Does FinTech Substitute for Banks? Evidence From the Paycheck Protection Program — NBER Working Paper 27659

Summary

NBER Working Paper No. 27659, August 2020, revised December 2020, by Isil Erel and Jack Liebersohn, titled Does FinTech Substitute for Banks? Evidence from the Paycheck Protection Program. The abstract says the authors find FinTech is disproportionately used in ZIP codes with fewer bank branches, lower incomes and a larger minority share of the population, and in industries with little ex ante small-business lending. It says more PPP provision by traditional banks causes statistically significant but economically small substitution away from FinTechs, implying that FinTech mostly expands the overall supply of financial services. The paper states that NBER working papers have not been peer-reviewed and that the views are the authors' own.

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

Full text

                              NBER WORKING PAPER SERIES




    DOES FINTECH SUBSTITUTE FOR BANKS? EVIDENCE FROM THE PAYCHECK
                         PROTECTION PROGRAM

                                             Isil Erel
                                         Jack Liebersohn

                                       Working Paper 27659
                               http://www.nber.org/papers/w27659


                     NATIONAL BUREAU OF ECONOMIC RESEARCH
                              1050 Massachusetts Avenue
                                 Cambridge, MA 02138
                          August 2020, Revised December 2020




We would like to thank Daniel Green, Greg Howard, Victor Lyonnet, Karen Mills, Bernadette
Minton, Claudia Robles-Garcia (discussant), René Stulz, Tejaswi Velayudhan, and participants at
the Stanford-Princeton Bendheim Center Corporate Finance and the Macroeconomy conference
and seminars at Cambridge, Indiana, Northeastern, Penn State, Oklahoma, Ohio State, and Zurich
for very helpful comments. Thanks to May Zhu, Jason Lee and David Xu for excellent research
assistance. All errors are our own. The views expressed herein are those of the authors and do not
necessarily reflect the views of the National Bureau of Economic Research.

NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official NBER publications.

© 2020 by Isil Erel and Jack Liebersohn. All rights reserved. Short sections of text, not to exceed
two paragraphs, may be quoted without explicit permission provided that full credit, including ©
notice, is given to the source.
Does FinTech Substitute for Banks? Evidence from the Paycheck Protection Program
Isil Erel and Jack Liebersohn
NBER Working Paper No. 27659
August 2020, Revised December 2020
JEL No. G00,G01,G2,G21,G23,G28,H12,H2,H3

                                         ABSTRACT

New technology promises to expand the supply of financial services to small businesses poorly
served by the banking system. Does it succeed? We study the response of FinTech to financial
services demand created by the introduction of the Paycheck Protection Program (PPP). We find
that FinTech is disproportionately used in ZIP codes with fewer bank branches, lower incomes,
and a larger minority share of the population, as well as in industries with little ex ante small-
business lending. FinTech’s role in PPP provision is also greater in counties where the economic
effects of the COVID-19 pandemic were more severe. We estimate that more PPP provision by
traditional banks causes statistically significant but economically small substitution away from
FinTechs, implying that FinTech mostly expands the overall supply of financial services, rather
than redistributing it.


Isil Erel
Fisher College of Business
Ohio State University
2100 Neil Avenue
Columbus, OH 43210
and NBER
erel@fisher.osu.edu

Jack Liebersohn
University of California Irvine
cjlieber@uci.edu
1         Introduction

The COVID-19 pandemic created “a crisis like no other,” with a projected global
economic contraction of 4.9 percent in 2020.1 To support small businesses through
the crisis, the U.S. government created the Paycheck Protection Program (PPP),
which offered guaranteed and potentially-forgivable small-businesses loans to “provide
a direct incentive for small businesses to keep their workers on the payroll.”2 Although
the program was administered by the Small Business Administration (SBA), approved
financial institutions received applications and distributed the funds. With $669
billion disbursed over a period of a few months, the PPP program was unprecedented
in speed and scale compared to other small business support programs.3 To meet
the extraordinary demand for PPP loans, the SBA made the last-minute decision
to approve a number of non-traditional lenders specializing in Financial Technology
(FinTech) to distribute PPP funds alongside traditional banks. How does credit
provision by these non-traditional lenders differ from traditional banks?
        This paper examines the effects of this multi-billion-dollar “experiment” to ap-
prove FinTech lenders for the PPP program, which provides a unique opportunity to
study how FinTech changes access to financial services. Non-traditional and FinTech
lenders are a relatively new but rapidly growing phenomenon. How they impact credit
access will have important consequences for credit access overall. For example, Fin-
Tech firms are currently not allowed to directly participate in important government
lending programs, but they may be allowed to do so in the future.4 We study how
    1
     World Economic Outlook Update, International Monetary Fund, June 2020.
    2
     PPP is an important part of the Coronavirus Aid, Relief, and Economic Security (CARES) Act:
See https://www.sba.gov/funding-programs/loans/coronavirus-relief-options/paycheck-protection-
program.
   3
     For comparison, the 7(a) program, which is the main guaranteed lending program from the SBA,
has guaranteed between $10 billion and $20 billion per year since 2010.
   4
     Non-Federally Regulated Lender laws make it hard for pure FinTech lenders to participate in
the SBA’s 7(a) and 504 programs even if they have a state banking license. Some FinTech firms are


                                               2
FinTechs differ from traditional banks in terms of not only reaching small businesses
with urgent funding needs, but also providing financial services to potential borrowers
with limited access to the banking system. We find that FinTech is disproportionately
used in areas with fewer bank branches, lower incomes, and a larger minority share
of the population, as well as in industries with little ex ante small-business lending
and where the economic effects of the COVID-19 pandemic were more severe.
   Why was the decision to approve FinTech lenders unusual? Although the role of
FinTech has increased in the financial services industry, allowing financial intermedia-
tion by FinTech lenders remains controversial.5 Regulators have argued that FinTech
lenders require more regulation (Brooks and Calomiris, 2020); and lawmakers have
been particularly concerned about whether FinTechs act in a more discriminatory way
than traditional banks.6 Yet FinTechs have argued that they are less discriminatory
because they do not rely on relationships or face-to-face interactions with customers.7
Understanding whether FinTechs serve traditionally under-served borrowers is impor-
tant for the future of bank regulation, made particularly relevant by the increased
concern that bank regulators have shown in racial and income disparities.
   Our findings have implications for how FinTech expansion, if it is allowed, might
affect the small business lending market, where SBA-backed loans have material im-
portance. Although we have some evidence on how FinTech lenders fill in the gap
created by traditional banks for loans secured with collateral, including mortgages
and equipment loans,8 we have limited information on how FinTech lenders would do
able to participate in these programs by partnering with a qualifying bank.
   5
     See, e.g., Buchak, Matvos, Piskorski and Seru (2018), Chernenko, Erel and Prilmeier (2019),
Gopal and Schnabl (2020), Stulz (2019), Liebersohn (2020), and Gopal and Schnabl (2020).
   6
     https://cleaver.house.gov/sites/cleaver.house.gov/files/Fintech Report 1.pdf;
https://www.occ.treas.gov/news-issuances/news-releases/2020/nr-occ-2020-112.html.
   7
     See https://www.fintechfutures.com/2020/03/can-fintech-eliminate-credit-discrimination/.
   8
     See Buchak et al. (2018) for evidence on mortgages and Gopal and Schnabl (2020) for loans in
the market for equipment loans.



                                               3
for the types of services that rely more on relationships. This setting also provides the
rare opportunity to study how past relationships affect the supply of a new financial
product. It is unclear if the low level of existing financial services in under-served
areas is driven by demand or supply. The exogenous demand shock created by the
PPP program and COVID-19 crisis provides an opportunity to understand how the
supply of services differs between banks and FinTechs. Further, we can study how
this varies for different types of borrowers in the same county at the same time.
      Our first question is whether FinTechs provided more access to PPP loans in
areas where these loans were needed more, relative to traditional banks. Traditional
financial institutions (i.e., depository institutions) have been shown to be inefficient in
their allocation of financial services across customers of different sizes, locations, and
demographics (Philippon, 2015; Harvey, Santoro and Ramachandran, 2020), and, in
the particular case of allocating PPP loans, have been heavily criticized by the popular
media for favoring their relationship borrowers at the expense of smaller firms that
were hit hardest by the pandemic.9 Granja, Makridis, Yannelis and Zwick (2020) show
that during Phase 1 of the PPP program, banks did not allocate credit to regions
where it was needed. We ask whether the same was true for FinTechs. We show
that during Phase 1 of the program, when traditional banks were most constrained,
FinTech lenders provided more PPP loans than traditional banks to areas with a
worse economic shock. Studying Phase 2 of the program leads to interesting findings
as well: While both traditional banks and FinTech lenders provided more PPP loans
to the areas with a higher COVID-19 case rate and more unemployment claims, the
response of the FinTech lenders was greater than ten times larger economically.
      Second, we study whether FinTech lenders provided PPP loans to businesses with
little previous access to the banking system, relative to traditional banks. We show
  9
      E.g., “Banks Gave Richest Clients ‘Concierge Treatment’ for Pandemic Aid,” NYT, April 2020.


                                                4
that borrowers with less local access to the traditional banking system were more
likely to get FinTech-enabled PPP loans. We also study how FinTechs serve low-
income and non-white borrowers. Comparing borrowers located in the same county,
borrowers in ZIP codes with few branches were more likely to get a loan from a
FinTech lender. FinTech is also disproportionately used in ZIP codes with lower
incomes and a larger minority share of the population. Relative to FinTech lenders,
traditional banks provided a higher fraction of PPP loans to firms in industries with
stronger ties to the banking system, measured using ex-ante demand for SBA loans
relative to new PPP demand. Moreover, one-person firms were more likely to rely on
FinTech loans. These findings support the view in the popular press that traditional
banks base their PPP loans on past relationships and are geographically constrained
by the location of their physical branches, unlike FinTech lenders, which operate
mainly online and where prior relationships are less relevant.
   Third, we ask whether FinTech lenders “expand the pie” of financial services
or merely redistribute it. Even for borrowers that could access traditional banks,
FinTechs lenders may be quicker or more convenient, as Buchak et al. (2018) show
for mortgage loans, especially during the early phases of the PPP program. But,
given our finding that FinTech lending of PPP loans was more common for firms
with weaker ties to the banking system, FinTechs may also have provided loans to
borrowers who otherwise would not have received them.
   To answer this question, we study whether small businesses substitute to FinTech
lenders when local banks are less responsive to their demand for PPP loans. A high
degree of substitution would mean that FinTech lenders are mostly redistributing
which borrowers get loans, whereas less substitution implies that FinTechs are opening
up new markets altogether. To measure how much substitution happens, we first
create a bank-level measure of PPP responsiveness at the national level by calculating

                                          5
how many PPP loans each traditional bank originates per branch. Then, using the
ex ante location of each bank’s branches or deposits, we predict how much PPP
origination we would expect based on the banks that happen to be located in each
ZIP code, in an approach that is akin to a shift-share (“Bartik”) design. Note that
using national lending patterns to predict local bank responsiveness yields variation in
traditional banks’ PPP lending that is independent of the magnitude of the COVID-
19 shock and concentrates only on ZIP codes where banks have branches.
   We find that FinTech lenders originate more PPP loans per business in ZIP codes
where our instrument predicts lower PPP lending by traditional banks. In other
words, borrowers respond to a lack of bank PPP provision by somewhat substituting
to these other types of financial institutions, despite the fact that many FinTech
lenders were granted authorization only during the last few days of the Phase 1 of the
PPP. But this substitution, which is significant statistically, is small economically.
The number of FinTech loans that are made as a result of this substitution is only 0.3
percent of the decrease in traditional bank lending. Overall, these findings show that
FinTech lenders expanded the access to the PPP program but did not close the gap in
financial services across regions where banks operate. But, the substitution is likely to
be substantially larger in ZIP codes that are under-served by the traditional banking
system as this estimate comes from ZIP codes where bank branches are located and
the estimates are weighted by the number of branches per ZIP code.
   It is important to note that the incentives in play for PPP loan origination are
different from standard credit. PPP funds are forgivable in many circumstances
and lenders do not bear credit risk. Therefore, our results do not speak to differ-
ences between FinTech lenders and traditional banks in evaluating borrowers’ credit-
worthiness, but rather, to the differences in their use of relationships to allocate
government-backed credit and in their use of new technology. Banks provide small

                                           6
businesses a variety of services, from treasury management to business checking and
money market accounts, that are mediated by their relationships with borrowers.
These types of services are still mostly provided by banks, but FinTechs are becom-
ing increasingly competitive. The effects of this change are unclear. Furthermore,
government-backed credit is an important part of small business borrowing, not only
in crisis times. Therefore, our findings have important implications for the poten-
tial effects of allowing (more) FinTech lenders to participate in any type of fully
or partially-guaranteed government loan program (e.g., SBA 7a loans).10 Allowing
greater FinTech participation has the potential to increase the efficiency of small
business lending even during non-crisis periods.
       The rest of the paper is organized as follows. Section 2 reviews the literature on
FinTech lending. Section 3 describes the PPP, discusses the data collection process
and presents summary statistics. In Section 4, we present our main results on geog-
raphy of online and nonbank lending. Section 5 addresses whether ZIP codes with
less bank branches had more PPP loans by FinTech lenders, controlling for local de-
mographics. In Section 6, we calculate predicted responsiveness of banks to the PPP
and then test whether borrowers were more likely to get a FinTech-enabled loan if
they are located in ZIP codes where local banks were unlikely to originate PPP loans.
Section 7 concludes.



2        Literature Review

The literature on FinTech lenders has focused mostly on household finance (see, e.g.,
Buchak et al. (2018) and Fuster, Plosser, Schnabl and Vickery (2019)), especially
mortgage lending, where the role of technology has increased significantly. Authors
  10
   See, e.g., Craig, Jackson and Thomson (2008) showing positive effects of SBA-backed loans on
employment in low-income areas.

                                              7
find that FinTech lenders process loan applications faster than traditional banks,
improving efficiency of financial intermediation in mortgage markets, but they do not
necessarily serve underserved borrowers with low access to finance.11 Ours is one
of the first papers to focus on how FinTechs can provide financial services to small
businesses, especially the ones in underserved areas.
       Small-business lending is important because it is cash-flow based, unlike mort-
gages, so it is an area that commercial banks are, theoretically, uniquely suited for
(see, e.g., seminal papers by Diamond (1984), Petersen and Rajan (1994)). Banks
create and maintain relationships with borrowers and these relationships are particu-
larly important for the types of small and privately held firms that the PPP program
targeted. If FinTech lenders can do as well as banks (or better) in serving small
businesses, then they are able to compete with traditional banking in one of their
core competencies provided they are allowed to. Not relying on branches and rela-
tionships, FinTech lenders can also “expand the lending pie” and serve underserved
businesses. This is one reason that we specifically focus on whether FinTech lenders
provide PPP loans to the types of firms that do not have preexisting relationships
with banks. We find that, indeed, FinTech lenders do this.
       One important obstacle in research about small businesses is the limited avail-
ability of loan-level data on FinTech lending in comparison to bank lending. But in
the case of PPP loans, data on the universe of lenders and customers is available,
which allows us to speak to the entirety of this market. One paper that studies
FinTech lending for small businesses is Gopal and Schnabl (2020), where the sam-
  11
    There is also a growing literature on peer-to-peer personal loans that use FinTech, testing various
predictions on lax screening/bottom fishing or cream skimming, comparing these loans with bank
loans (see, e.g., Morse (2015) for a review; de Roure, Pelizzon and Thakor (2018), Di Maggio and Yao
(2018), Tang (2019), Vallee and Zeng (2019), Basten and Ongena (2020) for more recent papers).
Carlin, Olafsson and Pagel (2020) study the introduction of a mobile application for a financial
aggregation platform in Iceland and show resulting reductions in high-interest, unsecured debt and
bank fees, improving consumers’ well-being.


                                                  8
ple includes only asset-backed loans to purchase durable goods. Unregulated finance
companies have traditionally been key players in this market and their role has in-
creased even more after the 2008 Financial Crisis, as shown by the authors.12 In the
case of PPP, most of the purely online FinTech lenders were introduced to a market
–government-subsidized loans– that they have never participated in before and com-
peted with traditional banks, which have traditionally the been primary participants
of these programs. We find that FinTech lenders helped significantly in reaching the
businesses and areas that were more severely hit by the COVID-19 shock.
       Insufficient access to bank credit is one important reason for borrowers to bank
with FinTech lenders. This reason has become more important as regulatory con-
straints during and after the financial crisis have contracted bank lending (Butler,
Cornaggia and Gurun, 2016; Mills and Dang, 2020). Balyuk, Berger and Hackney
(2020) study commercial loans through online lending platforms and show the impor-
tance of banking relationships for small businesses. The authors find that FinTech
lending platforms compete with banks on hard-information based loans by out-of-
market banks, but not necessarily on soft-information based relationship loans by
in-market banks to small businesses. We show, in a setting where only hard informa-
tion matters, that FinTech lenders reach a wider borrower base while banks remain
constrained within their branch networks.
       There are also several papers that study how FinTechs and nonbanks serve middle
market firms. Chernenko et al. (2019) show that nonbanks provide significantly more
credit to unprofitable public businesses and that this is because banks’ cash-flow
loans are subject to tighter regulatory scrutiny. In effect, they find that nonbank
  12
    See, also, Hanson, Shleifer, Stein and Vishny (2015), Cole, Cumming and Taylor (2019), and
Cortes, Demyanyk, Li, Loutskina and Strahan (2020) showing how various nonbank lenders have
been filling the gap when large commercial banks faced regulatory constraints and, therefore, had
to pull back from commercial lending.



                                               9
and traditional bank credit markets are highly segmented. We find that though the
FinTech-enabled PPP loans only partially substituted for traditional banks’ PPP
loans to small businesses, they significantly increased PPP loan availability where it
was needed the most.13
       FinTech banks have also been competing aggressively on the funding side of the
financial institutions’ balance sheet. Abrams (2019) points to the rapid growth in de-
posit contracts offered by online banks in the past decade: online banks now comprise
four of the 30 largest banks by deposits, pay higher deposit rates, and have about
the same amount of market power over their depositors as midsize banks do. Given
the way the PPP program is structured, having an existing relationship with a bank,
even through a simple commercial deposit account should matter.
       Lastly, we also contribute to the literature on government interventions – espe-
cially, directed lending programs. Such programs can run in a form of a direct subsidy
(e.g., Banerjee and Duflo (2014) using data from India) or an indirect subsidy as in a
loan guarantee (e.g., Claire, Sraer and Thesmar (2010) using data from France). PPP
is also a directed lending program, where the Small Business Administration offered
guaranteed and potentially forgivable loans to small businesses, but borrowers applied
for and received loans through the system of financial institutions. Therefore, the role
of these institutions in this process is essential. Some contemporaneous papers have
also studied the PPP program. Cororaton and Rosen (2020) study public firms that
got funding through the PPP and received significant media outrage as the program
  13
    Davydiuk, Marchuk and Rosen (2020) also study commercial lending to middle-market firms,
but only by Business Development Companies (BDCs). There are also papers using Dealscan data
on larger loans to study loans extended by or sold to nonbanks. For example, Carey, Post, and
Sharpe (1998) focus on loans arranged by finance companies. Berlin, Nini and Yu (2018), Lim,
Minton and Weisbach (2014), Nadauld and Weisbach (2012), Ivashina and Sun (2011), Massoud,
Nandy, Saunders and Song (2011), and Jiang, Li and Shao (2010), Biswas, Ozkan and Yin (2018),
Irani, Iyer, Meisenzahl and Peydro (2020) study participation by nonbanks in loans arranged and
syndicated by banks.



                                              10
was aimed to help small businesses. They document that only 13% of the eligible
public firms, which is half of the public firms, end up participating. Li and Strahan
(2020) focus on the role of close relationships with banks in accessing the PPP loans.
Using preliminary data, Granja et al. (2020) examine whether areas that were more
severely hit by the COVID-19 pandemic, as measured by declines in hours worked or
business shutdowns, end up getting more allocations. Barrios, Minnis, Minnis and
Sijthoff (2020) develop a payroll-based framework and provide preliminary analyses
that the state-level funds, which were granted till May 1st, were allocated as predicted
by their framework. Moreover, Autor, Cho, Crane, Goldar, Lutz, Montes, Peterman,
Ratner, Villar and Yildirmaz (2020) use administrative payroll data and find estimate
that the PPP boosted employment at eligible firms by 2% to 4.5%. Bartik, Bertrand,
Cullen, Glaeser, Luca and Stanton (2020) study the effects of PPP on small businesses
using a representative national survey.14 In this paper, we focus on the differential
effect of FinTech lenders in channeling PPP funds.



3      Payroll Protection Program and Data

The Paycheck Protection Program (PPP) authorized up to $669 billion toward job
retention by small businesses.15 The program provided loans to small businesses and
eligible nonprofit organizations to pay up to eight weeks of payroll costs including
benefits, interest on mortgages, rent, and utilities.16 With about $525 billion approved
— 5.2 million loans passed through 5,460 financial institutions— by the final approval
  14
     We also contribute to a broader literature studying the consequences of the COVID-19 crisis
on financial and capital markets (see e.g., Green and Loualiche (2020), Fahlenbrach, Rageth and
Stulz (2020), Pastor and Vorsatz (2020), Halling, Yu and Zechner (2020), and Falato, Goldstein and
Hortaçsu (2020)).
  15
     Funds were distributed in two phases. $349 billion was distributed in Phase 1 over April 3-16,
2020.
  16
     Tribal businesses, self-employed individuals, and independent contractors are also eligible if they
meet the PPP’s size standards.

                                                  11
date of August 8, the PPP was one of the largest economic stimulus programs in U.S.
history. According to data reported by program participants, it supported over 51
million jobs, a majority of small business employment in the United States.
       The program was administered by the Small Business Administration (SBA) but
loans were allocated through eligible financial institutions. Eligible institutions in-
cluded any SBA 7(a) lender, federally insured depository institution or credit union,
or any other lender that was approved by the SBA and enrolled in the program.
Lenders neither charged any fees nor asked for collateral to grant these small busi-
ness loans. Loans originated prior to June 5 had a maturity of 2 years while those
originated after June 5 had a maturity of 5 years. PPP loans carried an interest
rate of 1% but loan repayment was deferred for six months. Most importantly, the
loans were fully forgiven if the funds were used at least 60% for payroll costs, interest
on mortgages, rent, and utilities. The majority of loans granted were for less than
$150,000, with the overall average loan size being $100,729.
       Our main data source is the database of PPP loans released by the Small Busi-
ness Administration (SBA). It consists of loan-level data on all PPP loans that were
approved by the SBA. The data include some characteristics of borrowers and loans.
For loans with a value above $150,000, borrower names are available, but the loan
amounts are grouped into bins. For smaller loans, the exact dollar amount is available
but not the borrower names. The borrower’s ZIP code is included and industry infor-
mation is available at the 6-digit NAICS level for all loans. The SBA also provided
the names of the financial institutions (but no other identifiers) that facilitated the
loan applications and distributions.17
  17
    News reports have raised concerns about errors in some loans’ data fields, especially free-form
text fields and information about borrower demographics (Yanofsky, 2020). Our findings do not
rely on borrowers’ specific address or demographic information. Insofar as there are mistakes in
ZIP codes, this would create measurement error in our dependent variables and would not bias the
results.


                                                12
       We match this loan-level data to bank identifiers from the Federal Financial In-
stitutions Examination Council (FFIEC) using the lender names provided.18 Most of
the names are matched using automated name matching.19 Lenders which we are not
able to match automatically are a combination of non-bank lenders, banks that have
duplicate names, and banks that have idiosyncratic names. We therefore hand-match
all PPP lenders who originate over 500 PPP loans, classifying separately non-bank
lenders and banks which do not have a unique match in the FFIEC database. This
procedure allows us to match over 97% of all PPP loans in the sample. The remain-
ing lenders are mostly small community banks with non-standard names. For the
deposit-taking banks, we obtain bank-level characteristics, including bank size, from
June 2020 Call reports and data on the number of commercial bank branches by ZIP
code from the 2018 FDIC Summary of Deposits database. Then we classify lenders
into three categories: Large banks (with assets above $20bn), small banks and credit
unions (with assets below $20bn), and FinTech lenders. In most of our analyses, we
compare depository financial institutions (including savings institutions and credit
unions), henceforth referred to as traditional banks, with FinTech lenders. Note that
we treat any unclassified lender as a traditional bank.
       We identified Fintech lenders as any unregulated nonbank lender that participated
in the program as well as any regulated online direct bank. Specifically, nonbank
lenders are non-depository financial institutions, like Kabbage, that generally rely
on FinTech in their lending.20 These nonbank lenders are not subject to typical
bank regulation as they are not financed by deposits. Online banks, however, are
  18
      Specifically, we use the Attributes File from the end of June, 2020.
  19
      We start by searching for exact, unique name matches between the files. For unmatched lenders,
we try searching for common variants of their names, such as “N.A.” in place of “National Associa-
tion.” PPP lenders whose names match multiple banks are matched to the bank with more branches.
Names which remain unmatched are then matched by hand.
   20
      Some nonbank lenders in the sample may not necessarily be traditional FinTechs (such as
Business Development Corporations), but most are.


                                                13
regulated deposit-taking banks but with only one administrative branch. They also
rely heavily on FinTech for both their lending and deposit taking. Therefore, we
classify banks in our sample as online banks if they have one branch only but extended
more than 500 PPP loans per branch to exclude small traditional community banks
with single branches.21 To this sample, we also added a few banks with more than
one branch, as identified by Abrams (2019) as online banks. A few examples of these
additions are Axos Bank, Capital One Bank, and the TIAA Bank. A full list of our
FinTech lenders is provided in the Appendix B. We present results for online banks
and nonbank lenders combined as FinTechs, but, in the Appendix Table A3, we also
provide our main table with separate columns for these subgroups of institutions since
their regulatory treatment is different.
       One interesting observation is that two one-branch community banks (Cross River
bank and Celtic Bank) partnered heavily with FinTech firms and extended almost
350,000 PPP loans in total. These banks are classified as online banks in our sample.
As Stulz (2019) discusses, many purely online FinTech lenders (e.g., LendingClub and
Kabbage), which started as peer-to-peer lenders extending only personal loans, have
also moved to direct small-business lending over time, through a banking subsidiary
or a funding bank partner. Kabbage, which is a nonbank lender, extended about
200,000 loans, making it the fourth largest lender in terms of the number of PPP
loans granted. We have three FinTech lenders in top five and four FinTech lenders
in top ten PPP lenders by loan count.
       Many of our analyses will be at the ZIP code level, in which we aggregate PPP
lending based on the borrower’s ZIP code. Unless otherwise specified, all estimates
and summary statistics are weighted by the number of PPP loans per ZIP code. We
  21
    In Appendix Table A2, we also provide our main table using a smaller subset of online banks as
robustness. See the Appendix for the list of these FinTech lenders.



                                               14
measure the fraction of nonbank, online, and bank lending by ZIP code for borrowers
whose type we have classified.22
       We match this data to demographic information from the 2000 Decennial Cen-
sus and the 2014-2018 American Community Survey (ACS) (Manson, Schroeder,
Van Riper and Ruggles, 2017). From the Decennial Census, we measure the frac-
tion of the population that is white. From the ACS, we measure total population,
median household income, and travel time to work. We recode travel time to create
an indicator that measures the fraction of households that report a travel time of
over 45 minutes. Census variables are measured by ZIP Code Tabulation Area which
we match to ZIP codes. To measure the economic characteristics of firms —i.e., the
number and size of establishments— in each ZIP code, we use data from ZIP Busi-
ness Patterns 2017 data. The average size of establishments is calculated as the total
employment divided by the number of establishments in each ZIP code.
       We measure the magnitude of the economic shock by county using data from the
Opportunity Insights Track the Recovery web site (Chetty, Friedman, Hendren and
Stepner, 2020). We focus on two main measures. First, we measure the four-week
change in unemployment claims by county as of April 11, 2020. This measure covers
the last week before unemployment started rising until the peak level of unemploy-
ment claims nationally. Second, we measure the average of the daily count of COVID
cases by county in March per 100 people. See Chetty et al. (2020) for more details
on these measures.
       Summary Statistics by ZIP codes, weighted by PPP loans, are shown in Table
1.23 Since we do not have loan amounts for all types of loans — only those with a
value below $150,000 — our analysis focuses on the number of PPP loans rather than
  22
     Bank lending measured at the ZIP code level includes some lending by credit unions and saving
institutions, which are also depository institutions like commercial banks.
  23
     See Table A1 for unweighted statistics.


                                               15
on their dollar amount.24 However, in these types of direct government subsidies,
it is at least as important to understand how many firms ended up benefiting from
the loan program as the average value of the loan granted. 7% and 10% of all PPP
loans were processed by nonbank lenders and online banks, respectively. The mean
(median) ZIP code had 629 (515) PPP loans. In these ZIP codes, median income is
about $64,000, only 14% of the population commute at least 45 minutes per day to
work, and 80% of the population is white. In a typical ZIP code with PPP loans,
there are 9.4 bank branches but with a standard deviation of 7.2 branches. Also, note
that bank branch summary statistics are shown only for ZIP codes with a non-zero
number of branches. A given ZIP code in our sample has 924 establishments and a
total population of 13,736, on average. These areas also had 0.02% March COVID
case rate and 3.35% unemployment growth.25



4     The Geography of FinTech Lending

Figure 1 shows the number of PPP loans by lender type between April 3, when the
first Phase of the PPP started, and August 8, when the second Phase of the PPP
ends. The X-axis of this figure shows the approval date and the Y-axis shows the
number of PPP loans approved on each date by lender type. There is a gap between
April 16, when PPP Phase 1 ended, and April 27, when Phase 2 began. Most of the
PPP lending occurred in Phase 1 and in the beginning of Phase 2.
    Media reporting during PPP Phase 1 suggested that smaller banks were better
able to process PPP loans than larger banks. The evidence in the upper panel of
   24
      On average, FinTech lenders originate smaller loans than traditional banks do. The average loan
from FinTech lenders supports five jobs the average loan from traditional banks supports twelve jobs,
based on borrowers’ reports, suggesting that FinTech loans reach smaller businesses.
   25
      Unemployment rates, which we can measure only at the county level, are not available every-
where.



                                                 16
this figure supports this view: During the initial weeks of PPP Phase 1, there were
significantly more PPP loans arranged by small banks than by large ones.26 The
difference shrank towards the end of Phase 1, and by late Phase 2, large banks were
responsible for more PPP lending than small banks were.
       FinTech lenders were responsible for about 16% of PPP loans overall, with their
share increasing in Phase 2 of the program. The right panel of Figure 1 presents the
average FinTech fraction of PPP loans by week. The share of loans from FinTech
institutions started increasing during the last few days of Phase 1 and significantly
accelerated during Phase 2, reaching the majority of overall lending. It is important
to note here that many FinTech lenders were approved only during the last few days
of Phase 1, likely reducing the total number of PPP loans they could extend before
the program expired.
       We begin by studying where, on a national basis, FinTech lenders were more
widely used relative to traditional banks. We use an independent measure of regional
interest in online PPP lending based on Google searches to validate our measure.
Specifically, we use Google Trends to calculate, at the state level, variation across
states in searches for the phrase “apply for ppp loan online” from March 1, 2020 to
August 8, 2020. States with few searches are excluded from the Google Trends data.27
Figure 2 shows the relationship between Google searches for online lending and our
measure of actual PPP loans, with missing states located at zero. The relationship
is positive and statistically significant whether or not we include states that have too
few searches to include. This finding is supported by information we learned from
  26
      We believe that the unmatched banks in our sample are more likely to be credit unions and
community banks than larger national banks, because community banks have names which are
more difficult to match unambiguously (e.g., “First Bank” and “Farmers and Merchants Bank” each
refer to many possible banks). Therefore, the difference between small and large banks may actually
be understated in this figure.
   27
      The top state for online searches is Georgia. The largest nonbank FinTech lender in the sample
is Kabbage, Inc., which is based in Atlanta.


                                                17
discussions with FinTech lenders, that Google searches were an important way that
small businesses found out about FinTech loans.
   Next, we turn to the geographic distribution of PPP loan provision by FinTech
lenders measured using the SBA loan database and our classification of FinTech
lenders. The specifications in Table 2 explore the geographic correlates of PPP loan
provision. Our first question is whether FinTech PPP loans flowed unconditionally to
the areas that needed it most in both periods, and if they differed in this regard from
traditional banks. To measure which areas were most in need of PPP loans, we use
county-level variables collected by Chetty et al. (2020): the increase in unemployment
claims rate between the months of March and April and the average COVID-19
case rate per 100 people in March. Our other variables are measured at the ZIP
code level. We control for the log number of establishments by ZIP code to avoid
a mechanical relationship between the number of establishments and the number of
loans. Regressions are run at the ZIP code-level and estimates are weighted by total
PPP loans by ZIP code. Robust standard errors are reported. Our dependent variable
is the log of total PPP loans by traditional banks or by FinTech lenders.
   During Phase 1, traditional banks did not provide PPP financing to the regions
with higher case rates or higher unemployment, as already found by Granja et al.
(2020). In fact, traditional banks provided fewer PPP loans to counties which needed
it more, along both measures (see Column (2) of Table 2). By contrast, FinTech loans
did flow to areas with a worse COVID shock and a worse economic shock, despite
the fact that majority of them got approval in the last few days of Phase 1 (see
Column (1)). The difference between FinTech lenders and traditional banks is also
economically meaningful in Phase 1. According to the coefficients in Columns (1)
and (2), each additional COVID case per 100 people was associated with 4 log-points
less PPP lending from traditional banks but no less from FinTech lenders. One more

                                          18
percentage point increase in unemployment was associated with 0.07 log points less
traditional bank lending and 0.07 log points more FinTech lending.
   One possible explanation for the failure of traditional banks in reaching the areas
hit more by the COVID-19 shock during Phase 1 could be their ability or preference
to start PPP lending in areas with bank branches. If areas underserved by bank
branches are also hit worse, we would find the findings presented in Column (2)
troubling, given the intention of the PPP to reach to worse-hit businesses as quickly
as possible, but not surprising. In Column (3), we limit the sample of PPP loans
to those extended by traditional banks that have at least one branch in the same
ZIP code as the PPP loans, and hence focus on lending by local traditional banks.
Findings are similar: even local banks with close-by branches did not provide PPP
financing to the regions with higher case rates or higher unemployment during Phase
1 of the PPP.
   Another possible reason that traditional banks did not lend more in the worst-hit
areas may be that they faced different incentives than FinTech firms, for example
because stricter anti-money-laundering compliance made it more expensive for tra-
ditional banks to acquire new customers than FinTechs. However, many regulated
banks partnered with FinTech firms to originate PPP loans for their commercial
banking customers, suggesting that they wanted to originate PPP loans quickly but
did not have the capacity to do so themselves. These traditional banks may have
faced internal barriers to adopting new technologies even when they were motivated
to do so.
   During Phase 2, as shown in the last two columns of Table 2, PPP loans flowed
towards areas that needed more assistance both from traditional banks and from
online banks/online lenders. However, even then, FinTechs were more responsive
to financial need than traditional banks were. The coefficient on the average case

                                         19
rate is more than ten times larger for FinTech loans than traditional bank loans.
During Phase 2, each additional COVID case per 100 people was associated with 5
log points more FinTech loans and 0.3 log points more loans from traditional banks.
In Column 6, where we concentrate on the local traditional banks, the difference
drops but only slightly, with the order of magnitude of almost ten times smaller than
the FinTech lenders’ response. Supporting the widely-reported problems with banks’
ability to provide PPP loans to areas that needed it during Phase 1 of the program,
our findings seem to indicate that FinTech lenders were better able to respond to
local demand.


4.1    Demographics of FinTech Loans

Figure 3 is a county-level graph showing the fraction of PPP loans coming from each
type of institution for the entire United States. Here, we consider a combination of
Phase 1 and Phase 2 loans. There are clear patterns visible in this figure. Major
metropolitan areas, such Atlanta, Miami, Houston and Chicago, as well as both
coasts, have a high fraction of their PPP loans originated by FinTech lenders. Urban
parts of New Mexico, Colorado and Arizona also have significant FinTech PPP loan
origination.
   Our main question is whether FinTech lenders provided PPP loans to borrowers
poorly-served by the banking system. At a descriptive level, the national data sug-
gest that FinTech loans were most common in areas that are already well-served by
the banking system: coasts and major metro areas. However, there may be other
important factors that vary by state or county, such as differences in COVID-19
awareness or differences in historical FinTech marketing, that make this hard to in-
terpret. Therefore, our next step will be to study differences in FinTech PPP lending



                                         20
by borrowers located in the same county, in order to control for these sorts of regional
differences.
   To build intuition for our within-county results, we consider the geography of lend-
ing by online banks and nonbanks in the city of Chicago. Figure 4 shows the distribu-
tion of PPP loans for ZIP codes in Cook County, which includes most of the Chicago
metro population. We are interested in understanding the distribution of FinTech
loans in relation to demographic differences in ZIP codes within the metropolitan
area. As shown in the left panel of this figure, Cook County is characterized by large
differences in income by ZIP code. The North Shore is high-income and also mostly
white, as are the western parts of Cook County. South Chicago has lower median
incomes. Differences in income are sharp across neighborhood boundaries.
   These differences manifest themselves in differences in the proportion of PPP loans
that come from FinTech lenders. The right panel of Figure 4 shows the fraction of
PPP loans which we classify as coming from FinTechs. Businesses in the richer ZIP
codes of Chicago mostly get their loans from traditional banks and credit unions,
whereas the lower-income areas get a higher fraction of their loans from FinTech-
focused online banks and nonbanks. We have created similar maps for other major
metro areas and found similar patterns.
   One possible reason for local differences in FinTech PPP lending by ZIP code is
the variation in the location of traditional bank branches, a topic we now turn to. ZIP
codes with more bank branches are known to have more competitive banking markets
and hence better credit access. The relationship between bank branches, and FinTech
lending is shown in Figure 5, which is a binscatter plot. The left panel of this figure,
labeled “National”, uses pooled ZIP code data from the entire country. On the X
axis, we show the average (log) bank branches per ZIP code, where ZIP codes are
grouped into vintiles and the logarithms are in base-10 to make interpretation simpler.

                                          21
The Y axis shows the fraction of Fintech PPP loans for each vintile. Based on the
national patterns Figure 3, we should not be surprised to see that regions with more
bank branches also had a higher share of Fintech lending. Looking across regions,
relationship between bank branches and FinTech lending is generally upward-sloping
(although it is not perfectly linear).
    But when we look within-county, these patterns are reversed: ZIP codes with
fewer branches have a higher fraction of FinTech loans. The right panel of Figure 5
conditions on county fixed effects and hence uses only within-county variation in bank
branches by ZIP code. Here, there is a clear negative relationship bank branches and
the fraction of FinTech loans. In other words, although online banks and nonbank
lenders have a larger presence in parts of the country with more traditional banking,
they disproportionately serve under-resourced areas when we look within a county.
    We next turn to linear regressions to quantify this evidence and to distinguish
between the separate effects of bank branch location and demographic differences in
loan demand.



5     Branch Distance and FinTech Lending

A large body of research in banking shows that banks are more likely to have a
lending relationship with borrowers located physically close to their branches. If
banks prioritized PPP lending to customers with existing bank relationships, we would
expect to see a correlation between the number of bank branches and the likelihood
that borrowers got their PPP loans from a traditional bank.
    Confirming the pattern in Figure 5, this result is shown in Table 3. Bivariate
regressions, shown in Columns (1) and (2), show an economically and statistically
significant effect of bank branches on FinTech lending. A one log-point increase in


                                         22
the number of bank branches decreases the fraction of FinTech loans by about 0.02.
Since the median fraction of FinTech loans is about 15%, this means that doubling the
number of bank branches in a ZIP code is associated with a decrease in the FinTech
share of about 13.3% of the median.
   The finding in Table 3 controls for county fixed effects, which eliminates many
potential confounding variables that might vary by region. Nonetheless, we might
worry that the relationship between bank branches and traditional bank lending is due
to omitted demand factors, such as the types of businesses located there, rather than
due to the supply of loans from traditional banks. Yet, adding local demographic and
income controls slightly increases the estimated relationship between bank branches
and the FinTech share of borrowers. In Columns (3) and (4), we add median income,
the fraction of white population in a ZIP code, and the fraction of population with
a commute above 45 minutes. Within a county, areas with lower incomes, longer
commutes, and more non-white people have a larger FinTech share of PPP loans.
And conditional on these controls, the coefficient on log bank branches rises from
0.02 to 0.025. FinTech lenders, therefore, reached not only to areas more severely hit
by the virus, but also to under-served borrowers with limited access to the banking
system.
   As a robustness check, Table A4 in the Appendix shows results from similar spec-
ifications at the loan level. The results are similar. Loan level specifications allow us
to control for differences in borrower characteristics which may affect the riskiness of
the loans. All the specifications in this table control for borrower industry. Since we
use 6-digit NAICS industries, the NAICS industry likely proxies for many types of
borrower differences. We also add controls for average establishment size and local
demographics. Throughout, the effect of bank branches on the Fintech share remains
negative and statistically significant, and it does not change much even after including

                                           23
NAICS6 fixed-effects.
       Next, we exploit variation in PPP demand coming from regional variation in the
COVID-19 shock to show that, relative to FinTechs, banks’ supply of PPP loans is
more elastic in regions where there are more branches. To test this prediction, we
interact the magnitude of the shock with the density of bank branches and study the
supply response by FinTechs versus banks. We use the following regression specifica-
tion:


           F inT echSharezc = β1 Branchesz + β2 Shockc × Branchesz + δc + εz


       where z indexes ZIP codes, c indexes counties, and δc are county fixed effects.28
We are interested in β2 , the coefficient on interaction between the COVID-19 shock
and the number of bank branches. We expect that a high shock provokes a greater
response by banks relative to FinTechs in areas with many bank branches. Therefore,
we expect β2 to be negative. We again measure the COVID-19 shock using both the
number of cases per 100 people in March and the rise in unemployment at the county
level between the months of March and April.
       The results are shown in Table 4, where Columns (1) and (3) use Commuting
Zone rather than county fixed effects so that the Case Rate and Unemployment Rate
main effects — which are measured at the county level — are identified. We find that
the effect of COVID cases on FinTech lending is reduced in areas with more bank
branches (see Column (2)). Interpreting the coefficients in the first two columns,
one more case per 1,000 people means that FinTechs originate about 9 percentage
points more of the local PPP loans on average. But in areas with one log point
more branches, the effect of each COVID case on the FinTech fraction is reduced by
  28
    Since the COVID-19 shock is measured at the county level, county fixed effects mean we do not
include a main effect for the COVID shock in the baseline specification.


                                               24
about 10%. In other words, adding one log point more branches reduces the effect
of one COVID case per 1,000 from about 9 percentage points to about 8 percentage
points. Likewise, bank branches intermediate the effect of unemployment on PPP
loans. Shown in Columns (3) and (4), areas where the unemployment rate grew by
1 percentage point more had 1.5 percentage points more of their PPP loans coming
from FinTechs. But if the number of bank branches rises by one log-point, the effect
of a 1 percentage point increase in the unemployment rate on the FinTech share falls
to 1.0 percentage points. Overall, these results show that banks’ supply of PPP loans
was more elastic in regions with more branches, highlighting again the importance of
branch networks for banks’ provision of financial services.
      Borrowers in low-income regions learned about the PPP program from a variety
of sources, including messaging from their regular banks and from public service
announcements from the SBA and nonprofits.29 Informal discussions with FinTech
lenders suggest two important ways that borrowers learned about FinTech lenders
in particular. First, many borrowers, especially those who were denied loans by a
traditional bank, searched for PPP loans online and came across FinTech lenders.
Second, community banks without the capacity to make loans themselves may have
referred borrowers to the FinTechs, in some cases without informing the FinTech
lenders that they were doing this.


5.1      Banking Relationships and FinTech Lending

The statistics so far have provided evidence that in lower-income areas and in areas
with fewer banks, more borrowers turned to FinTech loans for their PPP. But, we
are also interested in directly answering the question of whether firms with less ex
 29
    One prominent nonprofit was Our Fair Share, which was founded by musician Sean Combs to
provide information about PPP and other government programs to minority entrepreneurs.



                                            25
ante exposure to the formal banking system were more likely to turn to these types
of lenders. Our results for branch density provide evidence that bank relationships
matter, but we would like to show this by showing the same thing for different types
of borrowers that are located in the same ZIP code.
   To measure exposure to the formal banking system, we measure pre-COVID bank-
ing system access at the industry level. To do this, we use SBA data on the 7(a)
program from the years 2018 and 2019. The 7(a) program is the main lending pro-
gram that the SBA uses to support small businesses. Since it is administered through
the same types of institutions as the PPP program, firms in industries which previ-
ously used 7(a) loans are likely to have ex ante banking relationships. Therefore,
we measure which industries disproportionately got PPP loans relative to how many
SBA loans they previously used. Small businesses in industries which demanded
many PPP loans, but previously had few SBA 7(a) loans, are unlikely to have strong
relationships with banks. On the other hand, small businesses in industries where
SBA 7(a) loans are common are more likely to have a formal banking relationship.
Therefore, we measure the log ratio of PPP loans in the sample relative to SBA 7(a)
loans from the previous two years. We construct this measure at the 6-digit NAICS
industry level.
   These estimates are shown in Table 5. As shown in Column (1) of this table,
businesses in industries with a higher PPP demand shock relative to the SBA 7(a)
lending quantity were more likely to go online or turn to nonbanks. Applying the
estimates in Column 1, firms in an industry with 10 percent more PPP loans than
SBA loans would get 0.2 percentage points more of their PPP loans from FinTechs,
relative to firms in an industry with the same number of PPP as SBA loans. While
0.2 percentage points sounds small, it is economically meaningful given that only
10 percent of all PPP loans are from FinTechs. Column (2) adds controls for fixed

                                        26
effects at the NAICS 2-digit level. We do this to ensure the robustness of the results
to differences in industry exposure to COVID at the sector level. When we add these
controls, the coefficient on the SBA loan access measure increases and remains highly
statistically significant.
   Another measure of exposure to traditional banking is presented in Columns (3)-
(4) of Table 5: whether firms that applied for the PPP loans are structured as a sole
proprietorship or as a self-employed individuals. These one-person firms, which are
unlikely to have a formal borrowing relationship with a traditional bank, are more
than 20 percentage points more likely to borrow from a FinTech PPP provider. As
expected, this effect is very significant not only statistically but also economically.
   In the last two columns of Table 5, we use industry-level employment growth,
which is a four-week change in unemployment insurance claims by April 11. We find
that firms from industries with larger growth around the COVID-19 shock rely less on
FinTech PPP loans. In other words, FinTech PPP loans have reached to harder-hit
industries that were underserved by banks.
   These results show that banks base their lending on past relationships and con-
strain themselves around their branches. FinTech lenders do not have geographic
constraints based on the presence of loan officers or physical bank branches. Despite
this observation, there are a few reasons to think that relationships, or something
akin to relationships, might matter for FinTechs. First, borrowers might not know
about the possibility of getting a PPP loan through an online bank unless they have
done it before. Therefore, areas with many FinTech borrowers in the past might be
disproportinately served by FinTech lenders during the PPP program. Second, small
businesses might use online banks for other types of financial services, such as de-
posits or credit cards. Such borrowers might also trust the same firms to supply PPP
loans for them. In both cases, we would expect areas with a large historical FinTech

                                           27
presence to have more PPP loans as well.
   To understand whether “relationships” matter for FinTechs, we measure how
many SBA 7(a) loans came from FinTechs in the years before the COVID crisis
and ask whether this is associated with borrowers getting PPP loans from FinTechs
as well. To do this, we match lender names from the 7(a) program to the classifica-
tion which we create for the PPP program. Less than 2% of 7(a) loans made from
2018-2019 come from lenders which did not make PPP loans, and which we therefore
do not classify. Among 7(a) loans we do classify, about 5% come from online banks
and about 1.5% come from nonbank lenders. Many of the most important nonbank
FinTech lenders, such as Kabbage, Inc., have no history of originating 7(a) loans at
all. There is substantial heterogeneity by ZIP code in terms of the share of loans
coming from FinTech lenders.
   The estimates in Table 6 Columns (1)-(2) present the relationship between the
share of 7(a) loans in each ZIP code coming from FinTech lenders and the share
of PPP loans coming from them. We find that geographic persistence matters for
FinTech lenders, but it is not the only important factor. On the one hand, the
estimated effect of 7(a) lending from FinTechs on the FinTech share of PPP loans
is statistically significant at the 5% level when we include control variables. On the
other hand, the point estimate is economically very small – not different from zero
without controls and 0.012 including controls. Moreover, the coefficient on log bank
branches is still significant when we include the FinTech fraction of 7(a) loans as a
control variable and the estimate is similar to the estimates in Table 3.
   Although we show that FinTech lenders serve underserved populations and areas,
an important potential constraint for borrowers’ access to FinTech lenders is the ac-
cess to information online. Therefore, we explore next whether FinTech lending is
larger in areas with larger fraction of population with a computer. Both regressions

                                          28
are run at the ZIP code level with country fixed effects included and presented in the
last two columns of Table 6. Interestingly, we find a negative and significant coef-
ficient in Column (3); however, in Column (4), this coefficient turns to be positive
and significant when we control for the median income and other demographic char-
acteristics of the Zip codes in addition to the number of bank branches. A possible
reason for this is that computer access is positively correlated with income, banking
system access, etc., so without controlling for these factors, the estimated relationship
between computer access and FinTech usage is spuriously negative.
   Another possible explanation for FinTech lending in areas underserved by banks is
selective and increased advertising in these areas during the COVID-19 shock. From
our discussions with FinTech lenders, we learned that FinTechs did not engage in
particular marketing pushes during this time relative to traditional banks. But we
would like to confirm this using the data. Using time-series data from Kantar Media,
as provided by Adspender, we analyze commercial and business financial services
advertising dollars by FinTech lenders and traditional Lenders through July, 2020.
We hand-matched our bank and top-30 FinTech PPP lenders to create this sample
and present monthly averages of advertisement dollars by quarter in the Appendix
Figure A1. We find a general decline in FinTech advertising dollars, which are mostly
dedicated to national, cable-TV advertising, in 2020 with no change in this trend
during COVID-19 period. If anything, we see an increase in spending on ads by
traditional banks in the third quarter of 2020.
   Overall, in this section, we find that geography matters more for traditional banks
than for FinTech lenders. While FinTech lenders do provide PPP loans in areas that
they have lent in the past, this effect is not strong. Rather, they focus on facilitating
transactions for any borrower with a computer.



                                           29
6     How Substitutable are FinTech Loans?

An important question is whether FinTechs expand the overall loan supply by provid-
ing PPP loans to new types of borrowers who otherwise would not have gotten PPP
loans, or their presence just causes substitution away from banks. We have shown
that borrowers in regions with fewer branches or with weaker ties to the banking
system are more likely to borrow from FinTech lenders. But this finding does not
necessarily imply that FinTechs lend to a new segment of borrowers. If borrowers
with weak ties to the banking system find FinTech lending more convenient, then
it is possible that they get PPP loans from FinTechs that they would have gotten
from banks if FinTech loans had not been available. On the other hand, if FinTechs
expand access into an altogether new market, then we would expect little substitution
between FinTech lenders and banks.
    To shed light on this question, we study whether borrowers substitute to FinTech
lenders in traditional banking regions, where traditional banks’ PPP loan supply is
constrained. If the FinTech and traditional bank markets are relatively segmented,
then we would not expect much substitution between FinTech and traditional bank
PPP loans in these regions with bank branches. On the other hand, if borrowers
easily substitute between banks and non-banks, then any reduction in traditional
bank lending should lead to a substantial substitution to FinTech lenders.
    Our identification strategy relies on differences in traditional banks’ overall supply
of PPP loans. According to widespread news reports around the time of Phase 1 of
PPP, some banks were able to handle the surge in PPP demand much better than
others. We exploit these differences at the national level and create a measure of pre-
dicted bank responsiveness that will allow us to distinguish the possible explanations
for our findings. The advantage of measuring predicted responsiveness, rather than


                                           30
realized responsiveness, is that the realized level of responsiveness of local banks may
be a function of the magnitude of the COVID shock in each region, which may also
have direct effects on the types of PPP loans that borrowers choose. By predicting
banks’ responsiveness based on their national lending patterns, we hope to create
a measure of traditional bank PPP lending that is independent of the number of
COVID-19 cases.
       We create our predicted-PPP measure in two steps. First, we measure PPP loans
per bank branch (PPP loans divided by the number of bank branches) at the bank
level nationally in all counties where a bank has branches. This is our measure of
bank “responsiveness” to the PPP program. In the second step, we calculate the
average responsiveness by ZIP code of banks located there. This yields a prediction
for the amount of PPP lending that will take place in each ZIP code. We take the log
of this measure to calculate the log predicted number of PPP loans by ZIP code.30
       Note that in constructing our predicted-PPP measure, we drop each state’s own
branches and loans and do the calculations in step 1 using loans from other states.
This leaving-one-out approach also ensures that our measure of bank responsiveness
in each state is independent of COVID-19 conditions in that state.31 In this way, we
create a measure akin to a shift-share shock (Bartik, 1991), where we quantify the
degree of responsiveness, at the bank level, to the PPP program.
       Following Goldsmith-Pinkham, Sorkin and Swift (2020), our identifying assump-
tion is that our measure is independent of unobserved variables which affect the
relative demand of FinTech versus traditional bank loans. To support this assump-
  30
     Since this variable is only available in ZIP codes with bank branches, estimates using predicted
loan amounts will have about half as many observations as the previous tables.
  31
     To see why this is important, consider the example of two community banks, each located in
only one state, where their states have different PPP demand shocks. The bank located in the
high-shock state would appear more “responsive” to the program because it would originate more
loans per branch. But, this would be coming from differences in loan demand, not loan supply.



                                                 31
tion, we show that predicted PPP lending is independent of the magnitude of the
COVID-19 shock. Table A5 in the Appendix verifies that the predicted lending mea-
sure is independent of our two proxies for the size of the COVID shock — the average
COVID case rate in March and the increase in the unemployment claims rate from
March 15 to April 11. Ideally, we would measure these variables at the ZIP code
level and use specifications with county fixed effects. Since these variables are only
available at the county level, this table uses Commuting Zone fixed effects instead.
We also include specifications which control for the number of bank branches per
ZIP.32
       Next, we verify that in ZIP codes with more responsive banks, more PPP lending
is provided overall. These results are shown in Appendix Table A7. For this table, we
include county fixed effects and weight estimates by the number of branches in the
ZIP code. We also add the total number of bank branches as an additional control
variable. The variable of interest is labeled “Predicted PPP” and it measures the log
predicted number of PPP loans by ZIP code. Since online banks have few branches
but many loans, they would appear highly responsive. To minimize the influence
of any online banks we may have missed in our classification, this table also shows
versions of the predicted measure at different levels of Winsorization. The coefficients
on the Winsorized measures are larger than the un-Winsorized ones, as we expect,
given that this measure minimizes the influence of online banks. Therefore, we use the
predicted PPP measure that is Winsorized at the bank level at the 95th percentile.
       Table 7 shows the effect of predicted PPP bank lending on the log number of
  32
    An implicit assumption of this approach is that banks are more likely to make PPP loans in ZIP
codes where their branches are located. Table A6 of the Appendix shows that this is true. Column
1 shows the results of bank-by-ZIP code level specifications for each ZIP code where banks have
branches. We also show results from a specification with ZIP code fixed effects in Column 2. The
large positive coefficient in this column means that, within ZIP code, banks with more branches
originate more PPP loans. Finally, Column 3 adds bank fixed effects, so the results are driven by
within-bank, cross-ZIP code variation. The coefficient does not vary much across columns.


                                               32
FinTech loans per establishment. In Column (1), we present the bivariate relationship;
in Column (2), we add the total number of bank branches as an additional control
variable; and, in Column (3), we add further demographic controls. Estimates are
weighted by establishments per ZIP code; but, weighting by PPP loans per ZIP code
or using log(PPP/Establishments) as a dependent variable yields similar results. The
negative, statistically significant coefficients in this table means that ZIP codes with
more responsive traditional banks have a lower number of loans from FinTech lenders.
In ZIP codes where local banks are predicted to make more PPP loans, there is less
FinTech PPP lending per establishment.33
       While the results in this table are statistically significant, the degree of substitution
between FinTechs and traditional banks is economically small. According to our
preferred specification, shown in Column (3), the elasticity of FinTech lending with
respect to predicted traditional bank lending is approximately -0.04. This coefficient
indicates that a 10 percent decrease in traditional bank lending causes approximately
a 0.4 percent increase in FinTech lending. Since FinTech lending is about 15 percent
of overall PPP lending, the number of FinTech loans that are made as a result is only
0.3 percent of the decrease in traditional bank lending.
       This finding indicates that substitution from traditional banks to FinTechs is
not enough to substantially replace traditional bank PPP loans that are not made
because banks were not responsive to the program. We conclude that the FinTech
and traditional bank lending markets are relatively segmented; and FinTech lenders
supply services to a new and relatively distinct market than traditional banks. An
important note, however, is that these estimates come from ZIP codes where bank
branches are located, and they are weighted by the number of branches per ZIP code.
  33
    In Table A8, we show that similar results hold when limited to PPP loans made during Phase
2 of the program, when FinTech lenders had full approval to participate.



                                                33
Substitution may be larger in ZIP codes where fewer branches operate –i.e., the ZIP
codes where FinTech loans are more common.



7     Conclusion

This papers studies whether FinTech lenders provide access to financial services for
regions and borrowers that are not served by the traditional banking system. When
we compare different regions of the country, these FinTech online banks and nonbank
lenders are concentrated in coastal areas and cities — regions that have better access
to banks and better access to financial services.
    Within counties, however, FinTech lenders disproportionately serve industries and
ZIP codes with less access to traditional finance. ZIP codes with fewer bank branches,
lower median income, and larger fraction of the minority population get more of their
PPP loans from these types of new lenders. Across industries, firms in industry codes
that previously got fewer SBA loans were more likely to get their PPP loans from
FinTech lenders. Finally, we show that in ZIP codes where traditional banks keep
branches but end up not participating in PPP origination, only a small fraction of
local small businesses turned to FinTech online banks and nonbanks instead.
    This paper systematically studies the first government program where traditional
banks and FinTech lenders have can be compared in terms of their responsiveness to
the demand by small businesses for exactly the same type of financial service. There-
fore, it has important policy implications, which speak to allowing (more) FinTech
lenders in a timely manner to participate in any type of fully or partially-guaranteed
government loan program (e.g., SBA 7a loans) to increase the efficiency of small-
business lending not only during crises but also in non-crises periods. Understanding
exact mechanisms, whereby FinTech companies are able to provide services to bor-


                                          34
rowers underserved by the traditional banking system, is a valuable topic for future
research. Moreover, in this paper, we have focused on online FinTech lenders which
do not engage in traditional banking. But traditional banks may also use technology-
enabled credit scoring or loan application mechanisms. Whether they do so in a
different way than specialized FinTech lenders is also fruitful area for future research.




                                           35
8     Author Affiliations

Isil Erel
Ohio State University, NBER, and ECGI
    Jack Liebersohn
University of California Irvine




                                    36
References
Abrams, Eliot, “Assessing Bank Deposit Market Power Given Limited Consumer
 Consideration,” 2019. University of Chicago Working Paper.
Autor, David, David Cho, Leland Crane, Mita Goldar, Byron Lutz, Joshua
 Montes, William Peterman, David Ratner, Daniel Villar, and Ahu
 Yildirmaz, “An Evaluation of the Paycheck Protection Program Using Admin-
 istrative Payroll Microdata,” 2020. MIT Working Paper.
Balyuk, Tetyana, Allen Berger, and John Hackney, “What is Fueling Fin-
 Tech Lending? The Role of Banking Market Structure,” 2020. Emory University
 Working Paper.
Banerjee, Abhijit and Esther Duflo, “Do Firms Want to Borrow More? Test-
 ing Credit Constraints Using a Directed Lending Program,” Review of Economic
 Studies, 2014, 81, 572–607.
Barrios, John, Michael Minnis, William Minnis, and Joost Sijthoff, “As-
 sessing the Payroll Protection Program: A Framework and Preliminary Results,”
 2020. University of Chicago Working paper.
Bartik, Alexander W, Marianne Bertrand, Zoë B Cullen, Edward L
 Glaeser, Michael Luca, and Christopher T Stanton, “How are small busi-
 nesses adjusting to covid-19? early evidence from a survey,” Technical Report,
 National Bureau of Economic Research 2020.
Bartik, Timothy J, “Who benefits from state and local economic development
 policies?,” 1991.
Basten, Christoph and Steven Ongena, “The impact of a FinTech platform
 matching mortgage lenders with borrowers online on bank competition, diversifi-
 cation and automation opportunities,” 2020. University of Zurich Working Paper.
Berlin, Mitchell, Greg Nini, and Edison Yu, “Concentration of Control Rights
 in Leveraged Loan Syndicates,” 2018. Federal Reserve Bank of Philadelphia Work-
 ing Paper.
Biswas, Swarnava, Neslihan Ozkan, and Junyang Yin, “Real Effects of Non-
  bank Lending,” 2018. University of Bristol Working Paper.
Brooks, Brian and Charles Calomiris, “Fintech Can Come Out of the Shadows,”
 WSJ Opinion, 2020.
Buchak, Greg, Gregor Matvos, Tomasz Piskorski, and Amit Seru, “Fin-
 tech, regulatory arbitrage, and the rise of shadow banks,” Journal of Financial
 Economics, 2018, 130, 453–483.

                                      37
Butler, Alexander, Jess Cornaggia, and Umit Gurun, “Do Local Capital Mar-
 ket Conditions Affect Consumers’ Borrowing Decisions?,” Management Science,
 2016, 63, 3999–4446.

Carlin, Bruce, Arna Olafsson, and Michaela Pagel, “FinTech and Consumer
 Well-Being in the Information Age,” 2020. UCLA Working Paper.

Chernenko, Sergey, Isil Erel, and Robert Prilmeier, “Why Do Firms Borrow
 Directly from Nonbanks?,” 2019. NBER Working Paper.

Chetty, Raj, John N Friedman, Nathaniel Hendren, and Michael Stepner,
 “How did covid-19 and stabilization policies affect spending and employment? a
 new real-time economic tracker based on private sector data,” Technical Report,
 National Bureau of Economic Research 2020.

Claire, Lelarge, David Sraer, and David Thesmar, Entrepreneurship and Credit
  Constraints Evidence from a French Loan Guarantee Program 2010. NBER Working
  Paper.

Cole, Rebel, Douglas Cumming, and Jon Taylor, “Does FinTech Compete
 with or Complement Bank Finance?,” 2019. Working paper.

Cororaton, Anna and Samuel Rosen, “Public Firm Borrowers of the US Paycheck
 Protection Program,” 2020. SMU Working Paper.

Cortes, Kristle, Yuliya Demyanyk, Lei Li, Elena Loutskina, and Philip
 Strahan, “Stress Tests and Small Business Lending,” Journal of Financial Eco-
 nomics, 2020, 136, 260–279.

Craig, Ben, William Jackson, and James Thomson, “Credit Market Failure
 Intervention: Do Government Sponsored Small Business Credit Programs Enrich
 Poorer Areas?,” Small Business Economics, 2008, 30, 345–360.

Davydiuk, Tetiana, Tatyana Marchuk, and Samuel Rosen, “Direct Lending
 in the U.S. Middle Market,” 2020. Carnegie Mellon University Working Paper.

de Roure, Calebe, Loriana Pelizzon, and Anjan Thakor, “P2P lenders ver-
  sus banks: Cream skimming or bottom fishing?,” 2018. Washington University
  Working Paper.

Diamond, Douglas, “Financial Intermediation and Delegated Monitoring,” The
 Review of Economic Studies, 1984, 51, 393–414.

Fahlenbrach, Rüdiger, Kevin Rageth, and René M Stulz, “How valuable
  is financial flexibility when revenue stops? Evidence from the Covid-19 crisis,”
  Technical Report, National Bureau of Economic Research 2020.

                                       38
Falato, Antonio, Itay Goldstein, and Ali Hortaçsu, “Financial Fragility in the
  COVID-19 Crisis: The Case of Investment Funds in Corporate Bond Markets,”
  Technical Report, National Bureau of Economic Research 2020.

Fuster, A., M. Plosser, P. Schnabl, and J. Vickery, “The role of technology in
  mortgage lending,” Review of Financial Studies, 2019, 32, 1854–1899.

Goldsmith-Pinkham, Paul, Isaac Sorkin, and Henry Swift, “Bartik Instru-
 ments: What, When, Why, and How,” American Economic Review, 2020.

Gopal, Manasa and Philipp Schnabl, “The Rise of Finance Companies and
 FinTech Lenders in Small Business Lending,” 2020. New York University Working
 Paper.

Granja, Joao, Christos Makridis, Constantine Yannelis, and Eric Zwick,
 “Did the Paycheck Protection Program Hit the Target?,” 2020. University of
 Chicago Working Paper.

Green, Daniel and Erik Loualiche, “Local Government Finances and Balanced
 Budgets in the COVID-19 Crisis,” Available at SSRN 3651605, 2020.

Halling, Michael, Jin Yu, and Josef Zechner, “How Did COVID-19 Affect
 Firms’ Access to Public Capital Markets?,” The Review of Corporate Finance Stud-
 ies, 2020.

Hanson, Samuel G., Andrei Shleifer, Jeremy C. Stein, and Robert W.
 Vishny, “Banks as patient fixed-income investors,” Journal of Financial Eco-
 nomics, 2015, 117, 449–469.

Harvey, Campbell, Joey Santoro, and Ashwin Ramachandran, “DeFi and
 the Future of Finance,” 2020. Duke University Working Paper.

Irani, Rustom, Rajkamal Iyer, Ralf Meisenzahl, and Jose-Luis Peydro,
  “The Rise of Shadow Banking: Evidence from Capital Regulation,” 2020. Working
  Paper.

Ivashina, Victoria and Zheng Sun, “Institutional demand pressure and the cost
  of corporate loans,” Journal of Financial Economics, 2011, 99, 500–522.

Jiang, Wei, Kai Li, and Pei Shao, “When shareholders are creditors: Effects of the
  simultaneous holding of equity and debt by non-commercial banking institutions,”
  Review of Financial Studies, 2010, 23, 3595–3637.

Li, Lei and Philip Strahan, “Who Supplies PPP Loans (And Does It Matter)?
  Banks, Relationships and the COVID Crisis,” 2020. Boston University Working
  Paper.

                                       39
Liebersohn, Jack, “Are Retail Banking Markets Still Local? Quasi-Experimental
  Evidence from Bank Mergers,” 2020. Ohio State University Working Paper.

Lim, Jongha, Bernadette A. Minton, and Michael S. Weisbach, “Syndi-
  cated loan spreads and the composition of the syndicate,” Journal of Financial
  Economics, 2014, pp. 45–69.

Maggio, Marco Di and Vincent Yao, “FinTech borrowers: Lax-Screening or
 cream-skimming?,” 2018. HBS Working Paper.

Manson, Steven, Jonathan Schroeder, David Van Riper, and Steven Rug-
 gles, “IPUMS National Historical Geographic Information System: Version 12.0
 [Database],” Minneapolis: University of Minnesota, 2017, 39.

Massoud, Nadia, Debarshi Nandy, Anthony Saunders, and Keke Song, “Do
 hedge funds trade on private information? Evidence from syndicated lending and
 short-selling,” , Journal of Financial Economics, 2011, 99, 477–499.

Mills, Karen and Annie Dang, “Building Small Business Utopia: How Artificial
 Intelligence and Big Data Can Increase Small Business Success In Big Data in
 Small Business,” 2020. Harvard Business School Working Paper.

Morse, Adair, “Peer-to-Peer Crowdfunding: Information and the Potential for Dis-
 ruption in Consumer Lending,” Annual Review of Financial Economics, 2015, 7,
 463–482.

Nadauld, Taylor D. and Michael S. Weisbach, “Did securitization affect the
 cost of corporate debt?,” Journal of Financial Economics, 2012, 105, 332–352.

Pastor, Lubos and M Blair Vorsatz, “Mutual fund performance and flows during
 the COVID-19 crisis,” Technical Report, National Bureau of Economic Research
 2020.

Petersen, Mitchell and Raghuram Rajan, “The Benefits of Lending Relation-
 ships: Evidence from Small Business Data,” Journal of Finance, 1994, 1, 3–37.

Philippon, Thomas, “Has the US Finance Industry Become Less Efficient? On
 the Theory and Measurement of Financial Intermediation,” American Economic
 Review, 2015, 105, 1408–1438.

Stulz, Rene, “FinTech, BigTech, and the Future of Banks,” Journal of Applied
  Corporate Finance, 2019, 31, 86–97.

Tang, H., “Peer-to-Peer Lenders versus Banks: Substitutes or Complements?,” The
  Review of Financial Studies, 2019, 32, 1900–1938.



                                      40
Vallee, B. and Y. Zeng, “Marketplace lending: A new banking paradigm?,” Review
 of Financial Studies, 2019, 32, 1939–1982.

Yanofsky, David, “Here’s what we know is wrong with the PPP data,” 2020.




                                     41
Figures

                                                              PPP Origination by Bank Type
                               PPP Phase 1     PPP Phase 2

                 50                                                                                         .5




  Total Daily # PPP (10,000's)

                                                                                             FinTech Fraction of Total, Weekly Avg



         20         30         40

                 10                                                                           .1        .2           .3         .4



                 0
                              Apr 01         May 01          Jun 01   Jul 01   Aug 01                       0
                                                               date                                                            Apr 01   Apr 29   May 27   Jul 01   Jul 29
                                                                                                                                                   week
                                             Small Bank/CU            Large Bank
                                             FinTech                                                                                      Nonbanks          Online Banks




Figure 1: PPP Lending by Day and Type of Institution. Panels show the number of
PPP loans given by various lenders daily between April 3 and August 8, 2020. “Large
banks” are banks with more than $20 billion in assets, “Small Bank/CU” includes
all other lenders except FinTech, including unclassified lenders. Source: Calculated
from SBA PPP Loan Databse.




                                                                                        42
                .3




Frac of PPP FinTech Loans in State




         .1            .2



                0
                                     0                 20               40             60                       80              100
                                                                      Google Search Intensity
                                     Source: Google Trends. Searches for ''apply ppp loan online'' from 3/1/20 to 8/8/20.
                                     Each circle represents a state. Weighted by the total number of PPP loans in each state.



                                             Figure 2: Google Searches for Online PPP Loans.




                                                                               43
                    FinTech Lending by County




                                                       13.96 − 59.87
                                                       8.78 − 13.96
                                                       5.85 − 8.78
                                                       3.59 − 5.85
                                                       1.49 − 3.59
                                                       0.00 − 1.49
                                                       No data


Figure 3: Fraction of PPP Loans from FinTech Lenders, U.S. Counties




                                44
             Median Income                 FinTech Lending by ZIP Code




98,358 − 212,750                       57.5 − 100.0
80,282 − 98,358                        37.3 − 57.5
65,590 − 80,282                        28.3 − 37.3
56,239 − 65,590                        22.1 − 28.3
44,150 − 56,239                        15.8 − 22.1
20,991 − 44,150                        0.0 − 15.8


Figure 4: Fraction of PPP Loans from FinTech Lenders, Chicago ZIP Codes.




                                  45
                                           PPP Loans and Log Bank Branches
                                             National                                                                Within-County
              .2                                                                           .2




  Fraction FinTech PPP Loans                                                   Fraction FinTech PPP Loans




 .14           .16        .18                                                 .14           .16        .18




              .12                                                                          .12
                                0          .5           1          1.5                                       0         .5           1          1.5
                                     Log-10 Bank Branches in ZIP                                                 Log-10 Bank Branches in ZIP

Source: Calculated from FDIC Summary of Deposits and SBA PPP Database.
Weighted by PPP loans per Zip code.



                                    Figure 5: Lending Share and Log Bank Branches.




                                                                         46
Tables

                Table 1: Summary Statistics by ZIP Code


                                 Mean     Std. Dev    Median    Count
   Frac Nonbank                  0.07        0.06      0.05   36,675
   Frac Online Bk                0.10        0.07      0.09   36,675
   Total FinTech PPP Fraction    0.16        0.12     0.15    36,675
   Frac Bk/CU                    0.83        0.12      0.84   36,675
   Num. PPP Lns                628.56       534.53   515.00   36,675
   Median Income              70,469.32   29,818.04 63,945.00 30,002
   Frac. 45m+ Commute           0.17         0.10      0.14   31,190
   Frac. White                   0.74        0.21      0.80   31,501
   Total Pop                  13,735.63    9,091.73 12,730.00 31,565
   Num. Bk Branches             9.44         7.21     8.00    19,404
   Avg COVID Case Rate           0.02        0.04      0.00   35,749
   Unemp. Growth                3.35         1.83      3.01   17,176
   Num. Estabs                 924.14       782.22   771.00   33,955
   Weighted by PPP loans per zip code. Bank branches are for ZIP codes
  that have at least one branch. Unemployment data is not available for
  all regions. Source: Calculated from SBA PPP database, FDIC SOD,
  Decennial Census/ACS, County Business Patterns.




                                   47
                           Table 2: Geographic Correlates of PPP Provision

                                     (1)        (2)        (3)        (4)        (5)        (6)
                                   Log PPP    Log PPP    Log PPP    Log PPP    Log PPP    Log PPP
                                   FinTech    Trad. Bk   Local Bk   FinTech    Trad. Bk   Local Bk
                                   Phase 1    Phase 1    Phase 1    Phase 2    Phase 2    Phase 2

      Avg Case Rate                 -0.035    -4.17∗∗∗   -2.97∗∗∗   4.94∗∗∗    0.28∗∗∗    0.54∗∗∗
                                    (0.25)     (0.23)     (0.32)    (0.19)     (0.077)    (0.15)
      Change in Unemployment       0.072∗∗∗ -0.071∗∗∗    -0.14∗∗∗    0.14∗∗∗   0.033∗∗∗   0.020∗∗∗
                                   (0.0078) (0.0056)     (0.012)    (0.0088)   (0.0025)   (0.0046)
      Log Establishments           1.07∗∗∗     0.89∗∗∗   1.00∗∗∗     1.07∗∗∗    0.97∗∗∗    1.24∗∗∗
48                                 (0.015)    (0.0066)   (0.017)    (0.0086)   (0.0033)   (0.0077)
      Constant                     -5.27∗∗∗   -0.85∗∗∗   -2.37∗∗∗   -3.47∗∗∗   -0.97∗∗∗   -3.41∗∗∗
                                   (0.094)    (0.040)     (0.10)    (0.055)    (0.020)    (0.049)
      Observations                  6432       14724      7742       10866      15639      8224
      R2                            0.562      0.748      0.392      0.714      0.913      0.778
      ZIP code level specifications showing the relationship between COVID-19 shock and the
     degree of PPP origination, for traditional banks and for FinTech lenders. “Local Banks” are
     traditional banks, where the sample of PPP loans is limited to those coming from a bank
     that has a branch in the same ZIP code as the PPP loan. Robust standard errors. Estimates
     are weighted by PPP loans per ZIP code. Data calculated from Chetty et al. (2020) and
     SBA PPP loan data.
Table 3: Number of Branches in the ZIP Code and FinTech PPP Lend-
ing Shares

                             (1)         (2)        (3)         (4)
                           FinTech      Bank      FinTech      Bank
                             PPP         PPP        PPP         PPP
                           Fraction    Fraction   Fraction    Fraction

 Log Branches              -0.019∗∗∗   0.020∗∗∗   -0.025∗∗∗   0.026∗∗∗
                           (0.0010)    (0.0010)   (0.0013)    (0.0013)
 Log Med. Inc                                     -0.012∗∗∗   0.014∗∗∗
                                                  (0.0035)    (0.0035)
 Frac Commute 45+m                                0.19∗∗∗     -0.20∗∗∗
                                                  (0.014)     (0.014)
 Frac White                                       -0.18∗∗∗     0.18∗∗∗
                                                  (0.0089)    (0.0088)
 Log Population                                   0.025∗∗∗    -0.025∗∗∗
                                                  (0.0011)    (0.0012)
 Establishments Per Cap.                           -0.0049     0.0059
                                                  (0.0051)    (0.0054)
 Constant                   0.20∗∗∗     0.79∗∗∗   0.22∗∗∗     0.75∗∗∗
                           (0.0022)    (0.0022)   (0.035)     (0.035)
 Observations               35937       35937      28850       28850
 R2                         0.706       0.684      0.808       0.790
 County FEs                   X           X          X           X
 ZIP code level specifications showing the relationship between the
share of FinTech PPP lending and the number of bank branches per
ZIP code. Robust standard errors. Estimates are weighted by PPP
loans per ZIP code. Data calculated from Chetty et al. (2020) and
SBA PPP loan data.




                                 49
     Table 4: Increases in County Unemployment and PPP Origination

                                    (1)        (2)         (3)          (4)
                                  FinTech    FinTech     FinTech      FinTech
                                    PPP        PPP         PPP          PPP
                                  Fraction   Fraction    Fraction     Fraction

 Case Rate                        0.89∗∗∗
                                  (0.24)
 Log Branches                    -0.017∗∗∗   -0.017∗∗∗   -0.018∗∗∗   -0.00074
                                 (0.0028)    (0.0031)    (0.0039)    (0.0080)
 Case Rate × Log Branches                    -0.097∗∗
                                             (0.047)
 Unemp. Chg                                              0.016∗∗∗
                                                         (0.0047)
 Unemp. Chg × Log Branches                                           -0.0054∗∗∗
                                                                      (0.0018)
 Constant                         0.18∗∗∗     0.20∗∗∗    0.16∗∗∗      0.22∗∗∗
                                 (0.0068)    (0.0060)    (0.016)     (0.0085)
 Observations                     35727       35749       17173        17176
 R2                               0.604       0.702       0.588        0.713
 County FEs                                     X                        X
 CZ FEs                              X                      X
 Specifications showing the relationship between the share of FinTech PPP
lending and ZIP code level statistics interacted with the change in unemploy-
ment and change in case rate. Columns (1) and (3) use Commuting Zone fixed
effects because the case rate and change in unemployment are observed at the
county level. Standard errors clustered by county. Estimates are weighted
by PPP loans per ZIP code. Data calculated from the FDIC Summary of
Deposits database and SBA PPP loan data.




                                     50
                   Table 5: FinTech Lending and Firm Relationships

                              (1)       (2)        (3)        (4)        (5)       (6)
                           FinTech    FinTech    FinTech    FinTech    FinTech   FinTech
                            PPP        PPP        PPP        PPP        PPP       PPP
                            Loan       Loan       Loan       Loan       Loan      Loan

 Log(PPP/SBA 7a)            0.023∗∗   0.029∗∗∗
                           (0.0097)   (0.011)
 One-Pers. Firm                                  0.22∗∗∗    0.21∗∗∗
                                                 (0.016)    (0.013)
 Industry Emp. Growth                                                  -0.22∗∗   -0.33∗∗∗
                                                                       (0.085)   (0.081)
 Constant                  0.076∗∗     0.050      0.12∗∗∗    0.12∗∗∗   0.35∗∗∗   0.44∗∗∗
                           (0.037)    (0.044)    (0.0055)   (0.0052)   (0.076)   (0.068)
 Observations              4863168    4863168    5076115    5076115    4035347   4035347
 R2                         0.113      0.133      0.161      0.173      0.121     0.142
 Zip FEs                      X          X          X          X          X         X
 NAICS2 FEs                              X                     X
 Loan-level specifications showing the relationship between FinTech loans and proxies for
borrowers’ ability to rely on loan relationships. Dependent variable is an indicator equal
to 1 if a loan is originated by a FinTech firm. Log(PPP/SBA 7a) measures the number
of PPP loans scaled by the number of SBA 7(a) loans in the years 2018-2019, measured
at the NAICS 6-digit level. One-Pers. firm is an indicator for sole proprieterships and
individuals as indicated in the PPP data. Industry emp. growth measures employment
growth by 3-digit NAICS industry between the March and April Current Employment
Statistics. Standard errors double-clustered by NAICS 6-digit industry and ZIP code.




                                         51
         Table 6: Local Technology Use and PPP Lending

                              (1)        (2)         (3)        (4)
                            FinTech    FinTech     FinTech    FinTech
                              PPP        PPP         PPP        PPP
                            Fraction   Fraction    Fraction   Fraction

 7(a) Share                 0.0048      0.012∗∗
                           (0.0075)    (0.0056)
 Fraction w Desktop                                -0.25∗∗∗   0.071∗∗∗
                                                   (0.014)    (0.020)
 Log Med. Inc                          -0.012∗∗∗              -0.026∗∗∗
                                       (0.0038)               (0.0044)
 Frac Commute 45+m                     0.20∗∗∗                0.19∗∗∗
                                       (0.017)                (0.014)
 Frac White                            -0.18∗∗∗               -0.18∗∗∗
                                       (0.0097)               (0.0094)
 Log Branches                          -0.026∗∗∗              -0.025∗∗∗
                                       (0.0014)               (0.0013)
 Log Population                        0.028∗∗∗               0.024∗∗∗
                                       (0.0015)               (0.0012)
 Establishments Per Cap.                -0.0014                -0.0059
                                       (0.0050)               (0.0054)
 Constant                   0.17∗∗∗    0.19∗∗∗     0.36∗∗∗    0.33∗∗∗
                           (0.0011)    (0.040)     (0.011)    (0.040)
 Observations                15376      14946      31316       28846
 R2                          0.690      0.812      0.720       0.809
 County FEs                    X          X          X           X
 Estimates show the relationship between ZIP-level measures of tech-
nology use and the FinTech fraction of PPP loans. “7(a) Share” mea-
sures the fraction of 7(a) loans in the ZIP code coming from FinTech
lenders as identified in the PPP data in the years 2018-2019. “Frac-
tion w Desktop” is the fraction of households in the ZIP code with
a desktop computer, as reported in the 2014-2018 ACS. Data cal-
culated from SBA 7(a) and PPP data, and US Census/ACS data.
Robust standard errors.



                                 52
Table 7: Effect of Predicted Bank PPP on Total PPP Per Establish-
ment

                             (1)              (2)              (3)
                        FinTech PPP/     FinTech PPP/     FinTech PPP/
                        Establishments   Establishments   Establishments

 Log Pred. PPP            -0.10∗∗∗         -0.043∗∗         -0.039∗∗
                          (0.021)          (0.020)          (0.018)
 Bk Branches                               -0.16∗∗∗         -0.24∗∗∗
                                           (0.0086)         (0.0093)
 Log Med. Inc.                                               0.18∗∗∗
                                                             (0.021)
 Frac Commute 45+m                                           1.00∗∗∗
                                                             (0.086)
 Frac White                                                 -0.79∗∗∗
                                                            (0.049)
 Log Pop                                                     0.20∗∗∗
                                                            (0.0088)
 Constant                 -2.26∗∗∗         -2.12∗∗∗         -5.35∗∗∗
                          (0.077)          (0.075)           (0.22)
 Observations              12695            12695            12547
 R2                        0.812            0.822            0.863
 County FEs                  X                X                X
 ZIP code level specifications showing the relationship between the
share of PPP lending per establishment and the predicted level of
PPP lending based on banks’ overall degree of PPP lending. Ro-
bust standard errors. Estimates are weighted by establishments
per ZIP code. Weighting by PPP loans per ZIP code or using
log(PPP/Establishments) as a dependent variable yields similar re-
sults. Data calculated from ZIP business patterns and SBA PPP loan
data.




                                   53
A        Appendix

                                               Business Financial Products Advertising

                                                                                                                    20000




     Monthly Avg, by Quarter, Dols 000's                                                                            15000




                  500               1000
                                                                                                                    10000




                      0                                                                                             5000
                                           2017q3   2018q1       2018q3     2019q1      2019q3      2020q1     2020q3
                                                                          Quarter

                                                    Fintech (left axis)              Traditional Bks (right axis)


Figure A1: Commercial and Business Financial Services Advertising Dollars by Fin-
Tech and Non-Fintech Lenders. “FinTech” includes the top-30 FinTech PPP lenders.
Monthly averages by quarter through July, 2020. Source: Compiled using data from
Kantar Media.




                                                                           54
       Table A1: Summary Statistics by ZIP Code (unweighted)


                                Mean     Std. Dev    Median    Count
 Frac Nonbank                  0.05         0.10        0.01   36,675
 Frac Online Bk                0.06         0.10        0.02   36,675
 Total FinTech PPP Fraction    0.10        0.15        0.06    36,675
 Frac Bk/CU                    0.88         0.16        0.92   36,675
 Num. PPP Lns                 142.12      262.93       26.00   36,675
 Median Income              59,457.12    25,252.21   54,286.00 30,002
 Frac. 45m+ Commute            0.17        0.12         0.15   31,190
 Frac. White                   0.83         0.20        0.92   31,501
 Total Pop                   4,558.81    6,742.19    1,306.00 31,565
 Num. Bk Branches              4.53        4.90        3.00    19,404
 Avg COVID Case Rate           0.01        0.02        0.00    35,749
 Unemp. Growth                 2.95         1.79        2.63   17,176
 Num. Estabs                  230.15      403.82       45.00   33,955
Bank branches are for ZIP codes that have at least one branch. Unem-
ployment data is not available for all regions. Source: Calculated from
SBA PPP database, FDIC SOD, Decennial Census/ACS, County Business
Patterns.




                                  55
Table A2: Number of Branches in the ZIP Code and PPP Lending
Shares, Alternative Measure of FinTech

                              (1)        (2)        (3)         (4)
                            FinTech    FinTech    FinTech     FinTech
                              PPP        PPP        PPP         PPP
                            Fraction   Fraction   Fraction    Fraction

 Log Branches              -0.020∗∗∗   0.020∗∗∗   -0.027∗∗∗   0.027∗∗∗
                           (0.0010)    (0.0010)   (0.0013)    (0.0013)
 Log Med. Inc                                     -0.014∗∗∗   0.014∗∗∗
                                                  (0.0035)    (0.0035)
 Frac Commute 45+m                                0.20∗∗∗     -0.20∗∗∗
                                                  (0.014)     (0.014)
 Frac White                                       -0.18∗∗∗     0.18∗∗∗
                                                  (0.0089)    (0.0089)
 Log Population                                   0.026∗∗∗    -0.026∗∗∗
                                                  (0.0012)    (0.0012)
 Establishments Per Cap.                           -0.0056     0.0056
                                                  (0.0053)    (0.0053)
 Constant                   0.20∗∗∗     0.80∗∗∗   0.23∗∗∗     0.77∗∗∗
                           (0.0022)    (0.0022)   (0.035)     (0.035)
 Observations                35884      35884      28834       28834
 R2                          0.684      0.684      0.798       0.798
 County FEs                    X          X          X           X
 ZIP code level specifications showing the relationship between the
share of FinTech PPP lending and ZIP code level statistics, including
the log number of bank branches. This table uses a narrow measure
of FinTech lenders as discussed in the text. Robust standard errors.
Estimates are weighted by PPP loans per ZIP code. Data calculated
from Chetty et al. (2020) and SBA PPP loan data.




                                 56
Table A3: Number of Branches in the ZIP Code and PPP Lending Shares,
Online and Nonbank Lenders

                                (1)           (2)         (3)          (4)
                             Online Bk     Nonbank     Online Bk     Nonbank
                                PPP          PPP          PPP          PPP
                              Fraction     Fraction     Fraction     Fraction

 Log Branches               -0.0063∗∗∗     -0.012∗∗∗   -0.010∗∗∗    -0.015∗∗∗
                            (0.00054)      (0.00057)   (0.00068)    (0.00070)
 Log Med. Inc                                          -0.0065∗∗∗   -0.0058∗∗∗
                                                        (0.0020)     (0.0018)
 Frac Commute 45+m                                     0.086∗∗∗      0.10∗∗∗
                                                       (0.0083)     (0.0075)
 Frac White                                            -0.070∗∗∗    -0.11∗∗∗
                                                       (0.0051)     (0.0046)
 Log Population                                         0.013∗∗∗     0.012∗∗∗
                                                       (0.00063)    (0.00064)
 Establishments Per Cap.                                -0.0038      -0.0011
                                                       (0.0026)     (0.0029)
 Constant                     0.11∗∗∗      0.093∗∗∗     0.11∗∗∗      0.11∗∗∗
                             (0.0011)      (0.0012)     (0.020)      (0.019)
 Observations                 35937         35937        28850        28850
 R2                           0.740         0.618        0.805        0.743
 County FEs                     X             X            X            X
 ZIP code level specifications showing the relationship between the share of
online and nonbank PPP lending and ZIP code level statistics, including
the log number of bank branches. Robust standard errors. Estimates are
weighted by PPP loans per ZIP code. Data calculated from Chetty et al.
(2020) and SBA PPP loan data.




                                      57
Table A4: Bank Branch Density and Fraction of
Online/Nonbank Loans, Loan Level Estimates

                               (1)         (2)
                             FinTech     FinTech
                              PPP         PPP
                              Loan        Loan

 Log Bk Branches            -0.022∗∗∗   -0.019∗∗∗
                            (0.0012)    (0.00096)
 Frac White                 -0.21∗∗∗    -0.18∗∗∗
                            (0.0068)    (0.0056)
 Frac Commute 45+m           0.14∗∗∗     0.12∗∗∗
                             (0.013)     (0.011)
 Log Pop                     0.025∗∗∗    0.019∗∗∗
                            (0.00092)   (0.00077)
 Log Est Size               -0.022∗∗∗   -0.014∗∗∗
                            (0.0016)    (0.0013)
 Constant                    0.16∗∗∗     0.18∗∗∗
                             (0.010)    (0.0086)
 Observations               5116435     4988784
 R2                          0.080       0.172
 County FEs                    X           X
 NAICS FEs                                 X
 Loan level specifications showing the relation-
ship between the likelihood that a PPP loan
is from an online bank/nonbank, and ZIP code
level statistics, including the log number of bank
branches. Standard errors clustered by ZIP
Code. Data calculated from the SBA PPP Loan
Database, ZIP Business Patterns, the 2014-2018
ACS and the 2010 Decennial Census.




                       58
             Table A5: Predicted Bank Lending and the Economic Shock

                                (1)         (2)          (3)            (4)
                              Average     Average    Unemployment   Unemployment
                             Case Rate   Case Rate      Growth         Growth

      Log Pred. PPP          -0.000082    -0.00055      -0.030         -0.026
                             (0.00043)   (0.00044)     (0.050)        (0.050)
      Bk Branches                        0.0012∗∗∗                     -0.011
                                         (0.00024)                    (0.022)

59    Constant                0.013∗∗∗   0.012∗∗∗      3.24∗∗∗        3.25∗∗∗
                              (0.0016)   (0.0016)      (0.18)         (0.18)
      Observations             13690      13690         6819           6819
      R2                       0.852      0.853         0.792          0.792
      Commuting Zone FEs                    X                            X
      ZIP code-by-bank level specifications showing the relationship between the
     predicted number of bank PPP loans, and the magnitude of the COVID-19
     economic shock. Robust standard errors. Estimates are weighted by PPP
     loans per ZIP code. Data calculated from the FDIC Summary of Deposits
     database, SBA PPP loan data, and data from Chetty et al. (2020).
Table A6: PPP Lending and Bank’s Own
Branch Locations

                  (1)       (2)       (3)
                Log PPP   Log PPP   Log PPP
                 Loans     Loans     Loans

 Bk Branches    2.12∗∗∗   1.87∗∗∗   1.99∗∗∗
                (0.070)   (0.099)   (0.046)
 Constant       1.36∗∗∗   1.64∗∗∗   1.47∗∗∗
                (0.086)   (0.093)   (0.036)
 Observations    58556     52360     57885
 R2              0.147     0.399     0.438
 Bank FEs                              X
 Zip FEs                     X
 ZIP code-by-bank level specifications show-
ing the relationship between the number of
PPP loans from a given bank, and the num-
ber of branches that that bank has in the ZIP
code. Robust standard errors. Estimates are
weighted by PPP loans per ZIP code. Data
calculated from the FDIC Summary of De-
posits database and SBA PPP loan data.




                     60
                      Table A7: Effect of Predicted Bank Lending on Overall Lending

                                             (1)       (2)        (3)        (4)       (5)        (6)
                                          Log PPP    Log PPP    Log PPP   Log PPP    Log PPP   Log PPP
                                           Loans      Loans      Loans     Loans      Loans     Loans

      Bk Branches                          0.83∗∗∗    0.72∗∗∗   0.83∗∗∗    0.72∗∗∗   0.83∗∗∗    0.72∗∗∗
                                           (0.012)   (0.0099)   (0.012)   (0.0099)   (0.012)   (0.0099)
      Log Predicted PPP (Winsor 90)        0.51∗∗∗   0.15∗∗∗
                                           (0.026)   (0.028)
      Log Predicted PPP (Winsor 95)                             0.42∗∗∗   0.12∗∗∗
                                                                (0.023)   (0.025)
61    Log Predicted PPP (Winsor 100)                                                 0.32∗∗∗   0.093∗∗∗
                                                                                     (0.021)   (0.021)
      Constant                             2.55∗∗∗   4.15∗∗∗    2.87∗∗∗   4.25∗∗∗    3.24∗∗∗    4.34∗∗∗
                                           (0.095)   (0.10)     (0.088)   (0.092)    (0.078)    (0.077)
      Observations                         13841      13814      13841     13814      13841     13814
      R2                                   0.510      0.789      0.507     0.789      0.502     0.789
      County FEs                                        X                    X                    X
      ZIP code-by-bank level specifications showing the relationship between the predicted number of
     bank PPP loans, and the total overall number of PPP loans. The PPP measures are Winsorized
     at the bank level before weighted averages are calculated at the ZIP code level, as discussed in the
     text. Robust standard errors. Estimates are weighted by PPP loans per ZIP code. Data calculated
     from the FDIC Summary of Deposits database and SBA PPP loan data.
Table A8: Effect of Predicted Bank PPP on Total PPP Per Establish-
ment, Phase 2 Only

                             (1)              (2)              (3)
                        FinTech PPP/     FinTech PPP/     FinTech PPP/
                        Establishments   Establishments   Establishments

 Log Pred. PPP             -0.11∗∗∗         -0.041∗          -0.036∗
                           (0.022)          (0.021)          (0.019)
 Bk Branches                               -0.18∗∗∗         -0.27∗∗∗
                                           (0.0090)         (0.0096)
 Log Med. Inc.                                               0.16∗∗∗
                                                             (0.022)
 Frac Commute 45+m                                           1.08∗∗∗
                                                             (0.091)
 Frac White                                                 -0.84∗∗∗
                                                            (0.050)
 Log Pop                                                     0.22∗∗∗
                                                            (0.0092)
 Constant                  -2.34∗∗∗        -2.19∗∗∗         -5.40∗∗∗
                           (0.082)         (0.079)           (0.22)
 Observations               12574           12574            12433
 R2                         0.802           0.814            0.860
 County FEs                   X               X                X
 ZIP code level specifications showing the relationship between the
share of PPP lending per establishment and the predicted level of
PPP lending based on banks’ overall degree of PPP lending. Depen-
dent variable is limited to PPP loans made during Phase 2 of the
program. Robust standard errors. Estimates are weighted by estab-
lishments per ZIP code. Weighting by PPP loans per ZIP code or
using log(PPP/Establishments) as a dependent variable yields similar
results. Data calculated from ZIP business patterns and SBA PPP
loan data.




                                   62
B     Lender Classifications

B.1   Non-Bank FinTech Lenders




                           63
Nonbank Lender                              PPP Loans

Kabbage, Inc.                                196402
Readycap Lending, LLC                         34261
MBE Capital Partners                          23945
Intuit Financing Inc.                         19086
Fundbox, Inc.                                 14281
Newtek Small Business Finance, Inc.           11677
New York Business Development Corporation      6468
FC Marketplace, LLC (dba Funding Circle)      6235
Harvest Small Business Finance, LLC            5353
CDC Small Business Finance Corporation         4095
BSD Capital, LLC dba Lendistry                 4076
Itria Ventures LLC                             3556
Fountainhead SBF LLC                           3453
Hope Enterprise Corporation                    2869
Accion                                         2483
CRF Small Business Loan Company, LLC           2398
Fund-Ex Solutions Group, LLC                   1416
Montana Community Development Corp.            1276
Mortgage Capital Development Corporation       1122
LiftFund, Inc.                                 1036
Opportunity Fund Community Development          990
Prestamos CDFI, LLC                             938
Centerstone SBA Lending, Inc.                  898
Trenton Business Assistance Corporation        894
Benworth Capital                                779
Colorado Enterprise Fund                       779
Arkansas Capital Corporation                    771
Grow America Fund, Incorporated                 715
Colorado Lending Source, Ltd.                   612
Accion East, Inc.                               560
American Lending Center                         555
Hana Small Business Lending, Inc.               504




                                   64
B.2   Online Bank FinTech Lenders

Online Bank Lender                                        PPP Loans

Cross River Bank                                           198738
Celtic Bank Corporation                                    147317
WebBank                                                     76578
Capital One, National Association                           15772
Live Oak Banking Company                                    11045
American Express National Bank                               6964
Signature Bank                                               6311
Radius Bank                                                  6224
First Bank of the Lake                                       4199
Transportation Alliance Bank, Inc. d/b/a TAB Bank, Inc.     1687
Pacific Enterprise Bank                                      1388
The Bancorp Bank                                             1288
Savoy Bank                                                   1165
American Business Bank                                       1072
First Secure Bank and Trust Co.                              1033
Union National Bank                                           956
Ally Bank                                                     943
T Bank, National Association                                  922
Primary Bank                                                  918
Endeavor Bank                                                 860
The MINT National Bank                                        859
Axos Bank                                                     853
LCA Bank Corporation                                          825
Signature Bank, National Associatioin                         777
Vinings Bank                                                  740




                                     65
Online Bank Lender                        PPP Loans

FinWise Bank                                 699
St. Louis Bank                               674
Solera National Bank                         665
Fresno First Bank                            656
Optus Bank                                   641
Continental Bank                             635
Loyal Trust Bank                             626
First Command Bank                           621
Lexicon Bank                                 616
Keystone Bank, National Association          612
Chain Bridge Bank, National Association      588
The Victory Bank                             579
Bankers’ Bank of Kansas                      559
Bank of San Francisco                        542
Buckeye State Bank                           531
Beacon Community Bank                        522
Small Business Bank                          520
New Valley Bank and Trust                    519
TIAA Bank, A Division of                     273




                                     66
B.3   Online Bank FinTech Lenders, Alternative Definition

Lender                                   PPP Loans

Cross River Bank                          198738
Celtic Bank Corporation                   147317
WebBank                                    76578
Capital One, National Association          15772
American Express National Bank              6964
Radius Bank                                6224
The Bancorp Bank                            1288
Ally Bank                                    943
Axos Bank                                    853
OneWest Bank, A Division of                 840
FinWise Bank                                699
First Internet Bank of Indiana               447
TIAA Bank, A Division of                    273
Green Dot Bank                                17




                                    67
C    Variable Definitions
   Variable              Source                               Description
Frac Nonbank       SBA PPP Database             Frac from lenders with above 750 loans
                                                with no match in the FFIEC Attributes
                                                File or otherwise using news reports.
Frac Online Bk     SBA PPP Database             Frac from lenders with one branch and
                                                at least 500 PPP loans or identified using
                                                news reports.
Total FinTech      SBA PPP Database             Fraction from either nonbank or online
PPP Fraction                                    bank PPP lenders.
One-Pers. Firm     SBA PPP Database             Indicator equal to 1 for borrowers with
                                                business type “Sole Proprietorship” or
                                                “Self-Employed Individuals”
Frac Traditional   SBA PPP Database             Fraction of loans from non-FinTech
Bk/CU                                           lenders that match to FFIEC Attributes
                                                file.
Median Income      American     Community       Median Income
                   Survey, 2014-2018
Frac     45m+      American     Community       Fraction of HHs with commute time
Commute            Survey, 2014-2018            greater than or equal to 45 minutes
Frac. White        American     Community       Fraction of individuals reporting “White”
                   Survey, 2014-2018            as only race
Total Pop          American     Community       Total population
                   Survey, 2014-2018
Fraction w Desk-   American     Community       Fraction of HHs with desktop computer
top                Survey, 2014-2018
Num.        Bk.    FDIC Summary of De-          Number of bank branches in ZIP code
Branches           posits Database
Avg      COVID     The New York Times, as       Avg number of active COVID cases per
Case Rate          collected by Chetty et al.   100 people in March, by county
                   (2020)
Unempl.            State Agencies, as col-      Four-week change in unemployment in-
Growth             lected by Chetty et al.      surance claims as of April 11, 2020, by
                   (2020)                       County
Num. Estabs        ZIP Business Patterns        Number of establishments in ZIP Code,
                                                2017
PPP/SBA 7(a)       SBA PPP        and   7(a)    Ratio of PPP loans to SBA 7(a) loans
                   databases                    from 2018-2019, by NAICS 5-digit indus-
                                                try
                                                                 Continued on next page


                                        68
                    Table A9 – continued from previous page
   Variable               Source                        Description
Fraction of 7(a)   SBA 7(a) data          Fraction of 7(a) loans where lender name
loans from Fin-                           matches a PPP FinTech lender
Tech




                                      69


File and source

File

does-fintech-substitute-for-banks-evidence-from-the-paycheck-protection-program.pdf

Size

879,118 bytes

SHA-256

d17ee5df34955adbab97b7c4a77a63711b34168bdcaf8402c0080adc11d17aee

Our copy

does-fintech-substitute-for-banks-evidence-from-the-paycheck-protection-program.pdf

Original

www.nber.org