggregate demand doom loop, in which declining
incomes and employment opportunities reinforced each other.

340
Brookings Papers on Economic Activity, Fall 2020
One way to address this concern would have been to lift lockdown orders.
But the public health effects of the virus and concern workers had about
getting sick would have made this strategy ineffective. The best option for
the federal government in a short, temporary shutdown was to make up a
large fraction of revenue businesses would have generated in normal times.
We return later to the challenges posed by longer-term partial shutdowns.
II.B.  Goals, Cost, and Key Design Features
The specific goals of such a program are to ensure small business
continuity and prevent a wave of bankruptcies and, during the period of the
shutdown, to preserve employment relationships. The overarching objec-
tive is to preserve as much of the productive capacity of the economy as
possible while short-term shelter-in-place orders are in effect and to help
the economy transition quickly to a new, post-shutdown equilibrium by
supporting labor demand over the medium term.
For firms, preventing wasteful liquidations allows the black box of pro-
ductive technologies and business relationships to remain intact. Profes-
sional networks are preserved, relationships with suppliers and customers
are maintained, and knowledge of local conditions and preferences can
continue to be put to productive use. For workers, the value of firm-specific
human capital is maintained, and maintaining employment relationships
means they continue to be paid by their employer, and they are in a posi-
tion to return to work immediately once shelter-in-place orders are lifted.
No separation takes place, not even a temporary furlough of workers. For
both workers and firms, productivity enhancing worker-firm matches are
maintained. And the economy is in a position to snap back quickly because
labor demand has been supported.4
This observation is especially true in a lockdown because the risk of
mass closures is so real. Without a program to support small business
continuity, a wave of closures would be followed by a period in which new
businesses started. Eventually, the economy would reach a new equilibrium.
4.  Papers that discuss the role of worker-firm matches include Mortensen and Pissarides
(1999) and Davis and von Wachter (2011). Jackson (2013) measures match quality directly
in the context of schools, estimating teacher, school, and match productivity on student
outcomes. He finds that teacher-school (worker-firm) match effects are important, estimating
that a one standard deviation increase in match quality increases math scores by an amount
roughly equal to two-thirds of the effect of a one standard deviation increase in teacher
quality. Using linked worker-firm data, Farooq, Kugler, and Muratori (2020) document an
important role for match quality and find that more generous unemployment insurance
benefits lead to higher quality matches. In our context, match quality likely matters the most
for larger PPP-eligible firms.

HUBBARD and STRAIN
341
But during the transition, labor demand would be depressed because there
would be fewer businesses looking for workers, which would lead to lengthy
spells of unemployment for millions of workers and a slower and more
sluggish recovery.
Because the (aggregate, present discounted value of) social benefits of
these businesses exceed their (aggregate, present discounted value of) costs,
a subsidy is justified under standard economic logic. Particularly given the
possibility of an aggregate demand doom loop and the lengthy period of
high unemployment it would cause, we argue that in the context of the
pandemic, the appropriate revenue replacement rate is large.
Once lockdown orders are lifted, partial revenue replacement may still be
needed. But it is no longer necessary or economically desirable to compel
firms to maintain pre-lockdown employment relationships or employ-
ment levels. After the economy has partially reopened, policy should not
introduce frictions into the process of reallocating labor (and capital) to
its post-lockdown most productive use, and policy should allow firms the
flexibility to reorganize their post-lockdown production functions to further
the key overall goal of a revenue replacement program: ensuring small
business continuity.
There is an inherent tension between a revenue replacement program’s
goal of maintaining employment relationships and keeping firms in business
and the goal of efficiently reallocating factor inputs and swiftly transition-
ing to a new, post-lockdown equilibrium. But for the reason we discussed
earlier, there is less to this tension than meets the eye in this case. A revenue
replacement program allows that transition to happen faster by preserving
many otherwise viable firms during the shutdown. Once the economy has
partially reopened, severing the link between program participation and
maintaining prepandemic employment levels is critical to minimizing this
tension. And a revenue replacement program should be of limited duration
following the reopening of the economy. A revenue replacement program
may also keep some businesses afloat that would have shut down in the
absence of the pandemic. Presumably most businesses that were not viable
prior to the pandemic will remain unviable once the revenue replacement
program has ended.
These considerations emphasize the need for the revenue replacement
program to focus on revenue, not simply on payroll costs. A separate reason
to focus on revenue rather than narrowly focusing on payroll costs is that
nonpayroll expenses, like rent in many cities, are significant. A program
replacing payroll costs but not overall revenue may not be sufficient to
keep many businesses in high-rent cities from closing.

342
Brookings Papers on Economic Activity, Fall 2020
Replacing small business revenue is an expensive proposition. Hubbard
and Strain (2020) estimate that replacing 80 percent of revenue for twelve
weeks for service sector businesses—that is, for businesses in industries
other than manufacturing, finance and insurance, health care, and educational
services—with fewer than 500 employees would cost $1.2 trillion.
Expensive as such an intervention is, the counterfactual would be even
costlier, with cascading business failures, wasteful liquidations, plunging
incomes, soaring unemployment, and little prospect for a rapid recovery
because of the devastating effects on the small business ecosystem. Another
budgetary consideration is the offsetting effects of less use of social insur-
ance programs, like unemployment insurance, and safety net programs,
like food stamps.
So far, our discussion of a small business revenue replacement program
has been general and could be applied to any situation in which small,
service sector businesses needed to shut down for a period of several weeks.
A key feature of the pandemic recession is that such a program did not
exist, and Congress needed to create one quickly. Given this context, it was
best for Congress to rely on the existing relationships many small businesses
have (via checking accounts or loans) with commercial banks rather than
to have had the government attempt to set up an entirely new direct transfer
program.
The government should have treated the banks essentially as conduits to
get money into business accounts as quickly as possible. Of course, such
an approach requires convincing banks that they will be held harmless in
the event of borrower misrepresentation, both by the current administration
and by future administrations. Strong assurances are necessary.5
To align better with an equity infusion, the revenue replacement grants
should be structured as loans that are forgivable if certain conditions are
met and should be fully backed by the government; that way banks assume
no risk. Banks should be allowed to charge fees, paid for by the government,
as an incentive to participate and for administrative costs.
5.  Prior to the 2008 financial crisis, large US banks routinely made Federal Housing
Administration (FHA) loans designed to help first-time home buyers and buyers with relatively
poor credit purchase houses. To reach these borrowers, the government encouraged lax lending
standards. This policy shift contributed to the housing bubble, and FHA’s solvency was in
question following the crash. The government imposed fines on banks, arguing they did not
adhere to FHA underwriting standards. The revenues from the fines helped to shore up FHA.
This episode has left many large banks skittish about using anything but strict underwriting
standards as part of government lending programs.

HUBBARD and STRAIN
343
Forgivable loans (i.e., grants) are necessary for the program to succeed.
The pandemic shutdown’s adverse consequences for firms’ collateralizable
net worth and cash flows require equity contributions. Loans, even with
low interest rates and long maturities, would likely be insufficient given
the need for equity financing. Service sector businesses permanently lose
revenue in a shutdown, and many would likely rather lay off workers than
take on additional debt. Even if debt service could be deferred for a period
of one or two years, many would be reluctant to take out a loan.6 These
businesses often have low profit margins, and a loan program would likely
have had an insufficient take-up rate to meet policymakers’ objectives.7
If the only concerns were cash flow challenges and a lack of access to
equity capital, then a lending program might be all that is justified. But as
we argued above, the divergence between the private and social value of
small business continuity suggests that subsidies are justified using standard
economic logic, particularly during the shutdown period. (In section VI,
we discuss how a lending program might complement grants once the
economy has partially reopened.)
A revenue replacement program should be broad-based and should
avoid too much targeting. In the fog-of-war atmosphere of the pandemic,
policymakers have limited knowledge of the virus’s spread, and crafting an
effective triggering mechanism based on public health metrics is difficult.
The government should avoid picking winners and losers by targeting the
program to select industries.
Revenue tests or demonstrations of hardship should also be avoided.
At the beginning of a sudden and unexpected lockdown, demonstrations
significantly slow down the process of getting funds to businesses, putting
the effectiveness of the program in jeopardy. Once the economy partially
reopens, it can be argued that revenue tests target assistance to firms that
need it most, as measured by revenue loss relative to normal circumstances.
But forward-looking revenue tests serve as a disincentive to earn revenue
by imposing implicit marginal tax rates on revenue. Backward-looking
revenue tests avoid this disincentive but are less generous to otherwise
6.  For a proposal that argues in favor of lending programs, see Ozimek and Lettieri (2020).
Hanson and others (2020a) argue for equity-like arrangements and grants to support small
business. Hanson and others (2020b) argue for payment assistance to impacted businesses to
meet recurring fixed obligations (e.g., interest, rent, and utilities) during the health emergency.
7.  At the time of this writing, the Federal Reserve’s Main Street Lending Facility has
very few loans, suggesting that even among midsize businesses taking on debt under terms
that are not borrower-friendly is not an attractive prospect.

344
Brookings Papers on Economic Activity, Fall 2020
identical firms that are doing better adjusting to the post-lockdown economic
circumstances.
The main appeal of revenue tests and hardship demonstrations are
lower program costs and targeting aid based on “need.” The problem
is that need is an amorphous concept in a partially reopened economy,
and revenue tests bring their own problems. The best targeting strategy
is broad-based, focusing on a large class of firms defined by size and
industry type.
II.C.  Addressing Moral Hazard Concerns
A program that replaces revenue for small businesses for a period of time
is an extraordinary government intervention in the private economy. It is
reasonable to be concerned that such a program would lead to excessive
risk taking or other imprudent behavior on the part of firms by potentially
creating the perception of a government “business revenue safety net.”
In normal public programs under normal circumstances, this concern is
certainly real. But in this instance, we are much less concerned about moral
hazard. The need to shut down large segments of the economy will occur
infrequently, and without advance notice. Businesses cannot purchase
shutdown insurance from private firms in the way they can insure against
risks from fires and floods. Government should communicate the extra­
ordinary nature of the assistance is driven by the extraordinary nature of
the threat. This step should mitigate moral hazard concerns.
II.D.  Policy Response in Other OECD Nations
Before turning to the Paycheck Protection Program, we briefly discuss
programs enacted by member countries in the Organisation for Economic
Cooperation and Development (OECD) during the pandemic recession.
See table A1 in the online appendix for specific program descriptions and
parameters for OECD countries.
Many European nations relied on a version of a wage subsidy scheme
in which workers saw their hours and pay reduced and their government
picked up a large part of the cost of employing them.8 This type of program
8.  Hamilton and Veuger (2020) argue that large expenditures to address the pandemic
will heighten concern about the public finances of some European Union member states,
implying that a broader European approach to fiscal policy is necessary. They suggest that
the eurozone issue Eurobonds to placate markets and to avoid issues associated with sovereign
debt overhang.

HUBBARD and STRAIN
345
was used by Germany (Kurzarbeit, or short-term work) during the Great
Recession and is widely credited with keeping the German unemployment
rate down during that period. The way it often worked was that firms paid
the benefit to their workers, which was typically somewhat lower than
wages, and the government reimbursed the firm (Blanchard, Philippon,
and Pisani-Ferry 2020). Austria implemented a similar program during the
pandemic, replacing up to 90 percent of covered wages.
A few examples: In the United Kingdom, the government reimbursed
firms for 80 percent of the wages of furloughed workers. Germany covered
60 percent of wages for childless workers on furlough and 67 percent for
furloughed workers with children. Depending on the month, the govern-
ment of France covered 84 percent or 72 percent (as of June) of wages for
workers on temporary layoff. Notably, these countries did not condition
eligibility based on firm size, in contrast to the United States’ emphasis
on small and midsize firms. Some European economies conditioned
subsidies on a demonstration of a significant decline in revenue (e.g., the
Netherlands, Estonia, and the Slovak Republic). Slovenia emphasized state-
funded bonuses for hazard pay in certain sectors.
These programs are similar to what we describe above. They maintain
the worker-firm relationship during the shutdown period, making it easier
for workers, firms, and the economy to recovery quickly once economic
activity partially resumes. Keeping workers paid by the firms also allows
government assistance to reach workers quickly. Such programs are similar
to standard unemployment insurance in that the government is helping sup-
port the incomes of workers who are underemployed, but unlike standard
unemployment insurance, they allow for part-time work.
At the same time, European programs have been more focused on
supporting workers in their current employment matches, rather than
smoothing a transition toward different employment matches. Programs
generally permitted workers receiving nonwork or part-time work benefits
to remain attached to the firm. As with the US Paycheck Protection Program,
the state effectively assumed a portion of payroll costs for covered workers,
albeit through payments made to firms.9 The US program formally worked
as a combination of loans and outright grants to firms and wage subsidies.
9.  Norway relied on layoffs, making it easier for firms to use temporary layoffs and
increasing the generosity of unemployment benefits for workers. Norway also instituted a
new compensation scheme for businesses that subsidized fixed costs. Alstadsæter and others
(2020) find that this program reduced firms’ economic distress by a similar magnitude to PPP
by reducing the negative effects of the crisis on profitability, liquidity, debt, and solvency.

346
Brookings Papers on Economic Activity, Fall 2020
As we describe later, a number of administrative challenges were “unforced
errors” in its implementation.
While some European pandemic unemployment or wage subsidy
schemes have faced fewer administrative challenges than in the United
States, they still raise concerns (to which we return later). Importantly,
they were and are designed to maintain employment relationships in a tem-
porary cyclical downturn (e.g., a moderate and short recession or a short
pandemic shutdown). In a reopening of the economy, policy shifts would
be needed to focus on rehiring workers and worker transitions by gradually
reducing wage subsidies and the generosity of unemployment benefits.
Employment policy responses in OECD countries outside Europe during
the pandemic have been varied. Canada, for example, focused on rehiring
workers previously laid off due to the COVID-19 pandemic, with subsidies
of up to 75 percent of all covered wages. Israel relied on relaxing require-
ments for unemployment benefits, direct and government-guaranteed loans
to businesses of all sizes, special support for high-risk businesses, grants
for small businesses, and a variety of measures to reduce the short-term
burden of business taxes. Australia, like large European economies, imple-
mented a wage subsidy for firms’ retention of employees. Japan financed
wage subsidies for retained workers, but only for small and midsize firms.
South Korea increased worker retention subsidies to up to 90 percent of
covered wages for three months for all employers. A less generous subsidy
to wages was provided in South Africa for firms whose operations were
at least partially curtailed as a consequence of the COVID-19 pandemic.10
In Latin America, Chile provided partial support for wage declines, and
Colombia assisted workers in firms with significant revenue declines with
support of 40 percent of the minimum wage.
III.  The Paycheck Protection Program
The Paycheck Protection Program (PPP) was created by the CARES Act,
the $1.8 trillion “phase 3” economic recovery package passed by Congress
and signed into law on March 27, 2020. In this section, we outline the
statutory design of PPP, the program’s implementation by the Department
of the Treasury and Small Business Administration, and the differences
between PPP and the features of a small business revenue replacement
program we discussed in the previous section.
10.  South African Government, “Support to Business,” https://www.gov.za/covid-19/
companies-and-employees/support-business#; accessed August 21, 2020.

HUBBARD and STRAIN
347
III.A.  PPP’s Design
PPP is a forgivable loan program. Businesses or nonprofits with 500 or
fewer employees; sole proprietors, independent contractors, or self-employed
individuals; and small businesses, 501(c)(19) veterans organizations, or
tribal business concerns that otherwise meet the Small Business Admin-
istration’s (SBA) size standards are eligible. Businesses in the accommo-
dation and food services sector (North American Industry Classification
System, or NAICS, code 72) may apply the 500 employee rule to each
physical location, not to the corporation as a whole. Congress appropriated
$349 billion for PPP in the CARES Act.
Under the program, businesses can borrow up to two and a half times
their average monthly payroll costs, capped at $10 million. Loans are issued
by banks and are guaranteed by the government.11 The amount of the loan
spent on payroll costs (including benefits), rent, utilities, and mortgage
interest during the twenty-four-week period (originally eight-week period)
after the loan is originated is forgiven—that is, it is converted to a grant—
provided that 60 percent (originally 75 percent) of the amount forgiven
is spent on payroll (a Treasury/SBA regulation not found in the CARES
Act) and that the business does not reduce headcount relative to precrisis
levels and does not reduce any employee’s compensation by more than
25 percent of his or her precrisis level. If headcount or compensation is
reduced beyond those parameters, the amount of the loan forgiven may
be reduced proportionately under some (but not all) circumstances. PPP
encouraged businesses that had already laid off workers due to the pandemic
to rehire them quickly without penalty.12
11.  Financial technology (fintech) played an important role, as well. Erel and Liebersohn
(2020) study the response of fintech to demand for financial services created by PPP. They
find that fintech was disproportionately used in zip codes with fewer bank branches, lower
incomes, and larger minority share of the population, in industries with less ex ante small
business lending, and in counties where the economic effect of the pandemic was more severe.
12.  Rules for loan forgiveness and for loan forgiveness reduction have been evolving.
At the time of this writing, loans can be fully forgiven if loan proceeds are spent and qualify-
ing costs are incurred during the covered period of the loan, which begins when the loan is
disbursed (or during an alternative covered period, depending on how the borrower manages
payroll); at least 60 percent of the loan amount (originally 75 percent) was used on payroll
costs; and staffing and compensation levels are maintained in the covered period relative to the
reference period. The covered period is twenty-four weeks for loans made after June 5, 2020.
For loans made before June 5, 2020, borrowers can choose between a twenty-four-week or
eight-week covered period. Borrowers can choose one of two reference periods: February 15,
2019, to June 30, 2019, or January 1, 2020, to February 29, 2020. (Seasonal employers have
different rules.) PPP also includes a safe harbor provision that allows borrowers to avoid
loan forgiveness reductions due to decreases in headcount or compensation that occurred

348
Brookings Papers on Economic Activity, Fall 2020
Borrowers do not need to demonstrate hardship in order to qualify for
a forgivable loan, which streamlines the process and allows banks to get
money to businesses quickly. Instead, they need to offer a series of good-
faith certifications, including: “Current economic uncertainty makes this loan
request necessary to support the ongoing operations of the Applicant.”13
Borrowers must also certify that the business intends to use the funds
received for payroll and other operating expenses and that they are not
applying for a duplicative loan. For a loan to be forgiven, in some cases,
businesses may need to present documentation to lenders demonstrating
that they complied with the terms of the loan. In other cases, businesses
simply need to attest to this.
To get funds to businesses quickly, PPP delegates authority to lenders
to determine borrower eligibility. Given the PPP’s structure, lenders do not
need to assess the ability of the borrower to repay the loan. No collateral
or personal guarantees from borrowers are required, and no credit else-
where tests are applied. Lenders simply need to establish that a business
was operational on February 15, 2020, and verify its payroll.
To entice banks to participate, the program allowed them to charge
generous fees—5 percent of principal on loans up to $350,000, 3 percent
on loans between $350,000 and $2 million, and 1 percent on loans above
$2 million up to $10 million. Lenders can charge an interest rate of 1 percent
on the portion of the loan that is not eligible for forgiveness, and loans have
zero weight in banks’ capital requirements. In the statute, lenders are “held
harmless” in the event of borrower misrepresentation, but the Treasury/
SBA did not waive requirements under the Bank Secrecy Act and required
anti–money laundering compliance programs.
The Paycheck Protection Program and Health Care Enhancement
Act was signed into law on April 24, 2020, and increased PPP funding by
between February 15, 2020, and April 26, 2020, provided that headcount and compensation
are restored by December 31, 2020 (originally June 30, 2020). Loan forgiveness will also
not be reduced if borrowers issue written offers to rehire workers who were employed on
February 15, 2020, and those offers are not accepted, or if borrowers document an inability
to rehire similarly qualified workers for vacancies as of December 31, 2020. Loan forgive-
ness will not be reduced if borrowers cannot maintain employment levels due to an inability
to return to the same level of business as of February 15, 2020, because they are complying
with coronavirus-related guidance for social distancing, sanitation, or worker or customer
safety requirements from various federal agencies and departments between March 1, 2020,
and December 31, 2020. On October 8, the Treasury/SBA issued additional guidance that
exempted borrowers with loans under $50,001 from any loan forgiveness reductions based
on failing to maintain headcount or wages.
13.  Paycheck Protection Program Borrower Application Form, revised June 24, 2020.

HUBBARD and STRAIN
349
$320 billion. The Paycheck Protection Program Flexibility Act (PPPFA)
was signed into law on June 5, 2020. The covered period of the forgiv-
able loan was extended from eight weeks to twenty-four weeks (or until
December 31, 2020). PPPFA also allowed businesses to spend 40 percent
of forgivable funds on nonpayroll expenses, rather than the 25 percent
previously established by Treasury/SBA regulation. The maturity of the
loans was increased from two years to five years for loans issued after June 5.
III.B.  Design Concerns
On the whole, PPP is well designed relative to objectives for financing
during a short-term shutdown we described earlier. It was able to get an
astonishing amount of money to millions of small businesses very quickly.
It relied on what are essentially grants and not loans. It took measures to
encourage banks to participate. It avoided revenue tests, and it did not
target select industries. Its goals—ensuring small business continuity and
preserving employment relationships—were the right ones.
But we have four concerns about some design elements. First, PPP is too
focused on payroll expenses. The goal should have been to replace revenue,
not simply to assist businesses with meeting payroll obligations. Even the
payroll share for forgiveness of 60 percent after PPPFA was enacted is too
high from this perspective.
Second, the program was designed with a short lockdown period in
mind. This approach was reasonable given widely held expectations
about the course of the pandemic in early March, and to some extent
this was addressed by PPPFA modifications to the program. Even so, the
program should be more flexible post-lockdown in allowing labor to be
reallocated across firms and industries, a problem given a longer period of
partial shutdown. PPP contains incentives that work against this needed
reallocation.
Third, a major flaw in PPP’s design was the original CARES Act
appropriation of $349 billion, and a major flaw in its execution was the
Treasury’s inability to convince banks that they would be held harmless
in the event of borrower misrepresentation. Both of these flaws led to the
reality and public perception that PPP funds were flowing to relatively
better resourced and less vulnerable small and midsize businesses.
Finally, Hubbard and Strain (2020) estimated that the PPP’s original
goals would require around $1 trillion. With only $349 billion originally
appropriated for PPP—and the intense demand for PPP loans in the early
days of the program—a perception developed that only businesses with
preexisting relationships with participating lenders would be able to access

350
Brookings Papers on Economic Activity, Fall 2020
the program. Lenders, in a rush to process applications and out of concern
that they would not be held harmless in all circumstances, focused lending
on existing bank customers.
IV.  Evaluating the PPP: Program Statistics, Implementation
Challenges, and Existing Evidence
In this section, we present basic statistics about PPP loans and discuss
implementation challenges. We also review current empirical evidence on
the effectiveness of PPP.
IV.A.  PPP Program Statistics
Table  1 presents PPP program statistics. As of August  8, PPP had
approved 5,212,128 loans representing a total of $525 billion provided
by 5,460 lenders. The average loan size is $101,000. The solid majority
of program dollars were included in loans of less than $2 million, and the
overwhelming majority of loans were for less than that amount. Loans
of over $2 million represent 0.6 percent of all loans and 20 percent of
all dollars loaned. In contrast, around 87 percent of all PPP loans were
made for less than $150,000, and 28 percent of all funds loaned were part
of loans of less than that amount. Figure 1 shows loan counts and loan
amounts over time.
Granja and others (2020) study the targeting of these loans across
geography and do not find evidence that the first round of PPP funds
went to parts of the country that saw the largest declines in hours worked
or business shutdowns. Further research is needed to study the targeting
of the full program. We also note that the entire country was affected by
shutdowns, and the degree to which different states were affected by the
Table 1.  Summary of PPP Lending, April 3–August 8
Cumulative lending
Loan count
Net loans ($)
Number of lenders
5,212,128
525,012,201,124
5,460
Distribution by loan size
% of count
% of amount
$150,000 and under
4,552,452
147,477,537,518
87.3
28.1
$150,000 to $2 million
630,694
272,228,531,130
12.1
51.9
Over $2 million
28,982
105,306,132,476
  0.6
20.1
Source: SBA Paycheck Protection Program.

HUBBARD and STRAIN
351
pandemic varied at different times, particularly as the nation entered the
summer months.14
IV.B.  Implementation Challenges
Table 2 presents a timeline of selected PPP events and includes some
implementation challenges. Before the program officially launched on
April 3, banks and other industry associations were warning of a chaotic
beginning to the program, arguing that borrower verification would be
onerous and would hamper the government’s objective of getting money
Source: SBA and Treasury Department micro data.
Notes: This figure displays cumulative loans and dollars lent during the operation of the PPP program
calculated as of August 20, 2020. Cumulative dollars lent are overstated in the micro data due to using
the midpoints of loan ranges provided for loans greater than $150,000. The shaded areas represent a
period of uncertainty over audits and the safe harbor deadline. The lightly shaded area covers the total
period of uncertainty over audits from April 28 (audits announced) to May 18 (final deadline to return
funds under safe harbor provision). The darker area covers the period of uncertainty over the safe harbor
deadline from May 7 (the original deadline) to May 18 (the final deadline).
4/17
5/1
5/15
5/29
6/12
6/26
7/10
7/24
8/7
1
2
3
4
5
Loans approved (millions)
100
200
300
400
500
Dollars lent (billions)
Initial
$349B
lent
New
$320B
approved
Flexibility Act
signed
Program
ends
Dollars approved
Loan count
Figure 1.  Cumulative Number of PPP Loans and Dollars Approved, April 3–August 8
14.  Figures A1 and A2 in the online appendix show loan counts and loan amounts by
state and PPP loans and employment losses by industry, respectively.

352
Brookings Papers on Economic Activity, Fall 2020
Table 2.  Timeline of Major Events in the PPP Program
Date
Description
March 27, 2020
CARES Act signed, appropriating $349 billion for PPP.
April 2
Treasury/SBA releases first interim final rule; 75 percent payroll
requirement; two-year repayment period; 0.5 percent interest rate;
eight weeks of covered expenses; application period to June 30.
Faced with complaints from small banks, the Treasury raises the
interest rate on PPP loans from 0.5 to 1 percent hours before the
program launch.
Bank associations, JPMorgan Chase Bank, and industry associations
warn of chaotic PPP launch; borrower verification requirements and
payroll cost calculations are unclear.
April 3
First round of PPP officially launches; only eight of twenty-five largest
SBA 7(a) lenders are taking applications. Bank of America and
JPMorgan Chase begin accepting applications but only for existing
customers.
April 16
First round of PPP ends; original $349 billion appropriation exhausted.
Thousands of submitted applications remain unapproved.
April 20
Small businesses sue large banks over allocation of loans. They claim
that banks violated first-come, first-served rules and gave priority
to larger applications that would generate more fees.
April 23
Treasury/SBA warns publicly traded companies and their subsidiaries
against seeking loans; sets May 7 deadline to return funds.
Treasury/SBA requires applicants to certify that the funds are necessary
due to the current economic uncertainty, as well as a lack of other
sources of funds to support their operations.
April 24
Paycheck Protection Program and Health Care Enhancement Act signed
into law authorizing an additional $320 billion for PPP.
April 27
Second round of PPP begins with $320 billion in new funding.
Treasury/SBA caps the dollar amount of loans that individual banks
can originate at $60 billion.
April 28
Secretary Steven Mnuchin announces full audits for loans > $2 million
and warns of criminal penalties for noncompliers.
April 29
SBA temporarily blocks large banks from submitting loans.
April 30
Justice Department launches probe of PPP.
IRS confirms that PPP loans are excluded from gross income, but
expenses paid for using PPP loans are not tax deductible.
May 5
Senate introduces Small Business Expense Protection Act to treat
expenses paid using PPP loans as ordinary deductible business
expenses.
Deadline for companies to return funds without penalty under safe
harbor provisions extended from May 7 to May 14.
May 8
SBA inspector general warns that the requirement of 75 percent payroll
costs and two-year repayment burdens borrowers and may not reflect
statutory intent.
May 13
SBA announces that loans below $2 million would be assumed to have
satisfied good-faith certification requirements; creates opportunity
for larger loans to be returned without penalty.
Deadline for companies to return funds without penalty under safe
harbor provisions extended from May 14 to May 18.

HUBBARD and STRAIN
353
May 14
Treasury says companies must use the total number of employees to
determine eligibility for PPP loans rather than full-time equivalent
as indicated previously.
May 22
Treasury/SBA warn that it may review PPP loans “of any size at any
time at SBA’s discretion”; borrowers required to retain documentation
for six years.
June 5
PPP Flexibility Act passed; covered period extended from eight weeks to
twenty-four weeks; repayment extended from two years to five years;
payroll costs allowed to be 60 percent of total loan forgiveness
amount, down from 75 percent.
June 12
For determining PPP eligibility, the look-back period for criminal
histories for nonfinancial felonies reduced from five years to one year.
June 30
Hours before program expiration and with $130 billion left, Congress
extends the PPP application period to August 8.
July 6
Under pressure from Congress, SBA releases the names of borrowers
and lenders and date of approval for loans of more than $150,000,
representing 15 percent of all approved loans and 75 percent of
dollars lent. Exact loan amounts are not disclosed.
July 7
Using data released by the SBA, researchers estimate that banks will
earn $24 billion in fees from PPP loans.
July 12
New York City comptroller report alleges that the city did not receive
its fair share of PPP loans.
July 17
Secretary Mnuchin asks Congress to consider automatically forgiving
all loans for less than $150,000, extending PPP, and suggesting
terms for PPP in a phase 4 economic recovery package.
August 4
Businesses, lobbyists, and professional organizations ask Congress to
exempt PPP income from tax reporting.
August 6
SBA releases guidelines on PPP loan forgiveness ahead of August 10
launch of forgiveness application platform. Many financial institu-
tions delay submitting applications until regulatory and legislative
uncertainty is resolved.
August 8
PPP application period closes with nearly $140 billion in reserve as
Congress debates “phase 4” economic recovery package.
Source: Authors’ compilation.
Table 2.  Timeline of Major Events in the PPP Program (Continued)
Date
Description
into the economy quickly, and due to confusion about basic program require-
ments like how lenders should calculate payroll costs. Due to confusion
about the program, on the day it launched only eight of the twenty-five
largest SBA 7(a) lenders were taking applications.
The early stage of PPP was also characterized by intense demand. By
the end of its second week, all $349 billion of CARES Act PPP appropria-
tions had been exhausted. Thousands of submitted applications remained
unapproved. There were accusations that large banks violated the first-
come, first-served structure of the program to favor large borrowers.

354
Brookings Papers on Economic Activity, Fall 2020
Articles in the press reported that some publicly traded companies or
their subsidiaries had received PPP loans. On April 23, the SBA released
guidance that publicly traded companies would likely find it difficult to
certify in good faith that they needed PPP loans.15 The Treasury/SBA
gave businesses until May 7 (later extended to May 14 and then May 18)
to return PPP funds without facing a penalty.16 On April 28, Treasury
Secretary Steven Mnuchin announced that a review of PPP loans in excess
of $2 million would take place. The secretary warned of potential criminal
penalties for borrowers found to have misrepresented themselves or not to
have complied with the terms of the loan.17 On May 13, the SBA attempted
to reassure borrowers and indicated that loans of less than $2 million would
be assumed to have made certifications of need in good faith.18
In our view, publicly traded firms or their subsidiaries should not have
been eligible for PPP loans. But confusion over eligibility for PPP loans,
which borrowers would be audited, and under what terms those audits
would take place had a profound effect on the program.
During the period of uncertainty discussed above, shown in the light and
dark gray bars in figure 1, the slope of both lines flattened. Dollars loaned
have increased more slowly since this period of Treasury-sown confusion
15.  See question 31, “Paycheck Protection Program Loans: Frequently Asked Questions,”
https://home.treasury.gov/system/files/136/Paycheck-Protection-Program-Frequently-Asked-
Questions.pdf (accessed August 21, 2020): “It is unlikely that a public company with substan-
tial market value and access to capital markets will be able to make the required certification
[of economic need] in good faith, and such a company should be prepared to demonstrate to
SBA, upon request, the basis for its certification.”
16.  See questions 43 and 47, “Paycheck Protection Program Loans: Frequently Asked
Questions,”
https://home.treasury.gov/system/files/136/Paycheck-Protection-Program-
Frequently-Asked-Questions.pdf.
17.  For example, Secretary Mnuchin made this statement on April 28 on CNBC: “I really
fault the borrowers who made these certifications. Now, there were some banks early on who
put things up on their website and prioritized their customers. We immediately told them
that was wrong. They took it down. So, you know, I want to be very clear: it’s the borrowers
who have criminal liability if they made this certification and it’s not true. And as I said,
we’re going to do a full audit of every loan over $2 million. This was a program designed
for small businesses, it was not a program that was designed for public companies that had
liquidity. Again, the certification was very clear in saying that if people had other sources of
liquidity, they could not take this loan”; https://www.cnbc.com/2020/04/28/cnbc-transcript-
treasury-secretary-steven-mnuchin-speaks-to-cnbcs-squawk-box-today.html.
18.  See question 46, “Paycheck Protection Program Loans: Frequently Asked Questions,”
accessed August 21, 2020. https://home.treasury.gov/system/files/136/Paycheck-Protection-
Program-Frequently-Asked-Questions.pdf: “Any borrower that, together with its affiliates,
received PPP loans with an original principal amount of less than $2 million will be deemed to
have made the required certification concerning the necessity of the loan request in good faith.”

HUBBARD and STRAIN
355
ended on May 18. New PPP loans continued to be made in the second half
of May and into June and July, but at a much slower rate than in April.
Of course, implementation shortcomings were inevitable to some degree
in setting up a program as ambitious as PPP in a short period of time
in the midst of a pandemic. But the Treasury’s muddled management of
PPP’s implementation is noteworthy because of its failure to take seri-
ously the advice it was given by a range of private sector participants and
policy experts, leading it to make mistakes that were both forecastable and
forecasted.
IV.C.  Brief Review of Existing Economic Research on the PPP
Study of the PPP by academic researchers is still in the working paper
stage, but some notable findings exist that shed light on the early effects
of the program. Bartik, Cullen, and others (2020) study the original
$349 billion of PPP funds. Using a survey of small businesses, they find
that PPP approval increased self-reported firm survival probability by
14 to 30 percentage points. They also find that banks allocated PPP funds
to firms with higher PPP treatment effects. But these firms were also more
likely to have stronger connections to banks, while firms with less cash on
hand were less likely to have their applications approved. They find that
PPP had a positive but statistically insignificant impact on employment.
Quite modest employment effects are also found by Chetty and others
(2020), who analyzed data from Earnin, a financial management applica-
tion.19 Granja and others (2020) also do not find evidence that the first round
of PPP had a substantial effect on employment or on other local economic
outcomes. Bartik, Bertrand, and others (2020) find that states that received
more PPP loans and those with more generous unemployment benefits had
labor markets whose declines were relatively less deep and whose recoveries
were relatively more rapid. Chodorow-Reich and others (2020) find that
PPP relaxed liquidity constraints facing firms, allowing some firms to pay
down existing credit line balances.
Autor and others (2020) use weekly data from ADP, Inc., payroll records
to study PPP’s effect on employment. Using a difference-in-differences
event study framework, they compare employment at firms above and
below the 500 employee PPP eligibility threshold. Through the first week
19.  Autor and others (2020) discuss limitations in the study by Chetty and others (2020),
including that Earnin data are focused on very low-wage workers, with median wages equal
to roughly the 10th percentile of wages in their industry, and that the absence of reported
standard errors makes the study results hard to interpret.

356
Brookings Papers on Economic Activity, Fall 2020
of June, they find that PPP increased employment by between 2 percent and
4.5 percent. After scaling by the take-up rate, they estimate PPP increased
aggregate payroll employment by 2.3 million workers, again through the
first week of June.
Autor and others (2020) divide total program expenditures by their
estimate of PPP’s effect on aggregate employment and report a cost per
job supported estimate of around $224,000. The paper notes that “while
this is a substantial cost per job supported, it would be premature to offer a
cost-benefit analysis of the PPP at this time” and points to the need to take
a longer-term view of PPP’s effects. We agree and would add that a short-
term cost-benefit analysis should include other factors. For example, many
workers who were kept on employer payrolls this spring would likely have
been receiving unemployment insurance benefits in the absence of PPP.
A short-term cost-benefit analysis should include cost savings from reducing
the demand for social insurance and safety net benefits.
More fundamentally, we disagree with Autor and others (2020) in that
we do not find cost per job supported to be a sufficient statistic to assess
PPP’s success. PPP is not exclusively a jobs program, and any evalua-
tion of its effectiveness per dollar of program expense—even a short-run
estimate—must include the benefit of preserving small businesses and
employment relationships holistically, including social benefits in excess
of private benefits and the benefits from hastening the economic recovery
by supporting labor demand over the medium term.
V.  Evaluating PPP: Empirical Analysis
We evaluate the effects of PPP on the employment, financial health, and
continuity of small businesses. To do this, we use data from the Dun &
Bradstreet Corporation (D&B), a company that provides commercial data
and analytics to businesses. We are able to identify businesses in the Dun
& Bradstreet data that applied for PPP loans of $150,000 or more. We do
not observe if those companies received a loan or the exact amount (above
$150,000 or more) of any loan received. We are not able to observe if a
business applied for a PPP loan of less than $150,000. Information on loan
applications comes from the SBA and is merged into the D&B data.
We estimate standard difference-in-differences models of the effect of
PPP application and of PPP eligibility based on size. We use several treat-
ment control groups in our analysis. We also estimate the dynamic effect
of PPP application and eligibility using event studies. We find evidence
that PPP increased employment, financial health, and continuity. We also

HUBBARD and STRAIN
357
find that the effect of PPP is unfolding, with effects on employment and
financial health growing over time and reaching their peak in August, the
last month for which we have data. In this section, we discuss the data, our
methods, and these results in further detail.
V.A.  Dun & Bradstreet
D&B is a global data and analytics company whose clients are busi-
nesses. The company was founded in 1841 as the Mercantile Agency and
became Dun & Bradstreet in 1933. It has extensive coverage, with over
355 million business records and data curated from tens of thousands of
sources, including public registries, newspapers, and websites, its own
investigations and telephone interviews, courts and legal filings, financial
statements, insolvency records, and its own network, making use of proprie-
tary and publicly available information. It is the world’s largest commercial
database and counts 90 percent of the Fortune 500 companies as clients,
along with every cabinet agency in the US government.
D&B is able to track whether businesses pay their bills on time through
its relationships with landlords, mortgage companies, credit card companies,
office suppliers, and the like. Its clients make use of D&B’s ability to
predict whether a particular establishment might be delinquent in order to
help clients manage financial risk. D&B has significant reach. For example,
the US government has historically required companies that want to
receive federal contracts to register with D&B, as does Apple for compa-
nies that want to distribute applications through its App Store. The Food
and Drug Administration uses a company’s D&B registration number as
a way to verify that importers of pharmaceutical products are legitimate
businesses and to confirm that applicant contact information is accurate
and complete.
V.B.  Sample, Variables, and Descriptive Statistics
Our sample includes all establishments in the D&B database active
as of October 2019 with one to 1,000 employees. We do not include sole
proprietorships, establishments with zero reported employees, establish-
ments with missing state and industry codes, and establishments with
modeled employee counts. We assign each establishment to a business
size category (one to 500 employees, 501 to 1,000 employees) based on
employment in February 2020. We also stratify establishments based on
whether they applied for a PPP loan worth $150,000 or more. (We are
only able to observe whether businesses applied for PPP loans of at least
$150,000.)

358
Brookings Papers on Economic Activity, Fall 2020
Table 3 presents summary means and standard deviations for key vari-
ables and the distribution of establishments over industry. Businesses that
applied for a PPP loan of $150,000 or more are nearly three times as
large as those that did not. This difference is likely due to the relatively
large size of the loan we are able to observe. Each group of businesses
has comparable PAYDEX scores (discussed below), and over the entire
sample period establishments with 501 to 1,000 employees are more likely
to go out of business. The group least likely to go out of business during
the sample period are establishments that we observe have applied for large
PPP loans, and by a wide margin.
Key variables for our analysis include PPP application (for loans of at
least $150,000), establishment employment, state, and industry. We use Dun
& Bradstreet’s PAYDEX variable as our measure of a business’s financial
health. PAYDEX is an indicator based on whether and how a business is
paying its bills. PAYDEX scores range from zero to 100. A PAYDEX score
of 80 denotes that payments made to D&B have generally been made within
the terms of the covered agreement. A PAYDEX score over 80 indicates
that payments reported to D&B have been made earlier than their terms
required. PAYDEX scores of 70, 60, 50, 40, 30, 20, or below 20 indicate
that businesses are 15, 22, 30, 60, 90, 120, or over 120 days late, respec-
tively, in paying their financial obligations. PAYDEX scores evolve slowly,
and for each business a given month’s PAYDEX score reflects transactions
that have taken place over the previous several months.
Examples of recent papers that have used D&B data to examine changes
in the financial health of small businesses include Barrot and Nanda
(2020), who study the impact of the 2011 federal QuickPay reform using
establishment-level employment data and PAYDEX scores from D&B.
Chava, Oettl, and Singh (2019) examine the effects of state minimum wage
increases on the financial health of small businesses. The authors use
the D&B PAYDEX score as their primary measure of financial health for
15.2 million establishments from 1989 to 2013.
D&B’s out-of-business indicator is our measure of business continuity.
It is a zero-one variable. D&B determines a business is out of business if it is
no longer engaging in transactions, through direct investigations, and in other
ways. Two separate authorities—for example, management or owners of the
company itself, the landlord at its address, its licensing body, and so on—
must confirm a business has closed for it to be recorded as out of business.
Panels A and B of figure 2 plot average establishment employment per
month for establishments with one to 500 employees in our analysis sample
and establishments with 501 to 1,000 employees. These plots indicate that

Table 3.  Summary Statistics, November 2019–March 2020
Group
1–500 employees
and applied for a
PPP loan ≥ $150,000
1–500 employees and
did not apply for a
PPP loan ≥ $150,000
All establishments
1–500 employees
All establishments
501–1,000 employees
Mean number of employees
per establishment
33.8
11.5
12.5
722.0
(47.6)
(35.2)
(36.1)
(167.0)
Mean PAYDEX score
73.9
72.6
72.7
70.0
(9.57)
(14.1)
(13.1)
(10.7)
Out of business (%)
0.010
0.157
0.150
0.325
(0.985)
(3.96)
(3.86)
(5.69)
Annual sales in 2019 ($)
5,603,688
2,168,570
2,338,007
66,242,380
(24,365,380)
(70,166,425)
(68,691,245)
(313,247,030)
Sectors (% share of employment)
    Agriculture
2.7
3.4
3.4
0.5
    Construction
14.2
8.0
8.2
1.7
    Finance, insurance, real estate
3.9
10.3
10.0
5.8
    Manufacturing
12.4
4.8
5.2
20.9
    Mining
0.5
0.3
0.3
0.7
    Public administration
0.1
3.1
2.9
16.5
    Retail trade
11.5
13.6
13.5
5.7
    Services
41.6
46.6
46.4
40.3
    Transportation, communications
4.6
4.9
4.9
5.1
    Wholesale trade
8.4
5.0
5.2
2.8
Source: Authors’ compilation using Dun & Bradstreet data.
Notes: This table displays means and standard deviations (in parentheses) in our pretreatment period, November–March, for the main establishment employee size groups
used in our analyses. We also calculate the distribution of employment across industries at the two-digit standard industrial classification level. The sample consists of all
establishments operating as of October 2019 that meet our sample selection criteria.

360
Brookings Papers on Economic Activity, Fall 2020
employment within the D&B sample is very stable. Among businesses with
one to 500 employees, employment decreased by 1.42 percent in August
relative to November. Panel B shows employment declines of 1.83 percent
among establishments with 501 to 1,000 employees. In contrast, employ-
ment reported in official statistics shows much larger losses. The summary
statistics we present suggest that employment evolves slowly among firms
of all sizes, and our analysis does not indicate any relationship between the
pace of evolution and PPP application. The relative stability of employ-
ment in the D&B data biases against finding a PPP employment effect, in
both our treatment-on-the-treated and intent-to-treat models. We interpret
all our estimates of PPP’s effects relative to trends in the D&B data.
We present the average PAYDEX score per month in panels C and D
of figure 2. These figures indicate that businesses’ financial health in our
sample is relatively stable, as well, falling in both panels by less than one
point. This apparent stability is most likely due to the relatively lengthy
look-back period for PAYDEX. As with the stability of employment, this
biases against finding an effect of PPP on financial health.
The share of establishments that went out of business is shown in panels E
and F of figure 2. The share of businesses with fewer than 500 employees
that went out of business increased by a factor of 16 between November
and August. Businesses with 501 to 1,000 employees saw closure rates
increase by a factor of 14.
V.C.  Estimation Strategy
To identify the effect of PPP on business outcomes, we estimate the
following equation:
(
)
= α + β
×
+ γ
+ δ
+ δ
+ ε
y
PPP
Post
PPP
im
ia
m
ia
sm
jm
im
(1)
,
where yim is an outcome experienced by business i in month m. Our analysis
sample covers ten months, November 2019 through August 2020, with
five months of pre-PPP period (the CARES Act was signed on March 27)
and five months of post-PPP period (PPP launched on April 3). PPPia is
an indicator as to whether business i applied for a PPP loan of at least
$150,000. This variable is our measure of PPP—we do not observe whether
businesses actually received PPP loans or, if they did receive loans, the
size of the loan. The variable dsm is a state-by-month effect, and djm is an
industry-by-month effect. The result of PPPia × Postm equals 1 if business i
applied for a PPP loan and the month is April, May, June, July, or August.
Standard errors are clustered by state.

HUBBARD and STRAIN
361
Source: Dun & Bradstreet; authors’ calculations.
Note: These graphs show average employment, PAYDEX score, and out-of-business rates from
November 2019 to August 2020 for establishments with 1–500 employees and 501–1,000 employees.
Establishments are assigned to an employment size group using February 2020 employment. Panels A,
C, and E include establishments with 1–500 employees. Panels B, D, and F include all establishments
with 501–1,000 employees.
E: Out of Business
1–500 Employees
0.3
0.4
0.1
0.2
F: Out of Business
501–1,000 Employees
0.50
0.75
0.25
C: PAYDEX
1–500 Employees
72.6
72.8
72.4
D: PAYDEX
501–1,000 Employees
69.5
70.0
69.0
A: Employment
1–500 Employees
12.5
12.4
B: Employment
501–1,000 Employees
715
720
710
Jan
Mar
May
Jul
Jan
Mar
May
Jul
Jan
Mar
May
Jul
Jan
Mar
May
Jul
Jan
Mar
May
Jul
Jan
Mar
May
Jul
Figure 2.  Average Establishment Employment, PAYDEX Scores, and Out-of-Business
Rates by Month

362
Brookings Papers on Economic Activity, Fall 2020
The coefficient of interest is β, which captures the effect of applying for
a PPP loan of $150,000 or greater on the outcome variable. The industry-
month effects capture time varying shocks to businesses in a given industry,
and the state-month effects capture time varying shocks to businesses in a
given state. The effects of the pandemic and the lockdowns varied substan-
tially across industries and states. Using within-state-by-month and within-
industry-by-month variation to estimate the effect of PPP application helps
ensure that our results are not driven by time varying public health or social
distancing policy differences between states and industries.
To trace the dynamics of PPP over the months since the CARES Act,
we estimate a difference-in-differences event study of the following form:
∑
(
)
= α +
β
× ϕ
+ γ
+ δ
+ δ
+ ε
=−
y
PPP
PPP
im
t
ia
t
t
ia
sm
jm
im
(2)
,
4
5
where βt is a vector of nine parameters estimating the dynamic effect of PPP,
jt is a month dummy, and everything else is the same as in equation (1).
The dynamics of the effect are interesting because of lags in receipt time,
the time it may take employers to bring workers back onto payroll, and
treatment control differences driven by the economic outcomes of control
businesses worsening over time because they do not have access to PPP
funds. The trend in the pre-PPP period coefficient vector is a partial check
against differential employment trends among businesses that applied for a
PPP loan and those that did not.
We observe whether a business applied for a PPP loan of $150,000 or
more. If some businesses that applied were turned down, then our estimates
of PPP’s effect are biased downward, because the treatment group would
be contaminated by control observations. Another important source of
downward bias in our estimates of PPP’s effect is that many businesses in
our control group applied for and received PPP loans of less than $150,000.
As presented in table 1, around 87 percent of all PPP loans were made for
$150,000 or less, and these loans accounted for 28 percent of all funds
disbursed. These are treatment-on-the-treated estimates and do not control
for selection into applying for PPP. Firms that did not apply could be very
different from those that did, perhaps thinking that they did not need the
funds to continue operating or, alternatively, perhaps thinking that the
situation was hopeless. They might have also been less financially savvy,
which could be correlated with other outcomes and characteristics.
Knowing how PPP affected firms that selected into participating is inter-
esting and important, but it confounds demand for PPP with PPP itself. To

HUBBARD and STRAIN
363
address this distinction, we estimate intent-to-treat models. In these models,
we do not use information on whether a business actually applied for a PPP
loan. Instead, we compare outcomes for establishments that were eligible for
PPP based on their size to establishments that were ineligible in a difference-
in-differences framework. Specifically, we estimate the following equation:
(
)
= α + β
×
+ γ
+ δ
+ δ
+ ε
y
PPP
Post
PPP
im
ie
m
ie
sm
jm
im
(3)
.
All variables in equation (3) are the same as in equation (1) except
PPPie, which equals 1 if a business is eligible for PPP based on its size, and
equals 0 otherwise. We also estimate intent-to-treat event studies analogous
to equation (2).
V.D.  Results
RESULTS FOR EMPLOYMENT  Table 4 presents estimates of equations (1)
and (3) for (the log of) employment. The specification in the first column
compares establishments with one to 500 employees that applied for a
PPP loan of $150,000 or more to establishments in the same size class
but that did not apply. PPP application is associated with a 0.90 percent
increase in employment. Columns 2 and 3 present the same specification,
but on smaller samples of establishments. Column 2 looks at establishments
with one to 250 employees and similarly finds a 0.94 percent increase in
employment from PPP. Column 3 analyzes a sample of establishments with
251 to 500 employees. Here, the effect on employment is negative, −3.2 per-
cent. This result might be driven by greater demand for larger PPP loans
within that size class among the treatment group, confounded by many control
firms taking out PPP loans that we do not observe. But in evaluating the
program as a whole, it is worth noting that there are approximately 81 million
establishment-months with one to 500 employees in our sample, and around
360,000 of those are establishment-months with 251 to 500 employees.
These estimates are valuable in part because they implicitly control for
establishment size category. But they are likely biased downward because
the treatment effect is defined as a business applying for a PPP loan of
$150,000 or greater, while most PPP loans were for less than this amount,
so PPP-treated establishments are in the control group. The specification
in column 4 attempts to address this by defining the treatment group as
establishments with less than 500 employees who applied for a PPP loan
of at least $150,000 and the control group as establishments with 501 to
1,000 employees. Here, we estimate a PPP employment effect of 1.78 percent,
substantially larger in magnitude than the coefficients discussed previously.

Table 4.  Estimating the Effect of PPP Loans on Establishment-Level Employment
(1)
(2)
(3)
(4)
(5)
(6)
Treated × Post × 100
0.902***
0.936***
−3.20***
1.78***
0.0772
1.38***
(0.0656)
(0.0655)
(0.470)
(0.234)
(0.366)
(0.258)
Treatment
1–500; loan
1–250; loan
251–500; loan
1–500; loan
400–475; all
    establishments
1–500; all
    establishments
Control
1–500; no loan
1–250; no loan
251–500; no loan
501–1,000; no loan
525–600; all
    establishments
501–1,000; all
    establishments
Observations
81,404,032
81,043,431
360,601
3,980,677
110,712
81,523,211
R2
0.1390
0.1373
0.0432
0.2343
0.3783
0.0966
Source: Authors’ compilation using Dun & Bradstreet data.
Note: This table reports difference-in-differences estimates for the impact of PPP on establishment-level employment. The sample consists of establishments operational
as of October 2019 that meet our sample selection criteria. For all regressions, the pretreatment period is November–March and the posttreatment period is April–August.
Each column uses a different treatment and control group, indicating the size of the establishment by employment in February; “loan” indicates that we observe that the
establishment applied for a PPP loan of at least $150,000; “no loan” indicates the opposite; “all establishments” indicates that we include all establishments in the analysis
sample regardless of whether they applied for a loan. All regressions include state, month, and two-digit standard industrial classification fixed effects as well as state-by-
month and industry-by-month fixed effects. Standard errors are clustered at the state level. Coefficients and standard errors are multiplied by 100 for ease of interpretation.
*** p < .01

HUBBARD and STRAIN
365
The estimates reported in columns 1–4 are treatment-on-the-treated
estimates. In the context of evaluating PPP, this is interesting because
estimating program outcomes conditional on selection is important and
relevant (program participation is voluntary) and survey evidence finds that
over 70 percent of small businesses participated in PPP.20 But the estimates
do confound the effect of demand for PPP with the effect of PPP, in addition
to the limitation that we only observe PPP loans of at least $150,000.
To address these limitations, column 6 reports intent-to-treat estimates in
which we define the treatment group purely based on size eligibility—that is,
we do not use information on whether a business applied for a PPP loan—
and the control group is establishments with 501 to 1,000 employees. We
estimate that PPP size eligibility increased employment by 1.38 percent. This
result might suggest an important role for smaller PPP loans in supporting
employment.
Column 5 also reports intent-to-treat effects but for firms close to the
500 employee cutoff (eliminating firms near the cutoff). The advantage
of this specification is that it directly controls for firm size. Comparing
firms in the 400–600 employee window, we do not find a PPP employment
effect. This result, along with the estimates reported in column 6, might
suggest that PPP was most effective in supporting employment among
smaller firms, at least through August.
The specification that estimates the effect of PPP within the 400–
600 employee window arguably offers the strongest basis for causal infer-
ence assuming that the effect of PPP loans on employment is similar for
firms of different sizes. But this assumption is very strong, and it is quite
likely that PPP loans have effects that vary by firm size. The estimates
reported in table 4 suggest this is the case, and the $10 million maxi-
mum for PPP loans also suggests that PPP would offer relatively more
assistance to smaller firms. In the D&B data, 2019 average annual sales
for firms with one to 500 employees were $2.4 million, while those for
firms with 400–475 employees were $46.4 million. This consideration
suggests that a holistic evaluation of PPP should include estimating its
effects on firms of all eligible sizes. Therefore, our preferred specifications
are presented in columns 4 and 6.
Our results contrast with Autor and others (2020), who find employment
effects for larger firms using ADP data. It is interesting to note that their
20.  The Small Business Pulse Survey of the US Census Bureau finds that 72.7 percent
of small businesses received financial assistance from PPP since March 13, 2020, as of
August 22, 2020.

366
Brookings Papers on Economic Activity, Fall 2020
estimates become less precise as the window around the 500 employee
eligibility cutoff shrinks. This finding may be due to sample size, or it could
indicate that PPP is relatively less effective at supporting employment for
larger firms in the ADP data.
We present event study graphs using our two preferred treatment and
control groups. Figure 3 presents results from equation (2). Panel A shows
the dynamic effect of PPP on employment when the treatment group is
establishments with one to 500 employees who applied for a PPP loan
of at least $150,000 and the control group is establishments with 501 to
1,000 employees. There is no trend in the pre-PPP period coefficients,
although the confidence interval on the negative coefficient in February does
not include zero. The absence of a pre-PPP period trend supports a causal
interpretation of the estimates. In the post-PPP period coefficients, the
effect of PPP increases over time, rising to 3.13 percent in August.
Panel B shows a similar effect of PPP on employment. Here, the dynamic
effect captures intent to treat, comparing establishments with 500 or fewer
employees to those with between 501 and 1,000 employees, regardless of
whether the firms applied for a PPP loan. Like panel A, there is no notice-
able trend in the pre-PPP period, and the strength of the effect increases in
the post-PPP period with each month. In August PPP eligibility is found to
increase employment by 3.83 percent.
To interpret the magnitude of these effects, consider that average
establishment employment fell by 1.6 percent in the D&B data for estab-
lishments with one to 1,000 employees over the sample period, between
November and August. In light of this change, the 1.78 percent increase in
employment reported in column 4 of table 4 and the 1.38 percent increase
reported in column 6 of table 4 are both substantial increases. The effects
for the month of August specifically—3.13 and 3.83 percent, respectively—
are even more substantial.
RESULTS FOR FINANCIAL HEALTH  Table 5 reports results for which the
outcome variable is financial health, as captured by Dun & Bradstreet’s
PAYDEX score. Table 5 is the same as table 4, except for the outcome
variable. The first three columns of table 5 report results from speci-
fications where the treatment and control groups are the same firm
employee size class. Taken together, they suggest that financial health
worsened for firms with one to 250 employees that applied for PPP loans
of at least $150,000. We think this puzzling finding is most likely the
result of PPP-treated observations (that is, establishments with less than
250 employees that applied for loans of less than $150,000) contaminat-
ing the control group.

HUBBARD and STRAIN
367
E: Out of Business
F: Out of Business
0
0.5
–0.5
C: PAYDEX
D: PAYDEX
0.5
0
A: Employment
B: Employment
2
4
0
Source: Dun & Bradstreet; authors’ calculations.
Note: These graphs show the results from event study regressions in equation (2) examining the impact
of the Paycheck Protection Program on establishment employment, financial health, and survival. Panels
A, C, and E examine PPP’s effect on employment, credit scores, and survival rates for establishments
with 1–500 employees that applied for a PPP loan of $150,000 or more compared to establishments with
501–1,000 employees. Panels B, D, and F examine the effect of PPP eligibility on the same outcomes,
comparing all establishments with 1–500 employees to all establishments with 501–1,000 employees
(i.e., dynamic intent-to-treat effects). Establishments are assigned to an employment size group using
February employment. Coefficients and standard errors for panels A, B, E, and F are multiplied by 100
to ease interpretation. Error bars represent 95 percent confidence intervals.
2
4
0
NovDec Jan Feb Mar AprMay Jun Jul Aug
0
0.5
–0.5
NovDec Jan Feb Mar AprMay Jun Jul Aug
0
0.5
–0.5
NovDec Jan Feb Mar AprMay Jun Jul Aug
NovDec Jan Feb Mar AprMay Jun Jul Aug
NovDec Jan Feb Mar AprMay Jun Jul Aug
NovDec Jan Feb Mar AprMay Jun Jul Aug
Figure 3.  Event Study Regressions

Table 5.  Estimating the Effect of PPP Loans on Establishment-Level Credit Scores
(1)
(2)
(3)
(4)
(5)
(6)
Treated × Post
−0.0392*
−0.0379*
−0.0349
0.305***
0.01
0.349***
(0.0220)
(0.0219)
(0.0969)
(0.0686)
(0.154)
(0.0616)
Treatment
1–500; loan
1–250; loan
251–500; loan
1–500; loan
400–475; all
    establishments
1–500; all
    establishments
Control
1–500; no loan
1–250; no loan
251–500; no loan
501–1,000; no loan
525–600; all
    establishments
501–1,000; all
    establishments
Observations
32,139,590
31,889,423
250,167
3,731,639
81,644
32,225,515
R2
0.013
0.013
0.021
0.024
0.027
0.012
Source: Authors’ compilation using Dun & Bradstreet data.
Note: This table reports difference-in-differences estimates for the impact of PPP on establishment-level employment. The sample consists of establishments operational
as of October 2019 that meet our sample selection criteria. For all regressions, the pretreatment period is November–March and the posttreatment period is April–August.
Each column uses a different treatment and control group, indicating the size of the establishment by employment in February; “loan” indicates that we observe that the
establishment applied for a PPP loan of at least $150,000; “no loan” indicates the opposite; “all establishments” indicates that we include all establishments in the analysis
sample regardless of whether they applied for a loan. All regressions include state, month, and two-digit standard industrial classification fixed effects as well as state-by-
month and industry-by-month fixed effects. Standard errors are clustered at the state level. Coefficients and standard errors are multiplied by 100 for ease of interpretation.
* p < .1, *** p < .01

HUBBARD and STRAIN
369
For reasons discussed previously, our preferred specifications are reported
in columns 4 and 6. The specification in column 4 compares firms with
500 or fewer employees that applied for PPP loans of at least $150,000 to
firms with 501 to 1,000 employees that were not eligible for PPP. PPP pre-
dicts a PAYDEX increase of about 0.31 points. Column 6 presents results
from an intent-to-treat specification. Here, PPP eligibility boosts PAYDEX
by about 0.35 points. Similar to our results for employment, PPP seems
to have had a larger impact on firms with fewer than 400 employees, as
suggested by comparing the results in column 5 with column 4.
Figure 3, panels C and D, present event study graphs that trace out the
dynamic effect of PPP for our two preferred specifications. As with employ-
ment, the effect of PPP on financial health (as measured by PAYDEX)
grows over time. Both figures show a flat trend centered on zero for the
pre-PPP period coefficients estimating the effect of PPP in November through
February relative to March. As with employment, this supports a causal
interpretation of our estimates. The effect of PPP application on financial
health was estimated imprecisely in April and precisely every month after.
The magnitude of the effect increased considerably over time, more than
doubling between June and August.
The dynamic intent-to-treat estimate is shown in panel D. As with the
results in panel C, PPP’s effect on financial health is estimated imprecisely
in April but precisely for the following four months. The magnitude of the
effect in August is more than double the effect in May. PPP eligibility is
estimated to have increased PAYDEX in August by 0.51 points.
The magnitude of the effect is substantial. For all firms with one to
1,000 employees, average monthly PAYDEX fell by 0.28 points from
November to August. A PPP PAYDEX effect of 0.31 (column 4) and 0.35
(column 5) represents a significant increase relative to the change in finan-
cial health of all firms during our sample period. As with employment, the
effect of PPP on PAYDEX in June is substantially larger than the post-PPP
period average, suggesting that the effects of PPP on financial health may
be increasing over time.
RESULTS FOR BUSINESS CONTINUITY  Table  6 reports results for D&B’s
out-of-business variable. Everything in table 6 is the same as in tables 4
and 5, except the outcome variable. PPP eligibility or application is esti-
mated to have reduced business closure in every specification at conven-
tional levels of statistical significance, except for column 5. Column 4
presents results from the specification that compares firms that applied for
a PPP loan of at least $150,000 to firms with 501 to 1,000 employees,
which were ineligible for PPP. PPP application is estimated to have reduced

Table 6.  Estimating the Effect of PPP Loans on the Probability an Establishment Goes Out of Business
(1)
(2)
(3)
(4)
(5)
(6)
Treated × Post × 100
−0.237***
−0.236***
−0.616***
−0.471***
0.0562
−0.219**
(0.0222)
(0.0224)
(0.0836)
(0.0853)
(0.124)
(0.0683)
Treatment
1–500; loan
1–250; loan
251–500; loan
1–500; loan
400–475; all
    establishments
1–500; all
    establishments
Control
1–500; no loan
1–250; no loan
251–500; no loan
501–1,000; no loan
525–600; all
    establishments
501–1,000; all
    establishments
Observations
81,625,920
81,262,585
363,335
3,982,131
111,512
81,745,730
R2
0.00805
0.00804
0.02154
0.00344
0.0166
0.00789
Source: Authors’ compilation using Dun & Bradstreet data.
Notes: This table reports difference-in-differences estimates for the impact of PPP on establishment-level employment. The sample consists of establishments operational
as of October 2019 that meet our sample selection criteria. For all regressions, the pretreatment period is November–March and the posttreatment period is April–August.
Each column uses a different treatment and control group, indicating the size of the establishment by employment in February 2020; “loan” indicates that we observe that the
establishment applied for a PPP loan of at least $150,000; “no loan” indicates the opposite; “all establishments” indicates that we include all establishments in the analysis
sample regardless of whether they applied for a loan. All regressions include state, month, and two-digit standard industrial classification fixed effects as well as state-by-
month and industry-by-month fixed effects. Standard errors are clustered at the state level. Coefficients and standard errors are multiplied by 100 for ease of interpretation.
** p < .05, *** p < .01

HUBBARD and STRAIN
371
the odds of business closure by 0.47 percentage points. Column 6 presents
results from our intent-to-treat model. Here, PPP eligibility is estimated to
reduce business closure odds by 0.22 percentage points. Column 5 reports
intent-to-treat results for a smaller window around the 500 employee cutoff.
As with employment and financial health, we do not find a significant effect
of PPP on business closure among firms with 400–475 employees.
Panels E and F of figure 3 present event studies for those two models.
The pre-PPP period coefficients show a trend, and these results should be
interpreted cautiously. The confidence interval on pre-PPP period coeffi-
cients includes zero in several cases. In the post-PPP period, the magnitude
of the effect is larger in June than in April or May. This pattern is similar
to our employment and PAYDEX results. The magnitude of these effects
is large.
To place the difference-in-differences estimates and June event study
coefficient estimates in context, the average establishment out-of-business
indicator in August was 0.42 percentage points higher than in November
for firms with one to 1,000 employees.
V.E.  Discussion and Conclusions
Our results point to PPP playing a significant role in the health and
viability of small businesses. Both applying for a PPP loan of $150,000 or
more and PPP eligibility as determined by firm size increase employment,
financial health, and business continuity. In addition, we find that it may
have taken a month or two for PPP to kick in. An alternative interpretation
is that PPP was more effective in a partially reopened economy (that is,
June–August) than during the lockdowns.
Several caveats are in order. We avoid making strong statements about
the success or failure of PPP because the program is so young, and we are
only analyzing the first five months of the program. PPP did have impor-
tant short-run goals, which included maintaining employment relationships
during the lockdowns and supporting consumer spending by allowing
workers to continue to be paid. But PPP has important medium-run goals
as well, and it is too early to say anything definitive about its success or
failure. Those goals include mitigating business closures after the economy
had partially reopened (which we observe for about one month), supporting
employment and reducing unemployment, and increasing productivity by
preserving firm-specific human capital, worker-firm matches, and networks.
Crucially, by preserving the productivity capacity of the small business
sector, PPP stands to quicken the recovery by supporting labor demand over
the medium run. In addition, the firms in the D&B data are not nationally

372
Brookings Papers on Economic Activity, Fall 2020
representative, and they exhibit employment and financial health indica-
tors that are likely more stable than typical firms. We also want to stress
the tentative nature of our conclusions. As shown in the dynamics of the
effect (in figure 3), the effect of PPP on employment, financial health,
and business continuity is evolving and is much stronger in July and
August than in April and May. The effects of PPP are unfolding, and it will
be particularly important to see what happens to businesses that received
PPP and the workers they employ once they have exhausted their forgiv-
able loan.
VI.  Retrospective and Lessons for the Future
Many of the common criticisms of the PPP as failed by design and effect
were too strong. Banks were skittish about participating, particularly in the
early days of the program. But program demand by lenders was sufficient
to allow the government to transfer funds in an amount roughly equal to
10 percent of a typical quarter’s GDP to small businesses. With the vast
majority of loans and the sizeable majority of program dollars going to
loans of less than $2 million, media coverage suggesting that PPP was in
the main offering grants to large and well-connected firms was overblown.
Many of the anecdotes in the media implying fraudulent participation in the
program actually pointed to firms that were eligible for PPP loans under
the statute. The criticism that the original CARES Act appropriation of
$349 billion was too small, obvious from the outset, was quickly proven
valid by events, but Congress rectified that swiftly.
Could policymakers have designed a more effective and cost-effective
intervention than a small business revenue replacement program? In theory,
one could argue that relying on the unemployment insurance (UI) system
to replace workers’ income and using a PPP-like program to help small busi-
nesses with nonpayroll cost has appeal to some economists and analysts.
But that plan would require worker-firm separations, albeit temporary, to
take place. It would change the default for small businesses from keeping
workers employed (as under a revenue replacement program) to recalling
workers following a separation, which is the wrong place for the default
to be during the shutdown. The UI system in many states was simply and
troublingly unable to handle the demands placed on it during the shutdown—
increasing those demands would not likely lead to the most successful
outcomes. Finally, having both UI and a small business revenue replace-
ment program is good policy design because it allows for redundancy, with
multiple programs operating to replace workers’ incomes.

HUBBARD and STRAIN
373
For the reasons we discussed previously, we do not view a loan program
as an adequate substitute for a small business revenue replacement program.
Many businesses would not want to add to their debt burdens, even under
very favorable lending conditions. Many would resort to layoffs, which
would disrupt other businesses, deepen the recession, and hurt workers’
employment and earnings opportunities.
Even though a small business revenue replacement program may have
been the best available option, the PPP could have been better designed and
better implemented in ways we previously discussed: it is too focused on
payroll expenses; banks should have been given stronger assurances that
they would be held harmless; and its initial appropriation was too small.
Much of the confusion about the program was driven by chaotic Treasury/
SBA management which weakened the program’s effectiveness, limited its
reach, and ultimately led to a falloff in demand for PPP funds.
PPP was designed for a short shutdown that would be followed by a
strong and rapid recovery. But the shutdown was longer than anticipated
and the recovery decelerated after a burst of improvement in May and June.
In addition, partial shutdowns may remain in some regions for an extended
period of time. Subsequent changes to PPP addressed these concerns, but
the program needed to facilitate the transition from the “freeze the economy
in place” stage to the “allow labor to reallocate across firms and industries”
stage. The economy overall, including workers, will benefit from a fast
transition from the pre- to post-lockdown equilibrium. PPP could facilitate
this transition by eliminating any link between PPP loan forgiveness and
precrisis employment levels.
We have argued that many small businesses needed equity or grants, and
not loans. But a lending program could—and in the future perhaps should—
exist alongside a revenue replacement program, particularly for a partially
reopened economy. An advantage of a lending program is that businesses
that expect to be nonviable in the post-pandemic economy would be less
likely to take out a loan than to accept a grant. This feature would keep the
cost of the program lower, channel funds more effectively, and allow for a
swifter transition to the post-pandemic equilibrium. A disadvantage—and
the reason we do not support this during the shutdown period—is that some
firms that might be viable in the absence of the loan could be tipped over
into insolvency by taking out a loan. More practically, in the shutdown, we
are concerned that few firms would participate in a lending program.
One way to structure such a lending program could be in two stages,
following a venture capital model preceded by a broadly available loan.
In the first stage, the Treasury Department could issue a small loan to firms

374
Brookings Papers on Economic Activity, Fall 2020
using limited underwriting standards, knowing that the loan will have a
high default rate. In the second stage, surviving firms could have access to
additional funding. This financing would help to give many firms a lifeline
for survival, while still well-stewarding taxpayer funds.21
An alternative approach would be a federal business interruption
insurance program for small and midsize firms (analogous to the federal
terrorism risk insurance program) layered on top of private business
interruption insurance. Linking a trigger to a pandemic shutdown could
require a shutdown order by a public official (for example, the governor
of the state).
Looking forward, there are broader lessons as well. For a situation in
which the government is shutting down large sections of the economy,
Congress and the White House need to be willing to tolerate stories of
“undeserving” beneficiaries of economic recovery programs. The alternative
is upfront targeting measures that slow down aid and worsen the downturn.
Another alternative is that programs are much less effective. PPP stands
a chance at succeeding because its relief was broad based. The Treasury
Department was much more conservative with putting taxpayer dollars at
risk when approving the terms of the PPP, limiting early take-up. The
Treasury’s conservative approach has extended to the Federal Reserve’s
Main Street Lending Facility, which received capital funds (along with
other Federal Reserve facilities under the CARES Act). As a consequence
of the Treasury’s aversion to putting that capital at risk, potentially driven
in part by concern about stories of undeserving borrowers, the facility is
not supporting the economic recovery yet because it, essentially, is not
making loans.
Another broader lesson is the need for government at the state and
federal levels to upgrade computer systems. Banks were needed as inter-
mediaries in part because the government’s IT constraint would not have
allowed for it to lend directly to banks in a timely fashion. Finally, the
government’s attempt to support small and midsize businesses in the
pandemic recession calls into question the nature of the division between
the Federal Reserve and the Treasury. Following the Dodd-Frank Act,
the Treasury is required to approve the terms of 13(3) lending programs,
including the Main Street programs. But these are labeled as Federal
21.  The Federal Reserve’s Main Street Lending Facility offers another lending vehicle for
small and midsize firms. While the facility’s design remains in flux, its structure could also
mimic better patient equity financing. Terms could include much longer maturity and very
low interest rates, for example.

HUBBARD and STRAIN
375
Reserve programs, creating confusion about which agency is ultimately
responsible for their success or failure. Furthermore, Congress appropriated
$454 billion in the CARES Act to the Treasury to support Federal Reserve
lending programs. At the time of this writing, little of those funds have
been put to use to support the recovery, despite congressional intent. If
the Treasury is unwilling to risk capital losses as part of Federal Reserve
lending programs, then Congress should consider whether an alternative
structure to support small and midsize businesses is advisable.
ACKNOWLEDGMENTS    We thank Duncan Hobbs for excellent research
assistance. We gratefully acknowledge the Dun & Bradstreet Corporation for
providing access to and assistance with its data; in particular, we thank Stephen
Daffron and Andrew Byrnes. We thank Jeffrey Clemens, Steven Davis, Stan
Veuger, and Eric Zwick for helpful comments.

376
Brookings Papers on Economic Activity, Fall 2020
References
Alstadsæter, Annette, Julie Brun Bjørkheim, Wojciech Kopczuk, and Andreas
Økland. 2020. “Norwegian and US Policies Alleviate Business Vulnerability
Due to the COVID-19 Shock Equally Well.” National Tax Journal 73, no. 3:
805–28.
Autor, David, David Cho, Leland D. Crane, Mita Goldar, Byron Lutz, Joshua
Montes, William B. Peterman, David Ratner, Daniel Villar, and Ahu Yildirmaz.
2020. “An Evaluation of the Paycheck Protection Program Using Administrative
Payroll Microdata.” Working Paper. Cambridge, Mass.: Massachusetts Institute
of Technology. https://economics.mit.edu/files/20094.
Barrot, Jean-Noël, and Ramana Nanda. 2020. “The Employment Effects of Faster
Payment: Evidence from the Federal Quickpay Reform.” Journal of Finance
75, no. 6: 3139–73.
Bartik, Alexander W., Marianne Bertrand, Feng Lin, Jesse Rothstein, and Matthew
Unrath. 2020. “Measuring the Labor Market at the Onset of the COVID-19
Crisis.” Brookings Papers on Economic Activity, Summer, 239–68.
Bartik, Alexander W., Zoe B. Cullen, Edward L. Glaeser, Michael Luca,
Christopher T. Stanton, and Adi Sunderam. 2020. “The Targeting and Impact of
Paycheck Protection Program Loans to Small Businesses.” Working Paper 27623.
Cambridge, Mass.: National Bureau of Economic Research. https://www.nber.
org/papers/w27623.
Blanchard, Olivier, Thomas Philippon, and Jean Pisani-Ferry. 2020. “A New Policy
Toolkit Is Needed as Countries Exit COVID-19 Lockdowns.” Policy Brief 20-8.
Washington: Peterson Institute for International Economics. https://www.piie.com/
publications/policy-briefs/new-policy-toolkit-needed-countries-exit-covid-19-
lockdowns.
Cajner, Tomaz, Leland D. Crane, Ryan A. Decker, John Grigsby, Adrian Hamins-
Puertolas, Erik Hurst, Christopher Kurz, and Ahu Yildirmaz. 2020. “The US Labor
Market during the Beginning of the Pandemic Recession.” Brookings Papers on
Economic Activity, Summer, 3–33.
Chava, Sudheer, Alexander Oettl, and Manpreet Singh. 2019. “Does a One-Size-
Fits-All Minimum Wage Cause Financial Stress for Small Businesses?” Working
Paper 26523. Cambridge, Mass.: National Bureau of Economic Research.
https://www.nber.org/papers/w26523.
Chetty, Raj, John N. Friedman, Nathaniel Hendren, Michael Stepner, and Oppor-
tunity Insights Team. 2020. “How Did COVID-19 and Stabilization Policies
Affect Spending and Employment? A New Real-Time Economic Tracker Based
on Private Sector Data.” Working Paper 27431. Cambridge, Mass.: National
Bureau of Economic Research. https://www.nber.org/papers/w27431.
Chodorow-Reich, Gabriel, Olivier Darmouni, Stephan Luck, and Matthew C.
Plosser. 2020. “Bank Liquidity Provision across the Firm Size Distribution.”
Working Paper 27945. Cambridge, Mass.: National Bureau of Economic
Research. https://www.nber.org/papers/w27945.

HUBBARD and STRAIN
377
Coibion, Olivier, Yuriy Gorodnichenko, and Michael Weber. 2020. “Labor
Markets during the COVID-19 Crisis: A Preliminary View.” Working Paper 27017.
Cambridge, Mass.: National Bureau of Economic Research. https://www.nber.
org/papers/w27017.
Davis, Steven J., and Till von Wachter. 2011. “Recessions and the Costs of Job Loss.”
Brookings Papers on Economic Activity, Fall, 1–55.
Erel, Isil, and Jack Liebersohn. 2020. “Does FinTech Substitute for Banks?
Evidence from the Paycheck Protection Program.” Working Paper 27659.
Cambridge, Mass.: National Bureau of Economic Research. https://www.nber.
org/papers/w27659.
Farooq, Ammar, Adriana D. Kugler, and Umberto Muratori. 2020. “Do Unemploy-
ment Insurance Benefits Improve Match Quality? Evidence from Recent U.S.
Recessions.” Working Paper 27574. Cambridge, Mass.: National Bureau of
Economic Research. https://www.nber.org/papers/w27574.
Farrell, Diana, Christopher Wheat, and Carlos Grandet. 2019. Place Matters: Small
Business Financial Health in Urban Communities. Washington: JPMorgan Chase
Institute. https://www.jpmorganchase.com/content/dam/jpmc/jpmorgan-chase-
and-co/institute/pdf/institute-place-matters.pdf.
Forsythe, Eliza, Lisa B. Kahn, Fabian Lange, and David Wiczer. 2020. “Labor
Demand in the Time of COVID-19: Evidence from Vacancy Postings and
UI Claims.” Journal of Public Economics 189, article no. 104238.
Gertler, Mark, and R. Glenn Hubbard. 1988. “Financial Factors in Business
Fluctuations.” In Economic Policy Symposium Proceedings: Financial Market
Volatility. Jackson Hole, Wyo.: Federal Reserve Bank of Kansas City.
Goolsbee, Austan, and Chad Syverson. 2020. “Fear, Lockdown, and Diversion:
Comparing Drivers of Pandemic Economic Decline 2020.” Working Paper 27432.
Cambridge, Mass.: National Bureau of Economic Research. https://www.nber.org/
papers/w27432.
Granja, João, Christos Makridis, Constantine Yannelis, and Eric Zwick. 2020.
“Did the Paycheck Protection Program Hit the Target?” Working Paper 27095.
Cambridge, Mass.: National Bureau of Economic Research. https://www.nber.
org/papers/w27095.
Hamilton, Steven, and Stan Veuger. 2020. “A Recession Is a Public Health
Necessity—Let’s Keep It Short.” London: Centre for Economic Policy Research,
VoxEU. https://voxeu.org/article/recession-public-health-necessity-let-s-keep-
it-short.
Hanson, Samuel G., Jeremy C. Stein, Adi Sunderam, and Eric Zwick. 2020a.
“Business Continuity Loans: Keeping America’s Lights on during the Pandemic.”
White Paper. Chicago: University of Chicago, Becker Friedman Institute. https://
bfi.uchicago.edu/wp-content/uploads/BFI_White-Paper_Zwick2_4.2020.pdf.
Hanson, Samuel G., Jeremy C. Stein, Adi Sunderam, and Eric Zwick. 2020b.
“Business Continuity Insurance: Keeping America’s Lights on during the
Pandemic.” Working Paper. Boston: Harvard Business School. https://www.

378
Brookings Papers on Economic Activity, Fall 2020
igmchicago.org/wp-content/uploads/2020/04/Business-Continuity-Insurance-
20200408-FINAL.pdf.
Hubbard, R. Glenn, and Michael R. Strain. 2020. A Business Fiscal Response to a
COVID-19 Recession. Washington: American Enterprise Institute. https://www.
aei.org/research-products/report/a-business-fiscal-response-to-covid-19-recession/.
Jackson, C. Kirabo. 2013. “Match Quality, Worker Productivity, and Worker
Mobility: Direct Evidence from Teachers.” Review of Economics and Statistics
95, no. 4: 1096–116.
Mortensen, Dale T., and Christopher A. Pissarides. 1999. “New Developments in
Models of Search in the Labor Market.” In Handbook of Labor Economics,
Volume 3, Part B, edited by Orley C. Ashenfelter and David Card. Amsterdam:
North-Holland.
Ozimek, Adam, and John Lettieri. 2020. “How to Rescue Main Street from Corona-
virus before It’s Too Late.” Blog post, March 18, Economic Innovation Group.
https://eig.org/news/main-street-rescue-and-resiliency-program.
Strain, Michael R. 2020. The Paycheck Protection Program: An Introduction.
Washington: American Enterprise Institute. https://www.aei.org/research-products/
report/the-paycheck-protection-program-an-introduction/.
US Small Business Administration, Office of Advocacy. 2019. “What’s New with
Small Business?” Infographic. Washington: US Small Business Administration.
https://cdn.advocacy.sba.gov/wp-content/uploads/2019/09/23172859/Whats-
New-With-Small-Business-2019.pdf.

379
Comment and Discussion
COMMENT BY
ERIC ZWICK  Hubbard and Strain offer a clear and comprehensive
assessment of the initial months of the Paycheck Protection Program (PPP).
My comment has two goals. First, I want to place the PPP into a framework
for evaluating the welfare effects of such programs. Second, I want to place
this analysis in the context of what we know from contemporaneous work
about whether the PPP succeeded. While I agree with the authors that it is
too soon to provide a complete grade of the PPP, we know enough now to
offer a provisional assessment that can guide ongoing policy debates and
future research on the program.
GRADING THE PPP: A RUBRIC  What kind of policy is the PPP? Many popu-
lar commentators have described the program as fiscal stimulus, but this
view is mistaken. The goal of the PPP was not to increase economic activity
immediately. In fact, if the initial crisis response was aimed at suppressing
the virus by reducing public interactions, the goal may well have been the
opposite: to encourage everyone to stay home to slow the virus’s spread
while supporting workers and firms during the lockdown.
Thus, the goal of the PPP was more to enable future economic activity—
by preserving firm-worker links and preventing permanent failures—than
to stimulate immediately. Hubbard and Strain are right to point out that in
this case we should not use metrics like cost per job or fiscal impact multi-
pliers to grade the program.
One might alternatively think of the program as support for capital
markets in the spirit of liquidity support programs aimed at the financial
system, which were pursued by the Federal Reserve and the Treasury both
during this crisis and during the Great Recession. Again, this is not the
right way to view the PPP. In particular, the lender-of-last-resort motivation
for policy intervention is not the right model when output evaporates and

380
Brookings Papers on Economic Activity, Fall 2020
there are real losses. In this world, loans are unattractive from a private and
public perspective because repayment is nonviable for many firms despite
having positive long-term prospects.1 In such a case, and in stark contrast
to the Great Recession, banks may be more unwilling than unable to lend
to firms that were hard hit by the crisis.
In my view, the closest policy analogy to the current situation is disaster
relief and insurance. Firms face a severe noneconomic shock that entails
an unusually low correlation between their short-run and long-run per­
formance. If these firms close, society will risk losing valuable firm-worker
matches, fixed startup costs already paid, and sweat equity already accu-
mulated. At the aggregate level, there is risk that congestion exter-
nalities in bankruptcy courts and in the labor market could exacerbate
economic losses.
In all of these respects, the pandemic is similar to a large storm that
devastates a local area, though this time the devastation is geographically
widespread, of uncertain duration, and wrought with lost revenues instead
of lost capital. Still, I believe that conceptualizing the PPP as a kind of
social insurance program helps to illuminate the framework we ought to
use to evaluate design and implementation.
REVENUE REPLACEMENT VERSUS BUSINESS CONTINUITY INSURANCE  An impor-
tant part of the paper is a discussion of the key design elements of a busi-
ness support program. If we think of the PPP as insurance, then we can
fruitfully debate these elements in terms of their insurance value beyond
being mere transfers. Let me contrast some of the authors’ preferred design
elements to our business continuity insurance proposal (Hanson and others
2020a), which was itself inspired by Hubbard and Strain’s earlier writing
on the program and by Emmanuel Saez and Gabriel Zucman’s buyer of last
resort proposal (Saez and Zucman 2020).
First is the question of what expenses to permit. The authors argue for
including payroll in the category of eligible expenses to be covered by the
PPP. Supporting payroll might help firms retain workers, thereby preserving
valuable firm-worker matches. It also might keep workers from claiming
unemployment insurance (UI) at a time when the risk of overwhelmed
UI systems is a concern. Both arguments amplify the insurance value of
the program.
1.  Hanson and others (2020b) argue why it makes more sense to think of the govern-
ment’s role as “venture capitalist of last resort,” that is, as needing to take on significant
credit risk in this crisis.

COMMENT and DISCUSSION
381
Nevertheless, I have several concerns with this element. First, including
payroll dramatically raises the cost of the program, thus limiting the amount
of time firms might be able to benefit from additional funds. Second, many
of the hardest hit firms were bars and restaurants in the service sector,
for which high rates of turnover in normal times imply the value of firm-
worker matches might be low. Third, it is somewhat unnatural to expect
firms to pay workers to be idle, when accounting and payroll systems are
based on hours worked and when many firms have some workers that
can still work. Fourth, though there were plenty of hiccups in the initial
rollout, the UI system actually worked pretty well in supporting more than
30 million workers! Last, as the authors note, including payroll deters
reallocation of workers across firms and industries by subsidizing newly
inefficient matches.
The second design question concerns how to deploy PPP funds. The
authors argue in favor of using banks as conduits to access the program.
Preexisting relationships between banks and firms might accelerate the
transfer of funds, and the underwriting infrastructure could help detect
fraud. These arguments speak to the efficiency and timeliness of the pro-
gram, both key aspects of good insurance.
At the same time, because banks do not have the same incentives as the
government, we have to pay them to participate. And banks might steer
preferred clients in one direction and new clients in another. While true
that many firms are connected to banks, there are many others without
prior relationships or whose banks were themselves disrupted by the lock-
down orders. My view is that the IRS could have been more involved in
implementing this policy, given its track record for large-scale stimulus in
other contexts (such as economic impact payments, refunds for net operating
losses, or the first-time homebuyer credit).
A third question concerns targeting. The authors argue that the program
should feature little or no targeting. Their logic is that, given the unknown
shock severity and duration, trying to narrow eligibility for the program in a
sophisticated way would fall prey to lobbying by connected industries. They
also worry that conditioning loan forgiveness on revenues might discourage
firms from reopening by subjecting them to high marginal tax rates.
Here, viewing the program through the lens of insurance is especially
instructive. Providing little or no targeting is both expensive and by defini-
tion allocates funds to low-insurance-value types. To the extent there is
a budget constraint at the federal level (which is debatable these days), we
are now in a position where benefits have been exhausted though help is
still needed. Moreover, it will surely strike many as unfair that firms that

382
Brookings Papers on Economic Activity, Fall 2020
were able to continue operating, or that operated in defiance of best public
health practices, received the same level of support as those that tempo-
rarily closed.
I should note that we agree on many design features and general prin-
ciples. For example, though the authors describe their proposal as “revenue
replacement,” it is better thought of as “net value added” replacement. In
other words, we agree to exclude profits, intermediates, and depreciation in
the list of eligible expenses. More fundamentally, we agree that the loans
should be closer to grants. We also agree that placing the demands solely
on the Small Business Administration (SBA) to implement this program
would not have made sense, given the agency’s size and the infrastructure
in place prior to the crisis.2 And we agree that the largest firms should be
treated less generously.
The bottom line in this discussion is that this area deserves more formal
study. Economics has a natural opportunity to contribute in this time of
crisis by improving our understanding of the optimal features of business
support policy.
DID FUNDS GO TO HIGH-INSURANCE-VALUE TYPES?  Ideally, we would grade
the PPP in terms of insurance value provided relative to the program’s cost,
but defining the notion of insurance value for firms is beyond the scope of
my discussion. As a first pass, let’s consider what we know about program
targeting and how firms used the funds.
Judged by its timeliness, the program receives high marks. More than
$500 billion in funds were distributed in just six weeks!
One reason why funds could be deployed so quickly is that nearly all
firms could apply. As a consequence, the targeting of the program was poor.
In work with João Granja, Christos Makridis, and Constantine Yannelis
(Granja and others 2020), I found that more of the program’s initial funds
actually flowed to regions that were less hard hit by the shock (figure 1, top
panel). This distribution is to a large extent due to differences across lenders
in their participation in the program. For example, the top four banks alone
account for 36 percent of total pre-policy small business loans but disbursed
less than 3 percent of all PPP loans in the first round of funding. Ultimately,
we find a weak correlation between initial shock severity and funding levels,
reflecting the program’s broad eligibility criteria (figure 1, bottom panel).
2.  Recently, the SBA inspector general released a report finding that the “unprecedented
demand for COVID-19 EIDLs [relief loans] and the equally unprecedented challenges SBA
had in responding to this pandemic combined with lowered controls resulted in billions
of dollars in potentially fraudulent loans and loans to potentially ineligible businesses”
(SBA 2020, 2).

COMMENT and DISCUSSION
383
0.25
0.30
0.35
Share of businesses shut down during 03/22 to 03/28 (pre-PPP)
Least
PPP/Est
2
3
4
5
6
7
8
9
Most
PPP/Est
B. Rounds 1 and 2
Quantile of PPP loans relative to all establishments (both rounds)
Quantile of PPP loans relative to all establishments (round 1)
A. Round 1
0.25
0.30
0.35
Share of businesses shut down during 03/22 to 03/28 (pre-PPP)
Least
PPP/Est
2
3
4
5
6
7
8
9
Most
PPP/Est
Source: Granja and others (2020).
Notes: The figure stratifies all businesses in Homebase in ten bins based on the fraction of
establishments in their zip code receiving PPP during the first round and during both rounds combined.
The figure plots for each bin the share of Homebase businesses that shut down in the week of
March 22–March 28, that is, prior to the PPP.
Figure 1.  Weak Geographic Targeting of the PPP

384
Brookings Papers on Economic Activity, Fall 2020
Additional evidence on targeting and design elements comes from firm
surveys (Alekseev and others 2020). Drawing on data from a large survey
of business owners on Facebook, Alekseev and others (2020) find that
30–40 percent of small businesses did not experience sales declines in the
first month of the crisis. Among the businesses that did experience declines,
the severity of the decline varies widely from declines of 10–20 percent
to nearly complete shutdowns. Moreover, only half of firms surveyed
reported struggling to pay obligated expenses (though presumably this share
increased over time). Such heterogeneity in experiences is at odds with the
one-size-fits-all design of the program’s forgiveness formula.
The evidence also points to other issues with the PPP’s rollout. First,
firms that do report struggling to make payments appear to struggle equally
to pay rent, wages, and interest on loans. This fact suggests the initial weight
of 75 percent on payroll expenses was likely too high. Second, half of
firms report not having preexisting relationships with banks as borrowers
(Alekseev and others 2020), which appears to have led to such firms initially
struggling to access the program and eventually switching lenders in order
to receive funds (Rudegeair 2020). Data from the SBA show that, across
participating banks, larger firms received funding first, and many smaller
borrowers had to wait several months to access the program (Granja and
others 2020).
WHAT DID FIRMS DO WITH THE MONEY?  The authors focus on a second
approach to evaluating the PPP’s success to date. They use data from Dun &
Bradstreet and a series of research designs to evaluate the impact of the
program on employment, financial performance, and business closures.
Their most compelling research design exploits the eligibility threshold of
500 workers; with some exceptions, firms above this threshold could not
apply for PPP loans, providing a natural control group.
The authors find modest but statistically significant effects on employ-
ment that appear to grow over the six-month period their data cover. These
employment effects are largely consistent with Autor and others (2020) and
Chetty and others (2020), who use the same research design and data from
different payroll processors. Using a bank exposure design at the regional
level, Granja and others (2020) find null effects of the program in April that
grow to modest effects in May and June. Across studies, the results appear
to imply very high cost-per-job estimates (about $200,000 per job), though
recall this metric is probably the wrong way to evaluate this program.
The authors find modest impacts of the program in reducing financial
vulnerability and business closures. These findings contrast somewhat with
Granja and others (2020), who find no impact on business shutdowns and

COMMENT and DISCUSSION
385
large effects on firms’ reported cash on hand and propensity to miss obli-
gated loan and other payments. On the other hand, Bartik and others (2020)
use a bank exposure design and find that the PPP had large effects on firms’
own forecasts of failure probabilities. Reconciling these contrasting find-
ings is a task for future research, as more data become available.
I do have a few concerns with the authors’ approach. First, in some
analyses, they report differences that compare loan applicants to non­
applicants. These differences will tend to overstate program impacts because
they do not isolate loan demand effects, which are likely to be correlated
with business expenditure plans. For these reasons, I prefer estimates using
the worker threshold as an instrument.
Second, the aggregate time series in the Dun & Bradstreet data are
extremely stable and indicate very limited aggregate impact of the pan­
demic and lockdowns on firms in the sample. These patterns stand in
sharp contrast to pretty much every other real-time data set available,
in which employment appears to fall by between 20 and 60 percent depend-
ing on the sample of interest. I worry that the patterns in the Dun &
Bradstreet data are an artifact of stale or incomplete measurement and
updating. Moreover, if such measurement issues are more pronounced
for small firms, this problem will confound estimates that compare firms
across size thresholds. My bet is the Dun & Bradstreet data will ultimately
be more useful for evaluating the question of permanent closures, once
data are comprehensively updated.
PENCILS DOWN  The ultimate grade for the PPP will depend on the medium-
term impacts that have yet to materialize. We see modest short-term employ-
ment effects but more significant improvement in firm balance sheets. The
PPP’s success will hinge on whether the cost of limited targeting is ulti-
mately offset by the gains from preventing a large number of firm failures.
If instead a large share of the funds prove inframarginal, the economic inci-
dence of the program will fall largely on business owners, many of whom
would have been able to weather the storm without this support.
REFERENCES FOR THE ZWICK COMMENT
Alekseev, Georgij, Safaa Amer, Manasa Gopal, Theresa Kuchler, J. W. Schneider,
Johannes Stroebel, and Nils C. Wernerfelt. 2020. “The Effects of COVID-19
on U.S. Small Businesses: Evidence from Owners, Managers, and Employees.”
Working Paper 27833. Cambridge, Mass.: National Bureau of Economic
Research. https://www.nber.org/papers/w27833.

386
Brookings Papers on Economic Activity, Fall 2020
Autor, David, David Cho, Leland D. Crane, Mita Goldar, Byron Lutz, Joshua
Montes, William B. Peterman, David Ratner, Daniel Villar, and Ahu Yildirmaz.
2020. “An Evaluation of the Paycheck Protection Program Using Administra-
tive Payroll Microdata.” Working Paper. Cambridge, Mass.: Massachusetts Insti-
tute of Technology. https://economics.mit.edu/files/20094.
Bartik, Alexander W., Zoe B. Cullen, Edward L. Glaeser, Michael Luca, Christopher
T. Stanton, and Adi Sunderam. 2020. “The Targeting and Impact of Pay­check
Protection Program Loans to Small Businesses.” Working Paper 27623. Cam-
bridge, Mass.: National Bureau of Economic Research. https://www.nber.org/
papers/w27623.
Chetty, Raj, John N. Friedman, Nathaniel Hendren, Michael Stepner, and Oppor-
tunity Insights Team. 2020. “How Did COVID-19 and Stabilization Policies
Affect Spending and Employment? A New Real-Time Economic Tracker Based
on Private Sector Data.” Working Paper 27431. Cambridge, Mass.: National
Bureau of Economic Research. https://www.nber.org/papers/w27431.
Granja, João, Christos Makridis, Constantine Yannelis, and Eric Zwick. 2020. “Did
the Paycheck Protection Program Hit the Target?” Working Paper 27095.
Cambridge, Mass.: National Bureau of Economic Research. https://www.nber.
org/papers/w27095.
Hanson, Samuel G., Jeremy C. Stein, Adi Sunderam, and Eric Zwick. 2020a.
“Business Continuity Insurance: Keeping America’s Lights on during the
Pandemic.” Working Paper. Boston: Harvard Business School. https://www.
igmchicago.org/wp-content/uploads/2020/04/Business-Continuity-Insurance-
20200408-FINAL.pdf.
Hanson, Samuel G., Jeremy C. Stein, Adi Sunderam, and Eric Zwick. 2020b.
“Business Credit Programs in the Pandemic Era.” In the present volume of
Brookings Papers on Economic Activity.
Rudegeair, Peter. 2020. “When Their PPP Loans Didn’t Come Through, These Busi-
nesses Broke Up with Their Banks.” Wall Street Journal, July 31. https://www.
wsj.com/articles/when-their-ppp-loans-didnt-come-through-these-businesses-
broke-up-with-their-banks-11596205736.
Saez, Emmanuel, and Gabriel Zucman. 2020. “Keeping Business Alive: The Gov-
ernment Will Pay.” Research Brief. Economics for Inclusive Prosperity. https://
econfip.org/wp-content/uploads/2020/03/20.Keeping-Businesses-Alive.pdf.
US Small Business Administration, Office of the Inspector General (SBA). 2020.
Inspection of Small Business Administration’s Initial Disaster Assistance
Response to the Coronavirus Pandemic. Report no. 21-02, October 28. https://
www.sba.gov/sites/default/files/2020-10/SBA%20OIG%20Report%2021-02.pdf.
GENERAL DISCUSSION    Caroline Hoxby began the discussion by
thanking the authors for a terrific paper and Eric Zwick for his excellent
discussion. She agreed with Zwick’s concern that the Dun & Bradstreet
data used in the paper are not very sensitive and are infrequently updated

COMMENT and DISCUSSION
387
in comparison to other monthly sources of data. Hoxby then remarked that
the specification in column 5 of tables 4 through 6, which compare firms
with 400 to 475 employees to firms with 525 to 600 employees, was most
persuasive because the control group and treatment group are similar and
therefore most comparable. Her concern, however, was that the results
for this specification showed very little. Hoxby closed her comments by
lamenting that the eligibility threshold was 500 employees because that
doesn’t allow the authors to identify the effects on very small firms, such as
restaurants and small family businesses. She wondered whether any eligi-
bility or identification strategy would allow the authors to look specifically
at very small firms.
Joshua Gotbaum asked, via the teleconferencing chat function, whether
shifting the program from a wage subsidy to a revenue subsidy made the
distributional effects of the Paycheck Protection Program (PPP) regressive,
at least in comparison to the European programs.
Gabriel Chodorow-Reich thanked the authors and the discussant and
followed up on Zwick’s question regarding what firms did with the PPP
money. He described his ongoing research with Darmouni, Luck, and
Plosser that finds midsize firms were making large repayments on credit
lines in the second quarter of 2020, which may be tied to PPP receipt.1
Chodorow-Reich wondered whether the authors’ view of the program’s
success might be affected if it is true that the same businesses who were
getting PPP loans were paying down credit lines.
Jason Furman stated that he would like to see the employment impact
of the PPP in December before gauging the PPP’s success. He noted that
workers could have been given unemployment insurance (UI), which
would have saved the firms the cost of their employment. Furthermore,
assuming employment is determined by comparing the marginal cost of
hiring someone to the firms’ marginal revenue product, then the PPP from
earlier in the year will have no effect on employment in December. Furman
stated that the effects the authors found could have been achieved more
cheaply and efficiently through UI. For the PPP to have a lasting effect, he
argued, it would need to keep some firms from going bankrupt such that
the firm is employing more people in December than they otherwise would
have. He concluded that whether a firm is hiring someone in December
1.  Gabriel Chodorow-Reich, Olivier Darmouni, Stephan Luck, and Matthew C. Plosser,
“Bank Liquidity Provision across the Firm Size Distribution,” Working Paper 27945
(Cambridge, Mass.: National Bureau of Economic Research, 2020).

388
Brookings Papers on Economic Activity, Fall 2020
is more likely to relate to the firm’s demand than whether they received a
lump sum transfer many months earlier.
Austan Goolsbee praised the authors’ work but cautioned them on the
potential confounds correlated with size. He described recent work with
Syverson using phone data of consumer visits to millions of businesses
which finds a strong trend of people shopping at smaller, less busy stores
correlated with the rise of the pandemic.2 Goolsbee noted that, given the
paper’s use of size to separate treatment and control groups, this would
confound the result of the paper; it may look like PPP is benefiting smaller
firms when in fact this is an unobserved effect related to increases in rela-
tive demand. Goolsbee concluded by offering to share the phone record
data with the authors and suggesting they might be able to match some
firms in their sample.
Steven Davis began by following up on Furman’s point regarding the
cost of the PPP program relative to other systems of distributing support to
workers. He restated that the cost per worker of using the PPP to support
workers is much larger than it would have been through the UI system.
He asked whether the PPP was better than UI at preserving worker-firm
matches in some way, and if so in what way. Relatedly, he pointed out
that it’s not clear that retaining worker-firm matches is valuable given the
kinds of employees who work for PPP-targeted firms. Davis noted that the
displaced worker literature largely focuses on mass layoff events affecting
high-tenure workers at large firms, typically in industrial jobs. He con-
trasted this with the jobs lost in the pandemic, which he argued are very
different, involving much less match-specific capital and match-specific
rents than workers studied in the displaced worker literature.
Davis went on to note that he knows little about the value of business
continuity for the population targeted by the PPP. He suggested that if there
is great evidence that business continuity is of high value beyond the value
to the business owner, the discussion should emphasize this. Davis also
wondered if the PPP, by sustaining incumbent firms, crowds out new firms.
He noted that the Census Bureau now draws on administrative records to
tabulate monthly business formation statistics by state and by industry. He
suggested it would be interesting to use these data to investigate whether
industries and states that received more PPP support show evidence of
crowding-out effects in the form of weaker business formation.
2.  Austan Goolsbee and Chad Syverson, “Fear, Lockdown, and Diversion: Comparing
Drivers of Pandemic Economic Decline,” Working Paper 27432 (Cambridge, Mass.: National
Bureau of Economic Research, 2020).

COMMENT and DISCUSSION
389
Davis concluded by noting that the PPP, the $600 federal benefit supple-
ment, and other recent expensive fiscal programs have been motivated by
the need to compensate for administrative weaknesses in the UI system. To
that end, he wondered what the cost of fixing the administrative weaknesses
would be and posited it might cost a few tens of billions of dollars, which
is much less than the trillion dollars spent on the current programs. Davis
remarked that this would likely belong in a separate paper from the authors’
current one.
Katharine Abraham seconded Davis’s point regarding the administrative
capacity of the UI system, noting that the system infrastructure is fragile
and inflexible. She lamented that many states’ UI systems still use COBOL
software and many more couldn’t easily be modified to pay workers a higher
replacement rate rather than a flat supplement to their benefit amount.
Abraham then followed up on Furman’s and Davis’s comments stating
that it would have been cheaper to allow individuals to be laid off and get
UI than to keep them attached to their jobs through the PPP. Besides the
value of business continuity and increasing the likelihood of workers still
being employed in December, she stated, there is also value in reducing
unemployment. Even beyond the associated loss of income, unemployment
imposes significant mental and physical costs on affected individuals.
Abraham argued that the value of the PPP thus includes the value to workers
of having a job, even if the job eventually ends. She concluded that this
should affect how the authors consider the policy’s effectiveness.
Glenn Hubbard thanked the discussant and the audience for their ques-
tions. In response to the comments regarding alternatives to the PPP, he
reiterated that the value of worker-firm matches and business continuity
is still unknown, making it difficult to decide whether it would have been
better to put people on UI, irrespective of the UI system’s administrative
deficiencies.
Hubbard then agreed that the authors hoped to find an effect using the
specification Hoxby noted she most preferred, found in column 5 of the
results tables. Hubbard added that the authors suspect this specification did
not show any significant results because this specification focuses on rela-
tively larger firms and the effects are mostly found in much smaller firms.
Hubbard concluded by addressing a question in the teleconferencing
chat function regarding targeting. Hubbard noted that it would have been
preferable to know exactly who could have been helped because that would
have been more efficient, but there is no way to know. Given that the value
of worker-firm matches and business continuity are also unknown, Hubbard
stated that it isn’t obvious what the alternatives to the PPP were.

390
Brookings Papers on Economic Activity, Fall 2020
Michael Strain also thanked the discussant and audience for their com-
ments. He then noted that the authors attempted to evaluate the program
empirically, using specifications with different treatment and control
groups, using treatment-on-the-treated effects and intent-to-treat effects.
One such specification, he notes, is comparing only firms with one to
250 employees, using loan application as the treatment. Strain argued that
this specification addresses some of the concerns about comparing firms
with one to 500 employees to firms with 501 to 1,000 employees. However,
he also acknowledged that their data only show firms that applied for a
loan greater than $150,000, which muddles the control group. Strain added
that there could also be issues with comparing groups based on eligibility
because the PPP likely affected firms with 450 employees very differently
from firms with 15 employees. The authors attempted to navigate these
empirical challenges accordingly, he said.
Strain then turned to Furman and others’ point regarding alternatives to
the PPP. He stated that part of the program’s goal was to preserve the pro-
ductive capacity of the economy over the medium term until the pandemic
had subsided. Moreover, the program was promoted as an employment pro-
gram which would make comparisons to UI natural, but he argued it was
more along the lines of a continuity program for businesses. For example,
a restaurant might survive the economic downturn due to the PPP, Strain
said. He added that it’s reasonable to believe that the PPP loan doesn’t affect
how many workers the restaurant hires in nine months; however, the con-
clusion shouldn’t be that the PPP failed, he argued, because the business
survived. Perhaps without the revenue replacement the restaurant received
from the grant the business otherwise wouldn’t be there at all, he said.
Strain concluded that when comparing the costs of the PPP to UI, it is
crucial to consider whether the period of economic weakness was shorter
because the PPP supported labor demand over the medium term during the
transition by averting millions of business closures that otherwise would
have happened. He noted that while this is difficult to quantify, comparing
UI costs to the PPP would need to account for these concerns holistically.

File and source

File

Hubbard-Strain_Brookings-BPEA_has-ppp-succeeded.pdf

Size

336,403 bytes

SHA-256

87b0fe1c748a5a5a8af78adb26668bf6ac050bb50ce289016e3598c7b709ee78

Our copy

Hubbard-Strain_Brookings-BPEA_has-ppp-succeeded.pdf

Original

www.brookings.edu