Discrimination in Lending? Evidence From the Paycheck Protection Program — Small Business Economics (2022)

Summary

Small Business Economics (2022) 58:843-865, Discrimination in lending? Evidence from the Paycheck Protection Program, is by Rachel Atkins, Lisa Cook and Robert Seamans, accepted 21 June 2021 and published online 5 July 2021. The abstract says the authors assess the role of race in loans made through the Paycheck Protection Program (PPP) and, using newly released data, find that Black-owned businesses received loans approximately 50% lower than observationally similar White-owned businesses. It says the effect is marginally smaller in areas with more bank competition and disappeared over time as changes to the program allowed entry by fintechs and other non-traditional lenders. The authors recommend that policy makers account for existing racial inequalities within banking or other systems in program design.

Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used

Issuer: Source documents
Document type: Discrimination In Lending Evidence From The Paycheck Protection Program
Case: Discrimination In Lending Evidence From The Paycheck Protection Program

Cite this document

“Discrimination in Lending? Evidence From the Paycheck Protection Program — Small Business Economics (2022).” Original at europepmc.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/discrimination-in-lending-evidence-from-the-paycheck-protection-program/, SHA-256 31e86da59e080523a4f5947304d16266fa13499749e2a70263b589ca5c90962c.

Full text

Small Bus Econ (2022) 58:843–865
https://doi.org/10.1007/s11187-021-00533-1

Discrimination in lending? Evidence from the Paycheck
Protection Program
Rachel Atkins · Lisa Cook · Robert Seamans

Accepted: 21 June 2021 / Published online: 5 July 2021
© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2021

Abstract We assess the role of race in loans made and began approving PPP loans. Loan size differences
through the Paycheck Protection Program (PPP). The were also slightly smaller in zip codes containing a
PPP program, created by the U.S. government as a larger number of bank branches. These results are
response to the Covid-19 pandemic, provides loans consistent with prior research which shows lending
to small businesses so they can keep employees on discrimination by commercial banks against Black
their payroll. We argue that the historical record and borrowers. It is also consistent with studies showing
PPP program design choices made it likely that many that greater access to and competition among banks
Black-owned businesses received smaller PPP loans and other lenders can reduce discrimination. In light
than White-owned businesses. Using newly released of these results we recommend that policy makers
data on the PPP program, we find that Black-owned account for existing racial inequalities within banking
businesses received loans that were approximately or other systems in their program design to produce
50% lower than observationally similar White-owned more equitable outcomes.
businesses. The effect is marginally smaller in areas
with more bank competition and disappeared over Keywords Minority-owned firms · Racial
time as changes to the PPP program were imple- discrimination · Discrimination in financial markets ·
mented allowing for entry by fintechs and other non- Overcoming discriminatory barriers
traditional lenders.
JEL Classifications G21 · J16 · L2 · L26
Plain English Summary We find that Black-owned
businesses received loans through the Paycheck Pro-
tection Program that were approximately 50 percent 1 Introduction
lower than White-owned businesses with similar
characteristics. However, this difference in loan size The Covid-19 pandemic caused one of the sharpest
shrank over time as more non-bank lenders such as economic downturns in U.S. history. Mandated and
fintechs were allowed to participate in the program voluntary stay-at-home orders decimated revenues for
small businesses, leaving them unable to meet their
R. Atkins · R. Seamans (*) expenses and causing massive increases in temporary
NYU Stern School of Business, New York City, USA and permanent firm closures. The Paycheck Protec-
e-mail: rcs10@stern.nyu.edu tion Program (PPP) was one of the many policy inter-
ventions by the U.S. government during the crisis.
L. Cook
Michigan State University and NBER, East Lansing, USA Between April 10, 2020 and August 8, 2020, PPP

Vol.:(0123456789)
13
844 R. Atkins et al.

provided nearly 5.2 million loans through nearly 5500 differentials in loan amounts. Black-White differences
lenders totaling $525 billion for small businesses in PPP loan amounts decrease in areas with more
(500 employees or less) that could be used for operat- commercial bank branch locations, but this difference
ing expenses including payroll, mortgage interest or is not statistically significant. Thus, we find limited
rent, utilities, and approved expenses.1 The program evidence consistent with Becker’s (1957) theory that
was administered using established financial institu- competition in commercial credit markets reduces
tions and initially relied heavily on commercial banks taste-based discrimination. Finally, we study how
to distribute the funding. changes to the PPP program affected Black-White dif-
Prior research provides much evidence of racial ferences in loan amounts. We find that over time the
discrimination against Black-owned businesses by Black-White difference in loan amounts decreases,
banks and other commercial lenders, raising con- and this seems due in particular to a change in the
cerns about whether Black-owned businesses would PPP program that approved more non-bank lenders to
also face discrimination while seeking PPP funds. participate in the program, including fintechs.
Early reports from media outlets and minority-owned One challenge with the PPP loan data is that race
business advocates highlighted the difficulties Black is reported for only 10%of the loans. To address this
business owners faced when attempting to access PPP challenge, we conduct our analyses on both the full
loans. Of surveyed Black business owners who sought set of loans and on the subset of loans for which race
assistance through the PPP, it is reported that 41% did is reported. When working with the full set of loans,
not receive the funds and an additional 21% never we treat “unanswered race” as the omitted category
heard back from lenders (Global Strategy Group, and compare loan sizes between Black and White
2020). A New York Times report based on data from business owners (both relative to the unanswered race
the Small Business Administration (SBA) and inter- category). When working with the subset of loans for
views with small businesses argues that a lack of con- which race is reported, we treat White business owner
nections with banks, together with some of the PPP’s as the omitted category and consider how Black loan
rules, contributed to this low level of access (Trevizo, sizes compare to the omitted White category. We
2020). However, Humphries et al. (2020) present data also include a Mills ratio derived from a first-stage
from the first round of PPP funding indicating that Heckman selection approach to account for selection
smaller businesses generally (not accounting for race) effects. The results are similar across both of these
were less aware of the program, faced longer process- approaches.
ing times, and were less likely to have their applica- Our analysis makes several contributions. To our
tions approved than larger businesses. They attribute knowledge, this is the first study to systematically
this skewness to the “first-come, first-served” nature examine whether there were differences in PPP loans
of the program design. Thus, it is unclear whether the to White- and Black-owned businesses or to inves-
reported difficulties that Black business owners faced tigate evidence of lending discrimination as a result
while trying to access PPP funds was due to race or of the PPP program. Black-owned businesses were
due to other factors such as firm size. among the hardest hit in the wake of the Covid-19
In this paper, we use newly released PPP loan crisis; hence, this study offers insight into whether the
data from the Small Business Administration (SBA) federal government’s primary intervention reached
to investigate whether Black business owners were one of the most vulnerable groups of business own-
disproportionately underserved by the program. Our ers. Estimates from the first few months of the cri-
results indicate that Black-owned businesses received sis indicate that business activity fell 41% among
loan amounts that were approximately 50% lower Black-owned businesses, the largest drop of any racial
than observationally similar White-owned businesses. sub-group (Fairlie, 2020). Nearly 20% of Black firm
We also investigate whether differences in the num- owners surveyed reported temporarily closing their
ber of banks in an area contributed to Black-White businesses due to Covid-19 (Global Strategy Group,
2020). Though language in the CARES Act indicates
that Congress explicitly intended for the PPP to prior-
1
https://www-sba-gov.proxy.library.nyu.edu/sites/default/files/ itize businesses owned by “socially and economically
2020-08/PPP_Report%20-%202,020-08-10-508.pdf. disadvantaged individuals” (U.S. Congress, 2020),

13
Discrimination in lending? Evidence from the Paycheck Protection Program 845

we show that the program fell short of meeting this features. Section 3 reviews literature on lending dis-
stated goal with respect to Black-owned businesses. crimination and access to bank financing and then
We contribute to the nascent body of literature on presents hypotheses. In Sect. 4 we describe our data
the PPP program (Fairlie & Fossen, 2021; Hubbard and analytical methods. We present results in Sect. 5
& Strain, 2020; Humphries et al., 2020; Li, 2020; and Sect. 6 concludes with recommendations for pol-
Shuai et al., 2020), small business outcomes during icy and future research.
Covid 19 (Fairlie, 2020; Kalogiannidis, 2020; Ong
et al. 2020), and government intervention during
ongoing economic crises (Feyrer & Sacerdote, 2011; 2 The Paycheck Protection Program
Chodorow-Reich et al., 2012; Parker et al., 2013) by
focusing on the challenges faced by Black business The Covid-19 pandemic caused one of the sharpest
owners and by governments seeking to reach them economic downturns in U.S. history. According to the
during crises. Prior research on racial equity in the Bureau of Economic Analysis (BEA), gross domes-
PPP program examined whether loans were allocated tic product dropped by about 33% in the 2nd quarter
to communities based on the share of minority resi- of 2020.2 Following guidelines from public health
dents or businesses in a geographic location. Calem officials, state and local governments issued a variety
and Freedman (2020) find a positive relationship of directives including stay-at-home orders, manda-
between a community’s minority share of residents tory business closures, and social distancing require-
and receipt of PPP loans. Likewise, Fairlie and Fossen ments. Even in locations where governments did not
(2021) report that the relationship between the minor- mandate tight restrictions, residents often voluntarily
ity share of a community’s residential or business followed public health guidance by social distanc-
population and PPP loan receipt was positive but ing, avoiding large gatherings, and working from
small in magnitude. However, they show that loan home where possible. These mandated and voluntary
size per employee had a negative relationship with actions decimated revenues for small businesses leav-
minority share of the population. We build on these ing them unable to meet their expenses causing mas-
existing studies by conducting our analysis at the firm sive increases in temporary and permanent firm clo-
rather than community level. sures (Fairlie, 2020). Chetty et al. (2020) show that
We also contribute to literature on racial discrimi- much of the effects were driven by voluntary actions;
nation in small business lending (Bates & Robb, state mandated re-openings had little effect on spend-
2013; Bates et al., 2018; Blanchard et al., 2008; Chen ing and employment. Many businesses were forced to
et al., 2021) by assessing whether government inter- furlough or lay off employees which led to massive
ventions in small business credit markets mirror or increases in unemployment (Bartik et al., 2020).
ameliorate discrimination and through what mecha- The U.S. federal government created multiple pol-
nisms. Our findings indicate that if governments uti- icy interventions in response to the pandemic under
lize existing systems to enact policy or administer the CARES Act, including the PPP. PPP was designed
programs without accounting for the pre-existing both to slow or stabilize the rate of small business
inequities in those systems, then the enacted poli- closures and to reverse the rise in unemployment
cies and programs will likely reproduce those same rates. In Round One of the PPP, Congress appropri-
inequities. The SBA relied heavily on commercial ated $349 billion for small businesses in the form
banks to administer the PPP. As we describe below, of loans that could be used for operating expenses
these are institutions with a long history of racial including payroll, mortgage interest or rent, utilities,
discrimination, and Black-owned firms therefore had and approved expenses. They attempted to curb rising
less access to PPP funds. When the SBA expanded unemployment by offering to forgive PPP loans pro-
the number and types of participating lenders, Black- vided firms retain their employees or quickly rehire
owned businesses gained access to more funding furloughed or previously laid-off workers. Moreover,
through the PPP.
The remainder of this article proceeds as follows:
Sect. 2 presents an overview of the PPP’s history and 2
https://www.bea.gov/news/2020/g ross-domestic-product-
2nd-quarter-2020-advance-estimate-and-annual-update.

13
846 R. Atkins et al.

to reach a broad swath of workers through the PPP, Chetty et al. (2020) estimate that the PPP loans
Congress made loans widely available to employers increased employment at small businesses by only
including businesses, nonprofits, Tribal businesses, about 2% (with an implied cost of $377,000 per job
and veterans’ organizations with 500 employees or saved). Reports speculate that since lenders were able
less. to extract more in fees from larger businesses more
The program was administered by the Federal quickly, they were dis-incentivized from granting
Small Business Administration (SBA). However, loans requested by smaller businesses (Egan, 2020).
as always, SBA does not make loans to businesses Data from the Census Small Business Pulse Survey
directly and relies on SBA-qualified lenders which (SBPS) also identify challenges many small busi-
include depository institutions, such as banks and nesses faced when trying to access PPP loans. While
credit unions, as well as non-depository institution 75% of small businesses applied for PPP loans, only
lenders, such as some Community Development 38% received them (Buffington et al., 2020). According
Financial Institutions (CDFIs). Though loan recipi- to Biz2Credit, prior to the pandemic small business
ents are not charged fees, SBA lenders can still charge loan approval rates were about 28% at large banks
fees associated with the loans which were paid by the and 50% at small banks.4 Furthermore, since the PPP
federal government. SBA began accepting PPP loan was designed as a “first-come-first-served program”
applications on April 3, 2020. By April 16, 2020, (Granja et al., 2020), small businesses may have been
funding had already run out. Congress acted by allo- at a disadvantage given their lag time in knowledge
cating an additional $310 billion in funding for a sec- about the program’s eligibility requirements and ben-
ond round of PPP loans. In Round Two, Congress efits. Indeed, preliminary survey evidence reveals that
made a few modifications, including allowing the smaller firms had less access to information about
smallest businesses and other businesses that were PPP and other programs when compared with large
initially to be targeted to access program funding first. firms and that this information gap did not close
The Federal Reserve created the Paycheck Protec- much over time (Humphries et al., 2020).
tion Program Liquidity Facility (PPPLF) to give a Black-owned firms, which were already in a more
financial boost to the PPP. The PPPLF provided credit fragile financial state relative to their White coun-
to SBA lenders using PPP loans as collateral. In doing terparts, were particularly hard hit by the Covid-19
so, the Federal Reserve supplied added liquidity and crisis. This is partly due to Black-owned firms being
thus expanded the capacity of financial institutions to more likely to operate in environments that produce
make PPP loans. When announcing that the PPPLF poor business outcomes (McKinsey, 2020). Prior
was fully operational on April 16, 2020, the Federal to the pandemic, 58%of Black-owned businesses
Reserve had only made the facility available to deposi- reported being at risk or financially distressed as com-
tory institutions. By April 30th, it had extended access pared with only 27% of White-owned firms (Mills &
to all SBA-authorized lenders. This later action pro- Battisto, 2020). Black-owned firms are heavily con-
vided credit to smaller non-depository institutions that centrated in sectors most negatively affected by the
are more likely to lend in communities underserved by pandemic—accommodation and food service, retail,
commercial banks, such as CDFIs and fintech opera- and healthcare and social assistance. Fairlie (2020)
tors (Liu & Volker, 2020). Furthermore, these non- reports findings from the April 2020 Current Popula-
traditional lenders benefited disproportionately from tion Survey that African American owned businesses
the PPPLF’s affordable credit thus bolstering the like- experienced a 41% decline in owner activity. These
lihood of their participation in the PPP. firms were among those most in need of relief under
Starting with the SBA Inspector General, many the PPP.
have highlighted the difficulties small business own-
ers faced when attempting to access PPP funds
(Global Strategy Group, 2020; Trevizo, 2020).3
4
https://www.globenewswire.com/news-release/2020/04/07/
3
In contrast, others have highlighted that many organizations 2012911/0/en/Small-Business-Loan-Approval-Rates-Plumm
received loans that probably should not have, including Shake eted- i n- M arch- 2 020- B iz2C r edit- S mall- B usin e ss- L endi n g-
Shack, Harvard University and others (Jeschke, 2020). Index.html.

13
Discrimination in lending? Evidence from the Paycheck Protection Program 847

Yet, early evidence suggests that minority busi- home equity were only associated with a higher prob-
ness owners faced unique challenges when attempt- ability of starting a business for White but not for
ing to access PPP loans (Flitter, 2020; Kurlyandchick, Black households.
2020). For example, of the 723 businesses in Salem Literature on discrimination presents two primary
MA that received PPP loans as of July 2020, only 7 types of discrimination: statistical and taste-based
identified as Black-owned (Leighton, 2020). One (Arrow, 1998; Fang & Moro, 2010). When engag-
local preschool franchise owner explained that his ing in statistical discrimination, loan officers may use
existing relationship with a commercial bank allowed race as a proxy for unobserved characteristics that are
him to acquire PPP funding but that “many Black correlated with both race and the probability of loan
owners don’t have relationships with banks.” He also default. We do not expect PPP lenders to engage in
noted that “banks would choose to focus on their statistical discrimination because the SBA guarantee
customers first” given the volume of loan requests on all PPP loans shifts 100% of the default risk from
brought on by Covid-19 and the PPP. Thus, one might private banks to the federal government. Instead, we
predict disproportionately lower access to PPP loans surmise that any observed discrimination is taste-
for Black-owned firms given that loans originated based, stemming from either conscious or uncon-
primarily with commercial banks (at least initially), scious racial biases rather than being statistical in
because these institutions traditionally underserve the nature.
Black community in at least two ways: (1) discrimi- Banks discriminate against potential borrow-
nation in lending decisions and (2) restricted access ers through at least four different mechanisms: loan
to bank branch locations. In the section that follows, application denial, discouraging borrowers, demand-
we describe prior literature that provides more detail ing a price premium, and supplying lower quanti-
on these two ways, and then draw from this literature ties of credit. Data from the 2014 Annual Survey of
to derive formal hypotheses about differences in PPP Entrepreneurs show that among loan applicants, 53%
loans for Black- and White-owned businesses. of Black firm owners were denied loans as compared
with only 24.7% of White firm owners. Studies using
the Survey of Small Business Finance (SSBF) find
3 Prior literature and hypothesis development that Black firm owners were twice as likely to be
denied credit than their White counterparts (Blanchflower
3.1 Discrimination in lending et al., 2003; Cavalluzzo & Wolken, 2005). Fairlie
et al., (2020) address critiques of analysis using these
Traditional commercial banks have a long history cross-sectional data sources by examining access to
of racial discrimination (Blanchflower, et al., 2003; capital among startups in the longitudinal Kauffman
Blanchard et al., 2008; Hu et al., 2011; Bates & Firm Survey. Controlling for both credit scores and
Robb, 2013). As a result, Black and other non-White net worth, they show that Black-owned ventures were
business owners enjoy less access to credit markets. three times less likely than Whites to report that their
Chatterji and Seamans (2012) attribute the finding loan applications were always approved.
that credit card deregulation had a positive effect on Fairlie et al., (2020) also estimate higher levels
Black entrepreneurship to Black entrepreneurs having of discouraged borrowing among Black firm owners
less access to capital from traditional sources such as even among those deemed “creditworthy.” Discour-
banks, where they faced discrimination in lending. aged borrowers report not applying for credit for fear
Their analysis shows greater post-deregulation rates they will be turned down. Though Fairlie is not able
of Black entrepreneurship in areas with higher levels to observe why discouraged borrowers believe they
of historical discrimination. Similarly, Atkins (2021) will be denied, accounts from audit studies illustrate
finds evidence that Black households are less able how Black firm owners are treated differently by
than Whites with similar amounts of home equity to bank employees. Lederer et al. (2020) conducted an
access commercial credit to finance new ventures. audit study of financial institutions in Washington DC
Since home equity is a useful source of collateral between April 27 and May 29, 2020. They sent 63
lowering lender risk, it should reduce statistical dis- matched pairs of Black and White testers to 32 bank
crimination. However, Atkins finds that increases in branches and found that in 44% of tests where the

13
848 R. Atkins et al.

testers requested PPP loans, White testers received to PPP loans for Black-owned firms. Since the federal
more favorable treatment. Black testers were more government assumes 100% of the default risk asso-
likely to be discouraged from applying for a loan or ciated with PPP loans, we argue that any observed
were more likely to be steered towards a home equity discrimination is more likely to stem from racial bias
line-of-credit product rather than SBA loan products rather than statistical correlations. We formally state
made available by PPP. this idea in the following hypothesis:
In cases of taste-based discrimination, lenders
may also operationalize discrimination by demand- H1: Black-owned businesses access less PPP loan
ing a price premium for transactions with less desired funds than White-owned businesses, all else equal.
clients. They can do so by charging higher interest
rates on loans. Empirical results on interest rate dif-
ferentials are mixed. Asiedu et al. (2012) use SSPF 3.2 Role of banks
data from1998 and 2003 to estimate racial differences
in interest rates on approved loans. Controlling for In addition to discrimination, banks have historically
owner and firm characteristics including credit his- provided Black business owners with less access to
tory they find no differences in interest rates charged commercial credit through location decisions. Prox-
to Black firm owners when compared with Whites imity to bank branch locations increases access to
though they do find differences for Hispanic borrow- commercial credit and prior literature suggests that
ers. Likewise, Blanchard et al. (2008) find no evi- traditional banks choose to locate branches in fewer
dence of interest rate discrimination on bank loans minority neighborhoods than White neighborhoods.
though they do find that other lenders such as mutual For example, several studies find negative correla-
fund and leasing companies exhibited discrimina- tions between the proportion of minority residences
tory behavior against Black-owned businesses. On and the probability of a commercial bank branch
the contrary, Hu et al. (2011) use propensity score location (Wheatly, 2010; Cover, 2011; Hegerty,
matching to estimate that Black business owners pay 2020). In a comparative case study, Hegerty (2016)
approximately 0.8 percentage point higher interest considers Milwaukee and Buffalo in search of bank-
rates than corresponding White owners on average. ing “deserts.” He finds that block groups with fewer
Bates and Robb (2013) find substantial differences banks are more likely to occur in areas with fewer
in loan sizes awarded by race. Among firms with less White residents. Kashian et al. (2018) examined the
than $500,000 in sales, the gap in average loan size growth of bank deserts during the Great Recession
between minority and non-minority-owned firms period. African Americans living in inner cities were
was approximately $35,000. Bates (1989), using data more likely to live in bank deserts and saw that prob-
from the Characteristics of Business Owners survey, ability increase over the course of the recession.
finds that firms located in minority neighborhoods Proximity to branch locations increase access to
were extended substantially smaller loans. Similarly, loans through at least two mechanisms. First, poten-
Fairlie et al., (2020), using data from the Kauffman tial borrowers are able to establish relationships with
Firm Survey, find persistent differences in the amount lenders through more frequent interactions to nearby
of bank loans offered to Black-owned businesses branches. Lack of access to bank branches thus hin-
when compared with White-owned businesses. ders Black business owners from establishing rela-
To summarize, historical and contemporary evi- tionships with lenders. Historically, Black households
dence reveals lending discrimination against Black are less likely to have a bank account as compared
firm owners by commercial banks. Though there is with White households. According to the Survey of
mixed evidence on interest rate discrimination, banks Consumer Finances, the percent of unbanked Black
consistently deny credit to Black-owned firms more and Hispanic households has decreased over time, but
frequently, discourage them from borrowing more still greatly outpaces the percent of unbanked White
often, and offer them smaller loan amounts than their households. A report by the Council of Economic
White counterparts. Thus, given the SBA’s heavy reli- Advisers (2016) reports that while the White–Non-
ance on these institutions to administer the PPP, we white gap has decreased from about 30 percentage
predict that lending discrimination will reduce access

13
Discrimination in lending? Evidence from the Paycheck Protection Program 849

points in 1989, it was still about 10 percentage points and should invite entry to secure uncaptured prof-
as of 2013. its. While Ferguson and Peters (2000) explain that
In addition, lack of proximity to bank branches discrimination may persist indefinitely even in com-
can hinder Black business owners from developing petitive markets when discrimination is not costly to
vital relationships with lenders. Through relationship the lender, Michell and Pearce (2011) provide some
lending, banks collect information about firms, their evidence that the types of lenders that enter a market
owners and their communities that they use to assess can influence the extent to which competition drives
risk and structure credit terms accordingly (Berger & down discrimination.
Udell, 2002). These relationships help address infor- In summary, proximity to bank locations increases
mation asymmetries and potentially increase credit access to commercial credit. Proximity increases
availability. Banks are able to collect this soft infor- access by more readily facilitating banking relation-
mation on firms at a lower cost when borrowers are ships and by increasing credit market competition.
close by (Elyasiani & Goldberg, 2004). Without an Since banks locate near Black-owned firms less often,
established relationship to a commercial bank many Black business owners have fewer opportunities to
Black households and Black business owners struggle establish banking relationships and therefore enjoy
to acquire credit from these lenders for business start- less access to commercial credit. Early reporting sug-
ups or other purposes. gested that banks had incentives to prioritize borrow-
Furthermore, early studies of the PPP program ers with whom they had existing relationships when
indicate that such relationships were vital to securing making PPP loans. Thus, we predict that the dispro-
the funds (Cororaton, & Rosen, 2020). Li and Strahan portionately low access to PPP loans among Black-
(2020) conclude that PPP lending reflects character- owned businesses will be most evident in neighbor-
istics of relationship lending including prior experi- hoods with fewer bank branches. We formally state
ence with the lender. Though the program poses none this idea in the following hypothesis:
of the default risks that relationships are intended
to shield banks from, Li and Strahan surmise that H2: The effect in H1 is stronger in areas with
banks may be using the program to ensure that their fewer bank branches.
borrowers survive the crisis. Additionally, the exist-
ing relationships lower the cost of the transaction.
Amiram and Rabetti (2020) show that borrowers 3.3 Role of non‑traditional lenders
benefitted from relationship lending when applying
for PPP loans. Firms with existing lending relation- Since many Black firm owners lack access to bank
ships received larger PPP loan amounts and faster branches and bank credit, Black-owned firms may
approvals. rely on non-traditional lenders such as fintech firms,
A second mechanism through which bank branch CDFIs, or other nonprofits for business loans more
locations facilitate lending is increased credit mar- often than White firm owners (Erel & Liebersohn,
ket competition. Areas with more bank branches are 2020). CDFIs typically raise capital via grants or low
associated with more credit market competition and interest loans from government entities, foundations,
less concentration (Cavalluzzo & Cavalluzzo, 1998). banks seeking to fulfill Community Reinvestment
In areas with fewer bank branches, lenders have Act (CRA) requirements, individual donors, and reli-
greater market power and are able to act on racial gious institutions (Benjamin, 2004). da Motta (2020)
preferences when rationing credit without sacrificing estimates that Black-owned businesses had 1.7 times
revenue. Cavalluzzo et al. (2002) evaluate racial dif- greater odds of applying for funding from a CDFI
ferences in several lending outcomes controlling for than White-owned firms.
a rich set of borrower characteristics. They find evi- Mitchell and Pearce (2020) explain that non-bank
dence that White male business owners fared better lenders often have fewer incentives to engage in sta-
than their Black counterparts but that outcome gaps tistical discrimination than banks. One reason is
narrowed in more competitive geographic credit mar- that these institutions are often better able to repos-
kets. The authors point to Becker’s theory (1957) sess and redeploy loan collateral in the event of
which posits that taste-based discrimination is costly default due to complementary business lines. Indeed,

13
850 R. Atkins et al.

Fig. 1 Histogram of loan amounts ($) Fig. 2 Histogram of Ln loan amounts ($)

Mitchell and Pearce analyze data from the National through working with local banks and fintech compa-
Survey of Small Business Finance and find that non- nies like Kabbage and Square.”5 Indeed, Howell et al.
banks gave equal treatment to minority and non- (2020) examine which institutions supplied PPP loans
minority borrowers seeking non- line-of-credit loans to businesses by race of the owner. They report that
while banks tended to deny loans to minority borrow- fintechs, minority development institutions, and non-
ers more often than White borrowers. Furthermore, profits accounted for the largest share of PPP loans
many non-profit lenders such as community develop- to Black-owned firms by far. Combined CDFI and
ment financial institutions were explicitly created to MDIs accounted for just 4.3% of the total number of
meet the funding needs of underserved communities. loans and fintech accounted for 4.8% of the total as
Having a variety of lender types may increase the of August 2020 (SBA, 2020). Most of these non-bank
supply of credit in a given market and increase access institutions were not approved to participate in the
to credit for Black-owned firms. Carey et al. (1998) program until the second round of PPP funding was
find that while banks were just as likely as financing authorized.
companies to offer credit to low information firms, Thus, since non-traditional lenders provided most
financing companies were willing to take on the risk- of the loans to Black-owned firms and non-traditional
ier borrowers by comparison. Daniels and Ramirez lenders were approved to participate after April 30th,
(2008) also examine lender specialization and find we predict that access to PPP funds likely increased
that non-bank lenders are more willing to generate for Black-owned businesses in round 2 of funding
loans to less transparent firms. Evidence from the UK after April 30th. In turn, this likely led to a decrease in
reiterates the role of non-bank lenders in increasing effects we hypothesize for H1. We formally state this
credit supply for minority business owners. Smallbone idea in the following hypothesis:
et al., (2003) find that ethnic minority businesses
were more likely than White-owned business to uti- H3: The effect of H1 decreases after April 30,
lize non-bank lenders having experienced higher rates 2020
of denial from traditional bank lenders.
Non-traditional lenders appear to have played a
role in delivering PPP loans to Black-owned busi-
nesses as well. According to a 2020 article in For-
tune, “black businesses had more success obtain- 5
https://fortune.com/2020/06/12/black-owned-business-coron
ing funds in the second round of PPP relief, in part avirus-sean-diddy-combs-enterprises-tarik-brooks/.

13
Discrimination in lending? Evidence from the Paycheck Protection Program 851

Table 1 Summary statistics
A: Full sample (N ~ 4.5 M) B: Sub-sample for race answered (N ~ 500 K)

Variable Mean St. dev Min Max Mean St. dev Min Max
Loan amount 89,498 206,692 1000 1,500,000 112,761 223,691 1000 1,500,000
Ln loan amount 10.252 1.438 6.908 14.221 10.579 1.445 6.908 14.221
Jobs reported 10.622 30.925 0.000 500.000 14.219 32.798 0.000 500.000
Asian owner 0.012 0.107 0.000 1.000 0.112 0.316 0.000 1.000
Black owner 0.003 0.058 0.000 1.000 0.032 0.176 0.000 1.000
Hispanic owner 0.008 0.090 0.000 1.000 0.078 0.269 0.000 1.000
White owner 0.080 0.271 0.000 1.000 0.771 0.420 0.000 1.000
Other race owner 0.000 0.006 0.000 1.000 0.000 0.018 0.000 1.000
Unanswered race owner 0.896 0.305 0.000 1.000
Veteran 0.007 0.081 0.000 1.000 0.032 0.176 0.000 1.000
Female owner 0.049 0.216 0.000 1.000 0.233 0.423 0.000 1.000
Corporation 0.293 0.455 0.000 1.000 0.339 0.473 0.000 1.000
Non-profit 0.035 0.183 0.000 1.000 0.019 0.137 0.000 1.000
LLC or LLP 0.283 0.450 0.000 1.000 0.310 0.463 0.000 1.000
Median income (zip code) 5445 4309 751 210,964 5341 3963 751 123,591
% Bachelors (zip code) 0.227 0.096 0.000 1.000 0.206 0.093 0.000 1.000
% Rural (zip code) 0.136 0.255 0.000 1.000 0.222 0.318 0.000 1.000

4 Data and methods of the loan amounts appear to have been recorded
incorrectly with values such as “0” or “1.” To address
4.1 Data these issues, we winsorize the loan amounts (any
loan less than $1000 is coded as $1000 and any loan
We test our hypotheses using several sources of data. more than $1.5 million is coded at $1.5 million) and
Data on individual PPP loans comes from the U.S. take the natural log of the loan amount. (In Appendix
Treasury.6 The PPP loan-level data provides us with Table 8, we present a set of results without winsoriz-
information about the size of the loan, which is the ing; the results are almost identical.) The PPP data
main dependent variable in most of our regressions. We also provides us with information about the number
use this dependent variable to study how race affects of jobs that the loan supports, the race of the owner
loan size. We use differences in race to study the role (if reported), the gender of the owner (if reported)
of discrimination. Since the PPP program sets interest and the veteran status of the owner (if reported). Data
at 1% for all loans, the price mechanism for rationing about bank branches comes from the FDIC.7 We
this credit is unavailable to banks; thus, we would not aggregate the bank branch data to the zip code level
expect to observe interest rate discrimination. We do to match to the PPP loan data. We use the FDIC data
not measure discrimination using loan denial rates since to create a count of the number of bank branches in
SBA does not provide data on loan applications. The the county. Demographic data at the zip code level
data does not capture the ways potential borrowers are comes from the U.S. Census Bureau.
treated by loan officers hence we are unable to measure Summary statistics of the variables are reported in
discrimination in borrower discouragement. Table 1. Panel A reports data from the full sample.
Loan amounts are highly skewed (see Figs. 1, 2 There are approximately 4.5 million loans in the full
and Tables 1, 2, 3, 4, 5, 6, and 7). In addition, some sample. The average loan was about $90,000. Most
loan recipients did not report race (unanswered race
6
https://home.treasury.gov/policy-issues/cares-act/assistance-
for-small-businesses/sba-paycheck-protection-program-loan-
7
level-data. https://www.fdic.gov/bank/statistical/.

13
852 R. Atkins et al.

Table 2 Prediction of unanswered race
(1) (2) (3) (4) (5)

Dependent variable: race unanswered
% Asian population (zip code) 0.253*** 0.231*** 0.221*** − 0.067*** − 0.067***
[0.010] [0.011] [0.011] [0.013] [0.013]
% Black population (zip code) 0.476*** 0.422*** 0.417*** 0.292*** 0.248***
[0.006] [0.006] [0.006] [0.007] [0.007]
% Hispanic population (zip code) 0.447*** 0.437*** 0.432*** 0.151*** 0.148***
[0.005] [0.005] [0.005] [0.007] [0.007]
% Native American population (zip code) − 0.327*** − 0.248*** − 0.197*** 0.311*** 0.309***
[0.035] [0.038] [0.038] [0.044] [0.044]
% Other race population (zip code) 0.802*** 0.921*** 0.940*** 0.813*** 0.793***
[0.024] [0.026] [0.026] [0.035] [0.035]
Median Income (zip code) 0.000*** 0.000*** 0.000*** 0.000*** 0.000***
[0.000] [0.000] [0.000] [0.000] [0.000]
% Bachelors (zip code) 1.127*** 1.156*** 1.017*** 0.883*** 0.872***
[0.011] [0.012] [0.012] [0.013] [0.013]
% Rural (zip code) − 0.341*** − 0.362*** − 0.299*** − 0.183*** − 0.173***
[0.004] [0.004] [0.004] [0.004] [0.004]
Jobs reported -0.001*** -0.000*** -0.000*** − 0.002***
[0.000] [0.000] [0.000] [0.000]
Loan amount − 0.000*** − 0.000*** − 0.000***
[0.000] [0.000] [0.000]
Veteran − 1.226*** − 1.235*** − 1.237*** − 1.237***
[0.007] [0.007] [0.007] [0.007]
Female owner − 1.325*** − 1.337*** − 1.303*** − 1.303***
[0.003] [0.003] [0.003] [0.003]
Corporate form dummies Yes Yes Yes Yes
2-Digit NAICS Dummies Yes Yes Yes
State dummies Yes Yes
Observations 4,833,101 4,516,504 4,403,731 4,403,709 4,403,709
Log likelihood

Standard errors in brackets
***
p < 0.01, **p < 0.05, *p < 0.1

Table 3 PPP national-level statistics
Race/ethnicity Number Percent Total loans ($) Total emp Loan ($) per org Emp per Org Loan per emp

American Indian or Alaska 2794 0.06 391,000,000 41,204 139,943 14.75 9489
Native
Asian 58,274 1.17 5,320,000,000 726,329 91,293 12.46 7325
Black or African American 16,453 0.33 1,360,000,000 177,989 82,660 10.82 7641
Hispanic 41,171 0.83 4,230,000,000 546,148 102,742 13.27 7745
Unanswered 4,450,201 89.44 421,000,000,000 40,000,000 94,602 8.99 10,525
White 406,663 8.17 53,400,000,000 5,600,000 131,313 13.77 9536

13
Discrimination in lending? Evidence from the Paycheck Protection Program 853

Table 4 Top lenders by Panel A
type, over time
Loans prior to April 30 Loans after April 30
Number Percent Number Percent
Top 5 bank 652,138 19% 265,335 19%
Top 5 alternative 177,869 5% 437,808 32%
Panel B
Loans prior to April 30 Loans after April 30
White Black White Black
Top 5 bank 1934 72 2803 298
Top 5 alternative 444 30 2525 636

Table 5 PPP loan-level results
(1) (2) (3) (4) (5) (6)

Dependent variable: Ln loan amount
Asian owner 0.247*** 0.234*** 0.074*** 0.044*** 0.028*** 0.022***
[0.006] [0.006] [0.005] [0.005] [0.005] [0.005]
Black owner (H1) − 0.144*** − 0.138*** − 0.158*** − 0.140*** − 0.137*** − 0.162***
[0.011] [0.011] [0.010] [0.010] [0.009] [0.009]
Hispanic owner 0.152*** 0.151*** 0.080*** 0.050*** 0.106*** 0.089***
[0.007] [0.007] [0.006] [0.006] [0.006] [0.006]
White owner 0.424*** 0.496*** 0.363*** 0.334*** 0.349*** 0.357***
[0.002] [0.002] [0.002] [0.002] [0.002] [0.002]
Other race owner − 1.113*** − 1.113*** − 0.752*** − 0.828*** − 0.281*** − 0.298***
[0.114] [0.116] [0.100] [0.098] [0.098] [0.098]
Median income (zip code) 0.000*** 0.000*** 0.000*** 0.000*** 0.000***
[0.000] [0.000] [0.000] [0.000] [0.000]
% Bachelors (zip code) 0.214*** 0.348*** 0.404*** 0.333*** 0.159***
[0.008] [0.007] [0.007] [0.007] [0.008]
% Rural (zip code) − 0.493*** − 0.313*** − 0.320*** − 0.310*** − 0.213***
[0.003] [0.003] [0.003] [0.003] [0.003]
Jobs reported 0.022*** 0.021*** 0.021*** 0.022***
[0.000] [0.000] [0.000] [0.000]
Veteran 0.294*** 0.268*** 0.256*** 0.895***
[0.007] [0.007] [0.007] [0.011]
Female owner − 0.123*** − 0.103*** − 0.098*** 0.549***
[0.003] [0.003] [0.003] [0.009]
Mills ratio − 1.079***
[0.015]
Corporate form dummies Yes Yes Yes Yes
2-Digit NAICS dummies Yes Yes Yes
State dummies Yes Yes
Observations 4,855,505 4,833,101 4,516,504 4,403,731 4,403,711 4,403,709
R-squared 0.007 0.016 0.298 0.333 0.340 0.341

Standard errors in brackets
***
p < 0.01, ** p < 0.05, *p < 0.1

13
854 R. Atkins et al.

Table 6 PPP loan results (1) (2) (3) (4)
split by number of bank
branches Dependent variable: Ln loan amount
Number of branches 1–3 4–6 7–9 10 +
Asian owner − 0.012 0.022* 0.055*** 0.023**
[0.015] [0.013] [0.015] [0.011]
Black owner (H2) − 0.241*** − 0.143*** − 0.143*** − 0.104***
[0.021] [0.022] [0.026] [0.020]
Hispanic owner 0.028 0.095*** 0.085*** 0.116***
[0.017] [0.018] [0.021] [0.017]
White owner 0.267*** 0.360*** 0.386*** 0.398***
[0.006] [0.007] [0.009] [0.006]
Other race owner − 0.098 − 0.300* − 0.734* − 0.531***
[0.115] [0.160] [0.424] [0.183]
Median income (zip code) 0.000*** 0.000*** 0.000*** 0.000***
[0.000] [0.000] [0.000] [0.000]
% Bachelors (zip code) − 0.066 − 0.108 − 0.086 0.074
[0.058] [0.066] [0.099] [0.069]
% Rural (zip code) − 0.149*** − 0.112*** − 0.156*** − 0.100***
[0.010] [0.019] [0.028] [0.032]
Jobs reported 0.023*** 0.022*** 0.022*** 0.021***
[0.000] [0.000] [0.000] [0.000]
Veteran 0.855*** 1.014*** 0.930*** 1.017***
[0.034] [0.043] [0.050] [0.038]
Female owner 0.519*** 0.646*** 0.598*** 0.672***
[0.033] [0.041] [0.047] [0.037]
Mills ratio − 0.967*** − 1.232*** − 1.166*** − 1.340***
[0.050] [0.067] [0.079] [0.064]
Corporate form dummies Yes Yes Yes Yes
2-Digit NAICS dummies Yes Yes Yes Yes
State dummies Yes Yes Yes Yes
Standard errors in brackets Observations 949,920 911,154 773,230 1,769,405
***
p < 0.01, **p < 0.05, R-squared 0.357 0.343 0.342 0.331
*
p < 0.1

of owner is about 90 percent). Approximately 0.3% In Table 2, we assess the demographic character-
of loan recipients self-reported as Black (represent- istics of the zip codes for the recipients that did not
ing approximately 16,000 loans), 0.8% self-reported report race, as well as other loan-level information.
as Hispanic (representing approximately 39,000 Column 1 includes only demographic information,
loans), 1.2% self-reported as Asian (representing Column 2 adds in additional loan-level information,
approximately 56,000 loans) and approximately 8% Column 3 adds indicators for industry (at the 2-digit
self-reported as White (representing approximately NAICS level), Column 4 adds in indicators for state in
388,000 loans). It is notable that most loan recipients which the loan recipient is based, and Column 5 rep-
did not report race. Panel B reports data from the sub- licates Column 4 after dropping the loan amount. We
sample of loans for which race is reported (of which exclude percent of the population that is White, so the
there are approximately 500,000). Some notable dif- variables for race can be interpreted as relative to White
ferences include the higher mean loan amount and population. Looking at the coefficients for each variable
higher mean jobs reported for the sub-sample of loans across the columns, there are some interesting patterns.
for which race is reported. This suggests a selection First, recipients appear less likely to report race when
issue that will need to be addressed in our economet- in higher income, more educated, and less rural areas.
rics, as described below. Second, many of the race variables drop in magnitude

13
Discrimination in lending? Evidence from the Paycheck Protection Program 855

Table 7 PPP loan results, by timing of grant
(1) (2) (3) (4) (5) (6) (7)

Dependent variable: Ln loan amount
Pre-May 1 Early May Late May Early June Late June Early July Late July
Asian owner − 0.037*** 0.069*** 0.043* 0.008 0.056** 0.014 0.046
[0.007] [0.018] [0.025] [0.030] [0.024] [0.057] [0.034]
Black owner (H3) − 0.099*** − 0.040 − 0.021 0.066 0.071** 0.129* 0.115**
[0.014] [0.027] [0.034] [0.045] [0.031] [0.066] [0.047]
Hispanic owner 0.133*** 0.093*** 0.098*** 0.144*** 0.170*** 0.011 0.096**
[0.010] [0.019] [0.025] [0.032] [0.024] [0.066] [0.039]
White owner 0.287*** 0.147*** 0.148*** 0.234*** 0.217*** 0.172*** 0.166***
[0.003] [0.009] [0.014] [0.018] [0.013] [0.031] [0.019]
Other race owner − 0.684 0.081 0.252* 0.14 0.482 0.489*
[0.899] [0.195] [0.151] [0.110] [0.435] [0.288]
Median income (zip code) 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000***
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
% Bachelors (zip code) 0.107*** 0.265*** 0.121*** 0.107* − 0.090** − 0.376*** − 0.432***
[0.036] [0.037] [0.043] [0.056] [0.044] [0.075] [0.052]
% Rural (zip code) − 0.366*** − 0.091*** − 0.091*** − 0.085*** − 0.133*** − 0.109*** − 0.088***
[0.008] [0.012] [0.017] [0.022] [0.016] [0.035] [0.021]
Jobs reported 0.020*** 0.019*** 0.022*** 0.022*** 0.020*** 0.030*** 0.024***
[0.000] [0.001] [0.001] [0.001] [0.001] [0.004] [0.001]
Veteran 0.803*** 0.695*** 0.756*** 0.660*** 0.585*** 1.017*** 1.012***
[0.021] [0.046] [0.070] [0.098] [0.065] [0.169] [0.100]
Female owner 0.480*** 0.367*** 0.355*** 0.309*** 0.206*** 0.670*** 0.655***
[0.021] [0.038] [0.057] [0.076] [0.051] [0.128] [0.078]
Mills ratio − 1.021*** − 0.748*** − 0.809*** − 0.760*** − 0.742*** − 1.436*** − 1.385***
[0.033] [0.064] [0.099] [0.136] [0.092] [0.220] [0.131]
Corporate form dummies Yes Yes Yes Yes Yes Yes Yes
2-Digit NAICS dummies Yes Yes Yes Yes Yes Yes Yes
State dummies Yes Yes Yes Yes Yes Yes Yes
Observations 3,142,619 429,427 170,402 95,896 227,700 47,551 133,746
R-squared 0.347 0.217 0.226 0.217 0.183 0.176 0.173

Standard errors in brackets
***
p < 0.01, **p < 0.05, *p < 0.1

or even switch signs when adding in the state dum- 4.2 Methods
mies. Third, the higher the loan amount and the more
jobs reported, the less likely the recipient is to leave To test our hypotheses, we run a series of regressions
race unanswered (i.e., the more likely the recipient using the data described above. The regressions are of
is to report race). We drop loan amount in Column 5 the following form:
because we use the results from this regression to con-
(1)
( )
struct a Heckman selection variable (Mills Ratio) that ln loanamountiz = 𝛽0 + 𝛽1 Blacki + Xiz B + 𝜀iz
we include in some of our regression results below to
for loan recipient i in zip code z. The dependent
account for selection into self-reporting of race.
variable, loan amount, has been transformed as has
been described in the prior sub-section. The main
independent variable of interest Black is an indicator
for whether the loan recipient self-reported as Black.

13
856 R. Atkins et al.

We include a variety of firm level, loan level, and we included county-level demographic variables
zip code level control variables X, including gender, in the construction of our Mills ratio that we do not
veteran status, jobs reported, and dummy variables include our regressions. We also provide results on
for industry (at the 2-digit NAICS code level) and the full sample, in which case unanswered race is
state. Data on jobs reported are an important control the excluded category, and (in an Appendix) on the
for variance in firm size that may be correlated with smaller sub-sample that provides information on race,
the race of business owners. Black-owned businesses in which case White is the excluded category.
tend to be smaller in size employing fewer workers
(Tareque et al., 2021). Since PPP loan amounts are
a function of firm payroll, we would expect smaller 5 Results
firms to qualify for smaller loans. By controlling for
reported jobs, we can estimate the average difference Before turning to regression results, we first aggre-
in loan amounts given to Black and White firms of gate the loan data to the national level. These results
comparable size. are presented in Table 3. The average loan per recipi-
It is also important to control for industry level ent (“Org”) for Black recipients is $82,660. This is
effects, given evidence that the pandemic has differ- about 13% less than recipients who report race as
entially affected sectors of the economy. For example, unanswered ($94,602) and about 37% less than White
data from Yelp suggests that retail establishments recipients ($131,313). Of course, these differences
have struggled to stay open, whereas professional ser- could be reflective of many other factors, includ-
vice establishments such as law firms and accountants ing the industry in which the business operates, the
have not.8 It is important to control for state, given region of the country, the size of the business, and
that states instituted different policies in response to others. In order to systematically account for these
the pandemic. We use robust standard errors, clus- factors, we next turn to our regression results.
tered at the zip code. For hypothesis 1, we expect that Next, we consider how the lender characteristics
β1 < 0. For hypotheses 2 and 3, we split the sample evolved over time. For this, we compare loans made by
(based on number of bank branches for hypothesis a “Top 5 bank” and a “Top 5 alternative lender” before
2 and timing of loan approval for hypothesis 3), run and after April 30, 2020. A “Top 5 bank” indicates
(1) separately for each sub-sample, and compare the that the lender was one of Bank of America, JP Mor-
coefficients on β1 across the sub-samples. gan Chase, Wells Fargo, U.S. Bank, or Citibank. These
There are two issues with our approach. First, our banks are the largest U.S. commercial banks by depos-
data is cross-sectional; it provides us with demo- its, and the first three of those accounted for approxi-
graphic, geographic, and industry characteristics of mately 12% of all PPP loans made. A “Top 5 alternative
the borrower, as well as the loan amount. We do not lender” indicates that the lender was one of Cross River
have a natural experiment that we can use to tease Bank, Kabbage, Celtic Bank, WebBank, or Customer’s
out causal effects of race on loan amounts. Instead, Bank. Kabbage is a fintech bank based in Georgia. The
we use our regressions to control for as many observ- other banks are the lenders behind leading fintech lend-
able differences across borrowers as we can. Second, ers such as Coinbase and RocketLoans (Cross River),
there is a lot of non-response on the race of owner Square and OnDeck (Celtic Bank), LoanBuilder.com
variable—as reported above, only about 10% of loans (WebBank), and Upstart (Customer’s Bank), to name
report race of owner. This potentially introduces a a few. We classify these five banks as “alternative” as
selection bias if, for example, Black borrowers that they are not traditional depository institutions like the
receive larger loans are systematically more likely Top 5 banks. Cross River and Kabbage together account
to not report race than White borrowers that receive for approximately 6% of all PPP loans made. We report
large loans. We have no reason to believe that to be results in Table 4. Panel A of this table indicates that
the case, but in order to address the selection issue prior to April 30, about 19% of all loans were made by a
and account for unanswered race, we include a Heck-
man selection correction term—the Mills ratio—in
all our regressions. The construction of the Mills 8
https://www.yelpeconomicaverage.com/business-closures-
ratio is described in the prior sub-section. Note that update-sep-2020.

13
Discrimination in lending? Evidence from the Paycheck Protection Program 857

Top 5 bank, whereas only about 5% were made by a Top statistics presented in Table 3, suggesting that com-
5 alternative. However, this changes after April 30. Top positional factors (such as location, size and indus-
5 banks still account for about 19% of all loans (though try) mask some of the difference in loans received by
there are fewer loans made after April 30), but Top 5 White and Black borrowers. Overall, the results from
alternative lenders account for approximately 30% of all Table 5 provide support for H1.
loans—a dramatic increase. Panel B of this table indi- Next, we consider how the main result varies
cates that prior to April 30, the Top 5 alternative lend- depending on the presence of bank branches. Table 6
ers made 444 loans to White borrowers and 30 to Black presents results from a series of split-samples. Col-
borrowers, whereas after April 30 the Top 5 alternative umn 1 replicates Column 6 from Table 5 on the sub-
lenders made 2525 loans to White borrowers and 636 to sample of zip codes that have 1–3 bank branches,
Black borrowers. This amounts to about a 6 × increase Column 2 does the same but on the sub-sample of zip
in loans to White borrowers and a 21 × increase in loans codes that have 4–6 bank branches, Column 3 does
to Black borrowers. Thus, the Top 5 alternative lenders the same on the sub-sample of zip codes that have 7–9
increased loans to both White and Black borrowers, but bank branches, and Column 4 is for the sub-sample of
much more to Black borrowers on a proportional basis. zip codes that have 10 or more bank branches. The
In Table 5, we present our main results (Eq. (1)). coefficient on Black in Column 4 (− 0.104) is larger
In Column 1 we include only loan-level race vari- than the coefficient on Black in Column 1 (− 0.241),
ables, in Column 2 we add in zip code level demo- suggesting that higher numbers of bank branches
graphics, in Column 3 we add in other loan-level var- helps to attenuate the main effect.
iables, in Column 4 we add in dummy variables for Some care needs to be taken when interpreting these
2-digit NAICS code, in Column 5 we add in dummy results, however. In Table 6, the coefficient on White
variables for the state, and in Column 6 we add in also increases when increasing the number of bank
the Mills Ratio to correct for selection. All the coef- branches (0.267 in Column 1 and 0.398 in Column 4),
ficients are quite stable across the various specifica- suggesting that everyone benefits from larger loans as
tions. The coefficient on Black is negative and statis- the number of bank branches increases. To understand
tically significant across all the specifications. Since whether the differential between Black and White bor-
our dependent variable is logged, we can interpret the rowers changes, we compare the differences in Black
coefficient in terms of percentage changes. The coef- and White coefficients in each case. The difference in
ficients suggest that Black borrowers received loans Column 1 (0.267–(− 0.241)) is 0.508. The difference
that were 14 to 16% smaller than those of borrowers in Column 4 (0.398–(− 0.104)) is 0.502. The difference
who did not answer their race. These estimates from between these two is quite small (0.006), and not statisti-
our loan-level regressions line up closely with what cally significant based on the standard errors. To see this
we observe from the national-level statistics pre- more directly, we can look at the coefficient on Black
sented in Table 3. To assess the difference between for the equivalent split-sample regressions on the sub-
Black and White borrowers, we can assess the differ- sample of loans for which race is reported, which is pre-
ence between the coefficients on these two variables. sented in Appendix Table 10. The coefficient on Black
The coefficient on Black in Column 6, the fully speci- goes from − 0.587 in Column 1 to − 0.507 in Column 4.
fied model, is − 0.162; the coefficient on White in Again the difference between these two is quite small
Column 6 is 0.357. The difference is approximately (0.08). Overall, the results presented in Table 6 and in
0.519, suggesting that loans to Black borrowers were Appendix Table 10 are directionally consistent with
about 50% lower than observationally similar White what we expect from H2, but not statistically significant.
borrowers. Appendix Table 9 replicates the specifica- Finally, we consider how the main result varies
tions in Table 5, but on the sample of borrowers that depending on timing. Table 7 Column 1 replicates Col-
report race (White is the excluded category in Appen- umn 6 from Table 5 on the sub-sample of loans that
dix Table 9). The coefficient on Black in Column 6 were approved before April 30, 2020. The coefficient
of Appendix Table 9 is − 0.550, suggesting that loans on Black in Column 1 is negative and statistically sig-
to Black borrowers were about 55% lower than obser- nificant. Recall from Table 4 that there are relatively
vationally similar White borrowers. These figures are few loans made by alternative lenders during this time
larger than the figures obtained from the aggregated period. Columns 2 7 repeat the analysis but for the six

13
858 R. Atkins et al.

2-week periods following April 30 until the end of July. Black firm owners were denied loans disproportion-
It is notable that the coefficient on Black almost mono- ately. While our results are consistent with lending
tonically increases in each successive time period. By discrimination, we cannot directly observe discrimi-
late June, the coefficient on Black is positive and sta- nation in the data. We are however able to show that
tistically significant. Over this same time period, the the PPP loan program produced disparate outcomes
coefficient on White remains positive and statistically for Black and White business owners.
significant, but declines in magnitude. By late July, the Our study makes several contributions to the litera-
coefficients on Black and White are nearly identical ture. It joins the growing corpus on the PPP program
from a statistical standpoint.9 Recall from Table 4 that and the larger body of work on government responses
there is a large increase in the percent of loans made to crises such as the Covid-19 pandemic. While prior
by alternative lenders during this time period, and that work examined whether PPP loans were reaching pre-
alternative lenders dramatically increased the propor- dominately Black neighborhoods at the zip code or
tion of loans to Black borrowers. Overall, the results census tract level (Calem & Freedman, 2020; Fairlie &
presented in Table 7 provide support for our H3. In Fossen, 2021), our study evaluates differences in PPP
Appendix 4 we replicate the results on the sub-sample loan sizes granted to Black business owners as com-
for which race was reported. pared with White business owners. Prior work shows
that Black-owned firms closed at a higher rate than oth-
ers due to the pandemic (Fairlie, 2020). Our study sug-
6 Conclusion gests that one possible explanation is inadequate access
to PPP funds, a vital source of relief for many small
We show that the PPP loan program produced different businesses. Our analysis highlights the need for policy
outcomes for Black and White business owners. Spe- makers to account for embedded inequalities when
cifically, Black-owned firms were approved for smaller using existing systems to administer relief programs.
loans than their White counterparts controlling for Our paper also contributes to the literature on minor-
observed characteristics. We also show that SBA reli- ity-owned business finance and access to capital. We
ance on commercial banks to administer the program confirm prior findings that Black-owned firms tend to
drives this outcome in two ways. First, Black–White be approved for smaller loans than White-owned firms,
disparities in PPP loans appear (marginally) greater in controlling for observed characteristics. We extend the
areas with fewer bank branches where Black firm own- literature by examining whether density of local bank
ers may have been less likely to have prior relationships branches and access to more non-bank lenders alter out-
with SBA approved lenders. Second, the Black–White comes for Black-owned firms. Future research should
disparities in PPP loans decreased during the second examine the mechanisms through which non-profit, com-
round of funding when more non-bank entities were munity development, minority development, and fintech
approved by SBA to participate in the program. lenders supply additional capital to Black business own-
Data limited the scope of our analysis. Data on ers in contrast to commercial banking institutions.
the race of business owner was only available for A few possible policy interventions seem warranted.
10% of loan recipients, potentially creating selection First, in preparation for the next financial crisis or
bias. We address this by conducting our analyses for national emergency, racial disparities in lending through
the entire sample as well as the sub-sample for which traditional financial institutions need to be investigated
race is reported, and by including a Mills ratio from and addressed by the various regulators who oversee
a first-stage Heckman selection approach to control these institutions. Discriminatory lending and banking
for potential selection bias. Also, we were unable to practices should not be allowed to stand anywhere in the
examine lending discrimination in loan approval rates financial system. Second, the next SBA Inspector Gen-
since SBA did not require reporting on all loan appli- eral’s report should include a systematic accounting of
cants. More research is needed to determine whether the factors negatively affecting the participation of Black
business owners in the PPP. In general, between finan-
9
The difference between Black and White coefficients is 0.05 cial and economic crises, the SBA infrastructure and
and the sum of the standard errors is 0.063. The ratio of the
two is less than 2, indicating one cannot reject that the differ- capacity should be increased such that all loans, whether
ence is 0. standard or emergency, can be processed significantly

13
Discrimination in lending? Evidence from the Paycheck Protection Program 859

more quickly than is currently the case. Fourth, ex ante
demographic data collection should be applied to PPP
loans and other emergency funding facilities going for-
ward. This is standard for other SBA loans and will
help with targeting funding better in the future. Finally,
serious consideration should be given to increasing the
capacity of CDFIs to reach and service businessowners
in underserved communities.

Appendix

Please see Tables 8, 9, 10, 11

Table 8 PPP loan-level results, on full sample, no Winsorizing
(1) (2) (3) (4) (5) (6)

Dependent variable: Ln loan amount (no winsorizing)
Asian owner 0.247*** 0.233*** 0.070*** 0.044*** 0.028*** 0.022***
[0.011] [0.011] [0.008] [0.007] [0.007] [0.007]
Black owner (H1) − 0.148*** − 0.142*** − 0.162*** − 0.144*** − 0.141*** − 0.165***
[0.016] [0.016] [0.012] [0.012] [0.011] [0.011]
Hispanic owner 0.150*** 0.148*** 0.074*** 0.044*** 0.102*** 0.085***
[0.015] [0.015] [0.011] [0.011] [0.010] [0.009]
White owner 0.426*** 0.499*** 0.360*** 0.330*** 0.345*** 0.353***
[0.006] [0.006] [0.004] [0.004] [0.003] [0.003]
Other race owner − 1.119*** − 1.120*** − 0.749*** − 0.824*** − 0.273*** − 0.290***
[0.088] [0.092] [0.087] [0.084] [0.088] [0.088]
Median income (zip code) 0.000*** 0.000*** 0.000*** 0.000*** 0.000***
[0.000] [0.000] [0.000] [0.000] [0.000]
% Bachelors (zip code) 0.219*** 0.360*** 0.415*** 0.344*** 0.172***
[0.050] [0.038] [0.036] [0.034] [0.036]
% Rural (zip code) − 0.499*** − 0.311*** − 0.319*** − 0.309*** − 0.213***
[0.012] [0.009] [0.008] [0.008] [0.008]
Jobs reported 0.023*** 0.023*** 0.022*** 0.023***
[0.000] [0.000] [0.000] [0.000]
Veteran 0.292*** 0.266*** 0.255*** 0.884***
[0.007] [0.007] [0.007] [0.021]
Female owner − 0.125*** − 0.104*** − 0.099*** 0.539***
[0.003] [0.003] [0.003] [0.020]
Mills ratio − 1.064***
[0.033]
Corporate form dummies Yes Yes Yes Yes
2-Digit NAICS dummies Yes Yes Yes
State dummies Yes Yes
Observations 4,855,504 4,833,100 4,516,503 4,403,730 4,403,710 4,403,708
R-squared 0.007 0.016 0.319 0.353 0.360 0.361

Standard errors in brackets
***
p < 0.01, **p < 0.05, *p < 0.1

13
860 R. Atkins et al.

Table 9 PPP loan-level results, on sub-sample reporting race
(1) (2) (3) (4) (5) (6)

Dependent variable: Ln loan amount
Asian owner − 0.177*** − 0.373*** − 0.380*** − 0.377*** − 0.445*** − 0.446***
[0.006] [0.006] [0.005] [0.005] [0.006] [0.006]
Black owner (H1) − 0.568*** − 0.728*** − 0.557*** − 0.534*** − 0.548*** − 0.550***
[0.012] [0.011] [0.009] [0.009] [0.009] [0.009]
Hispanic owner − 0.272*** − 0.448*** − 0.345*** − 0.350*** − 0.338*** − 0.340***
[0.008] [0.008] [0.006] [0.006] [0.006] [0.006]
Other race owner − 1.537*** − 1.742*** − 1.227*** − 1.265*** − 0.671*** − 0.670***
[0.115] [0.114] [0.094] [0.093] [0.093] [0.093]
Median income (zip code) 0.000*** 0.000*** 0.000*** 0.000*** 0.000***
[0.000] [0.000] [0.000] [0.000] [0.000]
% Bachelors (zip code) 0.409*** 0.460*** 0.580*** 0.517*** 0.482***
[0.025] [0.021] [0.021] [0.021] [0.022]
% Rural (zip code) − 1.026*** − 0.676*** − 0.597*** − 0.510*** − 0.495***
[0.008] [0.006] [0.006] [0.007] [0.007]
Jobs reported 0.022*** 0.021*** 0.021*** 0.021***
[0.000] [0.000] [0.000] [0.000]
Veteran 0.074*** 0.059*** 0.035*** 0.122***
[0.009] [0.009] [0.009] [0.018]
Female owner − 0.306*** − 0.275*** − 0.270*** − 0.182***
[0.004] [0.004] [0.004] [0.017]
Mills ratio − 0.139***
[0.026]
Corporate form dummies Yes Yes Yes Yes
2-Digit NAICS dummies Yes Yes Yes
State dummies Yes Yes
Observations 502,807 500,096 486,827 486,625 486,625 486,625
R-squared 0.008 0.061 0.375 0.408 0.419 0.419

Standard errors in brackets
***
p < 0.01, **p < 0.05, *p < 0.1

13
Discrimination in lending? Evidence from the Paycheck Protection Program 861

Table 10 PPP loan results, (1) (2) (3) (4)
alternate split by num
bank branches, on sample Dependent variable: Ln loan amount
reporting race
Number of branches: 1–3 4–6 7–9 10 +
Asian owner − 0.451*** − 0.457*** − 0.432*** − 0.440***
[0.016] [0.016] [0.016] [0.013]
Black owner (H2) − 0.587*** − 0.541*** − 0.558*** − 0.507***
[0.022] [0.024] [0.027] [0.019]
Hispanic owner − 0.339*** − 0.338*** − 0.368*** − 0.328***
[0.019] [0.019] [0.022] [0.016]
Other race owner − 0.387*** − 0.688*** − 1.295*** − 0.929***
[0.126] [0.155] [0.408] [0.213]
Median income (zip code) 0.000 0.000** 0.000*** 0.000***
[0.000] [0.000] [0.000] [0.000]
% Bachelors (zip code) 0.480*** 0.354*** 0.282*** 0.435***
[0.078] [0.093] [0.102] [0.070]
% Rural (zip code) − 0.430*** − 0.458*** − 0.565*** − 0.433***
[0.017] [0.030] [0.053] [0.048]
Jobs reported 0.024*** 0.021*** 0.020*** 0.020***
[0.000] [0.000] [0.000] [0.000]
Veteran 0.077* 0.256*** 0.097* 0.160***
[0.042] [0.050] [0.058] [0.037]
Female owner − 0.204*** − 0.061 − 0.187*** − 0.159***
[0.040] [0.046] [0.052] [0.034]
Mills ratio − 0.052 − 0.323*** − 0.151* − 0.206***
[0.058] [0.072] [0.085] [0.056]
Corporate form dummies Yes Yes Yes Yes
2-Digit NAICS dummies Yes Yes Yes Yes
State dummies Yes Yes Yes Yes
Observations 124,920 100,400 81,885 179,420
R-squared 0.450 0.419 0.404 0.389

Standard errors in brackets
***
p < 0.01, **p < 0.05, *p < 0.1

13
862 R. Atkins et al.

Table 11 PPP loan results by timing of grant, on sub-sample reporting race
(1) (2) (3) (4) (5) (6) (7)

Dependent variable: Ln loan amount
Pre-May 1 Early May Late May Early June Late June Early July Late July
Asian owner − 0.408*** − 0.199*** − 0.220*** − 0.338*** − 0.250*** − 0.260*** − 0.274***
[0.008] [0.022] [0.032] [0.039] [0.029] [0.071] [0.044]
Black owner (H3) − 0.385*** − 0.235*** − 0.197*** − 0.183*** − 0.142*** 0.031 0.008
[0.014] [0.029] [0.040] [0.050] [0.035] [0.080] [0.054]
Hispanic owner − 0.208*** − 0.100*** − 0.100*** − 0.164*** − 0.098*** − 0.184** − 0.127***
[0.010] [0.023] [0.031] [0.041] [0.030] [0.075] [0.047]
Other race owner − 1.099 − 0.139 − -0.092 − 0.119 0.171 0.217
[0.879] [0.204] [0.157] [0.125] [0.388] [0.285]
Median income (zip code) 0.000*** 0.000*** 0.000*** 0.000** 0.000** 0.000** 0
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
% Bachelors (zip code) 0.526*** 0.536*** 0.475*** 0.482*** 0.188 0.282 0.515***
[0.041] [0.097] [0.140] [0.178] [0.123] [0.287] [0.196]
% Rural (zip code) − 0.543*** − 0.242*** − 0.180*** − 0.189*** − 0.242*** − 0.320*** − 0.161**
[0.012] [0.031] [0.047] [0.063] [0.043] [0.104] [0.065]
Jobs reported 0.020*** 0.020*** 0.024*** 0.025*** 0.020*** 0.030*** 0.022***
[0.000] [0.002] [0.003] [0.004] [0.002] [0.008] [0.004]
Veteran 0.099*** 0.233*** 0.381*** 0.25 0.065 − 0.074 0.240*
[0.023] [0.075] [0.113] [0.154] [0.099] [0.255] [0.145]
Female owner − 0.189*** − 0.04 0.036 0.031 − 0.198** − 0.371* − 0.098
[0.022] [0.064] [0.099] [0.132] [0.085] [0.207] [0.116]
Mills ratio − 0.114*** − 0.271*** − 0.378** − 0.365 − 0.048 0.391 − 0.132
[0.034] [0.104] [0.168] [0.228] [0.148] [0.354] [0.198]
Corporate form dummies Yes Yes Yes Yes Yes Yes Yes
2-Digit NAICS dummies Yes Yes Yes Yes Yes Yes Yes
State dummies Yes Yes Yes Yes Yes Yes Yes
Observations 407,927 26,005 11,447 7086 15,659 2241 6720
R-squared 0.422 0.297 0.317 0.324 0.278 0.289 0.276

Standard errors in brackets
***
p < 0.01, **p < 0.05, *p < 0.1

13
Discrimination in lending? Evidence from the Paycheck Protection Program 863

References Science Research Network. https://doi.org/10.2139/ssrn.
3776794
Amiram, D., & Rabetti, D. (2020). The relevance of relation- Carey, M., Post, M., & Sharpe, S. A. (1998). Does corporate
ship lending in times of crisis (SSRN Scholarly Paper ID lending by banks and finance companies differ? Evidence
3701587). Social Science Research Network. https://doi. on specialization in private debt contracting. The Journal
org/10.2139/ssrn.3701587 of Finance, 53(3), 845–878. https://doi.org/10.1111/0022-
Arrow, K. J. (1998). What has economics to say about racial 1082.00037
discrimination? Journal of Economic Perspectives, Cavalluzzo, K. S., & Cavalluzzo, L. C. (1998). Market struc-
12(2), 91–100. https://doi.org/10.1257/jep.12.2.91 ture and discrimination: The case of small businesses.
Atkins, R. M. B. (2021). Homeownership, home equity, and Journal of Money, Credit and Banking, 30(4), 771–792.
Black-owned business starts: Examining the impact of https://doi.org/10.2307/2601128
racial disparities in housing assets on firm creation. 33. Cavalluzzo, K. S., Cavalluzzo, L. C., & Wolken, J. D. (2002).
Bartik, A. W., Bertrand, M., Cullen, Z. B., Glaeser, E. L., Competition, small business financing, and discrimina-
Luca, M., & Stanton, C. T. (2020). How are small busi- tion: Evidence from a new survey. The Journal of Busi-
nesses adjusting to COVID-19? Early evidence from ness, 75(4), 641–679. https://doi.org/10.1086/341638
a survey (No. w26989). National Bureau of Economic Cavalluzzo, K., & Wolken, J. (2005). Small business loan turn-
Research. https://doi.org/10.3386/w26989 downs, personal wealth, and discrimination. The Journal
Bates, T. (1989). Small business viability in the urban of Business, 78(6), 2153–2178. https://doi.org/10.1086/
ghetto*. Journal of Regional Science, 29(4), 625–643. 497045
https://doi.org/10.1111/j.1467-9787.1989.tb01248.x Cerqueiro, G., & Penas, M. F. (2017). How does personal
Bates, T., Bradford, W. D., & Seamans, R. (2018). Minority bankruptcy law affect startups? The Review of Financial
entrepreneurship in twenty-first century America. Small Studies, 30(7), 2523–2554. https://doi.org/10.1093/rfs/
Business Economics, 50(3), 415–427. https://doi.org/10. hhw081
1007/s11187-017-9883-5 Cerqueiro, G., Penas, M. F., & Seamans, R. (2019). Debtor
Bates, T., & Robb, A. (2013). Greater access to capital is protection and business dynamism. The Journal of Law
needed to unleash the local economic development and Economics, 62(3), 521–549. https://doi.org/10.1086/
potential of minority-owned businesses. Economic 704956
Development Quarterly, 27(3), 250–259. https://doi.org/ Chatterji, A. K., & Seamans, R. C. (2012). Entrepreneurial
10.1177/0891242413477188 finance, credit cards, and race. Journal of Financial Eco-
Bates, T., & Robb, A. (2015). Has the Community Reinvest- nomics, 106(1), 182–195. https://doi.org/10.1016/j.jfine
ment Act increased loan availability among small busi- co.2012.04.007
nesses operating in minority neighbourhoods? Urban Chen, T., Lin, C., & Sun, B. (2021). Racial disparities in small
Studies, 52(9), 1702–1721. https://doi.org/10.1177/ business lending (SSRN Scholarly Paper ID 3821442).
0042098014534903 Social Science Research Network. https://doi.org/10.2139/
Benjamin, L., Rubin, J. S., & Zielenbach, S. (2004). Commu- ssrn.3821442
nity development financial institutions: Current issues Chetty, R., Friedman, J. N., Hendren, N., Stepner, M., & Team,
and future prospects. Journal of Urban Affairs, 26(2), T. O. I. (2020). The economic impacts of COVID-19:
177–195. https://doi.org/10.1111/j.0735-2166.2004. Evidence from a new public database built using private
00196.x sector data (No. w27431). National Bureau of Economic
Berger, A. N., & Udell, G. F. (2002). Small business credit Research. https://doi.org/10.3386/w27431
availability and relationship lending: The importance of Chodorow-Reich, G., Feiveson, L., Liscow, Z., & Woolston,
bank organisational structure. The Economic Journal, W. G. (2012). Does state fiscal relief during recessions
112(477), F32–F53. https://doi.org/10.1111/1468-0297. increase employment? Evidence from the American
00682 Recovery and Reinvestment Act. American Economic
Blanchard, L., Zhao, B., & Yinger, J. (2008). Do lenders dis- Journal: Economic Policy, 4(3), 118–145. https://doi.org/
criminate against minority and woman entrepreneurs? 10.1257/pol.4.3.118
Journal of Urban Economics, 63(2), 467–497. https://doi. Chugh, D. (2004). Societal and managerial implications of
org/10.1016/j.jue.2007.03.001 implicit social cognition: Why milliseconds matter. Social
Blanchflower, D. G., Levine, P. B., & Zimmerman, D. J. Justice Research, 17(2), 203–222. https://doi.org/10.
(2003). Discrimination in the small-business credit mar- 1023/B:SORE.0000027410.26010.40
ket. The Review of Economics and Statistics, 85(4), 930– Cororaton, A., & Rosen, S. (2020). Public firm borrowers of
943. https://doi.org/10.1162/003465303772815835 the US Paycheck Protection Program (SSRN Scholarly
Buffington, C., Dennis, C., Dinlersoz, E., Foster, L., & Klimek, Paper ID 3590913). Social Science Research Network.
S. (2020). Measuring the effect of COVID-19 on US small https://doi.org/10.2139/ssrn.3590913
businesses. The Small Business Pulse Survey, No. 20–16. Couch, K. A., Fairlie, R. W., & Xu, H. (2020). Early evidence
https:// www.census. gov/libra r y/worki ng-papers/ 2020/ of the impacts of COVID-19 on minority unemployment.
adrm/CES-WP-20-16.html Journal of Public Economics, 192, 104287. https://doi.
Calem, P., & Freedman, A. (2020). Neighborhood demo- org/10.1016/j.jpubeco.2020.104287
graphics and the allocation of Paycheck Protection Pro- Council of Economic Advisers (2016). Financial inclusion in
gram Funds (SSRN Scholarly Paper ID 3776794). Social the United States. July 2016 Issue Brief. https://obama

13
864 R. Atkins et al.

whitehouse.archives.gov/sites/default/f iles/docs/20160 Hegerty, S. W. (2016). Commercial bank locations and “bank-
610_financial_inclusion_cea_issue_brief.pdf ing deserts”: A statistical analysis of Milwaukee and Buf-
Cover, J., Spring, A. F., & Kleit, R. G. (2011). Minorities on falo. The Annals of Regional Science, 56(1), 253–271.
the margins? The spatial organization of fringe bank- https://doi.org/10.1007/s00168-015-0736-3
ing services. Journal of Urban Affairs, 33(3), 317–344. Hegerty, S. W. (2020). “Banking Deserts”, bank branch losses,
https://doi.org/10.1111/j.1467-9906.2011.00553.x and neighborhood socioeconomic characteristics in the
da Motta-victor, V. E. (n.d.). Minority-owned enterprises and city of Chicago: A spatial and statistical analysis. The
access to capital from community development financial Professional Geographer, 72(2), 194–205. https://doi.org/
institutions. Retrieved May 25, 2021, from http://www. 10.1080/00330124.2019.1676801
anpad.org.br/abrir_pdf.php?e=Mjc4Mjc= Howell, S. T., Kuchler, T., & Stroebel, J. (n.d.). Which lenders
Daniels, K., & Ramirez, G. G. (2008). Information, credit had the highest minority share among their Payment Pro-
risk, lender specialization and loan pricing: Evidence tection Program (PPP) loans? 13.
from the DIP financing market. Journal of Financial Hubbard, R. G., & Strain, M. (2020). Has the Paycheck Pro-
Services Research, 34(1), 35–59. https://doi.org/10. tection Program succeeded? (No. w28032; p. w28032).
1007/s10693-008-0036-5 National Bureau of Economic Research. https://doi.org/
Elyasiani, E., & Goldberg, L. G. (2004). Relationship lend- 10.3386/w28032
ing: A survey of the literature. Journal of Economics Humphries, J. E., Neilson, C. A., & Ulyssea, G. (2020). Infor-
and Business, 56(4), 315–330. https://doi.org/10.1016/j. mation frictions and access to the Paycheck Protection
jeconbus.2004.03.003 Program. Journal of Public Economics, 190, 104244.
Erel, I., & Liebersohn, J. (2020). Does FinTech substitute for https://doi.org/10.1016/j.jpubeco.2020.104244
banks? Evidence from the Paycheck Protection Program Ivashina, V., & Scharfstein, D. (2010). Bank lending during the
(No. w27659). National Bureau of Economic Research. financial crisis of 2008. Journal of Financial Economics,
https://doi.org/10.3386/w27659 97(3), 319–338. https://doi.org/10.1016/j.jfineco.2009.12.
Fairlie, R. (2020). The impact of COVID-19 on small busi- 001
ness owners: Evidence from the first three months after Jeschke, M. (2020, April 24). Ineligible PPP borrowers
widespread social-distancing restrictions. Journal of granted safe harbor period to return money by May 7 |
Economics & Management Strategy, 29(4), 727–740. Insights | Holland & Knight. https://www.hklaw.com/en/
https://doi.org/10.1111/jems.12400 insig h ts/ p ubli c atio n s/ 2 020/ 0 4/ n ot- yours- to- keep- i neli
Fairlie, R., & Fossen, F. M. (2021). The early impacts gible-ppp-borrowers-granted-safe-harbor-period
of the COVID-19 pandemic on business sales. Kalogiannidis, S. (2020). Covid impact on small business.
Small Business Economics. https://doi.org/10.1007/ International Journal of Social Science and Economics
s11187-021-00479-4 Invention, 6(12), 387 to 391–387 391. https://doi.org/10.
Fairlie, R., Robb, A., & Robinson, D. (2020). Black and White: 23958/ijssei/vol06-i12/257
Access to capital among minority-owned startups (No. Kashian, R. D., Tao, R., & Drago, R. (2018). Bank deserts in
w28154; p. w28154). National Bureau of Economic the USA and the Great Recession: Geography and demo-
Research. https://doi.org/10.3386/w28154 graphics. Journal of Economic Studies, 45(4), 691–709.
Fang, H., & Moro, A. (2011). Chapter 5 - Theories of statisti- https://doi.org/10.1108/JES-05-2017-0121
cal discrimination and affirmative action: A Survey**We Kelly, M. B., & van Holm, E. J. (2021). Urbanization, commu-
thank Alberto Bisin, Jan Eeckhout, Yariv Fadlon, and nity demographics, and the distribution of Paycheck Pro-
Kevin Lang for providing comments on an initial draft of tection Program loans across counties (SSRN Scholarly
this chapter. We also thank our coauthors Glenn Loury Paper ID 3801348). Social Science Research Network.
and Peter Norman for providing many insights on statisti- https://doi.org/10.2139/ssrn.3801348
cal discrimination and affirmative action. In J. Benhabib, Kurlyandchic, M. (2020, July 16). 785 Michigan restaurants
A. Bisin, & M. O. Jackson (Eds.), Handbook of Social got big PPP loans. 1 listed as Black-owned [Newspaper].
Economics (Vol. 1, pp. 133–200). North-Holland. https:// Detroit Free Press. https://www.freep.com/story/enter
doi.org/10.1016/B978-0-444-53187-2.00005-X tainm ent/ dining/ mark- k urly andch ik/ 2020/ 07/ 16/ michi
FDIC. (2011). 2017 FDIC National Survey of Unbanked and gan-restaurants-payroll-protection-program-loans/53912
Underbanked Households. 96. 57002/
Figlio, D. N., & Genshlea, J. W. (1999). Bank consolidations Lederer, A., & Oros, S. (2020). Lending Discrimination within
and minority neighborhoods. Journal of Urban Econom- the Paycheck Protection Program. 18. https://www.ncrc.
ics, 45(3), 474–489. https://doi.org/10.1006/juec.1998. org/ lending- discr imina tion- within- t he-paych eck-prote
2099 ction-program/
Flitter, E. (2020, July 15). Black business owners had a harder Leighton, P. (2020, July 16). Few Black-owned businesses get
time getting federal aid, a study finds. The New York PPP loans | Local News | salemnews.com. https://www.
Times. https://www.nytimes.com/2020/07/15/business/ salemnews.com/news/local_news/few-black-owned-busin
paycheck-protection-program-bias.html esses-get-ppp-loans/article_764ff682-9ce9-529f-a94a-
Global Strategy Group. (2020). Federal stimulus survey find- 9a38ecdda75d.html
ings. Prepared for Color of Change and Unidos US. Li, L., & Strahan, P. E. (2020). Who supplies PPP loans (and
https://theblackresponse.org/wp-content/uploads/2020/05/ does it matter)? Banks, relationships and the COVID cri-
COC-UnidosUS-Abbreviated-Deck-F05.13.20.pdf sis. SSRN Electronic Journal. https://doi.org/10.2139/
Granja, J., Booth, C., Makridis, C., & Sloan, M. (n.d.). Did the ssrn.3710929
Paycheck Protection Program Hit the Target? 94.

13
Discrimination in lending? Evidence from the Paycheck Protection Program 865

Li, M. (2020). Did the small business administration’s Smallbone, D., Ram, M., Deakins, D., & Aldock, R. B. (2003).
COVID-19 assistance go to the hard hit firms and bring Access to finance by ethnic minority businesses in the
the desired relief? Journal of Economics and Business, UK. International Small Business Journal, 21(3), 291–
105969.https://doi.org/10.1016/j.jeconbus.2020.105969 314. https://doi.org/10.1177/02662426030213003
Liu, H., & Volker, D. (2020). The Paycheck Protection Pro- Sobey, R. (2020, April 21). Harvard under fire for accepting
gram Liquidity Facility (PPPLF) -Liberty Street Econom- more than $8M in coronavirus relief package. Boston
ics. https://libertystreeteconomics.newyorkfed.org/2020/ Herald. https://www.bostonherald.com/2020/04/20/harva
05/t he-paycheck-protection-program-liquidity-facility- rd-under-fire-for-accepting-more-than-8m-in-coronavirus-
ppplf.html relief-package/
McKinsey. (2020). Building supportive ecosystems for Black- Tareque, I., Orozco, M., Oyer, P., & Porras, J. (2021). U.S.
owned US businesses | McKinsey. https://www.mckinsey. Black-owned businesses: Pre-pandemic trends & chal-
com/industries/public-and-social-sector/our-insights/build lenges (SSRN Scholarly Paper ID 3841149). Social Sci-
ing-supportive-ecosystems-for-black-owned-us-businesses ence Research Network. https://papers.ssrn.com/abstract=
Mijid, N., & Bernasek, A. (2013). Decomposing racial and eth- 3841149
nic differences in small business lending: Evidence of dis- Trevizo, P. (2020, June 8). When small businesses can’t access
crimination. Review of Social Economy, 71(4), 443–473. PPP, local governments struggle to close the gap. San
Mills, C. K. (n.d.). Double jeopardy: COVID-19’s concentrated Antonio Express-News. https://www.expressnews.com/
health and wealth effects in Black communities. Federal news/local/article/When-small-businesses-can-t-access-
Reserve Bank of New York, 8. PPP-local-15323521.php
Mitchell, K., & Pearce, D. (2020). Availability of financing to Wheatley, W. P. (2010). Economic and regional determinants
small firms using the survey of small business finances. of the location of payday lenders and banking institutions
Ong, P., Comandom, A., DiRago, N., & Harper, L. (n.d.). in Mississippi: Reconsidering the role of race and other
COVID-19 impacts on minority businesses and systemic factors in firm location. The Review of Regional Studies,
inequality. 31. 40(1), 17.
Parker, J. A., Souleles, N. S., Johnson, D. S., & McClelland,
R. (2013). Consumer spending and the economic stimulus Publisher’s Note Springer Nature remains neutral with regard
payments of 2008. American Economic Review, 103(6), to jurisdictional claims in published maps and institutional
2530–2553. https://doi.org/10.1257/aer.103.6.2530 affiliations.
Robb, A., & Robinson, D. T. (2018). Testing for racial bias in
business credit scores. Small Business Economics, 50(3),
429–443. https://doi.org/10.1007/s11187-017-9878-2
Shuai, X., Chmura, C., & Stinchcomb, J. (2020). COVID-19,
labor demand, and government responses: Evidence from
job posting data. Business Economics. https://doi.org/10.
1057/s11369-020-00192-2

File and source

File: discrimination-in-lending-evidence-from-the-paycheck-protection-program.pdf
Size: 788,565 bytes
SHA-256: 31e86da59e080523a4f5947304d16266fa13499749e2a70263b589ca5c90962c
Our copy: discrimination-in-lending-evidence-from-the-paycheck-protection-program.pdf
Original: europepmc.org